EP3343496A1

EP3343496A1 - Method and system for energy management in a facility

Info

Publication number: EP3343496A1
Application number: EP16207155.9A
Authority: EP
Inventors: Domen Zupancic; Devid Palcic; Matjaz Gams; Jernej Zupancic
Original assignee: Robotina d o o
Current assignee: Robotina d o o
Priority date: 2016-12-28
Filing date: 2016-12-28
Publication date: 2018-07-04

Abstract

A method for energy management in a facility with a plurality of energy sources and/or electrical appliances comprises the steps of obtaining a plurality of operating parameters relating to energy production of said energy sources and/or energy consumption of said electrical appliances in said facility. Performing a plurality of optimization calculations for said energy management of said energy sources and/or electrical appliances based on said operating parameters. Said optimization calculations corresponding to a plurality of different target objectives, presenting results of said optimization calculations together with said corresponding target objectives to a user for user selection therefrom, and outputting an energy management strategy for said plurality of energy sources and/or electrical appliances based on said optimization calculations and said user selection.

Description

Field of the Invention

The invention relates to a method and system for energy management in a facility, said facility comprising a plurality of energy sources and/or electrical appliances. For instance, said facility may be an industrial facility, or a home or office building.

Background of the Invention

As users become more environmentally concerned and regulating authorities increasingly restrict the use of non-renewable energy, methods and systems for a smart electric energy management in industrial facilities or office buildings, in particular for the optimization of the control of energy flows, are becoming ever more important. Current solutions of intelligent homes use rather simple and predefined control mechanisms for home energy management. Even the solutions that learn user habits and adjust performance of the intelligent home accordingly, usually perform optimization in terms of one gain only, e.g. decreasing costs.
State-of-the-art research in energy management systems in intelligent buildings typically model the problem of energy management as a scheduling problem. First, predictive models for solar irradiation, consumption, prices, etc. are computed, which are then used in the optimization of an energy management schedule for the next time horizon (usually one day). In order for those techniques to work well, the predictions have to be extremely accurate, which is hard to achieve in real life due to uncertainty of external factors and parameters, such as weather predictions. Since the weather directly influences the production of energy in an energy management system, scheduling is not an adequate technique for optimal energy flow management.
Some researches present energy management systems that do not use scheduling: In their research article "Batch Reinforcement Learning for Smart Home Energy Management", Proceedings of the 24th International Conference on Artificial Intelligence, AAAI Press 2015, Berlink et al. use Markov decision processes. In their research article "Multiobjective Intelligent Energy Management for a Microgrid", IEEE Transactions on Industrial Electronics 60 (2013), pages 1688-1699, Chaouachi et al. employ fuzzy logic expert systems. Both research articles indicate the deficiency of using schedules for energy measurement optimization.
Some researchers deal with optimizing only one criteria, whereas others, such as Zhang et al., "A Novel Multiobjective Optimization Algorithm for a Home Energy Management System in Smart Grid", Mathematic Problems in Engineering 2015, acknowledge the need to optimize the system according to multiple contradictory optimization criteria. Often-used and practically relevant criteria such as energy consumption, carbon emission, self-consumption, and costs are usually at least in part contradictory, in the sense that improving one criterion can worsen another. For instance, if selling the energy is economically beneficial, then increasing the energy sales to the grid lowers the costs and thereby enhances the profit, but at the same time reduces the self-sufficiency rate. The prior art techniques usually transform the multiple criteria into one effective criterion, using a weighted sum approach, and then perform a one-criterion optimization.
Parameter-based optimization for energy management systems according to multiple criteria is described by Kitamura et al., "Multiobjective Energy Management System using Modified MOPSO", 2005 IEEE International Conference on Systems, Man and Cybernetics, Volume 4 (2005), pages 3497-3503, and by Yang et al., "Multi-Objective Optimization for Decision-Making of Energy and Comfort Management in Building Automation and Control", Sustainable Cities and Society 2 (2012), pages 1-7. Kitamura et al. perform daily optimization of the operational schedule, using optimized timetables, whereas Yang et al. dynamically perform optimization in searching for the optimal reference points for lower level controllers.
However, no system in the prior art can provide robust, decision tree-based optimization of an energy management system and multiple trade-off solutions in configurations with several conflicting criteria.

