Optical storage flexible DC operation prediction method based on deep learning

Abstract

The invention relates to the technical field of operation prediction, in particular to an optical storage flexible direct current operation prediction method based on deep learning, and the method achieves the precise evaluation of photovoltaic power fluctuation through the analysis of historical operation data and the combination of a power state conversion matrix, guarantees that the energy storage charging and discharging rate can be matched with the load demand distribution, and improves the prediction precision. Power scheduling is optimized, topology analysis and power flow path identification are carried out, topology weight calculation is carried out by using a graph convolutional network, power flow can be dynamically adjusted, the adaptability of an operation mode is improved, the influence of power fluctuation on system stability is reduced, power matching is carried out by adopting an adversarial generative network, photovoltaic output is optimized, and the system stability is improved. The energy storage release path is more reasonable, the power loss is reduced, the overall power distribution efficiency is improved, and the load compensation strategy is optimized through short-period fluctuation detection and long-period trend fitting. And variable correction is performed based on error calculation and anomaly detection, so that the reliability of regulation and control response is improved.

Description

Technical Field

[0001] The present invention relates to the field of operation prediction technology, and in particular to a method for predicting the operation of a PV-storage flexible direct current (DC-SCADA) system based on deep learning. Background Art

[0002] The field of operation prediction technology aims to use historical data and real-time monitoring information, combined with mathematical modeling and machine learning methods, to predict the future operating status of the system, optimize scheduling decisions, improve system stability, reduce operating costs, and enhance the safety, economy and reliability of the system.

[0003] The purpose of the deep learning-based PV-storage flexible DC operation prediction method is to establish a high-precision prediction model through learning and feature extraction of historical operation data, so as to predict the system operation status in advance, provide decision support for grid dispatching and energy management, improve the operation stability of the PV-storage flexible DC system, optimize energy dispatching, improve the new energy absorption capacity, and reduce the problems of wind and solar power curtailment, thereby improving the safety and economy of the power system.

[0004] When dealing with photovoltaic power fluctuations, existing technologies are unable to accurately capture complex power conversion relationships, resulting in a lack of targeted energy storage charging and discharging strategies. In terms of power flow optimization, there is a lack of in-depth analysis of the topological structure, resulting in limited ability to adjust the power flow path, affecting the flexibility of the overall operating mode, and making it difficult to cope with nonlinear changes, resulting in uneven energy storage release and an inability to adapt to dynamic changes under different load environments, causing power regulation strategies to lag. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a deep learning-based prediction method for the operation of PV-storage flexible direct current (DC-SSC).

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a method for predicting the operation of a PV-storage flexible direct current system based on deep learning, comprising the following steps: Step 1: Sensors are used to extract photovoltaic and energy storage load operating status data, calculate photovoltaic fluctuations, analyze energy storage charge and discharge rates, and calculate load demand distribution. Furthermore, power trend analysis and time window mean calculations are performed to generate a power state transition matrix. Step 2: Based on the power state transition matrix, perform PV power smoothing, energy storage charging and discharging strategy matching, and load power response adjustment, perform power path optimization and operation mode division, and generate the PV-storage operation mode; Step 3: Based on the PV-storage operation mode, perform topology analysis, power flow path identification, and node power exchange calculation. Use a graph convolutional network to calculate topology weights and adjust power flow directions to generate a PV-storage flexible direct current power flow adjustment matrix. Step 4: Based on the PV-storage flexible direct current power flow adjustment matrix, an adversarial generative network is used to perform energy storage charge and discharge matching, load power compensation, and photovoltaic output optimization, perform energy storage release path adjustment and power collaborative calculation, and generate a power optimization allocation matrix; Step 5: Based on the power optimization allocation matrix, perform short-term fluctuation detection, long-term trend fitting and regulation response calculation, perform power optimization and load compensation calculation, and generate photovoltaic power storage power control parameters; Step 6: Based on the photovoltaic storage power control parameters, perform photovoltaic power forecasting, load demand calculation and energy storage discharge matching, perform timing forecast correction and power regulation adjustment, and generate a photovoltaic storage flexible direct current operation forecast result; Step 7: Based on the PV-storage flexible DC operation prediction results, perform error calculation, anomaly detection and adjustment optimization, perform variable correction and power path optimization, and generate PV-storage flexible DC operation optimization prediction results.

[0007] As a further solution of the present invention, the specific steps of generating the power state transition matrix are: Sensors are used to extract photovoltaic and storage load operating status data, segment photovoltaic power generation data time series, calculate power output fluctuations, extract energy storage charging and discharging change trends, calculate rate distribution, extract load power demand curves, calculate load response time, and generate a basic power status data set. Based on the power state basic data set, the photovoltaic power fluctuation gradient is calculated, the short-term trend is extracted, the power conversion rate of the energy storage system is calculated, the charging and discharging mode intervals are divided, the load power adjustment amplitude is calculated, and the timing demand matching is performed to generate a power state transition parameter matrix; Based on the power state transition parameter matrix, the photovoltaic power timing is analyzed, short-cycle power smoothing is performed, the energy storage regulation response rate is calculated, load power compensation calculation is performed, the power conversion node distribution is calculated, and state transition optimization is performed to generate a power state transition matrix.

[0008] As a further solution of the present invention, the specific steps of generating the photovoltaic storage operation mode are: Based on the power state transition matrix, photovoltaic power smoothing adjustment is performed, short-term fluctuation suppression is performed, power change point distribution is calculated, mutation point buffering is optimized, power curve smoothness is calculated, and low-frequency fluctuation correction is performed to generate a photovoltaic power smoothing data set; Based on the photovoltaic power smoothing data set, the power regulation range of the energy storage system is calculated, and dynamic matching of charge and discharge conversion is performed, the energy storage compensation power demand is calculated, the charge and discharge strategy is adjusted, the response rate of the energy storage system is calculated, and power compensation optimization is performed to generate a photovoltaic storage matching adjustment data set; Based on the photovoltaic-storage matching adjustment data set, the load power demand change rate is calculated, and the load power compensation calculation is performed, the dynamic adjustment range of the energy storage system is calculated, and the load matching degree is optimized. The power scheduling mode is calculated, and the operation mode classification is performed to generate the photovoltaic-storage system operation mode.

[0009] As a further solution of the present invention, the specific steps of generating the PV-storage flexible DC power flow adjustment matrix are as follows: Based on the photovoltaic storage operation mode, topological node data is extracted, and the connectivity between nodes is calculated. The power flow path is calculated, and the power exchange amount between adjacent nodes is calculated. High-power transmission nodes are extracted and the energy exchange intensity is evaluated. The topological coupling relationship is calculated, and the node connection weight is allocated to generate the photovoltaic storage topology structure analysis results; Based on the results of the optical storage topology analysis, a graph convolutional network is used to extract the power transmission topology features, calculate the power loss of each path, identify high-loss paths, extract core power transmission nodes, calculate power scheduling priorities, calculate topology weight adjustment parameters, and optimize power paths to generate optimized parameters for the optical storage power flow path. Based on the optimization parameters of the photovoltaic storage power flow path, the power flow matrix is ​​calculated, and the topological path adaptation analysis is performed. The energy exchange stability index is calculated, and the low stability path adjustment is performed. The impact of power flow on the energy storage system is calculated, and the power compensation calculation of the energy storage system is performed. The optimized power transmission path is calculated, and the topological node adjustment is performed to generate the photovoltaic storage flexible direct current power flow adjustment matrix.

[0010] As a further solution of the present invention, the graph convolutional network is based on the formula: in: For the Feature representation of layer nodes, is the activation function, For nodes The set of neighbor nodes of is the normalization factor, For the The trainable weight matrix of the layer, For adjacent nodes In the The feature representation of the layer, For the current node In the The feature representation of the layer, For nodes With node The power transfer efficiency between For nodes The transmission power at It is the power level index adjustment node, reflecting the load adjustment strategy. is the power regulation coefficient, For nodes With node The physical distance between is the attenuation factor.

