A reliability analysis method and system for power distribution system

Abstract

The present invention relates to the technical field of reliability analysis of power distribution systems, and in particular to a method and system for reliability analysis of power distribution systems. The method comprises the following steps: collecting multidimensional data of the power distribution system through a sensor network, and performing multidimensional data fusion to obtain multidimensional correlation data; constructing a comprehensive operation data set for the multidimensional correlation data to obtain a comprehensive operation data set; segmenting time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slowly changing long-term trend sequence; performing time series decomposition on the fast fluctuation characteristic sequence and the slowly changing long-term trend sequence to obtain high-frequency fluctuation feature data and long-term trend feature data; performing feature component analysis on the high-frequency fluctuation feature data and the long-term trend feature data to obtain a multi-scale feature set. The present invention improves the reliability analysis accuracy of the power distribution system through risk factor association analysis and reliability network model construction.

Description

Technical Field

[0001] The present invention relates to the technical field of reliability analysis of power distribution systems, and in particular to a reliability analysis method and system for a power distribution system. Background Art

[0002] Early reliability analysis of distribution systems mainly relied on methods such as empirical statistics and fault tree analysis. These methods are based on historical fault data and expert experience, and evaluate system reliability through simple probability models or logical deductions. However, with the expansion of the scale and increase in complexity of distribution systems, the limitations of these methods have gradually become apparent. Later, with the development of computer technology, reliability assessment methods based on probability statistics have been widely used, such as Markov models and Monte Carlo simulations. These methods can consider factors such as equipment failure rate and repair time more finely, but often assume that equipment failures are independent of each other, ignoring the correlation between equipment and the impact of system operation status. In recent years, with the rise of artificial intelligence and big data technology, machine learning, deep learning and other methods have been applied to distribution system reliability analysis. These methods can automatically learn features from massive operating data and establish complex nonlinear models, improving the accuracy and intelligence level of reliability assessment.

[0003] Although the existing distribution system reliability analysis methods have made some progress, there are still the following deficiencies in feature extraction and risk identification:

[0004] 1. Single and shallow feature extraction: Existing methods lack the mining of deep features of time series data. The changes in the operating status of the distribution system often have multi-scale characteristics, including both rapid fluctuations and slow trends. Existing methods often only focus on a single time scale, ignoring the correlation and complementarity between features of different scales;

[0005] 2. Inaccurate and delayed risk identification: The operating status of the distribution system is dynamically changing. Existing methods often use static models, which are difficult to adapt to changes in the system's operating status, resulting in delayed risk identification and inability to provide timely warnings. Summary of the invention

[0006] Based on this, it is necessary to provide a distribution system reliability analysis method and system to solve at least one of the above technical problems.

[0007] To achieve the above object, a distribution system reliability analysis method includes the following steps:

[0008] Step S1: collect multi-dimensional data of the power distribution system through the sensor network, and perform multi-dimensional data fusion to obtain multi-dimensional related data; construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set;

[0009] Step S2: segmenting the time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slow-changing long-term trend sequence; performing time series decomposition on the fast fluctuation characteristic sequence and the slow-changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; performing feature component analysis on the high-frequency fluctuation characteristic data and the long-term trend characteristic data, and performing feature combination to obtain a multi-scale feature set; performing risk factor correlation analysis on the multi-scale feature set to obtain a correlation feature set; constructing a risk feature matrix on the correlation feature set to obtain a risk feature matrix;

[0010] Step S3: construct a reliability network model based on the comprehensive operation data set and the risk characteristic matrix to obtain a reliability network diagram; perform node importance assessment on the reliability network diagram to obtain node importance data; calculate node risk values ​​based on the node importance data and the reliability network diagram to obtain node risk assessment results; perform weak link classification based on the node risk assessment results, and generate a reliability weak point map to obtain a reliability weak point map;

[0011] Step S4: Train the reliability assessment model according to the reliability weak point map and the comprehensive operation data set, and optimize the model to obtain an optimized reliability assessment model; use the optimized reliability assessment model to perform a comprehensive reliability assessment to obtain a comprehensive reliability score.

[0012] The present invention integrates multi-source heterogeneous data, including sensor data, historical fault records, etc., to construct a comprehensive integrated operation data set, eliminate data islands, and provide a complete and high-quality data foundation for subsequent risk analysis and reliability assessment, thereby improving the accuracy and reliability of the assessment results. Through the multi-scale feature extraction method, the dynamic characteristics of the system are captured from different time scales, and through the risk factor association analysis, a risk feature matrix containing rich information is constructed, which can more comprehensively and deeply characterize the risk status of the system, and provide more effective feature input for subsequent reliability assessment. The risk characteristics are integrated into the network topology structure, and a reliability network model is constructed. Through node importance assessment and risk value calculation, the weak links in the system are identified, and finally a visual reliability weak point map is generated, which provides intuitive and easy-to-understand risk information for operation and maintenance personnel, and is convenient for quickly locating risk areas and formulating corresponding maintenance strategies. The reliability assessment model is trained and optimized using reinforcement learning technology, so that the model can adapt to different system states and operating environments, and finally outputs a comprehensive reliability score, providing an objective and quantitative reliability assessment result, which provides a scientific basis for system risk management and decision-making. Therefore, the present invention provides a distribution system reliability analysis method, which effectively solves the shortcomings of existing methods in feature extraction and risk identification through a series of technical steps such as multi-dimensional data fusion, multi-scale feature extraction, risk factor correlation analysis, reliability network model construction, and model training and optimization based on reinforcement learning, and improves the reliability analysis accuracy of the distribution system.

[0013] Preferably, step S1 comprises the following steps:

[0014] Step S11: collecting sensor data of the power distribution system through a sensor network to obtain raw sensor data;

[0015] Step S12: performing edge data preprocessing on the original sensor data to obtain preprocessed sensor data;

[0016] Step S13: uploading and centrally storing the pre-processed sensor data to obtain centralized data;

[0017] Step S14: performing semantic data modeling and fusion on the centralized data to obtain multi-dimensional associated data;

[0018] Step S15: construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set.

[0019] The present invention forms an original data set reflecting the operating status of the system by real-time collection of multidimensional data of each key node of the power distribution system, providing a basic data source for subsequent data analysis and reliability assessment. The collected original data is cleaned and preprocessed in real time through edge computing, effectively removing noise and outliers, improving data quality, reducing data transmission volume and the computational burden of subsequent processing, and ensuring the accuracy and efficiency of subsequent analysis. The preprocessed data is uploaded to the central server for centralized storage, and a unified data storage center is built, which facilitates unified management and access of data and provides the necessary data foundation for subsequent semantic data modeling and fusion. Through semantic data modeling and fusion technology, data from different sources and different types are associated, data redundancy and conflict are eliminated, multidimensional associated data is constructed, and a more comprehensive and in-depth understanding of the power distribution system is achieved, providing richer information for subsequent risk analysis and reliability assessment. By time alignment, sorting and missing value processing of multidimensional associated data, a structured, complete and easily accessible comprehensive operation data set is constructed, which provides high-quality data support for subsequent feature extraction, model training and reliability assessment, and improves the efficiency and accuracy of the entire analysis process.

[0020] Preferably, step S2 comprises the following steps:

[0021] Step S21: performing preliminary feature extraction based on the comprehensive operation data set to obtain a preliminary feature set;

[0022] Step S22: segmenting the comprehensive operation data set into time series data according to the preliminary feature set to obtain a fast fluctuation characteristic sequence and a slowly changing long-term trend sequence;

[0023] Step S23: performing time series decomposition on the fast fluctuation characteristic sequence and the slowly changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data;

[0024] Step S24: performing feature component analysis on the high-frequency fluctuation feature data and the long-term trend feature data, and performing feature combination to obtain a multi-scale feature set;

[0025] Step S25: performing multi-scale feature fusion on the multi-scale feature set to obtain a time series feature set;

[0026] Step S26: performing risk factor association analysis based on the preliminary feature set and the time series feature set to obtain an associated feature set;

[0027] Step S27: Perform feature screening and dimensionality reduction on the associated feature set to obtain a key feature set; construct a risk feature matrix on the key feature set to obtain a risk feature matrix.

