Intelligent battery energy storage system load forecasting method and system

Abstract

The present invention discloses an intelligent battery energy storage system load forecasting method and system, the method includes data collection, data preprocessing, dynamic clustering and battery energy storage system load forecasting. The present invention belongs to the field of battery load forecasting, and specifically refers to an intelligent battery energy storage system load forecasting method and system. This scheme constructs a multi-layer evaluation matrix, identifies key indicators based on indicator deviation, uses a weight matrix to perform consistency check, dynamically updates the final weight, performs waveform conversion on the load data, performs clustering processing, determines the optimal number of clusters and iteratively updates the cluster center, and finally introduces multi-objective optimization for iterative update, thereby ensuring dynamic adaptability to complex load forecasting data; by adaptively adjusting the step size, combining global and local searches, and automatically optimizing the clustering results through position mutation and dynamic weight adjustment, it ensures that the accuracy of the final load forecasting effect is high enough.

Description

Technical Field

[0001] The present invention relates to the technical field of battery load prediction, and in particular to an intelligent battery energy storage system load prediction method and system. Background Art

[0002] The load forecasting method of battery energy storage system refers to predicting the load demand of real-time battery energy storage system by analyzing various data of battery energy storage system. The main goal of this method is to ensure that the energy storage system can achieve efficient charging and discharging management under different load demands, so as to optimize energy utilization and distribution, improve system operation efficiency, and reduce operating costs. However, the general load method of battery energy storage system is seriously affected by noise, and cannot ensure the reasonable distribution of the importance of load forecasting indicators in the entire forecasting process, which leads to poor load forecasting effect; the general load method of battery energy storage system has slow convergence speed and poor parameter adjustment accuracy when adjusting model parameters, which leads to poor robustness of the model in complex load forecasting environment and low load forecasting effect. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent battery energy storage system load forecasting method and system. In view of the problem that the general battery energy storage system load method is seriously affected by noise and cannot ensure the reasonable distribution of the importance of load forecasting indicators in the entire forecasting process, which leads to poor load forecasting effect, this solution constructs a multi-layer evaluation matrix, identifies key indicators based on indicator deviation, uses a weight matrix for consistency verification, and dynamically updates the final weight through initial weights, distinction and uncertainty. After waveform conversion of the load data, clustering processing is performed based on data representations of different scales, the optimal number of clusters is determined, and the cluster centers are iteratively updated. Finally, by introducing Multi-objective optimization is iteratively updated to accurately locate the clustering center of the load data, thereby ensuring the correctness of the load forecasting results of the battery energy storage system and its dynamic adaptability to complex load forecasting data; the general battery energy storage system load method has the problems of slow convergence speed and poor parameter adjustment accuracy when adjusting the model parameters, which leads to poor robustness of the model in a complex load forecasting environment and low load forecasting effect. This solution adaptively adjusts the step size, combines global and local searches, and responds to noise and outliers in the load forecasting data through position mutation and dynamic weight adjustment, ensuring the robustness of the model in a complex load forecasting environment, automatically optimizing the clustering results, and ensuring that the final load forecasting effect is accurate enough.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent battery energy storage system load prediction method, the method comprising the following steps:

[0005] Step S1: data collection;

[0006] Step S2: data preprocessing;

[0007] Step S3: dynamic clustering;

[0008] Step S4: Battery energy storage system load forecasting.

[0009] Furthermore, in step S1, the data collection is to collect historical battery energy storage system load data and real-time battery energy storage system load data; the historical battery energy storage system load data and the real-time battery energy storage system load data both include battery status, environmental factors, time data and external load requirements; the historical battery energy storage system load data also includes load assessment level; the load assessment level includes normal, light load, moderate load and severe load; the load assessment level at the next moment is used as a data label; the data label does not participate in clustering operations.

[0010] Furthermore, in step S2, the data preprocessing is to perform data cleaning, data conversion and standardization on the collected data; the data cleaning is to remove missing values ​​and duplicate values; the data conversion is to convert the data into a vector form; and the standardization is to perform data standardization based on the maximum and minimum normalization method.

