Sensitivity analysis of systems and methods for optimizing energy management

Abstract

Embodiments of the present disclosure relate to sensitivity analysis of systems and methods for optimizing energy management. A system for optimizing field energy management, the system comprising: an energy optimizer adapted to determine one or more optimization objectives based on an optimization algorithm; wherein the data converter is adapted to provide at least one first energy data input and at least one second energy data input different from the first energy data input to the energy optimizer, and wherein the energy optimizer is adapted to receive the first energy data input and the second energy data input from the data converter, and determining a respective at least one first energy data output and a respective at least one second energy data output on the basis of the first energy data input and the second energy data input via an optimization algorithm, wherein a data sensitivity of the energy optimizer can be determined with respect to the first energy data input and the second energy data input and the first energy data output and the second energy data output.

Description

Technical Field

[0001] The present invention relates to a system and method, a computer program product and a computer readable medium for optimizing energy management at a site. Background Art

[0002] Factories, facilities, camps or industrial sites usually consist of large conglomerates of multiple energy flows (electricity, gas, heat) with multiple different energy consumers, grid connections and energy generation options. In such energy networks, energy management systems are usually provided to control the flow of power between generating and consuming assets, battery storage and the grid to facilitate overall energy management and resource utilization. However, in such complex systems, the impact of certain conditions and parameters and their potential impact on overall energy management is often not well assessed due to the complexity and interaction of the involved entities.

[0003] Therefore, there is a need for an improved transparent and understandable energy management system and method. Summary of the invention

[0004] This object is achieved by the invention according to the independent claim. Preferred embodiments of the invention are provided in the dependent claims, the description and the drawings.

[0005] The present invention relates to a system for optimizing energy management on site.

[0006] The system comprises an energy optimizer, wherein the energy optimizer is adapted to determine one or more optimization objectives based on an optimization algorithm.

[0007] The system further comprises a data transformer, wherein the data transformer is adapted to provide at least one first energy data input and at least one second energy data input to the energy optimizer, the second energy input being different from the first energy data input.

[0008] The system also includes an energy optimizer adapted to receive a first energy data input and a second energy data input from a data transformer.

[0009] The system further comprises an energy optimizer adapted to determine, via an optimization algorithm, the respective at least one first energy data output and at least one second energy data output based on the first energy data input and the second energy data input.

[0010] The system also includes that a data sensitivity of the energy optimizer is determinable with respect to the first and second energy data inputs and the first and second energy data outputs.

[0011] By utilizing data sensitivity analysis of an energy optimizer, it is possible to beneficially evaluate how the energy data output of the energy optimizer changes when the energy input data changes. In addition, the sensitivity analysis results can enable informed changes or improvements to the optimization algorithm. This can include weighted penalties to dampen the impact of changes to variables that are considered to be highly sensitive.

[0012] Furthermore, costs can be reduced by managing the corresponding energy inputs and outputs to select and adapt appropriately. For example, the purchase of electricity can be optimized during certain favorable days or nights, for example, during which cheaper tariffs can be offered.

[0013] The system can also provide transparency about the impact of different energy input parameter changes on the energy data output and can allow, for example, the selection of appropriate optimization methods accordingly. Thus, the operator of the system or the energy optimizer can appropriately take into account safety margins to prevent defects on the energy input or output side.

[0014] Furthermore, optimization of the energy optimizer may be performed without specific knowledge of the underlying prediction and / or optimization algorithms.The system according to the invention may also allow improved operation and control strategies, ie data-supported revisions of the adopted control strategies.

[0015] In this article, the field can be understood as any system or network including one or more energy consumers or producers. This can include, but is not limited to, one or more factories, microgrids, facilities, campuses and / or industrial sites. The elements of the system of the present invention (e.g., energy optimizers and / or data converters) can be arranged in the same location, or can be remotely located and connected to the system via a network.

[0016] The energy optimizer and data transformer may be implemented as virtual components, such as one or more software programs or applications working individually or in combination with each other. The energy optimizer may work based on artificial intelligence or machine learning and may include an optimization algorithm.

[0017] An optimization algorithm may be understood as a means for determining an output result based on one or more data inputs and certain boundary conditions to be considered when determining the output. The system may also include more than one optimization algorithm, which may be selected, for example, depending on the optimization goal to be achieved.

[0018] The first energy data input and the second energy data input are understood to be any suitable energy-related input data and may include, for example, the net load of the building and / or individual or group loads from the building, or the feed-in of one or more solar panels. However, the invention is not limited to these examples and may accordingly include any possible energy-related data suitable for optimization, such as data on energy production or consumption. A combination of different sources may also be used as energy data input.

[0019] The first energy data input and the second energy data input may include energy data of the same or different time steps or time periods, such as energy data of one or more different days, weeks or months. The energy input data may be stored locally or remotely on corresponding one or more data storage components (such as a database) and provided to the data converter and / or energy optimizer, for example, via a network.

[0020] The first energy data output and the second energy data output are understood to be any suitable energy related output data and may include energy data sets optimized by the energy optimizer based on the input data taking into account optimization objectives such as peak shaving. The energy data output may also include differences determined between different optimization objectives.

[0021] The data transformer may be implemented in the form of automated software for applying a perturbation to the first energy data input. Alternatively, the data transformer may be controlled by an operator to (semi-)manually apply a perturbation or change to a set of energy input data. The first energy data input and the second energy data input may include the same or different values ​​with respect to amplitude and / or corresponding time points or ranges.

