Distributed energy storage space-time coordinated regulation and control method, system and device and medium

Abstract

The invention belongs to the technical field of low-voltage power distribution network area regulation and control, and discloses a distributed energy storage space-time coordination regulation and control method, system and device and a medium, and the method comprises the steps: obtaining the operation data of a low-voltage optical storage area, and carrying out the preprocessing of the operation data; based on the preprocessed data, topology identification and line parameter estimation of the transformer area are executed; according to a two-stage space-time prediction training strategy, constructing a transformer area overvoltage risk space-time prediction model; predicting the preprocessed data by using a transformer area overvoltage risk space-time prediction model to obtain voltage risk probability information of transformer area nodes; according to the method, a topological structure, line parameters and overvoltage risk probability information of a transformer area are fused, a multi-objective optimization model is constructed, an NSGA-II algorithm is used for solving, and a distributed energy storage cooperative regulation and control strategy is obtained. The method has remarkable advantages in the aspects of data quality, topology identification, parameter estimation, overvoltage prediction, energy storage regulation and control and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of low-voltage distribution network area regulation, and particularly relates to a distributed energy storage space-time coordinated regulation method, system, device and medium. BACKGROUND

[0002] With the continuous access of high-proportion distributed photovoltaic and energy storage in the distribution area, the generation and consumption of the area power are affected by the space-time changes of sunlight and power demand, leading to voltage fluctuation of the area power grid and increasingly prominent overvoltage problems, which seriously affect the safe and stable operation of the power grid and the consumption capacity of photovoltaic power generation. The traditional overvoltage control method mainly relies on physical models and fixed control strategies, which are difficult to adapt to the complex and variable operation environment of the area. In the prior art, the methods based on centralized optimization or localized control often cannot balance photovoltaic consumption capacity and operation economy, leading to frequent problems such as voltage out-of-limit and increased network loss.

[0003] In terms of area topology structure and line parameter identification, the prior art faces many challenges. Traditional topology identification methods usually rely on manual inspection and static network models, which are difficult to cope with the dynamic changes of the topology structure of the area. For example, the access and exit of distributed resources such as photovoltaic and energy storage in the area, as well as line modification and other operations, will lead to changes in the topology structure. In addition, the accurate acquisition of line parameters also faces difficulties, and traditional parameter estimation methods mostly rely on offline measurement and historical data, which cannot reflect the changes of line impedance in real time, especially in the case of high-proportion distributed photovoltaic access, the line impedance is greatly affected by photovoltaic output fluctuation, further increasing the difficulty of parameter estimation.

[0004] In terms of overvoltage prediction, the existing methods are mostly based on single time series prediction models or spatial prediction models, which are difficult to capture the space-time characteristics of the area voltage. Traditional time series prediction models such as ARIMA and LSTM can handle time series data well, but ignore the spatial correlation between nodes; while spatial prediction models based on graph neural networks can capture the topological relationship between nodes, but have weak modeling ability for time series features. Therefore, there is an urgent need for a hybrid prediction model that can consider space-time features comprehensively to improve the accuracy of overvoltage prediction.

[0005] In terms of distributed energy storage regulation, the existing methods mostly adopt centralized optimization or localized control strategies, which are difficult to balance global optimization and local control. The centralized optimization method can achieve global optimization, but has high computational complexity and is difficult to meet real-time requirements; while the localized control strategy has fast response speed, but lacks a global perspective, which easily leads to local optimization and global suboptimization problems. In addition, the existing methods mostly use single-objective optimization, which is difficult to simultaneously consider multiple objectives such as photovoltaic consumption capacity, operation economy and voltage safety.

[0006] Therefore, how to provide a distributed energy storage space-time collaborative regulation method, system, device and medium is a problem to be solved at present. SUMMARY

[0007] The embodiment of the application provides a distributed energy storage space-time collaborative regulation method, system, device and medium to solve the problems in the prior art.

[0008] The following presents a simplified summary of some aspects of the disclosed embodiments in order to provide a basic understanding of such embodiments. This summary is not an extensive overview of the embodiments described in this document. It is not intended to identify key / critical elements of the embodiments or to delineate the scope of the embodiments. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is presented later.

[0009] According to a first aspect of the embodiment of the application, a distributed energy storage space-time collaborative regulation method is provided.

[0010] In one embodiment, the distributed energy storage space-time collaborative regulation method comprises the following steps:

[0011] Obtain the operation data of the low-voltage optical storage area, and pre-process the operation data;

[0012] Based on the pre-processed data, perform topology identification and line parameter estimation of the area; wherein the topology identification of the area is performed by using an improved DBSCAN clustering algorithm and a split expectation maximization algorithm, and the line parameter estimation of the area is performed according to a Bayesian filtering framework;

[0013] According to a two-stage space-time prediction training strategy, a space-time prediction model of the area overvoltage risk is constructed; the pre-processed data is predicted by using the space-time prediction model of the area overvoltage risk, and voltage risk probability information of the nodes of the area is obtained;

[0014] The topology structure, line parameter and overvoltage risk probability information of the area are fused to construct a multi-objective optimization model, and the NSGA-II algorithm based on non-dominated sorting and congestion calculation is used for solving, and a distributed energy storage collaborative regulation strategy is obtained.

[0015] According to a second aspect of the embodiment of the application, a distributed energy storage space-time collaborative regulation system is provided.

[0016] In one embodiment, the distributed energy storage space-time collaborative regulation system comprises:

[0017] The data pre-processing module is configured to obtain the operation data of the low-voltage optical storage area, and pre-process the operation data;

[0018] A topology and parameter identification module is configured to perform topology identification and line parameter estimation of the transformer area based on the preprocessed data, wherein the topology identification of the transformer area is performed by using an improved DBSCAN clustering algorithm and a split expectation maximization algorithm, and the line parameter estimation of the transformer area is performed according to a Bayesian filtering framework;

[0019] A overvoltage risk prediction module is configured to construct a transformer area overvoltage risk spatiotemporal prediction model according to a two-stage spatiotemporal prediction training strategy, and to perform prediction on the preprocessed data by using the transformer area overvoltage risk spatiotemporal prediction model to obtain voltage risk probability information of nodes of the transformer area.

