Molten salt energy storage system intelligent control method based on machine learning

Abstract

The invention discloses an intelligent control method for a fused salt energy storage system based on machine learning. The method comprises the following steps: determining historical control parameters of different historical frequency modulation instructions according to a historical frequency modulation log and calculating an application evaluation value; analyzing the historical control parameters and the application evaluation values based on a machine learning model to obtain an optimal control parameter set; determining an optimal control parameter according to the use cost minimization objective function, the frequency modulation efficiency maximization objective function and the optimal control parameter set; generating a similarity degree according to a first similarity coefficient of the real-time frequency modulation instruction and the historical frequency modulation instruction and a second similarity coefficient of the real-time operation condition and the historical operation condition; and determining a prediction optimal control parameter and a prediction control sub-strategy of the real-time frequency modulation instruction according to the similarity degree, and performing accurate regulation and control on the fused salt energy storage coefficient according to the real-time frequency modulation instruction and the prediction optimal control parameter, thereby improving the frequency modulation efficiency of the fused salt energy storage system, and improving the safety and stability of a thermal power generating unit and an electric power system.

Description

Technical Field

[0001] The present application relates to the field of molten salt energy storage technology, and in particular to an intelligent control method for a molten salt energy storage system based on machine learning. Background Art

[0002] Molten salt energy storage is a technology that converts chemical energy into thermal energy and then into electrical energy. The basic principle of the molten salt energy storage system is to heat the molten salt to a high temperature, store chemical energy, and convert thermal energy into electrical energy when electrical energy is needed. As a form of energy storage, the molten salt energy storage system has the ability to quickly adjust power. It can assist thermal power units in frequency regulation as a controllable load, store energy when the unit output is limited, and replace part of the unit's heating when the unit is under high load.

[0003] In the prior art, after the molten salt energy storage coefficient receives the frequency modulation instruction of the thermal power unit, it mostly relies on operating experience for manual operation, and cannot efficiently and accurately regulate the molten salt energy storage system according to the frequency modulation instruction, thereby reducing the safety, stability and economy of the thermal power unit and the power system. Summary of the invention

[0004] To solve the above technical problems, the present application provides an intelligent control method for a molten salt energy storage system based on machine learning. The historical control parameters of different historical frequency modulation instructions are determined through historical frequency modulation logs, and the application evaluation values ​​of the historical control parameters are calculated. The preferred control parameter set of different historical frequency modulation instructions is determined according to the machine learning model. The preferred control parameter set is analyzed by minimizing the usage cost objective function and maximizing the frequency modulation efficiency objective function to determine the optimal control parameters. The predicted optimal control parameters and the predicted control sub-strategy are determined according to the real-time frequency modulation instructions and real-time operating conditions of the molten salt energy storage system, and control instructions are generated. The molten salt energy storage coefficient is accurately regulated according to the real-time frequency modulation instructions and the predicted optimal control parameters, so as to improve the frequency modulation efficiency of the molten salt energy storage system, thereby improving the safety and stability of thermal power units and power systems.

[0005] In some embodiments of the present application, a method for intelligent control of a molten salt energy storage system based on machine learning is provided, comprising:

[0006] Obtain a historical frequency modulation log of the molten salt energy storage system, determine historical control parameters of different historical frequency modulation instructions according to the historical frequency modulation log, and calculate an application evaluation value of the historical control parameters;

[0007] Analyze historical control parameters of different historical frequency modulation instructions and corresponding application evaluation values ​​based on a machine learning model to obtain an optimal control parameter set of historical frequency modulation instructions;

[0008] Establish an objective function of minimizing the use cost and an objective function of maximizing the frequency modulation efficiency, and determine the optimal control parameters corresponding to the historical frequency modulation instructions in combination with the optimal control parameter set;

[0009] Acquire a real-time frequency modulation instruction and real-time operating conditions of multiple energy storage units in the molten salt energy storage system, and generate a similarity degree according to a first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction and a second similarity coefficient between the real-time operating condition and the historical operating condition;

[0010] The predicted optimal control parameters and the predictive control sub-strategy of the real-time frequency modulation command are determined according to the similarity, and the control command of the molten salt energy storage system is generated.

[0011] In some embodiments of the present application, determining historical control parameters of different historical frequency modulation instructions according to the historical frequency modulation log includes:

[0012] Obtaining a corresponding historical frequency modulation instruction according to the historical frequency modulation log, wherein the historical frequency modulation instruction includes a target frequency modulation value and a corresponding historical control duration;

[0013] A time reference line is established based on the historical modulation duration of each historical frequency modulation log, and multiple historical collection nodes are set according to preset time intervals;

[0014] Based on multiple historical collection nodes of each time reference line, historical actual frequency modulation values ​​and historical operation data of molten salt energy storage coefficient are collected and mapped to corresponding historical collection nodes;

[0015] According to the degree of synchronization change between the historical actual frequency modulation value and the historical operation data at each historical collection node in the same time reference line, if the degree of synchronization change is greater than the preset synchronization change degree threshold, the corresponding historical operation data is determined as the historical control parameter of the historical frequency modulation instruction corresponding to the current time reference line.