Overview of the Present Invention

This problem is addressed with a method for energy management in a facility according to independent claim 1, and a system for energy management in a facility according to independent claim 10. The dependent claims relate to preferred embodiments.
A method for energy management in a facility, said facility comprising a plurality of energy sources and/or electrical appliances, comprises the steps of obtaining a plurality of operating parameters relating to energy production of said energy sources and/or energy consumption of said electrical appliances in said facility, performing a plurality of optimization calculations for said energy management of said energy sources and/or electrical appliances based on said operating parameters, wherein said optimization calculations correspond to a plurality of different target objectives, presenting results of said optimization calculations together with said corresponding target objectives to a user for user selection therefrom, and outputting an energy management strategy or energy management rules for said plurality of energy sources and/or electrical appliances based on said optimization calculations and said user selection.
Performing a plurality of optimization calculations, wherein said optimization calculations correspond to a plurality of different target objectives, allows to provide the user with multiple trade-off solutions for energy management of the facility. The method according to the present invention may hence allow the user choose an optimized solution for energy management of said facility based on his choice or preferences. In accordance with the optimized solution selected by the user, an energy management strategy for said plurality of energy sources and/or electrical appliances may then be output. The method according to the present invention thereby provides a robust and fully multi-variate optimization of an energy management system.
Said energy management rules may correspond to, or may be translated into specific operation parameters for operating said facility, in particular for operating said energy sources and/or said electrical appliances.
In an embodiment, the method comprises a step of allowing said user to select said plurality of target objectives.
In particular, the user selection may occur prior to performing said plurality of optimization calculations.
In this configuration, the method allows to determine optimal trade-off solutions depending on those criteria that matter most for a given user, or in the specific circumstances or environmental conditions.
For instance, said target objectives may comprise a degree of energy self-sufficiency and/or a degree of user comfort and/or an estimated or projected energy cost.
The present invention can be employed in energy management of a large number of facilities, comprising industrial facilities, office facilities, or a home of said user.
Moreover, the present invention is highly versatile and can be employed in the control and management of energy flows in nearly every given environment or grid, wherein said environment or grid may comprise any number of energy sources and any number of electrical appliances.
In the sense of the present invention, an energy source may be considered as a device or connection that provides or delivers electrical energy to said facility.
For instance, said energy sources may comprise a connection to a (surrounding utility or remote) electrical grid and/or a photovoltaic source and/or a hydroelectric source and/or a wind power source and/or an electrical generator and/or an electrical energy storage device.
Energy appliances in the sense of the present invention may comprise any device or connection that uses electrical energy in said facility, hence a sink of electrical energy.
For instance, said energy appliances may comprise a ventilation system and/or a heating system and/or an air conditioning system and/or a water heating system and/or an electrical energy storage device.
Electrical energy storage devices, such as batteries or capacitors, may both store energy, and deliver previously stored energy, and may hence serve both as an energy source or as an energy appliance in the context of this invention.
Said user selection among said results of said optimization may be a manual user selection, or may be performed in an automated way base
Said results of said optimization calculation may be presented to said user on a user interface unit, in particular on a graphical display unit for said user selection.
In particular, said results of said optimization calculation may be presented to said user as a trade-off curve between different target objectives.
The method may comprise a step of receiving said user selection, in particular receiving said user selection from said user interface unit, in particular from said graphical display unit.
Said operating parameters relating to said energy production may be or may comprise any parameters that can influence the energy management in said facility.
In an embodiment, said operating parameters may pertain to a user context.
A user context may be a set of operating parameters, such as a set of operating parameters relating to a typical use scenario.
Said user context may preferably be selected by said user.
In an embodiment, said operating parameters comprise historical user data.
Alternatively or additionally, said operating parameters comprise current energy parameters, in particular current energy production and/or current energy selling prices and/or current energy buying prices.
The method may further comprise a step of operating said energy sources and/or electrical appliances in accordance with said energy management strategy or energy management rules.
In an embodiment, said results of said optimization calculations are non-dominated according to a Pareto dominance relation.
In an example, the method further comprises a step of characterizing said user-selected optimization results relative to optimization results not selected by said user.
In particular, said user-selected optimization results may be characterized relative to extremal values of said target objectives, such as minimum or maximum values of said target objectives.
Such a characterization may allow to adapt the optimized solution later in case said operating parameters change over time or in accordance with external events, and still provide a solution and energy management strategy that accurately reflects the user preferences.
For example, characterizing said user-selected optimization results may comprise a step of determining a distance between results of said optimization calculations and extremal values of said target objectives.
In an embodiment, the method further comprises a step of dynamic updating said optimization calculation and/or energy management strategy in accordance with a change of said operating parameters.
As an example, said dynamic updating may be based on a characterization of said user-selected optimization result relative to optimization results not selected by said user, in particular relative to extremal values of said target objectives, such as minimum or maximum values of said target objectives.
The invention further relates to a system for energy management in a facility, said facility comprising a plurality of energy sources and/or electrical appliances, wherein said system comprises a receiving unit adapted to obtain a plurality of operating parameters relating to energy production of said energy sources and/or energy consumption of said electrical appliances in said facility, a computing unit adapted to perform a plurality of optimization calculations for said energy management of said energy sources and/or electrical appliances based on said operating parameters, wherein said optimization calculations correspond to a plurality of different target objectives, a user interface unit adapted to display results of said optimization calculations together with said corresponding target objectives to a user, and to receive a user selection therefrom, and an output unit adapted to output an energy management strategy or energy management rules for said plurality of energy sources and/or electrical appliances based on said optimization calculation and said user selection.
The system may be adapted to perform a method with some or all of the steps described above.
In an example, said system may be partially or fully incorporated into an energy control unit or energy management unit of said facility.
It is a particular advantage of the method and system of the present invention that they can be readily applied to upgrade existing energy control units or energy management units as they are in use in industrial facilities, office facilities or smart homes.
In an embodiment, the system may further comprise a selection unit adapted to allow said user to select said plurality of target objectives, in particular prior to performing said plurality of optimization calculations.
Said system may further comprise a database unit.
For instance, said operating parameters may comprise historical user data, and said database unit may be adapted to store said historical user data, and to provide said historical user data to said computing unit.
In an embodiment, said operating parameters comprise current energy parameters, and said receiving unit may be adapted to receive said current energy parameters and to provide said current energy parameters to said computing unit.
Said output unit may be coupled to said energy sources and/or to said electrical appliances, and may be adapted to control said energy sources and/or said electrical appliances in accordance with said energy management strategy.
Said computing unit may be adapted to characterize said user-selected optimization results relative to optimization results not selected by said user, in particular relative to extremal values of said target objectives.
Said computing unit may be further adapted to dynamic update said optimization calculation and/or said energy management strategy in accordance with a change of said operating parameters.
The invention further relates to a computer program or to a computer program product comprising computer-readable instructions, such that said instructions, when read on a computer system, implement on said computer system a method with some or all of the features described above.

Brief Description of the Figures

The details and numerous advantages of the method and system according to the present invention will be best apparent from a description of preferred embodiments in accordance with the Figures, in which:

Fig. 1: is a schematic illustration of a facility and an energy management environment in which a method and system according to an embodiment of the present invention may be employed;
Fig. 2: schematically illustrates the data and information flow in the facility according to Fig. 1;
Fig. 3: schematically illustrates a process of automatic construction of a contextual hierarchical model template in accordance with an embodiment of the method according to the present invention;
Fig. 4: illustrates an example of a contextual hierarchical model template according to Fig. 3;
Fig. 5: is a schematic illustration of the process of optimization in a search for optimal solution controls for multiple objectives in accordance with an embodiment of a method according to the present invention;
Fig. 6: is a flow diagram illustrating a method of choosing a solution from a set of all (sub) optimal solutions found according to an embodiment;
Fig. 7: is an example of a trade-off curve that may be the result of an optimization calculation according to an embodiment of the method according to the present invention; and
Fig. 8: is a flow diagram illustrating the steps of a method for energy management according to an embodiment of the present invention.

