A multi-objective optimization configuration method for improving source-load carrying capacity of a transformer area energy storage

Abstract

The application provides a transformer area energy storage multi-objective optimization configuration method for improving source-load carrying capacity, and belongs to the technical field of transformer area energy storage, and comprises the following steps: firstly, continuously collecting operation data of loads, distribution network input, photovoltaic power generation, wind power generation and energy storage of a transformer area, and establishing a transformer area operation response equation set; obtaining a group of input and output data required for optimization through fitting and data expansion; then, constructing a multi-dimensional matrix from the data, and extracting main features by using tensor decomposition technology; based on the features, a multi-objective optimization model is constructed, a GPU accelerated multi-objective optimization algorithm is used for solving, and a non-inferior solution set is obtained; finally, the weighted score of each non-inferior solution is calculated, and the optimal solution corresponding to the energy storage configuration is selected as the output result. The process can fully utilize the actual operation data of the transformer area, construct a highly practical optimization model, find the optimal configuration of the transformer area energy storage, and solve the problem that the existing technology does not comprehensively consider the configuration optimization of the transformer area energy storage.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of transformer area energy storage, and in particular, relates to a transformer area energy storage multi-objective optimization configuration method for improving source-load bearing capacity. BACKGROUND

[0002] With the transformation of power systems to distributed and low-carbon, the access of distributed power sources such as photovoltaic and wind power to renewable energy is increasing, which brings many challenges to the power grid. On the one hand, the intermittency and volatility of renewable energy make the operation of the power grid face great uncertainty, affecting the safety and stability of the power grid and the reliability of power supply. On the other hand, large-scale renewable energy access will lead to power balance and voltage regulation problems in the power grid, and even excessive investment. Therefore, how to improve the flexibility and robustness of the distributed power grid has become a major problem in current power system planning and operation.

[0003] As an important means of regulating power balance and improving power quality, the electric energy storage system plays an increasingly important role in the distributed power grid. By reasonably configuring and optimizing the control of the electric energy storage system, the fluctuation of renewable energy output can be effectively absorbed, the reliability of power supply of the power grid can be improved, and the investment cost of the power grid can be reduced. However, since the configuration of the electric energy storage system is a complex optimization problem involving multiple objectives and multiple constraints, the existing configuration optimization methods for transformer area energy storage are mostly based on single objective optimization, such as minimizing cost or maximizing economic benefit, ignoring factors such as power grid efficiency, power supply reliability, and environmental impact, which leads to incomplete optimization results and cannot improve the overall performance of the power grid. SUMMARY

[0004] Therefore, the application provides a transformer area energy storage multi-objective optimization configuration method for improving source-load bearing capacity, which can solve the technical problem that the existing configuration optimization methods for transformer area energy storage are mostly based on single objective optimization and are not comprehensive enough.

[0005] The application is implemented as follows:

[0006] The application provides a transformer area energy storage multi-objective optimization configuration method for improving source-load bearing capacity, which includes the following steps:

[0007] S10, continuously collecting operation data of the transformer area, including load data, distribution network input data, photovoltaic power generation data, wind power generation data, and energy storage data;

[0008] S20, establishing a transformer area operation response equation set according to the operation data, including a power balance equation, a voltage balance equation, a power loss equation, an energy storage charge and discharge equation, a photovoltaic power generation equation, a wind power generation equation, a load response equation, and a power grid support equation;

[0009] S30, fit the operating data of the continuously collected transformer area to the transformer area operating response equation set to obtain a first equation set; and input the operating data into the first equation set to obtain a result set;

[0010] S40, expand the collected operating data in a data expansion manner to obtain expanded operating data, and input the expanded operating data into the first equation set to obtain an expanded result set; merge the operating data and the expanded operating data into an operating data set, and merge the result set and the expanded result set into a result data set;

[0011] S50, establish a multi-dimensional matrix according to the operating data set and the result data set; divide the multi-dimensional matrix into an input matrix X and an output matrix Y, wherein X contains all input parameters, and Y contains all output parameters related to optimization targets; and perform standardization processing on X and Y to eliminate the influence of different dimensions;

[0012] S60, use a tensor decomposition technique to reduce the dimension and extract features of the multi-dimensional matrix to obtain a simplified feature matrix F;

[0013] S70, construct a multi-objective optimization model based on the feature matrix F;

[0014] S80, solve the multi-objective optimization model to obtain a non-inferior solution set by using a GPU-accelerated multi-objective optimization algorithm;

[0015] S90, calculate the weighted score of each non-inferior solution in the non-inferior solution set, select the non-inferior solution with the highest weighted score as the optimal solution, and output the energy storage data corresponding to the optimal solution as the optimization configuration data of the transformer area energy storage.

[0016] Further, the power balance equation is specifically represented as follows:

[0017] ;

[0018] In the formula, is the total load power (kW); is the system loss power (kW); is the grid input power (kW); is the photovoltaic power (kW); is the wind power (kW); is the energy storage system power (kW, negative for charging, positive for discharging); , , , is the coefficient to be fitted; is the first error term.

[0019] Parameter acquisition method:

[0020] 、 、 、 and obtained by real-time measurement. obtained by calculation of power loss equation. 、 、 、 obtained by fitting method such as least square method.

[0021] The voltage balance equation is specifically represented as follows:

[0022] ;

[0023] In the formula, is the voltage (V) of node i; is the voltage (V) of power supply side; is the number of line sections from node i to power supply; is the resistance (Ω) of the kth line section; is the reactance (Ω) of the kth line section; is the current (A) of the kth line section; is the imaginary unit; 、 、 is the coefficient to be fitted; is the second error term.

[0024] Parameter acquisition method:

[0025] and obtained by voltage sensor measurement; and calculated according to line parameters; obtained by current sensor measurement. 、 、 obtained by fitting method such as complex least square method.

[0026] The power loss equation is specifically represented as follows:

[0027] ;

[0028] In the formula, is the total power loss (kW); is the number of line sections in the system; is the current (A) of the kth line section; is the resistance (Ω) of the kth line section; is the core loss coefficient; V is the voltage (V) of the kth line section; P is the switching loss coefficient; f is the switching frequency (Hz); S is the apparent power (kVA) of the kth line section; , , is the coefficient to be fitted; is the third error term.

[0029] Parameter acquisition method:

[0030] and obtained by real-time measurement; calculated according to the line parameters; and determined by equipment specifications and experiments; obtained according to system settings; calculated by , where and are the active power and the reactive power, respectively, obtained by measurement. , , obtained by fitting methods such as least squares.

[0031] The energy storage charging and discharging equation is specifically represented as follows:

[0032] ;

[0033] In the formula, is the state of charge at time t; is the state of charge at time t+1; is the charging efficiency; is the discharging efficiency; is the charging power at time t (kW); is the discharging power at time t (kW); is the time interval (h); is the rated capacity of the energy storage system (kWh); , , is the coefficient to be fitted; is the fourth error term.

[0034] Parameter acquisition method:

[0035] obtained by the battery management system (BMS) of the energy storage system; and determined according to the specifications of the energy storage system; and measured in real time by power measurement equipment; determined according to the data acquisition frequency; for the design parameters of the energy storage system. , , obtained by fitting methods such as least squares.

[0036] The photovoltaic power generation equation is specifically represented as follows:

[0037] ;

[0038] wherein, is the photovoltaic power generation power (kW); is the photovoltaic panel efficiency; is the photovoltaic panel area (m²); is the solar irradiance (kW / m²); is the temperature coefficient (usually -0.004 to -0.005 / °C); is the photovoltaic cell temperature (°C); is the reference temperature (usually 25°C); , , is the coefficient to be fitted; is the fifth error term.

[0039] Parameter acquisition method:

[0040] and is the photovoltaic system design parameter; obtained by irradiance sensor measurement; determined according to the photovoltaic panel specifications; obtained by temperature sensor measurement; is the fixed value 25°C. , , obtained by fitting methods such as nonlinear least squares.