[0028] The present invention extracts preliminary features reflecting the equipment status and operating environment from the comprehensive operating data set, such as load rate, current imbalance and number of operations, etc., to provide a basic feature set for subsequent in-depth analysis, which helps to preliminarily understand the operating status of the distribution system. The time series is segmented according to the fluctuation characteristics of the preliminary features, and the fast fluctuation characteristics and slow change characteristics are distinguished, so that different analysis methods can be used for different types of features in the future, so as to more effectively extract feature information and improve the accuracy and pertinence of feature extraction. Through methods such as wavelet transform and empirical mode decomposition, the feature information of different time scales is separated from the original time series, and high-frequency fluctuation feature data and long-term trend feature data are obtained, which provides a more refined data basis for subsequent multi-scale feature analysis and can more comprehensively characterize the dynamic characteristics of the system. By performing statistical analysis and spectral analysis on high-frequency and low-frequency feature data, a variety of feature components are extracted, and these components are combined into a multi-scale feature set, enriching the feature dimension, which can more comprehensively reflect the operating status and potential risks of the system. By fusing features of different time scales, such as fusing high-frequency fluctuation features with long-term trend features, a feature vector containing richer information is formed, which can more effectively characterize the dynamic change law of the system and improve the accuracy of subsequent risk analysis. Through physical models and simulation analysis, the correlation between different risk factors is revealed, such as the impact of load rate on temperature, and it is converted into new correlation features, which further enriches the feature set and makes risk assessment more comprehensive and in-depth. Through feature screening and dimensionality reduction, redundant features are removed, and the key features with the greatest impact on reliability are retained and organized into a risk feature matrix, which provides streamlined and effective data input for the subsequent construction of reliability network models and improves the efficiency and accuracy of the model.

[0029] Preferably, step S23 includes the following steps:

[0030] Step S231: setting wavelet transform parameters for the rapid fluctuation characteristic sequence to obtain wavelet transform parameters;

[0031] Step S232: setting a stop criterion for performing empirical mode decomposition on the slowly changing long-term trend sequence to obtain stop criterion data;

[0032] Step S233: extracting high-frequency fluctuation characteristics from the rapid fluctuation characteristic sequence according to the wavelet transform parameters to obtain high-frequency fluctuation characteristic data;

[0033] Step S234: extracting long-term trend features from the slowly changing long-term trend sequence according to the stopping criterion data to obtain long-term trend feature data.

[0034] The present invention ensures that the wavelet transform can effectively extract high-frequency information in the fast-fluctuation characteristic sequence by setting appropriate parameters, such as selecting the db4 wavelet basis function and determining the number of decomposition layers, laying the foundation for subsequent feature analysis. By setting reasonable stopping criteria, such as the energy proportion threshold and the IMF component number threshold, the process of EMD decomposition is effectively controlled, over-decomposition or under-decomposition is avoided, and the accuracy and effectiveness of the extracted long-term trend characteristics are guaranteed. The fast-fluctuation characteristic sequence is decomposed using preset wavelet transform parameters, and the high-frequency detail coefficients are extracted, which effectively captures the transient changes and fluctuation information in the data, providing important data support for subsequent risk analysis. The slowly changing long-term trend sequence is subjected to EMD decomposition using the preset stopping criteria, and the low-frequency IMF components and residual components are extracted, which effectively captures the long-term change trend in the data, providing important data support for subsequent risk analysis.

[0035] Preferably, step S24 comprises the following steps:

[0036] Step S241: Calculate the data standard deviation, data maximum change rate and data peak value of the high-frequency fluctuation characteristic data to obtain instantaneous fluctuation component data;

[0037] Step S242: performing data mean calculation, data slope calculation and data long-term offset calculation on the long-term trend characteristic data to obtain long-term evolution component data;

[0038] Step S243: performing main frequency component analysis on the instantaneous fluctuation component data and the long-term evolution component data to obtain main frequency component data;

[0039] Step S244: Calculate the high-frequency energy proportion of the instantaneous fluctuation component data to obtain high-frequency energy proportion data; calculate the low-frequency energy proportion of the long-term evolution component data to obtain low-frequency energy proportion data;

[0040] Step S245: performing spectrum entropy analysis on the long-term evolution component data to obtain spectrum entropy;

[0041] Step S246: Feature combination is performed on the instantaneous fluctuation component data, the long-term evolution component data, the main frequency component data, the high-frequency energy proportion data, the low-frequency energy proportion data and the spectrum entropy to obtain a multi-scale feature set.

[0042] The present invention quantifies the fluctuation degree and amplitude information of high-frequency fluctuation characteristic data by calculating statistics such as standard deviation, maximum change rate and peak value, effectively captures the instantaneous change characteristics in the data, and provides an important data basis for subsequent risk assessment. By calculating statistics such as mean, slope and long-term offset, the change trend and offset degree of long-term trend characteristic data are quantified, effectively capturing the long-term evolution characteristics in the data, and providing an important data basis for subsequent risk assessment. Through main frequency component analysis, the main frequency information in instantaneous fluctuation and long-term evolution component data is extracted, further revealing the periodic change law hidden in the data, which helps to understand the dynamic characteristics of the system more deeply. The high-frequency and low-frequency energy proportions are calculated respectively, and the energy distribution in different frequency ranges is quantified, which helps to identify different types of fluctuations and oscillations in the system and provide more comprehensive information for risk assessment. By calculating the spectral entropy, the distribution uniformity of frequency components in the long-term evolution component data is quantified. The higher the spectral entropy, the more uniform the frequency distribution, which helps to judge the stability and potential risks of the system. The various features extracted in the previous steps are combined to form a feature set containing multi-scale and multi-dimensional information, which comprehensively reflects the dynamic characteristics of the system and provides a rich data basis for subsequent risk factor association analysis.

[0043] Preferably, step S26 comprises the following steps:

[0044] Step S261: Select key equipment from the preliminary feature set to obtain a key equipment list;

[0045] Step S262: constructing a physical model according to the time series feature set and the key equipment list to obtain an equipment physical model set;

[0046] Step S263: setting simulation parameters for the device physical model set according to the preliminary feature set to obtain a simulation parameter set;

[0047] Step S264: performing physical simulation calculation on the device physical model set according to the simulation parameter set to obtain a simulation result set;

[0048] Step S265: extracting correlation features from the simulation result set to obtain correlation features; merging correlation features from the correlation features and the preliminary feature set to obtain a correlation feature set.

[0049] The present invention screens out key equipment that has a greater impact on system reliability by analyzing the preliminary feature set and the topological position of the equipment, which helps to concentrate resources to conduct in-depth analysis on these key equipment and improve the efficiency and pertinence of risk assessment. A physical model is constructed for key equipment, such as a thermal model of a transformer and a resistance model of a line, to link the physical characteristics of the equipment with the operating status, providing a basis for subsequent simulation analysis, and being able to more accurately assess the risk level of the equipment. Simulation parameters are set for the physical model according to actual operating data, so that the simulation is closer to the actual operating conditions, and the accuracy and reliability of the simulation results are improved. By performing simulation calculations on the physical model, the operating status data of the equipment under different working conditions, such as the transformer winding temperature and the line resistance value, are obtained, providing a data basis for subsequent associated feature extraction. Associated features are extracted from the simulation results, such as the degree of influence of the load rate on the winding temperature, and these associated features are merged with the preliminary feature set to form a more comprehensive associated feature set, which can more deeply reflect the interaction and influence between risk factors and improve the accuracy and comprehensiveness of risk assessment.

[0050] Preferably, step S3 comprises the following steps:

[0051] Step S31: constructing a power distribution system topology map according to the comprehensive operation data set to obtain a power distribution system topology map;

[0052] Step S32: integrating the risk characteristic matrix into the distribution system topology diagram, constructing a reliability network model, and obtaining a reliability network diagram;

[0053] Step S33: performing node importance evaluation on the reliability network diagram to obtain node importance data; performing node risk value calculation based on the node importance data and the reliability network diagram to obtain a node risk evaluation result;

[0054] Step S34: classify weak links according to the node risk assessment results to obtain a graded risk assessment result;

[0055] Step S35: Visualize the hierarchical risk assessment results into the power distribution system topology diagram, generate a reliability weak point map, and obtain a reliability weak point map.

[0056] The present invention constructs a topological diagram of the power distribution system according to the equipment connection relationship and geographical location information, clearly shows the connection relationship and spatial distribution of each device in the system, and provides a basic framework for the subsequent reliability network model construction. The risk feature matrix is ​​integrated into the topological diagram, the risk information of the equipment is combined with the network topological structure, and a reliability network diagram is constructed, which can more comprehensively reflect the risk status of each device in the system and the risk propagation relationship in the network. The node importance is evaluated by the improved PageRank algorithm, and the node risk value is calculated in combination with the node's own risk factor, which effectively identifies the key nodes that have a greater impact on the system reliability, and provides an important basis for the subsequent classification of weak links. According to the node risk value, the nodes are divided into different risk levels, such as high, medium and low levels, which helps to more clearly identify the weak links in the system and facilitate the operation and maintenance personnel to carry out more targeted maintenance and management. The hierarchical risk assessment results are visualized on the topological diagram, and a reliability weak point map is generated, which shows the risk level of different equipment in the system and the distribution of weak links in an intuitive way, which is convenient for the operation and maintenance personnel to quickly understand the risk status of the system and formulate corresponding maintenance strategies.

