A power supply prediction method and system based on the service life of power transmission and transformation equipment

Abstract

The present invention provides a power supply prediction method and system based on the service life of power transmission and transformation equipment, which relates to the technical field of power supply prediction, including: first, obtaining the basic information and historical operation and maintenance data of the power transmission and transformation equipment, and constructing a multi-level characteristic map combining the characteristics of insulating materials, cooling systems, switch devices, etc. Then, a hybrid graph convolutional network combined with a random walk algorithm is used to extract equipment map features and generate a map data set. Based on the data set, a dynamic Bayesian network is used to establish an equipment aging model, an aging-power supply data set is generated, and a time series analysis is performed to obtain power supply cost data. Finally, a power supply forecast report is generated in combination with historical power supply data and customer needs. The present invention improves the accuracy of power supply prediction and the efficiency of system operation.

Description

Technical Field

[0001] The present invention relates to the technical field of power supply prediction, and in particular to a power supply prediction method and system based on the service life of power transmission and transformation equipment. Background Art

[0002] With the widespread application of modern power transmission and transformation equipment in power systems, the stability and reliability of power supply systems are increasingly dependent on the operating status of these devices. However, traditional power transmission and transformation equipment management methods usually rely on regular maintenance and equipment replacement strategies, and fail to fully consider the aging of equipment, structural fatigue, and changes in operating frequency during long-term operation. These traditional methods often find it difficult to accurately predict the remaining life and power supply capacity of equipment, leading to problems such as sudden equipment failures and power outages. In addition, power supply prediction methods in the prior art are mostly based on simple time series analysis or linear regression, which cannot fully consider the complex characteristics of the equipment and its impact on power supply. When faced with complex and changeable power supply demands and equipment aging processes, these methods often show large prediction errors, which in turn increases the cost and risk of power supply operations.

[0003] Although existing technologies attempt to solve the above problems by introducing equipment monitoring systems and simple life assessment models, they still have some shortcomings. For example, existing life assessment models often ignore the mutual influence between the multi-dimensional characteristics of power transmission and transformation equipment, as well as the nonlinear impact of key components on power supply capacity at different aging stages. At the same time, existing power supply prediction methods cannot effectively integrate equipment aging information and customer power supply demand data, resulting in low adaptability and accuracy in dealing with equipment aging and power supply demand fluctuations.

[0004] Based on the shortcomings of the above-mentioned existing methods, there is an urgent need for a power supply prediction method and system based on the service life of power transmission and transformation equipment. Summary of the invention

[0005] The purpose of the present invention is to provide a power supply prediction method and system based on the service life of power transmission and transformation equipment to improve the above problems. In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] In a first aspect, the present application provides a power supply prediction method based on the service life of power transmission and transformation equipment, comprising:

[0007] Obtaining basic information data sets and historical power supply data for each type of power transmission and transformation equipment, wherein the basic information includes the type, rated capacity, technical specifications, and expected service life of the equipment, and the historical power supply data includes power supply time data and power supply quantity data for the operating area of ​​each type of power transmission and transformation equipment;

[0008] Performing a graph construction process based on all of the basic information data sets, wherein a multi-level characteristic graph is constructed by combining the properties of the insulating material, the configuration of the cooling system, the type of the switch device and the corresponding operating frequency;

[0009] Based on a preset hybrid graph convolutional network, the connection relationship between different levels in the multi-level characteristic map is processed, and the implicit features in the map are extracted in combination with a random walk algorithm to obtain a map data set for each power transmission and transformation equipment;

[0010] Model building is performed based on all the atlas data sets, and a device aging model is established by analyzing the aging path of the device using a dynamic Bayesian network to quantify the comprehensive impact of material aging, structural fatigue, and operating frequency changes on the device life. An aging-power supply data set is generated based on the device aging model, and the aging-power supply data set includes aging rate data of each key component and power supply demand data of the customer;

[0011] Performing time series analysis and processing on the aging-power supply data set and the historical power supply data to obtain power supply cost data;

[0012] A forecast is made based on the power supply cost data in combination with the historical power supply data and the customer's power supply demand data to generate a power supply forecast report.

[0013] In a second aspect, the present application also provides a power supply prediction system based on the service life of power transmission and transformation equipment, including:

[0014] An acquisition module is used to acquire a basic information data set and historical power supply data for each power transmission and transformation equipment, wherein the basic information includes the type, rated capacity, technical specifications and expected service life of the equipment, and the historical power supply data includes power supply demand data for the operating area of ​​each power transmission and transformation equipment;

[0015] A construction module, used for performing a graph construction process according to all the basic information data sets, wherein a multi-level characteristic graph is constructed by combining the properties of the insulating material, the configuration of the cooling system, the type of the switch device and the corresponding operating frequency;

[0016] An aggregation module is used to process the connection relationship between different levels in the multi-level characteristic map based on a preset hybrid graph convolutional network, and extract implicit features in the map in combination with a random walk algorithm to obtain a map data set for each power transmission and transformation equipment;

[0017] A generation module is used to perform model building processing according to all the atlas data sets, analyze the aging path of the equipment by using a dynamic Bayesian network, quantify the comprehensive impact of material aging, structural fatigue and operating frequency changes on the life of the equipment, establish an equipment aging model, and generate an aging-power supply data set based on the equipment aging model, wherein the aging-power supply data set includes aging rate data of each key component and power supply demand data of the customer;

[0018] An analysis module, configured to perform time series analysis and processing based on the aging-power supply data set and the historical power supply data to obtain power supply cost data;

[0019] The prediction module is used to make predictions based on the power supply cost data in combination with the historical power supply data and the customer's power supply demand data to generate a power supply prediction report.