[0011] As a further solution of the present invention, the specific steps of generating the power optimization allocation matrix are: Based on the PV-storage flexible direct current power flow adjustment matrix, a generative adversarial network is used to calculate the boundaries of the charge and discharge time periods, calculate the power compensation demand, extract the load-side compensation parameters, calculate the energy storage charge and discharge conversion rate, adjust the power scheduling priority, calculate the energy regulation period, perform power stability correction, and generate energy storage regulation matching parameters; Based on the energy storage regulation and matching parameters, the load demand range, power supply and demand relationship, and photovoltaic power adjustment range are calculated, and output power balancing control is performed. The power supply and demand matching degree is calculated, and dynamic load optimization is performed. The combined power output of energy storage and photovoltaics is calculated, and power matching optimization is performed to generate a power regulation matching matrix. Based on the power regulation matching matrix, the energy storage system release path is calculated, power is dynamically allocated, the combined output ratio of photovoltaic and energy storage is calculated, and power balancing control is performed. The load-side compensation capacity is calculated, and load supply and demand coordination is performed. The dynamic release adjustment amount of energy storage is calculated, and the power load is finally matched to generate a power optimization distribution matrix.

[0012] As a further solution of the present invention, the adversarial generative network is based on the formula: in: is the generator network, is the discriminator network, To counter the objective function, is the distribution of real power data, is the real power data sample, is the distribution of the latent variable, is a random noise input, Indicates the discriminator's response to the real data during the charging and discharging period and load demand power The predicted probability under the condition Represents the generator based on the noise input and charge-discharge conversion efficiency , charging power and energy regulation cycle The generated power data samples, represents the predicted probability of the discriminator for the generated power data under the corresponding conditions, Divide the boundaries of the charging and discharging time periods, is the load demand power, is the charge-discharge conversion efficiency, is the charging power, It is the energy regulation cycle.

[0013] As a further solution of the present invention, the specific steps of generating the photovoltaic power control parameter are: Based on the power optimization allocation matrix, short-term power fluctuation data is extracted, the time-series power change rate is calculated, and the fluctuation amplitude is divided into zones and statistics. The short-term power change points on the load side are extracted, the power adjustment rate of the mutation point is calculated, abnormal fluctuations are screened, the short-term power response time of the energy storage system is calculated, and the charge and discharge switching rate is corrected to generate the short-term fluctuation adjustment parameters. Based on the short-term fluctuation adjustment parameter, extract the long-term power change trend data, calculate the change rate of the historical power curve, and perform trend smoothing. Calculate the long-term power compensation capability of the energy storage system, perform charge and discharge cycle matching calculations, extract the load demand change trend, calculate the long-term load power adjustment rate, and perform load-side power supply and demand matching analysis to generate the long-term trend adjustment parameter. Based on the long-term trend adjustment parameters, the power regulation response rate data is extracted, the dynamic load response interval is calculated, and the power regulation rate is adjusted. The dynamic compensation parameters of energy storage charging and discharging are calculated, and the power flow balance calculation is performed. The dynamic adjustment range of load compensation is calculated, and the load regulation cycle is optimized to generate the photovoltaic storage power control parameters.

[0014] As a further solution of the present invention, the specific steps of generating the prediction result of the PV-storage flexible direct current operation are: Based on the photovoltaic storage power control parameters, historical photovoltaic power data is extracted, the time series change characteristics are calculated, and short-term trend calculations are performed. The photovoltaic power generation power mutation points are extracted, the fluctuation range change rate is calculated, and the power fluctuation range at future moments is predicted. The trend curve of photovoltaic power output is extracted, the photovoltaic output power adjustment amount in the short term is calculated, and the photovoltaic power prediction results are generated.

[0015] Based on the photovoltaic power prediction results, load demand change data is extracted, the time series change range of the load power demand is calculated, and the power demand interval calculation is performed. The available power distribution data is extracted, the charge and discharge adjustment capacity is calculated, and the energy storage supply and demand balance calculation is performed. The power supply and demand difference data is extracted, the power distribution error compensation range is calculated, and the load demand prediction parameters are generated.

[0016] Based on the load demand forecast parameters, the power timing forecast error data is extracted, the error compensation correction range is calculated, and the power forecast deviation is adjusted. The energy storage and photovoltaic combined power distribution data is extracted, the dynamic energy storage adjustment rate is calculated, and power distribution optimization is performed. The load power supply and demand balance analysis data is extracted, the optimal distribution path for power regulation is calculated, and the photovoltaic storage flexible direct current operation prediction result is generated.

[0017] As a further solution of the present invention, the specific steps of generating the optimization prediction results of the PV-storage flexible direct current operation are: Based on the PV-storage flexible direct current operation prediction results, the predicted power data and actual power data are extracted, the time distribution of the data error is calculated, and the error change rate analysis is performed. The error deviation interval is extracted, the error cumulative impact factor is calculated, and abnormal fluctuation screening is performed. The abnormal operation status data is extracted, the key parameters of the abnormal power fluctuation are calculated, and the error interval is classified to generate the prediction error and abnormal status data; Based on the prediction error and abnormal state data, extracting variables of power flow affected by the error, calculating the offset of the variables on the power flow, and calculating the variable adjustment amplitude, extracting corrected variable data, calculating the power output characteristics after the variable correction, and redistributing the power flow, extracting adjusted power path data, calculating the power transmission stability in the corrected path, and performing power transmission adjustment to generate variable correction and power path adjustment data; Based on the variable correction and power path adjustment data, the control parameters of the corrected power state are extracted, the corrected power transmission rate is calculated, the power flow balance is adjusted, the adjusted load power matching data is extracted, the load demand and power adjustment matching degree are calculated, and power flow optimization calculation is performed. The transmission parameters after power path optimization are extracted, the power flow adjustment range is calculated, and the optimization prediction results of the PV-storage flexible direct current operation are generated.

[0018] Compared with the prior art, the advantages and positive effects of the present invention are: 1. In this invention, by analyzing historical operating data and combining it with the power state transition matrix, an accurate assessment of photovoltaic power fluctuations is achieved, ensuring that the energy storage charge and discharge rate can match the load demand distribution and optimizing power scheduling; 2. In this invention, through topology analysis and power flow path identification, and using graph convolutional networks to calculate topological weights, power flow can be dynamically adjusted, improving the adaptability of operating modes and reducing the impact of power fluctuations on system stability; 3. This invention uses a generative adversarial network for power matching to optimize photovoltaic output, making the energy storage release path more rational, reducing power loss, and improving overall power distribution efficiency. It also optimizes load compensation strategies through short-term fluctuation detection and long-term trend matching. Variable correction is performed based on error calculation and anomaly detection, improving the reliability of control responses. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic diagram of the main steps of the present invention; Figure 2 This is a schematic diagram of the refinement of S1 of the present invention; Figure 3 This is a schematic diagram of the refinement of S2 of the present invention; Figure 4 This is a schematic diagram of the refinement of S3 of the present invention; Figure 5 This is a schematic diagram of the refinement of S4 of the present invention; Figure 6 This is a schematic diagram of the refinement of S5 of the present invention; Figure 7 This is a schematic diagram of the refinement of S6 of the present invention; Figure 8 This is a detailed schematic diagram of S7 of the present invention. DETAILED DESCRIPTION

[0020] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0021] See also Figure 1 The present invention provides a technical solution: a method for predicting the operation of a PV-storage flexible direct current (DC-SCADA) system based on deep learning, comprising the following steps: S1: Sensors are used to extract photovoltaic and energy storage load operating status data, calculate photovoltaic fluctuations, analyze energy storage charge and discharge rates, and calculate load demand distribution. Furthermore, power trend analysis and time window mean calculation are performed to generate a power state transition matrix. S2: Based on the power state transition matrix, PV power smoothing, energy storage charging and discharging strategy matching, and load power response adjustment are performed. Power path optimization and operation mode division are performed to generate the PV-storage operation mode. S3: Based on the PV-storage operation mode, perform topology analysis, power flow path identification, and node power exchange calculation. Graph convolutional networks are used to calculate topology weights and adjust power flows, generating a PV-storage flexible direct current power flow adjustment matrix. S4: Based on the PV-storage flexible direct current power flow adjustment matrix, an adversarial generative network is used to perform energy storage charge and discharge matching, load power compensation, and PV output optimization. Energy storage release path adjustment and power coordination calculation are performed to generate a power optimization allocation matrix. S5: Based on the power optimization allocation matrix, short-term fluctuation detection, long-term trend fitting, and regulation response calculation are performed. Power optimization and load compensation calculations are performed to generate PV-storage power control parameters. S6: Based on the PV-storage power control parameters, perform PV power forecasting, load demand calculation, and energy storage discharge matching, perform timing forecast correction and power regulation adjustment, and generate PV-storage flexible direct current operation forecast results; S7: Based on the PV-storage flexible DC operation prediction results, perform error calculation, anomaly detection and adjustment optimization, perform variable correction and power path optimization, and generate the PV-storage flexible DC operation optimization prediction results.