[0057] Preferably, step S33 includes the following steps:

[0058] Step S331: Initialize the node risk value of the reliability network graph to obtain the initial node risk value;

[0059] Step S332: Calculate the node load weight according to the comprehensive operation data set to obtain the node load weight;

[0060] Step S333: Calculate the node power supply range weight according to the reliability network diagram to obtain the node power supply range weight;

[0061] Step S334: performing iterative calculation of the node risk value according to the initial node risk value, the reliability network diagram, the node load weight, and the node power supply range weight to obtain an iterative node risk value;

[0062] Step S335: Output the node risk assessment result for the iterative node risk value to obtain the node risk assessment result.

[0063] The present invention provides a starting point for subsequent iterative calculations by assigning an initial risk value to each node, thereby ensuring that the calculation process of the node risk value can be carried out effectively. The load weight of each node is calculated to reflect the load bearing situation of the node in the system, and the load level is included in the risk assessment to make the risk assessment closer to the actual operation situation. The power supply range weight of each node is calculated to reflect the importance of the power supply of the node in the system. The power supply range is included in the risk assessment so that nodes with greater impact on the overall reliability of the system obtain higher risk weights. Through iterative calculations, multiple factors such as the node's own risk, the risk of adjacent nodes, the node load weight, and the power supply range weight are considered, and risk propagation is performed through the network topology structure, so that the calculation of the node risk value is more comprehensive and accurate, and the actual risk level of the node can be more accurately reflected. The final risk value obtained by iterative calculation is output as the node risk assessment result, which provides the final risk assessment data for the subsequent weak link classification and reliability weak point map generation.

[0064] Preferably, step S4 comprises the following steps:

[0065] Step S41: constructing an initial reliability assessment model according to the reliability weak point map and the comprehensive operation data set to obtain an initial reliability assessment model;

[0066] Step S42: using the initial reliability evaluation model as a reinforcement learning agent to construct a reinforcement learning environment to obtain a reinforcement learning environment;

[0067] Step S43: Perform reinforcement learning model training according to the reinforcement learning environment to obtain a trained reliability evaluation model;

[0068] Step S44: performing model verification and optimization on the trained reliability assessment model to obtain an optimized reliability assessment model;

[0069] Step S45: Acquire current system status data; use the optimized reliability evaluation model to perform a comprehensive reliability evaluation on the current system status data to obtain a comprehensive reliability score.

[0070] The present invention constructs an initial reliability assessment model by using a reliability weak point map and a comprehensive operation data set, provides a basic model for subsequent reinforcement learning training, avoids training from scratch, and improves training efficiency. A reinforcement learning environment is constructed, and the state space, action space, and reward function are defined, providing a platform for the reinforcement learning agent to interact with the environment, so that the agent can learn the optimal reliability assessment strategy through interaction with the environment. The initial reliability assessment model is trained by a reinforcement learning algorithm, such as DQN, so that the model can dynamically adjust the assessment strategy according to the state of the system, thereby improving the accuracy and adaptability of the reliability assessment. By using an independent test data set to verify and optimize the trained model, the generalization ability and robustness of the model are ensured, and the reliability of the model in practical applications is improved. The current system state is evaluated using the optimized reliability assessment model to obtain a comprehensive reliability score for the system, providing intuitive and quantitative reliability assessment results for operation and maintenance personnel, which is helpful for better risk management and decision-making.

[0071] Preferably, the present invention further provides a power distribution system reliability analysis system, which is used to execute the power distribution system reliability analysis method as described above, and the power distribution system reliability analysis system comprises:

[0072] The multi-dimensional data fusion module is used to collect multi-dimensional data of the power distribution system through the sensor network, and to fuse the multi-dimensional data to obtain multi-dimensional related data; to construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set;

[0073] The multi-scale feature extraction module is used to segment the time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slow-changing long-term trend sequence; perform time series decomposition on the fast fluctuation characteristic sequence and the slow-changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; perform feature component analysis on the high-frequency fluctuation characteristic data and the long-term trend characteristic data, and perform feature combination to obtain a multi-scale feature set; perform risk factor correlation analysis on the multi-scale feature set to obtain a correlation feature set; and construct a risk feature matrix on the correlation feature set to obtain a risk feature matrix;

[0074] The risk factor association analysis module is used to construct a reliability network model based on the comprehensive operation data set and the risk characteristic matrix to obtain a reliability network diagram; to evaluate the node importance of the reliability network diagram to obtain node importance data; to calculate the node risk value based on the node importance data and the reliability network diagram to obtain the node risk assessment result; to classify the weak links based on the node risk assessment result, and to generate a reliability weak point map to obtain a reliability weak point map;

[0075] The reliability network model building module is used to train the reliability assessment model according to the reliability weakness map and the comprehensive operation data set, and optimize the model to obtain the optimized reliability assessment model; the optimized reliability assessment model is used to perform a comprehensive reliability assessment to obtain a comprehensive reliability score. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 A schematic diagram of the steps of a reliability analysis method for a power distribution system is provided;

[0077] Figure 2 for Figure 1 Detailed implementation steps of step S2 in the flowchart;

[0078] Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG. DETAILED DESCRIPTION

[0079] See also Figures 1 to 3 , a distribution system reliability analysis method, comprising the following steps:

[0080] Step S1: collect multi-dimensional data of the power distribution system through the sensor network, and perform multi-dimensional data fusion to obtain multi-dimensional related data; construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set;

[0081] Step S2: segmenting the time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slow-changing long-term trend sequence; performing time series decomposition on the fast fluctuation characteristic sequence and the slow-changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; performing feature component analysis on the high-frequency fluctuation characteristic data and the long-term trend characteristic data, and performing feature combination to obtain a multi-scale feature set; performing risk factor correlation analysis on the multi-scale feature set to obtain a correlation feature set; constructing a risk feature matrix on the correlation feature set to obtain a risk feature matrix;

[0082] Step S3: construct a reliability network model based on the comprehensive operation data set and the risk characteristic matrix to obtain a reliability network diagram; perform node importance assessment on the reliability network diagram to obtain node importance data; calculate node risk values ​​based on the node importance data and the reliability network diagram to obtain node risk assessment results; perform weak link classification based on the node risk assessment results, and generate a reliability weak point map to obtain a reliability weak point map;

[0083] Step S4: Train the reliability assessment model according to the reliability weak point map and the comprehensive operation data set, and optimize the model to obtain an optimized reliability assessment model; use the optimized reliability assessment model to perform a comprehensive reliability assessment to obtain a comprehensive reliability score.

[0084] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a schematic diagram of the steps of the power distribution system reliability analysis method of the present invention. In this example, the power distribution system reliability analysis method includes the following steps:

[0085] Step S1: collect multi-dimensional data of the power distribution system through the sensor network, and perform multi-dimensional data fusion to obtain multi-dimensional related data; construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set;

[0086] In an embodiment of the present invention, raw data, including current, voltage, temperature, humidity, vibration, etc., are collected by a sensor network deployed at key nodes of the power distribution system, and transmitted to edge nodes through wireless communication protocols such as LoRaWAN. The edge node preprocesses the raw data, for example, by using a sliding window algorithm to remove outliers and analyzing harmonic components through FFT. The preprocessed data is uploaded to the central server and stored in the distributed database HBase. Using knowledge graph technology, data from different data sources are semantically associated, for example, device status data is associated with geographic location and historical fault records. Finally, the associated multidimensional data is aligned and sorted according to a unified timestamp, and the missing data is supplemented by a linear interpolation method, and stored in Parquet format to form a comprehensive operating data set containing device status, operating parameters, environmental data, and historical fault records.