[0022] The data sensitivity according to the present invention can be understood as an indicator showing how much influence the change of the data input has on the calculated data output. For example, the output may be only slightly affected by the change of the input data. Therefore, the sensitivity can be considered to be low. On the other hand, if the output is highly affected by the change of the input data, the sensitivity can therefore be considered to be high.

[0023] Furthermore, by utilizing sensitivity analysis, clear reasoning can be provided to the end user regarding the output energy strategy of the energy optimizer. As a result, the user is able to make decisions that better empower them to take control of energy management.

[0024] In a preferred embodiment, the system includes a parameter transformer as a supplement or alternative to the data transformer, wherein the parameter transformer is suitable for providing at least one first energy parameter input and at least one second energy parameter input, the second energy parameter input being different from the first energy parameter input, and wherein the energy optimizer is suitable for receiving the first parameter input and the second parameter input from the parameter transformer, and determining the corresponding at least one first energy parameter output and at least one second energy parameter output via an optimization algorithm based on the first energy parameter input and the second energy parameter input, wherein the parameter sensitivity of the energy optimizer can be determined with respect to the first energy parameter input and the second energy parameter input and the first energy parameter output and the second energy parameter output.

[0025] By utilizing parameter sensitivity analysis of an energy optimizer, it is possible to beneficially evaluate how the energy data output changes when energy parameters are changed accordingly. As a result, the optimization system of the present invention can be better tuned to a specific configuration, thereby increasing energy savings and / or reducing energy consumption. In addition, the sensitivity analysis results can enable informed changes or improvements to the optimization algorithm. This can include weighted penalties to suppress the impact of changes in variables that are considered to be highly sensitive.

[0026] It can be determined accordingly that one parameter (e.g., the initial battery state of charge) has a greater impact on the optimized output, while another parameter (e.g., the battery charging rate) has a smaller impact on the optimized output. Therefore, the energy optimizer can be analyzed and evaluated for different parameters, and the corresponding output can be optimized accordingly.

[0027] Thus, an energy parameter input may be understood as an input to an energy optimizer that is used to create a corresponding energy parameter output via an optimization algorithm. For example, a first energy parameter input may include a first optimization parameter that is provided to the energy optimizer and produces a first energy parameter output. Similarly, a second energy parameter input may be different from the first energy parameter input and may include a second optimization parameter that is correspondingly different from the first optimization parameter that is provided to the energy optimizer and produces a second energy parameter output.

[0028] The parameter changer may be implemented similarly to the data changer in the form of automated software for applying different parameters. Alternatively, the parameter changer may be controlled by an operator to (semi-)manually apply perturbations or changes to one or more energy parameters.

[0029] A parameter converter may be provided alternatively or in combination with the above-mentioned data converter. The system according to the present invention may then be adapted accordingly to determine the sensitivity with respect to the input data (i.e., data sensitivity) and / or to determine the sensitivity with respect to the input parameters (e.g., parameter sensitivity), as described herein. The energy data output and the parameter data output may be provided separately or in the form of a combined output.

[0030] In a preferred embodiment, the system includes a data transformer and a parameter transformer, and the combined data and parameter sensitivities of the energy optimizer can be determined with respect to the first energy data input and the second energy data input and the first energy data output and the second energy data output, and the first energy parameter input and the second energy parameter input and the first energy parameter output and the second energy parameter output.

[0031] Therefore, a parameter transformer may be provided in combination with the above-mentioned data transformer. The sensitivity of the energy optimizer to changes in the energy data input may be analyzed accordingly, while analyzing its sensitivity to different parameters, each of which individually or in combination affects the corresponding output of the energy optimizer. This may allow a complex combined sensitivity analysis of the energy optimizer and the corresponding optimization algorithm employed.

[0032] The variation of energy input data and parameters and the final determination of corresponding energy data output and energy parameter output may be performed subsequently or simultaneously. During the data sensitivity analysis, the optimization parameters may be constant. During the parameter sensitivity analysis, the energy input data (e.g., load amplitude) may be constant. During the parameter and data sensitivity analysis, the energy data input may be based in particular on a fixed time period or time window.

[0033] In a preferred embodiment, the at least one first energy data input is based on historical energy data, and the data transformer is preferably adapted to transform the historical energy data to create and provide the second energy data input.

[0034] Thus, the system can be correlated with known past energy data, which can be used as energy input data for subsequent determination of energy data output, e.g. for corresponding prediction or optimization targets. The historical data can preferably be further altered by a data transformer to create at least partially artificial or fictitious data, e.g. prediction data.

[0035] If historical data from the above different sources or sites is not available, historical data from another possibly similar source or site may also be used. This can enhance flexibility in determining data sensitivity and allow analysis and improvement of data output for sources for which insufficient historical data exists.

[0036] In addition, weights can be assigned to historical data to weight data that is more relevant, for example. This can be newer data, for example. Thus, in subsequent sensitivity assessments, data that may be less relevant (e.g., from past data) can be discarded or mitigated.

[0037] In a preferred embodiment, the historical energy data are based on energy data of at least one energy consumer and / or at least one energy producer.

[0038] Energy consumers and / or producers may be any device or system suitable for consuming or producing energy. Thus, energy consumers may include, for example, one or more electrically powered devices, such as lights, computers, etc., but may also include larger energy consuming entities, such as a network or university campus consisting of multiple energy consuming devices. Other examples of energy consumers may be heating, ventilation and air conditioning (HVAC) systems or electric vehicle charging, etc.