[0041] The wind power generation equation is specifically represented as follows:

[0042] ;

[0043] wherein, is the wind power generation power (kW); is the air density (kg / m³); is the wind wheel swept area (m²); is the wind energy utilization coefficient; is the wind speed (m / s); is the cut-in wind speed (m / s); Cut-off wind speed (m / s); Rated wind speed (m / s); Rated power (kW); , The coefficients to be fitted; This is the sixth error term.

[0044] Parameter acquisition method:

[0045] Determined based on local weather conditions; and Design parameters for wind turbine generators; Obtained by wind speed sensor measurement; , , and These are the rated parameters of the wind turbine. , It is obtained through fitting methods such as piecewise nonlinear least squares.

[0046] The load response equation is specifically expressed as follows:

[0047] ;

[0048] ;

[0049] In the formula, The actual load power (kW) at time t; Let t be the baseline load power (kW) at time t; The demand response regulation power (kW) at time t; This is the demand response sensitivity coefficient; Let t be the electricity price (yuan / kWh). Reference electricity price (RMB / kWh); , , The coefficients to be fitted; This is the seventh error term.

[0050] Parameter acquisition method:

[0051] Obtained through historical load data and load forecasting models; This was determined through statistical analysis and user behavior research; Obtained based on real-time electricity price information; It is a fixed reference electricity price. , , It is obtained through fitting methods such as nonlinear least squares.

[0052] The grid support equation is specifically represented as follows:

[0053] ;

[0054] In the formula, is the grid support capacity (kVA); is the maximum active power that the grid can provide (kW); is the actual active power input by the grid (kW); is the maximum reactive power that the grid can provide (kvar); is the actual reactive power input by the grid (kvar); , is the coefficient to be fitted; is the eighth error term.

[0055] Parameter acquisition method:

[0056] and determined according to transformer capacity and grid operation restrictions; and obtained through real-time measurement by power measurement equipment at the grid input point. 、 obtained through fitting methods such as least squares.

[0057] In the above equation set, all the coefficients to be fitted (such as 、 、 、 , etc.) and error terms ( 、 , etc.) need to be fitted and estimated through a large amount of historical data. The fitting process can use methods such as least squares, nonlinear least squares, maximum likelihood estimation, etc., and the specific choice depends on the characteristics of the equation and the distribution of the data. The error term is assumed to follow a normal distribution with a mean of 0, and the variance needs to be estimated through data.

[0058] In step S50, the multi-dimensional matrix standardization process is specifically represented as follows:

[0059] ;

[0060] ;

[0061] In the formula, and are the input matrix and the output matrix, respectively; and are the standardized input matrix and the output matrix, respectively; and The mean is and The standard deviation is and The mean is and The standard deviation is

[0062] The formula for calculating the mean and standard deviation is as follows:

[0063] ;

[0064] ;

[0065] where, is the number of samples, is the value of the th sample. The mean and standard deviation are calculated in the same way.

[0066] wherein the step of using tensor decomposition technology to reduce dimensionality and extract features of the multi-dimensional matrix to obtain a simplified feature matrix F is specifically: adopting a Tucker decomposition method to decompose the multi-dimensional matrix into a core tensor and a factor matrix; using the element size of the core tensor to identify important features and correlations; based on the decomposition result, constructing a simplified feature matrix F to retain the most important feature information.

[0067] wherein the Tucker decomposition is specifically represented as follows:

[0068] ;

[0069] wherein, is the original multi-dimensional matrix; is the core tensor; , and are factor matrices; represents an n-mode product.

[0070] The size of the core tensor is , wherein , , are the sizes of each dimension after dimensionality reduction. The sizes of the factor matrices , , are , , , respectively, wherein , , are the sizes of the three dimensions of the original matrix .

[0071] Wherein, the step of constructing a multi-objective optimization model based on the feature matrix F is specifically: taking the energy storage data as a decision variable vector, constructing a target function matrix, including: a system energy conversion efficiency maximization target function, an energy loss minimization target function, a maximum energy storage support for the power grid target function, a power supply reliability maximization target function and a total cost minimization target function; setting a constraint condition matrix, including energy storage capacity and power limit, voltage and frequency limit, load balance constraint, energy storage charge and discharge depth and frequency limit, renewable energy output limit and investment budget limit. Specifically, the target function matrix construction is specifically represented as follows:

[0072] ;

[0073] In the formula, is the target function matrix; is the decision variable vector; is the system energy conversion efficiency maximization target function; is the energy loss minimization target function; is the maximum energy storage support for the power grid target function; is the power supply reliability maximization target function; is the total cost minimization target function.

[0074] The specific expressions of each target function are as follows:

[0075] 1. System energy conversion efficiency maximization:

[0076] ;

[0077] 2. Energy loss minimization:

[0078] ;

[0079] 3. Maximum energy storage support for the power grid:

[0080] ;

[0081] 4. Power supply reliability maximization:

[0082] ;

[0083] 5. Total cost minimization:

[0084] ;

[0085] Wherein, is the total time step; is the time step; and Let be the output power and input power at time t, respectively; Let be the power loss at time t; Let be the grid support capacity at time t; and These represent the power supplied and the power demanded at time t, respectively. For investment costs; Let t be the operating cost at time t.

[0086] GPU-accelerated non-dominated sorting algorithms include the following steps:

[0087] 1. Dominance Relationship Matrix Calculation:

[0088] ;

[0089] In the formula, Elements of the dominance matrix; and The first The and the first One solution; For the first One objective function.

[0090] 2. Domination Count Calculation:

[0091] ;

[0092] In the formula, For the first The number of dominated solutions; The total number of solutions.

[0093] 3. Pareto front identified:

[0094] ;

[0095] In the formula, This is the first-level Pareto front, containing all solutions with a dominated count of 0.

[0096] 4. Crowding Distance Calculation:

[0097] ;

[0098] In the formula, For the first The crowding distance of each solution; The number of objective functions; and They were respectively in the second The function values ​​of two adjacent solutions on each objective; and The first max and min of the objective functions.

[0099] S90. Non-inferior solution comprehensive score calculation:

[0100] The non-inferior solution comprehensive score calculation is specifically represented as follows:

[0101] ;

[0102] wherein, is the comprehensive score of the i-th non-inferior solution; is the number of objective functions; is the weight of the j-th objective function, satisfying ; is the value of the i-th non-inferior solution on the j-th objective function; and are the minimum and maximum values of the j-th objective function among all non-inferior solutions, respectively. The determination of the weights can use methods such as analytic hierarchy process (AHP) or entropy weight method, and is calculated according to the preferences of the decision maker or the importance of the objective functions. Selection of optimal solution:

[0103] wherein,

[0104] is the optimal solution, i.e., the non-inferior solution with the highest comprehensive score.

[0105] ;

[0106] wherein, is the optimal solution, i.e., the non-inferior solution with the highest comprehensive score.

[0107] Compared with the prior art, the power supply and load carrying capacity improved substation energy storage multi-objective optimization configuration method provided by the present application has the beneficial effects that:

[0108] 1. A detailed response model covering all aspects of the substation is established. The present application establishes a comprehensive response equation set including power balance, voltage characteristics, loss analysis, energy storage behavior, renewable energy output, load response, and power grid support, etc. by continuously collecting the operation data of the load, distribution network, renewable energy, and energy storage system of the substation. This systematic modeling method can more accurately describe the actual operation characteristics of the substation, and provides a reliable basis for subsequent optimization decisions.

[0109] ​​​2. The generalization ability of the model is improved by using data augmentation technology. Based on the original operation data, the present application generates more abundant simulation operation data sets by using random interpolation, rolling window, time series model and other methods. This not only expands the coverage of the training data and enhances the adaptability of the model to future uncertainty, but also provides more extensive data support for subsequent optimization modeling.

[0110] 3. Advanced tensor decomposition technology is applied to extract key features. The present application uses Tucker decomposition method to reduce dimension and extract features of multi-dimensional operation data matrix, and obtains a simplified feature matrix. This feature extraction technology can effectively identify the most important information in the data, and provides high-quality input for subsequent multi-objective optimization modeling.

[0111] 4. A comprehensive optimization model considering multiple factors is constructed. Based on feature extraction, the present application constructs a multi-objective optimization model including five objective functions of system energy conversion efficiency, energy loss, power grid support, power supply reliability and total cost. This comprehensive optimization target can better reflect the performance indicators of power grid operation, and provide a more balanced and comprehensive basis for the final optimal energy storage configuration.