Detailed Description of Embodiments

The method and systems in accordance with the present invention will now be described for the example of an optimized control of an energy management system in a home or office facility that consists of an electric utility, one or more non-deferrable electrical energy consumption units, an electric energy storage unit, and an energy production unit for renewable energy sources and/or a deferrable electric energy consumption unit which has the possibility of remote control. As will be described in detail below, the method and system according to the present invention may allow to find optimized robust control models for controlling the electric energy flows in the system with respect to one or more objectives that may have influence on the operating scenarios for charging and discharging the electric energy storage unit and/or the deferrable electric energy consumption unit. The system may have access to measured or estimated environmental variables, such as historical electric energy production, historical electric energy consumption, etc. The system may further have access to variables related to the weather forecast, prediction of electric energy consumption, current buying and selling prices of electricity in the grid, etc. All these parameters may serve as operating parameters in the sense of the present invention, which can serve as an input for the optimization calculations.
Typically, a user may want to optimize the energy flow in the system according to multiple objectives, which are sometimes contradictory. Relevant energy management system objectives may include;

Running costs of the system (income from selling the electricity to the electric utility minus outcome from buying the electricity from the electric utility, cf. Eq. (1) where E_fromGrid (i) is the amount of energy sold from the electric utility in the time interval i, price_buy (i) is the buying price for the energy bought from the electric utility in the time interval i, E_toGrid (i) is the amount of energy sold to the electric utility in the time interval i, and price_sell (i) is the selling price for the energy sold to the electric utility in the time interval i. Running costs are computed for the observation horizon starting at time 0 and finishing at time M. $Costs = \sum_{i = 0}^{M} (E_{fromGrid} (i) \cdot {price}_{buy} (i) - E_{toGrid} (i) \cdot {price}_{sell} (i)),$
Self-sufficiency (one minus quotient with the numerator of the electricity bought from electric utility and denominator of total electricity consumed, cf. Eq. 2), where E_fromGrid (i) is the amount of energy sold from the electric utility in the time interval i, and E_consumed (i) is the amount of energy consumed by all devices and appliances in the system. $SelfSufficiency = 1 - \frac{\sum_{i = 0}^{M} (E_{fromGrid} (i))}{\sum_{i = 0}^{M} (E_{consumed} (i))},$

The costs according to Eq. (1) and the energy self-sufficiency according to Eq. (2) are typically contradicting or conflicting criteria: If selling the energy produced in the system, such as from solar panels or wind power is economically beneficial, the costs decreases (corresponding to increased profit), but the self-sufficiency rate decreases as well.
Other target objectives that a user may take into account are carbon (CO₂) emissions, maintenance costs, or user comfort.
The method and system according to the present invention allow to perform optimization calculations that may lead to trade-off curves for the conflicting target objectives. This enables the user to inspect the trade-offs among the solutions and choose a solution that best fits his preferences. When (sub)optimal solutions are found, their representations in objective space may be presented to the user on a graphical display. The user may choose the solution that best describes his or her trade-off between objectives. The solution can then be uploaded to the logic memory of the energy management system controller. This logic is then responsible for determining the electric energy flows in the system by turning the controllable devices on or off or to perform regulation at continuous set-point value and/or determining whether to charge or discharge the battery.
In some embodiments, the method according to the present invention may comprise two phases. The first phase deals with the optimization and is computationally demanding. This phase could run on a local or remote computer, depending on the setup of the system. The subsequent second phase concerns the control of the energy management, where an energy management strategy computed in the first phase is used in order to control the energy flows in the system. The second phase is typically not computationally demanding, and can usually be run on a standard controller device.
The optimization according to the method of the present invention may comprise four steps. In a first step, historical information about the system may be gathered (such as electric energy production and consumption trends, electric energy selling and buying prices, that can be inserted manually if not accessible otherwise), system characteristics, historical operational data of the system and historical environmental conditions such as relative air humidity and temperature.
This information is then used in the second step, where optimization is performed. Optimization may automatically find control models for electric energy management system control with respect to multiple objectives. Since there are typically multiple non-dominated optimal solutions for an optimization problem with multiple objectives, a third step may be needed in which the user interactively selects the preferred solution that is used for the control of the electric energy management system. The optimized solutions may be presented on a graphical display unit to the user, where the performance of every solution is shown with respect to all objectives.
The user may choose the solution that best represents his preferences regarding the trade-offs between the objectives. Because of the historical tracking of the user selections, the system can choose the solution that is closest to the user choice in history and best presents a user's previously chosen solution automatically, so that user intervention is not necessarily needed in response to each and every change of external parameters. If the user wants to choose another trade-off solution (for instance when going on holiday, the user requirements may change along with the preferred trade-off), he may do so at any time. The system enables the user to choose from solutions obtained in the latest optimization procedure.
In the fourth step, the chosen model may be loaded into the apparatus for electric energy management control, which is responsible for the directing of energy flows in the facility in real time. The apparatus may use the solution model for decision making in order to direct the energy flows. All of the above described steps may be constantly repeated during the operation of the apparatus.
In an embodiment, the apparatus for the optimized control of a system consists of an electric utility, one or more non-deferrable electrical energy consumption units, an electric energy storage unit, and at least one of the following units: an energy production unit from renewable energy sources, an energy storage unit, a deferrable electric energy consumption unit which has the possibility of remote operation, an electric utility that uses dynamic pricing techniques for buying and selling electric energy, some non-deferrable consumption units (such as lights and smaller electrical devices), solar panels, a battery, a heat-pump water heater and a washing machine. The apparatus for optimized control of an electric energy management system with production, consumption and storage of energy may receive information from other devices and sensors about current, historical and predicted future trends of electric energy consumption, current, historical and predicted future trends of electric energy production (energy calculated from the weather forecast for the geographical region of the installed system), as well as current, historical and predicted future trends of electric energy prices for buying and selling electricity from and to the electric utility. The apparatus may use the gathered information and optimized control model that was chosen by the user according to his/her preferences in order to decide whether to charge or discharge the energy storage unit and at what rate, and whether to power the remotely controlled devices on or off. Charging and discharging patterns for the electric energy storage unit therefore may change dynamically, depending on the past, present and predicted future environmental and operational parameters. A generalized description of similar states of past, present and predicted future environmental and operational parameters may be called a context in the present application. Likewise, the remotely controlled device can be turned on or off depending on the environmental parameters.
The control model used in the proposed procedure may be a hierarchical decision model, which consists of two levels. In the upper level the context of the operating state may be determined (an example of context is: morning of a sunny day with users present in the building) with respect to available information, and the user-chosen optimized control model. The lower level of the hierarchical decision model may consist of decision trees, wherein each decision tree belongs to one context. A decision tree can be a collection of IF-THEN-ELSE rules (e.g. IF1 sunny_tomorrow THEN1 feed_energy_to_grid ELSE1 store_energy) that are chained together to form a tree like structure of rules (or a flowchart), e.g.:

      IF1 sunny_today
              THEN1
              IF2 sunny_tomorrow
                     THEN2 feed_energy_to_grid
              ELSE2 store_energy
              END_IF2
       ELSE1
              IF2 some_energy_stored
                     THEN2 take_energy_from_storage
              ELSE2 take_energy_from_grid
              END_IF2
       END_IF1

The control action of the hierarchical decision model may be executed in two steps. In a first step, a (sub)optimal decision tree may be selected depending on the predicted context. In the second step the selected decision tree may be used in order to determine which action to take in real time.

Embodiments of the present invention will now be described in greater detail with reference to Figures 1 to 8.

Fig. 1 is a conceptual overview of an electrical facility 100, such as a user's home or office space, in which the method and system according to the present invention may be employed. The facility 100 comprises a plurality of energy sources, such as photovoltaic panels 101, a micro hydro electrical unit 102, a wind turbine 103, a connection to an electric power grid 104, and an electric backup generator 120.

The facility 100 further comprises a plurality of electrical appliances or power-consuming devices 114, such as a hot water heating unit 115, a heating ventilation and air conditioning unit (HVAC) 116, and a plurality of sensors 117. Further appliances 118 may be present in large numbers, depending on the size and configuration of the facility 100. They may be electrically connected to the local grid via controllable relays 119.

The facility 100 further comprises an electric storage unit 111, such as a battery or capacitor. The energy storage unit 111 may serve to store electric energy, and to feed the stored electric energy into the facility 100 at a later point in time. Depending on whether the electrical storage unit 111 is charged or discharged, it may either serve as an energy source or as an electrical appliance in the context of the present invention.

The facility 100 further comprises an apparatus for energy management 106. The apparatus 106 may be incorporated into a legacy control unit of the facility 100, but may also be a separate device. The apparatus 106 may comprise a user interface unit 107 and a computing unit 108, wherein the computing unit may comprise a controller unit 109 and an internal data storage database 110. The apparatus 108 may be connected to a remote control center 105, which may in turn be connected to the worldwide web 112 that may provide remote web services 113.

The user interface unit 107 may be an LED or LCD or CRT display with the capability of accepting user input, either in the form of a touch-sensitive display or a keyboard or a custom input device.

The controller unit 109 may be responsible for a real-time power distribution in the facility 100.

The database unit 110 may store historic records relating to energy parameters. For instance, information stored in the database unit 110 may comprise a timestamp and information about the electric energy consumption in a previous time interval, or the average electric power consumption in the last few minutes. The database unit 110 may also collect information about electric energy selling and buying prices, both current and in past time slots. Additional information may include weather-related data, such as solar radiation, wind speed and direction, temperature, a state of house appliances (whether they are on or off or operating at continuous set-point), occupancy and any other values of environmental parameters that can be measured by means of the sensor units 117, or can be retrieved from the worldwide web 112 by means of the remote control center 105.

The electrical storage unit 111 may comprise a plurality of lead-acid or nickel-metal-hydride or lithium ion storage batteries for storing electrical energy, and/or large capacitors, and/or other technology that can store electric energy when required and produce electric energy when required.

The facility 100 may further comprise power electronics, including inverters for converting DC electrical energy into AC energy, circuit breakers, phase converters, etc. However, these ancillary devices are not shown in Fig. 1, for ease of presentation.

The controller unit 109 may comprise a central processing unit, memory and peripherals, programmed with computer software for controlling the operation of the apparatus 106 in order to receive power from power sources 101-104 and energy storage 111, and distribute electrical power to devices 114-118, and possibly to electric grid 104. The control is set so that limitations and legal restrictions are respected. For example, in some countries the power fed-in to the grid must not exceed some percentage of the maximum peak power of the installed system. Further details of various steps that may be carried out by such software are described in more detail below.

The user interface unit 107 may be used in order to display information regarding the system operation, to enable the selection of various modes of operation (a mode of operation can be defined by the contextual hierarchical decision model and its estimated performance with respect to multiple objectives), and to enable the configuration of the system parameters.

In the configuration of Fig. 1, the apparatus 106 comprises the user interface unit 107 that may be attached to the housing of the computing device 108. However, in other embodiments a user interface to the controller unit 109 can also be enabled through the use of the remote control centre 105 and a web or a mobile application that can be accessed by means of a mobile handheld device or a stationary computer with the connection to the remote control centre 105.

In the configuration of Fig. 1, the remote control centre 105 may mostly serve for enabling a remote access to the apparatus 106, or to allow the apparatus 106 to invoke the web services 113. Remote control centre 105 may use web services 113 in order to regularly obtain information and/or custom-designed computer programs that retrieve information from structured or unstructured documents on the World Wide Web 112. The information could include and is not limited to prices of selling and buying electricity to the utility, and weather forecast (cloud coverage, wind speed and direction, temperature, etc.). However, in other embodiments the remote control centre 105 may also perform some or all of the optimization tasks when searching for optimal contextual hierarchical decision models in case the controller unit 109 does not have the computing power to perform optimization tasks on its own.

The apparatus 106 may be coupled to the electric grid 104 through a power interface (not shown), which may include surge suppressors, circuit breakers and other electronic devices. Electricity is provided in a form that is required by the system. Additionally, the backup generator 120 may be connected to the system and may be controlled by the apparatus 106 in order to provide electricity when needed. Alternative energy sources may be included in the system in order to provide electrical power to the system and the apparatus 106, but are not necessary. Such sources may include, but are not limited to the photovoltaic panels 101 that transform solar radiation to electric energy, the micro-hydroelectric power generators 102 that use the movement of the water to generate energy, and the wind turbines 103 that transform wind energy to electric energy. The information about the electric energy production by the alternative energy sources is regularly stored to the internal database unit 110. This information can then be used in the optimization process as will be described in detail later.