[0016] In some embodiments of the present application, calculating the application evaluation value of the historical control parameter includes:

[0017] Compare the historical control parameters with the preset standard control parameter interval of the corresponding historical frequency modulation instructions to obtain the number of historical control parameters that are not in the preset standard control parameter interval, the control parameter difference amount that the historical control parameters are not in the preset standard control parameter interval, and the comprehensive duration that the historical control parameters are not in the preset standard control parameter interval;

[0018] Generate a first application evaluation value according to the number, the control parameter difference and the comprehensive duration;

[0019] Determine a final frequency modulation value of a corresponding historical control parameter according to the historical frequency modulation log, and generate a second application evaluation value according to an absolute value of a difference between the final frequency modulation value and a target frequency modulation value;

[0020] Calculate the variance of historical control parameters within the corresponding historical regulation period, and set the compensation coefficient according to the variance;

[0021] An application evaluation value corresponding to the historical control parameter is generated according to the first application evaluation value, the second application evaluation value and the compensation coefficient of the same historical frequency modulation log.

[0022] In some embodiments of the present application, the calculation formula of the application evaluation value is:

[0023]

[0024] Wherein, W is the application evaluation value, a1 is the weight coefficient of the first application evaluation value, y1 is the first application evaluation conversion coefficient, m is the number of historical control parameters that are not in the preset standard control parameter range, n is the historical control parameter of the same historical frequency modulation log, t is the comprehensive duration of the historical control parameter not in the preset standard control parameter range, △Ps is the control parameter difference amount that the historical control parameter is not in the preset standard control parameter range, is the first application evaluation value, |△C| is the absolute value of the difference between the final frequency modulation value and the target frequency modulation value, y2 is the second application evaluation conversion coefficient, a2 is the weight coefficient of the second application evaluation value, and c is the compensation coefficient.

[0025] In some embodiments of the present application, historical control parameters of different historical frequency modulation instructions and corresponding application evaluation values ​​are analyzed based on a machine learning model, including:

[0026] Based on the machine learning model, historical control parameters of different frequency modulation instructions are divided to obtain historical control parameter groups of multiple historical frequency modulation logs of the same historical frequency modulation instruction;

[0027] Filter out historical frequency modulation logs whose historical control duration is less than a preset control duration threshold from multiple historical frequency modulation logs of the same historical frequency modulation instruction, and set the historical control parameter group corresponding to the historical frequency modulation log as the first subset of the current historical frequency modulation instruction, and sort the historical control parameter groups in the first subset according to the length of the historical control duration to obtain a first sorting result;

[0028] The historical control parameter groups for which the application evaluation value of the same historical frequency modulation instruction is greater than a preset application evaluation value threshold are set as a second subset, and the historical control parameter groups in the second subset are sorted according to the size of the application evaluation value to obtain a second sorting result.

[0029] In some embodiments of the present application, obtaining a preferred control parameter set of a historical frequency modulation instruction includes:

[0030] Compare the first subset and the first sorting result with the second subset and the second sorting result to obtain a first sorting value in the first subset and a second sorting value in the second subset for the same historical control parameter group;

[0031] Subtracting the first ranking value from the second ranking value to obtain an absolute value of the ranking difference, and eliminating the historical control parameter group whose absolute value of the ranking difference is greater than a preset absolute value of the ranking difference;

[0032] A preferred control parameter set corresponding to the historical frequency modulation instructions is constructed according to the remaining historical control parameter groups, multiple historical control parameter groups in the preferred control parameter set are set as multiple first centers, application evaluation values ​​corresponding to the multiple historical control parameter groups in the preferred control parameter set are set as multiple second centers, and a first comprehensive ranking value is determined according to the first ranking value and the second ranking value of the historical control parameter group, and the ranking results of the first center and the second center are determined according to the first comprehensive ranking value.

[0033] In some embodiments of the present application, a cost minimization objective function and a frequency modulation efficiency maximization objective function are established, and the optimal control parameters corresponding to the historical frequency modulation instructions are determined in combination with the preferred control parameter set, including:

[0034] The objective function of minimizing the usage cost is:

[0035] H1 = min[B(F, m, n)];

[0036] Among them, H1 is the objective function of minimizing the use cost, B(F, m, n) is the use cost from the mth first center to the corresponding nth second center in the Fth historical control instruction, and min is the minimum value symbol;

[0037] The objective function of maximizing frequency modulation efficiency is:

[0038] H2 = max[X(F, m, n)];

[0039] Wherein, H2 is the objective function of maximizing frequency modulation efficiency, X(F, m, n) is the frequency modulation efficiency from the mth first center to the corresponding nth second center in the Fth historical control instruction, and max is the maximum value symbol;