[0020] The beneficial effects of the present invention are:

[0021] The present invention obtains basic information data and historical power supply data of power transmission and transformation equipment, combines multi-level characteristic map construction, hybrid graph convolution network analysis and dynamic Bayesian network model, and comprehensively considers the impact of equipment aging path, operating frequency and material properties on equipment life and power supply capacity. Through multi-dimensional timing analysis, this method can not only accurately quantify the aging rate of key components, but also make accurate power supply predictions based on the customer's power supply needs, thereby greatly improving the accuracy of power supply cost predictions and reducing the risks of power outages and equipment failures. Compared with the prior art, the present invention overcomes the shortcomings of insufficient consideration of complex equipment characteristics and low prediction accuracy in traditional methods through innovative graph neural networks and dynamic Bayesian network modeling methods.

[0022] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or be understood by practicing the embodiments of the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A schematic flow chart of a power supply prediction method based on the service life of power transmission and transformation equipment according to an embodiment of the present invention;

[0025] Figure 2 It is a schematic diagram of the structure of a power supply prediction system based on the service life of power transmission and transformation equipment described in an embodiment of the present invention.

[0026] Markings in the figure: 1. Acquisition module; 2. Construction module; 21. First analysis unit; 22. Second analysis unit; 23. Third analysis unit; 24. First construction unit; 3. Aggregation module; 31. First processing unit; 32. First fusion unit; 33. Second processing unit; 34. First integration unit; 4. Generation module; 41. First extraction unit; 42. Third processing unit; 43. First calibration unit; 44. Fourth processing unit; 5. Analysis module; 51. Second extraction unit; 52. First classification unit; 53. Second construction unit; 54. First output unit; 6. Prediction module; 61. First prediction unit; 62. Second prediction unit; 63. Third prediction unit; 64. Fourth prediction unit. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0029] Embodiment 1: This embodiment provides a power supply prediction method based on the service life of power transmission and transformation equipment.

[0030] See also Figure 1 , the figure shows that the method includes steps S100 to S600.

[0031] Step S100, obtaining a basic information data set and historical power supply data of each power transmission and transformation equipment, wherein the basic information includes the type, rated capacity, technical specifications and expected service life of the equipment, and the historical power supply data includes power supply time data and power supply amount data of the operating area of ​​each power transmission and transformation equipment;

[0032] It is understandable that the basic information data set includes key parameters such as the type, rated capacity, technical specifications and expected service life of the equipment. These data are a direct reflection of the performance and applicability of the equipment and determine the adaptability and performance of the equipment in a specific application scenario. In addition, the historical power supply data covers the equipment's failure frequency, repair records and power supply maintenance plans. These data reflect the reliability and power supply requirements of the equipment in actual operation and are the basis for evaluating the health of the equipment and predicting future power supply requirements.

[0033] Step S200, performing a graph construction process according to all the basic information data sets, wherein a multi-level characteristic graph is constructed by combining the properties of the insulating material, the configuration of the cooling system, the type of the switch device and the corresponding operating frequency;

[0034] It is understandable that the multi-level characteristic map allows the diversity and complexity of equipment characteristics to be fully and accurately expressed, providing a rich information basis for subsequent analysis. Secondly, the multi-level characteristic map can capture the performance differences of the equipment under different operating conditions, which helps to understand the various situations that may occur in the actual use of the equipment. Finally, through the construction of this map, subsequent model analysis and optimization can be based on more detailed and comprehensive data, thereby improving the accuracy of full life cycle cost prediction and optimization.

[0035] Step S300: Processing the connection relationship between different levels in the multi-level characteristic map based on a preset hybrid graph convolutional network, and extracting implicit features in the map in combination with a random walk algorithm to obtain a map data set for each power transmission and transformation equipment;

[0036] It is understandable that the hybrid graph convolutional network can deeply explore the potential features in the multi-level feature map and improve the richness and accuracy of feature extraction. The introduction of the random walk algorithm helps to enhance the depth and breadth of the graph analysis, capture complex global features, and the extracted implicit features provide a solid data foundation for the subsequent aging model establishment and cost prediction, making the prediction results more accurate and comprehensive.

[0037] Step S400: Perform model building processing according to all the atlas data sets, analyze the aging path of the equipment by using a dynamic Bayesian network, quantify the comprehensive impact of material aging, structural fatigue and operating frequency changes on the equipment life, establish an equipment aging model, and generate an aging-power supply data set based on the equipment aging model, wherein the aging-power supply data set includes aging rate data of each key component and power supply demand data of the customer;

[0038] It can be understood that the dynamic Bayesian network can dynamically reflect the changes in the equipment status and provide accurate aging path analysis. Secondly, the causal analysis characteristics of the dynamic Bayesian network help to identify and understand the complex relationships between different aging factors.

[0039] Step S500: performing time series analysis and processing according to the aging-power supply data set and the historical power supply data to obtain power supply cost data;

[0040] Specifically, the aging-power supply data set contains the aging rate of key components and the corresponding power supply demand data. This information is obtained through the previous model construction and aging path analysis, reflecting the health status and expected power supply requirements of the equipment at different stages of use. The historical power supply data records the actual power supply frequency, power supply records and executed power supply activities of the equipment. These data comprehensively reflect the actual operation and power supply behavior of the equipment during its life cycle. When performing time series analysis, stationary time series models or non-stationary time series models can be applied to analyze trends and seasonal changes in the data. Specific methods include but are not limited to weighted average processing of equipment aging rate, fluctuation analysis of power supply frequency, and regression analysis of historical data. Through these methods, the periodicity and seasonal changes in equipment power supply demand can be identified, thereby accurately predicting future power supply costs.