[0020] A collaborative optimization decision module is configured to fuse the topology structure, line parameters and overvoltage risk probability information of the transformer area, construct a multi-objective optimization model, and solve the multi-objective optimization model by using a NSGA-II algorithm based on non-dominated sorting and congestion calculation to obtain a distributed energy storage collaborative regulation strategy.

[0021] According to a third aspect of the embodiments of the present application, a computer device is provided.

[0022] In some embodiments, the computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0023] According to a fourth aspect of the embodiments of the present application, a computer readable storage medium is provided.

[0024] In one embodiment, the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the above method.

[0025] The technical solutions provided by the embodiments of the present application can have the following beneficial effects:

[0026] 1. The present application realizes comprehensive perception of transformer area operation data by using multi-source data acquisition equipment such as transformer area intelligent electric meters, photovoltaic inverters and energy storage battery management systems (BMS).

[0027] 2、The application can dynamically adapt to the change of the transformer area topological structure, and update the line parameters in real time, significantly improving the accuracy and timeliness of topological identification and parameter estimation, providing a reliable data basis for overvoltage management;A GCN-LSTM-XGBoost hybrid prediction model is constructed, and a two-stage spatio-temporal prediction training strategy is proposed, which can capture the spatio-temporal characteristics of the transformer area voltage simultaneously, significantly improve the accuracy of overvoltage prediction, and further improve the prediction performance of the model through the Bayesian deep neural network optimization method, providing a scientific decision basis for energy storage collaborative regulation.

[0028] 3、The application proposes a multi-time scale multi-objective optimization model considering the spatio-temporal coupling, and solves it by using a multi-objective optimization method based on NSGA-II, which can dynamically adjust the charging and discharging strategy of the energy storage system under the premise of ensuring the safety of the transformer area voltage, significantly reduce the network loss, effectively solve the overvoltage problem caused by high proportion of distributed photovoltaic access to the distribution transformer area, and significantly improve the safety and economy of the transformer area, providing strong support for efficient operation of the distribution transformer area.

[0029] 4、The application has significant advantages in data quality, topological identification, parameter estimation, overvoltage prediction and energy storage regulation, and can better cope with the challenges brought by high proportion of distributed photovoltaic access.

[0030] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the application. BRIEF DESCRIPTION OF DRAWINGS

[0031] The drawings incorporated into the specification and forming part of the specification, show embodiments consistent with the application, and together with the specification, serve to explain the principles of the application.

[0032] Figure 1 is a flow chart of a distributed energy storage spatio-temporal collaborative regulation method according to an exemplary embodiment;

[0033] Figure 2 is a principle block diagram of a distributed energy storage spatio-temporal collaborative regulation system according to an exemplary embodiment;

[0034] Figure 3 is a structural schematic diagram of a computer device according to an exemplary embodiment;

[0035] Figure 4 is a core module architecture diagram in a distributed energy storage spatio-temporal collaborative regulation method according to an exemplary embodiment;

[0036] Figure 5is a two-stage distribution area topology identification flowchart based on split expectation maximization algorithm in a distributed energy storage space-time collaborative regulation method according to an exemplary embodiment;

[0037] Figure 6 is a GCN-LSTM-XGBoost overvoltage risk space-time prediction structure diagram in a distributed energy storage space-time collaborative regulation method according to an exemplary embodiment;

[0038] Figure 7 is a NSGA-II algorithm flowchart in a distributed energy storage space-time collaborative regulation method according to an exemplary embodiment;

[0039] Figure 8 is a distribution area distributed energy storage space-time collaborative regulation architecture schematic diagram in a distributed energy storage space-time collaborative regulation method according to an exemplary embodiment;

[0040] Figure 9 is a calculation example topology diagram in a distributed energy storage space-time collaborative regulation method according to an exemplary embodiment;

[0041] Figure 10 is a distribution area voltage comparison diagram before and after regulation in a distributed energy storage space-time collaborative regulation method according to an exemplary embodiment. DETAILED DESCRIPTION

[0042] The following description and drawings are illustrative of the specific embodiments herein and are not intended to be limiting. Parts and features of some embodiments can be included or replaced by parts and features of other embodiments. The scope of the embodiments herein includes the whole area of the claims and all available equivalents of the claims. Various embodiments are described in a progressive manner, each focusing on the differences from other embodiments, and the same or similar parts between various embodiments can be mutually referred to.

[0043] Each module in the device or system of the present application can be realized wholly or partially by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the above modules.

[0044] In the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other.

[0045] Figure 1 、 Figure 8 、 Figure 9 and Figure 10 shows one embodiment of the distributed energy storage space-time collaborative regulation method of the present application.

[0046] In this optional embodiment, the distributed energy storage spatiotemporal coordinated control method includes the following steps:

[0047] Step S101: Obtain the operating data of the low-pressure photovoltaic energy storage area and preprocess the operating data;

[0048] Step S102: Based on the preprocessed data, perform topology identification and line parameter estimation for the transformer substations; wherein, the improved DBSCAN clustering algorithm and split expectation maximization algorithm are used to identify the topology of the transformer substations, and the line parameters of the transformer substations are estimated according to the Bayesian filtering framework.

[0049] Step S103: Based on the two-stage spatiotemporal prediction training strategy, construct a spatiotemporal prediction model for transformer area overvoltage risk; use the spatiotemporal prediction model for transformer area overvoltage risk to predict the preprocessed data and obtain the voltage risk probability information of transformer area nodes.

[0050] Step S104: Integrate the topology, line parameters, and overvoltage risk probability information of the transformer area to construct a multi-objective optimization model. Solve the model using the NSGA-II algorithm based on non-dominated sorting and congestion calculation to obtain the distributed energy storage coordinated control strategy.

