Network construction energy storage locating and sizing method for supporting transient stability of high-proportion new energy power grid

Abstract

The invention discloses a network construction energy storage locating and sizing method for supporting transient stability of a high-proportion new energy power grid, and the method comprises the steps: obtaining power grid measurement data, extracting spatial-temporal features through segmented Hankel matrix recursive decomposition and dynamic weighted graph decomposition, and constructing a feature tensor through multi-scale tensor fusion; transient response features are analyzed based on hierarchical sparse principal component mapping, stability indexes are constructed through multi-scale coupling evaluation, and node scores are obtained through adaptive weight fusion; candidate nodes are screened through affinity propagation clustering, the supporting capacity is evaluated through multi-time-scale influence propagation, and capacity configuration is determined through fractional programming constraint optimization; real-time evaluation is realized based on recursive state estimation, a response strategy is generated through multi-time scale hierarchical decision, and cooperative control is realized by adopting distributed consistency optimization. The transient stability of the high-proportion new energy power grid is effectively improved.

Description

Technical Field

[0001] The present invention relates to a method for selecting a site and determining the capacity of a network energy storage, in particular to a method for selecting a site and determining the capacity of a network energy storage that supports transient stability of a high-proportion new energy power grid. Background Art

[0002] With the continuous increase in the proportion of installed capacity of new energy power generation such as wind power and photovoltaic power generation, the transient stability of the power system faces severe challenges. Due to the intermittent, volatile and random characteristics of new energy power generation, the system inertia continues to decline, and the frequency regulation ability is significantly weakened, which can easily cause system oscillation or even instability when a fault or disturbance occurs. Grid-type energy storage, with its advantages such as rapid response and two-way regulation, has become a key supporting technology for improving the transient stability of power grids with a high proportion of new energy. A reasonable energy storage site selection and sizing plan can not only effectively suppress the fluctuation of new energy output, but also provide the necessary transient support for the system in the event of a fault, which is of great significance to ensuring the safe and stable operation of the power grid.

[0003] At present, the research on energy storage site selection and capacity determination is mainly focused on economic evaluation and power balance. Traditional methods usually optimize the configuration based on economic indicators such as peak shaving and valley filling benefits and ancillary service revenue, or determine the energy storage capacity based on power balance constraints. In terms of transient support, existing research mainly adopts simplified energy storage response models and single time scale stability analysis methods to determine energy storage configuration plans through sensitivity analysis or heuristic algorithms. Although these methods have solved the energy storage planning problem to a certain extent, they have obvious limitations when dealing with transient stability issues of power grids with a high proportion of new energy.

[0004] In-depth analysis found that the current research still has the following technical difficulties: First, the transient feature extraction method is difficult to effectively handle the multi-time scale dynamic characteristics of renewable energy output, especially in the coupling analysis of millisecond transient processes and minute-level fluctuation processes; second, the existing stability assessment method fails to fully consider the dynamic change characteristics of the power grid topology, and it is difficult to accurately quantify the support effect of energy storage at different locations; third, the energy storage site selection optimization algorithm is prone to fall into local optimality when dealing with large-scale non-convex optimization problems, and fails to effectively balance multiple goals such as system stability, economy and engineering feasibility; fourth, the energy storage collaborative control strategy fails to achieve cross-time scale response characteristic optimization, and there is a trade-off problem between rapid transient support and continuous regulation; fifth, the existing methods lack the ability to accurately estimate and predict the system state, and it is difficult to support the real-time optimization control of the energy storage system. These technical problems seriously restrict the application effect of grid-type energy storage in power grids with a high proportion of renewable energy. Summary of the invention

[0005] The purpose of the invention is to provide a method for selecting the site and sizing the energy storage for network construction that supports the transient stability of a high-proportion new energy power grid, in order to solve at least one technical problem existing in the prior art.

[0006] The technical solution supports the transient stability of the high-proportion new energy power grid and the method of selecting the location and sizing of energy storage, including the following steps:

[0007] Step S1, obtaining grid node voltage data, node frequency data, node power data and new energy output data, composing these data into an original grid measurement data matrix, processing them according to the segmented Hankel matrix recursion method and the dynamic weighted spectrum method to obtain transient feature data; extracting the time feature matrix and the space feature matrix from the transient feature data, processing them according to the multi-scale tensor fusion method to obtain the feature tensor and the core tensor;

[0008] Step S2, obtaining the feature tensor and core tensor obtained in step S1, processing them according to the hierarchical sparse principal component mapping method, and obtaining a response feature matrix; based on the response feature matrix, various stability indicators are calculated by a multi-scale coupling evaluation method; according to these stability indicators and historical fault data, a conditional Wasserstein distance mapping method is used to obtain a probability feature matrix; the stability indicators and the probability feature matrix are processed by an adaptive weight fusion method to obtain a node stability score vector;

[0009] Step S3, obtaining the node stability score vector, power grid topology data and new energy distribution data obtained in step S2, and processing them according to the affinity propagation clustering method to obtain a candidate node set and a node weight vector; processing the candidate node set and the spatiotemporal feature tensor obtained in step S1 by a multi-time scale influence propagation method to obtain a node support capacity matrix; according to the node support capacity matrix and the system cost constraint, using the fractional programming constraint optimization method to obtain the optimal capacity configuration plan; processing the optimal capacity configuration plan with historical fault data and system parameters by a robustness probability evaluation method to obtain a verification report;

[0010] Step S4, obtain the real-time measurement data of the system, the optimal capacity configuration plan obtained in step S3 and the spatiotemporal feature tensor obtained in step S1, process them according to the recursive state estimation and prediction method, and obtain the system state estimation and predicted state; according to the system state estimation, predicted state and energy storage device parameters, adopt a multi-time scale hierarchical decision method to obtain the energy storage response strategy; process the energy storage response strategy, energy storage device state set and network topology information through a distributed consistency optimization method to obtain optimized control instructions; according to the optimized control instructions, actual response data and historical performance indicators, adopt an online learning and dynamic compensation method to obtain the adjusted control strategy and performance evaluation report.

[0011] Beneficial effect, effectively improving the transient stability of high-proportion new energy power grids. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It is a flow chart of the present invention.

[0013] Figure 2 It is a flow chart of step S1 of the present invention.

[0014] Figure 3 It is a flow chart of step S2 of the present invention.

[0015] Figure 4 It is a flow chart of step S3 of the present invention.

[0016] Figure 5 It is a flow chart of step S4 of the present invention. DETAILED DESCRIPTION

[0017] like Figure 1 As shown, according to one aspect of the present application, a method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid is provided, which includes the following steps:

[0018] Step S1, obtaining grid node voltage data, node frequency data, node power data and new energy output data, composing these data into an original grid measurement data matrix, processing them according to the segmented Hankel matrix recursion method and the dynamic weighted spectrum method to obtain transient feature data; extracting the time feature matrix and the space feature matrix from the transient feature data, processing them according to the multi-scale tensor fusion method to obtain the feature tensor and the core tensor;

[0019] Step S2, obtaining the feature tensor and core tensor obtained in step S1, processing them according to the hierarchical sparse principal component mapping method, and obtaining a response feature matrix; based on the response feature matrix, various stability indicators are calculated by a multi-scale coupling evaluation method; according to these stability indicators and historical fault data, a conditional Wasserstein distance mapping method is used to obtain a probability feature matrix; the stability indicators and the probability feature matrix are processed by an adaptive weight fusion method to obtain a node stability score vector;

[0020] Step S3, obtaining the node stability score vector, power grid topology data and new energy distribution data obtained in step S2, and processing them according to the affinity propagation clustering method to obtain a candidate node set and a node weight vector; processing the candidate node set and the spatiotemporal feature tensor obtained in step S1 by a multi-time scale influence propagation method to obtain a node support capacity matrix; according to the node support capacity matrix and the system cost constraint, using the fractional programming constraint optimization method to obtain the optimal capacity configuration plan; processing the optimal capacity configuration plan with historical fault data and system parameters by a robustness probability evaluation method to obtain a verification report;

[0021] Step S4, obtain the real-time measurement data of the system, the optimal capacity configuration plan obtained in step S3 and the spatiotemporal feature tensor obtained in step S1, process them according to the recursive state estimation and prediction method, and obtain the system state estimation and predicted state; according to the system state estimation, predicted state and energy storage device parameters, adopt a multi-time scale hierarchical decision method to obtain the energy storage response strategy; process the energy storage response strategy, energy storage device state set and network topology information through a distributed consistency optimization method to obtain optimized control instructions; according to the optimized control instructions, actual response data and historical performance indicators, adopt an online learning and dynamic compensation method to obtain the adjusted control strategy and performance evaluation report.

[0022] By building a complete "feature extraction-stability assessment-site selection optimization-coordinated control" technical system, the scientific configuration of network energy storage supporting the transient stability of high-proportion new energy power grids is realized. At the feature extraction level, the temporal and spatial characteristics of the system are accurately captured through multi-dimensional grid transient feature extraction and preprocessing; at the stability assessment level, the precise quantification of system transient characteristics is achieved through multi-scale coupled stability assessment; at the site selection optimization level, the optimal energy storage configuration scheme is determined through affinity propagation clustering and fractional programming constraint optimization; at the coordinated control level, the efficient coordinated operation of the energy storage system is achieved through multi-time scale hierarchical decision-making and distributed consistency optimization. This systematic technical solution is particularly suitable for solving the transient stability problems faced by high-proportion new energy power grids: first, through accurate feature extraction and stability assessment, the impact of new energy fluctuations on the system is scientifically quantified; second, through optimized site selection and capacity determination, the efficient use of energy storage resources is ensured; finally, through coordinated control strategies, the effective suppression of new energy fluctuations is achieved. The overall solution has significant advantages in improving system stability, enhancing new energy absorption capacity, and optimizing energy storage configuration efficiency, providing reliable technical support for the construction of new power systems.