[0011] Furthermore, in step S3, the dynamic clustering specifically includes the following steps:

[0012] Step S31: construct a sparse evaluation matrix; construct a multi-layer evaluation matrix, expressed as ; Sparse decomposition of the multi-layer evaluation matrix is ​​performed, expressed as: A≈ZW, and the constraints are expressed as: ; Get the sparse evaluation matrix A; where A k is the kth layer of the multi-layer evaluation matrix, each layer represents a different data dimension; is the element in the k-th layer evaluation matrix, representing the j-th indicator value of the i-th sample; Z is the simplified feature matrix; W is the weighting matrix; α is a constant that controls sparsity; is the L1 norm; m is the total number of samples;

[0013] Step S32: Calculate the index deviation; the formula used is as follows:

[0014] ;

[0015] ;

[0016] ;

[0017] In the formula, R i,j is the degree of deviation of the jth indicator of the ith sample from the mean; S i,jis the original data value of the jth indicator of the i-th sample; u j and σ j are the mean and standard deviation of the jth indicator respectively; is a smoothing term; D j is the discrimination of the jth indicator; E j is the uncertainty of the jth indicator;

[0018] Step S33: Calculate the final weight; initialize the weight matrix B; and perform consistency check, expressed as: ; and calculate the final weight, expressed as: ; Dynamic weights are introduced and used when weights are updated ; Where CI is the consistency of the weight matrix; is the maximum eigenvalue of the weight matrix; RI is the random consistency index; w j is the final weight; is the initial weight; α2 is the influencing parameter for controlling the difference; and are the final weights at the t+1th iteration and the tth iteration respectively; is the uncertainty at the tth iteration; is the weight factor;

[0019] Step S34: clustering processing; specifically including:

[0020] Step S341: data optimization; waveform conversion is performed on the data to obtain data representations of different scales, which are represented as follows: ;in, It is the original data; is the wavelet function; a i and b i are the scale parameter and the translation parameter; It is the data representation after waveform conversion;

[0021] Step S342: Determine the optimal number of clusters; the formula used is as follows:

[0022] ;

[0023] In the formula, CH(·) is the optimal cluster number index; tr(·) is the trace; B k and W k are the scatter matrix between samples and the scatter matrix within samples respectively; n is the total number of indicators;

[0024] Step S343: cluster iteration; randomly select the initial cluster center; calculate the distance from each non-cluster center to the cluster center, and assign the non-cluster center based on the minimum distance; iteratively update the cluster center and calculate the objective function; the distance calculation is expressed as: ;

[0025] ; Introduce the multi-objective optimization method to define the objective function, expressed as: ; Progressive learning is introduced during the update of the clustering center, and the clustering center update formula is expressed as: ; Among them, d m (·) is the distance metric; x i and x j are the i-th data point and the j-th data point respectively; S is the covariance matrix; ρe(·) is the local adaptive density; ρ(·) is the local density; N(·) is the set of neighborhood points, and i1 is the neighborhood point index; d(·) is the Euclidean distance; is the local density parameter; f(·) is the objective function; X and Y are the proportion of the label with the largest number of historical data and the proportion of other labels in the cluster respectively; and are the positions of the j1-th cluster at the (t + 1)-th and t-th iterations respectively; is the learning rate parameter; is the iteration upper limit; is the smoothing term; , and are the weight coefficients;

[0026] Step S35: Judgment; When the objective function converges, the clustering is completed; If only the iteration upper limit is reached, go to step S36; Otherwise, continue the iteration;

[0027] Step S36: Clustering optimization; Specifically including:

[0028] Step S361: Construct the parameter optimization space based on the weight matrix, the influence parameter of the control difference degree, the local density parameter, the learning rate parameter, the scale parameter and the translation parameter; Initialize the parameter optimization population, and use the objective function when the parameter optimization individual reaches the iteration upper limit or converges as the individual fitness value; The formula for initializing the optimization population is as follows:

[0029] ;