[0051] The distributed energy storage spatiotemporal coordinated control method provided by the present invention will be described in detail below.

[0052] like Figure 4 The diagram illustrates the core module architecture of the distributed energy storage spatiotemporal coordinated control method. This invention achieves precise perception of the operating status of distribution substations through data-driven substation topology identification and line parameter estimation; it establishes a hybrid prediction model based on GCN-LSTM-XGBoost to accurately predict overvoltage risks in the next hour; and it dynamically adjusts the charging and discharging strategies of the energy storage system through a multi-timescale, multi-objective optimization model to balance substation overvoltage with operational economy. The purpose of this invention is to solve the overvoltage problem caused by high-proportion photovoltaic (PV) grid connection in distribution substations, improve substation voltage qualification rate, reduce network losses, extend energy storage life, and simultaneously enhance PV absorption capacity, thereby providing technical support for the safe, economical, and efficient operation of distribution substations.

[0053] The distributed energy storage spatiotemporal coordinated control method of the present invention comprises the following specific steps:

[0054] I. Data Acquisition and Preprocessing.

[0055] 1. Collecting low-voltage photovoltaic storage area operation data, including voltage, current, power, and other electrical quantity measurement data, photovoltaic output, energy storage system, and load power, and meteorological data including light intensity and temperature, by using intelligent electric meters, photovoltaic inverters, and battery management systems (BMS).

[0056] 2. Data cleaning, alignment, and standardization processing of the collected data. Specifically, abnormal data is identified and removed using the 3σ criterion, missing values are interpolated based on the K-nearest neighbor algorithm, and input features are normalized.

[0057] II. Identification of area topology and parameter estimation.

[0058] 1. Analyzing the voltage-power correlation characteristics of the area based on the improved DBSCAN clustering algorithm, and identifying the area topology structure using the split expectation-maximization (split-EM) method.

[0059] Specifically, the following improvements are made based on the original DBSCAN clustering algorithm:

[0060] 1) Correlation analysis of voltage-power is performed, and the result is used as a weight coefficient to adjust the distance measurement in the clustering process.

[0061]

[0062] where ρ V,P is the correlation coefficient between voltage V and power P, used to measure the degree of linear correlation between them, with a value range of [-1, 1], and the closer the absolute value is to 1, the stronger the correlation; σ U is the standard deviation of voltage V, describing the dispersion of voltage data; and σ P is the standard deviation of power P, describing the dispersion of power data.

[0063] 2) Introducing dynamic adjustment of local density neighborhood radius r, i.e.:

[0064]

[0065] where d(x i , x j ) is the Euclidean distance between samples x i and x j ; k is the number of samples; r i is the dynamic adjustment neighborhood radius of the local density of sample x i ; and j is the number of sample x j .

[0066] 3) Introduce dynamic neighbor point adjustment strategy, set minimum sample number according to local density, that is:

[0067]

[0068] In the formula, MinPts i is the minimum field sample number of sample x i ; n i,neighbor is the number of neighborhood samples of sample x i . The steps of analyzing the voltage-power correlation characteristics of the transformer area based on the improved DBSCAN clustering algorithm are as follows:

[0069] 1) Initialize parameters neighborhood radius, minimum field sample number, prepare input data set, and node voltage-power data set.

[0070] 2) Calculate the voltage-power correlation coefficient as the weight, and adaptively adjust the neighborhood radius r and the minimum field sample number MinPts according to formula (2) and formula (3).

[0071] 3) Perform DBSCAN clustering operation, solve the distance corresponding to the sample with the fastest change rate as the parameter r of the DBSCAN clustering algorithm, verify the reliability through the silhouette coefficient, that is, formula (4), generate MinPts, and output the voltage-power correlation analysis result.

[0072]

[0073] In the formula, S(i) is the silhouette coefficient of sample x i , which is used to evaluate the reliability and compactness of the clustering result, and the value range is [0, 1]; a(i) represents the average distance of i and all other data points in the same cluster; b(i) represents the minimum average distance of i to any remaining cluster. S(i) value close to 1 indicates that the data point has been correctly clustered, and S(i) close to zero means that the data point is located near the boundary of two natural clusters.

[0074] As shown in Figure 5 , it is a two-stage topology identification framework.

[0075] In stage 1, for historical data identification, the unlabeled records can be divided into several smaller samples, each sample including several records, and a record refers to the node measurement value at a time point. The split-EM method is used for topology identification operation in a parallel manner to improve efficiency. For each record [s ij , y ij ], it is the node measurement value of all nodes in the power distribution network at a time point, the vector s ij includes voltage amplitude and phase angle, and the vector y ijThe injected active and reactive power. The topology class T is determined after executing the topology identification procedure ij , which is added to the original record to generate the labeled data [s ij , y ij , T ij ].

[0076] The machine learning training will generate several topology classifiers using all the historical labeled data. The input of the classifier is the node measurements, and the output is the corresponding topology class. In phase 2, the trained topology classifier is used for real-time data recognition. The trained classifier is used to label the real-time measurements, and a reliability analysis is performed to prevent rare cases where the classifier cannot correctly label certain measurements under new topology parameters that may never appear in the historical data. Based on the results of the reliability analysis, reliable labeled data will be sent to the historical labeled database to update the topology classifiers for machine learning training, while unreliable data will be re-identified using a Bayesian recursive model and sent to the labeled database for updating.

[0077] 2. On-line estimation of line parameters using a Bayesian filtering framework to establish a probability distribution model of line impedance.

[0078] On-line estimation of line impedance using a Bayesian probability model to continuously update the posterior distribution and improve the accuracy of parameter estimation. The steps are as follows:

[0079] 1) Parameter initialization, based on typical line parameters in the area, set the prior distribution P(θ) parameter, initialize the posterior distribution. Set the probability distribution of line parameters, the invention believes that the line resistance and reactance parameters both follow normal distribution, i.e.

