A data-driven real-time monitoring system for line loss in a photovoltaic grid-connected substation area

Abstract

The present invention discloses a real-time monitoring system for line loss in a photovoltaic grid-connected substation area based on data driving, which relates to the field of power systems. It includes a data collection and preprocessing module, a line loss anomaly detection module, a line loss cause identification module, a line loss prediction module, and an optimization control module. The clustering algorithm is applied for line loss detection, a comprehensive power grid model and a photovoltaic power generation model are constructed using a cloud big data platform, and a time series analysis method is used to mine the line loss trend of a distributed photovoltaic grid-connected substation area, realizing the prediction and early warning of line loss. An optimization control strategy is formulated according to the prediction trend, the current power grid operation state, and historical optimization records, thereby improving the stability and reliability of power grid operation; it solves the problem that it is difficult to quickly and accurately identify and locate the line loss in a photovoltaic grid-connected substation area and there is a lack of corresponding line loss optimization strategies. The present invention can realize the real-time monitoring of the line loss in a distributed photovoltaic grid-connected substation area, effectively reduce the line loss rate, and ensure the stable operation of the power grid.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and particularly to a real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving. Background Art

[0002] With the wide application of photovoltaic power stations and the large-scale development of photovoltaic power, distributed photovoltaic power generation, as a clean and renewable energy technology, has been widely used globally. However, in the process of the rapid development of distributed photovoltaic power generation, the operation of the power grid also faces many challenges. Especially in the aspect of line loss management in the substation area, the line loss of the substation area refers to the power loss generated during the process of power grid transmitting and distributing electric energy, and is an important indicator to measure the operation efficiency of the power grid. The traditional management methods are difficult to meet the new requirements brought by the increasingly complex power grid structure and the access of photovoltaic power generation. Therefore, developing a system that can monitor the line loss of the photovoltaic grid-connected substation area in real time has become the core requirement for improving the efficiency of photovoltaic power generation, ensuring power quality, and reducing operating costs.

[0003] The line loss problem of the distributed photovoltaic grid-connected substation area is affected by various factors, including abnormal cable resistance, electromagnetic field vortex effect, alternating magnetic field action, power grid management loopholes, and metering device failures. These factors are intertwined, making the line loss problem complex and changeable. The traditional methods often rely on manual inspections and empirical judgments, and it is difficult to quickly and accurately identify and locate the line loss problem, resulting in the problem not being solved in a timely manner. Moreover, due to the voltage and frequency fluctuations in the local power grid connected by distributed photovoltaic power generation, there is a lack of corresponding line loss optimization strategies in the current distribution network management, which affects the safe and stable operation of the power grid.

[0004] Therefore, a real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving is needed to solve the above problems. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the present invention discloses a real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving. Through the organic combination of functions such as real-time data collection, precise anomaly detection, cause identification, trend prediction, and optimization control, it realizes the all-round and intelligent monitoring and management of the line loss of the distributed photovoltaic grid-connected substation area, with high automation and intelligence.

[0006] The present invention adopts the following technical solutions:

[0007] A real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving, the system includes:

[0008] A data collection and preprocessing module, which uses an edge-aware monitoring and interconnection device to collect the operation status data of a distributed photovoltaic grid-connected substation area in real time, and transmits the real-time collected data to a cloud big data platform through a wireless communication network for data preprocessing, storage, and analysis operations. The operation status data at least includes photovoltaic power generation, voltage, current, and load characteristics;

[0009] A line loss anomaly detection module, which uses the K-Medoids clustering algorithm to perform clustering analysis on the historical detected substation area line loss data to identify the clustering centers of the normal working mode and the abnormal line loss mode of the distributed photovoltaic grid-connected substation area, and calculates the Euclidean distance between the real-time collected data and the clustering center of the abnormal line loss mode. When the Euclidean distance between the real-time collected data and the clustering center of the abnormal line loss mode exceeds a preset low threshold, the real-time collected data is marked as abnormal line loss data;

[0010] A line loss cause identification module, which constructs a comprehensive power grid model and a photovoltaic power generation model of the distributed photovoltaic grid-connected substation area through the cloud big data platform. The cloud big data platform compares and calculates the time-sharing line loss between the simulated physical state data of the comprehensive power grid model and the photovoltaic power generation model and the abnormal line loss data, and identifies the cause of the abnormal line loss based on the calculation results. The comprehensive power grid model is used to simulate the operation characteristics of the power grid in the distributed photovoltaic grid-connected substation area, and the photovoltaic power generation model is used to simulate the output characteristics of the distributed photovoltaic grid-connected substation area under different weather conditions;

[0011] A line loss prediction module, which uses the time series analysis method to mine the time series characteristics of photovoltaic power generation, voltage change, and load fluctuation to predict the line loss trend of the distributed photovoltaic grid-connected substation area;

[0012] An optimization control module, which formulates an optimization control strategy according to the line loss prediction trend, the current power grid operation status, and the historical optimization records.