[0022] See also Figure 2 , the specific steps to generate the power state transition matrix are: S101: Extracting photovoltaic and energy storage load operating status data through sensors, segmenting photovoltaic power generation data time series, calculating power output fluctuations, extracting energy storage charging and discharging change trends, and performing rate distribution calculations. Extracting load power demand curves and calculating load response times generate a basic power status data set. S102: Based on the power state basic data set, calculate the photovoltaic power fluctuation gradient, extract the short-term trend, calculate the energy storage system power conversion rate, divide the charging and discharging mode intervals, calculate the load power adjustment range, and perform time sequence demand matching to generate a power state transition parameter matrix; S103: Based on the power state transition parameter matrix, analyze the photovoltaic power time series, perform short-cycle power smoothing, calculate the energy storage regulation response rate, perform load power compensation calculation, calculate the power conversion node distribution, and perform state transition optimization to generate a power state transition matrix; S101: Based on the PV-storage load operating status data, a K-means clustering algorithm is used. The Euclidean distance formula is used to calculate the similarity between samples, and the PV power generation data time series is clustered and segmented. The power output fluctuation is calculated. The discrete wavelet transform method is used to perform multi-scale decomposition of the time series, and the high-frequency components are extracted to represent short-term fluctuations. The least squares method is used to fit the curve to model the energy storage charging and discharging change trend, and the linear fitting parameters, including the slope and intercept, are calculated. The distribution probability density function is used to fit the charge and discharge rate distribution. The polynomial fitting method is used to construct the load power demand curve, and the least squares regression is performed to calculate the fitting coefficient. The sliding window method is used to calculate the load response time, with the window size set to 30 seconds. The time series offset is calculated to generate the power status basic data set. S102: Based on the power state basic data set, calculate the photovoltaic power fluctuation gradient, use the five-point central difference method to perform numerical differentiation on the time series data, calculate the gradient of adjacent time points, and generate a gradient sequence to extract short-term trends. Use the fast Fourier transform to perform spectral analysis on the power sequence, extract the main frequency component, and calculate the amplitude and phase. Calculate the power conversion rate of the energy storage system. Use the double exponential smoothing method with smoothing coefficients α=0.3 and β=0.7 to smooth the energy storage power time series and divide the charging and discharging mode intervals. Use the dynamic time warping algorithm to perform pattern matching on historical power data and calculate the optimal alignment path. Calculate the load power adjustment amplitude. Use the autoregressive integral moving average model to predict future power demand and perform time series demand matching. Use the dynamic time warping method to calculate the minimum Euclidean distance between the current load demand sequence and the historical optimal matching sequence, and generate the power state transition parameter matrix. S103: Based on the power state transition parameter matrix, the photovoltaic power time series is analyzed. The wavelet packet decomposition method is used to decompose the photovoltaic power into three layers, and the low-frequency trend component is extracted to perform short-period power smoothing. The Kalman filter algorithm is used to set the state transfer matrix A and the observation matrix H, and the optimal estimated power value is calculated. The energy storage regulation response rate is calculated. The second-order difference method is used to calculate the energy storage output power and generate a response rate time series. The load power compensation calculation is performed. The linear programming method is used to construct a set of constraint equations and solve the optimal power compensation value. The power conversion node distribution is calculated. The hierarchical clustering algorithm is used to set a distance threshold of 0.1 to group the power conversion events and perform state transition optimization. The Markov decision process is used to solve the optimal strategy and generate the power state transition matrix.

[0023] See also Figure 3 , the specific steps to generate the PV storage operation mode are: S201: Based on the power state transition matrix, the photovoltaic power is smoothly adjusted and short-term fluctuations are suppressed. The power change point distribution is calculated, and the mutation point buffer is optimized. The power curve smoothness is calculated and low-frequency fluctuation correction is performed to generate a photovoltaic power smoothing data set. S202: Based on the PV power smoothing dataset, calculate the energy storage system power regulation range, perform dynamic matching of charge and discharge conversion, calculate the energy storage compensation power requirement, adjust the charge and discharge strategy, calculate the energy storage system response rate, and perform power compensation optimization to generate a PV-storage matching adjustment dataset. S203: Based on the PV-storage matching adjustment data set, calculate the load power demand change rate, perform load power compensation calculation, calculate the dynamic adjustment range of the energy storage system, optimize the load matching degree, calculate the power scheduling mode, and classify the operation mode to generate the PV-storage system operation mode; S201: Based on the power state transition matrix, photovoltaic power smoothing adjustment is performed. The empirical mode decomposition algorithm is used, and the decomposition layer number is set to ten. The photovoltaic power time series data is decomposed, the intrinsic mode function components are extracted, and the high-frequency intrinsic mode function components are reconstructed to suppress short-term fluctuations. The median filter algorithm is used, and the window size is set to five. The reconstructed power data is smoothed and the power change point distribution is calculated. The edge detection algorithm is used, and the low threshold is set to 0.1 and the high threshold is set to 0.3. The first-order derivative of the power time series data is calculated, and the mutation point is detected. The mutation point buffer is optimized. The Gaussian smoothing filter algorithm is used to smooth and fit the change point data, and the power curve stability is calculated. The mean square error calculation method is used to evaluate the error of the smoothed power series and perform low-frequency fluctuation correction. The wavelet transform is used, and the mother wavelet is set to the fourth-order Dai Bessie wavelet. Four-layer decomposition is performed, and the low-frequency component is reconstructed by inverse transform to generate a photovoltaic power smoothing data set. S202: Based on the photovoltaic power smoothing data set, calculate the power regulation range of the energy storage system, use the variational mode decomposition algorithm, set the number of decomposition modes to three, decompose the power time series data, and calculate the power range of each mode, perform dynamic matching of charge and discharge conversion, use the long short-term memory network, set the number of hidden layer nodes to 64 and the time step to 10, train the historical charge and discharge data, and predict the current energy storage charge and discharge conversion point, calculate the energy storage compensation power demand, use the Bayesian estimation method, set the prior distribution to Gaussian distribution, fit the historical load demand data, and calculate the expected value of compensation power, adjust the charge and discharge strategy, use the dynamic programming algorithm, set the state transfer matrix, and calculate the optimal charge and discharge path, calculate the response rate of the energy storage system, use the autoregressive sliding average model, set the order to two and one, fit the energy storage response time series data, and perform power compensation optimization, use the Lagrangian relaxation optimization algorithm, set the constraints, and solve the optimal power compensation distribution to generate the photovoltaic storage matching adjustment data set; S203: Based on the photovoltaic storage matching adjustment data set, calculate the load power demand change rate, adopt the exponentially weighted moving average algorithm, set the weight factor to 0.4, perform weighted calculation on the load power data, and extract the change rate time series to perform load power compensation calculation, adopt the optimal power allocation algorithm, set the objective function, and solve the optimal compensation power allocation scheme, calculate the dynamic adjustment range of the energy storage system, adopt the Markov prediction model, set the state transition probability matrix, and calculate the future adjustment range, optimize the load matching degree, adopt the divergence calculation method, calculate the deviation between the current load demand distribution and the historical best matching distribution, and adjust the load power distribution, calculate the power scheduling mode, adopt the clustering algorithm, set the number of cluster centers to four, classify the power scheduling data, and classify the operation mode, adopt the support vector machine classification algorithm, set the kernel function to the radial basis kernel, and perform classification training on the scheduling mode to generate the photovoltaic storage system operation mode.