The power-consuming devices 114 through 118 may be controlled by and receive power from the apparatus 106. These devices may include sensors 117, such as indoor and outdoor temperature sensors, occupancy sensors, air quality sensors and others. If available, the sensors 117 produce data that is made available to the controller unit 109, which uses the sensor information in order to decide on what action to take in real time. Further, sensor data can also be stored either in the internal database unit 110 of the apparatus 106 (where the stored information can be used when performing optimization) or in a remote location, such as the web 112, to which either the controller unit 109 or the remote control centre 105 has access. Further, the information about electric energy consumption and production can regularly be stored either in the internal database unit 110 or a remotely accessible location, such as the web 112. Devices such as the hot water heaters 115 and HVAC 116 that can be remotely controlled and can receive the command to turn on or off, or to change the operational load set-point multiple times a day. However, the internal logic and safety measures implemented on the device can prevent the device from turning on or off or changing an operational load set-point. The controller could for instance decide to extra heat the water in the hot water heater 115 in cases when the electric energy obtained from the alternative sources cannot be directed elsewhere.

Some of the appliances 118, called "connected" or "smart" appliances, may already be connected to the internet. Such smart appliances 118 may receive a control signal from the controller 109 directly in order to turn on. Appliances 118 that are not connected to the internet by default may be controlled using one or more controllable relays 119.

FIG. 2 shows a data flow in the facility 100. A parameters monitoring service 204 may be run on the controller unit 109 and may be responsible for providing an interface for accessing the information stored in the internal data storage device 110, world wide web 112 and sensors 117. Further, the parameters monitoring service 204 may store the information from the sensors 117 or the World Wide Web 112 to the internal data storage 110 or World Wide Web 112.

A system configuration service 205 may run on the controller unit 109 and may enable the access to the system parameters, such as energy storage capacity, energy storage charge or discharge capacity and efficiency, energy storage self-discharge, peak power production of the photovoltaic panels, energy consumption profile of controllable appliances and devices etc., provided by the vendors. The system configuration service 205 may provide enough information to enable the simulation of the whole system based on given information about power consumption and power production (for which the estimate can also be computed from solar radiation, wind speed and direction, if needed), and energy prices (all those variables are generally independent). Simulation may be used in order to enable an optimization procedure 207 run on the controller unit 109, which will be described in detail later. The optimization procedure 207 also requires the contextual hierarchical model template data, provided by the contextual hierarchical model template generator 206 running on the controller unit 109.

A contextual hierarchical model template may be a combination of: a model that predicts the future context (type of a day) based on current and historic information; and placeholders for control strategies (e.g., decision trees) that are responsible for real-time energy management. It is called a template because while the model for prediction of context may be already generated, the trees that belong to each context are generally not. The trees can be found in the optimization step of the method and can be inserted into the template to provide a contextual hierarchical model, which may present the logic of the optimized controllers. The simplest contextual hierarchical model template may require the search for only one decision tree (an implemented decision tree generates a solution), such as when only one context is defined and applied for any condition. Solutions found by the optimization procedure 207 can be presented in a visualization and solution selection module 208 running on the controller unit 109. Each solution represents the evaluated control operation according to the corresponding contextual hierarchical model.

Solutions can be presented to the user on the display of the user interface unit 107 in the form of FIG. 7, where various trade-offs between conflicting criteria of cost and self-sufficiency is observable. Since all presented solutions are the best ones found (non-dominated according to the Pareto dominance relation), an increase in one objective results in a decrease in the second objective. When the user chooses the solution that best presents her preferences, the solution position with regards to other solutions is stored into the system configuration service 205. Storing this information enables the system to automatically choose the solution the next time the optimization procedure 207 is launched and new solutions is required to be selected. The system can then choose the solution that lays in the same position (or is close to) with regards to other solutions as preferred by the user, as will be described in more details below.

After the solution is selected, the corresponding contextual hierarchical model is loaded by an upload controller service 209 running on the controller unit 109 into the memory of the controller unit 109 that is reserved for instructions on how to control the system. The controller unit 109 is able to interpret the model for context selection and corresponding decision trees in order to control the system in real time which is to execute the decision tree statements in a solution execution service 210 running on the controller unit 109. If information about current, past and predicted future environmental variables are required by the controller unit 109, parameters monitoring service 204 provides it. A model for context selection may be activated either every N hours, where N is a user defined constant and could be for instance 6 or 12, or it is activated when the parameters monitoring service 204 determines that the parameters are outside the boundaries normal to currently chosen context.

FIG. 3 shows a procedure for contextual hierarchical model template generation in further detail. In the first phase of the template generation, a context definition module 304 running on the controller unit 109 determines the context. First, it retrieves the system configuration parameters from the system configuration service 205, where constraints on contexts are defined. For instance, a context constraint can be: "one context is computed for the whole day", or "a context can be computed for half a day", etc. This is to ensure that context does not change too often, which could lead to system instability. Second, information about environmental parameters is gathered from the parameters monitoring service 204. This information may include past electric energy production, past electric energy consumption, past outdoor temperatures, past wind speeds, past occupancy information etc. It is advantageous that the timespans of obtained data overlaps as much as possible in order to obtain valid context. Third, the data is combined and transformed so that one data instance consists of all gathered information that can fall into one context as constrained by the system configuration parameters. One instance could comprise information about temperature, cloud coverage, electric energy buy and sell price for every 15 minutes for the whole day, which means 4 · 4 · 24 = 384 entries for each instance. Those instances are then clustered together by means of a clustering algorithm such as K-mean clustering, Affinity propagation, Mean-shift, Spectral clustering, Ward hierarchical clustering, Agglomerative clustering, DBSCAN, Gaussian mixtures, Birch or others. A Cluster id is then associated with each instance. Each cluster corresponds to a different context. Sunny summer days when lots of electric energy is available will likely be clustered into a different cluster than sunny winter days, so the context will be different. Therefore, a context can be defined for every instance by the context definition module 304. The contextual hierarchical model template generator 206 then generates a model for context prediction using a context prediction module 305 running on the controller unit 109. The context prediction module 305 may use the context definition data generated by the context definition module 304. It may further use parameters from the system configuration that define what information can be used in order to build a context prediction model.