[0112] 5. GPU-accelerated multi-objective optimization algorithm is used for solving. The present application uses the advantages of GPU parallel computing, and adopts efficient multi-objective optimization algorithms such as non-dominated sorting, which greatly improves the solving efficiency. This algorithm can find the non-inferior solution set of the energy storage system in a limited time, providing multiple alternative schemes for decision makers to meet the time requirements in practical applications.

[0113] In summary, the multi-objective optimization configuration method of the present application for district energy storage can fully utilize real-time monitoring data for comprehensive modeling, and use advanced feature extraction and optimization algorithms to obtain high-quality configuration decisions, which not only improves the performance of power grid operation, but also provides effective support for the flexibility and reliability of distributed power grid. The present application solves the technical problem that the existing configuration optimization method for district energy storage is mostly based on single objective optimization and does not consider comprehensively. BRIEF DESCRIPTION OF DRAWINGS

[0114] Figure 1 The flowchart of the method provided by the present application is shown. DETAILED DESCRIPTION

[0115] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application.

[0116] As Figure 1The diagram shown is a flowchart of a multi-objective optimization configuration method for transformer substation energy storage to improve source load carrying capacity, provided by the present invention. The specific implementation methods for each step of this method are described in detail below:

[0117] The specific implementation of step S10 is: continuously collecting the operating data of the transformer area. This includes the following sub-steps:

[0118] First, collect load data, including active power, reactive power, voltage, current, and power factor. This data reflects the electricity consumption characteristics of users in the distribution area and is an important input for subsequent optimization. It is recommended to set the data collection frequency to 1 to 15 minutes to capture dynamic changes in the distribution area's load.

[0119] Secondly, data on the distribution network input is collected, including feeder voltage, current, active power, reactive power, and power factor. This data reflects the power supply quality on the grid input side and can be used to analyze the grid's support capacity. The data collection frequency is consistent with the load data.

[0120] Secondly, photovoltaic power generation data is collected, including output active power, reactive power, voltage, current, irradiance, and temperature. This data reflects the power generation status of the photovoltaic system and can be used to analyze the changing characteristics of renewable energy.

[0121] Simultaneously, wind power generation data is collected, including output active power, reactive power, voltage, current, wind speed, and wind direction. This data reflects the power generation status of the wind power system and can be used to analyze the changing characteristics of renewable energy.

[0122] Finally, data from the energy storage system is collected, including charge / discharge power, state of charge, voltage, current, and temperature. This data reflects the operating status of the energy storage system and serves as direct input for optimizing energy storage configuration.

[0123] In summary, the purpose of step S10 is to continuously collect various important operational data from the distribution area, providing a detailed data foundation for subsequent modeling, optimization, and decision-making. This data covers various aspects such as load, power grid, renewable energy, and energy storage systems, laying the groundwork for building a comprehensive distribution area response model.

[0124] The specific implementation of step S20 is as follows: Based on the collected operational data, establish a set of operational response equations for the transformer area. Specifically, this includes the following equations:

[0125] 1. Power balance equation:

[0126] ;

[0127] This equation describes the balance between the total load and power loss of the transformer substation and the power supplied from the grid, solar power, wind power, and energy storage systems. (Coefficients) to and error terms Need to be obtained by historical data fitting.

[0128] 2. Voltage balance equation:

[0129] ;

[0130] This equation describes the relationship between node voltage and power supply voltage, line resistance, reactance, and current. Coefficients , and Need to be obtained by historical data fitting.

[0131] 3. Power loss equation:

[0132] ;

[0133] This equation describes the total power loss of the system, including line loss, core loss, and switching loss. Coefficients to Need to be obtained by historical data fitting.

[0134] 4. Energy storage charging and discharging equation:

[0135] ;

[0136] This equation describes the charging and discharging process of the energy storage system, and the influence on the battery state of charge SOC. Coefficients to Need to be obtained by historical data fitting.

[0137] 5. Photovoltaic power generation equation:

[0138] ;

[0139] This equation describes the relationship between photovoltaic power generation and irradiance and battery temperature. Coefficients to Need to be obtained by historical data fitting.

[0140] 6. Wind power generation equation:

[0141] ;

[0142] This equation describes the relationship between wind power generation and wind speed, including the piecewise characteristics of cut-in, rated, and cut-out wind speeds. Coefficients and Need to be obtained by historical data fitting.

[0143] 7. Load response equation:

[0144] ;

[0145] ;

[0146] This equation describes the load response characteristics to the price signal, including the base load and the demand response adjustment part. The coefficient to Need to be obtained by historical data fitting.

[0147] 8. Grid support equation:

[0148] ;

[0149] This equation describes the active and reactive power support capacity that the grid can provide. The coefficient and Need to be obtained by historical data fitting.

[0150] In summary, the purpose of step S20 is to establish a response model covering various aspects of the substation, including power balance, voltage characteristics, loss analysis, energy storage behavior, renewable energy output, load response, and grid support. These equation sets constitute a complete substation simulation model, providing a basis for subsequent optimization decisions.

[0151] The specific implementation of step S30 is to use continuously collected substation operation data to fit the established substation response equation set to obtain the first set of equation parameters, and to bring the operation data into the equation set to calculate the result set.

[0152] First, for the eight response equations established in step S20, use least squares method, nonlinear least squares method or maximum likelihood estimation method, combined with actual collected operation data to fit the coefficients of each equation, to obtain the first set of parameters. In this way, a preliminary substation response model is established.

[0153] Next, the collected actual operation data, including load, grid, renewable energy and energy storage system data, are brought into the equation set obtained by fitting to calculate. In this way, a set of result data consistent with the actual operation is obtained, denoted as the result set.

[0154] The purpose of this step is: 1) to obtain the initial parameters of the substation response model through fitting; 2) to verify whether the established model can accurately describe the operation characteristics of the actual substation. The result set provides a benchmark and reference for subsequent data expansion and optimization.

[0155] The specific implementation of step S40 is to expand the collected operation data using data expansion, obtain expanded operation data, and input the equation set obtained by fitting to calculate an expanded result set. Then, the original operation data and the expanded operation data are merged into an operation data set, and the result set and the expanded result set are merged into a result data set.

[0156] Specifically, first, some data expansion techniques, such as random interpolation, rolling window, time series model, etc., are used to expand and enhance the original operation data collected in step S10. In this way, a set of expanded simulation operation data is obtained, which better covers various possible situations.

[0157] Next, these expanded operation data are input into the substation response equation set obtained in step S30 to calculate the corresponding expanded result set.

[0158] Finally, the original operation data and the expanded operation data are merged into a more abundant operation data set, and the result set and the expanded result set are merged into a result data set.

[0159] The purpose of this step is: 1) to expand the coverage of the training data set through data expansion, and to improve the generalization ability of the model; 2) to generate more abundant input-output data pairs, and to lay a solid foundation for subsequent optimization modeling.

[0160] The specific implementation of step S50 is to establish a multi-dimensional matrix according to the operation data set and the result data set, and to standardize the input matrix X and the output matrix Y.

[0161] First, the operation data set and the result data set are organized into a multi-dimensional matrix . Specifically, the rows of the matrix represent different time points, the columns represent different data types and parameters, and the depth dimension represents different scenarios or expanded data sets. In this way, a rich multi-dimensional data structure is constructed, laying a foundation for subsequent feature extraction and optimization modeling.

[0162] Next, the multi-dimensional matrix is divided into an input matrix X and an output matrix Y. X contains all the input parameters, such as measured data of load, power grid, renewable energy and energy storage system; Y contains all the output parameters related to the optimization target, such as energy conversion efficiency, energy loss, power grid support, etc.

[0163] Finally, the input matrix X and the output matrix Y are standardized to eliminate the influence of different dimensions. The standardization formula is:

[0164] ;

[0165] ;

[0166] where, , , and are the mean and standard deviation of X and Y, respectively.

[0167] The purpose of this step is: 1) to build a multi-dimensional data structure to better organize and represent complex operational data; 2) to standardize input and output data, providing a standardized data foundation for subsequent feature extraction and optimization modeling.