[0041] Step S600: Perform prediction based on the power supply cost data in combination with the historical power supply data and the customer's power supply demand data to generate a power supply prediction report.

[0042] Preferably, in the specific forecasting process, a variety of statistical and machine learning methods can be used, such as regression analysis, decision tree models, or cost allocation models based on historical data. These methods can help identify the main drivers of various costs and analyze their changing trends at different stages of use. This step can provide enterprises with a comprehensive power supply cost estimate, covering the entire power supply cycle from equipment procurement to final retirement. This forecast helps enterprises conduct long-term financial planning and budgeting, thereby reducing cost risks and improving overall benefits in future operations.

[0043] Further, step S200 includes step S210 to step S240.

[0044] Step S210: performing material property cluster analysis according to the basic information data set, clustering the chemical properties, electrical properties, withstand voltage grades and thermal stability of different insulating materials by using a K-means algorithm to obtain material clusters with similar properties, and generating a material property cluster data set;

[0045] Specifically, this step clusters the chemical properties, electrical properties, withstand voltage grades and thermal stability of the insulating materials, with the aim of meaningfully grouping the materials so that these material properties can be better understood and applied in subsequent analysis. Further, in order to optimize the number of clusters K in the K-means algorithm, this embodiment adopts a comprehensive method of the silhouette coefficient and the elbow rule. The silhouette coefficient is used to evaluate the quality of the clustering results and help select the K value that maximizes the similarity of the insulating material within its cluster and minimizes the similarity with other clusters. For power transmission and transformation equipment, there are large differences in the characteristics of insulating materials. The silhouette coefficient can effectively evaluate the clustering effect of these characteristics in practical applications and ensure the internal consistency of each material cluster. The elbow rule selects the optimal K value by drawing the total clustering error sum of squares curve. In power transmission and transformation equipment, different insulating materials have similar but not completely identical characteristics, so it is necessary to accurately select K to avoid over-clustering or under-clustering. The elbow rule helps determine the point where the total clustering error sum of squares curve slows down, that is, the optimal K value, thereby balancing the accuracy and complexity of clustering. In addition, the present application introduces a density-based spatial clustering algorithm into the K-means algorithm for comparison. This density-based clustering technology is particularly suitable for processing abnormal data in power transmission and transformation equipment, such as occasional material property anomalies or noise data. The density-based spatial clustering algorithm does not require a preset K value, but identifies dense areas of data by setting density thresholds (such as the minimum number of points and the neighborhood radius), so that abnormal points and noise in the material data can be discovered and processed. This can significantly improve the robustness and accuracy of clustering when processing the insulation material data of actual power transmission and transformation equipment. Through the comprehensive application of the silhouette coefficient, the elbow rule and the density-based spatial clustering algorithm, this step not only optimizes the clustering effect of the insulating material, but also improves the ability to handle abnormal data, thereby providing more accurate and reliable material property data for the full life cycle cost prediction of power transmission and transformation equipment.

[0046] Step S220: performing cooling system configuration mode analysis based on the basic information data set, extracting the medium type, flow control mode and heat exchanger type configuration features of the cooling system by principal component analysis, and performing dimensionality reduction and feature selection processing to obtain a cooling system feature vector set;

[0047] It should be noted that principal component analysis, as a linear dimensionality reduction technique, has two main meanings in this step. First, principal component analysis can identify and retain the most explanatory principal components in the cooling system characteristics. These principal components represent the main direction of change in the data, so that high-dimensional data can be mapped to low-dimensional space while retaining the information of the original data as much as possible. This provides a simplified and effective analysis tool for complex configuration parameters in the cooling system, especially when comparing different devices. Secondly, principal component analysis helps feature selection. By analyzing the variance contribution rate of each principal component, it can be determined which features have the greatest impact on the cooling system configuration mode. This step can effectively remove redundant information and noise, thereby obtaining a clearer and more streamlined feature vector set. In the cooling system of power transmission and transformation equipment, the thermal conductivity of a specific medium, the accuracy of flow control, and the efficiency of the heat exchanger are key influencing factors. Principal component analysis can be used to discover the potential correlation and interaction between these features. In the specific implementation, the application of principal component analysis is not only for dimensionality reduction, but also for understanding the physical meaning of different configurations in the cooling system. For example, the use of different media may correspond to different thermal management strategies, different flow control modes may affect the response speed and stability of the equipment, and the design of the heat exchanger is directly related to the heat dissipation efficiency of the equipment. Through these analyses, the final cooling system feature vector set is not only an important indicator of equipment performance, but also an important input for the subsequent power supply cost analysis of power transmission and transformation equipment.

[0048] Step S230: performing association analysis based on the basic information data set, mining potential patterns of high-frequency operations and specific types of switch devices using a frequent pattern mining algorithm, and analyzing the matching degree between the design load and actual usage of the switch device to obtain a switch device characteristic data set;

[0049] Specifically, the original data related to the switchgear is first extracted from the basic information data set, including its type, design load, operating frequency, etc. Among these data, the operating frequency refers to the number of times the switchgear is turned on and off in a specific time period, while the design load refers to the maximum current or voltage that the switchgear can withstand when it is designed. Through the frequent pattern mining algorithm, it can be revealed which types of switchgear are frequently operated in actual use, and the relationship between these operating frequencies and their design loads. In frequent pattern mining, the algorithm first identifies frequently occurring operating patterns, such as the number of times a certain switchgear is repeatedly operated in a specific time period exceeding a preset frequency threshold. Then, by analyzing these frequent patterns, the load conditions of different switchgears in actual use can be further understood. In particular, by comparing the differences between actual use and design load, switchgears that may be at risk of overload or overuse can be identified. This has important guiding significance for the power supply and safe operation of equipment. In addition, association analysis can also help identify whether a specific operating mode of a switchgear has potential associations with other equipment characteristics (such as material type, cooling system configuration, etc.). This analysis can provide a basis for the subsequent power supply strategy formulation, ensuring that the equipment can reasonably distribute the load in actual operation and avoid failures caused by overload. Through these analyses, the final switch device characteristic data set not only contains the basic information of various switch devices, but also includes the matching data between their operating frequency in actual operation and the designed load.