For example, the system could be configured in such a way that the context prediction module 305 can use all available information about the system that can be accessed by means of the parameters monitoring service 204 for a period of 24 hours before a decision about the context can be made, and that the context prediction should take place every day at 00:00 and 12:00. That means that a context for one day is predicted at the beginning of the day, and the whole gathered information about the previous day is used in order to determine the context of the coming day. Since the context of the following period has already been defined by the context definition module 304, this information can be used as a label or a target variable that the context prediction model 305 is to predict. Therefore, a classification model can be generated using one of the classification methods such as Decision tree, Random forest, Nearest neighbours, Support vector machines, Naive Bayes, Artificial neural networks or others. In order to find a good performing combination of a classification method and its parameters, a model selection technique can be used such as sequential Bayes optimisation or an evolutionary algorithm approach or others. Cross-validation accuracy score can be used to evaluate classification methods. The chosen model is then inserted into a contextual hierarchical model template 306.

FIG. 4 shows an example of a detailed contextual hierarchical model template 306 that can be generated by the process described above with reference to FIG. 3. The contextual hierarchical model template 306 comprises a context prediction model 403 and placeholders for decision tree controller 407 to 409 that are later found by the optimization procedure described in more detail herein. The context prediction model 403 uses the information provided by parameters monitoring service 204 in order to predict contexts 404 to 406.

FIG. 5 shows a procedure for the optimization 207 of control of the energy management system in greater detail. A search technique is used to find (sub)optimal solutions. In one embodiment, an evolutionary based search technique as presented in FIG. 5 can be used. Since an evolutionary technique is a computationally intensive process, it may be advantageous to execute it on the remote control centre 105. However, if the controller is able to execute the search locally, it can do so as well.

The evolutionary search technique may consist of seven building blocks: solution candidate initialization 505, solution candidate evaluation 506, parameter optimization 507, solution subset selection 508 and evolutionary operators of recombination 509 and mutation 510.

The solution candidate initialization 505 can be performed using the ramped-half-and-half method described by J. Koza in "Genetic Programming - On the Programming of Computers by Natural Selection", Harvard University Press 2000, or any other method for random generation of decision trees. For each solution one tree is generated for every tree placeholder in the contextual hierarchical model template. Therefore, one solution is a contextual hierarchical model with several optimized decision trees that correspond to each context type. The controller uses the historical and current information to predict the context and applies decision tree (corresponding to the predicted context) statements in order to perform appropriate control. A binary tree can be used, where inner nodes present conditions and outer nodes present actions. In that case a tree can be represented as a list of nodes - quadruples

(nodeID, finalNode?, parameterName, parameterValue),

where nodeID is the ID of the node that uniquely defines the node position in a tree, finalNode? is a Boolean value that is either True when the node does not have any children or False when the node is an inner node that has children, parameterName is the name of the condition to be checked when traversing the tree and parameterValue are the parameters that are required when checking the condition parameterName.

A nodeID can be defined recursively as follows: The nodeID value of a root node of a tree is 0, every inner node of a tree has two children, when the condition of an inner node with a nodeID = n evaluates to True then the child following the "True" branch has a nodeID = 2 · n + 1, when the condition evaluates to False a child following the "False" branch has a nodeID = 2 · n + 2.

Each solution candidate may be evaluated 506 in order to determine its performance according to multiple criteria. For the evaluations 506 a numerical simulator can be used, that tries best to mimic real electric energy flow dynamics in the electric energy management system. For instance, the energy management system simulator can take as input the historical information about electric energy production, electric energy consumption, electric energy buying and selling prices, all provided through the parameters monitoring service 204. When historical information is not available, the parameters monitoring service 204 may try to locate the information on the internet (for a case of weather, electric energy prices and estimated consumption). If the information is not available on the internet, the energy management system provider may supply information that best relate to the system. Further, information about system configuration 205 can be used to determine the technical specifications about installed energy management system components, e.g. solar panel peak power production, wind power peak production, electric energy storage device capacity, its charging and discharging efficiencies, self-discharge rate and minimal state of charge, consumption patterns or their estimates for controllable devices and appliances.

In the parameter optimization operator 507, a subset of newly generated solution candidate decision trees may be selected, and optimization may be performed in order to optimize the parameters of the selected decision trees. Any numerical multi-objective optimization algorithm can be used for the parameter optimization. An exemplary description of such an algorithm is presented herein below. An overview of the algorithm is as follows:

1. Pick a random subset of current solutions.
2. Generate a subpopulation of mutants from each picked solution.
3. Perform multi-objective optimization on each subpopulation until the stopping criteria is met.
4. Perform a subset selection of optimized solutions from the union of the subpopulations.

Algorithm 1: Parameter optimization pseudocode

       // PARAMETER OPTIMIZATION
       require evaluation_fn // this is the evaluation
                                                         // function
       require new_population // the set of solution candidates
                                                         // from which a subset of
                                                         // candidates whose parameters
                                                         // will be optimized is chosen
       require subset_selection_operator // selection operator to select
                                                         // the subset of candidates
       // paramter optimization parameters
       require sub_population_size
       require selection_operator
       require crossover_operator
       require crossover_rate
       require mutation_operator
       require mutation_rate
       require stopping_criteria
       chosen_subset = subset_selection_operator(new_population)
       chosen_subpopulations = []
       for individual in chosen_subset:
         chosen_subpopulations.append([mutation_operator(individual) for
       subpopulation in range(sub_population_size)])
       for subpopulation in subpopulations:
         for ind in subpopulation:
            evaluation_fn(ind)
      while not stopping_criteria:
         for subpopulation in chosen_subpopulations:
            subpopulation_c = copy(shuffle(subpopulation))
            for ind1, ind2 in zip(subpopulation_c[::2],subpopulation_c[1::2]):
               if random.random() < crossover_rate:
                  crossover_operator (ind1, ind2)
            for ind in subpopulation_c:
               if random.random() < mutation_rate:
                  mutation_operator(ind)
            subpopulation = selection_operator(subpopulation +
                                                                        subpopulation_c)
      return selection_operator(union(chosen_subpopulations))

Selection operator 508 may choose the subset of solutions with regards to multiple criteria according to multi-objective selection methods, such as non-dominated sorting and crowding distance based selection from NSGA-II, strength Pareto based selection from SPEA-II, goal based non-dominated sorting from NSGA-III or other. Choosing a subset of solutions is beneficial in order to avoid population size growth, which would slow down the solution search process.