[0080] 2) Based on the measurement data D, calculate the likelihood function P(D|θ) and update the posterior distribution P(θ|D), the specific calculation formula is as follows:

[0081]

[0082] P(θ|D)∝P(D|θ)P(θ) (6)

[0083] In the formula, n D is the number of data, is the voltage prediction value; V i is the i-th measured voltage value.

[0084] 3) Iterative update, stop iteration when the parameter change rate is less than the threshold value α = 0.01, output the line impedance estimation matrix and the parameter confidence interval.

[0085]

[0086] where θ t is the parameter value at time t; θ t-1 is the parameter value at time t-1.

[0087] 3. Regularly update the topology and line parameters to ensure the real-time and accuracy of the model.

[0088] III. Construction of the spatial-temporal prediction model of overvoltage risk in the transformer area.

[0089] 1. Construction of the GCN-LSTM-XGBoost hybrid prediction model. As shown in FIG. 1, it is the spatial-temporal prediction structure diagram of the overvoltage risk in the GCN-LSTM-XGBoost hybrid prediction model. Figure 6

[0090] 1. LSTM time series feature prediction.

[0091] The long short-term memory (LSTM) is based on the recurrent neural network (RNN) structure form with only short-term memory. The addition of input gate, forget gate and output gate and internal memory unit in the hidden layer solves the long-term dependence of memory information transmission, and the advantage of reverse propagation without layer transmission calculation solves the gradient explosion problem of RNN algorithm. Specifically, the forget gate processes the current input information x t and the short-term memory information K t 1 at the last time and outputs the forgetting vector f t ; then x t and K t1 are superimposed, the long-term memory candidate state is updated through the Tanh function, the input gate decides the retention amount i t of the long-term memory; the forget useless new information and the addition of useful new information complete the update of the cell state, as shown in equation (10); finally, the output gate outputs the current unit long-term memory information and short-term memory, as shown in equation (11):

[0092] f t = σ (W f x t + c f K t-1 + b f ) (8)

[0093] i t = σ (W i × [K t-1 , x t ] + b i ) (9)

[0094] ​

[0095] In the formula, σ∈[0,1] is the forgetting gate parameter, used to filter and remove irrelevant historical information; W f For K t-1 and x t The weights; b × U is the bias parameter; t This is the updated cell state.

[0096] Historical voltage sequences, load power change trends, operating status of distributed photovoltaic and energy storage systems, and meteorological data are input into the LSTM model to obtain the output of the LSTM hidden layer, which is the predicted voltage of the transformer substation nodes for the next 1 hour. This is used as one of the input features of the XGBoost model.

[0097] 2) GCN spatial feature extraction.

[0098] A power system consists of transmission lines, electrical equipment, and buses, and can be represented by a graph of edges and nodes. During GCN operation, node information is transmitted through edge connections. Changes in bus status are also transmitted along the lines to other buses and electrical equipment. The relationship between transmission lines and buses can be converted into a relationship between edges and nodes. Bus voltage, branch current, and line impedance are fundamental characteristics of a power system. By treating buses as nodes and lines as edges, the distribution network can be converted into a graph structure. Each node's information includes voltage, current, and phase characteristics aggregated to the bus; distances between nodes can be considered edge characteristics. The graph topology data and graph characteristic data are represented as G = (Ω... B ,E,A), where Ω B Let E be the set of nodes in the distribution network, E be the set of edges, and A be the adjacency matrix. The hidden layer formula is:

[0099]

[0100] In the formula, H represents the hidden layer result, i.e., the feature representation of each layer of the graph neural network; W h I The weight matrix, identity matrix, and weight matrix of the h-th layer are respectively. The weight matrix; H k This is the input feature matrix of the k-th hidden layer, used in the iterative calculation of the hidden layers of the graph neural network to pass on the feature information of the previous layer and support the subsequent feature transformation and update process.

[0101] 3) Node overvoltage risk prediction based on XGBoost.

[0102] The XGBoost algorithm is an ensemble learning algorithm based on decision trees, essentially an additive model, that is:

[0103]

[0104] where L is the total number of integrated number models; x is the model input vector, i.e., the output of LSTM time series feature prediction and the output of GCN spatial feature extraction; b l (x) is the basic tree model of the lth training; is the model prediction output value, i.e., the overvoltage risk probability of the transformer node.

[0105] The algorithm is used to fuse the node voltage, power and other time series features extracted by LSTM and the spatial features such as transformer node voltage correlation matrix and topology connection, and to predict the node overvoltage risk in the next 1h. The objective function of the XGBoost algorithm is:

[0106]

[0107] where n L is the number of samples; y i is the label value of the ith sample; is the model prediction output value of the ith sample; g is the loss function, which is the mean square error function here; Ω(f l ) is the regularization term, which is used to measure the complexity of the model; n T is the number of leaf nodes of a single tree model; w is the leaf weight; γ and λ are the regularization coefficients; J is the objective function of the XGBoost algorithm; b i is the ith decision tree, which is the basic component unit in the XGBoost model.

[0108] 2. A two-stage spatio-temporal prediction training strategy is proposed.

[0109] The training process of the GCN-LSTM-XGBoost spatio-temporal prediction model is as follows:

[0110] 1) LSTM time series prediction: initialize network parameters, set the number of LSTM layers to 2 and the number of neurons to 64, use the ADAM optimizer, and set the learning rate to 0.001; calculate the prediction value by forward propagation, calculate the loss function and update the network parameters by back propagation, and iterate until convergence; extract time series features such as voltage and power from historical data to predict node voltage values in the next 1h.

[0111] 2) GCN spatial feature extraction: fix the LSTM parameters, initialize the GCN layer parameters, initialize the weight matrix, set the optimizer to Adam and the learning rate to 0.001; the input layer is the node feature matrix, including transformer graph topology data and graph feature data; set 2 layers of graph convolution layers, each with a dimension of 64; train the model by forward propagation and update the node features by formula (12); calculate the gradient by back propagation and update the weight matrix, and iterate until convergence; output the node embedding representation and spatial correlation matrix.