[0023] According to one aspect of the present application, step S1 specifically comprises:

[0024] Step S11, obtaining grid node voltage data, grid node frequency data, grid node power data and new energy output data, and arranging these data into an original grid measurement data matrix according to time series; calculating the quantile value of each data item according to the quantile distribution characteristics of each data item in the data matrix; normalizing the original data based on the calculated quantile value to obtain a standardized data matrix;

[0025] Step S12, obtaining the standardized data matrix obtained in step S11, setting the time window length and sliding step length, and constructing a sliding Hankel matrix; performing singular value decomposition on the Hankel matrix in each time window, extracting the main eigenvalues ​​and eigenvectors; recursively reconstructing the data according to the changing rules of the eigenvalues ​​and eigenvectors, and obtaining a time feature matrix;

[0026] Step S13, obtaining the adjacency matrix data of the power grid topology structure and the time characteristic matrix obtained in step S12, and calculating the weight coefficient of each node; constructing a weighted Laplace matrix according to the weight coefficient and the adjacency matrix; performing eigenvalue decomposition on the weighted Laplace matrix, extracting the main eigenvectors, and obtaining a spatial characteristic matrix;

[0027] Step S14, obtain the time feature matrix obtained in step S12 and the space feature matrix obtained in step S13, combine the two feature matrices according to the node dimension, and construct a third-order tensor; perform Tucker decomposition on the third-order tensor to obtain basis matrices in three directions respectively; and calculate the feature tensor and the core tensor according to the product relationship of the basis matrices.

[0028] Through a systematic method of multi-dimensional grid transient feature extraction and preprocessing, the complex spatiotemporal features of high-proportion renewable energy grids are accurately captured. The method first processes the original data through adaptive quantile normalization, effectively eliminating the influence of different dimensions and scales; then the segmented Hankel matrix recursive decomposition method is used to process the time series data, which not only maintains the temporal characteristics of the data, but also significantly reduces the computational complexity; then the dynamic weighted graph decomposition method is used to extract spatial features, accurately reflecting the dynamic change characteristics of the network topology; finally, the multi-scale tensor fusion method is used to achieve an organic combination of spatiotemporal features, so that the extracted features contain both the transient dynamic characteristics of the system and the spatial topological structure information. This systematic feature extraction method provides a comprehensive and accurate data basis for the subsequent energy storage site selection and capacity determination, especially when dealing with complex scenarios such as renewable energy output fluctuations, load changes, and dynamic changes in network topology, it shows excellent adaptability and reliability.

[0029] According to one aspect of the present application, step S2 is specifically:

[0030] Step S21, obtaining the feature tensor and core tensor obtained in step S1, constructing a multi-layer mapping matrix, wherein the dimension of each layer of the mapping matrix is ​​determined according to preset parameters; transforming the feature tensor data through each layer of the mapping matrix in turn to obtain intermediate features; performing principal component analysis on the intermediate features to extract the main direction vectors and corresponding weight coefficients; reconstructing the data according to the direction vectors and weight coefficients to obtain a response feature matrix;

[0031] Step S22, obtaining the response characteristic matrix obtained in step S21, calculating the characteristic deviation value of each node at different time scales; performing weighted summation on the characteristic deviation values ​​according to the preset time weight coefficient to obtain a transient stability index; calculating the time derivative of the response characteristic, and integrating it in the time domain to obtain a dynamic stability index; calculating the covariance of the response characteristics between the node pairs, and multiplying it by the coupling coefficient to obtain a collaborative stability index; combining these three types of indicators into a stability index set;

[0032] Step S23, obtaining the stability index set and the system historical fault data obtained in step S22, and performing conditional probability density estimation on the stability index according to the fault data; calculating the Wasserstein distance of the probability distribution of the stability index between different nodes; forming a distance matrix with the calculated distance values; performing dimensionality reduction processing on the distance matrix to obtain a probability feature matrix;

[0033] Step S24, obtain the stability index set obtained in step S22 and the probability feature matrix obtained in step S23, construct an adaptive weight function, and calculate the dynamic weight coefficient of each index; multiply the stability index and probability feature by the corresponding weight coefficient respectively; sum the weighted index and feature to obtain the comprehensive score of each node; and form a node stability score vector with the scores of all nodes.

[0034] By constructing a multi-dimensional transient stability assessment and quantification system, a systematic analysis of the stability characteristics of a high-proportion renewable energy power grid is achieved. This method first extracts transient response characteristics through hierarchical sparse principal component mapping, thereby accurately characterizing the dynamic characteristics of the system; the mapping hierarchy based on the Wasserstein distance and the geodesic flow feature transformation enable the feature extraction process to accurately capture the random fluctuation characteristics of renewable energy output. Secondly, a stability index system is constructed through a multi-scale coupled stability assessment method, in which the stability functional of the Lyapunov exponent and the dynamic evolution model of the Koopman operator are combined to accurately describe the nonlinear dynamic characteristics of the system; through sliding Prony analysis and synchronous phase coupling evaluation, accurate identification of the system's dominant mode and node coupling relationship is achieved. Thirdly, the conditional Wasserstein distance mapping method is used to process probabilistic features, and the accurate quantification of system stability is achieved through information fusion of von Neumann entropy and sensitivity analysis of Malliavin derivative. Finally, various indicators are integrated through an adaptive weight fusion method, in which The multi-dimensional correlation model of the transformed dynamic weight function and the Copula function ensures the scientificity and reliability of the evaluation results. This multi-level stability evaluation system can not only fully reflect the impact of new energy access on system stability, but also provide a reliable quantitative basis for energy storage site selection and capacity determination.

[0035] According to one aspect of the present application, step S3 is specifically:

[0036] Step S31, obtaining the node stability score vector, power grid topology data and new energy distribution data obtained in step S2, calculating the characteristic distance and topological distance between nodes; constructing a node affinity matrix based on the characteristic distance and topological distance; calculating the responsibility matrix between nodes, in which each element represents the suitability of a node as a cluster center; calculating the availability matrix between nodes, in which each element represents the suitability of a node to select other nodes as cluster centers; iterating and updating the responsibility matrix and the availability matrix until convergence to obtain a set of candidate nodes; calculating the influence range and importance of each candidate node to obtain a node weight vector;

[0037] Step S32, obtain the candidate node set obtained in step S31 and the spatiotemporal feature tensor obtained in step S1, set multiple feature time scales; construct a normalized Laplace matrix, calculate the basic influence matrix; for each time scale, calculate the influence propagation matrix; multiply the influence propagation matrix at different time scales with the corresponding time weight; calculate the cumulative support effect, and obtain the support capacity value of each node at different time scales; combine these support capacity values ​​into a node support capacity matrix;

[0038] Step S33, obtaining the candidate node set obtained in step S31, the node support capacity matrix obtained in step S32, and the system cost constraint data, constructing a fractional objective function, wherein the numerator is the support effect and the denominator is the cost function; setting upper and lower limit constraints of energy storage capacity, total cost constraints, and response time constraints; converting the fractional programming problem into a parameterized quadratic programming problem; iteratively solving the parameterized quadratic programming problem, updating the parameters until convergence, and obtaining the optimal capacity configuration solution;

[0039] Step S34, obtaining the optimal capacity configuration scheme, system historical fault data and system parameter data obtained in step S33, generating multiple groups of disturbance scenarios according to the historical fault data and system parameters; calculating the system performance index for each group of disturbance scenarios; counting the probability that the performance index is less than the threshold in all scenarios, and obtaining the robustness index of the scheme; calculating the expected value of the degree of constraint violation, and obtaining the feasibility index of the scheme; combining the robustness index and the feasibility index into a verification report.

[0040] By constructing a multi-objective optimization configuration method for networked energy storage, the optimal configuration of energy storage resources is achieved. The method first screens candidate nodes through affinity propagation clustering evaluation, in which the characteristic distance calculation of Optimal Transport and the topological feature extraction of Persistent Homology accurately identify the key nodes in the system; the critical point detection of the MessagePassing framework and Morse theory ensures the scientific nature of node screening. Secondly, the multi-time scale influence propagation method is used to evaluate the transient support capacity, in which the multi-scale stratification of Folias theory and the dynamic distance measurement of Finsler geometry accurately describe the spatiotemporal distribution of the energy storage support effect; the Whitney embedding and Lie bracket evaluation are used to achieve accurate quantification of the energy storage control efficiency. Finally, the optimal capacity configuration is determined by the fractional programming constrained optimization method, in which the Pontryagin dual objective function and the Maslov form constraint processing effectively balance the economic and technical feasibility; the Newton-YY method and the solution refinement of the Bregman divergence ensure the convergence and stability of the optimization results. This systematic optimization configuration method can not only achieve the economy of energy storage investment while meeting the system stability requirements, but also ensure the engineering feasibility of the configuration plan through robust probabilistic evaluation.

[0041] According to one aspect of the present application, step S4 is specifically:

[0042] Step S41, obtain the real-time measurement data of the system, the optimal capacity configuration plan obtained in step S3 and the spatiotemporal characteristic tensor obtained in step S1, construct the state observation equation, calculate the deviation between the measured value and the predicted value; calculate the gain matrix according to the measurement noise covariance; multiply the gain matrix by the measurement deviation to update the state estimate; calculate the predicted state according to the updated state estimate and the control input; add the model uncertainty compensation term to the predicted state to obtain the system state estimate and the predicted state;

[0043] Step S42, obtaining the system state estimation, predicted state and energy storage device parameter data obtained in step S41, calculating the millisecond-level rapid response control quantity, including the state deviation term and the compensation term; solving the medium-term regulation optimization problem at the time scale of seconds to obtain the medium-term control sequence; solving the long-term optimization problem at the time scale of minutes to obtain the long-term control sequence; calculating the dynamic weight coefficients of the control quantities at the three time scales; combining the control quantities at the three time scales according to the dynamic weight coefficients to obtain the energy storage response strategy;

[0044] Step S43, obtaining the energy storage response strategy, energy storage device state set and network topology information obtained in step S42, constructing the local objective function of each energy storage device, including the tracking error term and the state constraint term; calculating the local gradient and dual variable of each energy storage device; exchanging the state information of adjacent energy storage devices according to the network topology information; updating the control amount and dual variable of each energy storage device; repeating the iterative calculation until convergence, and obtaining the optimized control instruction;

[0045] Step S44, obtain the optimized control instruction, the actual system response data and the historical performance index data obtained in step S43, calculate the control instruction tracking error and the energy storage loss, and obtain the performance evaluation index; calculate the gradient of the control parameter according to the performance evaluation index; update the control parameter to obtain a new parameter set; generate a dynamic compensation amount based on the updated parameter set and the historical performance index; combine the optimized control instruction with the dynamic compensation amount to obtain an adjusted control strategy; combine the performance evaluation index, the parameter update result and the compensation effect into a performance evaluation report.