[0039] Likewise, an energy producer may be considered any entity that is allowed to provide electrical energy, which may include a single solar PV cell as well as a wind turbine park including a plurality of wind turbines.

[0040] Some devices can be both energy producers and consumers, depending on their operation. For example, a rechargeable battery can be an energy producer when it is discharged at certain times of the day (e.g., at night), and an energy consumer when it is charged at other times of the day (e.g., during the day).

[0041] In a preferred embodiment, the first energy parameter input is based on an energy parameter, and the parameter converter is preferably adapted to change the energy parameter to create and provide the second energy parameter input based on the changed energy parameter. In a preferred embodiment, the energy parameter is one or more of the following:

[0042] The initial state of charge of each battery storage system,

[0043] the limits or capacity of each battery storage system,

[0044] The charging rate and efficiency of each battery storage system,

[0045] the limits of the energy that can be drawn from the grid,

[0046] The look-ahead time of the prediction algorithm.

[0047] Energy parameters may be related to physical, chemical or electrical boundary conditions of energy producers or consumers that may affect the optimized output. Depending on the different parameters, different results of the optimization algorithm may be expected even if the same data input may be provided. Another example of a suitable energy parameter may be an optimization objective, such as peak clipping or profile smoothing.

[0048] Alternatively, the energy parameter may be related to certain variables of the algorithm used for prediction (see look-ahead time), which may change the result of the optimization. It should be clear that the above examples are not limiting and other different parameters may also be used.

[0049] In a preferred embodiment, the system further comprises an interpreter unit adapted to provide an interpretation regarding the sensitivity of the energy optimizer based on the energy data input / output, and / or the energy parameter input / output.

[0050] After obtaining output from the data or parameter sensitivities based on the changes provided by the parameter transformer and / or data transformer, the interpreter unit can help make the calculations of the energy optimizer transparent to the end user.

[0051] A set of parameters and optimization results can be stored for each data or parameter change. Therefore, the system of the present invention can realize scenario-specific sensitivity analysis, and the sensitivity of the optimization results with respect to, for example, the time, magnitude and duration of the peak load can be determined accordingly. In addition, with respect to data sensitivity analysis, for each set of data, the difference and / or error value (or its aggregation, such as the average value) and the subsequent optimization results can be stored.

[0052] Explanations can be generated by training a machine learning regressor on this data and applying an explainable AI (XAI) method (e.g., LIME). This may result in feature contributions, which can serve as a means of conveying the importance or sensitivity of each parameter with respect to the optimized output.

[0053] When computing the above explanations for a specific data set, this may result in "local" explanations. When this operation is repeated for several windows or time periods of historical data, the results are presented as "global" explanations, i.e. not specific to a particular time period, but aggregated over a larger time period. Such changes over a larger time period can be represented by a block diagram or can be shown as changes in the corresponding chart. After computing these explanations, the results can be presented to the end user in various ways. These explanations can also be aggregated by hour.

[0054] The output of the energy optimizer can be combined with the sensitivity analysis and presented to the end user in a combined manner accordingly. Providing corresponding explanations to the end user can provide the end user with the necessary information for clarifying what drives the results of the optimizer.

[0055] Because energy optimizer output may depend on many factors, the impact of each factor may not initially be obvious to the end user. Therefore, providing an explanation of how the optimization algorithm works may overcome potential confusion or rejection of the solution provided by the energy optimizer.

[0056] Furthermore, the quantified explanations could potentially be converted into natural language for better understanding by the end-users.

[0057] The invention also relates to a method for optimizing energy management at a site.

[0058] The method comprises the steps of providing at least one first energy data input and a second energy data input, and / or at least one first energy parameter input and a second energy parameter input to an energy optimizer.

[0059] The method also includes a step in which the energy optimizer determines, based on the first energy data input and the second energy data input, and / or the first energy parameter input and the second energy parameter input, a corresponding at least one first energy data output and at least one second energy data output, and / or at least one first energy parameter output and at least one second energy parameter output.

[0060] The method further comprises the step of determining a data sensitivity and / or a parameter sensitivity of the energy optimizer based on the energy data input and / or the energy parameter input and based on the energy data output and / or the energy parameter output.

[0061] Thus, the method allows optimizing energy management based on data sensitivity analysis regarding the energy optimizer's corresponding output regarding different data inputs. In other words, the method of the invention allows determining the impact or importance of different energy data inputs on changes in the corresponding energy data output.

[0062] Likewise, the method of the present invention allows optimizing energy management based on parameter sensitivity analysis regarding the impact or importance of different energy parameter changes on the energy parameter output.

[0063] The determination of the sensitivity with respect to the changing input data and with respect to the changing energy parameter can be performed separately or simultaneously in the form of a combined sensitivity determination taking into account the changing data input and the changing energy parameter.

[0064] The present method can also be used with existing prediction and optimization algorithms, thereby providing an agnostic approach to energy management analysis.

[0065] In a preferred embodiment, the first energy data input is formed by historical energy data, and the method further comprises the following steps:

[0066] determining, by an energy optimizer, a first energy data forecast based on historical energy data,

[0067] providing a first energy data forecast to an optimization algorithm,

[0068] Determine a first optimization target based on the first energy data prediction and optimization parameters by an optimization algorithm,

[0069] Providing ground truth data to the optimization algorithm, and

[0070] A second optimization objective is determined by an optimization algorithm based on the ground truth data and the optimization parameters.