[0050] Step S240: Graph construction is performed based on the material property cluster data set, cooling system feature vector set and switch device characteristic data set. Graph neural networks are used to aggregate node features and learn multi-relationship edge weights. The relationship strength and influence between different characteristics are quantified to construct a multi-level characteristic graph that reflects the comprehensive characteristics of the power transmission and transformation equipment.

[0051] It should be noted that the constructed multi-level characteristic map includes multiple levels, each of which represents the key characteristics of different aspects of power transmission and transformation equipment. These levels include material characteristic layer, cooling system configuration layer, switch device characteristic layer and its related operating frequency, etc. These levels fully reflect the comprehensive characteristics of power transmission and transformation equipment through complex relationships and interactions. The material characteristic layer contains the characteristics of various insulating materials used in the equipment, such as chemical properties, electrical properties, withstand voltage level and thermal stability, etc. These characteristics directly affect the electrical performance and safety of the equipment. For example, the thermal stability of insulating materials under high temperature environment is an important factor in determining the long-term operation reliability of the equipment. The cooling system configuration layer involves various configuration characteristics of the cooling system in the equipment, such as medium type, flow control mode and heat exchanger type, etc. These characteristics directly affect the temperature management and heat dissipation efficiency of the equipment. The performance of the cooling system is crucial to ensure the stable operation of the equipment under high load conditions. Therefore, its configuration characteristics are closely related to the material characteristic layer, such as the impact of a certain cooling medium on the performance of a specific insulating material. The switchgear characteristic layer covers various characteristics of the switchgear in the equipment, including design load, actual usage, operating frequency, etc. These characteristics not only affect its own reliability and life, but also have an important impact on the stability of the entire system. By analyzing the operation records and design parameters of the equipment, the overload or frequent operation conditions that the equipment may encounter in actual operation are revealed. Between these layers, the graph connects and integrates the information of each layer through complex relationships and interactions. Multi-relation edge weight learning is used to quantify and represent the relationship strength between different characteristics. For example, the relationship between the cooling system configuration layer and the material characteristic layer can be reflected in the long-term aging effect of the cooling medium on the material, and the relationship between the switchgear characteristic layer and the cooling system configuration layer may be reflected in the effect of certain cooling systems on the heat dissipation efficiency of the switchgear. These relationships are captured and quantified through node feature aggregation and multi-relation edge weight learning of graph neural networks. Graph neural networks can not only aggregate the characteristics of nodes at each level, but also capture and understand the interaction effects between these characteristics across levels. For example, the nodes of the material characteristic layer obtain more information about the overall performance of the equipment through the edges with the cooling system configuration layer and the switchgear characteristic layer. This information interaction can reveal the complex interaction between material selection, cooling system configuration and operation mode. Through this multi-level characteristic map, equipment purchasers can have a more comprehensive understanding of the various key characteristics of the equipment and their interrelationships, thereby making more accurate and reasonable judgments in the power supply cost forecast of the equipment.

[0052] Further, step S300 includes step S310 to step S340.

[0053] Step S310: performing graph convolution processing according to the node features and edge connection relationships of the multi-level feature map, transferring and aggregating feature information between different layers by applying a hybrid graph convolutional network, taking the neighbor features of the node as input, and performing convolution operations in combination with a specific weight matrix to capture the interactive relationship between different features, thereby obtaining a node feature matrix after convolution;

[0054] Specifically, the neighbor features of each node are used as input, and a convolution operation is performed through a specific weight matrix to generate a new feature representation. The weight matrix is ​​learned during the network training process and is used to optimally combine and transmit feature information. For the evaluation of power transmission and transformation equipment in this embodiment, this graph convolution processing can effectively capture the complex interactions between different levels such as material properties, cooling system configuration, and switch device characteristics. For example, how the characteristics of a certain insulating material affect the thermal management of the equipment together with the configuration of the cooling system, and how these characteristics affect the operating frequency and reliability of the switch device. Through the convolution operation, the potential correlation between these multi-dimensional characteristics can be identified, especially in the complex interactions between cross-level characteristics, such as the impact of material aging on the efficiency of the cooling system, or how the frequent operation of the switch device accelerates the degradation of the material. This method is beneficial to understand the performance of the equipment in the actual operating environment because it considers the global interaction between multi-level characteristics, rather than just local, independent characteristic performance. Ultimately, the result of the graph convolution processing is a convolved node feature matrix, which contains the new feature representation of each node (i.e., each key feature of the equipment) after aggregating the neighbor features. These representations not only integrate the node’s own information, but also include the information of its neighboring nodes, providing a feature representation from a global perspective.