[0030] In the formula, X i2,j2 (·) is the position of the i2-th optimization individual in the j2-th dimension; 0 represents the 0-th search, that is, initialization; rand is a random number belonging to LB to UB; UB and LB are the search upper limit and lower limit of the j2-th dimension respectively;

[0031] Step S362: Exploration stage; When it < itmax / 2, update the step size control, expressed as: ; ; The position update is expressed as: ; Among them, ; W1(·) and W2(·) are individual movement weights; i2 is the index; hy(·) is the individual demand value; SHhy is the sum of the demand values ​​of all individuals; r, r1, r2 are all random numbers between 0 and 1; N is the number of individuals in the population; randn is a random number sampled from a standard normal distribution; X b,j2 (·) is the position of the optimal individual in the j2th dimension. The optimal individual is the individual with the lowest fitness value; R is a random number generated in the range [-a, a], a=2·(1-it / itmax); Hs is the additional demand value of the individual, Hs=Lhs·(1+r); Lhs is the lower bound demand value; Allf(·) is the individual fitness value; BF is the optimal individual fitness value; itmax is the maximum number of searches;

[0032] Step S363: Utilization phase; when it ≥ it / 2, the location update is expressed as: ;in, ; RB is a random number generated from Gaussian distribution; is the average fitness value of the population; r3 is a random number between 0 and 1;

[0033] Step S364: mutation; perform position mutation on the updated individual. If the fitness value of the position after mutation decreases, retain the mutation position, otherwise retain the original position. The formula used for mutation is as follows:

[0034] ;

[0035] In the formula, is the position after mutation; r4 and r5 are random numbers between 0 and 1; U is a binary vector; X roll1,j2 (·) and X roll2,j2 (·) is the position of the random individual in the j2th dimension;

[0036] Step S365: optimization judgment; a fitness threshold is set in advance. When there is an individual fitness value lower than the fitness threshold, the optimization ends, the individual position, i.e., the parameter setting, is used to obtain the clustering result; if the maximum number of searches is reached, return to step S361; otherwise, return to step S362.

[0037] Further, in step S4, the battery energy storage system load prediction is based on the clustering result, and the label with the largest number of historical battery energy storage system load data is used as the cluster label, and the cluster label to which the real-time battery energy storage system load data belongs is used as the prediction result of the real-time battery energy storage system load data.

[0038] The present invention provides an intelligent battery energy storage system load prediction system, comprising a data acquisition module, a data preprocessing module, a dynamic clustering module and a battery energy storage system load prediction module;

[0039] The data acquisition module collects historical battery energy storage system load data and real-time battery energy storage system load data, and sends the data to the data preprocessing module;

[0040] The data preprocessing module performs data cleaning, data conversion and standardization on the collected data; and sends the data to the dynamic clustering module;

[0041] The dynamic clustering module constructs a multi-layer evaluation matrix, identifies key indicators based on indicator deviation, uses a weight matrix for consistency verification, and dynamically updates the final weight through initial weights, distinguishability, and uncertainty. After waveform conversion of the load data, clustering is performed based on data representations of different scales, the optimal number of clusters is determined, and cluster centers are iteratively updated. Finally, iterative updates are performed by introducing multi-objective optimization; and the global and local searches are combined by adaptively adjusting the step size, and parameters are adjusted through position mutation and dynamic weight adjustment, so as to finally achieve dynamic clustering; and the data is sent to the battery energy storage system load prediction module;

[0042] The battery energy storage system load prediction module realizes load prediction on real-time battery energy storage system load data based on clustering results.

[0043] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0044] (1) In view of the problem that the general battery energy storage system load method is seriously affected by noise and cannot ensure the reasonable distribution of the importance of load forecasting indicators in the entire forecasting process, which leads to poor load forecasting results, this scheme constructs a multi-layer evaluation matrix, identifies key indicators based on indicator deviation, uses a weight matrix for consistency verification, and dynamically updates the final weight through initial weights, distinction and uncertainty. After waveform conversion of the load data, clustering is performed based on data representations of different scales, the optimal number of clusters is determined, and the cluster centers are iteratively updated. Finally, multi-objective optimization is introduced for iterative updating to accurately locate the cluster centers of the load data, thereby ensuring the correctness of the load forecasting results of the battery energy storage system and its dynamic adaptability to complex load forecasting data.