[0071] Thus, different optimization goals can be achieved based on ground truth data on the one hand and on historical data on the other hand. The prediction can be performed by any suitable algorithm or calculation method. The prediction can be based on energy data for one or more time points or time windows and can provide a prediction of the time development of the above energy data.

[0072] Ground truth data can be understood as known (past) real data. That is, historical data may not be simulated, fictitious or predicted data. For example, such ground truth data may be energy data for last year.

[0073] In addition, in another embodiment, energy data input (e.g., historical data) can be directly provided to the optimization algorithm without prediction. In this case, historical energy data can be used instead of predicted data for sensitivity analysis.

[0074] According to the invention, the optimization objectives may be, for example, peak (load) reduction, load shifting, cost reduction, maximization or minimization of energy input and / or output. One optimization objective may be to achieve fast charging of energy storage systems provided on site. Another optimization objective may be to reduce power consumption at certain times of the day, e.g., at which times local or global power consumption is at a maximum. The invention is of course not limited to the above examples, but is applicable to additional and different optimization objectives that may be acquired by the energy optimizer.

[0075] The present invention may provide the advantage that when peak loads are predicted, the sensitivity analysis output may be used to estimate what-if scenarios for better planning, such as developing appropriate fallback strategies. This may enhance and protect energy supply at the site.

[0076] In a preferred embodiment, the method further comprises the following steps:

[0077] Change historical energy data,

[0078] providing the changed energy data to an energy optimizer,

[0079] determining, by the energy optimizer, one or more changed energy data predictions based on the changed energy data,

[0080] providing one or more modified energy data forecasts to an optimization algorithm, and

[0081] One or more changed optimization objectives are determined by the optimization algorithm based on the one or more changed energy data predictions, wherein the optimization parameters remain constant.

[0082] Thus, a sensitivity analysis with respect to the changing data points can be performed so that the importance or impact of the data changes on the optimization algorithm output can be estimated. Providing constant optimization parameters can allow for analysis of the optimization algorithm's influence on changes in energy input data to be defined. In other words, the difference between the optimization target based on the ground truth data and the changed optimization target predicted based on the changed energy data can be directly assignable to the changed energy data.

[0083] Therefore, the sensitivity determination of different energy data can be achieved by performing data sensitivity analysis as follows:

[0084] 1. Select a fixed window / time period of historical data.

[0085] 2. For the selected data, perturb the time series data (i.e., add disturbances) within an appropriate range to create different variations of the historical data - these can also be assumed to represent various forecasts.

[0086] ○ For each of these predictions:

[0087] ■ Calculate the difference between the original historical data and the perturbed version for the selected time window

[0088] ■Running optimization algorithm

[0089] ○Store the difference (or an aggregation thereof, such as the average) and the subsequent optimization results

[0090] 3. Repeat (1) and (2) for several windows / time periods of historical data.

[0091] In a preferred embodiment, the method further comprises the following steps:

[0092] determining at least one prediction difference between the ground truth data and each of the one or more energy data predictions,

[0093] At least one optimization objective difference between a second optimization objective based on the ground truth data and each of the optimization objectives based on each of the energy data predictions is determined.

[0094] Thus, the relevant difference between the prediction and the ground truth data, and the final optimization target based on the ground truth data and the corresponding prediction can be obtained. In some embodiments, the method can also include providing only one of the above-mentioned determination steps, i.e., determining at least one prediction difference between the ground truth data and each of the one or more energy data predictions, or determining at least another optimization target difference between a second optimization target based on the ground truth data and each optimization target based on each energy data prediction.

[0095] By determining the corresponding difference, the quality of the prediction can be evaluated because the prediction can be compared with historical data (i.e., real past data). For example, energy data for one month last year can be provided to the energy optimizer as historical data. The energy optimizer can then provide a prediction for the next month based on the above historical data. The predicted data can then be compared with known real historical ground truth data for the corresponding month last year. Therefore, the quality of the prediction, such as accuracy, can be determined.

[0096] Furthermore, the differences between different optimization objectives based on the optimization algorithm results can provide insights into how changes in the input data based on different predictions affect the optimization algorithm output.

[0097] In a preferred embodiment, the first energy data input is formed by historical energy data, the method further comprising the steps of:

[0098] The energy optimizer determines the constant energy data forecast based on the historical energy data,

[0099] Provide constant energy data predictions to the optimization algorithm,

[0100] Determine the first optimization target based on constant energy data prediction and optimization parameters through the optimization algorithm,

[0101] Change the optimization parameters,

[0102] providing the changed optimization parameters to an energy optimizer,

[0103] One or more changed optimization targets are determined by the optimization algorithm based on the constant energy data prediction and the one or more changed optimization parameters.

[0104] Thus, a sensitivity analysis on the changed optimization parameters can be performed so that the importance or impact of the parameter changes on the optimization algorithm output can be estimated. Thus, it can be determined how sensitive the energy optimizer is to changes in the optimization parameters. Providing a constant energy data prediction can allow for a limited analysis of the optimization algorithm's impact on energy parameter changes. In other words, the difference between the optimization target based on the ground truth data and the changed optimization target based on the changed optimization parameters can be directly assignable to the changed optimization parameters.