[0024] See also Figure 4 The specific steps for generating the PV-storage flexible DC power flow adjustment matrix are as follows: S301: Based on the PV-storage operation mode, extract topological node data, calculate inter-node connectivity, calculate power flow paths, calculate power exchange between adjacent nodes, extract high-power transmission nodes, evaluate energy exchange intensity, calculate topological coupling relationships, and assign node connection weights to generate PV-storage topological structure analysis results; S302: Based on the results of the PV-storage topology analysis, a graph convolutional network is used to extract the power transmission topology features, calculate the power loss of each path, identify high-loss paths, extract core power transmission nodes, calculate power scheduling priorities, calculate topology weight adjustment parameters, and optimize power paths to generate optimized parameters for the PV-storage power flow path. S303: Based on the optimized parameters of the PV-storage power flow path, the power flow matrix is ​​calculated, and topological path adaptation analysis is performed. The energy exchange stability index is calculated, and low-stability path adjustment is performed. The impact of power flow on the energy storage system is calculated, and power compensation calculation of the energy storage system is performed. The optimized power transmission path is calculated, and topological node adjustments are performed to generate the PV-storage flexible direct current power flow adjustment matrix. S301: Based on the photovoltaic storage operation mode, extract the topological node data, use the depth-first search algorithm, set the starting node number to zero, traverse the grid topology, store the connection relationship between nodes, calculate the connectivity between nodes, use the adjacency matrix calculation method to build the node connection matrix, and calculate the matrix eigenvalue, calculate the power flow path, use the shortest path search algorithm, set the path weight to the impedance value between nodes, and calculate the power transmission amount of each path, calculate the power exchange amount between adjacent nodes, use the energy conservation equation, set the input and output power balance conditions of each node, and solve the exchange power value , extract high-power transmission nodes, use the node degree centrality calculation method, set the threshold to twice the total power mean, and screen high-power nodes to evaluate the energy exchange intensity, use the power flow density calculation method, set the measurement time window to sixty seconds, and calculate the power change rate per unit time, calculate the topological coupling relationship, use the Granger causality analysis method, set the lag order to two, and calculate the coupling coefficient between nodes, perform node connection weight allocation, use the adaptive weight allocation method based on node power flow, set the power flow intensity as the weighting factor, and calculate the final connection weight to generate the optical storage topology analysis results; S302: Based on the results of the photovoltaic storage topology analysis, a graph convolutional network is used to extract the power transmission topology features. The input data is set as the node connection matrix, the number of hidden layer neurons is 64, and the gradient descent method is used for parameter optimization. The power loss of each path is calculated. The DC power flow calculation method is used, the node injection power is set as a known variable, and the line loss value is solved to identify high-loss paths. The threshold segmentation method is used, the loss threshold is set to 1.5 times the total loss mean, and the high-loss paths are marked. The core power transmission nodes are extracted. The betweenness centrality calculation method is used to calculate the betweenness value of each node, and the top 10% of the nodes are selected to calculate the power scheduling priority. The fuzzy hierarchical analysis method is used to set the weight factors including power flow, node load and topological centrality, and calculate the priority score. The topology weight adjustment parameters are calculated. The Lagrange multiplier method is used, the constraint conditions are set as the power conservation equation, and the optimal weight adjustment scheme is solved to optimize the power path. The distributed optimization algorithm is used, the objective function is set to minimize power loss, and the optimal path allocation scheme is calculated to generate the photovoltaic storage power flow path optimization parameters. S303: Based on the optimization parameters of the photovoltaic power flow path, the power flow matrix is ​​calculated. The power flow calculation method is used to set the node injection power and branch impedance parameters, and the power flow is solved. The topological path adaptation analysis is performed. The shortest path matching method is used to set the path weight as the line impedance, and the fitness of each path is calculated. The energy exchange stability index is calculated. The dynamic timing stability analysis method is used to set the time window to 120 seconds and calculate the energy fluctuation variance. The low-stability path adjustment is performed. The heuristic search method is used to set the adjustment target to minimize power fluctuations, and the path weight is adjusted to calculate the impact of power flow on the energy storage system. In order to understand the impact of energy storage, a power distribution model based on discrete event simulation is adopted, and the event trigger condition is set as energy storage charging and discharging switching. The change in energy storage state is calculated to perform power compensation calculation for the energy storage system. An optimization scheduling algorithm is adopted, and the objective function is set to minimize the energy storage regulation cost. The optimal compensation strategy is calculated to calculate the optimized power transmission path. An incremental update algorithm is adopted to set the initial path as the state before optimization, and the optimal path after adjustment is calculated to adjust the topological nodes. A node reconstruction method based on power flow balance is adopted, and the adjustment rule is set to maximum power load balancing. The node connection relationship is updated to generate a power flow adjustment matrix for the PV-storage flexible direct current system.

[0025] Graph convolutional network, according to the formula: in: For the Feature representation of layer nodes, is the activation function, For nodes The set of neighbor nodes of is the normalization factor, For the The trainable weight matrix of the layer, For adjacent nodes In the The feature representation of the layer, For the current node In the The feature representation of the layer, For nodes With node The power transfer efficiency between For nodes The transmission power at It is the power level index adjustment node, reflecting the load adjustment strategy. is the power regulation coefficient, For nodes With node The physical distance between is the attenuation factor; Execution process: First, the topology modeling of the PV-storage flexible DC network is carried out, the connection relationship of each power node is defined, the adjacency matrix is ​​constructed, and the initial characteristics of the node are determined. and adjacent node features And set the trainable weight matrix To optimize the feature map, and then calculate the normalization factor To balance the impact of different nodes on power transmission characteristics and utilize power transmission efficiency Adjust the influence weight of adjacent nodes on the target node, By line resistance and inductance Jointly decide to use exponential decay function to express the influence of line impedance on power loss, and introduce power level index adjustment term To dynamically adjust the importance of power nodes, According to the load balancing requirements, ensure that high-power nodes have appropriate weights in the topology feature extraction process, and the physical distance As an impact factor, combined with the attenuation factor Calculate the spatial attenuation effect of power flow, finally sum the weighted features of each adjacent node, and calculate the next layer node representation based on the target node’s own features , through the activation function Perform nonlinear transformation to obtain the output after optimizing the power topology characteristics, which is used for high-loss path identification, core power transmission node extraction and power scheduling priority calculation. It also optimizes the power transmission path and generates the optimized parameters of the photovoltaic power flow path, thereby improving the accuracy and stability of the flexible direct current system operation prediction.