[0040] According to the objective function of minimizing the use cost and the objective function of maximizing the frequency modulation efficiency, the preferred control parameter set of each historical frequency modulation instruction is analyzed to obtain the use cost and frequency modulation efficiency of multiple historical control parameter groups in the preferred control parameter set, and the second comprehensive ranking value of the corresponding historical control parameter group is generated according to the use cost and the frequency modulation efficiency, and the first historical control parameter group selected by the objective function of minimizing the use cost and the second historical control parameter group selected by the objective function of maximizing the frequency modulation efficiency are obtained;

[0041] Compare the first historical control parameter group with the second historical control parameter group, and if they are the same historical control parameter group and the first comprehensive ranking value is greater than the preset ranking value threshold, set the corresponding historical control parameter group as the optimal control parameter of the corresponding historical frequency modulation instruction;

[0042] If they are not the same historical control parameter group, determine whether the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group and the second historical control parameter group are both greater than the preset ranking value threshold; if the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group or the second historical control parameter group are both greater than the preset ranking value threshold, set the corresponding historical control parameter group as the optimal control parameter of the corresponding historical frequency modulation instruction; if the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group and the second historical control parameter group are not both greater than the preset ranking value threshold, determine the third comprehensive ranking value according to the first comprehensive ranking value and the second comprehensive ranking value, and set the historical control parameter group with the largest third comprehensive ranking value in the preferred control parameter set as the optimal control data of the corresponding historical frequency modulation instruction.

[0043] In some embodiments of the present application, the similarity degree is generated according to the first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction and the second similarity coefficient between the real-time operating condition and the historical operating condition, including:

[0044] Subtracting the target frequency modulation value of the real-time frequency modulation instruction from the target frequency modulation value of the historical frequency modulation instruction to obtain a target frequency modulation difference, and setting a first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction according to the target frequency modulation difference;

[0045] Acquire real-time characteristic operation data of the real-time operation condition, and compare it with the historical characteristic operation data of the historical operation condition, and generate a second similarity coefficient according to the comparison result;

[0046] A similarity degree is generated based on the first similarity coefficient and the second similarity coefficient.

[0047] In some embodiments of the present application, the predicted optimal control parameters and the predicted control sub-strategy of the real-time frequency modulation instruction are determined according to the similarity, including:

[0048] sorting the multiple similarities, and setting the optimal control parameter corresponding to the first-ranked similarity as the predicted optimal control parameter of the current real-time frequency modulation instruction;

[0049] According to the historical operating conditions of the energy storage units corresponding to the optimal control parameters in the historical frequency regulation log, an operating condition-control parameter mapping table is constructed. Based on the operating condition-control parameter mapping table, the predicted optimal control parameters are allocated to the energy storage units matching the real-time operating conditions. According to the predicted optimal control parameters corresponding to each energy storage unit, multiple predictive control sub-strategies are generated;

[0050] Generate control commands based on the time sequence of predictive control sub-strategies.

[0051] Compared with the prior art, the intelligent control method of a molten salt energy storage system based on machine learning in the embodiment of the present application has the following beneficial effects:

[0052] The historical control parameters of different historical frequency modulation instructions are determined through historical frequency modulation logs, and the application evaluation values ​​of the historical control parameters are calculated. The preferred control parameter set of different historical frequency modulation instructions is determined according to the machine learning model. The preferred control parameter set is analyzed by minimizing the usage cost objective function and maximizing the frequency modulation efficiency objective function to determine the optimal control parameters. The predicted optimal control parameters and predictive control sub-strategies are determined according to the real-time frequency modulation instructions and real-time operating conditions of the molten salt energy storage system, and control instructions are generated. The molten salt energy storage coefficient is accurately regulated according to the real-time frequency modulation instructions and the predicted optimal control parameters to improve the frequency modulation efficiency of the molten salt energy storage system, thereby improving the safety and stability of thermal power units and power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of an intelligent control method for a molten salt energy storage system based on machine learning in a preferred embodiment of the present application. DETAILED DESCRIPTION

[0054] The specific implementation methods of the present application are further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present application but are not intended to limit the scope of the present application.

[0055] In the description of the present application, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.