[0168] The specific implementation of step S60 is to use the Tucker decomposition method to reduce the dimensionality and extract features of the multi-dimensional matrix, obtaining a simplified feature matrix F.

[0169] Tucker decomposition is a multi-dimensional data decomposition method that can approximately decompose the original multi-dimensional matrix into a core tensor and three factor matrices , , :

[0170] ;

[0171] where, denotes the n-mode product. The size of the core tensor is , reflecting the most important features and correlations in the data. The sizes of the factor matrices , , are , , , respectively, containing the principal components of the data in each dimension.

[0172] Through Tucker decomposition, key features in multi-dimensional data can be effectively extracted, and the data dimension can be greatly compressed. In this problem, the results of Tucker decomposition can be used to construct a simplified feature matrix F, which only contains the most important feature information, providing high-quality input for subsequent optimization modeling.

[0173] The purpose of this step is: 1) to use tensor decomposition techniques such as Tucker decomposition to extract key features from multi-dimensional data; 2) to construct a simplified feature matrix F, which greatly reduces the data dimension while preserving the main information, providing efficient input for optimization modeling.

[0174] The specific implementation of step S70 is to construct a multi-objective optimization model based on the feature matrix F.

[0175] Specifically, the decision variable vector of the energy storage system is denoted as , the objective function matrix is constructed, including the following five objective functions:

[0176]

[0177] 1. System energy conversion efficiency maximization objective function

[0178]

[0179] 2. Energy loss minimization objective function

[0180]

[0181] 3. Energy storage maximum support for the power grid objective function

[0182]

[0183] 4. Power supply reliability maximization objective function

[0184]

[0185] 5. Total cost minimization objective function

[0186]

[0187] At the same time, the corresponding constraint condition matrix is set, including energy storage capacity and power limit, voltage and frequency limit, load balance constraint, energy storage charge and discharge depth and times limit, renewable energy output limit and investment budget limit, etc.

[0188] In summary, the purpose of step S70 is to construct a multi-objective optimization model based on the feature extraction results, considering system efficiency, loss, reliability, cost and other factors, to provide a basis for subsequent optimal energy storage configuration.

[0189] The specific implementation of step S80 is to use a GPU-accelerated multi-objective optimization algorithm to solve the above multi-objective optimization model to obtain a non-inferior solution set.

[0190] Specifically, first, the objective function matrix and the constraint condition matrix are calculated in parallel on the GPU. Then, using the non-dominated sorting algorithm implemented by the GPU, the multi-objective optimization problem is solved through the following steps: ​​​​​​​​​​​

[0191] 1. Calculate the dominance matrix Determine whether each solution is dominated by other solutions.

[0192] 2. Parallel computation of the dominance count for each solution. And the dominated set.

[0193] 3. Determine the Pareto frontier That is, the solution set whose dominated count is 0.

[0194] 4. Calculate the crowding distance for each solution. This reflects the distribution of solutions in the target space.

[0195] 5. Based on non-dominated sorting and crowding distance, perform selection, crossover, and mutation operations for iterative optimization.

[0196] 6. Repeat steps 2-5 until the termination condition is met, and obtain the final non-dominated solution set.

[0197] This algorithm fully utilizes the parallel computing power of GPUs, significantly improving the efficiency of multi-objective optimization. Non-dominated sorting and crowding distance calculation guarantee the convergence and dispersion of the final solution set, satisfying the requirements of multi-objective optimization.

[0198] The purpose of this step is to use GPU acceleration technology to solve complex multi-objective optimization problems, obtain the non-dominated solution set of the energy storage system in the transformer substation, and provide alternative solutions for subsequent optimal selection.

[0199] The specific implementation of step S90 is as follows: select the solution with the highest comprehensive score from the set of non-dominated solutions as the optimal energy storage configuration scheme.

[0200] Specifically, the objective function values ​​for each non-dominated solution are first normalized to unify their numerical range to the [0,1] interval:

[0201] ;

[0202] in, For the first The overall score of each non-dominated solution. For the first The weights of each objective function, For the first The non-inferior solution is at the th . The values ​​on the objective function, and The first The minimum and maximum values ​​of an objective function in all non-dominated solutions.

[0203] Weight The Analytic Hierarchy Process (AHP) or entropy weight method can be used to determine the importance of the decision maker's preferences or objective functions, satisfying .

[0204] Finally, the non-inferior solution with the highest comprehensive score is selected as the optimal solution , and the corresponding energy storage system capacity, power and optimization values of each objective function are output.

[0205] The purpose of this step is: 1) According to the preferences of the decision maker, calculate the comprehensive score of each non-inferior solution; 2) Select the solution with the highest score from the non-inferior solution set as the optimal energy storage configuration scheme, providing a basis for subsequent implementation.

[0206] In this multi-objective optimization configuration method for substation energy storage, the multi-dimensional matrix is a key data structure for organizing and representing complex operational data. The following analyzes this multi-dimensional matrix step by step:

[0207] 1. Matrix dimensions:

[0208] Rows: Represent different time points;

[0209] Columns: Represent different data types and parameters;

[0210] Depth: Represents different scenarios or augmented data sets;

[0211] 2. Specific meaning of depth:

[0212] The depth dimension here represents different scenarios or augmented data sets. This means that the matrix is not just two-dimensional, but contains multiple "layers" or "slices". Each layer represents a specific scenario or a set of augmented data.

[0213] 3. Role of depth:

[0214] Scenario representation: Each layer may represent a different operational scenario, such as a normal weekday, weekend, holiday, or different seasons, etc.

[0215] Data augmentation: Additional data sets generated through data augmentation techniques (such as mentioned in the S40 step), each augmented data set may occupy a depth layer of the matrix.

[0216] Rich information: The depth dimension allows the model to consider more variations and possibilities, improving the robustness and adaptability of the optimization results.

[0217] 4. Example explanation:

[0218] Suppose there is a 3x3x3 multi-dimensional matrix:

[0219] 3 rows represent 3 time points (e.g. morning, noon, evening);

[0220] 3 columns represent 3 data types (e.g. load, photovoltaic, energy storage);

[0221] 3 depth layers represent 3 scenarios (e.g. sunny, cloudy, rainy);

[0222] This matrix can be represented as:

[0223] Depth 1 (sunny): [[load1, photovoltaic1, energy storage1], [load2, photovoltaic2, energy storage2], [load3, photovoltaic3, energy storage3]];

[0224] Depth 2 (cloudy): [[load1', photovoltaic1', energy storage1'], [load2', photovoltaic2', energy storage2'], [load3', photovoltaic3', energy storage3']];

[0225] Depth 3 (rainy): [[load1", photovoltaic1", energy storage1"], [load2", photovoltaic2", energy storage2"], [load3", photovoltaic3", energy storage3]];

[0226] 5. Advantages of multi-dimensional matrix:

[0227] Data integration: integrates a large amount of complex operation data into a unified structure.

[0228] Scenario analysis: can consider multiple operation scenarios and conditions at the same time.

[0229] Optimization basis: provides a rich data basis for subsequent feature extraction and multi-objective optimization.

[0230] In summary, this method of optimizing the carrying capacity of source and load in the distribution area by configuring energy storage provides a systematic and comprehensive solution for the optimization configuration of the distribution area energy storage system by fully utilizing the detailed operation data of the distribution area, establishing a response model covering various aspects, expanding the coverage of the training data set using data augmentation technology, extracting key features using Tucker decomposition, constructing a multi-objective optimization model, and solving the non-inferior solution set using a GPU-accelerated multi-objective optimization algorithm, and finally selecting the optimal energy storage configuration scheme with the highest comprehensive score.

[0231] Specifically, the principle of the present application is:

[0232] Firstly, the method establishes a detailed response model covering all aspects of the substation by continuously collecting various types of operating data, including load, distribution network, real-time monitoring data of renewable energy and energy storage system. These response equations include power balance, voltage characteristics, power loss, energy storage charging and discharging, renewable energy output and grid support, which can accurately describe the actual operating characteristics of the substation. Compared with existing methods that only focus on a single target, this comprehensive modeling approach lays a solid foundation for subsequent multi-objective optimization.