[0046] By constructing a system state-based energy storage collaborative control method, the efficient operation of the energy storage system is achieved. First, the method realizes the real-time evaluation of the system state through recursive state estimation and prediction. The noise preprocessing of Martingale decomposition and the state transition model of Fokker-Planck equation accurately capture the random dynamic characteristics of the system; the error dynamics of Symplectic geometry and the predictor of Stratonovich integral ensure the accuracy and reliability of state estimation. Secondly, the multi-time scale hierarchical decision method is used to generate the energy storage response strategy. The time scale decomposition of Cartan moving frame and the fast response control of Beltrami flow realize the precise control of disturbances of different time scales; the economy and life characteristics of energy storage are optimized through the long-term planning of Hamilton-Jacobi equation and Exponential family. Finally, the collaborative control of energy storage equipment is realized through the distributed consistency optimization method. The complex variable objective function of Wirtinger derivative and the constraint decomposition of Frank-Wolfe flow handle the complex operation constraints; the global optimality of the control strategy is ensured through the refinement optimization of Krasnoselskii-Mann iteration and Mirror descent. This hierarchical collaborative control method not only enables the energy storage system to quickly respond to and continuously adjust to fluctuations in renewable energy, but also significantly improves the overall stability of the system through the coordination of multiple energy storage devices.

[0047] According to one aspect of the present application, step S12 is specifically:

[0048] S121, obtaining the standardized data matrix obtained in step S11, calculating the optimal embedding dimension and time delay parameter according to the time correlation of the data; segmenting the data matrix, constructing a submatrix for each segment according to the calculated parameters; rearranging all submatrices in time order to obtain an initial Hankel matrix;

[0049] S122, obtaining the initial Hankel matrix obtained in step S121, calculating the condition number and rank of the matrix; constructing a regularization parameter sequence according to the condition number; for each regularization parameter, calculating the corresponding matrix completion term; selecting the optimal regularization parameter, and obtaining an enhanced Hankel matrix;

[0050] S123, obtaining the enhanced Hankel matrix obtained in step S122, constructing an iterative decomposition framework based on the Krylov subspace; calculating the eigenvalue basis vector of each iteration; orthogonalizing the eigenvalue basis vector; repeating the iteration until convergence, and obtaining the principal component matrix;

[0051] S124, obtaining the principal component matrix obtained in step S123, establishing a recursive prediction model; calculating the confidence interval of the model parameters; screening reliable eigenvectors according to the confidence interval; combining the screened eigenvectors to obtain a reduced-dimensional feature matrix;

[0052] S125. Obtain the reduced dimension feature matrix obtained in step S124 and the initial Hankel matrix obtained in step S121, and construct a reconstruction error function; calculate the local reconstruction error of each time window; determine the optimal reconstruction parameters according to the error distribution; apply the reconstruction parameters to reconstruct the data and obtain the time feature matrix.

[0053] By introducing the recursive decomposition method of the segmented Hankel matrix, efficient processing of power grid time series data is achieved. First, this method ensures that the reconstructed phase space can retain the system dynamic characteristics to the greatest extent through adaptive calculation of the optimal embedding dimension and time delay parameters; secondly, the regularization processing based on the condition number is adopted to effectively solve the ill-conditioned problem of the traditional Hankel matrix when processing high-dimensional data; thirdly, the introduction of the Krylov subspace iterative decomposition framework significantly reduces the computational complexity of feature extraction while ensuring the accuracy of the results; finally, through the introduction of the recursive prediction model and confidence interval, a quantitative evaluation of the reliability of the features is achieved, thereby ensuring the effectiveness of the extracted features. This hierarchical and progressive processing method not only improves the computational efficiency, but also enhances the reliability of the results. It is particularly suitable for processing fast transient processes caused by the high proportion of new energy access.

[0054] According to one aspect of the present application, step S13 is specifically:

[0055] S131, obtaining the adjacency matrix data of the power grid topology structure and the time characteristic matrix obtained in step S12, calculating the initial weight coefficient according to the degree distribution of the node; calculating the sliding variance matrix for the time characteristic matrix; mapping the elements of the variance matrix to the weight space to obtain the dynamic weight coefficient matrix;

[0056] S132, obtaining the dynamic weight coefficient matrix and adjacency matrix obtained in step S131, constructing a distance metric function based on the Riemann manifold; calculating the geodesic distance between nodes; constructing a distance matrix according to the geodesic distance; adaptively smoothing the distance matrix to obtain a weighted distance matrix;

[0057] S133, obtaining the weighted distance matrix obtained in step S132, constructing a K nearest neighbor relationship graph; calculating the local connectivity index of the graph; adjusting the weights of the edges according to the connectivity index; combining the adjusted weights with the original adjacency relationship to obtain an enhanced adjacency matrix;

[0058] S134, obtaining the enhanced adjacency matrix obtained in step S133, calculating the generalized degree matrix; constructing a clustering structure based on a stochastic block model; extracting connection features between blocks and within blocks; combining the connection features to obtain a multi-scale topological matrix;

[0059] S135, obtaining the multi-scale topological matrix obtained in step S134, constructing a spectral decomposition optimization model based on Schur complement; iteratively calculating eigenvalues ​​and eigenvectors; determining a truncation threshold according to the eigenvalue distribution; extracting the main eigenvectors to obtain a spatial feature matrix.

[0060] By introducing the dynamic weighted graph decomposition method, the topological features of the power grid are accurately extracted. This method first constructs a distance measurement function based on the Riemann manifold, so that the geodesic distance between nodes can accurately reflect the physical connection characteristics of the system; secondly, by introducing the K nearest neighbor relationship graph and the local connectivity index, the description ability of the local topological structure is enhanced; thirdly, the Stochastic Block Model is used for clustering structure identification, which effectively captures the network module characteristics at different scales; finally, through the spectral decomposition optimization based on Schur complement, the efficient extraction of topological features is achieved. This multi-level topological feature extraction method can not only accurately reflect the static structural characteristics of the power grid, but also effectively capture the dynamic topological changes caused by the access of new energy, providing a reliable spatial feature basis for energy storage site selection.

[0061] According to one aspect of the present application, step S14 is specifically:

[0062] S141, obtaining the time feature matrix obtained in step S12 and the space feature matrix obtained in step S13, constructing a feature correlation matrix; calculating the mutual information between features; screening significant features according to the mutual information value; standardizing the significant features to obtain a preprocessing feature matrix;

[0063] S142, obtaining the preprocessing feature matrix obtained in step S141, constructing a feature mapping function based on the Grassmann manifold; calculating the principal curvature of the feature subspace; adjusting the feature weights according to the curvature information; combining the weighted features to obtain an enhanced feature matrix;

[0064] S143, obtaining the enhanced feature matrix obtained in step S142, constructing a multilinear tensor space; calculating the basis vectors of the tensor space; orthogonalizing the basis vectors; mapping the features to the tensor space to obtain an initial feature tensor;

[0065] S144, obtaining the initial feature tensor obtained in step S143, constructing a multidimensional optimization model based on Tucker decomposition; calculating the rank of the core tensor; determining the compression rate according to the distribution of the rank; compressing the tensor to obtain a compressed feature tensor;

[0066] S145. Obtain the compressed feature tensor obtained in step S144, and construct a reconstruction model based on TT decomposition (Tensor-train); calculate the reconstruction error of the model; adjust the reconstruction parameters according to the error distribution; reconstruct the data to obtain the final feature tensor and core tensor.

[0067] By introducing the multi-scale tensor fusion method, the organic unity of temporal and spatial features is achieved. This method first performs feature screening based on feature correlation matrix and mutual information calculation, effectively reducing data redundancy; secondly, by introducing the feature mapping function of Grassmann manifold, the geometric association between different feature subspaces is established, so that the feature fusion process can maintain the intrinsic structure of the original data; thirdly, the combination method of multilinear tensor space and Tucker decomposition is adopted to achieve dimensional compression while maintaining data integrity, significantly improving the computational efficiency of subsequent processing; finally, through the application of Tensor-train reconstruction model, the retention of key information in the compression process is ensured. This systematic feature fusion method not only solves the dimensional disaster problem of traditional methods when processing high-dimensional spatiotemporal data, but also ensures the physical interpretability of the fused features, which is particularly suitable for describing the rapidly changing transient processes and complex spatial coupling relationships in new energy power grids.

[0068] According to one aspect of the present application, step S21 is specifically:

[0069] S211, obtaining the feature tensor and core tensor obtained in step S1, constructing a feature projection operator based on the matrix manifold; calculating the projection coefficients of the feature tensor in different directions; constructing a directional sensitivity matrix according to the projection coefficients; normalizing the sensitivity matrix to obtain a directional feature matrix;

[0070] S212, obtaining the directional feature matrix obtained in step S211, constructing a mapping hierarchy structure based on Wasserstein distance; calculating the transmission distance between each layer; optimizing the inter-layer connection weights according to the transmission distance; combining the optimized inter-layer weights to obtain a multi-layer mapping matrix;

[0071] S213, obtaining the multi-layer mapping matrix obtained in step S212, constructing a feature transformation function based on geodesic flow; calculating the evolution trajectory of the feature on the manifold; extracting the key points of the trajectory; combining the key points into a transformation sequence to obtain an intermediate feature sequence;

[0072] S214, obtaining the intermediate feature sequence obtained in step S213, constructing a sparse constraint based on the Frobenius norm; calculating a sparse representation of the feature sequence; selecting the main components according to the sparsity; orthogonalizing the main components to obtain a main component feature matrix;

[0073] S215, obtaining the principal component feature matrix obtained in step S214, constructing an importance evaluation model based on Fisher information; calculating the Fisher score of the feature; sorting the features according to the score; selecting high-scoring features to reconstruct the data, and obtaining a response feature matrix.

[0074] By introducing the hierarchical sparse principal component mapping method, the transient response characteristics are accurately characterized. This method first realizes directional sensitivity analysis based on the feature projection operator of the matrix manifold, accurately capturing the response characteristics of the system to different disturbances; secondly, the transmission relationship between features is established through the mapping hierarchy of the Wasserstein distance, which effectively describes the evolution of the system state; thirdly, the feature transformation function of the geodesic flow is used to achieve accurate characterization of the nonlinear transient process; then, the main response mode is extracted through the sparse constraint of the Frobenius norm, reducing the redundancy of the features; finally, the importance evaluation based on Fisher information ensures the scientific nature of feature selection. This multi-level feature analysis method can not only accurately describe the rapid fluctuation characteristics of renewable energy power generation, but also effectively identify the key response modes of the system, providing a reliable theoretical basis for the formulation of energy storage configuration plans.