[0013] Further, the working method of the K-Medoids clustering algorithm includes the following steps:

[0014] S1. Randomly select K data points from the historical detected substation area line loss data set as the initial clustering centers of the abnormal line loss mode, set the maximum number of iterations and the threshold of the change amount of the clustering center, and then record the initial clustering center index or the actual data point value;

[0015] S2. Calculate the Euclidean distance between the real-time collected data points and the clustering centers of the abnormal line loss mode, and assign each data point to the closest clustering center;

[0016] S3. Recalculate the sum of the distances from all points within each cluster to the cluster center, and select the cluster center with the minimum sum of distances as the new abnormal line loss pattern cluster center, then update the index of the cluster center or the actual data point value;

[0017] S4. Repeat the operation of S2, and reassign the data points using the new abnormal line loss pattern cluster center until the change amount of the abnormal line loss pattern cluster center is less than the threshold or the maximum number of iterations is reached;

[0018] S5. Use the silhouette coefficient and CH index clustering quality evaluation indicators to evaluate whether the clustering result conforms to the expected abnormal line loss pattern. If not, adjust the clustering parameters and repeat the operation of S1.

[0019] Further, in the S2, first extract the time series features and spatial series features of the preprocessed real-time collected data points, and perform regularization and eigenvector representation on the extracted features. The time series features and spatial series feature vectors extracted are expressed as:

[0020]

[0021] In formula (1), X represents the set of feature vectors of the real-time collected data points, is the time series feature vector of the i-th real-time collected data point, 1 ≤ i ≤ n, i is the ordinal number of the real-time collected data point, n is the number of time series feature vectors, t represents the time series feature vector, is the spatial series feature vector of the i-th real-time collected data point, y represents the spatial series feature vector, and the abnormal line loss pattern cluster center set is V = {v1, v2,..., v g ,..., v k}, v g represents the g-th abnormal line loss pattern cluster center, k represents the number of abnormal line loss pattern cluster centers, and the Euclidean distance metric method is used to calculate the distances between the time series features and spatial series features and the abnormal line loss pattern cluster centers respectively. The output function formula is:

[0022]

[0023] In formula (2), is the distance from the time series feature vector of the i-th real-time collected data point to the abnormal line loss pattern cluster center, ξ is the clustering hyperparameter of the time series feature vector, 0 < ξ ≤ 1, is the distance from the spatial series feature vector of the i-th real-time collected data point to the abnormal line loss pattern cluster center, δ is the clustering hyperparameter of the spatial series feature vector, 0 < δ ≤ 1, and the Euclidean distance calculation formula between the i-th real-time collected data point and the abnormal line loss pattern cluster center is:

[0024]

[0025] In formula (3), D i represents the Euclidean distance between the i-th real-time collected data point and the clustering center of the abnormal line loss pattern.

[0026] Furthermore, the integrated power grid model and the photovoltaic power generation model use a physical engine to simulate the operating states of the distributed photovoltaic grid-connected substation area under different time periods and different working conditions, and perform real-time calculation of the time-of-use line loss based on the simulation results. When the calculated result of the time-of-use line loss exceeds the set time-of-use line loss threshold, an alarm is automatically triggered, and then the physical state data monitored in real time and the time-of-use line loss data are subjected to correlation analysis to locate the cause of the abnormal line loss.

[0027] Furthermore, the time series analysis method uses a long short-term memory network (LSTM) to construct a power prediction model, trains the model by minimizing the error between the predicted value and the actual value, and then uses the trained model to predict the short-term power. The long short-term memory network (LSTM) uses the mean square error as the loss function and uses the root mean square error and mean absolute error indexes to evaluate the prediction performance of the trained model.

[0028] Furthermore, the optimization control module uses a game theory model to analyze the electricity consumption behavior of users to identify abnormal electricity consumption patterns, and formulates an optimization control strategy according to the results of the game theory analysis. The game theory model quantifies the benefits or costs of different electricity consumption strategies for the power grid operator and users by establishing a utility function, and identifies the optimal strategy combination of users and operators by calculating the Nash equilibrium.