[0233] Secondly, in order to enhance the adaptability of the model to future uncertainties, the invention uses data augmentation technology. By using methods such as random interpolation, rolling window and time series prediction, a more diverse simulated operating data set is generated. This not only expands the coverage of the training data, but also provides more extensive data support for subsequent optimization modeling. Compared with existing methods that rely only on limited historical data, this data augmentation technology can better predict and respond to uncertainties in power grid operation.

[0234] Thirdly, the invention applies advanced Tucker decomposition technology to reduce the dimensionality and extract features of the multi-dimensional operating data matrix. This tensor decomposition method can effectively identify the most important features and correlations in the data, significantly compressing the data dimensionality and providing high-quality input for subsequent multi-objective optimization modeling. Compared with existing methods that simply rely on empirical feature selection, this data-driven feature extraction technology is more objective, systematic and efficient.

[0235] Finally, based on the extracted key features, the invention constructs a comprehensive optimization model considering multiple factors. The model includes five objective functions: system energy conversion efficiency, energy loss, grid support, power supply reliability and total cost, which can more comprehensively reflect various performance indicators of power grid operation. Compared with existing methods that only focus on single target optimization, this multi-objective optimization approach can seek a more balanced solution among the objectives, providing a more scientific decision basis for the final optimal energy storage configuration.

[0236] In order to efficiently solve this complex multi-objective optimization problem, the invention uses a non-dominated sorting algorithm based on GPU acceleration. This algorithm fully utilizes the parallel computing capability of GPU, greatly improving the solving efficiency. By calculating the dominance relationship and crowding distance of each solution, the algorithm can find the non-inferior solution set of the energy storage system in a limited time, providing multiple alternative solutions for decision makers. This GPU-accelerated optimization algorithm meets the demand for rapid decision-making in practical applications.

[0237] In summary, the transformer area energy storage multi-objective optimization configuration method provided in the application solves the problems existing in the existing electric energy storage configuration method systematically on the basis of a plurality of key technologies such as comprehensive modeling, data expansion, feature extraction, multi-objective optimization and GPU accelerated optimization algorithm.

[0238] In order to better understand and implement the application, one specific embodiment 1 of the application is provided below, and the specific implementation of each step in the embodiment 1 is described in detail as follows: the specific implementation of step S10 is as follows:

[0239] The purpose of this step is to continuously collect various operation data of the transformer area, to provide detailed data basis for subsequent modeling and optimization. The collected data includes load data, distribution network input data, photovoltaic power generation data, wind power generation data and energy storage data.

[0240] Firstly, for the load data, the following parameters are collected:

[0241] Active power : reflects the electricity demand of the transformer area user; reactive power : reflects the reactive power consumption of the transformer area user; voltage : reflects the voltage level of the transformer area user side; current : reflects the current change of the transformer area user side; power factor : reflects the power factor of the transformer area user side.

[0242] These load parameters can comprehensively describe the electricity consumption characteristics of the transformer area user. The data collection frequency is recommended to be set to minutes, so as to capture the dynamic change of the load.

[0243] Secondly, for the distribution network input data, the following parameters are collected: feeder voltage : reflects the voltage level of the power grid input side; feeder current : reflects the current change of the power grid input side; active power : reflects the active power of the power grid input side; reactive power : reflects the reactive power of the power grid input side; power factor : reflects the power factor of the power grid input side.

[0244] These distribution network input parameters can reflect the power supply quality and support capacity of the power grid side. The data collection frequency is also set to minutes.

[0245] Thirdly, for the photovoltaic power generation data, the following parameters are collected:

[0246] Active power : reflects the output active power of the photovoltaic system;

[0247] reactive power : reflects the output reactive power of the photovoltaic system;

[0248] voltage : reflects the output voltage of the photovoltaic system;

[0249] current : reflects the output current of the photovoltaic system;

[0250] irradiance : reflects the lighting conditions;

[0251] temperature : reflects the operating temperature of the photovoltaic cell.

[0252] These photovoltaic power generation parameters can comprehensively reflect the power generation status of the photovoltaic system and provide a basis for analyzing the change characteristics of renewable energy.

[0253] At the same time, wind power generation data is also collected, including:

[0254] active power : reflects the output active power of the wind power generation system;

[0255] reactive power : reflects the output reactive power of the wind power generation system;

[0256] voltage : reflects the output voltage of the wind power generation system;

[0257] current : reflects the output current of the wind power generation system;

[0258] wind speed : reflects the wind conditions;

[0259] wind direction : reflects the changes in wind direction.

[0260] These wind power generation parameters can comprehensively reflect the power generation status of the wind system and provide a basis for analyzing the change characteristics of renewable energy.

[0261] Finally, for energy storage system data, the following parameters are collected:

[0262] charge and discharge power : reflects the charge and discharge status of the energy storage system, positive value indicating discharge, negative value indicating charge;

[0263] state of charge : reflects the battery charge level of the energy storage system;

[0264] voltage This reflects the output voltage of the energy storage system;

[0265] Current This reflects the output current of the energy storage system;

[0266] temperature This reflects the operating temperature of the energy storage system.

[0267] These energy storage data can comprehensively reflect the operating status of the energy storage system, providing key input for subsequent optimization and configuration.

[0268] In summary, this step, through the continuous collection of the aforementioned operational data, provides a detailed data foundation for subsequent modeling, optimization, and decision-making, laying the groundwork for the efficient configuration of the energy storage system in the distribution area.

[0269] The specific implementation method of step S20 is as follows:

[0270] The purpose of this step is to establish a set of response equations covering various aspects of the transformer substation based on the collected operational data, providing a foundational model for subsequent optimization modeling. The established set of equations includes:

[0271] 1. Power balance equation:

[0272] ;

[0273] in, Total load power ; Total system power loss ; Input power to the power grid ; Photovoltaic power generation ; Wind power generation capacity ; For energy storage system power Charging is negative, discharging is positive; , , , The coefficients to be fitted; This is the error term. The equation describes the balance between the total load and power loss of the transformer substation and the power supplied by the grid, renewable energy, and energy storage systems.

[0274] 2. Voltage balance equation:

[0275] ;

[0276] in, For nodes voltage ; Power supply side voltage ; For nodes Number of line segments to the power source; For the first resistance of a section of the line ; For the first Reactance of the line segment ; For the first Current in the line segment ; The imaginary unit; , , The coefficients to be fitted; This represents the error term. The equation describes the relationship between node voltages and source voltages, line parameters, and currents.

[0277] 3. Power loss equation:

[0278] ;

[0279] in, Total power loss ; This represents the number of line segments in the system. For the first Current in the line segment ; For the first resistance of a section of the line ; This is the core loss coefficient; For the first voltage of the line segment ; This is the switching loss factor; Switching frequency ; For the first Apparent power of the line segment ; , , The coefficients to be fitted; This is the error term. The equation describes the total power loss of the system, including line losses, core losses, and switching losses.

[0280] 4. Energy storage charging and discharging equations:

[0281] ;

[0282] in, for State of charge at time t is State of charge at time t is charging efficiency is discharging efficiency is charging power at time t ; is discharging power at time t ; is time interval ; is rated capacity of energy storage system ; , , are coefficients to be fitted is error term. This equation describes the charging and discharging process of the energy storage system and its impact on the battery state of charge .

[0283] 5. Photovoltaic power generation equation:

[0284] ;

[0285] where, is photovoltaic power generation power ; is photovoltaic panel efficiency ; is photovoltaic panel area ; is solar irradiance ; is temperature coefficient ; is photovoltaic cell temperature ; is reference temperature ; ; , , are coefficients to be fitted is error term. This equation describes the variation of photovoltaic power generation power with irradiance and cell temperature.

[0286] 6. Wind power generation equation:

[0287] ;

[0288] where, is wind power generation power ; is air density ; is wind wheel swept area ; Wind energy utilization coefficient; Wind speed ; To cut in wind speed ; To cut off the wind speed ; Rated wind speed ; Rated power ; , The coefficients to be fitted; This is the error term. This equation describes the piecewise characteristics of wind power generation as a function of wind speed.