[0045] (2) In view of the fact that the general battery energy storage system load method has the problems of slow convergence speed and poor parameter adjustment accuracy when adjusting the model parameters, which leads to poor robustness of the model in complex load forecasting environments and low load forecasting effect, this scheme adaptively adjusts the step size, combines global and local search, and copes with noise and outliers in load forecasting data through position mutation and dynamic weight adjustment, ensuring the robustness of the model in complex load forecasting environments, automatically optimizing the clustering results, and ensuring that the final load forecasting effect is accurate enough. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 A schematic diagram of a flow chart of a load forecasting method for an intelligent battery energy storage system provided by the present invention;

[0047] Figure 2 A schematic diagram of an intelligent battery energy storage system load forecasting system provided by the present invention;

[0048] Figure 3 It is a schematic diagram of the process of step S3.

[0049] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0051] In the description of the present invention, it is necessary to understand that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred system or element must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0052] Example 1, see Figure 1 The present invention provides an intelligent battery energy storage system load forecasting method, the method comprising the following steps:

[0053] Step S1: data collection, collecting historical battery energy storage system load data and real-time battery energy storage system load data;

[0054] Step S2: Data preprocessing, data cleaning, data conversion and standardization of the collected data;

[0055] Step S3: Dynamic clustering, by constructing a multi-layer evaluation matrix, identifying key indicators based on indicator deviation, using a weight matrix for consistency check, and dynamically updating the final weight through initial weight, distinguishability and uncertainty, after waveform conversion of load data, clustering processing is performed based on data representations of different scales, determining the optimal number of clusters and iteratively updating the cluster centers, and finally iteratively updating by introducing multi-objective optimization; and adaptively adjusting the step size, combining global and local searches, adjusting parameters through position mutation and dynamic weight adjustment, and finally realizing dynamic clustering;

[0056] Step S4: Battery energy storage system load forecasting, realizing load forecasting for real-time battery energy storage system load data based on clustering results.

[0057] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, both the historical battery energy storage system load data and the real-time battery energy storage system load data include battery status, environmental factors, time data and external load requirements; the historical battery energy storage system load data also includes load assessment levels; the load assessment levels include normal, light load, moderate load and severe load; the load assessment level at the next moment is used as a data label; the data label does not participate in the clustering operation.

[0058] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, data cleaning is to remove missing values ​​and duplicate values; data conversion is to convert data into vector form; and standardization is to standardize data based on the maximum and minimum normalization method.

[0059] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, dynamic clustering specifically includes the following steps:

[0060] Step S31: construct a sparse evaluation matrix; construct a multi-layer evaluation matrix, expressed as ; Sparse decomposition of the multi-layer evaluation matrix is ​​performed, expressed as: A≈ZW, and the constraints are expressed as: ; Get the sparse evaluation matrix A; where A k is the kth layer of the multi-layer evaluation matrix, each layer represents a different data dimension; is the element in the k-th layer evaluation matrix, representing the j-th indicator value of the i-th sample; Z is the simplified feature matrix; W is the weighting matrix; α is a constant that controls sparsity; is the L1 norm; m is the total number of samples;

[0061] Step S32: Calculate the index deviation; the formula used is as follows:

[0062] ;

[0063] ;

[0064] ;

[0065] In the formula, R i,j is the degree of deviation of the jth indicator of the ith sample from the mean; S i,j is the original data value of the jth indicator of the i-th sample; u j and σ j are the mean and standard deviation of the jth indicator respectively; is a smoothing term; D j is the discrimination of the jth indicator; E j is the uncertainty of the jth indicator;

[0066] Step S33: Calculate the final weight; initialize the weight matrix B; and perform consistency check, expressed as: ; and calculate the final weight, expressed as: ; Dynamic weights are introduced and used when weights are updated ; Where CI is the consistency of the weight matrix; is the maximum eigenvalue of the weight matrix; RI is the random consistency index; w j is the final weight; is the initial weight; α2 is the influencing parameter for controlling the difference; and are the final weights at the t+1th iteration and the tth iteration respectively; is the uncertainty at the tth iteration; is the weight factor;