[0055] Step S320: performing multi-scale feature fusion processing according to the node feature matrix, by introducing a multi-scale feature fusion mechanism in the hybrid graph convolutional network, weightedly combining the features output by different layers to integrate information of different scales, and obtaining a fused first feature matrix;

[0056] In this process, the node feature matrix output by each layer is first obtained, and then these matrices are weighted and combined to form a fused feature representation. This weighted combination is usually achieved by learning specific weight coefficients that reflect the importance of different layers in the final feature representation. This method can effectively integrate information from different scales to generate a more comprehensive and richer feature representation. In particular, in the life cycle cost prediction scenario of power transmission and transformation equipment, multi-scale feature fusion processing can combine different levels of information such as material properties, cooling system configuration, and switchgear type, which have different effects on equipment performance and life at different scales. For example, shallower features may highlight the performance of a certain insulating material under specific operating conditions, while deeper features may reveal the complex relationship between material aging and operating frequency in the entire system. Through multi-scale feature fusion, these different levels of information are integrated into a unified feature space, so that the feature representation has not only the accuracy of local details, but also contextual understanding.

[0057] Step S330, performing random walk processing according to the feature matrix, simulating multiple random paths on the graph and capturing long-distance dependencies and structural information in the graph, to obtain a second feature matrix containing random walk features;

[0058] It can be understood that the random walk process selects the node to visit next in a probabilistic manner by simulating random paths between nodes on the graph multiple times. This method can capture not only the relationship between directly connected nodes, but also the connection between nodes on longer paths, thereby identifying structural information in the graph that is difficult to capture through simple proximity relationships. Ultimately, the second feature matrix generated by the random walk process includes richer and more comprehensive feature information that integrates the local and global structures in the graph, which can not only reveal the complex interactions between the components of the device, but also better characterize the state of the entire system.

[0059] Step S340: perform feature integration processing according to the second feature matrix, and perform dimension reduction and standardization processing on the integrated features to obtain a graph data set for each power transmission and transformation equipment.

[0060] Further, step S400 includes step S410 to step S440.

[0061] Step S410: performing feature extraction processing according to the atlas data set, and obtaining an aging feature set by screening out representative features that affect device aging from the atlas data set;

[0062] It should be noted that the feature extraction process first needs to define the key factors that affect equipment aging, including material aging characteristics, structural fatigue characteristics, operating frequency characteristics, etc. Preferably, statistical analysis and machine learning methods are used to identify these key features. For example, a feature importance evaluation method based on a tree model, such as a random forest or a gradient boosting tree, is used to screen out important features by evaluating the contribution of each feature in predicting the target variable (such as the degree of equipment aging).

[0063] Step S420: Performing structural learning processing according to the aging feature set, mining conditional dependencies between features in time series data by applying a dynamic Bayesian network and establishing a model structure of the equipment aging path, thereby obtaining an aging model structure diagram;

[0064] It should be noted that in equipment aging analysis, different aging characteristics change over time and are interdependent. The dynamic Bayesian network can capture the conditional dependencies in these time series to build a model structure that describes the equipment aging path. The specific steps are as follows:

[0065] Step S421: Determine the nodes in the model according to the variables in the aging feature set.

[0066] In the dynamic Bayesian network, each node represents a random variable, which indicates a feature in the aging feature set (e.g., material aging rate, usage frequency, environmental conditions, etc.). Assume that the feature set in the aging feature set is , each feature can take different discrete states.

[0067] Step S422: Use the time series in the historical data to preliminarily model the dynamic Bayesian network, and estimate the parameters of the conditional dependency relationship between the nodes at each time step.

[0068] A dynamic Bayesian network consists of a series of time slices, each of which represents the state of the system at a specific time step. Assume that the time slices start from arrive , for each time ,feature The state is represented by . Initial time step The joint probability distribution of is:

[0069] ;

[0070] in, Represents all node time steps The joint state of Indicates the number of feature variables; represents conditional probability; Indicates that at the initial time step When each characteristic variable The specific status.

[0071] Step S423: Determine the directed edges between nodes in the model through structural learning to obtain the causal relationship between features.

[0072] In a dynamic Bayesian network, conditional dependencies between nodes are represented by directed edges. The direction of the edge reflects the causal relationship. Assume that at time step ,feature and If there is a conditional dependency relationship between them, the conditional probability is:

[0073] ;

[0074] in, Represents feature variables At time step The parent node set of represents conditional probability; Indicated in Time step feature variable Status; Indicated in Time step feature variable Status; and Indicates the sequence number of the feature variable.

[0075] Step S424: adjust and verify the model parameters, use historical data for cross-validation, and evaluate the prediction performance and stability of the model.

[0076] For each edge in the DBN, the corresponding conditional probability distribution needs to be estimated. This is achieved through maximum likelihood estimation. Given a training data set, the conditional probability of each feature can be expressed as:

[0077] ;

[0078] in, Indicates that the condition is met in the training set , , The number of samples; Indicates that the condition is met in the training set , The number of samples; represents conditional probability; Indicates data points; Indicates data points.

[0079] Step S425: Use the expectation maximization algorithm to perform structure learning to obtain a network structure that maximizes the likelihood function. The formula is as follows:

[0080] ;

[0081] in, represents the optimal conditional probability parameter, Represents the network structure, Represents a set of parameters. Represents a dataset containing multiple samples.

[0082] The final joint probability distribution is expressed as:

[0083] ;

[0084] in, represents the total number of time steps; Indicates time slice from arrive The set of feature variables.

[0085] Step S430: performing parameter learning and model calibration processing according to the aging model structure diagram, estimating parameter values ​​of each conditional probability distribution in the equipment aging model by combining historical data, and obtaining a calibration model;

[0086] This step minimizes the difference between the model's predictions and actual observed data by repeatedly adjusting parameters, thereby improving the accuracy and stability of the model.

[0087] Step S440: quantify the impact of aging according to the calibration model, simulate the time series prediction in the aging model, quantify the aging rate of each key component and the preset power supply demand of the customer, and integrate to obtain the aging-power supply data set.