[0289] 7. Load response equation:

[0290] ;

[0291] ;

[0292] in, for Actual load power at time ; for Reference load power at time ; for Demand response power adjustment at any time ; This is the demand response sensitivity coefficient; for Electricity price at any time ; For reference electricity price ; , , The coefficients to be fitted; This represents the error term. This equation describes the dynamic response characteristics of the load to electricity price signals.

[0293] 8. Power grid support equations:

[0294] ;

[0295] in, Support capacity for the power grid ; The maximum active power that the power grid can provide. ; Active power input to the actual power grid ; The maximum reactive power that the power grid can provide. ; reactive power input for the actual power grid ; , are coefficients to be fitted; is an error term. This equation describes the active and reactive support capability that the power grid can provide.

[0296] In summary, this step establishes a detailed response model covering all aspects of the transformer area, providing a basis for subsequent optimization modeling and decision-making. All coefficients to be fitted and error terms need to be estimated through a large amount of historical data.

[0297] The specific implementation of step S30 is as follows:

[0298] The purpose of this step is to use continuously collected transformer area operation data to fit the transformer area response equation set established in step S20, obtain preliminary model parameters, and verify the accuracy of the model.

[0299] Specifically, first, using the least squares method, nonlinear least squares method, or maximum likelihood estimation method, combined with the actual operation data collected in step S10, the eight response equations in step S20 are fitted to obtain the first set of fitting coefficients: ;

[0300] In this way, a preliminary transformer area response model is established.

[0301] Next, the actual operation data collected in step S10, including

[0302] , etc., are substituted into the equation set obtained by the above fitting to calculate a set of result data that agrees with the actual situation, denoted as the result set.

[0303] Through this step, a preliminary transformer area response model is established, and it is verified whether the model can accurately describe the operation characteristics of the actual transformer area. This lays a foundation for subsequent data augmentation and optimization modeling.

[0304] The specific implementation of step S40 is as follows:

[0305] The purpose of this step is to use data augmentation techniques to expand the coverage of the training data set and improve the generalization ability of the model, providing more abundant data support for subsequent optimization modeling.

[0306] Specifically, first, the following data augmentation techniques are used to expand and enhance the original operation data collected in step S10:

[0307] 1. Random interpolation: Randomly interpolate missing points in the collected data to generate new data samples.

[0308] 2. Rolling window method: Roll the time window of the collected data to generate more time series data.

[0309] 3. Time series model: Train a time series prediction model based on the collected data to generate simulated data.

[0310] Through the above techniques, a set of augmented and expanded simulated operation data is obtained, denoted as augmented and expanded operation data.

[0311] Next, these augmented and expanded operation data are substituted into the district response equation set fitted in step S30, and the results are calculated and recorded to obtain an augmented and expanded result set.

[0312] Finally, the original operation data collected in step S10 and the augmented and expanded operation data generated above are merged into a more comprehensive operation data collection, and the result set of step S30 and the augmented and expanded result set are merged into a result data collection.

[0313] Through this step, the coverage of the training data set is expanded, the generalization ability of the model is improved, and a more solid data foundation is laid for subsequent optimization modeling.

[0314] The specific implementation of step S50 is as follows:

[0315] The purpose of this step is to construct a multi-dimensional data matrix based on the operation data collection and the result data collection, and to standardize the input and output data, providing a basis for subsequent feature extraction and optimization modeling.

[0316] First, the operation data collection and the result data collection are organized into a multi-dimensional matrix , where:

[0317] Row represents different time points ;

[0318] Column represents different data types and parameters, such as

[0319] , etc.

[0320] Depth represents different scenarios or augmented and expanded data sets.

[0321] In this way, a rich multi-dimensional data structure is constructed, laying a foundation for subsequent feature extraction and optimization modeling.

[0322] Next, the multi-dimensional matrix is divided into input matrix and output matrix . Among them, contains all the input parameters, such as measured data of load, power grid, renewable energy and energy storage system; contains output parameters related to optimization objectives, such as energy conversion efficiency, energy loss, power grid support, etc.

[0323] Finally, the input matrix and the output matrix are standardized to eliminate the influence of different dimensions:

[0324] ;

[0325] ;

[0326] Among them, and are the mean and standard deviation of , respectively, and are the mean and standard deviation of .

[0327] Through this step, a well-organized multi-dimensional data structure is constructed, and the input and output data are standardized, providing a standardized data basis for subsequent feature extraction and optimization modeling.

[0328] The specific implementation of step S60 is as follows:

[0329] The purpose of this step is to use the Tucker decomposition method to reduce the dimension of the multi-dimensional matrix constructed in step S50 and extract features, obtaining a simplified feature matrix , providing high-quality input for subsequent optimization modeling.

[0330] Tucker decomposition is a multi-dimensional data decomposition method that can approximately decompose the original multi-dimensional matrix into a core tensor and three factor matrices , , :

[0331] ;

[0332] Among them, denotes n-mode multiplication. The size of the core tensor is This reflects the most important features and correlations in the data. Factor matrix , , The sizes are respectively , , It contains the principal components of the data across various dimensions.

[0333] Specifically, firstly, for multidimensional matrices Perform Tucker decomposition to obtain the core tensor. sum factor matrix , , Among them, the core tensor The size of each element in the data reflects the importance of each feature.

[0334] Next, according to The size of the elements in the middle, select the most important one. 1. Features, construct a simplified feature matrix This approach preserves the core information of the data while significantly reducing its dimensionality, providing high-quality input features for subsequent optimization and modeling.

[0335] This step effectively extracts key features from multidimensional data using tensor decomposition techniques such as Tucker decomposition, laying the foundation for subsequent optimization modeling.

[0336] The specific implementation method of step S70 is as follows:

[0337] The purpose of this step is to leverage the feature matrix extracted in step S60. A multi-objective optimization model is constructed to provide a basis for subsequent optimal energy storage configuration.

[0338] Specifically, the decision variable vector of the energy storage system is denoted as... Construct the following 5 objective functions:

[0339] 1. Objective function for maximizing system energy conversion efficiency:

[0340] ;

[0341] in, for Output power at any moment for Input power at time 10:00 For time step, This represents the total number of time steps. The objective function aims to maximize the energy conversion efficiency of the system.

[0342] 2. Energy loss minimization objective function:

[0343] ;

[0344] wherein, is the power loss at time t. The objective function aims to minimize the total energy loss of the system.

[0345] 3. Energy storage maximum support to grid objective function:

[0346] ;

[0347] wherein, is the grid support capacity at time t. The objective function aims to maximize the support of the energy storage system to the grid.

[0348] 4. Power supply reliability maximization objective function:

[0349] ;

[0350] wherein, is the power supply power at time t, is the demand power at time t. The objective function aims to maximize the power supply reliability.

[0351] 5. Total cost minimization objective function:

[0352] ;

[0353] wherein, is the investment cost, is the operation cost at time t. The objective function aims to minimize the total cost.

[0354] At the same time, corresponding constraints are set, including energy storage capacity and power limit, voltage and frequency limit, load balance constraint, energy storage charge and discharge depth and frequency limit, renewable energy output limit, and investment budget limit, etc.

[0355] In summary, this step constructs a multi-objective optimization model containing 5 objective functions and multiple constraints, which provides a basis for subsequent optimal energy storage configuration.

[0356] The specific implementation of step S80 is as follows:

[0357] The purpose of this step is to solve the multi-objective optimization model constructed in step S70 using a GPU-accelerated multi-objective optimization algorithm to obtain a non-inferior solution set. ​​​​​

[0358] First, the objective function matrix is ​​computed in parallel on the GPU. and constraint matrix .

[0359] Next, a non-dominated sorting algorithm implemented using GPUs is used to solve the multi-objective optimization problem. The specific steps are as follows:

[0360] 1. Dominance Relationship Matrix Calculation:

[0361] ;

[0362] in, For elements of the dominance relation matrix, and The first The and the first One solution, For the first There are several objective functions. This step calculates the dominance relationships between each solution.

[0363] 2. Domination Count Calculation:

[0364] ;

[0365] in, For the first The number of dominated solutions, This represents the total number of solutions. This step calculates the number of times each solution is dominated by other solutions.