[0029] Furthermore, the optimization control strategy includes the following aspects:

[0030] (1) For the voltage fluctuation problem, an intelligent substation device, coordinated voltage control, and reactive power compensation methods are used to control and regulate the voltage fluctuation;

[0031] (2) For the problem of unbalanced power consumption load, load balance is achieved by adjusting the distribution network structure and power consumption load;

[0032] (3) For the problem of distributed energy access and distribution, the photovoltaic access position and access method are adjusted, and an inverter and a monitoring device are used to monitor the photovoltaic power generation amount and the power grid load condition;

[0033] (4) For the problem of power equipment failure, the power equipment failure is eliminated by performing online fault monitoring, prediction, and fault diagnosis on the power equipment.

[0034] Furthermore, the cloud big data platform stores a large amount of line loss data using distributed storage technology and performs streaming processing on the real-time line loss data using a real-time computing framework to achieve real-time monitoring and analysis of line loss.

[0035] The beneficial effects of the present invention are as follows:

[0036] 1. The present invention collects the operation status data of the distributed photovoltaic grid-connected substation area in real time through the edge perception monitoring and interconnection device, ensuring the timeliness and accuracy of the data, and using the K-Medoids clustering algorithm to perform clustering analysis on the line loss data, which can quickly identify abnormal line loss patterns, and by calculating the Euclidean distance between the real-time data and the clustering center of the abnormal pattern, the rapid marking of abnormal data is realized, improving the accuracy and timeliness of abnormal detection.

[0037] 2. The present invention constructs a comprehensive power grid model and a photovoltaic power generation model to simulate the operation characteristics of the power grid and the output characteristics of photovoltaic power generation, providing a scientific basis for the accurate identification of the causes of abnormal line loss, and by comparing the simulated physical state data with the abnormal line loss data and performing time-sharing line loss calculation, the specific causes of abnormal line loss can be more accurately identified, providing strong support for subsequent optimization control.

[0038] 3. The present invention uses the time series analysis method to deeply explore the time series characteristics of photovoltaic power generation, voltage change and load fluctuation, improving the accuracy and reliability of line loss prediction, and the prediction model constructed based on the time series characteristics can accurately predict the line loss trend of the distributed photovoltaic grid-connected substation area, providing an important reference for the planning and dispatching of the power grid.

[0039] 4. The present invention formulates an optimization control strategy according to the line loss prediction trend, the current operation state of the power grid and the historical optimization records, realizing the intelligent management of the power grid operation, and by implementing the optimization control strategy, the line loss rate of the distributed photovoltaic grid-connected substation area can be effectively reduced, improving the energy utilization efficiency and reducing energy waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram of the architecture of the overall system of the present invention;

[0041] Figure 2 It is a schematic diagram of the working process of the K-Medoids clustering algorithm in the present invention;

[0042] Figure 3 It is a schematic diagram of the working process of the time series analysis method in the present invention;

[0043] Figure 4 It is a schematic diagram of the working process of the data-driven model predictive control MPC strategy in the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0044] Next, in combination with the accompanying drawings in the embodiments of the present invention, Figure 1 to Figure 4 , the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0045] An embodiment of the present invention discloses a data-driven real-time monitoring system for line loss in a photovoltaic grid-connected substation area. The system includes:

[0046] Data collection and preprocessing module: This module uses existing edge perception monitoring and interconnection devices to install sensor devices such as power meters, voltage meters, current meters, temperature sensors, etc., to collect photovoltaic power generation, voltage fluctuations, load characteristics, and line loss data in real time, and perform data cleaning, missing value filling, and outlier detection and other processing operations to ensure the accuracy and integrity of the data, and transmit these data to the cloud big data platform through a wireless communication network.

[0047] Line loss anomaly detection module: This module performs clustering analysis on the historical line loss data detected in the substation area by using the K-Medoids clustering algorithm to identify the clustering centers of normal working modes and abnormal line loss modes. Specifically, the algorithm performs multiple clustering operations on the historical data to obtain all clustering centers, and calculates the Euclidean distance between the real-time collected data within a specific time and the clustering center of the abnormal line loss mode based on this. When it exceeds the preset low threshold, this data is marked as abnormal line loss data and needs to be processed.

[0048] Line loss cause identification module: This module first constructs a comprehensive power grid model and a photovoltaic power generation model for the distributed photovoltaic grid-connected substation area with the help of the cloud big data platform. Specifically, the comprehensive power grid model is used to simulate the operating characteristics of the distributed photovoltaic grid-connected substation area power grid, and the photovoltaic power generation model is used to simulate the output characteristics of the distributed photovoltaic grid-connected substation area under different weather conditions. Secondly, the simulated physical state data is compared with the abnormal line loss data to calculate the time-sharing line loss. Finally, the cause of the abnormal line loss is identified according to the calculation results.