[0105] Therefore, the sensitivity determination of different parameters can be achieved by performing parameter sensitivity analysis as follows:

[0106] 1. Select a fixed window / time period of historical data.

[0107] Either simply use this historical data or pass it to a forecaster to generate a forecast for a given timeframe

[0108] 2. For the selected data, vary the above input parameters within the appropriate range to create different setups / configurations / scenario arrangements.

[0109] ○ For each configuration, run the optimization algorithm

[0110] ○Storage of configuration and subsequent optimization results

[0111] 3. Repeat (1) and (2) for several windows / time periods of historical data.

[0112] The invention also relates to a computer program product comprising instructions which, when executed by a computer, cause the computer to perform any of the methods disclosed herein.

[0113] The features of the system according to the present invention can be realized by corresponding suitable digital or computing components, which may include, for example, one or more computers, applications and / or networks. The method may be at least partially implemented by a computer, and may be implemented in software or hardware, or in software and hardware. In addition, the method may be implemented by computer program instructions running on components providing data processing functions. The data processing component may be a suitable computing component, such as an electronic control module, etc., which may also be a distributed computer system. The data processing component or computer may include one or more of a processor, a memory, a data interface, etc., respectively.

[0114] The present invention also relates to a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to perform any of the methods disclosed herein.

[0115] The features and advantages outlined above in the context of systems and methods also apply to the computer program products and computer-readable media described herein.

[0116] The computer program may be stored / distributed on a suitable medium, such as an optical storage medium or a solid-state medium, provided together with or as a part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0117] Additional features, examples and advantages will become apparent from the following detailed description of the preferred embodiments and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0118] In order to better understand the present invention and illustrate its practicality, the following drawings are provided and reference is made to them. It should be understood that the drawings represent only exemplary embodiments and therefore in no way limit the scope of the present invention. The same or similar functional elements are always represented by the same reference numerals. Any reference numerals in the claims should not be interpreted as limiting the scope of the claims.

[0119] In the accompanying drawings,

[0120] Figure 1 A system for optimizing energy management on site according to the present invention is schematically shown;

[0121] Figure 2 is a flow chart illustrating a method for optimizing energy management at a site according to the present invention;

[0122] Figure 3 is a flow chart illustrating a method for optimizing energy management at a site according to the present invention;

[0123] Figure 4 The perturbations of the historical energy data are schematically shown;

[0124] Figure 5 Parameter sensitivity results for different optimization parameters are schematically shown. DETAILED DESCRIPTION

[0125] Figure 1 A system 100 for optimizing energy management on site according to the present invention is shown. In this article, historical energy data 22 from different sources (such as, for example, load historical energy data at a given time point or time range, solar historical energy data, and other unspecified historical energy data) are provided to a data converter 20. The historical energy data is then provided (see step S1) as energy data input to the energy optimizer 10. The energy optimizer then determines (see step S6) a prediction for each energy data provided, such as a prediction for the next moment. The energy prediction data 26 or historical data is then provided to the optimization algorithm 14 (see step S7). The optimization algorithm 14 is also provided with an energy parameter input (see step S3). The energy optimizer then determines (see step S2) an energy data output, which is provided to a data storage device 50, where the result data obtained from the energy optimizer 10 is saved.

[0126] The historical energy data 22 from different sources may also be changed by the data transformer 20 (see step S9). In this context, a disturbance / change in the energy data at a given moment is applied to a data set. The data set thus modified may then also be provided to the energy optimizer 10, for example for creating a corresponding prediction model. The resulting energy data prediction 26 may then be provided to the optimization algorithm 14. Alternatively, the disturbed energy data may be provided directly to the optimization algorithm 14 as input. Thus, no prediction is used as data input. The energy optimizer 10 then calculates the corresponding energy data output based on the disturbance data, which is also provided to the data storage device 50.

[0127] Likewise, the parameter transformer 30 may provide different optimization parameters 32 to the energy optimizer 10 (see step S3). Thus, for example, different optimization parameters 32 may be used together with a constant or changed energy data forecast 26 or with a constant or changed historical energy data 22. The energy optimizer 10 accordingly determines one or more energy parameter outputs (see step S4), which may be provided accordingly and stored in the data storage device 50.

[0128] Any information about the data input (e.g., the applied changes) may also be saved or stored in the data storage device 50 (see step S14). Likewise, any information about the parameters (e.g., the applied changes) may also be saved or stored in the data storage device 50 (see step S15).

[0129] Based on the obtained energy data output and the obtained parameter output, the sensitivity of the corresponding data and / or parameters can be determined (see step S5), which sensitivity can be provided to the interpreter unit 40. The energy data output and the parameter output can be, for example, different optimization results obtained from the energy optimizer 10 for different energy data inputs and / or different optimization parameters.

[0130] The interpreter unit 40 may be a machine learning or explainable AI unit that is trained by the stored input and output data. Thus, the interpreter unit 40 may clarify information to the user about the potential impact of changes in input data on output data and / or the potential impact of changes in different optimization parameters on output data.

[0131] Furthermore, a computer program product 210 and a computer readable medium 220 are shown, each comprising instructions which, when executed by the computer 200, cause the computer 200 to perform a method according to the invention.