[0366] 3. Pareto front identified:

[0367] ;

[0368] This step determines the first Pareto front, which is the solution set with a dominated count of 0.

[0369] 4. Crowding Distance Calculation:

[0370] ;

[0371] in, For the first The crowding distance of each solution The number of objective functions, and They were respectively in the second The function values ​​of two adjacent solutions on each objective. and The first The maximum and minimum values ​​of each objective function are calculated. This step also calculates the crowding distance of each solution in the objective space.

[0372] 5. Selection, crossover and mutation:

[0373] Selection, crossover and mutation operations are performed based on non-dominated sorting and crowding distance, and iterative optimization is performed.

[0374] 6. Termination condition check:

[0375] Repeat steps 2-5 until the termination condition is reached to obtain the final non-inferior solution set.

[0376] The algorithm fully utilizes the parallel computing capability of the GPU, greatly improving the efficiency of multi-objective optimization. At the same time, the calculation of non-dominated sorting and crowding distance ensures the convergence and dispersion of the final solution set, meeting the requirements of multi-objective optimization.

[0377] In order to better understand and implement the present application, the following provides an embodiment 2 of a specific application scenario of the present application: a certain power enterprise in a certain city, its distribution network is composed of a 110kV substation and several 10kV distribution transformers, supplying about 20,000 households and commercial users. In recent years, the substation has gradually connected photovoltaic power generation and wind power generation systems, with a renewable energy installed capacity of about 10MW. In order to smooth the volatility of renewable energy and improve the flexibility and reliability of power supply, the power enterprise plans to deploy a storage system in the substation.

[0378] Based on the substation storage multi-objective optimization configuration method proposed in the present application, the power enterprise has carried out detailed data collection and modeling analysis on the substation, and finally determined the optimal storage system configuration scheme. The specific implementation process is as follows:

[0379] I. Data collection and preliminary modeling

[0380] 1. Operation data collection

[0381] The power enterprise has deployed a large number of monitoring equipment in the substation and has continuously collected various operation data for 1 year, including:

[0382] Table 1. Substation operation data collection

[0383]

[0384] Through 1 year of continuous collection, the power enterprise has obtained comprehensive operation data of the substation.

[0385] 2. Preliminary response model establishment

[0386] According to the collected operation data, the power enterprise has established a response model covering various aspects of the substation, mainly including:

[0387] (1) Power balance equation:

[0388] ;

[0389] in, Total load power; This refers to the system's power loss. Input power to the power grid; Photovoltaic power generation; For wind power generation capacity; Power of the energy storage system; This is the error term.

[0390] (2) Voltage balance equation:

[0391] ;

[0392] in, Node voltage; This refers to the power supply side voltage. and The first Resistance and reactance of the circuit segment; For the first The current in the section of the line; This is the error term.

[0393] (3) Power loss equation:

[0394] ;

[0395] in, Total power loss; This refers to the number of line segments in the system. For the first Sectional line current; For the first Section line resistance; This is the core loss coefficient; For the first Section line voltage; This is the switching loss factor; Switching frequency; For the first Apparent power of the line segment; This is the error term.

[0396] (4) Energy storage charging and discharging equations:

[0397] ;

[0398] in, for State of charge at time t, For time intervals; For charging efficiency; For discharge efficiency; and are the charging and discharging power, respectively, with the value range of ; is the rated capacity of the energy storage system; is the error term.

[0399] (5) Photovoltaic power generation equation:

[0400] ;

[0401] where, is the photovoltaic power generation power; is the photovoltaic panel efficiency; is the photovoltaic panel area; is the solar irradiance; is the temperature coefficient; is the photovoltaic cell temperature; is the error term.

[0402] (6) Wind power generation equation:

[0403] ;

[0404] where, is the wind power generation power; is the air density; is the wind wheel swept area; is the wind energy utilization coefficient; is the wind speed; , , are the cut-in, rated and cut-out wind speeds, respectively; is the rated power; is the error term.

[0405] (7) Load response equation:

[0406] ;

[0407] where, is the actual load power; is the reference load power; is the demand response sensitivity coefficient; is the electricity price; is the error term.

[0408] (8) Grid support equation:

[0409] ;

[0410] where, is the grid support capacity; and Pmax and Qmax are the maximum active and reactive power outputs of the grid, respectively; and P and Q are the actual active and reactive power inputs of the grid, respectively; e is the error term.

[0411] By fitting the above 8 response equations, the power company established a preliminary operation model for the substation.

[0412] 3. Model verification and data augmentation

[0413] To verify the accuracy of the established response model, the power company substituted the actual collected operation data into the model for calculation and compared the results with the measured results. After repeated adjustments, the fitting accuracy of each response equation reached more than 95%.

[0414] To further enhance the model's adaptability to future uncertainties, the power company used data augmentation techniques to generate more abundant simulated data sets based on the original operation data. The specific methods include:

[0415] (1) Random interpolation method: random interpolation is performed on some missing points in the original data to generate new data samples.

[0416] (2) Rolling window method: time window rolling is performed on the original data to generate more time series data.

[0417] (3) Time series model: based on the original data, an ARIMA time series prediction model is trained to generate simulated data.

[0418] Through the above methods, the power company expanded the coverage of the training data set, including various possible situations such as different load levels, weather conditions, and price changes. These augmented data, together with the original collected data, constitute the substation operation data set and the result data set, providing a broad foundation for subsequent optimization modeling.

[0419] II. Optimization modeling and solution

[0420] 1. Data organization and feature extraction

[0421] The power company organized the above operation data set and result data set into a multi-dimensional matrix , where:

[0422] Row represents different time points ;

[0423] Column represents different data types and parameters, such as , , , , , , , , , , , , , , , , , , , , , , , , , , wait;

[0424] depth Representing different scenarios or expanded datasets.

[0425] Then, the power company applied this multidimensional matrix Tucker decomposition was performed to obtain the core tensor. sum factor matrix , , .according to The size of the elements in the middle, power companies extracted the most important ones , , A simplified feature matrix was constructed from these features. .

[0426] 2. Construction of Multi-Objective Optimization Model

[0427] Based on the above feature matrix The power company constructed a multi-objective optimization model with the following five objective functions:

[0428] (1) Maximize system energy conversion efficiency:

[0429] ;

[0430] in, for Output power at any moment for Input power at time 10:00 This represents the total number of time steps. The objective function aims to maximize the energy conversion efficiency of the system.

[0431] (2) Minimize energy loss:

[0432] ;

[0433] in, for The power loss at time t. The objective function aims to minimize the total energy loss of the system.

[0434] (3) Maximizing the supporting role of energy storage in the power grid:

[0435] ;

[0436] in, for The grid support capacity at any given time. The objective function aims to maximize the grid support provided by the energy storage system.

[0437] (4) Maximize power supply reliability:

[0438] ;

[0439] in, for Power supply at any time for The required power at any given time. This objective function aims to maximize power supply reliability.

[0440] (5) Minimize total cost:

[0441] ;

[0442] in, For investment costs, for The operating cost at time step 1. The objective function aims to minimize the total cost.

[0443] At the same time, power companies have also set the following constraints:

[0444] Energy storage system capacity and power limitations: , ;

[0445] Voltage and frequency limitations: , ;

[0446] Load balance constraints: ;

[0447] Limitations on the depth and number of charge / discharge cycles for energy storage: , , ;

[0448] Renewable energy output limits: , ;

[0449] Investment budget constraint: .

[0450] 3. GPU-accelerated multi-objective optimization solution

[0451] Based on the above multi-objective optimization model, the power enterprise uses a GPU-accelerated non-dominated sorting algorithm for solution, the main steps are as follows:

[0452] (1) Parallel computing the objective function matrix on GPU and the constraint condition matrix .

[0453] (2) Calculate the dominance relation matrix , judge whether each solution is dominated by other solutions.

[0454] (3) Parallel computing the dominance count and dominated set of each solution.

[0455] (4) Determine the Pareto front , that is, the solution set with a dominance count of 0.

[0456] (5) Calculate the crowding distance of each solution , reflecting the distribution of solutions in the objective space.

[0457] (6) Based on non-dominated sorting and crowding distance, select, cross and mutate, and iterate optimization.