[0087] Step S2: segmenting the time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slow-changing long-term trend sequence; performing time series decomposition on the fast fluctuation characteristic sequence and the slow-changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; performing feature component analysis on the high-frequency fluctuation characteristic data and the long-term trend characteristic data, and performing feature combination to obtain a multi-scale feature set; performing risk factor correlation analysis on the multi-scale feature set to obtain a correlation feature set; constructing a risk feature matrix on the correlation feature set to obtain a risk feature matrix;

[0088] In an embodiment of the present invention, preliminary features are extracted from a comprehensive operation data set, such as transformer load rate, line current imbalance, circuit breaker operation times, etc. Then, the time series data is segmented according to the fluctuation characteristics of the features, for example, the daily load rate curve is used as a slowly changing long-term trend sequence, and the minute current imbalance sequence is used as a fast fluctuation characteristic sequence. The fast fluctuation characteristic sequence is decomposed by db4 wavelet in three layers, and the high-frequency detail coefficient is extracted as high-frequency fluctuation feature data; the slowly changing long-term trend sequence is decomposed by EMD, and the low-frequency IMF component and residual component are extracted as long-term trend feature data. Statistical features and spectral features are calculated for high-frequency and low-frequency feature data, respectively, and these features are combined to form a multi-scale feature set. Finally, the multi-scale features are fused, for example, by using a feature splicing method, and risk factor association analysis is performed, for example, the influence of load rate on temperature is analyzed by physical model simulation, and the obtained associated features are merged with the preliminary features to form an associated feature set. The PCA method is used to reduce the dimension of the associated feature set, and key features are selected to finally construct a risk feature matrix.

[0089] Step S3: construct a reliability network model based on the comprehensive operation data set and the risk characteristic matrix to obtain a reliability network diagram; perform node importance assessment on the reliability network diagram to obtain node importance data; calculate node risk values ​​based on the node importance data and the reliability network diagram to obtain node risk assessment results; perform weak link classification based on the node risk assessment results, and generate a reliability weak point map to obtain a reliability weak point map;

[0090] In an embodiment of the present invention, a distribution system topology is constructed based on the equipment connection relationship and geographic location information in the comprehensive operation data set. Then, the risk feature matrix is ​​integrated into the topology, the node risk feature is added to the node attribute, and the weight of the edge is calculated according to the mean of the risk factor of the adjacent nodes to form a reliability network diagram. Using the improved PageRank algorithm, the node load level is used as the damping factor to calculate the node importance. The node risk value is calculated by multiplying the mean of the node's own risk factor by the node importance. According to the node risk value, the K-means algorithm (K=3) is used to divide the nodes into three risk levels: high, medium, and low. Finally, the grading results are visualized in the topology, and different colors and sizes are used to represent the node risk level and risk value to generate a reliability weak point map.

[0091] Step S4: training a reliability assessment model according to the reliability weak point map and the comprehensive operation data set, and optimizing the model to obtain an optimized reliability assessment model; performing a comprehensive reliability assessment using the optimized reliability assessment model to obtain a comprehensive reliability score;

[0092] In an embodiment of the present invention, an initial reliability assessment model based on SVM is constructed based on the reliability weakness map and the historical power outage data in the comprehensive operation data set. The model input is the node risk level and the historical power outage data, and the output is the node reliability score. Then, a reinforcement learning environment is constructed, the state space is the node risk level and the historical power outage data, the action space is the preventive maintenance measures, and the reward function is the overall reliability level of the system (the weighted average of the node reliability score, and the weight is the node power supply range weight). The DQN algorithm is used to train the reinforcement learning agent, and the goal is to maximize the cumulative reward. The model performance is verified using an independent test set, and the model is optimized by adjusting the DQN parameters until the mean square error MSE meets the requirements. Finally, the current system state data is input into the optimized model, the reliability score of each node is calculated, and the weighted average is used to obtain the comprehensive reliability score of the system.

[0093] Step S1 includes the following steps:

[0094] Step S11: collecting sensor data of the power distribution system through a sensor network to obtain raw sensor data;

[0095] Step S12: performing edge data preprocessing on the original sensor data to obtain preprocessed sensor data;

[0096] Step S13: uploading and centrally storing the pre-processed sensor data to obtain centralized data;

[0097] Step S14: performing semantic data modeling and fusion on the centralized data to obtain multi-dimensional associated data;

[0098] Step S15: construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set.

[0099] In an embodiment of the present invention, a sensor network is deployed at key nodes (transformers, circuit breakers, lines, etc.) of the power distribution system. Current transformers and voltage transformers measure line current and voltage respectively, temperature sensors monitor transformer winding temperature, vibration sensors monitor transformer core vibration, and environmental monitoring units collect data such as ambient temperature, humidity, and wind speed. All sensor data is collected once per second and transmitted to edge computing nodes in real time through wireless communication protocols (such as LoRaWAN) to form a raw sensor data set corresponding to timestamps and values. For example, if the temperature value collected by a transformer temperature sensor at 10:00:00 on August 27, 2024 is 60°C, it is recorded as (2024-08-27, 10:00:00, 60).

[0100] After receiving the raw sensor data, the edge computing node immediately performs preprocessing. For temperature data, a sliding window algorithm is used, and the window size is set to 5 seconds. It is determined whether the current data is within the range of the mean of the data in the window plus or minus three times the standard deviation. If it exceeds the range, it is determined to be an outlier and replaced with the previous valid value. For current and voltage data, the fast Fourier transform (FFT) is used to analyze its harmonic components, and the harmonic amplitude is added to the data set as a new feature. All preprocessed data is stored in the local database of the edge node and uploaded to the central server every minute. For example, the 5-second data collected by a line current sensor is [100, 102, 101, 99, 150]. The sliding window algorithm determines that 150 is an outlier and replaces it with the previous valid value 99. The preprocessed data is [100, 102, 101, 99, 99].

[0101] The edge node uploads the pre-processed sensor data to the central server through a secure and encrypted network connection (such as VPN). The central server uses a distributed database (such as Hadoop HBase) to store all data. The database table structure is designed as: timestamp, device ID, data type, data value. After the data is uploaded, the system automatically verifies the integrity and consistency of the data. If the data is missing or wrong, a retransmission request is sent to the corresponding edge node. For example, the data record uploaded by an edge node is (2024-08-27, 10:01:00, Transformer_001, Temperature, 61), which will be stored in the corresponding table of the central database.

[0102] The semantic model of the distribution system is constructed using knowledge graph technology. Entities such as equipment, lines, and regions are taken as nodes, and the connection relationship, functional dependency relationship, and risk factor association relationship between them are taken as edges. For example, there is a physical connection relationship between the transformer and the line, and there is a risk factor association relationship between the transformer and the ambient temperature. The SPARQL query language is used to query and associate centralized data, such as querying the temperature, load current, and ambient temperature of a transformer in a specific time period, and merging the query results into a multi-dimensional associated data record. For example, the query result is: (2024-08-27, 10:00:00-10:05:00, Transformer_001, Temperature: [60, 61, 62, 61, 60], Current: [100, 102, 101, 99, 99], Ambient Temperature: [30, 31, 31, 30, 30]).

[0103] Align and sort multidimensional related data according to a unified timestamp. For data with different sampling frequencies, linear interpolation is used for data completion. All data is stored in Parquet format and indexed to improve query efficiency. The comprehensive operation data set contains all related data, such as equipment status data, operating parameters, environmental data, and historical fault records, and is arranged in chronological order to form a complete data time series for subsequent analysis and modeling. For example, a record of the comprehensive operation data set may contain the measurement values ​​of all sensors at a certain point in time, the equipment operating status, environmental data, and the corresponding historical fault information. This record example: (2024-08-27, 10:00:00, Transformer_001, Temperature: 60, Current: 100, Ambient Temperature: 30, Humidity: 70, Wind Speed: 5, Operating Status: Normal, Historical Fault: None, Line_001: Current: 150, Voltage: 10kV, ...). This dataset will provide a complete data foundation for subsequent feature extraction, model training, and reliability assessment.

[0104] Step S2 includes the following steps:

[0105] Step S21: performing preliminary feature extraction based on the comprehensive operation data set to obtain a preliminary feature set;

[0106] Step S22: segmenting the comprehensive operation data set into time series data according to the preliminary feature set to obtain a fast fluctuation characteristic sequence and a slowly changing long-term trend sequence;

[0107] Step S23: performing time series decomposition on the fast fluctuation characteristic sequence and the slowly changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data;

[0108] Step S24: performing feature component analysis on the high-frequency fluctuation feature data and the long-term trend feature data, and performing feature combination to obtain a multi-scale feature set;

[0109] Step S25: performing multi-scale feature fusion on the multi-scale feature set to obtain a time series feature set;

[0110] Step S26: performing risk factor association analysis based on the preliminary feature set and the time series feature set to obtain an associated feature set;

[0111] Step S27: Perform feature screening and dimensionality reduction on the associated feature set to obtain a key feature set; construct a risk feature matrix on the key feature set to obtain a risk feature matrix.