[0132] Figure 2is a flow chart illustrating a method according to the present invention, in particular with respect to sensitivity analysis based on changes in energy input data. In this document, in step S1a, historical energy data 22 are provided as a first energy data input to the energy optimizer 10. The historical energy data 22 may be, for example, past energy consumption at a site. In step S6a, the energy optimizer 10 determines a first energy data forecast 26a based on the data input. For example, this may be a forecast of energy consumption for the next day. The energy data forecast may then be compared with ground truth data 24 acquired based on the historical data 22. For example, this is known energy consumption for the day. Then in step S10, the difference between the first energy data forecast 26a and the ground truth data 24 is determined. In step S7a, the first energy data forecast 26a is then provided to the optimization algorithm 14. Then, in step S8a, the optimization of the algorithm 14 uses the optimization parameter 32 to determine the first optimization target 12a, such as peak load reduction.

[0133] Likewise, in step S7c, the ground truth data 24 is provided to the optimization algorithm 14. Then, in step S8c, the optimization of the algorithm 14 determines a second optimization objective 12c, such as different peak load shedding, using the optimization parameters 32. The optimization parameters 32 are kept constant.

[0134] Then, in step S11, the first optimization target 12a and the second optimization target 12c are compared, and the difference is determined. Figure 1 As shown, historical energy data from different sources, such as load, solar energy, etc., can be used.

[0135] As shown in step S9, the historical energy data 22 can be changed by the data transformer 20. For example, disturbances can be included to change the data, such as the load of a specific time window. Then in step S1b, the changed data is provided as the second energy data input of the energy optimizer 10. Then, as shown in step S6b, a second energy data prediction 26b, also referred to as a changed energy data prediction, can be determined. In step S10, the difference between the changed energy data prediction 26b and the ground truth data 24 based on the changed energy data is determined accordingly. It should be noted that the changed historical data can also be used instead of the prediction for comparison with the ground truth data 24. In step S7b, the second energy data prediction 26b (or the changed historical data) is then provided to the optimization algorithm 14. Then, in step S8b, the optimization algorithm 14 determines the changed optimization target, i.e., the changed data optimization target 12b, such as peak load reduction, based on the changed historical energy data using the optimization parameter 32 that remains constant.

[0136] Then in step S11 , the changed data optimization target 12 b and the second optimization target 12 c are compared and the respective differences are determined.

[0137] Therefore, the sensitivity of the energy optimizer to disturbances or changes in the energy data input can be determined accordingly. In other words, if a changed energy data input is provided, it can be determined how much the energy data output changes.

[0138] The difference between the first energy data prediction 26a determined in step S10 and the ground truth data 24 may be stored in a table such as Figure 2 As shown in the lower part (indicated by "*"). Likewise, the difference between the modified energy data prediction 26b and the ground truth data 24 determined in step S10 may be stored in a table.

[0139] The difference determined in step S10 may be determined for historical energy data from different sources, such as load, solar energy, etc. Thus, as shown in the second and third columns of the table shown, the average difference over time steps between actual historical data and a given forecast may be determined and saved, for example with respect to "energy load" and "solar energy". It should be noted that the sources are not limited to these examples, and other sources of energy data may also be evaluated, which may then be stored in corresponding other additional columns.

[0140] In step S2a, the result of the difference between the first optimization target 12a based on the first prediction determined in step S11 and the second optimization target 12c based on the ground truth data 24 can be determined and stored in the right column of the table shown. Similarly, in step S2b, the result of the difference between the (multiple) changed optimization targets based on the changed energy data prediction 26b determined in step S11 and the second optimization target based on the ground truth data 24 can be determined and saved and stored in the right column of the table.

[0141] Thus, the difference determined in steps S10 and S11 for the corresponding changed energy data can be determined and saved for each source. The applied change is indicated by an identifier ("data change index") stored in the left column. Thus, the data sensitivity of the impact or significance of the change of the data input to the corresponding data output can be estimated.

[0142] Then, the result matrix of the table can be used to train the interpreter AI or machine learning unit accordingly. Here, the difference obtained in step S10 can be used as the input or feature of the model training, and the difference obtained in step S11 can also be used as the output of the feature of the model training.

[0143] Figure 3 is a flow chart illustrating a method according to the invention, in particular with regard to sensitivity analysis based on variations of the optimization parameters 32. Note that, except for reference Figure 2 In addition to the method described, this method can also be provided.

[0144] and Figure 2 Similar to the method of , in step S1a, the historical energy data 22 is provided as the first energy data input of the energy optimizer 10. However, Figure 2 In contrast to the method of , the historical data is not changed, but remains unchanged. In step S6d, the energy optimizer 10 determines one or more predictions, such as the constant energy data prediction 26d, based on the input data based on the constant historical data. In step S7d, the constant energy data prediction 26d is then provided to the optimization algorithm 14. The historical energy data 22 can also be provided directly to the optimization algorithm 14. Then, in step S8a, the optimization of the algorithm 14 uses the optimization parameters 32 provided in step S13 to determine the first optimization target 12a. Therefore, a first energy parameter output can be determined and stored in a table, as shown in step S4a.

[0145] As shown in step S12, one or more optimization parameters 32 may be changed by the parameter transformer 30. As shown in step S13a, the changed optimization parameters 32a may be provided to the energy optimizer 10. Then, in step S8d, the optimization algorithm 14 employs the changed optimization parameters 32a to determine a changed parameter optimization target 12d based on the changed optimization coefficients 32a. Thus, a second energy parameter output may be determined and stored in a table, as shown in step S4b.