[0026] See also Figure 5 , the specific steps to generate the power optimization allocation matrix are: S401: Based on the PV-storage flexible direct current power flow adjustment matrix, a generative adversarial network is used to calculate the boundaries of the charge and discharge time periods, calculate the power compensation demand, extract the load-side compensation parameters, calculate the energy storage charge and discharge conversion rate, adjust the power scheduling priority, calculate the energy regulation period, perform power stability correction, and generate the energy storage regulation matching parameters. S402: Based on the energy storage regulation matching parameters, the load demand range, power supply and demand relationship, and PV power adjustment range are calculated, and output power balancing control is performed. The power supply and demand matching degree is calculated, and dynamic load optimization is performed. The combined power output of energy storage and PV is calculated, and power matching optimization is performed to generate a power regulation matching matrix. S403: Based on the power regulation matching matrix, the energy storage system release path is calculated, power is dynamically allocated, the combined output ratio of photovoltaic and energy storage is calculated, power balancing is controlled, the load-side compensation capacity is calculated, load supply and demand are coordinated, the energy storage dynamic release adjustment amount is calculated, and the power load is finally matched to generate the power optimization allocation matrix. S401: Based on the PV-storage flexible direct current power flow adjustment matrix, an adversarial generative network is used, the number of generator and discriminator network layers is set to three, the activation function is the hyperbolic tangent function, the historical charge and discharge data are trained, and the boundaries of the charge and discharge time periods are generated to calculate the power compensation demand. The optimization method based on Lagrange multipliers is used to set the objective function to minimize the energy storage compensation cost, and calculate the optimal compensation demand. The load side compensation parameters are extracted, and the parameter optimization method based on gradient descent is used. The learning rate is set to 0.01, the load demand data is fitted, and the optimal compensation parameters are calculated. Calculate the energy storage charge and discharge conversion rate, use the autoregressive integral moving average model, set the order parameters to two and one, model the charge and discharge conversion process, calculate the conversion rate, adjust the power scheduling priority, use the fuzzy comprehensive evaluation method, set the fuzzy membership function, calculate the energy storage scheduling priority, calculate the energy regulation period, use the dynamic time warping algorithm, set the time step to 60 seconds, and calculate the optimal regulation period, perform power stability correction, use the least squares fitting method to model the historical power fluctuation data, calculate the adjusted power sequence, and generate the energy storage regulation matching parameters; S402: Based on the energy storage adjustment matching parameters, calculate the load demand interval, use the quantile regression method, set the quantile parameter to 0.5, perform interval estimation on the load data, and calculate the upper and lower bounds, calculate the power supply and demand relationship, use the supply and demand balance model based on linear regression, set the regression coefficient to the least squares estimate of the historical supply and demand data, and calculate the current power supply and demand matching degree, calculate the photovoltaic power adjustment range, use the adaptive adjustment factor method, set the adjustment step to 0.05, and calculate the photovoltaic output correction value, perform output power balance control, use the model predictive control method, set the control time domain to 30 seconds, and solve the optimal power distribution, calculate Calculate the power supply and demand matching degree, use the cosine similarity calculation method to calculate the similarity between the current power distribution and the ideal matching distribution, and perform dynamic load optimization. Use a nonlinear optimization algorithm, set the objective function to minimize load fluctuations, and calculate the optimal load adjustment parameters. Calculate the combined power output of energy storage and photovoltaics. Use a method based on the power allocation matrix, set the weight parameter to the historical output ratio of energy storage and photovoltaics, and calculate the current combined output value. Perform power matching optimization. Use a genetic algorithm with a population size of 50, a crossover rate of 0.8, and a mutation rate of 0.05. Calculate the optimized power matching parameters and generate a power regulation matching matrix. S403: Based on the power regulation matching matrix, calculate the release path of the energy storage system, adopt the optimization method based on network flow, set the constraint condition as the maximum discharge power of energy storage, and calculate the optimal release path to perform dynamic power allocation. Adopt the allocation method based on convex optimization, set the optimization goal as minimizing energy storage loss, and solve the dynamic energy storage allocation scheme, calculate the combined output ratio of photovoltaic and energy storage, adopt the calculation method based on weighted average method, set the weight coefficient as the historical output ratio of photovoltaic and energy storage, and calculate the current optimal combined output ratio, perform power balancing control, adopt the hierarchical control method, set the first layer as local power regulation, the second layer as global power balance, and calculate the control parameters. Calculate the load side compensation capacity, adopt a method based on load elasticity analysis, set the elasticity coefficient as the least squares fitting value of the historical load adjustment response curve, and calculate the current load compensation capacity, coordinate load supply and demand, adopt a supply and demand matching algorithm based on dynamic adjustment, set the adjustment step size to 0.05, and calculate the current optimal load matching strategy, calculate the dynamic release adjustment amount of energy storage, adopt a method based on Markov decision process, set the state transition probability matrix, and calculate the future energy storage release adjustment value, perform the final power load matching, adopt a matching method based on optimal scheduling, set the optimization goal to minimize the power mismatch, and solve the optimal power matching solution to generate a power optimization allocation matrix.

[0027] Adversarial generation network, according to the formula: in: is the generator network, is the discriminator network, To counter the objective function, is the distribution of real power data, is the real power data sample, is the distribution of the latent variable, is a random noise input, Indicates the discriminator's response to the real data during the charging and discharging period and load demand power The predicted probability under the condition Represents the generator based on the noise input and charge-discharge conversion efficiency , charging power and energy regulation cycle The generated power data samples, represents the predicted probability of the discriminator for the generated power data under the corresponding conditions, Divide the boundaries of the charging and discharging time periods, is the load demand power, is the charge-discharge conversion efficiency, is the charging power, It is the energy regulation cycle; Execution process: First, based on the PV-storage flexible DC power flow adjustment matrix, a real power data sample set is constructed. , and define , discriminator Receive real samples and the boundaries of the charging and discharging time periods and load demand power As input, by calculating its discriminant probability Evaluate the authenticity of input data, generator With random noise As the initial input, combined with the charge and discharge conversion efficiency In order to reflect the energy loss of the energy storage system, the charging power is introduced As a key variable to optimize the charge and discharge conversion rate, and adjust the energy cycle As a time dimension adjustment factor, simulated power data is generated on this basis , the minimum and maximum game optimization objective function is used during training , and continuously iteratively update the parameters of the generator and the discriminator, so that the generated data gradually approaches the real power data distribution, which is used for the calculation of the boundary of the charging and discharging time period, the calculation of power compensation requirements, the extraction of load-side compensation parameters, the calculation of energy storage charging and discharging conversion rate, the adjustment of power scheduling priority, the calculation of energy regulation cycle, the power stability correction and the generation of energy storage regulation matching parameters, thereby improving the accuracy and stability of the operation prediction of the photovoltaic storage flexible direct current system.

[0028] See also Figure 6 ,The specific steps for generating the photovoltaic storage power control parameters are: S501: Based on the power optimization allocation matrix, extract short-term power fluctuation data, calculate the time-series power change rate, perform fluctuation amplitude partition statistics, extract short-term power change points on the load side, calculate the power adjustment rate at the mutation point, filter abnormal fluctuations, calculate the short-term power response time of the energy storage system, and perform charge and discharge switching rate correction to generate short-term fluctuation adjustment parameters. S502: Based on the short-term fluctuation adjustment parameters, extract long-term power change trend data, calculate the rate of change of the historical power curve, perform trend smoothing, calculate the long-term power compensation capability of the energy storage system, perform charge and discharge cycle matching calculations, extract the load demand change trend, calculate the long-term load power adjustment rate, perform load-side power supply and demand matching analysis, and generate long-term trend adjustment parameters. S503: Based on the long-term trend adjustment parameters, extract the power regulation response rate data, calculate the dynamic load response range, and adjust the power regulation rate. Calculate the dynamic compensation parameters for energy storage charging and discharging, perform power flow balance calculations, calculate the dynamic adjustment range of load compensation, optimize the load regulation cycle, and generate the PV storage power control parameters. S501: Based on the power optimization allocation matrix, extract the short-period power fluctuation data, use the wavelet transform method, set the mother wavelet to the fourth-order Dai Bessie wavelet, decompose the power time series data into three layers, extract the high-frequency component, calculate the time series power change rate, use the finite difference method, set the time step to 0.05 seconds, calculate the first-order derivative of the short-period power sequence, and obtain the change rate, perform fluctuation amplitude partition statistics, use the histogram distribution statistics method, set the number of partitions to ten, and calculate the fluctuation frequency of each interval, extract the short-term power change point on the load side, use the Canny edge detection algorithm, set the high threshold to 0.3 and the low threshold to 0.1, and perform load power The gradient of the sequence is calculated and the change points are screened. The power adjustment rate of the mutation point is calculated. The sliding window regression method is used, the window length is set to five data points, and the local change rate is calculated. Abnormal fluctuations are screened. The three sigma criterion is used, the threshold is set to the mean plus or minus three times the standard deviation, and abnormal data is marked. The short-cycle power response time of the energy storage system is calculated. The autoregressive sliding average model is used, and the order parameters are set to two and one. The response time series of the energy storage system is fitted and the time delay is calculated. The charge and discharge switching rate is corrected. The Kalman filter method is used, the state transfer matrix and the observation matrix are set, and the corrected charge and discharge rate is calculated to generate the short-cycle fluctuation adjustment parameter. S502: Based on the short-term fluctuation adjustment parameters, extract the long-term power change trend data, use the empirical mode decomposition method, set the decomposition level to five, perform modal decomposition on the power sequence, extract the low-frequency component, calculate the rate of change of the historical power curve, use the difference operation method, set the difference order to one, and calculate the rate of change of each time step, perform trend smoothing, use the exponential weighted moving average method, set the weight factor to 0.2, and calculate the smoothing trend, calculate the long-term power compensation capacity of the energy storage system, use the Markov prediction model, set the state transfer matrix, and calculate the future energy storage compensation capacity, and perform charging and discharging. The power cycle matching calculation uses the dynamic time warping method, sets the time step to 60 seconds, and calculates the optimal matching cycle. The load demand change trend is extracted. The long short-term memory network is used, and the number of hidden layer nodes is set to 64. Historical load data is trained and future trends are predicted. The long-term load power adjustment rate is calculated. The multivariate regression analysis method is used, and the input variables are set as historical load data and external environmental parameters. The optimal adjustment rate is calculated. The load-side power supply and demand matching analysis is performed. The Lagrangian optimization method is used, and the constraint condition is set as power supply and demand balance. The matching results are calculated and the long-term trend adjustment parameters are generated. S503: Based on the long-term trend adjustment parameters, the power regulation response rate data is extracted. The Fourier transform method is used to set the sampling frequency to 50 Hz and extract the main frequency component to calculate the dynamic load response interval. The sliding window variance analysis method is used to set the window size to 30 seconds and calculate the variance of the load response interval to adjust the power regulation rate. The particle swarm optimization algorithm is used to set the number of particles to 50 and the maximum number of iterations to 100, and the optimal regulation rate is solved. The dynamic compensation parameters of energy storage charging and discharging are calculated. The Bayesian estimation method is used to set the prior distribution to Gaussian distribution and calculate the optimal compensation parameters. The power flow balance calculation is performed. The network flow calculation method is used to set the line impedance and node injection power and calculate the power flow. The dynamic adjustment range of load compensation is calculated. The fuzzy control method is used to set the fuzzy membership function and calculate the dynamic adjustment interval. The load regulation cycle is optimized. The reinforcement learning method is used to set the reward function to minimize the load regulation cost and train the optimal regulation strategy to generate the photovoltaic storage power control parameters.