[0075] According to one aspect of the present application, step S22 is specifically:

[0076] S221, obtaining the response characteristic matrix obtained in step S21, constructing a stability functional based on the Lyapunov exponent; calculating the local Lyapunov exponent of the system; constructing a stability domain according to the exponential distribution; segmenting the stability domain to obtain a segmented stability index;

[0077] S222, obtaining the segmented stability index obtained in step S221, constructing a dynamic evolution model based on the Koopman operator; calculating the spectral decomposition sequence of the system; extracting the characteristic pattern of the spectral sequence; combining the characteristic patterns to obtain dynamic stability characteristics;

[0078] S223, obtaining the dynamic stability characteristics obtained in step S222, constructing an integral optimizer based on sliding Prony analysis; calculating the complex frequency components of the characteristics; determining the integral interval according to the frequency components; integrating the interval data to obtain a time integral index;

[0079] S224, obtaining the time integral index obtained in step S223, constructing a coupling evaluation function based on the synchronous phase; calculating the phase correlation between nodes; constructing a coupling network according to the correlation; clustering the coupling network to obtain a coupling area index;

[0080] S225. Obtain the segmented stability index obtained in step S221, the time integral index obtained in step S223, and the coupling area index obtained in step S224, and construct an index fusion model based on the entropy weight method; calculate the information entropy of each index; determine the combination weight according to the entropy value; combine the indicators according to the weights to obtain a stability index set.

[0081] By introducing a multi-scale coupled stability assessment method, a comprehensive quantification of the transient stability of the system is achieved. This method first constructs a stability functional based on the Lyapunov exponent to accurately describe the local stability characteristics of the system; secondly, the dynamic evolution model of the Koopman operator is used to capture the nonlinear dynamic characteristics of the system, making the stability assessment more accurate; thirdly, the sliding Prony analysis is used for integral optimization to effectively extract the dominant mode and key frequency components of the system; then, the dynamic interaction between nodes is analyzed through the coupled evaluation function of the synchronous phase; finally, the scientific fusion of multiple stability indicators is achieved based on the entropy weight method. This systematic stability assessment method can not only accurately reflect the impact of new energy access on system stability, but also quantitatively evaluate the stabilization effect of energy storage configurations at different locations, providing a reliable stability assessment basis for energy storage site selection and capacity determination.

[0082] According to one aspect of the present application, step S24 is specifically:

[0083] S241, obtain the stability index set obtained in step S22 and the probability feature matrix obtained in step S23, and construct a The dynamic weight function of the transformation is used; the time-varying eigenvalues ​​of the indicators are calculated; a nonlinear mapping is constructed based on the eigenvalues; the mapping results are normalized to obtain the basic weight sequence.

[0084] S242, obtaining the basic weight sequence obtained in step S241, constructing a weight optimization model based on the Multinomial Dirichlet process; calculating the posterior distribution of weights; extracting optimal weight samples according to distribution characteristics; smoothing the samples to obtain an optimized weight matrix.

[0085] S243, obtaining the optimized weight matrix obtained in step S242, constructing an information fusion function based on von Neumann entropy; calculating the mutual information between indicators; adjusting the weight coefficients according to the mutual information; combining the adjusted weights to obtain an adaptive weight vector.

[0086] S244, obtaining the adaptive weight vector and original index data obtained in step S243, constructing a sensitivity analysis model based on Malliavin derivative; calculating the influence of the index on the system response; modifying the weight according to the influence; and obtaining a modified weight matrix.

[0087] S245. Obtain the modified weight matrix obtained in step S244, and construct a multidimensional correlation model based on the Copula function; calculate the conditional correlation coefficient between the indicators; optimize the combination method according to the correlation coefficient; combine the indicators to obtain a node stability score vector.

[0088] By introducing the adaptive weight fusion method, an accurate assessment of the comprehensive stability of the system is achieved. The dynamic weight function is constructed by transformation, which accurately captures the nonlinear change characteristics of the indicator weights; secondly, the weight distribution is optimized through the Multinomial Dirichlet process, making the weight distribution more statistically significant; thirdly, the information fusion function of von Neumann entropy is used to evaluate the information redundancy between indicators and improve the reliability of the score; then, the sensitivity analysis is performed through the Malliavin derivative to quantify the influence of each indicator on the system response; finally, a multidimensional correlation model is constructed based on the Copula function to accurately describe the complex dependency relationship between indicators. This systematic scoring method can not only accurately reflect the comprehensive impact of new energy fluctuations on system stability, but also scientifically evaluate the overall effect of different energy storage configuration schemes, providing a reliable quantitative basis for the final site selection and capacity determination decision.

[0089] According to one aspect of the present application, step S31 is specifically:

[0090] S311. Obtain the node stability score vector, power grid topology data and new energy distribution data obtained in step S2, and construct a feature distance calculator based on Optimal Transport; calculate the transmission distance of the node feature; construct a similarity matrix based on the distance; standardize the similarity matrix to obtain a node similarity vector.

[0091] S312, obtaining the node similarity vector obtained in step S311, constructing a topological feature extractor based on Persistent Homology; calculating a simple complex sequence; extracting a persistence graph; analyzing the persistence features, and obtaining a topological feature matrix.

[0092] S313, obtaining the topological feature matrix obtained in step S312, constructing an affinity propagation model based on Message Passing; calculating the responsibility between nodes; updating the availability of nodes; iterating and optimizing until convergence, and obtaining a cluster center matrix.

[0093] S314, obtaining the cluster center matrix obtained in step S313, constructing a critical point detector based on Morse theory; calculating the Morse index of the node; identifying stable and unstable manifolds; analyzing the intersection points of the manifolds to obtain a set of key nodes.

[0094] S315, obtaining the key node set obtained in step S314, constructing an influence assessment model based on the Perron-Frobenius operator; calculating the main eigenvector of the node; analyzing the convergence of the eigenvector; determining the influence range according to the convergence characteristics, and obtaining the candidate node set and the node weight vector.

[0095] By introducing the affinity propagation clustering evaluation method, scientific screening of candidate energy storage nodes is achieved. This method first constructs a feature distance calculator based on the Optimal Transport theory to accurately quantify the dynamic similarity between nodes; secondly, the topological features are extracted through the Persistent Homology method to effectively identify the persistent structure and key connection mode in the system; thirdly, the Message Passing framework is used to realize the dynamic propagation of affinity, making the node classification more accurate; then, critical point detection is performed through the Morse theory to accurately identify the stable points and unstable manifolds in the system; finally, the node influence is evaluated based on the Perron-Frobenius operator, and the system importance of each node is scientifically quantified. This systematic node screening method can not only accurately identify the key nodes around the new energy access point, but also effectively evaluate the contribution of each node to the system stability, providing a scientific set of candidate locations for energy storage site selection, which is particularly suitable for dealing with changes in system dynamic characteristics caused by a high proportion of new energy access.

[0096] According to one aspect of the present application, step S32 is specifically:

[0097] S321. Obtain the candidate node set obtained in step S31 and the spatiotemporal feature tensor obtained in step S1, and construct a multi-scale classifier based on Folianations theory; calculate the characteristic leaf layers of the system; extract the cross-sectional structure between the leaf layers; parameterize the cross-sectional structure to obtain a time scale decomposition matrix.

[0098] S322. Obtain the time scale decomposition matrix obtained in step S321, construct a dynamic distance metric based on Finsler geometry; calculate the asymmetric distance between nodes; construct an influence propagation path based on the distance; weight the propagation path to obtain a basic influence matrix.

[0099] S323. Obtain the basic influence matrix obtained in step S322, and construct a network transmission model based on Kirchhoff flow; calculate the maximum flow capacity between nodes; analyze the flow distribution characteristics; construct a transmission efficiency matrix based on the flow characteristics, and obtain a transmission capacity indicator.

[0100] S324, obtaining the transmission capacity index obtained in step S323, constructing a state space mapper based on Whitney embedding; calculating the controllability matrix of the system; analyzing the controllability ellipsoid; extracting the main axis direction of the ellipsoid, and obtaining the control effectiveness matrix.

[0101] S325. Obtain the control effectiveness matrix and transmission capacity index obtained in step S324, construct a dynamic coupling evaluator based on Lie brackets; calculate the coupling strength of the system; analyze the time-varying characteristics of the coupling; combine the evaluation results to obtain the node support capacity matrix.

[0102] By introducing a multi-time scale influence propagation method, an accurate assessment of the transient support capacity of candidate nodes is achieved. This method first constructs a multi-scale stratifier based on the Folies theory to accurately separate the fast and slow dynamic characteristics of the system; secondly, a dynamic distance metric is established through Finsler geometry to effectively describe the asymmetric influence propagation characteristics between nodes; thirdly, the Kirchhoff flow model is used to analyze the network transmission capacity and quantify the spatial distribution of the energy storage support effect; then, the scientific mapping of the state space is achieved through Whitney embedding to accurately evaluate the control efficiency of energy storage; finally, the dynamic coupling characteristics are evaluated based on the Lie bracket, and the spatiotemporal coupling effect of the energy storage support is systematically analyzed. This multi-dimensional support capacity assessment method can not only accurately reflect the suppression effect of energy storage on new energy fluctuations, but also scientifically evaluate its contribution to the transient stability of the system, providing reliable technical support for the optimal configuration of energy storage capacity.

[0103] According to one aspect of the present application, step S33 is specifically:

[0104] S331. Obtain the candidate node set obtained in step S31, the node support capacity matrix obtained in step S32, and the system cost constraint data, and construct an objective function generator based on Pontryagin duality; calculate the Hamiltonian of the system; analyze the critical trajectory; construct optimality conditions, and obtain the optimization objective function.

[0105] S332, obtaining the optimization objective function obtained in step S331, constructing a constraint processor based on the Maslov form; converting hard constraints into soft penalty terms; calculating the constraint violation degree; constructing an augmented Lagrangian function, and obtaining a constrained optimization model.