[0067] Step S34: clustering processing; specifically including:

[0068] Step S341: data optimization; waveform conversion is performed on the data to obtain data representations of different scales, which are represented as follows: ;in, It is the original data; is the wavelet function; a i and b i are the scale parameter and the translation parameter; It is the data representation after waveform conversion;

[0069] Step S342: Determine the optimal number of clusters; the formula used is as follows:

[0070] ;

[0071] In the formula, CH(·) is the optimal cluster number index; tr(·) is the trace; B k and W k are the scatter matrix between samples and the scatter matrix within samples respectively; n is the total number of indicators;

[0072] Step S343: cluster iteration; randomly select the initial cluster center; calculate the distance from each non-cluster center to the cluster center, and assign the non-cluster center based on the minimum distance; iteratively update the cluster center and calculate the objective function; the distance calculation is expressed as: ;

[0073] ; Introduce the multi-objective optimization method to define the objective function, expressed as: ; Progressive learning is introduced when updating the cluster center, and the cluster center update formula is expressed as: ; where d m (·) is the distance metric; x i and x j are the i-th data point and the j-th data point respectively; S is the covariance matrix; ρe(·) is the local adaptive density; ρ(·) is the local density; N(·) is the domain point set, i1 is the neighborhood point index; d(·) is the Euclidean distance; is the local density parameter; f(·) is the objective function; X and Y are the proportion of the label with the largest number of historical data in the cluster and the proportion of other labels, respectively; and are the positions of the j1th cluster at the t+1th and tth iterations respectively; is the learning rate parameter; is the iteration upper limit; is a smoothing term; , and is the weight coefficient;

[0074] Step S35: Determination; when the objective function converges, clustering is completed; if only the iteration upper limit is reached, go to step S36; otherwise, continue iteration;

[0075] Step S36: cluster optimization.

[0076] By performing the above operations, for the general battery energy storage system load method, there are problems such as being severely affected by noise, being unable to ensure the reasonable allocation of the importance of load prediction indicators throughout the prediction process, and thus resulting in poor load prediction effects. In this solution, by constructing a multi-layer evaluation matrix, key indicators are identified based on indicator deviations, consistency tests are carried out using a weight matrix, and the final weights are dynamically updated through initial weights, discriminability, and uncertainty. After waveform conversion of the load data, clustering processing is performed based on data representations at different scales, the optimal number of clusters is determined and the cluster centers are iteratively updated. Finally, through the introduction of multi-objective optimization for iterative update, the cluster centers of the load data are accurately located, thereby ensuring the correctness of the battery energy storage system load prediction results and the dynamic adaptability to complex load prediction data.

[0077] Embodiment 5, refer to Figure 1 and Figure 3 , based on the above embodiment, in step S36, the clustering optimization specifically includes:

[0078] Step S361: Construct a parameter optimization space based on the weight matrix, the influence parameter of the control difference degree, the local density parameter, the learning rate parameter, the scale parameter, and the translation parameter; initialize the parameter optimization population, and use the objective function when the parameter optimization individual reaches the iteration upper limit or converges as the individual fitness value; the formula for initializing the optimization population is as follows:

[0079] ;

[0080] In the formula, X i2,j2 (·) is the position of the i2-th optimization individual in the j2-th dimension; 0 represents the 0th search, that is, initialization; rand is a random number belonging to LB to UB; UB and LB are the search upper limit and lower limit of the j2-th dimension respectively;

[0081] Step S362: Exploration stage; when it < itmax / 2, update the step size control, which is expressed as: ; ; The position update is expressed as: ; where ; W1(·) and W2(·) are individual movement weights; i2 is the index; hy(·) is the individual demand value; SHhy is the sum of the demand values of all individuals; r, r1, r2 are all random numbers from 0 to 1; N is the number of individuals in the population; randn is a random number sampled from the standard normal distribution; X b,j2(·) is the position of the optimal individual in the j2th dimension. The optimal individual is the individual with the lowest fitness value; R is a random number generated in the range [-a, a], a=2·(1-it / itmax); Hs is the additional demand value of the individual, Hs=Lhs·(1+r); Lhs is the lower bound demand value; Allf(·) is the individual fitness value; BF is the optimal individual fitness value; itmax is the maximum number of searches;