[0049] Line loss prediction module: In time series analysis, this module uses time series analysis methods to reveal the laws of photovoltaic power generation, voltage changes, and load fluctuations, mine the time series characteristics of photovoltaic power generation, voltage changes, and load fluctuations, and predict the line loss trend of the distributed photovoltaic grid-connected substation area based on these characteristics.

[0050] Optimization control module: This module formulates an optimization control strategy based on the line loss prediction trend, the current power grid operation status, and historical optimization records. Specifically, according to the line loss prediction trend, the system predicts the possible line loss situation in the future for a period of time and formulates corresponding optimization control plans. At the same time, based on the current power grid operation status, it analyzes the changes in parameters such as power grid load for targeted optimization control. Finally, using historical optimization records and model prediction results, it conducts continuous optimization and update to ensure the effectiveness and practicality of the optimization control strategy.

[0051] In summary, based on data-driven real-time monitoring and analysis, this system uses a variety of technical means to solve the line loss problem in the distributed photovoltaic grid-connected substation area, with advantages such as high efficiency, accuracy, and reliability.

[0052] The output end of the data collection and preprocessing module is connected to the input ends of the line loss anomaly detection module, the line loss cause identification module, and the line loss prediction module. The output end of the line loss anomaly detection module is connected to the input end of the line loss cause identification module. The output ends of the line loss anomaly detection module, the line loss cause identification module, and the line loss prediction module are connected to the input end of the optimization control module.

[0053] The working method of the K-Medoids clustering algorithm includes the following steps:

[0054] S1: Initialize the clustering centers;

[0055] Data preparation: Obtain data from the historical detected substation area line loss dataset.

[0056] Random selection: Use the random selection method to select K different data points from this group of data as the initial clustering centers of abnormal line loss patterns.

[0057] Set parameters: Set the maximum number of iterations and the threshold of the change amount of the clustering centers.

[0058] Record the initial clustering centers: Record the indices or actual data point values of these K initial clustering centers for subsequent comparison and update.

[0059] S2: Assign data points;

[0060] Distance calculation: For each data point in the real-time collected dataset, calculate its Euclidean distance from all K clustering centers.

[0061] Assign data points: Assign each data point to the clustering center with the smallest Euclidean distance from it to form K clusters.

[0062] S3: Update the clustering centers;

[0063] Calculate the within-cluster distance sum: For each cluster, calculate the sum of the distances from all the data points within it to the current cluster center.

[0064] Select the new cluster center: In each cluster, select the data point that minimizes the within-cluster distance sum as the new cluster center. This is usually done by traversing all the data points within the cluster, calculating the within-cluster distance sum after replacing each point with the cluster center, and then selecting the point that gives the minimum sum.

[0065] Update the cluster center: Update the index or the actual data point value of the cluster center to the new selection.

[0066] S4: Iterative optimization;

[0067] Repeat assignment: Go back to S2, recalculate the distances from each data point to the cluster centers using the new cluster centers, and reassign the data points to the nearest cluster centers.

[0068] Termination condition: Check whether the change in the cluster centers is less than a preset threshold or whether the maximum number of iterations has been reached. If either condition is met, stop the iteration; otherwise, continue to go back to S3 for the next iteration.

[0069] S5: Cluster quality assessment;

[0070] Calculate the evaluation metrics: Use cluster quality evaluation metrics such as the silhouette coefficient and the CH index (Calinski-Harabasz Index) to evaluate the quality of the clustering results.

[0071] The silhouette coefficient measures the tightness of the data points within a cluster and the separation of the data points between clusters. Its value ranges from -1 to 1, and a larger value indicates a better clustering effect.

[0072] The CH index evaluates the clustering effect by calculating the ratio of the trace of the within-class scatter matrix to the trace of the between-class scatter matrix. A larger value also indicates a better clustering effect.

[0073] Result judgment: If the clustering result does not conform to the expected abnormal line loss pattern (e.g., verified by expert judgment or business rules), then adjust the clustering parameters (such as the K value, the number of iterations, the threshold, etc.), and repeat the operations from S1 to S4 until the clustering result is satisfactory.