[0029] See also Figure 7 The specific steps to generate the prediction results of PV-storage flexible direct current operation are as follows: S601: Based on the photovoltaic power storage control parameters, extract the photovoltaic power historical data, calculate the time series change characteristics, and perform short-term trend calculation. Extract the photovoltaic power generation power mutation point, calculate the fluctuation range change rate, and predict the power fluctuation range at future moments. Extract the photovoltaic power output trend curve, calculate the photovoltaic output power adjustment amount in the short term, and generate the photovoltaic power prediction result. S602: Based on the photovoltaic power forecast results, extract load demand change data, calculate the time series change range of the load power demand, perform power demand interval calculation, extract available power allocation data, calculate the charge and discharge adjustment capacity, perform energy storage supply and demand balance calculation, extract power supply and demand difference data, calculate the power allocation error compensation range, and generate load demand forecast parameters; S603: Based on the load demand forecast parameters, extract the power time series forecast error data, calculate the error compensation correction range, and adjust the power forecast deviation. Extract the energy storage and photovoltaic combined power allocation data, calculate the dynamic energy storage adjustment rate, and optimize the power allocation. Extract the load power supply and demand balance analysis data, calculate the optimal power allocation path, and generate the PV-storage flexible direct current operation forecast result. S601: Based on the photovoltaic power storage power control parameters, extract the photovoltaic power historical data, use the sliding window segmentation method, set the window size to 60 seconds, divide the photovoltaic power time series data into windows, and extract the power mean in each time window, calculate the time series change characteristics, use the first-order difference method, set the difference step to one time step, perform differential operation on the photovoltaic power data, calculate the time series change rate, perform short-term trend calculation, use the fast Fourier transform method, set the sampling frequency to 50 Hz, perform frequency domain analysis on the time series data, extract the main frequency components, extract the photovoltaic power mutation point, use the edge detection algorithm, set the low threshold value to 0.1 and the high threshold value to 0.3, calculate the gradient of the power time series data, identify the mutation point, and calculate the wave The dynamic range change rate is calculated by using the sliding mean square error calculation method, setting the window size to five data points, performing statistical analysis on the power change rate, calculating the fluctuation amplitude, and predicting the power fluctuation range in the future. The long short-term memory network method is used, setting the number of hidden layer neurons to 64 and the time step to 10, training the historical photovoltaic power data, and predicting the future power fluctuation range. The trend curve of photovoltaic power output is extracted. The cubic spline interpolation method is used, setting the number of interpolation nodes to 10, interpolating the photovoltaic power data, and generating a trend curve. The photovoltaic output power adjustment amount in the short term is calculated. The regression analysis method is used, setting the regression order to 2, modeling the photovoltaic power trend data, calculating the adjustment amount, and generating the photovoltaic power prediction result. S602: Based on the photovoltaic power forecast results, extract the load demand change data, use the autoregressive sliding average method, set the order parameters to two and one, fit the load demand time series data, and extract the change trend. Calculate the time series change range of the load power demand. Use the quantile regression method, set the quantile parameter to 0.5, perform interval estimation on the load data, calculate the upper and lower bounds, calculate the power demand interval, use the dynamic programming method, set the state transfer matrix, and calculate the optimal demand interval. Extract the available power allocation data, use the least squares estimation method, set the target variable to photovoltaic and energy storage power output, and calculate the optimal allocation parameters. , calculate the charge and discharge adjustment capability, use the particle swarm optimization algorithm, set the number of particles to fifty, the maximum number of iterations to one hundred, and solve the optimal charge and discharge adjustment parameters, perform energy storage supply and demand balance calculation, use the Lagrangian optimization method, set the constraint condition to energy storage power conservation, and calculate the supply and demand balance state, extract the power supply and demand difference data, use the time series difference calculation method, set the time step to sixty seconds, calculate the energy storage and load power, and extract the supply and demand difference, calculate the power allocation error compensation range, use the error correction method based on Kalman filtering, set the state transfer matrix and observation matrix, and calculate the error compensation amount, generate the load demand forecast parameters; S603: Based on the load demand forecast parameters, extract the power time series forecast error data, use the mean square error calculation method to calculate the error between the forecast result and the actual power data, generate an error sequence, calculate the error compensation correction range, use the Markov decision process method to set the state transition probability matrix, and calculate the error correction value, adjust the power forecast deviation, use the dynamic time warping method, set the time step to 60 seconds, and calculate the optimal correction path, extract the energy storage and photovoltaic combined power distribution data, use the Bayesian estimation method, set the prior distribution to Gaussian distribution, and calculate the optimal combined power output. Calculate the dynamic energy storage regulation rate, adopt the adaptive filtering method, set the filter gain to 0.1, and calculate the regulation rate, perform power distribution optimization, adopt the nonlinear constraint optimization method, set the objective function to minimize the energy storage loss, and solve the optimal distribution plan, extract the load power supply and demand balance analysis data, adopt the supply and demand balance analysis method based on the entropy weight method, set the weight factor to the historical load data, and calculate the current supply and demand balance state, calculate the optimal distribution path for power regulation, adopt the weighted shortest path search method, set the path weight to the power transmission loss, and calculate the optimal path, generate the PV storage flexible direct current operation prediction result.