[0082] Step S363: Utilization phase; when it ≥ it / 2, the location update is expressed as: ;in, ; RB is a random number generated from Gaussian distribution; is the average fitness value of the population; r3 is a random number between 0 and 1;

[0083] Step S364: mutation; perform position mutation on the updated individual. If the fitness value of the position after mutation decreases, retain the mutation position, otherwise retain the original position. The formula used for mutation is as follows:

[0084] ;

[0085] In the formula, is the position after mutation; r4 and r5 are random numbers between 0 and 1; U is a binary vector; X roll1,j2 (·) and X roll2,j2 (·) is the position of the random individual in the j2th dimension;

[0086] Step S365: optimization judgment; a fitness threshold is set in advance. When there is an individual fitness value lower than the fitness threshold, the optimization ends, the individual position, i.e., the parameter setting, is used to obtain the clustering result; if the maximum number of searches is reached, return to step S361; otherwise, return to step S362.

[0087] By performing the above operations, the general battery energy storage system load method has the problems of slow convergence speed and poor parameter adjustment accuracy when adjusting model parameters, which leads to poor robustness of the model in complex load forecasting environments and low load forecasting effect. This solution adaptively adjusts the step size, combines global and local searches, and copes with noise and outliers in load forecasting data through position mutation and dynamic weight adjustment, ensures the robustness of the model in complex load forecasting environments, automatically optimizes the clustering results, and ensures that the final load forecasting effect is accurate enough.

[0088] Example 6, see Figure 1This embodiment is based on the above embodiment. In step S4, the battery energy storage system load prediction is based on the clustering result. The label with the largest number of historical battery energy storage system load data is used as the cluster label, and the cluster label to which the real-time battery energy storage system load data belongs is used as the prediction result of the real-time battery energy storage system load data.

[0089] Embodiment 7, see Figure 2 , this embodiment is based on the above embodiment, and the present invention provides an intelligent battery energy storage system load forecasting system, including a data acquisition module, a data preprocessing module, a dynamic clustering module and a battery energy storage system load module;

[0090] The data acquisition module collects historical battery energy storage system load data and real-time battery energy storage system load data, and sends the data to the data preprocessing module;

[0091] The data preprocessing module performs data cleaning, data conversion and standardization on the collected data; and sends the data to the dynamic clustering module;

[0092] The dynamic clustering module constructs a multi-layer evaluation matrix, identifies key indicators based on indicator deviation, uses a weight matrix for consistency verification, and dynamically updates the final weight through initial weights, distinguishability, and uncertainty. After waveform conversion of the load data, clustering is performed based on data representations of different scales, the optimal number of clusters is determined, and cluster centers are iteratively updated. Finally, iterative updates are performed by introducing multi-objective optimization; and the global and local searches are combined by adaptively adjusting the step size, and parameters are adjusted through position mutation and dynamic weight adjustment, so as to finally achieve dynamic clustering; and the data is sent to the battery energy storage system load prediction module;

[0093] The battery energy storage system load realizes load prediction for real-time battery energy storage system load data based on clustering results.