[0112] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes:

[0113] Step S21: performing preliminary feature extraction based on the comprehensive operation data set to obtain a preliminary feature set;

[0114] In an embodiment of the present invention, predefined preliminary features are extracted from a comprehensive operation data set. For each transformer, the load rate per hour is calculated, and the formula is: load rate = current load / rated capacity × 100%. For each line, the three-phase current imbalance is calculated, and the formula is: (maximum phase current-minimum phase current) / average phase current × 100%. The daily operation times of each circuit breaker are counted. Environmental features are obtained directly from the environmental monitoring unit. All extracted features are stored according to timestamp, device ID and feature type to form a preliminary feature set. For example, the load rate of transformer `Transformer_001` at 10:00:00 on August 27, 2024 is 80%, which is recorded as (2024-08-27, 10:00:00, Transformer_001, load rate, 80).

[0115] Step S22: segmenting the comprehensive operation data set into time series data according to the preliminary feature set to obtain a fast fluctuation characteristic sequence and a slowly changing long-term trend sequence;

[0116] In the embodiment of the present invention, the time series is segmented according to the fluctuation characteristics of the preliminary features. Taking the transformer load rate as an example, it is segmented according to 24 hours a day to generate a daily load rate curve. The line three-phase current imbalance data is segmented according to each minute to generate a current imbalance sequence per minute. The daily load rate curve is regarded as a slowly changing long-term trend sequence, and the current imbalance sequence per minute is regarded as a fast fluctuation characteristic sequence. Other features are also segmented accordingly according to their fluctuation characteristics.

[0117] Step S23: performing time series decomposition on the fast fluctuation characteristic sequence and the slowly changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data;

[0118] In the embodiment of the present invention, the wavelet transform is used to decompose the fast fluctuation characteristic sequence, the wavelet basis function selects `db4`, and the decomposition level is set to 3. Taking the line current imbalance as an example, high-frequency detail coefficients and low-frequency approximation coefficients are obtained after decomposition. The high-frequency detail coefficients reflect the instantaneous fluctuation of the current and serve as high-frequency fluctuation characteristic data. For the slowly changing long-term trend sequence, empirical mode decomposition (EMD) is used for decomposition, and the stopping criterion is set to the energy proportion of the IMF component is less than 0.01. Taking the daily transformer load rate curve as an example, multiple IMF components and a residual component are obtained after decomposition, wherein the low-frequency IMF component and the residual component reflect the long-term trend of the load rate and serve as the long-term trend characteristic data.

[0119] Step S24: performing feature component analysis on the high-frequency fluctuation feature data and the long-term trend feature data, and performing feature combination to obtain a multi-scale feature set;

[0120] In the embodiment of the present invention, for the high-frequency fluctuation characteristic data, the statistical features such as the root mean square value, peak factor and waveform factor are calculated. For the long-term trend characteristic data, the statistical features such as the mean value, standard deviation and trend slope are calculated. In addition, the original time series is subjected to a fast Fourier transform (FFT) to extract the spectrum features, such as the main frequency and spectrum energy distribution. All the extracted features are combined to form a multi-scale feature set, for example, the high-frequency fluctuation features and spectrum features of the line current imbalance are combined together.

[0121] Step S25: performing multi-scale feature fusion on the multi-scale feature set to obtain a time series feature set;

[0122] In an embodiment of the present invention, features of different time scales are fused. For example, the long-term trend feature of the daily load rate of the transformer and the high-frequency fluctuation feature of the current imbalance degree per minute in the corresponding time period are fused. The fusion method uses feature concatenation to connect different feature vectors into a longer feature vector. The vector contains feature information at different time scales to form a time series feature set.

[0123] Step S26: performing risk factor association analysis based on the preliminary feature set and the time series feature set to obtain an associated feature set;

[0124] In the embodiment of the present invention, the correlation between the preliminary features and the time series features is studied by using a physical model and simulation analysis. For example, a physical model between transformer temperature and load rate is established, and the influence of load rate on temperature is quantified through simulation analysis. The simulation results are compared and verified with the actual data, and the verified correlation is added as a new feature to the feature set to form an associated feature set. For example, "sensitivity of transformer temperature to load rate changes" can be used as an associated feature.

[0125] Step S27: performing feature screening and dimensionality reduction on the associated feature set to obtain a key feature set; constructing a risk feature matrix on the key feature set to obtain a risk feature matrix;

[0126] In an embodiment of the present invention, the principal component analysis (PCA) method is used to reduce the dimension of the associated feature set, and the principal components that can explain 95% of the data variance are selected. The selected principal components and other important associated features form a key feature set. Finally, the key feature set is organized into a risk feature matrix. The rows of the matrix represent different devices or lines, and the columns represent different key features. The value of each cell represents the value of the corresponding device or line on the corresponding feature. For example, a row in the matrix may represent `Transformer_001`, and its corresponding column values ​​are the values ​​of the key features of the transformer, such as load rate, temperature, and aging degree. This matrix is ​​the risk feature matrix, which is used for subsequent reliability analysis.

[0127] Step S23 includes the following steps:

[0128] Step S231: setting wavelet transform parameters for the rapid fluctuation characteristic sequence to obtain wavelet transform parameters;

[0129] Step S232: setting a stop criterion for performing empirical mode decomposition on the slowly changing long-term trend sequence to obtain stop criterion data;

[0130] Step S233: extracting high-frequency fluctuation characteristics from the rapid fluctuation characteristic sequence according to the wavelet transform parameters to obtain high-frequency fluctuation characteristic data;

[0131] Step S234: extracting long-term trend features from the slowly changing long-term trend sequence according to the stopping criterion data to obtain long-term trend feature data.

[0132] In the embodiment of the present invention, for fast fluctuation characteristic sequences, such as line current imbalance data per minute, discrete wavelet transform (DWT) is used. The wavelet basis function selects Daubechies 4 (db4) wavelet because it performs well in capturing transient characteristics of power system signals. The number of decomposition layers is set to 3 to extract detailed information of different frequency bands. These parameters (wavelet basis function: db4, number of decomposition layers: 3) together constitute wavelet transform parameters and serve as input parameters for subsequent wavelet decomposition.

[0133] For slowly changing long-term trend series, such as daily transformer load rate curves, the empirical mode decomposition (EMD) method is used. The stopping criterion is set as follows: when the energy proportion of the IMF component is less than the preset threshold of 0.01, or when the number of IMF components reaches 10, the decomposition is stopped. The energy proportion is calculated as: the energy of the current IMF component divided by the energy of the original signal. These two stopping criteria (energy proportion threshold: 0.01, IMF component number threshold: 10) together constitute the stopping criterion data, which is used to control the EMD decomposition process.

[0134] The fast fluctuation characteristic sequence is subjected to discrete wavelet transform using wavelet transform parameters (db4 wavelet, decomposition level 3). After decomposition, three layers of detail coefficients (D1, D2, D3) and one approximate coefficient (A3) are obtained. Since the goal is to extract high-frequency fluctuation characteristics, detail coefficients D1, D2 and D3 with higher decomposition levels and richer frequency information are selected as high-frequency fluctuation characteristic data. These detail coefficients are stored at the corresponding time points of the original time series to form high-frequency fluctuation characteristic data.

[0135] The empirical mode decomposition (EMD) of the slowly changing long-term trend series is performed using the stopping criteria (energy ratio threshold 0.01, IMF component number threshold 10). The decomposition process continues until the stopping criteria are met. After decomposition, multiple IMF components and a residual component are obtained. The IMF components with lower frequencies (for example, the last three IMF components) and the residual components are selected as long-term trend feature data because they represent the long-term change trend of the original sequence. These selected IMF components and residual components are stored at the corresponding time points of the original time series to form long-term trend feature data.

[0136] Step S24 includes the following steps:

[0137] Step S241: Calculate the data standard deviation, data maximum change rate and data peak value of the high-frequency fluctuation characteristic data to obtain instantaneous fluctuation component data;

[0138] Step S242: performing data mean calculation, data slope calculation and data long-term offset calculation on the long-term trend characteristic data to obtain long-term evolution component data;

[0139] Step S243: performing main frequency component analysis on the instantaneous fluctuation component data and the long-term evolution component data to obtain main frequency component data;

[0140] Step S244: Calculate the high-frequency energy proportion of the instantaneous fluctuation component data to obtain high-frequency energy proportion data; calculate the low-frequency energy proportion of the long-term evolution component data to obtain low-frequency energy proportion data;

[0141] Step S245: performing spectrum entropy analysis on the long-term evolution component data to obtain spectrum entropy;

[0142] Step S246: Feature combination is performed on the instantaneous fluctuation component data, the long-term evolution component data, the main frequency component data, the high-frequency energy proportion data, the low-frequency energy proportion data and the spectrum entropy to obtain a multi-scale feature set.