[0030] See also Figure 8 The specific steps to generate the optimization prediction results of PV-storage flexible direct current operation are as follows: S701: Based on the PV-storage flexible direct current (FDC) operation prediction results, extract the predicted power data and actual power data, calculate the time distribution of the data error, perform error change rate analysis, extract the error deviation interval, calculate the error cumulative impact factor, perform abnormal fluctuation screening, extract abnormal operation status data, calculate the key parameters of abnormal power fluctuations, classify the error intervals, and generate prediction error and abnormal status data. S702: Based on the prediction error and abnormal state data, extract the variables affected by the error in the power flow, calculate the offset of the variables on the power flow, calculate the variable adjustment amplitude, extract the corrected variable data, calculate the power output characteristics after the variable correction, redistribute the power flow, extract the adjusted power path data, calculate the power transmission stability in the corrected path, perform power transmission adjustment, and generate variable correction and power path adjustment data; S703: Based on the variable correction and power path adjustment data, extract the control parameters of the corrected power state, calculate the corrected power transmission rate, adjust the power flow balance, extract the adjusted load power matching data, calculate the matching degree between load demand and power adjustment, perform power flow optimization calculation, extract the transmission parameters after power path optimization, calculate the power flow adjustment range, and generate the PV-storage flexible direct current operation optimization prediction results; S701: Based on the prediction results of the PV-storage flexible direct current operation, the predicted power data and the actual power data are extracted. The time series alignment method is used, and the time step is set to 60 seconds. The two sets of data are time series aligned and the power error after alignment is calculated. The time distribution of the data error is calculated. The sliding window mean calculation method is used, and the window size is set to five time steps. The error data is smoothed and an error time distribution curve is constructed. The error change rate analysis is performed. The first-order difference calculation method is used to perform differential operations on the error time series data and calculate the error change rate. The error deviation interval is extracted. The percentile quantile analysis method is used, and the upper boundary is set to the ninety-fifth percentile and the lower boundary is the fifth percentile. The error deviation interval is screened and the error cumulative impact factor is calculated. The exponential weighting is used. Using the moving average method, the weight factor is set to 0.3, the error data is weighted and summed, and the cumulative impact factor is calculated to screen abnormal fluctuations. Using the three sigma criterion, the threshold is set to the mean plus or minus three times the standard deviation, and abnormal error data is screened out. Abnormal operating state data is extracted. Using the Markov hidden state estimation method, the state transition probability matrix is ​​set, and abnormal operating states are identified. The key parameters of abnormal power fluctuations are calculated. Using the Fourier transform method, the sampling frequency is set to 50 Hz, and the abnormal power fluctuation data is subjected to spectral analysis, and the main frequency components are calculated. Error intervals are classified. Using the K-means clustering method, the number of cluster centers is set to three, and cluster analysis is performed on the error interval data. Different error types are classified to generate prediction error and abnormal state data. S702: Based on the prediction error and abnormal state data, extract the variables affected by the error in the power flow, adopt the method based on Granger causality analysis, set the lag order to two, and calculate the degree of influence of the error data on the power flow variables, calculate the offset of the variables on the power flow, adopt the least squares regression method, set the independent variable as the error data and the dependent variable as the power flow parameter, and calculate the regression coefficient, calculate the variable adjustment amplitude, adopt the adaptive step gradient descent method, set the initial learning rate to 0.01, optimize and adjust the power flow variables, extract the corrected variable data, adopt the Kalman filter method, set the state transfer matrix and observation matrix, filter the variable data, and output the corrected variable sequence, calculate the power output characteristics after the variable correction, and adopt dynamic regression. Regression prediction method, set the regression window size to ten time steps, and predict the adjusted power output trend, redistribute the power flow, use the network flow optimization calculation method, set the node power injection constraint, and calculate the corrected power flow, extract the adjusted power path data, use the shortest path weighted optimization method, set the path weight to power loss, and screen the optimal transmission path, calculate the power transmission stability in the corrected path, use the stability analysis method based on fuzzy membership calculation, set the stability membership function, and calculate the stability parameters of each path, adjust the power transmission, use the path optimization method based on genetic algorithm, set the population size to fifty, the crossover rate to 0.8, the mutation rate to 0.05, and solve the optimal power transmission path, generate variable correction and power path adjustment data; S703: Based on the variable correction and power path adjustment data, the control parameters of the corrected power state are extracted, and a power control method based on an adaptive adjustment factor is adopted. The adjustment factor is set as the corrected variable data, and the control parameters are calculated. The corrected power transmission rate is calculated. The second-order difference calculation method is used to perform a differential operation on the corrected power transmission data and calculate the power transmission rate. The power flow balance is adjusted. The Lagrangian relaxation optimization method is used to set the objective function to minimize the power flow imbalance and calculate the balance adjustment parameters. The adjusted load power matching data is extracted. The matching method based on supply and demand ratio analysis is used to set the load demand and supply ratio and calculate the matching. The degree parameter is used to calculate the matching degree between load demand and power adjustment. The cosine similarity calculation method is used to set the load data before and after power adjustment and calculate the similarity. The power flow optimization calculation is performed. The nonlinear constraint optimization method is used to set the constraint condition as power supply and demand balance, and the optimal power allocation scheme is solved. The transmission parameters after power path optimization are extracted. The hierarchical clustering method is used to set the number of clustering layers to three. The optimized power path data is clustered and the cluster center value is calculated. The power flow adjustment range is calculated. The method based on Markov decision process is used to set the state transition probability matrix and calculate the adjustment range of power flow to generate the optimization prediction results of PV-storage flexible direct current operation.

[0031] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A method for predicting the operation of a PV-storage flexible direct current system based on deep learning, characterized in that: The following steps are involved: Step 1: Extract the operating status data of photovoltaic and energy storage loads through sensors, calculate photovoltaic fluctuations, analyze the energy storage charge and discharge rates, and calculate the load demand distribution. Then, perform power trend analysis and time window mean calculation to generate a power state transition matrix. Step 2: Based on the power state transition matrix, perform PV power smoothing, energy storage charging and discharging strategy matching, and load power response adjustment, perform power path optimization and operation mode division, and generate the PV-storage operation mode. Step 3: Based on the PV-storage operation mode, perform topology analysis, power flow path identification, and node power exchange calculation. Use a graph convolutional network to calculate topology weights and adjust power flow directions to generate a PV-storage flexible direct current power flow adjustment matrix. Step 4: Based on the PV-storage flexible direct current power flow adjustment matrix, an adversarial generative network is used to perform energy storage charge and discharge matching, load power compensation, and photovoltaic output optimization, perform energy storage release path adjustment and power collaborative calculation, and generate a power optimization allocation matrix; Step 5: Based on the power optimization allocation matrix, perform short-term fluctuation detection, long-term trend fitting and regulation response calculation, perform power optimization and load compensation calculation, and generate photovoltaic power storage power control parameters; Step 6: Based on the photovoltaic storage power control parameters, perform photovoltaic power forecasting, load demand calculation and energy storage discharge matching, perform timing forecast correction and power regulation adjustment, and generate a photovoltaic storage flexible direct current operation forecast result; Step 7: Based on the PV-storage flexible DC operation prediction results, perform error calculation, anomaly detection and adjustment optimization, perform variable correction and power path optimization, and generate PV-storage flexible DC operation optimization prediction results.

2. The deep learning-based PV-storage flexible direct current operation prediction method according to claim 1 is characterized in that: The specific steps of generating the power state transition matrix are: Sensors are used to extract photovoltaic and storage load operating status data, segment photovoltaic power generation data time series, calculate power output fluctuations, extract energy storage charging and discharging change trends, calculate rate distribution, extract load power demand curves, calculate load response time, and generate a basic power status data set. Based on the power state basic data set, the photovoltaic power fluctuation gradient is calculated, the short-term trend is extracted, the power conversion rate of the energy storage system is calculated, the charging and discharging mode intervals are divided, the load power adjustment amplitude is calculated, and the timing demand matching is performed to generate a power state transition parameter matrix; Based on the power state transition parameter matrix, the photovoltaic power timing is analyzed, short-cycle power smoothing is performed, the energy storage regulation response rate is calculated, load power compensation calculation is performed, the power conversion node distribution is calculated, and state transition optimization is performed to generate a power state transition matrix.

3. The deep learning-based PV-storage flexible DC operation prediction method according to claim 1 is characterized in that: The specific steps of generating the solar-storage operation mode are as follows: Based on the power state transition matrix, photovoltaic power smoothing adjustment is performed, short-term fluctuation suppression is performed, power change point distribution is calculated, mutation point buffering is optimized, power curve smoothness is calculated, and low-frequency fluctuation correction is performed to generate a photovoltaic power smoothing data set; Based on the photovoltaic power smoothing data set, the power regulation range of the energy storage system is calculated, and dynamic matching of charge and discharge conversion is performed, the energy storage compensation power demand is calculated, the charge and discharge strategy is adjusted, the response rate of the energy storage system is calculated, and power compensation optimization is performed to generate a photovoltaic storage matching adjustment data set; Based on the photovoltaic-storage matching adjustment data set, the load power demand change rate is calculated, and the load power compensation calculation is performed, the dynamic adjustment range of the energy storage system is calculated, and the load matching degree is optimized. The power scheduling mode is calculated, and the operation mode classification is performed to generate the photovoltaic-storage system operation mode.