[0094] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0095] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0096] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent battery energy storage system load forecasting method, characterized in that: The method comprises the following steps: Step S1: data collection, collecting historical battery energy storage system load data and real-time battery energy storage system load data; Step S2: data preprocessing; Step S3: Dynamic clustering, by constructing a multi-layer evaluation matrix, identifying key indicators based on indicator deviation, using a weight matrix for consistency check, and dynamically updating the final weight through initial weight, distinguishability and uncertainty, after waveform conversion of load data, clustering processing is performed based on data representations of different scales, determining the optimal number of clusters and iteratively updating the cluster centers, and finally iteratively updating by introducing multi-objective optimization; and adaptively adjusting the step size, combining global and local searches, adjusting parameters through position mutation and dynamic weight adjustment, and finally realizing dynamic clustering; Step S4: Battery energy storage system load forecasting, realizing load forecasting for real-time battery energy storage system load data based on clustering results; In step S3, the dynamic clustering includes the following steps: Step S31: construct a sparse evaluation matrix; construct a multi-layer evaluation matrix, expressed as ; Sparse decomposition of the multi-layer evaluation matrix is ​​performed, expressed as: A≈ZW, and the constraints are expressed as: ; Get the sparse evaluation matrix A; where A k is the kth layer of the multi-layer evaluation matrix, each layer represents a different data dimension; is the element in the k-th layer evaluation matrix, representing the j-th indicator value of the i-th sample; Z is the simplified feature matrix; W is the weighting matrix; α is a constant that controls sparsity; is the L1 norm; m is the total number of samples; Step S32: Calculate the index deviation; the formula used is as follows: ； ； ； In the formula, R i,j is the degree of deviation of the jth indicator of the ith sample from the mean; S i,j is the original data value of the jth indicator of the i-th sample; u j and σ j are the mean and standard deviation of the jth indicator respectively; is a smoothing term; D j is the discrimination of the jth indicator; E j is the uncertainty of the jth indicator; Step S33: Calculate the final weight; initialize the weight matrix B; and perform consistency check, expressed as: ; The final weight is expressed as: ; Dynamic weights are introduced and used when weights are updated ; Where CI is the consistency of the weight matrix; is the maximum eigenvalue of the weight matrix; RI is the random consistency index; w j is the final weight; is the initial weight; α2 is the influencing parameter for controlling the difference; and are the final weights at the t+1th iteration and the tth iteration respectively; is the uncertainty at the tth iteration; is the weight factor; n is the total number of indicators; Step S34: clustering processing; specifically including: Step S341: data optimization; waveform conversion is performed on the data to obtain data representations of different scales, which are represented as follows: ;in, It is the original data; is the wavelet function; a i and b i are the scale parameter and the translation parameter; It is the data representation after waveform conversion; Step S342: Determine the optimal number of clusters; the formula used is as follows: ； In the formula, CH(·) is the optimal cluster number index; tr(·) is the trace; B k and W k are the scatter matrix between samples and the scatter matrix within samples respectively; n is the total number of indicators; Step S343: cluster iteration; randomly select the initial cluster center; calculate the distance from each non-cluster center to the cluster center, and assign the non-cluster center based on the minimum distance; iteratively update the cluster center and calculate the objective function; the distance calculation is expressed as: ; ; Introduce the multi-objective optimization method to define the objective function, expressed as: ; Progressive learning is introduced when updating the cluster center, and the cluster center update formula is expressed as: ; where d m (·) is the distance metric; x i and x j are the i-th data point and the j-th data point respectively; S is the covariance matrix; ρe(·) is the local adaptive density; ρ(·) is the local density; N(·) is the domain point set, i1 is the neighborhood point index; d(·) is the Euclidean distance; is the local density parameter; f(·) is the objective function; X and Y are the proportion of the label with the largest number of historical data in the cluster and the proportion of other labels, respectively; and are the positions of the j1th cluster at the t+1th and tth iterations respectively; is the learning rate parameter; is the iteration upper limit; is a smoothing term; , and is the weight coefficient; Step S35: Determination; when the objective function converges, clustering is completed; if only the iteration upper limit is reached, go to step S36; otherwise, continue iteration; Step S36: cluster optimization; In step S1, the historical battery energy storage system load data and the real-time battery energy storage system load data both include battery status, environmental factors, time data and external load requirements; the historical battery energy storage system load data also includes a load assessment level; the load assessment level includes normal, light load, moderate load and severe load; the load assessment level at the next moment is used as a data label; the data label does not participate in the clustering operation; In step S4, the battery energy storage system load prediction is based on the clustering result, taking the label with the largest number of historical battery energy storage system load data as the cluster label, and taking the cluster label to which the real-time battery energy storage system load data belongs as the prediction result of the real-time battery energy storage system load data.