[0106] S333. Obtain the constrained optimization model obtained in step S332, construct a solver based on the Newton-YY method; calculate the search direction; construct an adaptive step sequence; perform iterative optimization to obtain an initial capacity solution.

[0107] S334, obtaining the initial capacity solution obtained in step S333, constructing a solution refiner based on Bregman divergence; calculating the stability index of the solution; analyzing the sensitivity parameters; optimizing the solution parameters, and obtaining a modified capacity solution.

[0108] S335. Obtain the modified capacity plan obtained in step S334, construct a plan verifier based on Moreau-Yosida regularization; calculate the feasibility measure of the plan; analyze the robustness of the plan; optimize the configuration parameters, and obtain the optimal capacity configuration plan.

[0109] By introducing the fractional programming constrained optimization method, the optimal configuration of energy storage capacity is achieved. This method first constructs the objective function based on Pontryagin duality, accurately describing the nonlinear relationship between energy storage investment benefits and system stability; secondly, the constraints are processed in the Maslov form, effectively balancing the influence of hard constraints and soft constraints; thirdly, the innovative Newton-YY method is used for optimization and solution, which significantly improves the convergence speed of the algorithm; then, the solution is refined by Bregman divergence to enhance the stability of the optimization results; finally, the feasibility of the scheme is verified based on Moreau-Yosida regularization to ensure the engineering practicality of the configuration scheme. This systematic capacity optimization method can not only optimize system stability under cost constraints, but also ensure the practical feasibility of the optimization results. It is particularly suitable for solving energy storage planning problems under the conditions of high proportion of new energy access.

[0110] According to one aspect of the present application, step S41 is specifically:

[0111] S411, obtain the real-time measurement data of the system, the optimal capacity configuration plan obtained in step S3 and the spatiotemporal feature tensor obtained in step S1, and construct a noise preprocessor based on Martingale decomposition; calculate the local fluctuation of the measurement data; extract the deterministic component; decompose the random component to obtain the preprocessed measurement data.

[0112] S412, obtaining the preprocessed measurement data obtained in step S411, constructing a state transition model based on the Fokker-Planck equation; calculating the probability flow of the system; analyzing the probability density evolution; extracting state transition features, and obtaining a state prediction matrix.

[0113] S413, obtaining the state prediction matrix obtained in step S412, constructing an error dynamics model based on Symplectic geometry; calculating the invariants of Hamiltonian flow; analyzing the conservation characteristics of the system; constructing error compensation terms, and obtaining a modified prediction matrix.

[0114] S414, obtain the modified prediction matrix obtained in step S413, and construct a Stochastic differential observer; calculate measurement update gain; analyze innovation sequence characteristics; optimize filter parameters to obtain state estimates.

[0115] S415, obtaining the state estimation value obtained in step S414, constructing a predictor based on the Stratonovich integral; calculating the integral value of the prediction window; analyzing the prediction error distribution; combining the estimation results to obtain the system state estimation and predicted state.

[0116] By introducing the recursive state estimation and prediction method, the real-time and accurate evaluation of the system state is achieved. This method first pre-processes the measurement data based on Martingale decomposition, effectively separating the deterministic component and the random disturbance; secondly, the state transition characteristics are described by the Fokker-Planck equation, accurately describing the probabilistic evolution process of the system; thirdly, the error dynamics model is constructed using Symplectic geometry to ensure the stability of the numerical calculation; then, The state observer is designed by stochastic differentiation to improve the accuracy of state estimation; finally, state prediction is realized based on Stratonovich integral to enhance the reliability of prediction results. This systematic state assessment method can not only accurately capture the random fluctuation characteristics of renewable energy output, but also reliably predict the dynamic response of the system, providing accurate state information for real-time control of energy storage.

[0117] According to one aspect of the present application, step S42 is specifically:

[0118] S421. Obtain the system state estimation, predicted state and energy storage device parameter data obtained in step S41, construct a multi-time scale decomposer based on the Cartan moving frame; calculate the characteristic time of the system; analyze the time scale coupling; construct a decomposition basis function, and obtain a multi-scale decomposition matrix.

[0119] S422. Obtain the multi-scale decomposition matrix obtained in step S421, construct a fast response controller based on Beltrami flow; calculate the geodesic of the system; analyze the curvature characteristics; design the control law, and obtain a fast response strategy.

[0120] S423, obtaining the rapid response strategy obtained in step S422, constructing a mid-term optimizer based on the Hamilton-Jacobi equation; calculating the value function gradient; analyzing the optimal trajectory; solving the optimization problem, and obtaining a mid-term control sequence.

[0121] S424, obtaining the medium-term control sequence obtained in step S423, constructing a long-term planner based on the Exponential family; calculating the natural parameters of the system; analyzing the parameter evolution law; optimizing the planning objectives, and obtaining the long-term control strategy.

[0122] S425, obtaining the quick response strategy obtained in step S422, the medium-term control sequence obtained in step S423, and the long-term control strategy obtained in step S424, and constructing a strategy fuser based on Riemannian metric; calculating the distance between strategies; analyzing the fusion path; optimizing the combination weights, and obtaining the energy storage response strategy.

[0123] By introducing a multi-time scale hierarchical decision-making method, the accurate generation of energy storage response strategy is achieved. This method first constructs a multi-time scale decomposer based on the Cartan moving frame to accurately identify the characteristic time scale and coupling relationship of the system; secondly, a fast response controller is designed through the Beltrami flow to achieve rapid suppression of transient disturbances; thirdly, the Hamilton-Jacobi equation is used to optimize the medium-term control sequence to ensure the stable operation of the system on a second time scale; then, the Exponential family method is used to plan the long-term control strategy to optimize the economy and life characteristics of energy storage; finally, the scientific integration of multi-scale strategies is achieved based on the Riemannian metric to ensure the continuity and smoothness of the control strategy. This hierarchical response strategy generation method can not only provide corresponding control countermeasures for the different time scale fluctuation characteristics of new energy, but also coordinate the balance between fast response and continuous regulation of energy storage, which is particularly suitable for dealing with system stability control problems under the condition of high proportion of new energy access.

[0124] According to one aspect of the present application, step S43 is specifically:

[0125] S431, obtaining the energy storage response strategy, energy storage device state set and network topology information obtained in step S42, constructing a complex variable objective function generator based on Wirtinger derivative; calculating the analytical gradient of the system; analyzing the singular points of the gradient; constructing a local objective function to obtain an objective function set.

[0126] S432, obtaining the objective function set obtained in step S431, constructing a constraint decomposer based on Frank-Wolfe flow; calculating the tightness index of the constraint; analyzing the feasible domain structure; converting the constraint form to obtain the decomposed constraint matrix.

[0127] S433, obtaining the decomposition constraint matrix obtained in step S432, constructing a consistency optimizer based on Krasnoselskii-Mann iteration; calculating the consistency error between nodes; analyzing the convergence characteristics; constructing an iterative format, and obtaining a consistency constraint vector.

[0128] S434, obtaining the consistency constraint vector obtained in step S433, constructing a distributed solver based on Nesterov momentum; calculating the search direction; constructing a momentum accumulation sequence; performing alternating iterations to obtain a local optimal solution.

[0129] S435, obtaining the local optimal solution obtained in step S434, constructing a global coordinator based on the Proximal point method; calculating the proximal gradient; analyzing the gradient mapping characteristics; performing global optimization, and obtaining the optimized control instructions.

[0130] S436, obtain the optimized control instructions obtained in step S435, and construct a Kurdyka-based Stability analyzer for inequalities; compute Lyapunov functions; analyze energy dissipation rates; verify convergence properties and obtain stability indicators.

[0131] S437, obtain the optimized control instruction obtained in step S435 and the stability index obtained in step S436, construct a refined optimizer based on Mirror Descent; calculate the dual gradient; analyze the dual gap; optimize the control parameters to obtain the final control instruction.

[0132] By introducing the distributed consistency optimization method, the coordinated control of energy storage equipment is realized. This method first constructs a complex variable objective function based on the Wirtinger derivative, accurately describing the dynamic characteristics of energy storage control; secondly, the constraint decomposition is performed through the Frank-Wolfe flow, effectively handling the complex operating constraints; thirdly, the Krasnoselskii-Mann iteration is used to achieve consistency optimization, improving the convergence performance of distributed optimization; then, the optimization process is accelerated by the Nesterov momentum method, significantly improving the computational efficiency of the algorithm; thirdly, global coordination is performed based on the Proximal point method to ensure the global optimality of the optimization results; then, the Kurdyka- The stability of inequality analysis ensures the robustness of the control strategy; finally, the mirror descent method is used for fine-tuning and optimization to further improve the control effect. This systematic collaborative control method can not only achieve the coordination of multiple energy storage devices, but also ensure the stability and economy of the overall control strategy, which is particularly suitable for solving the collaborative control problem of large-scale distributed energy storage systems.

[0133] In a certain scenario, the data processing flow is as follows:

[0134] S1: Multi-dimensional grid transient feature extraction and preprocessing, extracting transient characteristic information from the original grid data and establishing spatiotemporal feature tensors;

[0135] S11: Transient data acquisition and standardization:

[0136] Original grid measurement data D(t); including: node voltage n is the number of nodes; node frequency Node power New energy output m is the number of new energy access points.

[0137] Adaptive quantile normalization (AQN) is used to obtain a standardized data matrix

[0138] S12: Time series decomposition and reconstruction:

[0139] Innovative method: introducing "Piecewise Hankel Matrix Recursive Decomposition" (PHMRD)

[0140] Read the normalized data matrix X(t).

[0141] Constructing the sliding Hankel matrix Where L is the embedding dimension and K is the time delay;

[0142] Apply recursive decomposition to each time window w: H(τ) = U(τ)Σ(τ)V(τ)^T;

[0143] Extract the main eigenvector matrix r is the characteristic dimension;

[0144] Get the time feature matrix

[0145] S13: Spatial topological feature extraction, processed using dynamic weighted graph decomposition (DWSD):

[0146] Read the grid topology adjacency matrix Time feature matrix T(t).

[0147] Construct a dynamic weight matrix W(t)=diag(|T(t)|);

[0148] Calculate the weighted Laplacian matrix L(t)=W(t)–A;

[0149] Apply spectral decomposition to obtain eigenvalues ​​λ_i and eigenvectors φ_i;

[0150] Get the spatial feature matrix k is the number of selected features.