[0143] In an embodiment of the present invention, for high-frequency fluctuation characteristic data (for example, wavelet transform detail coefficients D1, D2, D3 of line current imbalance), the standard deviation, maximum change rate and peak value in each time window are calculated respectively. The standard deviation is calculated using a standard formula. The maximum change rate is defined as the maximum absolute value of the difference between two adjacent data points in the time window. The peak value is defined as the maximum absolute value of the data in the time window. The calculated standard deviation, maximum change rate and peak value are arranged in the order of the time window to form instantaneous fluctuation component data.

[0144] For long-term trend characteristic data (for example, the low-frequency IMF component and residual component after EMD decomposition of transformer load factor), the mean, slope and long-term offset in each time window are calculated respectively. The mean is calculated using the standard formula. The slope calculation method is to perform a linear fit on the data in the time window, and the slope of the fitted line is the slope value of the time window. The long-term offset is defined as the difference between the last data point and the first data point in the time window. The calculated mean, slope and long-term offset are arranged in the order of the time window to form the long-term evolution component data.

[0145] Perform fast Fourier transform (FFT) on the instantaneous fluctuation component data and the long-term evolution component data respectively. For the spectrum data after FFT transformation, find the frequency component with the largest amplitude, which is the main frequency. Extract the main frequency of each time window and arrange it in the order of time windows to form the main frequency component data. Among them, the main frequency analysis here is performed on the characteristic data obtained by S241 and S242, rather than on the original time series.

[0146] The spectrum of the instantaneous fluctuation component data is analyzed to calculate the energy proportion of the high-frequency part. The frequency range of the high-frequency part is predefined, for example, set to a frequency range higher than 0.1 Hz. The high-frequency energy proportion is calculated as follows: the energy of the high-frequency part is divided by the total energy. The spectrum of the long-term evolution component data is analyzed to calculate the energy proportion of the low-frequency part. The frequency range of the low-frequency part is predefined, for example, set to a frequency range lower than 0.01 Hz. The low-frequency energy proportion is calculated as follows: the energy of the low-frequency part is divided by the total energy.

[0147] Perform spectrum analysis on the long-term evolution component data and calculate its spectrum entropy. The spectrum entropy is calculated as follows: first, normalize the spectrum energy, then use the normalized spectrum energy as the probability distribution and calculate its information entropy. The spectrum entropy reflects the distribution of spectrum energy. The more uniform the spectrum energy distribution, the greater the entropy value.

[0148] The instantaneous fluctuation component data, long-term evolution component data, main frequency component data, high-frequency energy proportion data, low-frequency energy proportion data and spectrum entropy calculated in steps S241 to S245 are spliced ​​together according to the time window to form a new feature vector. Each time window corresponds to a feature vector, and these feature vectors are combined to form a multi-scale feature set. The feature set contains comprehensive information of different time scales and different feature types, which is used for subsequent risk factor association analysis.

[0149] Step S26 includes the following steps:

[0150] Step S261: Select key equipment from the preliminary feature set to obtain a key equipment list;

[0151] Step S262: constructing a physical model according to the time series feature set and the key equipment list to obtain an equipment physical model set;

[0152] Step S263: setting simulation parameters for the device physical model set according to the preliminary feature set to obtain a simulation parameter set;

[0153] Step S264: performing physical simulation calculation on the device physical model set according to the simulation parameter set to obtain a simulation result set;

[0154] Step S265: extracting correlation features from the simulation result set to obtain correlation features; merging correlation features from the correlation features and the preliminary feature set to obtain a correlation feature set.

[0155] In the embodiment of the present invention, the key equipment that has a greater impact on system reliability is determined based on the data in the preliminary feature set, such as the number of historical faults, the range of load rate changes, and the topological position of the equipment in the power distribution network, such as the main transformer of the substation, the tie switch, etc. The ID information of these key equipment is recorded to form a list of key equipment. For example, the list of key equipment may include transformers Transformer_001, Transformer_002, circuit breakers Breaker_001, Breaker_002, lines Line_001, Line_002, etc.

[0156] For each device in the key device list, build a corresponding physical model. For example, for a transformer, build a thermal model based on its nameplate parameters and structural information. This model describes the relationship between the transformer winding temperature and the load current and ambient temperature. For a line, build a resistance model based on its length, cross-sectional area, and material parameters. This model describes the relationship between the resistance value of the line and the temperature. Summarize all the constructed physical models to form a set of equipment physical models.

[0157] According to the data in the preliminary feature set, set the simulation parameters for the physical model of the equipment. For example, the load current, ambient temperature and other data of the transformer are used as input parameters and substituted into the transformer thermal model. The temperature data of the line is used as an input parameter and substituted into the line resistance model. All the set simulation parameters are summarized to form a simulation parameter set. Each parameter set corresponds to a specific time point and a specific key equipment.

[0158] Use numerical calculation methods to simulate the physical model of the device. For example, use the finite element analysis method to calculate the temperature distribution of the transformer winding under different load currents and ambient temperatures. Use the circuit analysis method to calculate the resistance and loss of the line at different temperatures. Record the results of each simulation calculation, such as the maximum temperature of the transformer winding, the resistance value of the line, etc., as well as the corresponding simulation parameters and timestamps to form a simulation result set.

[0159] Extract associated features from the simulation result set. For example, extract the relationship between the maximum temperature of the transformer winding and the load rate, and quantify the influence of the load rate on the winding temperature. Extract the relationship between the line resistance value and the ambient temperature, and quantify the influence of the ambient temperature on the line resistance. Merge the extracted associated features with the preliminary feature set to form a new associated feature set. For example, add "sensitivity of the maximum temperature of the transformer winding to changes in the load rate" as a new associated feature to the associated feature set.

[0160] Step S3 includes the following steps:

[0161] Step S31: constructing a power distribution system topology map according to the comprehensive operation data set to obtain a power distribution system topology map;

[0162] Step S32: integrating the risk characteristic matrix into the distribution system topology diagram, constructing a reliability network model, and obtaining a reliability network diagram;

[0163] Step S33: performing node importance evaluation on the reliability network diagram to obtain node importance data; performing node risk value calculation based on the node importance data and the reliability network diagram to obtain a node risk evaluation result;

[0164] Step S34: classify weak links according to the node risk assessment results to obtain a graded risk assessment result;

[0165] Step S35: Visualize the hierarchical risk assessment results into the power distribution system topology diagram, generate a reliability weak point map, and obtain a reliability weak point map.

[0166] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes:

[0167] Step S31: constructing a power distribution system topology map according to the comprehensive operation data set to obtain a power distribution system topology map;

[0168] In an embodiment of the present invention, the equipment connection relationship and geographic location information are extracted from the comprehensive operation data set. Each device, such as a transformer, a circuit breaker, a line, etc., is represented as a node in a graph. The node attributes include information such as device ID, type, rated parameters, and geographic coordinates. If there is a physical connection between two devices, an edge is added to the graph to connect the two nodes. The attributes of the edge include information such as connection type, length, impedance, etc. The constructed topology map accurately reflects the physical connection structure of the distribution system, for example, the connection relationship between substations, lines, and switches, as well as their geographical location relationship.

[0169] Step S32: integrating the risk characteristic matrix into the distribution system topology diagram, constructing a reliability network model, and obtaining a reliability network diagram;

[0170] In an embodiment of the present invention, the characteristic data in the risk characteristic matrix is ​​added to the node attributes of the distribution system topology map. The attributes of each node now contain the basic information of the equipment and the corresponding risk characteristics, such as load rate, aging degree, etc. In addition, the weight of the edge is calculated according to the risk characteristics of the adjacent nodes. The weight calculation method of the edge is: 1 / (mean value of the risk factor of the adjacent nodes). By integrating the risk characteristics into the topology map, a reliability network map is constructed, which reflects the degree of risk association between different devices in the distribution system.

[0171] Step S33: performing node importance evaluation on the reliability network diagram to obtain node importance data; performing node risk value calculation based on the node importance data and the reliability network diagram to obtain a node risk evaluation result;

[0172] In an embodiment of the present invention, an improved PageRank algorithm is used to calculate the importance of each node in the reliability network diagram. Based on the traditional PageRank algorithm, the improved PageRank algorithm uses the load level of the node as a damping factor. The higher the load level of the node, the higher its importance. The calculated node importance data is stored. The calculation method of the node risk value is: the mean value of the node's own risk factor × the node importance. The calculated node risk value is stored to form a node risk assessment result.

[0173] Step S34: classify weak links according to the node risk assessment results to obtain a graded risk assessment result;

[0174] In the embodiment of the present invention, the nodes are divided into risk levels according to the node risk assessment results. The K-means clustering algorithm is used to divide all nodes into three clusters, representing high risk, medium risk and low risk levels respectively. The parameter K of the K-means algorithm is set to 3, and the initial clustering center selects the three nodes with the highest risk value, the middle risk value and the lowest risk value. After the clustering is completed, each node is marked as a high, medium or low risk level to form a graded risk assessment result.