[0056] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0057] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0058] like Figure 1 As shown, a preferred embodiment of the present application embodiment is a molten salt energy storage system intelligent control method based on machine learning, including:

[0059] Step S101: obtaining a historical frequency modulation log of the molten salt energy storage system, determining historical control parameters of different historical frequency modulation instructions according to the historical frequency modulation log, and calculating an application evaluation value of the historical control parameters;

[0060] Step S102: Analyze historical control parameters of different historical frequency modulation instructions and corresponding application evaluation values ​​based on a machine learning model to obtain a preferred control parameter set of the historical frequency modulation instructions;

[0061] Step S103: establishing an objective function of minimizing the use cost and an objective function of maximizing the frequency modulation efficiency, and determining the optimal control parameters corresponding to the historical frequency modulation instructions in combination with the preferred control parameter set;

[0062] Step S104: obtaining a real-time frequency modulation instruction and real-time operating conditions of multiple energy storage units in the molten salt energy storage system, and generating a similarity degree according to a first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction and a second similarity coefficient between the real-time operating condition and the historical operating condition;

[0063] Step S105: Determine the predicted optimal control parameters and the predicted control sub-strategy of the real-time frequency modulation instruction according to the similarity, and generate the control instruction of the molten salt energy storage system.

[0064] In some embodiments of the present application, determining historical control parameters of different historical frequency modulation instructions according to the historical frequency modulation log includes:

[0065] Obtaining a corresponding historical frequency modulation instruction according to the historical frequency modulation log, wherein the historical frequency modulation instruction includes a target frequency modulation value and a corresponding historical control duration;

[0066] A time reference line is established based on the historical modulation duration of each historical frequency modulation log, and multiple historical collection nodes are set according to preset time intervals;

[0067] Based on multiple historical collection nodes of each time reference line, historical actual frequency modulation values ​​and historical operation data of molten salt energy storage coefficient are collected and mapped to corresponding historical collection nodes;

[0068] According to the degree of synchronization change between the historical actual frequency modulation value and the historical operation data at each historical collection node in the same time reference line, if the degree of synchronization change is greater than the preset synchronization change degree threshold, the corresponding historical operation data is determined as the historical control parameter of the historical frequency modulation instruction corresponding to the current time reference line.

[0069] In this embodiment, the historical frequency modulation instruction includes a target frequency modulation value. The historical control time required to reach the target frequency modulation value and multiple historical actual frequency modulation values ​​within the historical control time are determined according to the historical frequency modulation log. A time reference line is constructed according to the historical control time, and the historical actual frequency modulation value at the historical collection node and the historical operation data of the molten salt energy storage system are obtained. The degree of synchronous change refers to the degree of change of the historical actual frequency modulation value at the adjacent historical collection node and the degree of change of multiple historical operation data. If both are greater than the preset degree of change threshold, that is, the degree of synchronous change is greater than the preset degree of synchronous change threshold, then it is set as the historical control parameter corresponding to the historical frequency modulation instruction.

[0070] In this embodiment, according to the historical actual frequency modulation values ​​and the degree of synchronous change of historical operating data within the historical control period, historical control parameters that are strongly associated with each historical frequency modulation instruction are screened out, thereby reducing the subsequent data processing volume and improving the analysis efficiency. This lays a data foundation for the subsequent determination of the preferred control data set for each historical frequency modulation instruction, improves the frequency modulation efficiency of the molten salt energy storage system, and thereby improves the safety and stability of thermal power units and power systems.

[0071] In some embodiments of the present application, calculating the application evaluation value of the historical control parameter includes:

[0072] Compare the historical control parameters with the preset standard control parameter interval of the corresponding historical frequency modulation instructions to obtain the number of historical control parameters that are not in the preset standard control parameter interval, the control parameter difference amount that the historical control parameters are not in the preset standard control parameter interval, and the comprehensive duration that the historical control parameters are not in the preset standard control parameter interval;

[0073] Generate a first application evaluation value according to the number, the control parameter difference and the comprehensive duration;

[0074] Determine a final frequency modulation value of a corresponding historical control parameter according to the historical frequency modulation log, and generate a second application evaluation value according to an absolute value of a difference between the final frequency modulation value and a target frequency modulation value;

[0075] Calculate the variance of historical control parameters within the corresponding historical regulation period, and set the compensation coefficient according to the variance;

[0076] An application evaluation value corresponding to the historical control parameter is generated according to the first application evaluation value, the second application evaluation value and the compensation coefficient of the same historical frequency modulation log.

[0077] In some embodiments of the present application, the calculation formula of the application evaluation value is:

[0078]

[0079] Wherein, W is the application evaluation value, a1 is the weight coefficient of the first application evaluation value, y1 is the first application evaluation conversion coefficient, m is the number of historical control parameters that are not in the preset standard control parameter range, n is the historical control parameter of the same historical frequency modulation log, t is the comprehensive duration of the historical control parameter not in the preset standard control parameter range, △Ps is the control parameter difference amount that the historical control parameter is not in the preset standard control parameter range, is the first application evaluation value, |△C| is the absolute value of the difference between the final frequency modulation value and the target frequency modulation value, y2 is the second application evaluation conversion coefficient, a2 is the weight coefficient of the second application evaluation value, and c is the compensation coefficient.