[0458] (7) Repeat steps (2)-(6) until the termination condition is reached, and get the final non-inferior solution set.

[0459] The entire solution process is parallel computing on GPU, which greatly improves the computing efficiency. The power enterprise finally gets a set of 20 non-inferior solution alternatives.

[0460] III. Optimal energy storage configuration scheme

[0461] 1. Non-inferior solution comprehensive evaluation

[0462] In order to select the optimal scheme from the non-inferior solution set, the power enterprise conducted a comprehensive evaluation of each non-inferior solution. The specific steps are as follows:

[0463] (1) Normalize the values of the five objective functions to unify the numerical range to the interval [0, 1]:

[0464] ;

[0465] Among them, is the normalized value of the i-th objective function. the overall score of the non-inferior solution, is the weight of each objective function, and is the minimum and maximum value of the i-th objective function among all non-inferior solutions, respectively.

[0466] (2) Select the non-inferior solution with the highest overall score as the optimal solution . .

[0467] 2. Optimal energy storage configuration scheme

[0468] After the above evaluation, the power enterprise selects the following optimal energy storage configuration scheme:

[0469] Table 2 Optimal energy storage configuration scheme

[0470]

[0471] The values of this scheme on the five objective functions are respectively: , , , , , and the overall score is the highest.​

Claims

1. A method for multi-objective optimization configuration of transformer substation energy storage to enhance source-load carrying capacity, characterized in that, Includes the following steps: S10. Continuously collect the operating data of the distribution area, including load data, distribution network input data, photovoltaic power generation data, wind power generation data, and energy storage data; S20. Based on the aforementioned operating data, establish a set of operating response equations for the transformer substation, including power balance equations, voltage balance equations, power loss equations, energy storage charging and discharging equations, photovoltaic power generation equations, wind power generation equations, load response equations, and grid support equations. S30. The operating data of the continuously collected transformer area is used to fit the operating response equation set of the transformer area to obtain the first equation set. The running data is then input into the first set of equations to obtain the result set; S40. The collected running data is augmented using a data augmentation method to obtain augmented running data, which is then substituted into the first set of equations for calculation to obtain an augmented result set. The running data and the expanded running data are merged into a running data set, and the result set and the expanded result set are combined into a result data set. S50. Establish a multidimensional matrix based on the set of running data and the set of result data; The multidimensional matrix is ​​divided into an input matrix X and an output matrix Y, where X contains all input parameters and Y contains all output parameters related to the optimization objective; X and Y are standardized to eliminate the influence of different dimensions. S60. Use tensor decomposition technology to reduce the dimensionality and extract features from the multidimensional matrix to obtain a simplified feature matrix F; S70. Based on the feature matrix F, construct a multi-objective optimization model, specifically: using the energy storage data as the decision variable vector, construct an objective function matrix, including: an objective function to maximize system energy conversion efficiency, an objective function to minimize energy loss, an objective function to maximize the supporting role of energy storage in the power grid, an objective function to maximize power supply reliability, and an objective function to minimize total cost; set a constraint matrix, including energy storage capacity and power limits, voltage and frequency limits, load balance constraints, energy storage charging and discharging depth and number of times limits, renewable energy output limits, and investment budget limits; S80. The multi-objective optimization model is solved using a GPU-accelerated multi-objective optimization algorithm to obtain a set of non-dominated solutions; S90. Calculate the weighted score of each non-dominated solution in the non-dominated solution set, select the non-dominated solution with the highest weighted score as the optimal solution, and output the energy storage data corresponding to the optimal solution as the optimized configuration data of the energy storage in the distribution area. The power balance equation is specifically expressed as follows: ; In the formula, Total load power; This refers to the system's power loss. Input power to the power grid; Photovoltaic power generation; This refers to the power output of wind power generation. Power of the energy storage system; , , , The coefficients to be fitted; This is the first error term; The voltage balance equation is specifically expressed as follows: ; In the formula, Let be the voltage at node i; This refers to the power supply side voltage. Let i be the number of line segments from node i to the power source; Let be the resistance of the k-th segment of the line; Let K be the reactance of the k-th segment of the line; Let be the current in the k-th segment of the line; The imaginary unit; , , The coefficients to be fitted; This is the second error term; The power loss equation is specifically expressed as follows: ; In the formula, This represents the number of line segments in the system. Let be the current in the k-th segment of the line; Let be the resistance of the k-th segment of the line; This is the core loss coefficient; Let be the voltage of the k-th line segment; This is the switching loss factor; The switching frequency; Let be the apparent power of the k-th line segment; , , The coefficients to be fitted; This is the third error term; The energy storage charging and discharging equations are specifically expressed as follows: ; In the formula, Let be the state of charge at time t; The state of charge at time t+1; For charging efficiency; For discharge efficiency; Let be the charging power at time t; Let be the discharge power at time t; For time intervals; This refers to the rated capacity of the energy storage system. , , The coefficients to be fitted; This is the fourth error term; The photovoltaic power generation equation is specifically expressed as follows: ; In the formula, Photovoltaic power generation; For photovoltaic panel efficiency; The area of ​​the photovoltaic panel; Solar irradiance; Temperature coefficient; The temperature of the photovoltaic cell; For reference temperature; , , The coefficients to be fitted; This is the fifth error term; The wind power generation equation is specifically expressed as follows: ; In the formula, This refers to the power output of wind power generation. air density; The area swept by the wind turbine; The wind energy utilization coefficient; Wind speed; To cut in wind speed; To cut off the wind speed; Rated wind speed; Rated power; , The coefficients to be fitted; This is the sixth error term; The load response equation is specifically expressed as follows: ; ; In the formula, The actual load power at time t; Let be the baseline load power at time t; The demand response adjustment power at time t; This is the demand response sensitivity coefficient; Let be the electricity price at time t; For reference electricity prices; , , The coefficients to be fitted; This is the seventh error term; The power grid support equation is specifically expressed as follows: ; In the formula, To support the capacity of the power grid; This represents the maximum active power that the power grid can provide. The active power input to the actual power grid; This represents the maximum reactive power that the power grid can provide. The reactive power input to the actual power grid; , The coefficients to be fitted; This is the eighth error term.

2. The method for multi-objective optimization configuration of transformer substation energy storage to enhance source-load carrying capacity according to claim 1, characterized in that, in, The sampling frequency is 1 minute to 15 minutes.

3. The method for multi-objective optimization configuration of transformer substation energy storage to enhance source-load carrying capacity according to claim 1, characterized in that, The load data specifically includes active power, reactive power, voltage, current, and power factor.

4. The method for multi-objective optimization configuration of transformer substation energy storage to improve source-load carrying capacity according to claim 1, characterized in that, The specific data input to the distribution network includes feeder voltage, current, active power, reactive power, and power factor.

5. The method for multi-objective optimization configuration of transformer substation energy storage to improve source-load carrying capacity according to claim 1, characterized in that, The photovoltaic power generation data specifically includes output active power, reactive power, voltage, current, irradiance, and temperature.

6. The method for multi-objective optimization configuration of transformer substation energy storage to improve source-load carrying capacity according to claim 1, characterized in that, The wind power generation data specifically includes output active power, reactive power, voltage, current, wind speed, and wind direction.

7. The method for multi-objective optimization configuration of transformer substation energy storage to enhance source-load carrying capacity according to claim 1, characterized in that, The energy storage data specifically includes charge / discharge power, state of charge, voltage, current, and temperature.

8. The method for multi-objective optimization configuration of transformer substation energy storage to improve source-load carrying capacity according to claim 7, characterized in that, The rows of the multidimensional matrix represent different points in time, the columns represent different data types and parameters, and the depth represents different scenarios or expanded datasets.

9. A method for multi-objective optimization configuration of transformer substation energy storage to enhance source-load carrying capacity according to claim 1, characterized in that, The step of using tensor decomposition technology to reduce the dimensionality of the multidimensional matrix and extract features to obtain a simplified feature matrix F specifically involves: using the Tucker decomposition method to decompose the multidimensional matrix into a core tensor and a factor matrix; and using the element size of the core tensor to identify important features and correlations. Based on the decomposition results, a simplified feature matrix F is constructed, retaining important feature information.

Images