[0088] Specifically, the calibrated dynamic Bayesian network model is first used to predict the time series of power transmission and transformation equipment, including simulating the state evolution of the equipment at future time points through the model, especially predicting the aging rate of each key component. The model infers the aging state of the equipment at different time points through the aging path and conditional probability distribution of each component. Then, the predicted aging state is analyzed to quantify the aging rate of each key component. The aging rate reflects the deterioration rate of the component in a specific time period, which provides a basis for accurately evaluating the remaining life of the equipment. At the same time, according to the aging degree of different components and their impact on the stability and performance of the overall system, the corresponding power supply demand is determined. In this process, the actual working conditions and historical fault data of each component need to be taken into account to ensure the accuracy of the prediction results. Finally, the aging rate data of each key component and the corresponding power supply demand are integrated to form an aging-power supply data set. This data set records in detail the aging process of the equipment throughout its life cycle and the corresponding power supply strategy.

[0089] Further, step S500 includes step S510 to step S540.

[0090] Step S510: performing time series feature extraction processing according to the aging-power supply data set, analyzing the time dependence of the aging rate data by using an autoregressive integrated moving average model, extracting trend, seasonality and residual time series features, and obtaining time series data;

[0091] In this step, by applying the autoregressive integrated moving average model to the aging rate data, three characteristics of the time series can be extracted: trend, seasonality, and residual. The trend part reflects the long-term upward or downward trend of the data, the seasonal part captures the cyclical fluctuations of the data, and the residual part is the random error after the trend and seasonality are removed. These characteristics help understand the long-term change law and cyclical behavior of the equipment aging process, and provide an important basis for further analysis and prediction.

[0092] Step S520, performing classification processing according to the time series data, and obtaining a classification result by dividing the time series data with similar aging patterns into at least two groups;

[0093] Specifically, each time series data is converted into a vector representation, where each dimension of the vector corresponds to a feature of the time series or a numerical value after feature extraction. Then, for each pair of time series data, the cosine similarity is calculated. Based on the calculated cosine similarity, the time series data is divided into different groups or categories, and data with higher similarity are classified into the same category.

[0094] Step S530: construct a model according to the classification result, quantify the contribution of different features to the power supply cost by using a Bayesian linear regression algorithm, and generate an estimated value of the power supply cost under each category to obtain a power supply cost model;

[0095] It should be noted that the Bayesian linear regression algorithm can effectively handle the dependencies between features and can estimate model parameters through appropriate prior information, which is suitable for processing multiple features and complex data distribution. In this process, it is necessary to quantify the impact of different features on the power supply cost and generate power supply cost estimates for each category. Ultimately, the power supply cost model obtained through these steps can provide the ability to predict the power supply cost for different categories of equipment.

[0096] Step S540: output a standardized power supply cost data set based on the power supply cost model.

[0097] This step uses the electricity cost model that has been constructed and calibrated to standardize the electricity cost estimates for each category or group. Standardization involves converting the data into a specific scale or distribution to facilitate subsequent analysis and comparison. This process ensures that the electricity cost data for different categories are comparable and consistent, making further data analysis and decision-making processes more accurate and effective.

[0098] Further, step S600 includes step S610 to step S640.

[0099] Step S610: Input the historical power supply data and power supply cost data as input information into the BP neural network model to perform power supply prediction for each customer, and obtain the power supply prediction result for each customer;

[0100] It can be understood that this step captures the implicit relationship between the historical power supply mode and power supply cost of each customer in the power supply system through the self-learning ability of the neural network. In this step, the historical power supply data can reflect the power consumption pattern of each customer, while the power supply cost data includes the impact of factors such as equipment maintenance and energy consumption on power supply. By inputting these two sets of data into the BP neural network model, the model can gradually learn the complex nonlinear relationship between these factors and the customer's power supply needs. The BP neural network uses a feedforward method to pass the input information to the hidden layer, then extracts features through the activation function, and finally outputs the customer's power supply prediction results. In this process, the multi-layer structure of the BP neural network can effectively capture the nonlinear mapping relationship of the input features, thereby improving the prediction accuracy.

[0101] Step S620: Compare the power supply forecast of each customer with the power supply demand data of the customer to obtain a comparison difference;

[0102] It can be understood that this step provides a basis for subsequent model optimization. By analyzing the difference between power supply prediction and actual demand, the prediction error of the BP neural network can be quantified, and this difference will be used for the next step of gradient descent algorithm optimization.

[0103] Step S630, adjusting the weight and bias information of the BP neural network based on the gradient descent algorithm and the contrast difference until the contrast difference is less than a preset threshold, thereby obtaining a trained BP neural network model;

[0104] It is understandable that the gradient descent algorithm calculates the gradient of the loss function relative to the parameters of the neural network and gradually adjusts the weights and biases according to these gradients to minimize the prediction error. In this process, the weights and biases updated after each iteration will make the BP neural network closer to the optimal solution until the comparison difference is less than the preset threshold. This weight update mechanism based on error feedback is the core of BP neural network training, enabling it to adapt to changes in power supply demand of different customers and improve prediction accuracy.

[0105] Step S640: Summarize the power supply forecast results for each customer to obtain a power supply forecast report for all customers.

[0106] It is understandable that the BP neural network can make efficient predictions in large-scale and diverse customer power demand data through adaptive learning and continuous optimization, and then plan the power supply for each customer through power supply prediction to ensure the reliability and economy of the power supply system. At the same time, the introduction of the gradient descent algorithm further ensures the accuracy of the model, making the power supply prediction more in line with actual needs, avoiding power surplus or shortage caused by prediction deviation, and thus optimizing the operation of the overall power distribution system.