[0112] 3) XGBoost fusion LSTM, GCN layer output features, and input meteorological features, photovoltaic, energy storage and load state, train the model; initialize parameters, set the number of trees to 100, the learning rate to 0.1, the maximum tree depth to 6, and the regularization coefficient to 0.1; set the initial prediction value Iterative training and updating of the prediction value according to the rules of formula (14) - formula (15) and output the node overvoltage risk probability within 1h in the future.

[0113] 3. Bayesian deep neural network optimization method is used to optimize hyperparameters and improve the spatiotemporal prediction performance of the model.

[0114] To improve the expression ability of GCN-LSTM-XGBoost model for uncertain data, Bayesian optimization method is introduced to optimize the model hyperparameters. The process is as follows:

[0115] 1) The hyperparameters have a great influence on the neural network training results. Input meteorological factors and other related data, and output the distributed photovoltaic and load prediction results of the substation area through Bayesian deep neural network. The hidden layer adopts a double-layer probability fully connected layer with 16 units and 8 units. Through efficient learning of adaptive parameters, the robustness of initial learning rate, training rounds and other parameters is improved. The early stopping strategy is used to set the continuous 3 rounds of network loss value change rate less than or equal to 1% to stop training to determine the total training rounds.

[0116] 2) Use variational inference method to train the network to solve the posterior distribution P(W|D), where W is the weight set and D is the training data set. The steps are as follows:

[0117] ① Construct the distribution q(W|θ) to fit P(W|D), and convert the network training target to the minimum difference between q(W|θ) and P(W|D) θ = θ * , that is:

[0118]

[0119] In the formula, θ * is the distribution parameter after q training; divergence[] is the distance function.

[0120] ② Introduce KL divergence to measure the difference between the two, that is:

[0121]

[0122] ③ Through the KL divergence expansion formula, solve the maximum value obtained when the minimum KL divergence is equivalent to L(D, θ), that is:

[0123] L(D, θ) = E q(W|θ)[log P(D|W) + log P(W) - log q(W|θ)] (18)

[0124] wherein E q(W|θ) represents the solution by approximating the conditional likelihood; log P(W) represents the prior probability distribution of the weight distribution following the standard Gaussian distribution; and log q(W|θ) is the probability distribution of the distribution q(W|θ). Assuming that the weights are subject to the Gaussian distribution with 0 as the mean value and independence, the expectation value is obtained by solving the distribution log mean value, and then only the distribution q variance needs to be updated and trained. Based on the optimal gradient descent method for solving the extreme value, the weight distribution variance can be updated, that is:

[0125] σ' m = σ m -g[L(D m ,q),σ m ] (19)

[0126] wherein σ m is the distribution variance of the weight m; and g[] is the calculation iteration factor.

[0127] Four, a distributed energy storage collaborative regulation method considering space-time coupling is proposed.

[0128] 1. Taking the minimum voltage deviation and the minimum network loss as the target, considering the power balance constraint, the node voltage constraint, the branch power flow constraint, the energy storage system constraint, the photovoltaic capacity constraint and the line capacity constraint, a multi-time scale multi-objective optimization model is established.

[0129] 1) The expression formula of the minimum voltage deviation is as follows:

[0130]

[0131] wherein V i represents the voltage per unit of node i; V lmt is the upper and lower limit value of the substation voltage, when V i > 1.07 p.u., V lmt = 1.07 p.u., when V i < 0.93 p.u., V lmt = 0.93 p.u., when the node voltage fluctuates within the allowable range, V lmt = V i ; and T is the regulation period.

[0132] 2) The expression formula of the minimum network loss is as follows:

[0133]

[0134] wherein I ij,t is the current flowing from node i to node j at t.ij This represents the resistance between the lines at nodes i and j.

[0135] A multi-objective optimization model is established using a weighted sum method. Due to the overvoltage control requirements in the photovoltaic-storage area, α = 0.8 and β = 0.2 are chosen, i.e.:

[0136] min f=αf1+βf2 (22)

[0137] The constraints are as follows:

[0138] 1) The formula for expressing the power balance constraint is as follows:

[0139]

[0140] In the formula, V j,t G represents the per-unit voltage value of node j at time t; ij B ij These represent the conductance and susceptance of branch ij, respectively; θ ij,t P is the voltage phase angle between node i and node j at time t; i,t Q i,t P represents the active and reactive power of the load at node i at time t; PV,i,t Q PV,i,t P represents the active and reactive power of the photovoltaic system at node i at time t; ESS,i,t Let be the active power of energy stored at node i at time t. It is positive when energy is discharging and negative when it is charging.

[0141] 2) The formulas for the upper and lower voltage limits are as follows:

[0142] V i,min ≤V i,t ≤V i,max |I ij,t | ≤I ij,max (twenty four)

[0143] In the formula, V i,max V i,min V represents the minimum and maximum allowable voltage values ​​for node i, respectively; i,t Let be the per-unit voltage value of node i at time t;

[0144] 3) The formula for expressing branch capacity constraints is as follows:

[0145] |I ij,t |≤I ij,max (25)

[0146] In the formula, I ij,max I represents the maximum allowable current value through branch ij; ij,t Let be the current value of branch ij at time t;

[0147] 4) The expression formula of energy storage capacity and charge-discharge constraints is as follows:

[0148]

[0149] In the formula, P ESS,i,t is the active power of the energy storage of node i at time t, which is positive when the energy storage is discharged and negative when the energy storage is charged; P ESS,i,max is the maximum charge-discharge power of the energy storage of node i; SOC i,t and ΔSOC i,t are the state of charge and the change of the energy storage device of node i at time t; SOC i,max and SOC i,min are the upper and lower limits of the state of charge of the energy storage device of node i; S ESS,i is the capacity of the energy storage device of node i; Δt is the scheduling time step; SOC i,t+1 and ΔSOC i,t+1 are the state of charge and the change of the energy storage device of node i at time t+1.