[0175] Step S35: Visualize the hierarchical risk assessment results into the power distribution system topology diagram, generate a reliability weak point map, and obtain a reliability weak point map;

[0176] In an embodiment of the present invention, the hierarchical risk assessment results are visualized in a distribution system topology map. Different colors are used to represent different risk levels, for example, red represents high risk, yellow represents medium risk, and green represents low risk. Nodes are marked on a geographic location map and assigned corresponding colors according to their risk levels. In addition, the size of the risk value can be indicated by adjusting the size of the node. The larger the risk value, the larger the display size of the node. The reliability weakness map finally generated intuitively shows the risk level of different equipment in the distribution system and the distribution of weak links.

[0177] Step S33 includes the following steps:

[0178] Step S331: Initialize the node risk value of the reliability network graph to obtain the initial node risk value;

[0179] Step S332: Calculate the node load weight according to the comprehensive operation data set to obtain the node load weight;

[0180] Step S333: Calculate the node power supply range weight according to the reliability network diagram to obtain the node power supply range weight;

[0181] Step S334: performing iterative calculation of the node risk value according to the initial node risk value, the reliability network diagram, the node load weight, and the node power supply range weight to obtain an iterative node risk value;

[0182] Step S335: Output the node risk assessment result for the iterative node risk value to obtain the node risk assessment result.

[0183] In an embodiment of the present invention, for each node in the reliability network diagram, the initial risk value is calculated according to the risk factor corresponding to it in the risk feature matrix. The calculation method is: take the average value of all risk factors of the node. For example, the risk factors of the node `Transformer_001` include load rate, aging degree and historical failure number, then the initial risk value is (load rate + aging degree + historical failure number) / 3. The initial risk values ​​of all nodes are stored to form an initial node risk value set.

[0184] According to the load data in the comprehensive operation data set, the load weight of each node is calculated. The calculation method is: node load / total system load. For example, the load of node `Transformer_001` is 10MW, and the total system load is 100MW, then the load weight of the node is 10 / 100=0.1. The load weights of all nodes are stored to form a node load weight set.

[0185] According to the topological structure of the reliability network diagram, the power supply range weight of each node is calculated. The calculation method is: count the sum of the loads of all nodes downstream of the node, and then divide it by the total load of the whole system. The downstream node refers to the node that is powered by the node. For example, there are two transformers downstream of the node `Breaker_001`, with loads of 5MW and 7MW respectively, and the total load of the whole system is 100MW. Then the power supply range weight of the circuit breaker is (5+7) / 100=0.12. The power supply range weights of all nodes are stored to form a node power supply range weight set.

[0186] Use the following formula to iteratively calculate the node risk value:

[0187] Node risk value (t+1) = α × initial node risk value + β × Σ [adjacent node risk value (t) × edge weight] + γ × node load weight + δ × node power supply range weight;

[0188] Wherein, t represents the number of iterations, and α, β, γ and δ are predefined weight coefficients, such as α=0.2, β=0.5, γ=0.2, δ=0.1. The edge weight is the weight of the edge calculated in step S32. The iteration process continues until the risk value change of all nodes is less than a preset threshold (e.g., 0.001) or the maximum number of iterations (e.g., 100) is reached. The final node risk value is stored to form an iterated node risk value set.

[0189] The iterated node risk value is output as the node risk assessment result. The risk assessment result of each node contains the node ID and the corresponding risk value. For example, the risk assessment result of the node `Transformer_001` is (Transformer_001, 0.85). These data will be used for subsequent weak link classification and reliability weak point map generation.

[0190] Step S4 includes the following steps:

[0191] Step S41: constructing an initial reliability assessment model according to the reliability weak point map and the comprehensive operation data set to obtain an initial reliability assessment model;

[0192] Step S42: using the initial reliability evaluation model as a reinforcement learning agent to construct a reinforcement learning environment to obtain a reinforcement learning environment;

[0193] Step S43: Perform reinforcement learning model training according to the reinforcement learning environment to obtain a trained reliability evaluation model;

[0194] Step S44: performing model verification and optimization on the trained reliability assessment model to obtain an optimized reliability assessment model;

[0195] Step S45: Acquire current system status data; use the optimized reliability evaluation model to perform a comprehensive reliability evaluation on the current system status data to obtain a comprehensive reliability score.

[0196] In an embodiment of the present invention, an initial reliability assessment model based on a support vector machine (SVM) is constructed using the risk level of each node in the reliability weakness map and the historical power outage data in the comprehensive operation data set. The input features of the model are the risk level of the node (high, medium, low), as well as the number of power outages and the duration of power outages of the node in the past year. The output of the model is the reliability score of the node, which ranges from 0 to 1. The higher the score, the higher the reliability. The SVM model is trained using historical data, the radial basis function (RBF) is selected as the kernel function, and the penalty coefficient C is set to 1.0.

[0197] A reinforcement learning environment is constructed to train and optimize the initial reliability assessment model. The state space of the environment is defined as the risk level and historical power outage data of all nodes. The action space is defined as the preventive maintenance measures taken for each node, such as replacing equipment, adding redundant lines, etc. The reward function is defined as the overall reliability level of the system. The higher the reliability level, the higher the reward. The reliability level is calculated as the weighted average of the reliability scores of all nodes, with the weight being the power supply range weight of the node.

[0198] The reinforcement learning agent is trained using the Deep Q Network (DQN) algorithm. The neural network structure of DQN is a three-layer fully connected network, with the number of input layer nodes equal to the dimension of the state space and the number of output layer nodes equal to the dimension of the action space. The experience replay mechanism is used to store the training data, and the target network is used to improve the training stability. The goal of training is to maximize the cumulative reward obtained by the agent in the reinforcement learning environment. After the training is completed, a trained reliability assessment model is obtained, which can select the best preventive maintenance measures according to the state of the system.

[0199] The trained reliability assessment model is verified using an independent test data set. The data in the test data set is not used in model training. The evaluation indicator is the mean square error (MSE) between the reliability score predicted by the model and the actual reliability score. If the MSE is greater than the preset threshold (e.g. 0.01), the parameters of the DQN algorithm, such as the learning rate, discount factor, etc., are adjusted and retrained. This process is repeated until the MSE of the model meets the requirements. The final model is the optimized reliability assessment model.

[0200] The current system status data is obtained from the sensor network and database, including the risk level of each node, historical power outage data, etc. These data are input into the optimized reliability assessment model to calculate the reliability score of each node. The system's comprehensive reliability score is calculated as the weighted average of the reliability scores of all nodes, with the weight being the power supply range weight of the node. For example, if there are three nodes in the system, with reliability scores of 0.9, 0.8, and 0.7, respectively, and power supply range weights of 0.5, 0.3, and 0.2, respectively, the system's comprehensive reliability score is 0.9×0.5+0.8×0.3+0.70.2=0.83.

[0201] The present invention also provides a power distribution system reliability analysis system, which is used to execute the power distribution system reliability analysis method as described above. The power distribution system reliability analysis system comprises:

[0202] The multi-dimensional data fusion module is used to collect multi-dimensional data of the power distribution system through the sensor network, and to fuse the multi-dimensional data to obtain multi-dimensional related data; to construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set;

[0203] The multi-scale feature extraction module is used to segment the time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slow-changing long-term trend sequence; perform time series decomposition on the fast fluctuation characteristic sequence and the slow-changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; perform feature component analysis on the high-frequency fluctuation characteristic data and the long-term trend characteristic data, and perform feature combination to obtain a multi-scale feature set; perform risk factor correlation analysis on the multi-scale feature set to obtain a correlation feature set; and construct a risk feature matrix on the correlation feature set to obtain a risk feature matrix;

[0204] The risk factor association analysis module is used to construct a reliability network model based on the comprehensive operation data set and the risk characteristic matrix to obtain a reliability network diagram; to evaluate the node importance of the reliability network diagram to obtain node importance data; to calculate the node risk value based on the node importance data and the reliability network diagram to obtain the node risk assessment result; to classify the weak links based on the node risk assessment result, and to generate a reliability weak point map to obtain a reliability weak point map;

[0205] The reliability network model building module is used to train the reliability assessment model according to the reliability weakness map and the comprehensive operation data set, and optimize the model to obtain the optimized reliability assessment model; the optimized reliability assessment model is used to perform a comprehensive reliability assessment to obtain a comprehensive reliability score.