[0080] In this embodiment, the variance refers to the fluctuation of the historical control parameters of the same historical frequency modulation log. The larger the variance, that is, the larger the fluctuation of this set of historical control parameters and the more unstable it is, the smaller the compensation coefficient is. The smaller the variance, that is, the smaller the fluctuation of this set of historical control parameters and the more stable it is, the larger the compensation coefficient is. The value range of c is (0.75, 1.25).

[0081] In this embodiment, the first application evaluation value is determined by comparing the historical control parameters with the preset standard control parameter interval, the second application evaluation value is determined according to the absolute value of the difference between the target frequency modulation value and the final frequency modulation value, and the compensation coefficient is determined according to the variance of the historical control parameters. By comprehensively analyzing the operation evaluation of the historical control parameters, the result evaluation of the frequency modulation value, and the fluctuation evaluation of the historical control parameters, the application evaluation value of the historical control parameters is determined, which lays a data foundation for the subsequent screening of the preferred control data set, reduces the data processing volume, and ensures the accuracy of the preferred control data of each historical control indicator, thereby improving the frequency modulation efficiency and control effect of the molten salt energy storage coefficient when facing the frequency modulation instruction.

[0082] In some embodiments of the present application, historical control parameters of different historical frequency modulation instructions and corresponding application evaluation values ​​are analyzed based on a machine learning model, including:

[0083] Based on the machine learning model, historical control parameters of different frequency modulation instructions are divided to obtain historical control parameter groups of multiple historical frequency modulation logs of the same historical frequency modulation instruction;

[0084] Filter out historical frequency modulation logs whose historical control duration is less than a preset control duration threshold from multiple historical frequency modulation logs of the same historical frequency modulation instruction, and set the historical control parameter group corresponding to the historical frequency modulation log as the first subset of the current historical frequency modulation instruction, and sort the historical control parameter groups in the first subset according to the length of the historical control duration to obtain a first sorting result;

[0085] The historical control parameter groups for which the application evaluation value of the same historical frequency modulation instruction is greater than a preset application evaluation value threshold are set as a second subset, and the historical control parameter groups in the second subset are sorted according to the size of the application evaluation value to obtain a second sorting result.

[0086] In this embodiment, historical control parameters are divided according to historical frequency modulation instructions and historical frequency modulation logs based on a machine learning model to obtain historical control parameter groups of different historical frequency modulation logs for the same historical frequency modulation instruction. The historical control parameter group refers to all historical control parameters included in the historical frequency modulation log.

[0087] In this embodiment, the historical control parameter group is analyzed through a machine learning model to obtain the first subset and the second subset, as well as the sorting results of the first subset and the second subset, thereby laying a foundation for the subsequent determination of the preferred control parameter set for each historical frequency modulation instruction and ensuring the accuracy of the preferred control parameter set of the historical frequency modulation instructions.

[0088] In some embodiments of the present application, obtaining a preferred control parameter set of a historical frequency modulation instruction includes:

[0089] Compare the first subset and the first sorting result with the second subset and the second sorting result to obtain a first sorting value in the first subset and a second sorting value in the second subset for the same historical control parameter group;

[0090] Subtracting the first ranking value from the second ranking value to obtain an absolute value of the ranking difference, and eliminating the historical control parameter group whose absolute value of the ranking difference is greater than a preset absolute value of the ranking difference;

[0091] A preferred control parameter set corresponding to the historical frequency modulation instructions is constructed according to the remaining historical control parameter groups, multiple historical control parameter groups in the preferred control parameter set are set as multiple first centers, application evaluation values ​​corresponding to the multiple historical control parameter groups in the preferred control parameter set are set as multiple second centers, and a first comprehensive ranking value is determined according to the first ranking value and the second ranking value of the historical control parameter group, and the ranking results of the first center and the second center are determined according to the first comprehensive ranking value.

[0092] In this embodiment, corresponding historical control parameters are eliminated according to the absolute value of the sorting difference, thereby excluding historical control parameters with too long regulation time, low application evaluation value or regulation time that does not match the application evaluation value, reducing the data processing volume and ensuring the accuracy of the historical control parameter set of each historical frequency modulation instruction.

[0093] In some embodiments of the present application, a cost minimization objective function and a frequency modulation efficiency maximization objective function are established, and the optimal control parameters corresponding to the historical frequency modulation instructions are determined in combination with the preferred control parameter set, including:

[0094] The objective function of minimizing the usage cost is:

[0095] H1 = min[B(F, m, n)];

[0096] Among them, H1 is the objective function of minimizing the use cost, B(F, m, n) is the use cost from the mth first center to the corresponding nth second center in the Fth historical control instruction, and min is the minimum value symbol;

[0097] The objective function of maximizing frequency modulation efficiency is:

[0098] H2 = max[X(F, m, n)];

[0099] Wherein, H2 is the objective function of maximizing frequency modulation efficiency, X(F, m, n) is the frequency modulation efficiency from the mth first center to the corresponding nth second center in the Fth historical control instruction, and max is the maximum value symbol;