[0074] In S2, first extract the time series features and spatial series features of the preprocessed real-time collected data points, and perform regularization and eigenvector representation on the extracted features. The extracted time series features and spatial series feature vectors are represented as:

[0075]

[0076] In formula (1), X represents the set of feature vectors of real-time collected data points, is the time-series feature vector of the i-th real-time collected data point, 1 ≤ i ≤ n, i is the ordinal number of the real-time collected data point, n is the number of time-series feature vectors, t represents the time-series feature vector, is the spatial-series feature vector of the i-th real-time collected data point, y represents the spatial-series feature vector, and the abnormal line loss pattern clustering center set is V = {v1, v2,..., v g ,..., v k}, v g represents the g-th abnormal line loss pattern clustering center, k represents the number of abnormal line loss pattern clustering centers, and the Euclidean distance metric method is used to calculate the distances between the time-series features and spatial-series features and the abnormal line loss pattern clustering centers respectively. The output function formula is:

[0077]

[0078] In formula (2), is the distance from the time-series feature vector of the i-th real-time collected data point to the abnormal line loss pattern clustering center, ξ is the clustering hyperparameter of the time-series feature vector, 0 < ξ ≤ 1, is the distance from the spatial-series feature vector of the i-th real-time collected data point to the abnormal line loss pattern clustering center, δ is the clustering hyperparameter of the spatial-series feature vector, 0 < δ ≤ 1, and the Euclidean distance calculation formula between the i-th real-time collected data point and the abnormal line loss pattern clustering center is:

[0079]

[0080] In formula (3), D i represents the Euclidean distance between the i-th real-time collected data point and the abnormal line loss pattern clustering center.

[0081] The hardware working environment of the K-Medoids clustering algorithm includes:

[0082] Processor (CPU): A stronger processor performance can accelerate data processing and calculation speed. Especially for large-scale data sets, a high-performance CPU is necessary. A multi-core CPU can process multiple tasks in parallel and improve the execution efficiency of the algorithm.

[0083] Memory (RAM): Sufficient memory is crucial for processing large data sets. The K-Medoids algorithm needs to frequently access and update the data set during the iteration process. Therefore, sufficient RAM can reduce disk I / O operations and improve the running speed of the algorithm.

[0084] Storage device: Although the K-Medoids algorithm mainly relies on in-memory data processing, larger datasets may not be fully loaded into memory. In such cases, a fast storage device (such as an SSD) can reduce data loading time and improve overall performance.

[0085] Graphics Processing Unit (GPU): For some K-Medoids algorithm implementations that support GPU acceleration, a powerful GPU can significantly improve the execution speed of the algorithm. GPUs are good at handling large-scale parallel computing and are particularly suitable for accelerating operations such as distance calculation and cluster update.

[0086] Network conditions (for distributed or cloud environments): If the K-Medoids algorithm runs in a distributed system or cloud environment, network bandwidth and latency will affect data transmission and the communication efficiency between nodes. Therefore, good network conditions are crucial for ensuring the stable operation of the algorithm.

[0087] According to the annual substation area line loss work requirements of State Grid Sanmenxia Power Supply Company, a series of detailed on-site surveys were carried out in Shanzhou Power Supply Company to understand the actual operation of distributed photovoltaic grid-connected substations. The research content includes photovoltaic power generation, voltage fluctuation, load characteristics, and line loss data, etc. Through these data, the main sources and influencing factors of substation area line loss were initially identified.

[0088] A Core i9 series computer with 64 + 1T memory was used for data analysis. The on-site experimental environment was set up with a simulated data accuracy of 95% and an algorithm running error not exceeding 2.5%. Comparative experiments were carried out using the K-Medoids algorithm (Group A) and the linear regression method (Group B) respectively. Four groups of historical line loss and no-line-loss mixed data were manually input, with each group of data capacity being 500KB. Simulated work was carried out in the experimental environment, and the detection accuracy and time consumption were recorded. The experiment was repeated 5 times and the average value was calculated and recorded in Table 1.

[0089] Table 1 Result Statistical Table

[0090]

[0091] The results show that on the test set, the K-Medoids algorithm has a higher detection accuracy and a lower error rate than the linear regression method. At the same time, the running time of the K-Medoids algorithm is also lower than that of the linear regression method, with higher computational efficiency. Therefore, under this specific experimental condition, the performance of the K-Medoids algorithm is superior to the linear regression method.