[0151] S14: Construction of spatiotemporal feature tensors, processed using the multi-scale tensor fusion (MTF) algorithm:

[0152] Read the time feature matrix T(t) and the space feature matrix S(t).

[0153] Constructing a third-order tensor

[0154] Apply Tucker decomposition: Ψ = G × 1U1 × 2U2 × 3U3;

[0155] Get the feature tensor Ψ and the core tensor G.

[0156] S2: Multi-dimensional transient stability assessment and quantification, based on the spatiotemporal characteristic tensor of S1, a quantitative indicator system for transient stability is constructed;

[0157] S21: Transient response characteristic analysis, using hierarchical sparse principal component mapping HSPCM for processing:

[0158] Read the feature tensor Ψ (from S14) and the core tensor G (from S14).

[0159] Constructing a multi-layer sparse mapping matrix l is the number of layers;

[0160] Calculate the inter-layer mapping: Y_l = M_l·Y_(l-1), where Y_0 is the input feature;

[0161] Extract the principal component direction v_i and the corresponding weight w_i;

[0162] Get the response feature matrix p is the feature dimension.

[0163] S22: Transient stability index construction, using multi-scale coupled stability evaluation MCSE algorithm for processing:

[0164] Read the response feature matrix R(t);

[0165] Define the transient stability index: IST(i) = Σ(α_k·|ΔR_k(i)| / R_ref), i is the node number, α_k is the weight coefficient, ΔR_k is the characteristic deviation; define the dynamic stability index: DST(i) = ∫(β_t·|dR(i) / dt|)dt, β_t is the time scale weight; define the collaborative stability index: CST(i,j) = γ·cov(R(i),R(j)) where γ is the coupling coefficient and (i,j) is a node pair;

[0166] Get the stability index set Θ = {IST, DST, CST};

[0167] S23: Probabilistic feature extraction and quantification, using conditional Wasserstein distance CWDM mapping for processing:

[0168] Read stability indicator set Θ; historical fault data set F;

[0169] Construct the conditional probability density function p(Θ|F);

[0170] Calculate Wasserstein distance: W(p_i,p_j)=inf{E[d(X,Y)]:Xp_j};

[0171] Build a distance matrix

[0172] Get the probability feature matrix q is the probability feature dimension;

[0173] S24: Comprehensive stability score, processed using adaptive weight fusion AWF:

[0174] Read stability indicator set Θ; probability feature matrix P;

[0175] Construct an adaptive weight function: w(t) = σ(λ·Θ(t) + μ·P(t)), where σ is the activation function, and λ and μ are adaptation coefficients;

[0176] Calculate the comprehensive score: S(i) = Σ(w_k · [θ_k(i) + ρ · p_k(i)]) where ρ is the balance factor;

[0177] Get the node stability score vector

[0178] S3: Multi-objective optimization configuration of network energy storage, based on the characteristic tensor of S1 and the stability score of S2, to determine the optimal energy storage configuration scheme;

[0179] S31: Candidate node screening, using affinity propagation clustering evaluation APCE for processing:

[0180] Read the node stability score vector S (from S24); the grid topology A; the new energy distribution information RE;

[0181] Construct node affinity matrix: Aff(i,j)=exp(-||x_i-x_j|| 2 / 2σ 2 )·δ(d_ij) where x_i is the feature vector of node i, d_ij is the topological distance, and δ is the decay function;

[0182] Calculate the responsibility matrix R and the availability matrix A: R(i,k)=s(i,k)-max{a(i,k')+s(i,k'):k'≠k} Identify cluster centers and spheres of influence;

[0183] Get the candidate node set C and node weight vector W;

[0184] S32: transient support capacity assessment, processed using multi-time scale influence propagation MTSIP;

[0185] Read the candidate node set C; the spatiotemporal feature tensor Ψ (from S14); define the time scale set T = {t1, t2, ..., t k};

[0186] Construct the influence propagation matrix: I(t) = [1-α·L]^(-1)·B(t), where L is the normalized Laplace matrix, α is the attenuation coefficient, and B(t) is the basic influence;

[0187] Calculate the cumulative support effect: E(i) = Σ_t[β(t)·I(t)·φ(t)], where β(t) is the time weight and φ(t) is the support effect function;

[0188] Get the node support capacity matrix E;

[0189] S33: Capacity optimization configuration, using fractional programming constraint optimization FPC for processing:

[0190] Read candidate node set C; node support capacity matrix E; cost constraint Cost_max;

[0191] Construct the objective function: max f(x) = [Σ(w_i·E_i·xi)] / [Σ(c_i·xi)+b], where x_i is the energy storage capacity, c_i is the unit cost, and b is the fixed cost; set the constraints: capacity constraint: x_min≤x_i≤x_max; budget constraint: Σ(c_i·xi)≤Cost_max; response time constraint: τ_i(x_i)≤τ_max; solve the optimization problem: use the Dinkelbach algorithm to transform it into parameterized quadratic programming;

[0192] Get the optimal capacity configuration plan X*;

[0193] S34: Configuration solution verification, processed using robustness probability assessment RPA:

[0194] Read the optimal capacity configuration plan X*; historical fault data F; system parameter set P;

[0195] Generate a perturbation scenario set Ω: ω_i = g(F,P,ε) where g is the scenario generation function and ε is the random perturbation;

[0196] Calculate the robustness index of the scheme: R(X*)=min_ω∈Ω{Pr(φ(X*,ω)≥η)}, where φ is the performance function and η is the threshold;

[0197] Evaluate the feasibility of the solution: V(X*) = E_ω[max(0,h(X*,ω))], where h is the constraint violation metric function;

[0198] The obtained verification report V contains robustness indicators and feasibility assessment.

[0199] S4: Energy storage collaborative control based on system status. Based on the configuration results of S1-S3, design a collaborative control strategy for energy storage equipment.

[0200] S41: Real-time evaluation of system status, using recursive state estimation and prediction (RSEP) for processing:

[0201] Read real-time measurement data M(t); configuration scheme X* (from S33); spatiotemporal characteristic tensor Ψ (from S14);

[0202] Construct the state observation equation: z(t)=h(x(t))+v(t), where h is the observation function and v is the measurement noise; design the recursive estimator: x*(t|t)=x*(t|t-1)+K(t)[z(t)-h(x*(t|t-1))], where K(t) is the gain matrix, which is adaptively adjusted based on the measurement noise covariance; state prediction: x*(t+1|t)=f(x*(t|t),u(t))+Δ(t), where f is the state transfer function and Δ is the model uncertainty compensation term;

[0203] Obtain the system state estimate x*(t) and predicted state x*(t+1|t).

[0204] S42: Energy storage response strategy generation, using multi-time scale hierarchical decision-making MSHD for processing:

[0205] Read system state estimate x*(t); predict state x*(t+1|t); energy storage device parameter set B;

[0206] Construct a hierarchical decision-making framework: rapid response layer (ms level): u_f(t) = K_f [x*(t)-x_ref] + μ_f(t); medium-term adjustment layer (s level): u_m(t) = arg min J_m(x*(t:t+T_m), u(t:t+T_m)); long-term optimization layer (min level): u_l(t) = arg min J_l(x*(t:t+T_l), u(t:t+T_l)); coordinate decisions at different levels: u(t) = w_f(t)·u_f(t)+w_m(t)·u_m(t)+w_l(t)·u_l(t) where w_i(t) is the dynamic weight coefficient;

[0207] The energy storage response strategy u(t) is obtained.

[0208] S43: Collaborative optimization of energy storage equipment, using the proposed distributed consistency optimization DCO for processing:

[0209] Read the energy storage response strategy u(t); energy storage device status set S(t); network topology information A;

[0210] Construct local objective function: f_i(x_i)=||x_i-u_i(t)|| 2+λ·g(x_i,S_i(t)); Design distributed optimization algorithm: x_i(k+1)=Π_X[x_i(k)-α_k(LA f_i(x_i(k))+ρ_k·p_i(k))]p_i(k+1)=p_i(k)+β_k·Σ_j a_ij(x_i(k)-x_j(k)) where Π_X is the projection operator, α_k, β_k are the step sequence; Consistency constraint processing: Add coupling constraint h(x_i,x_j)≤0,

[0211] Get the optimized control instruction u*(t).

[0212] S44: Performance evaluation and adaptive adjustment, using online learning and dynamic compensation OLDC for processing:

[0213] Read optimization control instruction u*(t); actual response data R(t); historical performance index H;

[0214] Construct performance evaluation indicators: η(t) = φ(R(t), u*(t)) + ψ(ΔE(t)), where φ is the tracking error function and ψ is the energy loss function; online parameter adjustment: θ(t+1) = θ(t) + γ(t)·LAη(t), where θ is the control parameter and γ is the learning rate; dynamic compensation generation: c(t) = f(η(t), H, θ(t)); LA is the differential operator.

[0215] Get the adjusted control strategy and Performance Evaluation Report P.

[0216] In this embodiment, part of the calculation process is as follows:

[0217] In the Hankel matrix recursive decomposition, the numerical stability guarantee mechanism is as follows:

[0218] During the recursive decomposition process, condition number constraints and regularization terms are introduced:

[0219] H*(τ)=H(τ)+λR(τ);

[0220] Where: H*(τ) is the modified Hankel matrix; H(τ) is the original Hankel matrix; λ is the regularization parameter, λ = α·σmax / κ(H); R(τ) is the regularization matrix; α is the adjustment coefficient (0<α<1); σmax is the maximum singular value; κ(H) is the matrix condition number;

[0221] The eigenvalue adaptive screening criteria are as follows: η(i) = μσ + β·σstd·log(d / i); where:

[0222] η(i) is the screening threshold of the i-th eigenvalue; μσ is the mean of the eigenvalue; σstd is the standard deviation of the eigenvalue; d is the total number of eigenvalues; β is the adaptive coefficient (1≤β≤2); i is the eigenvalue index;

[0223] Dynamic weighted graph method enhancements, specifically:

[0224] The adaptive weight update frequency is introduced as;

[0225] Where: f_update is the weight update frequency; fmax is the maximum update frequency; ρ is the proportional coefficient;

[0226] ΔE is the system energy change rate; Eth is the energy change threshold; is the ceiling function;

[0227] Weight smooth transition strategy, specifically:

[0228] The exponential moving average is: W(t+1)=γW(t)+(1-γ)W_new(t);

[0229] Where: W(t) is the weight matrix at time t; W_new(t) is the newly calculated weight matrix; γ is the smoothing coefficient, γ = exp(-Δt / τ); τ is the characteristic time constant; Δt is the sampling time interval.