[0150] 5) Since the PV inverter has reactive power support capability, the residual reactive power capacity of the photovoltaic inverter is used in cooperation with the energy storage device, so that the voltage out-of-limit problem of the distribution network can be solved, and the following constraints need to be met, and the expression formula of the photovoltaic capacity constraint is as follows:

[0151] 0≤P PV,i,t ≤P PV,i,max (27)

[0152]

[0153] In the formula, S PV,i is the PV capacity configured at node i; P PV,i,t and Q PV,i,t are the active and reactive power of the photovoltaic of node i at time t; P PV,i,max is the maximum active power of the photovoltaic configured at node i.

[0154] 2, a multi-objective optimization method based on non-dominated sorting genetic algorithm (NSGA-II) is used for solving, to find the Pareto optimal solution set, to obtain the long-time scale, short-time scale and real-time collaborative control strategy.

[0155] The core idea is to maintain the diversity of the population through non-dominated sorting and crowding distance calculation, and find the Pareto optimal solution set. In the solving process, the algorithm can maintain the diversity of the population through non-dominated sorting and crowding distance, and continuously optimize the individual fitness in the evolution process. In each generation, the NSGA-II algorithm sorts the individuals in the population into different levels through non-dominated sorting, ensures that the individuals in each level are superior to the individuals in the next level in the dominance relationship, and realizes the effective optimization sorting of the individuals in the population. Based on this, the optimization algorithm can calculate the crowding distance and evaluate the crowding degree of the individual, so as to maintain the diversity of the population. The flow chart of the NSGA-II algorithm is shown in Figure 7 , and the specific steps are as follows:

[0156] 1) Initialization of population: randomly generate an initial summary population, each individual represents a possible solution, that is, the control strategy of the energy storage system; set the population size N = 100; perform real number coding on each individual to represent the charging and discharging power of each energy storage device, and the coding length is the number of energy storage devices M.

[0157] 2) Non-dominated sorting: non-dominated sorting is performed on the individuals in the population, and multiple Pareto frontiers are divided; the dominance relationship of each individual is calculated, and the rule is to calculate the objective function value of each individual. If there is at least one objective function value of a certain individual p not worse than that of another individual q, it is said that p dominates q, otherwise, q dominates p.

[0158] 3) Crowding degree calculation: calculate the crowding degree of the individuals in each Pareto frontier, and the larger the crowding degree, the better the diversity of the solution.

[0159] 4) Selection operation: binary tournament selection method is used to select parent individuals, and individuals with higher non-dominated level or larger crowding degree are selected.

[0160] 5) Crossover and mutation: when the dispersion degree of the early population is high, a small population crossover probability is used, and when the population converges to the optimal Pareto frontier, the crossover probability can be increased to prevent the population from falling into local optimum. The expression of the adaptive crossover operator is shown in equation (29); the mutation operation can effectively expand the search range of the population, and increasing the mutation probability in the later stage can enhance the explorability of the algorithm and avoid prematureness of the algorithm. The expression of the adaptive mutation operator is shown in equation (30):

[0161]

[0162] In the formula, P c ∈[P c,min ,P c,max ] and P v ∈[P v,min ,P v,max ] are the crossover probability and the mutation probability respectively; n is the current cycle; n maxP is the maximum iteration number. Generally, P = 0.1, P = 0.5, P = 0.1, and P = 0.7. c,min c,max v,min v,max

[0163] 6) Merge the parent and child populations: merge the parent and child populations to generate a new population, and perform non-dominated sorting and crowding calculation on the new population, and select the first N individuals as the next generation population by using the improved elitist reservation strategy.

[0164] 7) Determine the termination condition: if the iteration number n reaches the maximum or the objective function converges, output the Pareto optimal solution set, otherwise n = n + 1, and return to step 2).

[0165] Figure 2 An embodiment of the distributed energy storage space-time collaborative regulation method, system, device and medium of the application is shown.

[0166] In this optional embodiment, the distributed energy storage space-time collaborative regulation system comprises:

[0167] The data preprocessing module 201 is configured to acquire operation data of a low-voltage light storage area and pre-process the operation data.

[0168] The topology and parameter identification module 202 is configured to perform topology identification and line parameter estimation of the area based on the pre-processed data; wherein the topology identification of the area is performed by using an improved DBSCAN clustering algorithm and a split expectation maximization algorithm, and the line parameter estimation of the area is performed according to a Bayesian filtering framework.

[0169] The overvoltage risk prediction module 203 is configured to construct an area overvoltage risk space-time prediction model according to a two-stage space-time prediction training strategy, and to predict the pre-processed data by using the area overvoltage risk space-time prediction model to obtain voltage risk probability information of nodes in the area.

[0170] The collaborative optimization decision module 204 is configured to fuse the topology structure, line parameters and overvoltage risk probability information of the area, construct a multi-objective optimization model, and solve the model by using an NSGA-II algorithm based on non-dominated sorting and crowding calculation to obtain a distributed energy storage collaborative regulation strategy.

[0171] In one embodiment, a computer device is provided, which can be a server, and the internal structure diagram of the computer device can be as shown in Figure 3 ​​​​As shown in the figure. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is used to store static information and dynamic information data. The network interface of the computer device is used to communicate with the external terminal through the network connection. The computer program is executed by the processor to implement the steps in the above method embodiments.

[0172] Those skilled in the art can understand that, Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0173] In addition, the present application also provides a computer device comprising a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0174] In addition, the present application also provides a computer readable storage medium having a computer program stored thereon, which is executed by a processor to implement the steps in the above method embodiments.

[0175] Those skilled in the art can understand that all or part of the processes in the above embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above embodiments. Any reference to memory, storage, database or other medium used in the embodiments of the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0176] The present application is not limited to the structures described above and shown in the drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present application is only limited by the appended claims.