[0092] The integrated power grid model and the photovoltaic power generation model simulate the operation status of the distributed photovoltaic grid-connected substation area based on a physical engine, including parameters such as voltage, current, power, and electricity quantity, and adjust the photovoltaic power generation model under different weather conditions to simulate the change of photovoltaic power generation. During the simulation process, the integrated power grid model and the photovoltaic power generation model collect and store the simulation result data in real time through a cloud big data platform, including parameters such as time-sharing electricity quantity, time-sharing power, time-sharing voltage, and time-sharing current. The simulation result data is correlated with the real-time monitored physical state data in real time to locate the cause of abnormal line loss. If the calculated result of the time-sharing line loss exceeds the set time-sharing line loss threshold, an alarm is triggered. Based on the analysis result, the optimization control module will formulate an optimization control strategy, such as adjusting the operation mode of the inverter, adjusting the load, etc., to reduce the time-sharing line loss. At the same time, the line loss prediction module will mine the time series characteristics of photovoltaic power generation, voltage change, and load fluctuation to predict the line loss trend of the distributed photovoltaic grid-connected substation area, and the optimization control module will formulate a more targeted optimization control strategy by combining the line loss prediction trend and the current power grid operation status.

[0093] The time series analysis method uses a long short-term memory network (LSTM) to construct a power prediction model, trains the model by minimizing the error between the predicted value and the actual value, and then uses the trained model to predict the short-term power. The long short-term memory network (LSTM) uses the mean square error as the loss function and uses the root mean square error and mean absolute error indicators to evaluate the prediction performance of the trained model.

[0094] The time series analysis method is a short-term power prediction algorithm based on a recurrent neural network, which can capture the nonlinear and dynamic characteristics in the process of photovoltaic power generation. As shown Figure 3 below, the implementation details of the time series analysis method are as follows:

[0095] (S1). Data preprocessing: including data cleaning, standardization, and feature engineering to improve the generalization ability of the model.

[0096] (S2). Model construction: Use a long short-term memory network (LSTM), a variant of the RNN, to construct a power prediction model. The LSTM can solve the gradient vanishing problem in the traditional RNN and improve the learning efficiency.

[0097] (S3). Training process: Train the model by minimizing the error between the predicted value and the actual value, usually using the mean square error (MSE) as the loss function.

[0098] (S4). Prediction and evaluation: Use the trained model to predict the short-term power and use indicators such as the root mean square error (RMSE) and mean absolute error (MAE) to evaluate the prediction performance.

[0099] The optimization control module uses a game theory model to analyze users' electricity consumption behaviors, identify abnormal electricity consumption patterns, and optimize electricity consumption strategies. Based on the results of the game theory analysis, it formulates optimization strategies to reduce line losses and improve the operating efficiency of the power grid. The game theory model quantifies the benefits or costs of different electricity consumption strategies for power grid operators and users by establishing utility functions, and identifies the optimal strategy combinations for users and operators by calculating the Nash equilibrium.

[0100] In addition, data-driven methods such as reinforcement learning and policy optimization methods play important roles in optimization control. The data-driven model predictive control (MPC) strategy can utilize historical data to eliminate the influence brought by parameter mismatch, thereby improving control performance. As shown Figure 4 below, the implementation details of the MPC strategy include:

[0101] Step 1. Establish a photovoltaic system model: According to the physical characteristics and operating parameters of the photovoltaic system, establish an accurate system model.

[0102] Step 2. Define the objective function: Usually includes minimizing the tracking error and the smoothness of the control input.

[0103] Step 3. Optimization process: Use a solver for online optimization to find the optimal control sequence that satisfies the system constraints.

[0104] Step 4. Real-time update: Continuously update the model and optimization process according to real-time data to adapt to the changes in the state of the photovoltaic system.

[0105] The utilization of local energy storage technology compensates for the difference between photovoltaic power and load through the existing power grid, better adapting to the power flow of local grid connection. The control algorithm of the energy storage system can dynamically adjust the charge and discharge strategies of the energy storage according to the needs of the power grid and the volatility of photovoltaic power generation to achieve optimal energy allocation and management.

[0106] The optimization control strategies include the following aspects:

[0107] (1) For voltage fluctuation problems, intelligent substation devices, coordinated voltage control, and reactive power compensation methods are adopted to control and regulate voltage fluctuations;

[0108] (2) For the problem of unbalanced electricity loads, load balance is achieved by adjusting the distribution network structure and electricity loads;

[0109] (3) For the problems of distributed energy access and distribution, adjust the photovoltaic access location and access method, and use inverters and monitoring devices to monitor the photovoltaic power generation and grid load conditions;

[0110] (4) For power equipment failure problems, eliminate power equipment failures through online fault monitoring, prediction, and fault diagnosis of power equipment.