[0230] Step S2 also includes a stability assessment,

[0231] The dimension of the mapping matrix is ​​adaptively determined.

[0232] Where: di is the dimension of the i-th layer; d0 is the initial dimension; λd is the dimension attenuation coefficient; ε is the adjustment coefficient; SVR(i) is the singular value ratio of the i-th layer; i is the layer index.

[0233] The sparse constraint parameters are adjusted dynamically, specifically: μ(t) = μ0·(1+δ·||LA L(t)|| / ||LA L0||);

[0234] Where: μ(t) is the sparse constraint parameter at time t; μ0 is the baseline constraint parameter; δ is the sensitivity coefficient; LA L(t) is the current loss gradient; LA is the Laplace operator. LA L0 is the initial loss gradient; ||·|| is the norm operator.

[0235] Multi-scale coupling evaluation method, specifically:

[0236] Introduce the coupling matrix: C(t1,t2)=exp(-|t1-t2| / τc)·cos(ω·|t1-t2|);

[0237] Where: C(t1,t2) is the coupling coefficient between different time scales; t1,t2 are different time scales; τc is the coupling characteristic time; ω is the oscillation frequency.

[0238] Comprehensive stability index: SI = ∑∑wij·C(ti,tj)·φ(xi,xj);

[0239] Among them: SI is the comprehensive stability index; wij is the weight coefficient matrix; φ(xi,xj) is the stability evaluation function; xi,xj are state vectors at different time scales.

[0240] Step S3 optimizes the configuration method, including:

[0241] During the affinity propagation clustering evaluation enhancement process, the number of cluster centers is adaptively determined as follows:

[0242] Where: Nc is the number of cluster centers; n is the total number of nodes; β is the basic coefficient; θ is the entropy weight coefficient; H(X) is the data entropy; X is the node feature matrix.

[0243] Node similarity calculation: s(i,j)=exp(-||xi-xj|| 2 / 2σ 2 )·exp(-dij / dc); where: s(i,j) is the similarity between nodes i and j; xi,xj are node feature vectors; dij is the topological distance; σ is the feature scale parameter; dc is the distance feature length.

[0244] Fractional programming constrained optimization, specifically: objective function weight: wi = pi · (1 + αi · Vi / Vmax) · (1 + βi · Si / Smax);

[0245] Where: wi is the weight of the i-th target; pi is the basic weight; αi, βi are adjustment coefficients; Vi is the voltage sensitivity; Si is the power sensitivity; Vmax, Smax are standardization coefficients.

[0246] In step S4, a collaborative control method is also included.

[0247] Prediction error compensation mechanism in recursive state estimation: x*(t+1)=f(x*(t))+K(t)·e(t)+Δ(t);

[0248] Where: x*(t) is the state estimate; f(·) is the state transfer function; K(t) is the gain matrix; e(t) is the prediction error; Δ(t) is the compensation term.

[0249] Model uncertainty quantification, specifically: Δ(t) = Γ(t)·ψ(x*(t))+ω(t);

[0250] Where: Γ(t) is the uncertainty gain matrix; ψ(·) is the nonlinear mapping function; ω(t) is the random disturbance term.

[0251] The control strategy coordination mechanism of multi-time scale hierarchical decision-making is as follows: u(t) = ∑λi(t)·ui(t)·φi(x(t));

[0252] Among them: u(t) is the comprehensive control instruction; λi(t) is the dynamic weight coefficient; ui(t) is the control instruction of each layer; φi(·) is the control effectiveness function; x(t) is the system state.

[0253] Dynamic weight adjustment, specifically: λi(t+1)=λi(t)+η·LA J(λi(t));

[0254] Where: λi(t) is the weight of the i-th layer; η is the learning rate; LA J is the performance gradient; J is the control performance index.

[0255] Since all parameters need to be adjusted and optimized according to the actual engineering scenario, in practical applications, offline parameter setting is first performed, and then online adaptive adjustment is performed to optimize the control effect.

[0256] The preferred embodiments of the present invention are described in detail above; however, the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A method for selecting a site and determining the capacity of a network energy storage system that supports transient stability of a high-proportion new energy power grid, characterized in that: The steps include: Step S1, obtaining grid node voltage data, node frequency data, node power data and new energy output data, forming these data into an original grid measurement data matrix, processing them according to the segmented Hankel matrix recursive method and the dynamic weighted spectrum method, and obtaining transient characteristic data; Extracting the time feature matrix and the space feature matrix from the transient feature data, processing them according to the multi-scale tensor fusion method to obtain the feature tensor and the core tensor; Step S2, obtaining the feature tensor and core tensor obtained in step S1, and processing them according to the hierarchical sparse principal component mapping method to obtain a response feature matrix; Based on the response characteristic matrix, a multi-scale coupling evaluation method is used to calculate various stability indicators. According to these stability indicators and historical fault data, the conditional Wasserstein distance mapping method is used to obtain the probability characteristic matrix. The stability indicators and the probability characteristic matrix are processed by the adaptive weight fusion method to obtain the node stability score vector. Step S3, obtaining the node stability score vector, power grid topology data and new energy distribution data obtained in step S2, and processing them according to the affinity propagation clustering method to obtain a candidate node set and a node weight vector; processing the candidate node set and the spatiotemporal feature tensor obtained in step S1 by a multi-time scale influence propagation method to obtain a node support capacity matrix; according to the node support capacity matrix and the system cost constraint, using the fractional programming constraint optimization method to obtain the optimal capacity configuration plan; processing the optimal capacity configuration plan with historical fault data and system parameters by a robustness probability evaluation method to obtain a verification report; Step S4, obtaining the system real-time measurement data, the optimal capacity configuration scheme obtained in step S3 and the spatiotemporal characteristic tensor obtained in step S1, and processing them according to the recursive state estimation and prediction method to obtain the system state estimation and predicted state; according to the system state estimation, predicted state and energy storage device parameters, adopting the multi-time scale hierarchical decision method to obtain the energy storage response strategy; The energy storage response strategy, energy storage device status set and network topology information are processed through a distributed consistency optimization method to obtain optimized control instructions; based on the optimized control instructions, actual response data and historical performance indicators, online learning and dynamic compensation methods are used to obtain the adjusted control strategy and performance evaluation report.

2. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S1 is specifically as follows: Step S11, obtaining grid node voltage data, grid node frequency data, grid node power data and new energy output data, and arranging these data into an original grid measurement data matrix according to time series; calculating the quantile value of each data item according to the quantile distribution characteristics of each data item in the data matrix; normalizing the original data based on the calculated quantile value to obtain a standardized data matrix; Step S12, obtaining the standardized data matrix obtained in step S11, setting the time window length and sliding step length, and constructing a sliding Hankel matrix; performing singular value decomposition on the Hankel matrix in each time window, extracting the main eigenvalues ​​and eigenvectors; recursively reconstructing the data according to the changing rules of the eigenvalues ​​and eigenvectors, and obtaining a time feature matrix; Step S13, obtaining the adjacency matrix data of the power grid topology structure and the time characteristic matrix obtained in step S12, and calculating the weight coefficient of each node; constructing a weighted Laplace matrix according to the weight coefficient and the adjacency matrix; performing eigenvalue decomposition on the weighted Laplace matrix, extracting the main eigenvectors, and obtaining a spatial characteristic matrix; Step S14, obtain the time feature matrix obtained in step S12 and the space feature matrix obtained in step S13, combine the two feature matrices according to the node dimension, and construct a third-order tensor; perform Tucker decomposition on the third-order tensor to obtain basis matrices in three directions respectively; and calculate the feature tensor and the core tensor according to the product relationship of the basis matrices.

3. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S2 is specifically as follows: Step S21, obtaining the feature tensor and core tensor obtained in step S1, constructing a multi-layer mapping matrix, wherein the dimension of each layer of the mapping matrix is ​​determined according to a preset parameter; transforming the feature tensor data through each layer of the mapping matrix in turn to obtain an intermediate feature; Perform principal component analysis on the intermediate features to extract the main direction vectors and corresponding weight coefficients; reconstruct the data based on the direction vectors and weight coefficients to obtain the response feature matrix; Step S22, obtaining the response characteristic matrix obtained in step S21, calculating the characteristic deviation value of each node at different time scales; performing weighted summation on the characteristic deviation values ​​according to a preset time weight coefficient to obtain a transient stability index; calculating the time derivative of the response characteristic, and integrating it in the time domain to obtain a dynamic stability index; calculating the covariance of the response characteristics between the node pairs, and multiplying it by the coupling coefficient to obtain a collaborative stability index; combining these three types of indicators into a stability index set; Step S23, obtaining the stability index set and the system historical fault data obtained in step S22, and performing conditional probability density estimation on the stability index according to the fault data; calculating the Wasserstein distance of the probability distribution of the stability index between different nodes; forming a distance matrix with the calculated distance values; performing dimensionality reduction processing on the distance matrix to obtain a probability feature matrix; Step S24, obtain the stability index set obtained in step S22 and the probability feature matrix obtained in step S23, construct an adaptive weight function, and calculate the dynamic weight coefficient of each index; multiply the stability index and probability feature by the corresponding weight coefficient respectively; sum the weighted index and feature to obtain the comprehensive score of each node; and form a node stability score vector with the scores of all nodes.

4. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S3 is specifically as follows: Step S31, obtaining the node stability score vector, power grid topology data and new energy distribution data obtained in step S2, calculating the characteristic distance and topological distance between nodes; constructing a node affinity matrix according to the characteristic distance and topological distance; calculating the responsibility matrix between nodes, in which each element represents the suitability of the node as a cluster center; Calculate the availability matrix between nodes, where each element represents the suitability of a node to select other nodes as cluster centers; iterate and update the responsibility matrix and availability matrix until convergence to obtain a set of candidate nodes; calculate the influence range and importance of each candidate node to obtain a node weight vector; Step S32, obtaining the candidate node set obtained in step S31 and the spatiotemporal feature tensor obtained in step S1, and setting multiple feature time scales; Construct the normalized Laplace matrix and calculate the basic influence matrix; For each time scale, calculate the influence propagation matrix; multiply the influence propagation matrix at different time scales by the corresponding time weight; Calculate the cumulative support effect and obtain the support capacity value of each node at different time scales; These support capacity values ​​are combined into a node support capacity matrix; Step S33, obtaining the candidate node set obtained in step S31, the node support capability matrix obtained in step S32, and the system cost constraint data, constructing a fractional objective function, where the numerator is the support effect and the denominator is the cost function; setting upper and lower limit constraints of energy storage capacity, total cost constraints, and response time constraints; The fractional programming problem is transformed into a parameterized quadratic programming problem. The optimal capacity configuration solution is obtained by iteratively solving the parameterized quadratic programming problem and updating the parameters until convergence. Step S34, obtaining the optimal capacity configuration scheme, system historical fault data and system parameter data obtained in step S33, generating multiple groups of disturbance scenarios according to the historical fault data and system parameters; calculating the system performance index for each group of disturbance scenarios; counting the probability that the performance index is less than the threshold in all scenarios, and obtaining the robustness index of the scheme; calculating the expected value of the degree of constraint violation, and obtaining the feasibility index of the scheme; combining the robustness index and the feasibility index into a verification report.

5. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S4 is specifically as follows: Step S41, obtain the real-time measurement data of the system, the optimal capacity configuration plan obtained in step S3 and the spatiotemporal characteristic tensor obtained in step S1, construct the state observation equation, calculate the deviation between the measured value and the predicted value; calculate the gain matrix according to the measurement noise covariance; multiply the gain matrix by the measurement deviation to update the state estimate; calculate the predicted state according to the updated state estimate and the control input; add the model uncertainty compensation term to the predicted state to obtain the system state estimate and the predicted state; Step S42, obtaining the system state estimation, predicted state and energy storage device parameter data obtained in step S41, calculating the millisecond-level rapid response control quantity, including the state deviation term and the compensation term; solving the medium-term regulation optimization problem at the second-level time scale to obtain the medium-term control sequence; solving the long-term optimization problem at the minute-level time scale to obtain the long-term control sequence; calculating the dynamic weight coefficients of the control quantities at the three time scales; The control quantities of the three time scales are combined according to the dynamic weight coefficients to obtain the energy storage response strategy; Step S43, obtaining the energy storage response strategy, energy storage device state set and network topology information obtained in step S42, constructing the local objective function of each energy storage device, including the tracking error term and the state constraint term; calculating the local gradient and dual variable of each energy storage device; exchanging the state information of adjacent energy storage devices according to the network topology information; updating the control amount and dual variable of each energy storage device; repeating the iterative calculation until convergence, and obtaining the optimized control instruction; Step S44, obtaining the optimized control instruction, the actual system response data and the historical performance index data obtained in step S43, calculating the control instruction tracking error and the energy storage loss, and obtaining the performance evaluation index; calculating the gradient of the control parameter according to the performance evaluation index; updating the control parameter to obtain a new parameter set; generating a dynamic compensation amount based on the updated parameter set and the historical performance index; combining the optimized control instruction with the dynamic compensation amount to obtain an adjusted control strategy; The performance evaluation indicators, parameter update results and compensation effects form a performance evaluation report.

6. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S12 is specifically as follows: S121, obtaining the standardized data matrix obtained in step S11, calculating the optimal embedding dimension and time delay parameter according to the time correlation of the data; segmenting the data matrix, constructing a submatrix for each segment according to the calculated parameters; rearranging all submatrices in time order to obtain an initial Hankel matrix; S122, obtaining the initial Hankel matrix obtained in step S121, calculating the condition number and rank of the matrix; constructing a regularization parameter sequence according to the condition number; for each regularization parameter, calculating the corresponding matrix completion term; selecting the optimal regularization parameter, and obtaining an enhanced Hankel matrix; S123, obtaining the enhanced Hankel matrix obtained in step S122, constructing an iterative decomposition framework based on the Krylov subspace; calculating the eigenvalue basis vector of each iteration; Orthogonalize the eigenvalue basis vectors; repeat the iteration until convergence to obtain the principal component matrix; S124, obtaining the principal component matrix obtained in step S123, establishing a recursive prediction model; calculating the confidence interval of the model parameters; screening reliable eigenvectors according to the confidence interval; combining the screened eigenvectors to obtain a reduced-dimensional feature matrix; S125. Obtain the reduced dimension feature matrix obtained in step S124 and the initial Hankel matrix obtained in step S121, and construct a reconstruction error function; calculate the local reconstruction error of each time window; determine the optimal reconstruction parameters according to the error distribution; apply the reconstruction parameters to reconstruct the data and obtain the time feature matrix.

7. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S13 is specifically as follows: S131, obtaining the adjacency matrix data of the power grid topology structure and the time characteristic matrix obtained in step S12, calculating the initial weight coefficient according to the degree distribution of the node; calculating the sliding variance matrix for the time characteristic matrix; mapping the elements of the variance matrix to the weight space to obtain the dynamic weight coefficient matrix; S132, obtaining the dynamic weight coefficient matrix and the adjacency matrix obtained in step S131, constructing a distance metric function based on the Riemann manifold; calculating the geodesic distance between nodes; and constructing a distance matrix according to the geodesic distance; Adaptively smooth the distance matrix to obtain a weighted distance matrix; S133, obtaining the weighted distance matrix obtained in step S132, constructing a K nearest neighbor relationship graph; calculating the local connectivity index of the graph; adjusting the weights of the edges according to the connectivity index; combining the adjusted weights with the original adjacency relationship to obtain an enhanced adjacency matrix; S134, obtaining the enhanced adjacency matrix obtained in step S133, calculating the generalized degree matrix; constructing a clustering structure based on the stochastic block model; extracting the connection features between blocks and within blocks; combining Connect the features to obtain a multi-scale topological matrix; S135, obtaining the multi-scale topological matrix obtained in step S134, constructing a spectral decomposition optimization model based on Schur complement; iteratively calculating eigenvalues ​​and eigenvectors; determining a truncation threshold according to the eigenvalue distribution; extracting the main eigenvectors to obtain a spatial feature matrix.

8. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S14 is specifically as follows: S141, obtaining the time feature matrix obtained in step S12 and the space feature matrix obtained in step S13, and constructing a feature correlation matrix; Calculate mutual information between features; select significant features based on mutual information values; Standardize the significant features to obtain the preprocessing feature matrix; S142, obtaining the preprocessing feature matrix obtained in step S141, and constructing a feature mapping function based on the Grassmann manifold; Calculate the principal curvature of the feature subspace; adjust the feature weights based on the curvature information; combine Weighted features to obtain enhanced feature matrix; S143, obtaining the enhanced feature matrix obtained in step S142, and constructing a multilinear tensor space; Calculate the basis vectors of the tensor space; orthogonalize the basis vectors; map the features to the tensor space to obtain the initial feature tensor; S144, obtaining the initial feature tensor obtained in step S143, constructing a multidimensional optimization model based on Tucker decomposition; calculating the rank of the core tensor; determining the compression rate according to the distribution of the rank; compressing the tensor to obtain a compressed feature tensor; S145. Obtain the compressed feature tensor obtained in step S144, build a reconstruction model based on TT decomposition (Tensor-train); calculate the reconstruction error of the model; adjust the reconstruction parameters according to the error distribution; reconstruct the data to obtain the final feature tensor and core tensor.

9. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S21 is specifically as follows: S211, obtaining the feature tensor and core tensor obtained in step S1, and constructing a feature projection operator based on the matrix manifold; Calculate the projection coefficients of the feature tensor in different directions; construct a directional sensitivity matrix based on the projection coefficients; normalize the sensitivity matrix to obtain a directional feature matrix; S212, obtaining the directional feature matrix obtained in step S211, constructing a mapping hierarchy structure based on Wasserstein distance; calculating the transmission distance between each layer; optimizing the inter-layer connection weights according to the transmission distance; combining the optimized inter-layer weights to obtain a multi-layer mapping matrix; S213, obtaining the multi-layer mapping matrix obtained in step S212, and constructing a characteristic transformation function based on geodesic flow; Calculate the evolution trajectory of features on the manifold; extract the key points of the trajectory; combine the key points into a transformation sequence to obtain an intermediate feature sequence; S214, obtaining the intermediate feature sequence obtained in step S213, constructing a sparse constraint based on the Frobenius norm; calculating a sparse representation of the feature sequence; and selecting the main components according to the sparsity; Orthogonalize the main components to obtain the main component feature matrix; S215, obtaining the principal component feature matrix obtained in step S214, and constructing an importance evaluation model based on Fisher information; Calculate the Fisher score of the feature; sort the features according to the score; select high-scoring features to reconstruct the data and obtain the response feature matrix.

10. The method for selecting a site and determining the capacity of a network energy storage system to support transient stability of a high-proportion new energy power grid according to claim 1, characterized in that: The step S22 is specifically as follows: S221, obtaining the response characteristic matrix obtained in step S21, constructing a stability functional based on the Lyapunov exponent; calculating the local Lyapunov exponent of the system; and constructing a stability domain according to the exponential distribution; The stability domain is segmented to obtain the segmented stability index; S222, obtaining the segmented stability index obtained in step S221, constructing a dynamic evolution model based on the Koopman operator; calculating the spectral decomposition sequence of the system; extracting the characteristic pattern of the spectral sequence; combination Characteristic mode, obtain dynamic stability characteristics; S223, obtaining the dynamic stability characteristics obtained in step S222, constructing an integral optimizer based on sliding Prony analysis; calculating the complex frequency components of the characteristics; and determining the integral interval according to the frequency components; Integrate the interval data to obtain the time integral index; S224, obtaining the time integral index obtained in step S223, constructing a coupling evaluation function based on the synchronous phase; calculating the phase correlation between nodes; and constructing a coupling network according to the correlation; Cluster the coupled network and obtain the coupling area index; S225. Obtain the segmented stability index obtained in step S221, the time integral index obtained in step S223, and the coupling area index obtained in step S224, and construct an index fusion model based on the entropy weight method; calculate the information entropy of each index; determine the combination weight according to the entropy value; combine the indicators according to the weights to obtain a stability index set.

Images