4. The method for predicting the operation of a PV-storage flexible direct current system based on deep learning according to claim 1, characterized in that: The specific steps of generating the PV-storage flexible DC power flow adjustment matrix are as follows: Based on the photovoltaic storage operation mode, topological node data is extracted, and the connectivity between nodes is calculated. The power flow path is calculated, and the power exchange amount between adjacent nodes is calculated. High-power transmission nodes are extracted and the energy exchange intensity is evaluated. The topological coupling relationship is calculated, and the node connection weight is allocated to generate the photovoltaic storage topology structure analysis results; Based on the results of the optical storage topology analysis, a graph convolutional network is used to extract the power transmission topology features, calculate the power loss of each path, identify high-loss paths, extract core power transmission nodes, calculate power scheduling priorities, calculate topology weight adjustment parameters, and optimize power paths to generate optimized parameters for the optical storage power flow path. Based on the optimization parameters of the photovoltaic storage power flow path, the power flow matrix is ​​calculated, and the topological path adaptation analysis is performed. The energy exchange stability index is calculated, and the low stability path adjustment is performed. The impact of power flow on the energy storage system is calculated, and the power compensation calculation of the energy storage system is performed. The optimized power transmission path is calculated, and the topological node adjustment is performed to generate the photovoltaic storage flexible direct current power flow adjustment matrix.

5. The method for predicting the operation of a PV-storage flexible direct current system based on deep learning according to claim 1, characterized in that: The graph convolutional network, according to the formula: in: For the Feature representation of layer nodes, is the activation function, For nodes The set of neighbor nodes of is the normalization factor, For the The trainable weight matrix of the layer, For adjacent nodes In the The feature representation of the layer, For the current node In the The feature representation of the layer, For nodes With node The power transfer efficiency between For nodes The transmission power at It is the power level index adjustment node, reflecting the load adjustment strategy. is the power regulation coefficient, For nodes With node The physical distance between is the attenuation factor.

6. The deep learning-based PV-storage flexible DC operation prediction method according to claim 1 is characterized in that: The specific steps of generating the power optimization allocation matrix are: Based on the PV-storage flexible direct current power flow adjustment matrix, a generative adversarial network is used to calculate the boundaries of the charge and discharge time periods, calculate the power compensation demand, extract the load-side compensation parameters, calculate the energy storage charge and discharge conversion rate, adjust the power scheduling priority, calculate the energy regulation period, perform power stability correction, and generate energy storage regulation matching parameters; Based on the energy storage regulation and matching parameters, the load demand range, power supply and demand relationship, and photovoltaic power adjustment range are calculated, and output power balancing control is performed. The power supply and demand matching degree is calculated, and dynamic load optimization is performed. The combined power output of energy storage and photovoltaics is calculated, and power matching optimization is performed to generate a power regulation matching matrix. Based on the power regulation matching matrix, the energy storage system release path is calculated, power is dynamically allocated, the combined output ratio of photovoltaic and energy storage is calculated, and power balancing control is performed. The load-side compensation capacity is calculated, and load supply and demand coordination is performed. The dynamic release adjustment amount of energy storage is calculated, and the power load is finally matched to generate a power optimization distribution matrix.

7. The deep learning-based PV-storage flexible DC operation prediction method according to claim 1 is characterized in that: The adversarial generation network is based on the formula: in: is the generator network, is the discriminator network, To counter the objective function, is the distribution of real power data, is the real power data sample, is the distribution of the latent variable, is a random noise input, Indicates the discriminator's response to the real data during the charging and discharging period and load demand power The predicted probability under the condition Represents the generator based on the noise input and charge-discharge conversion efficiency , charging power and energy regulation cycle The generated power data samples, represents the predicted probability of the discriminator for the generated power data under the corresponding conditions, Divide the boundaries of the charging and discharging time periods, is the load demand power, is the charge-discharge conversion efficiency, is the charging power, It is the energy regulation cycle.

8. The method for predicting PV-storage flexible direct current operation based on deep learning according to claim 1, characterized in that: The specific steps of generating the photovoltaic power control parameters are as follows: Based on the power optimization allocation matrix, short-term power fluctuation data is extracted, the time-series power change rate is calculated, and the fluctuation amplitude is divided into zones and statistics. The short-term power change points on the load side are extracted, the power adjustment rate of the mutation point is calculated, abnormal fluctuations are screened, the short-term power response time of the energy storage system is calculated, and the charge and discharge switching rate is corrected to generate the short-term fluctuation adjustment parameters. Based on the short-term fluctuation adjustment parameter, extract the long-term power change trend data, calculate the change rate of the historical power curve, and perform trend smoothing. Calculate the long-term power compensation capability of the energy storage system, perform charge and discharge cycle matching calculations, extract the load demand change trend, calculate the long-term load power adjustment rate, and perform load-side power supply and demand matching analysis to generate the long-term trend adjustment parameter. Based on the long-term trend adjustment parameters, the power regulation response rate data is extracted, the dynamic load response interval is calculated, and the power regulation rate is adjusted. The dynamic compensation parameters of energy storage charging and discharging are calculated, and the power flow balance calculation is performed. The dynamic adjustment range of load compensation is calculated, and the load regulation cycle is optimized to generate the photovoltaic storage power control parameters.

9. The deep learning-based PV-storage flexible DC operation prediction method according to claim 1, characterized in that: The specific steps for generating the prediction results of the PV-storage flexible direct current operation are: Based on the photovoltaic storage power control parameters, historical photovoltaic power data is extracted, the time series change characteristics are calculated, and short-term trend calculations are performed. The photovoltaic power generation power mutation points are extracted, the fluctuation range change rate is calculated, and the power fluctuation range at future moments is predicted. The trend curve of photovoltaic power output is extracted, the photovoltaic output power adjustment amount in the short term is calculated, and the photovoltaic power prediction results are generated. Based on the photovoltaic power prediction results, load demand change data is extracted, the time series change range of the load power demand is calculated, and the power demand interval calculation is performed. The available power distribution data is extracted, the charge and discharge adjustment capacity is calculated, and the energy storage supply and demand balance calculation is performed. The power supply and demand difference data is extracted, the power distribution error compensation range is calculated, and the load demand prediction parameters are generated. Based on the load demand forecast parameters, the power timing forecast error data is extracted, the error compensation correction range is calculated, and the power forecast deviation is adjusted. The energy storage and photovoltaic combined power distribution data is extracted, the dynamic energy storage adjustment rate is calculated, and power distribution optimization is performed. The load power supply and demand balance analysis data is extracted, the optimal distribution path for power regulation is calculated, and the photovoltaic storage flexible direct current operation prediction result is generated.

10. The method for predicting PV-storage flexible direct current operation based on deep learning according to claim 1, characterized in that: The specific steps for generating the optimization prediction results of the PV-storage flexible direct current operation are as follows: Based on the PV-storage flexible direct current operation prediction results, the predicted power data and actual power data are extracted, the time distribution of the data error is calculated, and the error change rate analysis is performed. The error deviation interval is extracted, the error cumulative impact factor is calculated, and abnormal fluctuation screening is performed. The abnormal operation status data is extracted, the key parameters of the abnormal power fluctuation are calculated, and the error interval is classified to generate the prediction error and abnormal status data; Based on the prediction error and abnormal state data, extracting variables of power flow affected by the error, calculating the offset of the variables on the power flow, and calculating the variable adjustment amplitude, extracting corrected variable data, calculating the power output characteristics after the variable correction, and redistributing the power flow, extracting adjusted power path data, calculating the power transmission stability in the corrected path, and performing power transmission adjustment to generate variable correction and power path adjustment data; Based on the variable correction and power path adjustment data, the control parameters of the corrected power state are extracted, the corrected power transmission rate is calculated, the power flow balance is adjusted, the adjusted load power matching data is extracted, the load demand and power adjustment matching degree are calculated, and power flow optimization calculation is performed. The transmission parameters after power path optimization are extracted, the power flow adjustment range is calculated, and the optimization prediction results of the PV-storage flexible direct current operation are generated.

Images