[0111] Regarding the voltage fluctuation problem, the automatic voltage control system was improved to reduce voltage fluctuations. Regarding the complexity of line loss, by optimizing the PV access location and access method, the line loss was reduced. To solve the problem of massive IoT data, using big data analysis technology, data sharing and comprehensive utilization between different systems were realized, and the potential value of the data was mined. In the specific implementation of the control strategy, a variety of technical means and algorithms were adopted. For example, in the distributed photovoltaic power generation grid-connected substation area line loss accurate calculation method for electric pile harmonics, by analyzing the impacts caused by problems such as voltage, harmonics, and islanding effect, the accurate calculation of the actual power loss value of the line was realized. The distributed PV line loss abnormal perception algorithm based on K-Medoids clustering, combined with the local outlier factor (LOF) algorithm and granular computing to optimize the clustering center, effectively avoided the influence of isolated points on the abnormal perception effect, and accurately and effectively perceived the distributed PV substation area line loss abnormality. In addition, the distributed PV grid-connected capacity matching strategy based on network loss and voltage constraints was also studied. From the perspectives of safety and economy, the PV grid-connected capacity calculation formula applicable to practical engineering applications was obtained, and its correctness and effectiveness were verified through simulation.

[0112] The cloud big data platform uses distributed storage technologies such as Hadoop and HBase to store massive line loss data, ensuring the high availability and high reliability of the data. And it adopts real-time computing frameworks such as Spark Streaming to perform streaming processing on the real-time line loss data, realizing the real-time monitoring and analysis of line loss.

[0113] The application effects of the distributed PV grid-connected substation area line loss governance technology are remarkable in multiple scenarios. In industries, commercial buildings, agricultural facilities, and residential areas, the distributed PV grid-connected system can effectively reduce the line loss rate, improve the power supply efficiency and stability. For example, Liaoyang Power Supply Company significantly reduced the line loss rate by adjusting the line. In the microgrid application, the distributed PV grid-connected system jointly supplies power with energy storage, diesel generators, etc., improving the stability and reliability of the system.

[0114] Public facilities such as airports have achieved green and sustainable power supply by installing distributed PV grid-connected systems. In high-load places, the distributed PV grid-connected system effectively absorbs the impacts brought by the volatility of new energy power generation, ensuring the stable operation of the power grid. These application cases show that through reasonable planning and technological innovation, the distributed PV grid-connected substation area line loss governance technology can effectively improve the safety, stability, and economy of power grid operation.

[0115] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easily understood by those skilled in the art that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

Claims

1. A data-driven real-time monitoring system for line loss in a photovoltaic grid-connected substation area, characterized in that, The system includes: A data collection and preprocessing module, which uses edge-aware monitoring and interconnection devices to collect the operation status data of a distributed photovoltaic grid-connected substation area in real time, and transmits the real-time collected data to a cloud big data platform through a wireless communication network for data preprocessing, storage, and analysis operations. The operation status data at least includes photovoltaic power generation, voltage, current, and load characteristics; A line loss anomaly detection module, which uses the K-Medoids clustering algorithm to perform clustering analysis on the historical detected substation area line loss data to identify the clustering centers of the normal working mode and abnormal line loss mode of the distributed photovoltaic grid-connected substation area, and calculates the Euclidean distance between the real-time collected data and the clustering center of the abnormal line loss mode. When the Euclidean distance between the real-time collected data and the clustering center of the abnormal line loss mode exceeds a preset low threshold, the real-time collected data is marked as abnormal line loss data; A line loss cause identification module, which constructs a comprehensive power grid model and a photovoltaic power generation model of the distributed photovoltaic grid-connected substation area through the cloud big data platform. The cloud big data platform compares and calculates the time-sharing line loss between the simulated physical state data of the comprehensive power grid model and the photovoltaic power generation model and the abnormal line loss data, and identifies the cause of the abnormal line loss based on the calculation result. The comprehensive power grid model is used to simulate the operation characteristics of the power grid in the distributed photovoltaic grid-connected substation area, and the photovoltaic power generation model is used to simulate the output characteristics of the distributed photovoltaic grid-connected substation area under different weather conditions; The comprehensive power grid model and the photovoltaic power generation model use a physical engine to simulate the operation status of the distributed photovoltaic grid-connected substation area at different times and under different working conditions, and perform real-time calculation of the time-sharing line loss based on the simulation results. When the time-sharing line loss calculation result exceeds the set time-sharing line loss threshold, an alarm is automatically triggered, and then the correlation analysis is performed on the real-time monitored physical state data and the time-sharing line loss data to locate the cause of the line loss anomaly; A line loss prediction module, which uses time series analysis methods to mine the time series characteristics of photovoltaic power generation, voltage changes, and load fluctuations to predict the line loss trend of the distributed photovoltaic grid-connected substation area; An optimization control module, which formulates an optimization control strategy according to the line loss prediction trend, the current power grid operation status, and the historical optimization records.