[0206] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0207] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for analyzing the reliability of a power distribution system, characterized in that: The following steps are involved: Step S1: collect multi-dimensional data of the power distribution system through the sensor network, and perform multi-dimensional data fusion to obtain multi-dimensional correlation data; Constructing a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set; Step S2: segmenting the time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slowly changing long-term trend sequence; Perform time series decomposition on the fast fluctuation characteristic series and the slowly changing long-term trend series to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; Perform feature component analysis on high-frequency fluctuation feature data and long-term trend feature data, and perform feature combination to obtain a multi-scale feature set; perform risk factor correlation analysis based on the multi-scale feature set to obtain a correlation feature set; Constructing a risk feature matrix for the associated feature set to obtain a risk feature matrix; Step S3: construct a reliability network model based on the comprehensive operation data set and the risk characteristic matrix to obtain a reliability network diagram; Conduct node importance evaluation on the reliability network diagram to obtain node importance data; Calculate the node risk value based on the node importance data and the reliability network diagram to obtain the node risk assessment result; classify the weak links based on the node risk assessment result, and generate a reliability weak point map to obtain a reliability weak point map; Step S4: Train the reliability assessment model according to the reliability weak point map and the comprehensive operation data set, and optimize the model to obtain an optimized reliability assessment model; use the optimized reliability assessment model to perform a comprehensive reliability assessment to obtain a comprehensive reliability score.

2. The power distribution system reliability analysis method according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: collecting sensor data of the power distribution system through a sensor network to obtain raw sensor data; Step S12: performing edge data preprocessing on the original sensor data to obtain preprocessed sensor data; Step S13: uploading and centrally storing the pre-processed sensor data to obtain centralized data; Step S14: performing semantic data modeling and fusion on the centralized data to obtain multi-dimensional associated data; Step S15: construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set.

3. The power distribution system reliability analysis method according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing preliminary feature extraction based on the comprehensive operation data set to obtain a preliminary feature set; Step S22: segmenting the comprehensive operation data set into time series data according to the preliminary feature set to obtain a fast fluctuation characteristic sequence and a slowly changing long-term trend sequence; Step S23: performing time series decomposition on the fast fluctuation characteristic sequence and the slowly changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; Step S24: performing feature component analysis on the high-frequency fluctuation feature data and the long-term trend feature data, and performing feature combination to obtain a multi-scale feature set; Step S25: performing multi-scale feature fusion on the multi-scale feature set to obtain a time series feature set; Step S26: performing risk factor association analysis based on the preliminary feature set and the time series feature set to obtain an associated feature set; Step S27: Perform feature screening and dimensionality reduction on the associated feature set to obtain a key feature set; construct a risk feature matrix on the key feature set to obtain a risk feature matrix.

4. The power distribution system reliability analysis method according to claim 3, characterized in that: Step S23 includes the following steps: Step S231: setting wavelet transform parameters for the rapid fluctuation characteristic sequence to obtain wavelet transform parameters; Step S232: setting a stop criterion for performing empirical mode decomposition on the slowly changing long-term trend sequence to obtain stop criterion data; Step S233: extracting high-frequency fluctuation characteristics from the rapid fluctuation characteristic sequence according to the wavelet transform parameters to obtain high-frequency fluctuation characteristic data; Step S234: extracting long-term trend features from the slowly changing long-term trend sequence according to the stopping criterion data to obtain long-term trend feature data.

5. The power distribution system reliability analysis method according to claim 3, characterized in that: Step S24 includes the following steps: Step S241: Calculate the data standard deviation, data maximum change rate and data peak value of the high-frequency fluctuation characteristic data to obtain instantaneous fluctuation component data; Step S242: performing data mean calculation, data slope calculation and data long-term offset calculation on the long-term trend characteristic data to obtain long-term evolution component data; Step S243: performing main frequency component analysis on the instantaneous fluctuation component data and the long-term evolution component data to obtain main frequency component data; Step S244: Calculate the high-frequency energy proportion of the instantaneous fluctuation component data to obtain high-frequency energy proportion data; calculate the low-frequency energy proportion of the long-term evolution component data to obtain low-frequency energy proportion data; Step S245: performing spectrum entropy analysis on the long-term evolution component data to obtain spectrum entropy; Step S246: Perform feature combination on the instantaneous fluctuation component data, the long-term evolution component data, the main frequency component data, the high-frequency energy proportion data, the low-frequency energy proportion data and the spectrum entropy to obtain a multi-scale feature set.

6. The power distribution system reliability analysis method according to claim 4, characterized in that: Step S26 includes the following steps: Step S261: Select key equipment from the preliminary feature set to obtain a key equipment list; Step S262: constructing a physical model according to the time series feature set and the key equipment list to obtain an equipment physical model set; Step S263: setting simulation parameters for the device physical model set according to the preliminary feature set to obtain a simulation parameter set; Step S264: performing physical simulation calculation on the device physical model set according to the simulation parameter set to obtain a simulation result set; Step S265: extracting correlation features from the simulation result set to obtain correlation features; merging correlation features from the correlation features and the preliminary feature set to obtain a correlation feature set.

7. The power distribution system reliability analysis method according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: constructing a power distribution system topology map according to the comprehensive operation data set to obtain a power distribution system topology map; Step S32: integrating the risk characteristic matrix into the distribution system topology diagram, constructing a reliability network model, and obtaining a reliability network diagram; Step S33: performing node importance evaluation on the reliability network diagram to obtain node importance data; performing node risk value calculation based on the node importance data and the reliability network diagram to obtain a node risk evaluation result; Step S34: classify weak links according to the node risk assessment results to obtain a graded risk assessment result; Step S35: Visualize the hierarchical risk assessment results into the power distribution system topology diagram, generate a reliability weak point map, and obtain a reliability weak point map.

8. The power distribution system reliability analysis method according to claim 7, characterized in that: Step S33 includes the following steps: Step S331: Initialize the node risk value of the reliability network graph to obtain the initial node risk value; Step S332: Calculate the node load weight according to the comprehensive operation data set to obtain the node load weight; Step S333: Calculate the node power supply range weight according to the reliability network diagram to obtain the node power supply range weight; Step S334: performing iterative calculation of the node risk value according to the initial node risk value, the reliability network diagram, the node load weight, and the node power supply range weight to obtain an iterative node risk value; Step S335: Output the node risk assessment result for the iterative node risk value to obtain the node risk assessment result.

9. The power distribution system reliability analysis method according to claim 1, characterized in that: Step S4 includes the following steps: Step S41: constructing an initial reliability assessment model according to the reliability weak point map and the comprehensive operation data set to obtain an initial reliability assessment model; Step S42: using the initial reliability evaluation model as a reinforcement learning agent to construct a reinforcement learning environment to obtain a reinforcement learning environment; Step S43: Perform reinforcement learning model training according to the reinforcement learning environment to obtain a trained reliability evaluation model; Step S44: performing model verification and optimization on the trained reliability assessment model to obtain an optimized reliability assessment model; Step S45: Acquire current system status data; use the optimized reliability evaluation model to perform a comprehensive reliability evaluation on the current system status data to obtain a comprehensive reliability score.

10. A distribution system reliability analysis system, characterized in that: Used to execute the power distribution system reliability analysis method as claimed in claim 1, the power distribution system reliability analysis system comprises: The multi-dimensional data fusion module is used to collect multi-dimensional data of the power distribution system through the sensor network, and to fuse the multi-dimensional data to obtain multi-dimensional related data; to construct a comprehensive operation data set for the multi-dimensional related data to obtain a comprehensive operation data set; The multi-scale feature extraction module is used to segment the time series data according to the comprehensive operation data set to obtain a fast fluctuation characteristic sequence and a slow-changing long-term trend sequence; perform time series decomposition on the fast fluctuation characteristic sequence and the slow-changing long-term trend sequence to obtain high-frequency fluctuation characteristic data and long-term trend characteristic data; perform feature component analysis on the high-frequency fluctuation characteristic data and the long-term trend characteristic data, and perform feature combination to obtain a multi-scale feature set; perform risk factor correlation analysis on the multi-scale feature set to obtain a correlation feature set; and construct a risk feature matrix on the correlation feature set to obtain a risk feature matrix; The risk factor association analysis module is used to construct a reliability network model based on the comprehensive operation data set and the risk characteristic matrix to obtain a reliability network diagram; to evaluate the node importance of the reliability network diagram to obtain node importance data; to calculate the node risk value based on the node importance data and the reliability network diagram to obtain the node risk assessment result; to classify the weak links based on the node risk assessment result, and to generate a reliability weak point map to obtain a reliability weak point map; The reliability network model building module is used to train the reliability assessment model according to the reliability weakness map and the comprehensive operation data set, and optimize the model to obtain the optimized reliability assessment model; the optimized reliability assessment model is used to perform a comprehensive reliability assessment to obtain a comprehensive reliability score.

Images