[0100] According to the objective function of minimizing the use cost and the objective function of maximizing the frequency modulation efficiency, the preferred control parameter set of each historical frequency modulation instruction is analyzed to obtain the use cost and frequency modulation efficiency of multiple historical control parameter groups in the preferred control parameter set, and the second comprehensive ranking value of the corresponding historical control parameter group is generated according to the use cost and the frequency modulation efficiency, and the first historical control parameter group selected by the objective function of minimizing the use cost and the second historical control parameter group selected by the objective function of maximizing the frequency modulation efficiency are obtained;

[0101] Compare the first historical control parameter group with the second historical control parameter group, and if they are the same historical control parameter group and the first comprehensive ranking value is greater than the preset ranking value threshold, set the corresponding historical control parameter group as the optimal control parameter of the corresponding historical frequency modulation instruction;

[0102] If they are not the same historical control parameter group, determine whether the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group and the second historical control parameter group are both greater than the preset ranking value threshold; if the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group or the second historical control parameter group are both greater than the preset ranking value threshold, set the corresponding historical control parameter group as the optimal control parameter of the corresponding historical frequency modulation instruction; if the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group and the second historical control parameter group are not both greater than the preset ranking value threshold, determine the third comprehensive ranking value according to the first comprehensive ranking value and the second comprehensive ranking value, and set the historical control parameter group with the largest third comprehensive ranking value in the preferred control parameter set as the optimal control data of the corresponding historical frequency modulation instruction.

[0103] In this embodiment, the objective function of minimizing the usage cost refers to minimizing the sum of the usage costs of the historical control parameters from the third center in each historical frequency modulation instruction and the corresponding fourth center, and the objective function of maximizing the frequency modulation efficiency refers to maximizing the frequency modulation efficiency of the historical control parameters from the third center in each historical frequency modulation instruction and the corresponding fourth center.

[0104] In this embodiment, under the premise of shortening the regulation time and improving the application effect, by determining the objective function of minimizing the usage cost and maximizing the frequency modulation efficiency, unnecessary usage costs can be effectively reduced and the frequency modulation efficiency can be improved. The optimal control parameters of each frequency modulation instruction are comprehensively determined to ensure the frequency modulation effect and the precise regulation of the molten salt energy storage system.

[0105] In some embodiments of the present application, the similarity degree is generated according to the first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction and the second similarity coefficient between the real-time operating condition and the historical operating condition, including:

[0106] Subtracting the target frequency modulation value of the real-time frequency modulation instruction from the target frequency modulation value of the historical frequency modulation instruction to obtain a target frequency modulation difference, and setting a first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction according to the target frequency modulation difference;

[0107] Acquire real-time characteristic operation data of the real-time operation condition, and compare it with the historical characteristic operation data of the historical operation condition, and generate a second similarity coefficient according to the comparison result;

[0108] A similarity degree is generated based on the first similarity coefficient and the second similarity coefficient.

[0109] In this embodiment, the real-time operating condition refers to the actual energy of each energy storage unit in the molten salt energy storage system when it is in heating, heat storage, heat release, antifreeze or maintenance conditions. The real-time characteristic operation data refers to data that has a strong correlation with the real-time operating condition and can represent the change results of the real-time operating condition. When the target frequency modulation difference is closer to 0, the first similarity coefficient is larger, and when the real-time characteristic operation data is more similar to the historical characteristic operation data, the second similarity coefficient is larger.

[0110] In this embodiment, the historical frequency modulation instructions and historical operating conditions with the greatest similarity are determined based on the real-time operating conditions and the real-time frequency modulation instructions, so as to determine the optimal control data of the real-time operating conditions and the real-time frequency modulation instructions, realize precise control of the molten salt energy storage system, and improve the frequency modulation effect.

[0111] In some embodiments of the present application, the predicted optimal control parameters and the predicted control sub-strategy of the real-time frequency modulation instruction are determined according to the similarity, including:

[0112] sorting the multiple similarities, and setting the optimal control parameter corresponding to the first-ranked similarity as the predicted optimal control parameter of the current real-time frequency modulation instruction;

[0113] According to the historical operating conditions of the energy storage units corresponding to the optimal control parameters in the historical frequency regulation log, an operating condition-control parameter mapping table is constructed. Based on the operating condition-control parameter mapping table, the predicted optimal control parameters are allocated to the energy storage units matching the real-time operating conditions. According to the predicted optimal control parameters corresponding to each energy storage unit, multiple predictive control sub-strategies are generated;

[0114] Generate control commands based on the time sequence of predictive control sub-strategies.

[0115] In this embodiment, the predicted optimal control parameters of the real-time frequency modulation instructions are determined according to the degree of similarity, and the predicted optimal control parameters are allocated to the corresponding energy storage units, so as to obtain the predictive control sub-strategy and generate the control instructions, and the predicted optimal control parameters are allocated to the real-time frequency modulation instructions to improve the frequency modulation efficiency and realize the precise control of the molten salt energy storage system.