Selected solutions can then be used to produce new solution candidates. First, solutions are combined 509 by the use of a crossover operator, such as a subtree crossover as described in J. Koza (cited above), either on every pair of decision trees belonging to the same context, or on the whole solution, since the whole solution is a decision tree (contextual hierarchical model template with a number of decision subtrees). The operator is applied with some probability (crossover rate). If no crossover is used (operator is not applied), the new solutions are only the exact copies of the parent solutions. Small disruptions operators 510 (mutations) are applied next with some probability (mutation rate). A subtree mutation operator can be used in a combination with other genetic programming mutation operators (node replacement mutation, shrink mutation, hoist mutation, etc.). The mutation operator can provide new parts of solutions (sub-controls) that are currently not present in the available solution candidates. Newly created candidate solutions may then be evaluated, and the whole process of new solution candidate generation is repeated. The loop in the optimization procedure 207 is executed until a stopping criteria 511 is reached. Typical examples of the stopping criteria are: maximum amount of time made available for the optimization, maximum number of evaluations, no improvement of solutions detected, user stopped the optimization manually.

When the optimization procedure 207 stops, the solutions 512 are stored and presented to the user. The user is to choose the preferred one, or the system automatically chooses the solution that best corresponds to the previously chosen solution by the user.

FIG. 6 is a flow diagram that illustrates the process of solution selection and solution handling in the electric energy management system according to an embodiment. The user can demand the solution selection 601 anytime during the operation of the energy management system. Available solutions from the most recent completed optimization process run are selected in order to proceed with the solution selection process. Solutions can then be visualised 602 to the user in an easy to understand way. An example for a visualization is the trade-off curve of Fig. 7, in which all the points of the curve correspond to optimal (non-Pareto dominated) set points within the given context, for different values of the two conflicting target objectives costs and self-sufficiency.

Solutions can be presented either on the user interface unit 107 that is attached to the apparatus, on the remote control centre 105 or on a mobile computation device belonging to the user that can access the remote control centre 105. On the graphical presentation of solutions, the user has the access to possible solutions and their estimated performance according to all the objectives defined. The user interface enables the user to choose the preferred solution 603. The chosen solution is then uploaded 606 to the controller unit 109 for real time execution and energy management.

Additionally, a relative position metric can be used in order to calculate the preferred trade-off of the selected solution with respect to all other available solutions. The relative position metric can be helpful in dynamic multi-objective optimization scenarios where many solutions exist for one problem instance and the solutions change under different conditions. In order to preserve the trade-off required by the user, a normalization of the solutions by the use of some relative position metric can be used after the optimization procedure is terminated.

An example of such relative position metric could be the following. Assume that the system is performing a minimization optimization task. Let m ₁, m ₂, ..., m_n be the minimum values achieved for each of the n objectives, and let M ₁ , M ₂ , ..., M_n be the maximum values achieved for each of the n objectives. Let a ₁ , a ₂ , ... , a_n be criteria values of the chosen solution for each of the n objectives. Let (1₁, 1₂, ... , 1 _n ) be a vector of ones of length n. The relative position metric or a reference point R can be computed as follows (using intermediary position metric T, which is a non-normalized version of a reference point R) where all operations are by the component (element by element).

T = (t_{1}, t_{2}, \dots, t_{n}) = \frac{(a_{1}, a_{2}, \dots, a_{n}) - (m_{1}, m_{2}, \dots, m_{n})}{((M_{1}, M_{2}, \dots, M_{n}) - (m_{1}, m_{2}, \dots, m_{n}))} - (1_{1}, 1_{2}, \dots, 1_{n}),

R = (r_{1}, r_{2}, \dots, r_{n}) = \frac{(t_{1}, t_{2}, \dots, t_{n})}{\sqrt (t_{1}^{2} + t_{2}^{2} + \dots + t_{n}^{2})},

This operation may fail when any difference M_i - m_i or the norm of T is zero, but when the objectives are contradictory, such a situation never happens. When the objectives are not contradictory, the optimization problem criteria can be reduced to contradictory criteria. Other relative position metrics may be used that can uniquely determine the position of one chosen point relative to other non-dominated points. This relative position metric value R is then stored into system configuration 605 for future use. When a new frontier of optimized solutions is found, the relative metric values of solutions are calculated and a solution with relative metric closest to the relative metric of the previously chosen solution can be chosen for energy management system control.

Fig. 8 is a flow diagram summarizing a method for energy management in the facility 100 according to an embodiment of the invention.

In a first step S10, a plurality of operating parameters relating to energy production of said energy sources 101 to 104, 120 and/or energy consumption of said electrical appliances 114 to 118 in said facility 100 is obtained.

In a subsequent step S12, a plurality of optimization calculations for said energy management of said energy sources 101 to 104, 120 and/or electrical appliances 114 to 118 based on said operating parameters is performed, wherein said optimization calculation corresponds to a plurality of different target objectives.

In step S 14, the results of said optimization calculation are presented together with said corresponding target objectives to a user for user selection therefrom.

In step S16, an energy management strategy is output for said plurality of energy sources 101 to 104, 120 and/or said electrical appliances 104 to 118 based on said optimization calculations and said user selection.

One idea of the invention is to design several flexible strategies in advance in the form of decision trees that perform (sub)optimal according to user preferences under specific circumstances and in real time using the best flexible strategy given concrete circumstances.