Claims

1. A method for spatiotemporal coordinated regulation of distributed energy storage, characterized in that, The method includes the following steps: Acquire operational data of the low-pressure photovoltaic energy storage area and preprocess the operational data; Based on the preprocessed data, topology identification and line parameter estimation of the transformer substations are performed. Specifically, the improved DBSCAN clustering algorithm and split expectation-maximization algorithm are used for topology identification of the transformer substations, and the line parameters of the transformer substations are estimated based on the Bayesian filtering framework. Based on a two-stage spatiotemporal prediction training strategy, a spatiotemporal prediction model for transformer area overvoltage risk is constructed. Using the spatiotemporal prediction model for transformer area overvoltage risk, the preprocessed data is predicted to obtain the voltage risk probability information of transformer area nodes. By integrating the topology, line parameters, and overvoltage risk probability information of the transformer substation, a multi-objective optimization model is constructed. The NSGA-II algorithm based on non-dominated sorting and congestion calculation is used to solve the model, resulting in a distributed energy storage coordinated control strategy.

2. The distributed energy storage spatiotemporal coordinated control method according to claim 1, characterized in that, The topology identification of transformer substations using the improved DBSCAN clustering algorithm and the split expectation-maximization algorithm includes the following steps: An improved DBSCAN clustering algorithm is obtained by improving the original DBSCAN algorithm. An improved DBSCAN clustering algorithm was executed to perform voltage-power correlation feature analysis on the preprocessed data and obtain the clustering results. Based on the clustering results, a two-stage topology identification framework is used, combined with the split expectation maximization algorithm, to complete the topology identification of the transformer area.

3. The distributed energy storage spatiotemporal coordinated control method according to claim 2, characterized in that, The improved DBSCAN clustering algorithm is used to perform voltage-power correlation feature analysis on the preprocessed data to obtain the clustering results, including the following steps: Initialize the neighborhood radius and minimum neighborhood sample size parameters, and use the preprocessed data as input; The voltage-power correlation coefficient is calculated as a weight, and the neighborhood radius and minimum number of neighborhood samples are dynamically adjusted according to the local density. Perform DBSCAN clustering, select the distance corresponding to the point where the sample distance changes most rapidly as the final clustering parameter, and use the silhouette coefficient to verify the reliability of the clustering results, and output the clustering results.

4. The distributed energy storage spatiotemporal coordinated control method according to claim 2, characterized in that, The process of identifying transformer area topology based on clustering results, using a two-stage topology identification framework, and combining the split expectation-maximization algorithm includes the following steps: In the historical data identification phase, unlabeled records are divided into sample sets containing measurement values ​​of several time points, and topology identification is performed in parallel using the split expectation-maximization algorithm; after topology identification is completed, a topology category label is assigned to each record to generate labeled data; A topology classifier is trained using all historical labeled data. The topology classifier takes node measurement values ​​as input and outputs the corresponding topology category. In the real-time data identification stage, the trained topology classifier is used to label the real-time measurement values, and the credibility analysis of the labeling results is performed to obtain the topology identification results of the transformer area.

5. The distributed energy storage spatiotemporal coordinated control method according to claim 1, characterized in that, The estimation of line parameters for a transformer area based on the Bayesian filtering framework includes the following steps: Based on typical line parameters in the transformer area, a prior probability distribution model of line resistance and reactance is constructed. The likelihood function is calculated based on the preprocessed data, and the posterior probability distribution of the line parameters is updated iteratively using Bayes' theorem. When the rate of change of the parameter estimate is less than the preset threshold, the iteration terminates and the estimated result and confidence interval of the line impedance are output.

6. The distributed energy storage spatiotemporal coordinated control method according to claim 1, characterized in that, The construction of the spatiotemporal prediction model for transformer area overvoltage risk based on the two-stage spatiotemporal prediction training strategy includes the following steps: A hybrid prediction model is constructed by fusing the LSTM temporal feature prediction layer, the GCN spatial feature extraction layer, and the XGBoost risk prediction layer. A hybrid prediction model is trained based on a two-stage spatiotemporal prediction training strategy. By using the Bayesian deep neural network optimization algorithm, the hyperparameters of the trained hybrid prediction model are optimized to obtain a spatiotemporal prediction model for overvoltage risk in transformer areas.

7. The distributed energy storage spatiotemporal coordinated control method according to claim 6, characterized in that, The training of the hybrid prediction model based on the two-stage spatiotemporal prediction training strategy includes the following steps: In the first stage, the LSTM temporal feature prediction layer and the GCN spatial feature extraction layer are trained separately. The LSTM temporal feature prediction layer adopts a network structure with two hidden layers, each with 64 neurons, and is trained using the Adam optimizer with a learning rate of 0.

001. The GCN spatial feature extraction layer is configured with two graph convolutional layers, each with a dimension of 64, and is trained using the Adam optimizer. In the second stage, the parameters of the LSTM temporal feature prediction layer and the GCN spatial feature extraction layer are fixed, and the XGBoost risk prediction layer is trained. The number of trees is set to 100, the learning rate is 0.1, the maximum tree depth is 6, and the regularization coefficient is 0.