[0116] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and substitutions can be made without departing from the technical principles of the present application. These improvements and substitutions should also be regarded as the scope of protection of the present application.

Claims

1. A method for intelligent control of a molten salt energy storage system based on machine learning, characterized in that: include: Obtain a historical frequency modulation log of the molten salt energy storage system, determine historical control parameters of different historical frequency modulation instructions according to the historical frequency modulation log, and calculate an application evaluation value of the historical control parameters; Analyze historical control parameters of different historical frequency modulation instructions and corresponding application evaluation values ​​based on a machine learning model to obtain an optimal control parameter set of historical frequency modulation instructions; Establish an objective function of minimizing the use cost and an objective function of maximizing the frequency modulation efficiency, and determine the optimal control parameters corresponding to the historical frequency modulation instructions in combination with the optimal control parameter set; Acquire a real-time frequency modulation instruction and real-time operating conditions of multiple energy storage units in the molten salt energy storage system, and generate a similarity degree according to a first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction and a second similarity coefficient between the real-time operating condition and the historical operating condition; The predicted optimal control parameters and the predictive control sub-strategy of the real-time frequency modulation command are determined according to the similarity, and the control command of the molten salt energy storage system is generated.

2. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 1, characterized in that: Determining historical control parameters of different historical frequency modulation instructions according to the historical frequency modulation log includes: Obtaining a corresponding historical frequency modulation instruction according to the historical frequency modulation log, wherein the historical frequency modulation instruction includes a target frequency modulation value and a corresponding historical control duration; A time reference line is established based on the historical modulation duration of each historical frequency modulation log, and multiple historical collection nodes are set according to preset time intervals; Based on multiple historical collection nodes of each time reference line, historical actual frequency modulation values ​​and historical operation data of molten salt energy storage coefficient are collected and mapped to corresponding historical collection nodes; According to the degree of synchronization change between the historical actual frequency modulation value and the historical operation data at each historical collection node in the same time reference line, if the degree of synchronization change is greater than the preset synchronization change degree threshold, the corresponding historical operation data is determined as the historical control parameter of the historical frequency modulation instruction corresponding to the current time reference line.

3. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 2, characterized in that: Calculate the application evaluation value of historical control parameters, including: Compare the historical control parameters with the preset standard control parameter interval of the corresponding historical frequency modulation instructions to obtain the number of historical control parameters that are not in the preset standard control parameter interval, the control parameter difference amount that the historical control parameters are not in the preset standard control parameter interval, and the comprehensive duration that the historical control parameters are not in the preset standard control parameter interval; Generate a first application evaluation value according to the number, the control parameter difference and the comprehensive duration; Determine a final frequency modulation value of a corresponding historical control parameter according to the historical frequency modulation log, and generate a second application evaluation value according to an absolute value of a difference between the final frequency modulation value and a target frequency modulation value; Calculate the variance of historical control parameters within the corresponding historical regulation period, and set the compensation coefficient according to the variance; An application evaluation value corresponding to the historical control parameter is generated according to the first application evaluation value, the second application evaluation value and the compensation coefficient of the same historical frequency modulation log.

4. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 3, The calculation formula of the application evaluation value is: in, W is the application evaluation value, a1 is the weight coefficient of the first application evaluation value, y1 is the first application evaluation conversion coefficient, m is the number of historical control parameters that are not in the preset standard control parameter range, n is the historical control parameter of the same historical frequency modulation log, t is the comprehensive duration of the historical control parameter not in the preset standard control parameter range, △Ps is the control parameter difference amount that the historical control parameter is not in the preset standard control parameter range, is the first application evaluation value, |△C| is the absolute value of the difference between the final frequency modulation value and the target frequency modulation value, y2 is the second application evaluation conversion coefficient, a2 is the weight coefficient of the second application evaluation value, and c is the compensation coefficient.

5. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 4, characterized in that: Analyze the historical control parameters of different historical frequency modulation instructions and the corresponding application evaluation values ​​based on the machine learning model, including: Based on the machine learning model, historical control parameters of different frequency modulation instructions are divided to obtain historical control parameter groups of multiple historical frequency modulation logs of the same historical frequency modulation instruction; Filter out historical frequency modulation logs whose historical control duration is less than a preset control duration threshold from multiple historical frequency modulation logs of the same historical frequency modulation instruction, and set the historical control parameter group corresponding to the historical frequency modulation log as the first subset of the current historical frequency modulation instruction, and sort the historical control parameter groups in the first subset according to the length of the historical control duration to obtain a first sorting result; The historical control parameter groups for which the application evaluation value of the same historical frequency modulation instruction is greater than a preset application evaluation value threshold are set as a second subset, and the historical control parameter groups in the second subset are sorted according to the size of the application evaluation value to obtain a second sorting result.

6. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 5, characterized in that: The optimal control parameter set of the historical frequency modulation instruction is obtained, including: Compare the first subset and the first sorting result with the second subset and the second sorting result to obtain a first sorting value in the first subset and a second sorting value in the second subset for the same historical control parameter group; Subtracting the first ranking value from the second ranking value to obtain an absolute value of the ranking difference, and eliminating the historical control parameter group whose absolute value of the ranking difference is greater than a preset absolute value of the ranking difference; A preferred control parameter set corresponding to the historical frequency modulation instructions is constructed according to the remaining historical control parameter groups, multiple historical control parameter groups in the preferred control parameter set are set as multiple first centers, application evaluation values ​​corresponding to the multiple historical control parameter groups in the preferred control parameter set are set as multiple second centers, and a first comprehensive ranking value is determined according to the first ranking value and the second ranking value of the historical control parameter group, and the ranking results of the first center and the second center are determined according to the first comprehensive ranking value.

7. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 6, characterized in that: Establish an objective function of minimizing the cost of use and maximizing the frequency modulation efficiency, and determine the optimal control parameters corresponding to the historical frequency modulation instructions in combination with the optimal control parameter set, including: The objective function of minimizing the usage cost is: H1 = min B(F, m, n)]; Among them, H1 is the objective function of minimizing the use cost, B(F, m, n) is the use cost from the mth first center to the corresponding nth second center in the Fth historical control instruction, and min is the minimum value symbol; The objective function of maximizing frequency modulation efficiency is: H2 = max X(F, m, n)[; Wherein, H2 is the objective function of maximizing frequency modulation efficiency, X(F, m, n) is the frequency modulation efficiency from the mth first center to the corresponding nth second center in the Fth historical control instruction, and max is the maximum value symbol; According to the objective function of minimizing the use cost and the objective function of maximizing the frequency modulation efficiency, the preferred control parameter set of each historical frequency modulation instruction is analyzed to obtain the use cost and frequency modulation efficiency of multiple historical control parameter groups in the preferred control parameter set, and the second comprehensive ranking value of the corresponding historical control parameter group is generated according to the use cost and the frequency modulation efficiency, and the first historical control parameter group selected by the objective function of minimizing the use cost and the second historical control parameter group selected by the objective function of maximizing the frequency modulation efficiency are obtained; Compare the first historical control parameter group with the second historical control parameter group, and if they are the same historical control parameter group and the first comprehensive ranking value is greater than the preset ranking value threshold, set the corresponding historical control parameter group as the optimal control parameter of the corresponding historical frequency modulation instruction; If they are not the same historical control parameter group, determine whether the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group and the second historical control parameter group are both greater than the preset ranking value threshold; if the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group or the second historical control parameter group are both greater than the preset ranking value threshold, set the corresponding historical control parameter group as the optimal control parameter of the corresponding historical frequency modulation instruction; if the first comprehensive ranking value and the second comprehensive ranking value of the first historical control parameter group and the second historical control parameter group are not both greater than the preset ranking value threshold, determine the third comprehensive ranking value according to the first comprehensive ranking value and the second comprehensive ranking value, and set the historical control parameter group with the largest third comprehensive ranking value in the preferred control parameter set as the optimal control data of the corresponding historical frequency modulation instruction.

8. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 7, characterized in that: Generate a similarity degree according to a first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction and a second similarity coefficient between the real-time operating condition and the historical operating condition, including: Subtracting the target frequency modulation value of the real-time frequency modulation instruction from the target frequency modulation value of the historical frequency modulation instruction to obtain a target frequency modulation difference, and setting a first similarity coefficient between the real-time frequency modulation instruction and the historical frequency modulation instruction according to the target frequency modulation difference; Acquire real-time characteristic operation data of the real-time operation condition, and compare it with the historical characteristic operation data of the historical operation condition, and generate a second similarity coefficient according to the comparison result; A similarity degree is generated based on the first similarity coefficient and the second similarity coefficient.

9. The intelligent control method for a molten salt energy storage system based on machine learning according to claim 8, characterized in that: The predicted optimal control parameters and the predicted control sub-strategy of the real-time frequency modulation instruction are determined according to the similarity, including: sorting the multiple similarities, and setting the optimal control parameter corresponding to the first-ranked similarity as the predicted optimal control parameter of the current real-time frequency modulation instruction; According to the historical operating conditions of the energy storage units corresponding to the optimal control parameters in the historical frequency regulation log, an operating condition-control parameter mapping table is constructed. Based on the operating condition-control parameter mapping table, the predicted optimal control parameters are allocated to the energy storage units matching the real-time operating conditions. According to the predicted optimal control parameters corresponding to each energy storage unit, multiple predictive control sub-strategies are generated; Generate control commands based on the time sequence of predictive control sub-strategies.

Images