The benefit of the approach was empirically verified and stems from the flexibility of the designed strategies that enable following the same strategy over say one-half day period if the conditions do not change too much. Namely, the benefit of one strategy often relies also on actions that have a prolong effect, and changing control strategy too fast is usually not beneficial in this sense. If the circumstances differ from the predicted (anticipated, forecasted, simulated) for too much and too long, another more suitable strategy is applied.

In summary, a method according to an embodiment of the present invention may combine the following advantageous properties:

1. The output is a hierarchical control logic for an energy management controller.
2. The optimization takes into account several objectives, therefore producing multiple solutions that are non-dominated according to the Pareto dominance relation.
3. The hierarchical control logic dynamically chooses optimized control strategy according to the context.
4. The control logic produced by the method is not computationally demanding and can be uploaded to a standard controller.

The method and system thereby allow to provide a robust, decision tree-based and multi-objective optimization of an energy management system that provides multiple trade-off solutions, and allows a user to select a preferred solution based on context, environmental parameters and user preferences.

The examples described above and the drawings merely serve to illustrate the invention, but should not be understood to imply any limitation. The scope of the invention is to be determined based on the appended claims.

Method and System for Energy Management in a Facility Reference Signs

100: facility
101: photovoltaic panels
102: micro-hydro electrical unit
103: wind turbine
104: electric grid connection
105: remote control centre
106: apparatus for energy management
107: user interface unit
108: computing unit
109: controller unit
110: database unit
111: electrical storage unit
112: world wide web
113: web services
114: electrical appliances/ power consuming devices
115: hot water heating unit
116: heating, ventilation, air conditioning (HVAC) unit
117: sensors
118: appliances
119: controllable relays
120: electric backup generator
204: parameters monitoring service
205: system configuration service
206: contextual hierarchical model template generator
207: optimization procedure
208: visualization and solution selection module
209: upload controller service
210: solution execution service
304: context definition module
305: context prediction module
306: contextual hierarchical model template
403: context prediction model
404, 405, 406: contexts
407, 408, 409: decision tree placeholders
505: solution candidate initialization
506: solution candidate evaluation
507: parameter optimization
508: solution subset selection
509: evolutionary operators of recombination
510: small disruption operators/ mutation
511: stopping criteria
512: optimization solutions
601: demand of solution selection
602: visualization of solutions
603: choice of preferred solution
604: calculation of tradeoff
605: storing of preferred tradeoff
606: upload of selected solution

Claims

A method for energy management in a facility (100), said facility (100) comprising a plurality of energy sources (101-104, 111, 120) and/or electrical appliances (114-118), said method comprising the following steps:
obtaining a plurality of operating parameters relating to energy production of said energy sources (101-104, 111, 120) and/or energy consumption of said electrical appliances (114-118) in said facility;

performing a plurality of optimization calculations for said energy management of said energy sources (101-104, 111, 120) and/or electrical appliances (114-118) based on said operating parameters, wherein said optimization calculations correspond to a plurality of different target objectives;

presenting results of said optimization calculations together with said corresponding target objectives to a user for user selection therefrom; and

outputting an energy management strategy for said plurality of energy sources (101-104, 111, 120) and/or electrical appliances (114-118) based on said optimization calculations and said user selection.
The method according to claim 1, further comprising a step of allowing said user to select said plurality of target objectives, in particular prior to performing said plurality of optimization calculations.
The method according to claim 1 or 2, wherein said results of said optimization calculation are presented to said user on a graphical display unit (107) for said user selection.
The method according to any of the preceding claims, wherein said operating parameters pertain to a user context (404-406), wherein said user context (404-406) may preferably be selected by said user.
The method according to any of the preceding claims, further comprising a step of operating said energy sources (101-104, 111, 120) and/or electrical appliances (114-118) in accordance with said energy management strategy.
The method according to any of the preceding claims, wherein said results of said optimization calculations are non-dominated according to a Pareto dominance relation.
The method according to any of the preceding claims, further comprising a step of characterizing said user-selected optimization results relative to optimization results not selected by said user, in particular relative to extremal values of said target objectives.
The method according to any of the preceding claims, further comprising a step of dynamically updating said optimization calculation and/or energy management strategy in accordance with a change of said operating parameters.
The method according to claim 8, wherein said dynamically updating is based on a characterization of said user-selected optimization result relative to optimization results not selected by said user, in particular relative to extremal values of said target objectives.
A system for energy management in a facility (100), said facility (100) comprising a plurality of energy sources (101-104, 111, 120) and/or electrical appliances (114-118), said system comprising:
a receiving unit (110, 204, 205) adapted to obtain a plurality of operating parameters relating to energy production of said energy sources (101-104, 111, 120) and/or energy consumption of said electrical appliances (114-118) in said facility (100);

a computing unit (108) adapted to perform a plurality of optimization calculations for said energy management of said energy sources (101-104, 111, 120) and/or electrical appliances (114-118) based on said operating parameters, wherein said optimization calculations correspond to a plurality of different target objectives;

a user interface unit (107) adapted to display results of said optimization calculations together with said corresponding target objectives to a user, and to receive a user selection therefrom; and

an output unit (109, 606) adapted to output an energy management strategy for said plurality of energy sources (101-104, 111, 120) and/or electrical appliances (114-118) based on said optimization calculations and said user selection.
The system according to claim 10, further comprising a selection unit (107) adapted to allow said user to select said plurality of target objectives, in particular prior to performing said plurality of optimization calculations.
The system according to claim 10 or 11, wherein said operating parameters comprise historical user data, and said system comprises a database unit (110) adapted to store said historical user data, and to provide said historical user data to said computing unit (108).
The system according to any of the claims 10 to 12, wherein said operating parameters comprise current energy parameters, and said receiving unit (204) is adapted to receive said current energy parameters and to provide said current energy parameters to said computing unit (108).
The system according to any of the claims 10 to 13, wherein said output unit (109, 606) is coupled to said energy sources (101-104, 111, 120) and/or to said electrical appliances (114-118), and is adapted to control said energy sources (101-104, 111, 120) and/or said electrical appliances (114-118) in accordance with said energy management strategy.
A computer program comprising computer-readable instructions such that said instructions, when read on a computer system, implement on said computer system a method according to any of the claims 1 to 9.