[0146] Thus, the sensitivity of the energy optimizer to data disturbances or changes can be determined. In other words, when a changed energy parameter input is provided, it can be determined how the energy parameter output changes.

[0147] In the depicted embodiment, the different parameters 32, 32a exemplarily relate to "battery initial charge state", "grid limit" and "battery capacity" (each indicated by "*"), which are stored in the second and third columns of the table, respectively, as shown in FIG. Figure 3 As shown in the lower part of . Obviously, any other suitable optimization parameters can also be added in other columns. Similarly, the corresponding results of the first optimization target 12a and the changed optimization target 12d determined for each optimization parameter in steps S4a and S4b can be stored or saved in the table (see the right column). Therefore, the parameter sensitivity with respect to the influence or importance of different energy parameters on the change of the corresponding output can be estimated.

[0148] Therefore, the result matrix of the table can be used to train the interpreter AI or machine learning unit. In this article, different optimization parameters can be used as input or features of model training, and the corresponding first optimization target and second optimization target can be used as output of features of model training.

[0149] Figure 4Two examples (i and ii) of an energy data set 28 for a specific time period t are shown. On the x-axis, the time is shown and on the y-axis the amplitude is shown, which can be, for example, the power consumption. As indicated by the small double arrows, the amplitude of the data set can be varied accordingly to produce a disturbance in the energy data set. The disturbance can have, for example, a constant amplitude variation range, as shown in the upper figure i), or a different amplitude variation range, as shown in the lower figure ii). The amplitude variation can be applied to every data point or only to certain data points. These variations can also be applied by providing a corresponding constant offset or by creating random deviations from the original data set. Any type of suitable disturbance or variation is conceivable.

[0150] Figure 5 An example of a parameter sensitivity result 34 is shown, which can be indicated to the user via the AI ​​interpreter. The figure shows a visualization of the importance or sensitivity of each of the optimization parameters with respect to the optimization output. In this example, the battery initial state of charge shows a high value. Therefore, the optimization output is highly affected by this optimization parameter. In contrast, the battery charging rate only shows a low value and is therefore less important and has little impact on the optimization output. In addition, as shown in the figure, a corresponding error tolerance can also be provided for each parameter.

[0151] Reference numerals list

[0152] 10: Energy Optimizer

[0153] 12a: First Optimization Goal

[0154] 12b: Changed data optimization goals

[0155] 12c: Second optimization goal

[0156] 12d: Changed parameter optimization target

[0157] 14: Optimization Algorithm

[0158] 20: Data Transformer

[0159] 22: Historical energy data

[0160] 24: Ground truth data

[0161] 26: Energy data prediction

[0162] 26a: First Energy Data Prediction

[0163] 26b: Changed Energy Data Forecast

[0164] 26d: Constant Energy Data Prediction

[0165] 28: Energy Dataset

[0166] 30: Parameter Transformer

[0167] 32: Optimize parameters

[0168] 32a: Changed optimization parameters

[0169] 34: Parameter sensitivity results

[0170] 40: Interpreter unit

[0171] 50: Data storage device

[0172] 100: System

[0173] 200: Computer

[0174] 210: Computer program products

[0175] 220: Computer readable medium

[0176] S1: Provision of energy data input

[0177] S1a: Providing the first energy data input

[0178] S1b: Provision of the second energy data input

[0179] S2: Determination of energy data output

[0180] S2a: Determination of the first energy data output

[0181] S2b: Determination of the second energy data output

[0182] S3: Provision of energy parameter input

[0183] S4: Determination of energy parameter output

[0184] S4a: Determination of the first energy parameter output

[0185] S4b: Determination of the second energy parameter output

[0186] S5: Determination of sensitivity

[0187] S6: Confirmation of prediction

[0188] S6a: Determination of the first prediction

[0189] S6b: Determination of the second prediction

[0190] S7: Provision of forecast or historical data

[0191] S7a: Providing the first energy data prediction

[0192] S7b: Providing the second energy data prediction

[0193] S7c: Provision of ground truth data

[0194] S7d: Provision of constant energy data forecast

[0195] S8a: Determination of the first optimization goal

[0196] S8b: Determination of a changed optimization target based on the changed data

[0197] S8c: Determination of the second optimization objective

[0198] S8d: Determination of a changed optimization target based on the changed parameters

[0199] S9: Changes in energy data

[0200] S10: Determination of prediction error

[0201] S11: Determination of optimization target difference

[0202] S12: Changes in optimization parameters

[0203] S13: Provision of optimization parameters

[0204] S13a: Providing the changed optimization parameters

[0205] S14: Storage data information

[0206] S15: Storage parameter information

Claims

1. A system (100) for optimizing energy management at a site, the system (100) comprising: An energy optimizer (10), wherein the energy optimizer (10) is adapted to determine one or more optimization objectives (12) based on an optimization algorithm (14), a data converter (20), wherein the data converter (20) is adapted to provide at least one first energy data input and at least one second energy data input to the energy optimizer (10), the second energy data input being different from the first energy data input, wherein the energy optimizer (10) is adapted to receive the first energy data input and the second energy data input from the data converter (20), and to determine, via the optimization algorithm (14), corresponding at least one first energy data output and at least one second energy data output based on the first energy data input and the second energy data input, The data sensitivity of the energy optimizer (10) can be determined with respect to the first energy data input and the second energy data input and the first energy data output and the second energy data output.