2. The real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving according to claim 1, wherein The working method of the K-Medoids clustering algorithm includes the following steps: S1. Randomly select K data points from the historical detected substation area line loss data set as the initial abnormal line loss mode clustering centers, set the maximum number of iterations and the threshold of the clustering center change amount, and then record the initial clustering center index or the actual data point value; S2. Calculate the Euclidean distance between the real-time collected data points and the abnormal line loss mode clustering centers, and assign each data point to the closest clustering center; S3. Recalculate the sum of the distances from all points within each cluster to the clustering center, and select the clustering center with the smallest sum of distances as the new abnormal line loss mode clustering center, and then update the clustering center index or the actual data point value; S4. Repeat the operation of S2, and reassign data points with the new clustering center of the abnormal line loss pattern until the change amount of the clustering center of the abnormal line loss pattern is less than the threshold or the maximum number of iterations is reached; S5. Use the silhouette coefficient and the CH index clustering quality evaluation indicators to evaluate whether the clustering result conforms to the expected abnormal line loss pattern. If not, adjust the clustering parameters and repeat the operation of S1.

3. A real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving according to claim 2, characterized in that, In S2, first extract the time series features and spatial series features of the preprocessed real-time collected data points, and regularize and represent the extracted features as feature vectors. The time series features and spatial series feature vectors extracted are as follows: In formula (1), X represents the set of feature vectors of real-time collected data points, is the time series feature vector of the i-th real-time collected data point, 1 ≤ i ≤ n, i is the ordinal number of the real-time collected data point, n is the number of time series feature vectors, and t represents the time series feature vector, is the spatial series feature vector of the i-th real-time collected data point, y represents the spatial series feature vector, and the abnormal line loss pattern clustering center set is V = {v1, v2,..., v g ,..., v k}, v g represents the g-th abnormal line loss pattern clustering center, k represents the number of abnormal line loss pattern clustering centers, and the Euclidean distance metric method is used to calculate the distances between the time series features and spatial series features and the abnormal line loss pattern clustering centers respectively. The output function formula is: In formula (2), is the distance from the time series feature vector of the i-th real-time collected data point to the clustering center of the abnormal line loss pattern. ξ is the clustering hyperparameter of the time series feature vector, where 0 < ξ ≤ 1. is the distance from the spatial series feature vector of the i-th real-time collected data point to the clustering center of the abnormal line loss pattern. δ is the clustering hyperparameter of the spatial series feature vector, where 0 < δ ≤ 1. The Euclidean distance calculation formula between the i-th real-time collected data point and the clustering center of the abnormal line loss pattern is: In formula (3), D i represents the Euclidean distance between the i-th real-time collected data point and the clustering center of the abnormal line loss pattern.

4. A real-time monitoring system for line loss in a photovoltaic grid-connected substation area based on data-driven, characterized in that, The time series analysis method uses a long short-term memory network (LSTM) to construct a power prediction model, trains the model by minimizing the error between the predicted value and the actual value, and then uses the trained model to predict the short-term power. The long short-term memory network (LSTM) uses the mean square error as the loss function and uses the root mean square error and mean absolute error indicators to evaluate the prediction performance of the trained model.

5. A real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving according to claim 1, characterized in that The optimization control module uses a game theory model to analyze the electricity consumption behavior of users to identify abnormal electricity consumption patterns, and formulates an optimization control strategy based on the game theory analysis results. The game theory model quantifies the benefits or costs of different electricity consumption strategies for the power grid operator and users by establishing a utility function, and identifies the optimal strategy combination of users and operators by calculating the Nash equilibrium.

6. The real-time monitoring system for line loss of a photovoltaic grid-connected substation area based on data driving according to claim 5, wherein, The optimization control strategy includes the following aspects: (1) For the voltage fluctuation problem, use intelligent substation devices, coordinated voltage control, and reactive power compensation methods to achieve the control and regulation of voltage fluctuations; (2) For the problem of unbalanced electricity load, achieve load balance by adjusting the distribution network structure and electricity load; (3) For the problem of distributed energy access and distribution, adjust the photovoltaic access location and access method, and use inverters and monitoring devices to monitor the photovoltaic power generation and the power grid load situation; (4) For the problem of power equipment failures, eliminate power equipment failures through online fault monitoring, prediction, and fault diagnosis of power equipment.

7. A real-time monitoring system for line loss in a photovoltaic grid-connected substation area based on data-driven, characterized in that, The cloud big data platform uses a distributed storage technology to store a large amount of line loss data, and uses a real-time computing framework to perform streaming processing on the real-time line loss data to achieve real-time monitoring and analysis of line loss.

Images