2. The intelligent battery energy storage system load forecasting method according to claim 1, characterized in that: In step S36, the cluster optimization specifically includes: Step S361: construct a parameter optimization space based on the weight matrix, the influence parameter of the control difference, the local density parameter, the learning rate parameter, the scale parameter and the translation parameter; initialize the parameter optimization population, and use the objective function when the parameter optimization individual reaches the iteration upper limit or convergence as the individual fitness value; the formula used to initialize the optimization population is as follows: ； Where, X i2,j2 (·) is the position of the j2th dimension of the i2th optimization individual; 0 indicates the 0th search, i.e. initialization; rand is a random number from LB to UB; UB and LB are the upper and lower limits of the search in the j2th dimension, respectively; Step S362: Exploration phase; when it < itmax / 2, update the step size control, expressed as: ; ; The position update is expressed as: ; where, ; W1(·) and W2(·) are individual movement weights; i2 is the index; hy(·) is the individual demand value; SHhy is the sum of the demand values of all individuals; r, r1, r2 are all random numbers between 0 and 1; N is the number of individuals in the population; randn is a random number sampled from the standard normal distribution; X b,j2 (·) is the position of the optimal individual in the j2 dimension, and the optimal individual is the individual with the lowest fitness value; R is a random number generated in the range [-a, a], a = 2·(1 - it / itmax); Hs is the additional demand value of the individual, Hs = Lhs·(1 + r); Lhs is the lower bound demand value; Allf(·) is the individual fitness value; BF is the optimal individual fitness value; itmax is the maximum number of searches; Step S363: Utilization phase; when it ≥ it / 2, the location update is expressed as: ;in, ; RB is a random number generated from Gaussian distribution; is the average fitness value of the population; r3 is a random number between 0 and 1; Step S364: mutation; perform position mutation on the updated individual. If the fitness value of the position after mutation decreases, retain the mutation position, otherwise retain the original position. The formula used for mutation is as follows: ； In the formula, is the position after mutation; r4 and r5 are random numbers between 0 and 1; U is a binary vector; X roll1,j2 (·) and X roll2,j2 (·) is the position of the random individual in the j2th dimension; Step S365: optimization judgment; a fitness threshold is set in advance. When there is an individual fitness value lower than the fitness threshold, the optimization ends, the individual position, i.e., the parameter setting, is used to obtain the clustering result; if the maximum number of searches is reached, return to step S361; otherwise, return to step S362.

3. The intelligent battery energy storage system load forecasting method according to claim 1 is characterized in that: In step S2, the data preprocessing is to perform data cleaning, data conversion and standardization on the collected data; the data cleaning is to remove missing values ​​and duplicate values; the data conversion is to convert the data into a vector form; the standardization is to standardize the data based on the maximum and minimum normalization method.

4. An intelligent battery energy storage system load forecasting system, used to implement an intelligent battery energy storage system load forecasting method as described in any one of claims 1 to 3, characterized in that: It includes data acquisition module, data preprocessing module, dynamic clustering module and battery energy storage system load forecasting module; The data acquisition module collects historical battery energy storage system load data and real-time battery energy storage system load data, and sends the data to the data preprocessing module; The data preprocessing module performs data cleaning, data conversion and standardization on the collected data; And send the data to the dynamic clustering module; The dynamic clustering module constructs a multi-layer evaluation matrix, identifies key indicators based on indicator deviation, uses a weight matrix for consistency verification, and dynamically updates the final weight through initial weights, distinguishability, and uncertainty. After waveform conversion of the load data, clustering is performed based on data representations of different scales, the optimal number of clusters is determined, and cluster centers are iteratively updated. Finally, iterative updates are performed by introducing multi-objective optimization; and the global and local searches are combined by adaptively adjusting the step size, and parameters are adjusted through position mutation and dynamic weight adjustment, so as to finally achieve dynamic clustering; and the data is sent to the battery energy storage system load prediction module; The battery energy storage system load prediction module realizes load prediction on real-time battery energy storage system load data based on clustering results.

Images