[0107] Embodiment 2, as Figure 2 As shown, this embodiment provides a power supply prediction system based on the service life of power transmission and transformation equipment, and the system includes:

[0108] Acquisition module 1, acquires basic information data set and historical power supply data of each power transmission and transformation equipment, wherein the basic information includes the type, rated capacity, technical specifications and expected service life of the equipment, and the historical power supply data includes power supply time data and power supply amount data of the operating area of ​​each power transmission and transformation equipment;

[0109] Construction module 2, used for performing graph construction processing according to all the basic information data sets, wherein a multi-level characteristic graph is constructed by combining the properties of the insulating material, the configuration of the cooling system, the type of the switch device and the corresponding operating frequency;

[0110] Aggregation module 3, based on a preset hybrid graph convolutional network, processes the connection relationship between different levels in the multi-level characteristic graph, and extracts implicit features in the graph in combination with a random walk algorithm to obtain a graph data set for each power transmission and transformation equipment;

[0111] A generation module 4 is used to perform model building processing according to all the atlas data sets, analyze the aging path of the equipment by using a dynamic Bayesian network, quantify the comprehensive impact of material aging, structural fatigue and operating frequency changes on the equipment life, establish an equipment aging model, and generate an aging-power supply data set based on the equipment aging model, wherein the aging-power supply data set includes aging rate data of each key component and power supply demand data of the customer;

[0112] An analysis module 5, configured to perform time series analysis processing according to the aging-power supply data set and the historical power supply data to obtain power supply cost data;

[0113] The prediction module 6 is used to make predictions based on the power supply cost data in combination with the historical power supply data and the customer's power supply demand data to generate a power supply prediction report.

[0114] In a specific embodiment disclosed in the present application, building block 2 includes:

[0115] A first analysis unit 21 is used to perform material property cluster analysis based on the basic information data set, cluster the chemical properties, electrical properties, withstand voltage grades and thermal stability of different insulating materials by using a K-means algorithm to obtain material clusters with similar properties, and generate a material property cluster data set;

[0116] A second analysis unit 22 is used to perform cooling system configuration mode analysis based on the basic information data set, extract the medium type, flow control mode and heat exchanger type configuration features of the cooling system by using principal component analysis, and perform dimension reduction and feature selection processing to obtain a cooling system feature vector set;

[0117] The third analysis unit 23 is used to perform association analysis based on the basic information data set, mine potential patterns of high-frequency operations and specific types of switch devices using a frequent pattern mining algorithm, and analyze the matching degree between the design load and actual usage of the switch device to obtain a switch device characteristic data set;

[0118] The first construction unit 24 is used to perform graph construction processing based on the material property cluster data set, the cooling system feature vector set and the switch device characteristic data set, and to perform node feature aggregation and multi-relationship edge weight learning by using a graph neural network, and to quantify the relationship strength and influence between different characteristics, so as to construct a multi-level characteristic graph reflecting the comprehensive characteristics of the power transmission and transformation equipment.

[0119] In a specific implementation disclosed in the present application, the aggregation module 3 includes:

[0120] The first processing unit 31 is used to perform graph convolution processing according to the node features and edge connection relationships of the multi-level feature map, by applying a hybrid graph convolution network to transfer and aggregate feature information between different layers, taking the neighbor features of the node as input, and performing convolution operations in combination with a specific weight matrix to capture the interactive relationship between different features, thereby obtaining a node feature matrix after convolution;

[0121] A first fusion unit 32 is used to perform multi-scale feature fusion processing according to the node feature matrix, by introducing a multi-scale feature fusion mechanism in the hybrid graph convolutional network, weightedly combining the features output by different layers to integrate information of different scales, and obtaining a fused first feature matrix;

[0122] A second processing unit 33 is used to perform random walk processing according to the feature matrix, and obtain a second feature matrix containing random walk features by simulating multiple random paths on the graph and capturing long-distance dependencies and structural information in the graph;

[0123] The first integration unit 34 is used to perform feature integration processing according to the second feature matrix, and perform dimension reduction and standardization processing on the integrated features to obtain a graph data set for each power transmission and transformation equipment.

[0124] In a specific implementation disclosed in the present application, the generation module 4 includes:

[0125] A first extraction unit 41 is used to perform feature extraction processing according to the atlas data set, and obtain an aging feature set by screening out representative features that affect device aging from the atlas data set;

[0126] The third processing unit 42 is used to perform structural learning processing according to the aging feature set, and to mine the conditional dependency relationship between features in the time series data by applying a dynamic Bayesian network and to establish a model structure of the equipment aging path, so as to obtain an aging model structure diagram;

[0127] A first calibration unit 43, configured to perform parameter learning and model calibration processing according to the aging model structure diagram, and obtain a calibration model by estimating parameter values ​​of each conditional probability distribution in the device aging model in combination with historical data;

[0128] The fourth processing unit 44 is used to perform aging impact quantification processing according to the calibration model, by simulating the time series prediction in the aging model, and quantifying the aging rate of each key component and the preset customer power supply demand, and integrating to obtain the aging-power supply data set.

[0129] In a specific embodiment disclosed in the present application, the analysis module 5 includes:

[0130] A second extraction unit 51 is used to perform time series feature extraction processing according to the aging-power supply data set, and extract trend, seasonality and residual time series features by using an autoregressive integrated moving average model to obtain time series data;

[0131] A first classification unit 52 is used to perform classification processing according to the time series data, and obtain a classification result by dividing the time series data with similar aging patterns into at least two groups;

[0132] A second construction unit 53 is used to perform model construction processing according to the classification result, quantify the contribution of different features to the power supply cost by using a Bayesian linear regression algorithm, and generate an estimated value of the power supply cost under each category to obtain a power supply cost model;

[0133] The first output unit 54 outputs a standardized power supply cost data set based on the power supply cost model.