2. The system (100) according to the preceding claim, The system (100) comprises a parameter converter (30), which is a supplement or alternative to the data converter (20). wherein the parameter converter (30) is adapted to provide at least one first energy parameter input and at least one second energy parameter input, the second energy parameter input being different from the first energy parameter input, and The energy optimizer (10) is adapted to: receiving the first parameter input and the second parameter input from the parameter converter (30), and determining at least one first energy parameter output and at least one second energy parameter output via the optimization algorithm (14) based on the first energy parameter input and the second energy parameter input, The parameter sensitivity of the energy optimizer (10) can be determined with respect to the first energy parameter input and the second energy parameter input and the first energy parameter output and the second energy parameter output.

3. The system (100) according to the preceding claim, wherein the system (100) comprises the data transformer (20) and the parameter transformer (30), and Wherein the combined data and parameter sensitivity of the energy optimizer (10) can be determined with respect to a combination of the first energy data input and the second energy data input and the first energy data output and the second energy data output, and the first energy parameter input and the second energy parameter input and the first energy parameter output and the second energy parameter output.

4. A system (100) according to any of the preceding claims, wherein the at least one first energy data input is based on historical energy data (22), and wherein the data transformer (20) is preferably adapted to change the historical energy data (22) to create and provide the second energy data input.

5. The system (100) according to the preceding claim, wherein the historical energy data (22) are based on energy data of at least one energy consumer and / or at least one energy producer.

6. A system (100) according to any one of the preceding claims 2 to 5, wherein the first energy parameter input is based on an energy parameter, and wherein the parameter converter (30) is preferably adapted to change the energy parameter to create and provide the second energy parameter input based on the changed energy parameter.

7. The system (100) according to the preceding claim, wherein the energy parameter is one or more of the following: The initial state of charge of each battery storage system, The limits / capacity of each battery storage system, The charging rate and efficiency of each battery storage system, the limits of the energy that can be drawn from the grid, • The look-ahead time of the prediction algorithm.

8. The system (100) according to any of the preceding claims, further comprising an interpreter unit (40), which is suitable for providing an explanation about the sensitivity of the energy optimizer (10) based on the energy data input / the energy data output, and / or the energy parameter input / the energy parameter output.

9. A method for optimizing energy management at a site, comprising the steps of: providing (S1a / S1b) at least one first energy data input and at least one second energy data input, and / or at least one first energy parameter input and at least one second energy parameter input to an energy optimizer (10), The energy optimizer (10) determines (S2a / S2b) at least one first energy data output and at least one second energy data output based on the first energy data input and the second energy data input, and / or the first energy parameter input and the second energy parameter input, and / or determines (S4a / S4b) at least one first energy parameter output and at least one second energy parameter output, and A data sensitivity and / or a parameter sensitivity of the energy optimizer (10) is determined based on the energy data input and / or the energy parameter input and based on the energy data output and / or the energy parameter output.

10. The method according to the preceding claim, wherein the first energy data input is formed by historical energy data (22), The method further comprises the following steps: determining (S6a) by the energy optimizer (10) a first energy data forecast (26a) based on the historical energy data (22), providing (S7a) said first energy data prediction (26a) to an optimization algorithm (14), determining (S8a) a first optimization target (12a) by means of the optimization algorithm (14) based on the first energy data prediction (26a) and the optimization parameter (32), providing (S7c) ground truth data (24) to the optimization algorithm (14), and A second optimization objective (12c) is determined (S8c) by the optimization algorithm (14) based on the ground truth data (24) and the optimization parameters (32).

11. The method according to the preceding claim further comprises the following steps: changing (S9) the historical energy data (22), providing (S1b) the changed energy data to the energy optimizer (10), determining (S6b) by the energy optimizer (10) based on the changed energy data one or more changed energy data predictions (26b), providing (S7b) the one or more modified energy data forecasts (26b) to the optimization algorithm (14), and One or more changed data optimization targets (12b) are determined (S8b) by the optimization algorithm (14) based on the one or more changed energy data predictions (26b), wherein the optimization parameter (32) remains constant.

12. The method according to the preceding claim further comprises the following steps: determining (S10) at least one prediction difference between the ground truth data (24) and each of the one or more energy data predictions (26a, 26b), Determine (S11) at least one optimization objective difference between the second optimization objective (12c) based on ground truth data and each of the optimization objectives (12a, 12b) based on each of the energy data predictions (26a, 26b).

13. The method according to any one of the preceding claims, Wherein the first energy data input is formed by historical energy data (22), the method further comprises the following steps: determining (S6d) by the energy optimizer (10) a constant energy data forecast (26d) based on the historical energy data (22), providing (S7d) said constant energy data prediction to an optimization algorithm (14), determining (S8a) a first optimization target (12a) by means of the optimization algorithm (14) based on the constant energy data prediction (26d) and the optimization parameter (32), changing (S12) the optimization parameter (32), providing (S13) the changed optimization parameters to the energy optimizer (10), One or more changed optimization targets (12d) are determined (S8d) by the optimization algorithm (14) based on the constant energy data prediction (26d) and one or more changed optimization parameters (32a).

14. A computer program product (210) comprising instructions which, when executed by a computer (200), cause the computer (200) to perform the method according to any one of claims 9 to 13.

15. A computer readable medium (220) comprising instructions which, when executed by a computer (200), cause the computer (200) to perform the method according to any one of claims 9 to 13.

Images