1. The final prediction value is obtained by iterative update.

8. The distributed energy storage spatiotemporal coordinated control method according to claim 1, characterized in that, Based on the topology, line parameters, and overvoltage risk probability information of the integrated transformer area, a multi-objective optimization model is constructed, including: By integrating the topology, line parameters, and overvoltage risk probability information of the transformer substation, and taking into account power balance constraints, node voltage constraints, branch power flow constraints, energy storage system constraints, photovoltaic capacity constraints, and line capacity constraints, a multi-objective optimization model is constructed with the goals of minimizing voltage deviation and network loss. The power balance constraint is expressed by the following formula: In the formula, V j,t G represents the per-unit voltage value of node j at time t; ij B ij These represent the conductance and susceptance of branch ij, respectively; θ ij,t P is the voltage phase angle between node i and node j at time t; i,t Q i,t P represents the active and reactive power of the load at node i at time t; PV,i,t Q PV,i,t P represents the active and reactive power of the photovoltaic system at node i at time t; ESS,i,t Let be the active power of the energy stored at node i at time t. It is positive when the energy is discharging and negative when it is charging. The formulas for expressing the upper and lower voltage limits are as follows: V i,min ≤V i,t ≤V i,max |I ij,t |≤I ij,max ; In the formula, V i,max V i,min V represents the minimum and maximum allowable voltage values ​​for node i, respectively; i,t Let be the per-unit voltage value of node i at time t; The formula for expressing the branch capacity constraint is as follows: |I ij,t |≤I ij,max ; In the formula, I ij,max I represents the maximum allowable current value through branch ij; ij,t Let be the current value of branch ij at time t; The energy storage capacity and charge / discharge constraints are expressed by the following formulas: In the formula, P ESS,i,t P represents the active power of energy stored at node i at time t, which is positive during energy discharge and negative during charging; ESS,i,max Maximum charge / discharge power for energy storage at node i; SOC i,t ΔSOC i,t These represent the state of charge (SOC) and its change at time t for energy storage device i; i,max SOC i,min These are the upper and lower limits of the state of charge of the energy storage device at node i, respectively; S ESS,i Δt represents the capacity of the energy storage device at node i; Δt represents the scheduling time step; SOC i,t+1 ΔSOC i,t+1 For the energy storage device at node i, the state of charge and its change at time t+1; The formula for expressing the photovoltaic capacity constraint is as follows: 0≤P PV,i,t ≤P PV,i,max ; In the formula, S PV,i P represents the photovoltaic capacity configured at node i; PV,i,t Q PV,i,t P represents the active and reactive power of the photovoltaic system at node i at time t; PV,i,max This represents the maximum active power of the photovoltaic system configured at node i.

9. The distributed energy storage spatiotemporal coordinated control method according to claim 8, characterized in that, The process of using the NSGA-II algorithm based on non-dominated sorting and congestion calculation to obtain the distributed energy storage coordinated control strategy includes the following steps: Initialize the population, randomly generate individuals representing the energy storage control strategy, and use real number encoding to represent the charging and discharging power of each energy storage device; Non-dominated ranking of individuals in the population is used to classify Pareto front ranks; Calculate individual crowding to maintain population diversity; The offspring population is generated through binary tournament selection, adaptive crossover, and mutation operations. Merge parent and offspring populations and employ an elite preservation strategy to select the next generation; The evolutionary process is repeated until the termination condition is met, and the Pareto optimal solution set is output as a distributed energy storage coordinated control strategy.

10. A distributed energy storage spatiotemporal coordinated control system, characterized in that, The system includes: The data preprocessing module is used to acquire the operating data of the low-voltage photovoltaic storage area and preprocess the operating data. The topology and parameter identification module is used to perform topology identification and line parameter estimation of transformer substations based on preprocessed data. Specifically, the improved DBSCAN clustering algorithm and split expectation-maximization algorithm are used for topology identification of transformer substations, and the line parameters of transformer substations are estimated based on the Bayesian filtering framework. The overvoltage risk prediction module is used to construct a spatiotemporal prediction model for transformer area overvoltage risk based on a two-stage spatiotemporal prediction training strategy; and to use the spatiotemporal prediction model for transformer area overvoltage risk to predict the voltage risk probability information of transformer area nodes. The collaborative optimization decision module is used to integrate the topology of the transformer area, line parameters and overvoltage risk probability information, construct a multi-objective optimization model, and solve it using the NSGA-II algorithm based on non-dominated sorting and congestion calculation to obtain the distributed energy storage collaborative control strategy.

11. The distributed energy storage spatiotemporal coordinated control system according to claim 10, characterized in that, The topology identification of transformer substations using the improved DBSCAN clustering algorithm and the split expectation-maximization algorithm includes the following steps: An improved DBSCAN clustering algorithm is obtained by improving the original DBSCAN algorithm. An improved DBSCAN clustering algorithm was executed to perform voltage-power correlation feature analysis on the preprocessed data and obtain the clustering results. Based on the clustering results, a two-stage topology identification framework is used, combined with the split expectation maximization algorithm, to complete the topology identification of the transformer area.

12. The distributed energy storage spatiotemporal coordinated control system according to claim 11, characterized in that, The improved DBSCAN clustering algorithm is used to perform voltage-power correlation feature analysis on the preprocessed data to obtain the clustering results, including the following steps: Initialize the neighborhood radius and minimum neighborhood sample size parameters, and use the preprocessed data as input; The voltage-power correlation coefficient is calculated as a weight, and the neighborhood radius and minimum number of neighborhood samples are dynamically adjusted according to the local density. Perform DBSCAN clustering, select the distance corresponding to the point where the sample distance changes most rapidly as the final clustering parameter, and use the silhouette coefficient to verify the reliability of the clustering results, and output the clustering results.

13. The distributed energy storage spatiotemporal coordinated control system according to claim 11, characterized in that, The process of identifying transformer area topology based on clustering results, using a two-stage topology identification framework, and combining the split expectation-maximization algorithm includes the following steps: In the historical data identification phase, unlabeled records are divided into sample sets containing measurement values ​​of several time points, and topology identification is performed in parallel using the split expectation-maximization algorithm; after topology identification is completed, a topology category label is assigned to each record to generate labeled data; A topology classifier is trained using all historical labeled data. The topology classifier takes node measurement values ​​as input and outputs the corresponding topology category. In the real-time data identification stage, the trained topology classifier is used to label the real-time measurement values, and the credibility analysis of the labeling results is performed to obtain the topology identification results of the transformer area.

14. The distributed energy storage spatiotemporal coordinated control system according to claim 10, characterized in that, The estimation of line parameters for a transformer area based on the Bayesian filtering framework includes the following steps: Based on typical line parameters in the transformer area, a prior probability distribution model of line resistance and reactance is constructed. The likelihood function is calculated based on the preprocessed data, and the posterior probability distribution of the line parameters is updated iteratively using Bayes' theorem. When the rate of change of the parameter estimate is less than the preset threshold, the iteration terminates and the estimated result and confidence interval of the line impedance are output.

15. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 9.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 9.

Images