[0134] In a specific implementation disclosed in the present application, the prediction module 6 includes:

[0135] The first prediction unit 61 is used to input the historical power supply data and the power supply cost data as input information into the BP neural network model to perform power supply prediction for each customer, and obtain the power supply prediction result for each customer;

[0136] The second prediction unit 62 is used to compare the power supply prediction of each customer with the power supply demand data of the customer to obtain a comparison difference;

[0137] A third prediction unit 63, used to adjust the weight and bias information of the BP neural network based on the gradient descent algorithm and the contrast difference, until the contrast difference is less than a preset threshold, to obtain a trained BP neural network model;

[0138] The fourth prediction unit 64 is used to summarize the power supply prediction results of each customer to obtain a power supply prediction report for all customers.

[0139] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A power supply prediction method based on the service life of power transmission and transformation equipment, characterized in that: include: Obtaining a basic information data set and historical power supply data for each type of power transmission and transformation equipment, wherein the basic information includes the type, rated capacity, technical specifications and expected service life of the equipment, and the historical power supply data includes power supply time data and power supply quantity data for the operating area of ​​each type of power transmission and transformation equipment; Performing a graph construction process based on all of the basic information data sets, wherein a multi-level characteristic graph is constructed by combining the properties of the insulating material, the configuration of the cooling system, the type of the switch device and the corresponding operating frequency; Based on a preset hybrid graph convolutional network, the connection relationship between different levels in the multi-level characteristic map is processed, and the implicit features in the map are extracted in combination with a random walk algorithm to obtain a map data set for each power transmission and transformation equipment; Model building is performed based on all the atlas data sets, and a device aging model is established by analyzing the aging path of the device using a dynamic Bayesian network to quantify the comprehensive impact of material aging, structural fatigue, and operating frequency changes on the device life. An aging-power supply data set is generated based on the device aging model, and the aging-power supply data set includes aging rate data of each key component and power supply demand data of the customer; Performing time series analysis and processing on the aging-power supply data set and the historical power supply data to obtain power supply cost data; According to the power supply cost data, combined with the historical power supply data and the customer's power supply demand data, a forecast is made to generate a power supply forecast report; The graph construction process is performed based on all the basic information data sets, including: Performing material property clustering analysis based on the basic information data set, clustering the chemical properties, electrical properties, withstand voltage grades and thermal stability of different insulating materials by using a K-means algorithm to obtain material clusters with similar properties, and generating a material property cluster data set; Performing cooling system configuration mode analysis based on the basic information data set, extracting the medium type, flow control mode and heat exchanger type configuration features of the cooling system by principal component analysis, and performing dimension reduction and feature selection processing to obtain a cooling system feature vector set; Performing association analysis based on the basic information data set, mining potential patterns of high-frequency operations and specific types of switch devices using a frequent pattern mining algorithm, and analyzing the matching degree between the design load and actual usage of the switch device to obtain a switch device characteristic data set; A graph is constructed based on the material property cluster dataset, cooling system feature vector set and switch device property dataset. Node feature aggregation and multi-relationship edge weight learning are performed using a graph neural network. The relationship strength and influence between different characteristics are quantified to construct a multi-level characteristic graph that reflects the comprehensive characteristics of the power transmission and transformation equipment.

2. The power supply prediction method based on the service life of power transmission and transformation equipment according to claim 1 is characterized in that: Based on the preset hybrid graph convolutional network, the connection relationship between different levels in the multi-level characteristic map is processed, and the implicit features in the map are extracted in combination with the random walk algorithm to obtain the map data set of each power transmission and transformation equipment, including: Performing graph convolution processing according to the node features and edge connection relationships of the multi-level feature graph, transferring and aggregating feature information between different layers by applying a hybrid graph convolutional network, taking the neighbor features of the node as input, and performing convolution operations in combination with a specific weight matrix to capture the interactive relationship between different features, thereby obtaining a node feature matrix after convolution; Performing multi-scale feature fusion processing according to the node feature matrix, by introducing a multi-scale feature fusion mechanism in the hybrid graph convolutional network, weightedly combining the features output by different layers to integrate information of different scales, and obtaining a fused first feature matrix; Performing random walk processing according to the fused first feature matrix, simulating multiple random paths on the graph and capturing long-distance dependencies and structural information in the graph, thereby obtaining a second feature matrix containing random walk features; Feature integration processing is performed according to the second feature matrix, and the integrated features are subjected to dimensionality reduction and standardization processing to obtain a graph data set for each power transmission and transformation equipment.

3. The power supply prediction method based on the service life of power transmission and transformation equipment according to claim 1 is characterized in that: Model building is performed based on all of the atlas data sets. The aging path of the equipment is analyzed by using a dynamic Bayesian network to quantify the comprehensive impact of material aging, structural fatigue, and operating frequency changes on the equipment life to establish an equipment aging model. Based on the equipment aging model, an aging-power supply data set is generated, including: Performing feature extraction processing according to the atlas data set, and obtaining an aging feature set by screening out representative features that affect device aging from the atlas data set; Performing structural learning processing according to the aging feature set, mining conditional dependencies between features in time series data by applying a dynamic Bayesian network and establishing a model structure of the equipment aging path, thereby obtaining an aging model structure diagram; Performing parameter learning and model calibration processing according to the aging model structure diagram, estimating parameter values ​​of each conditional probability distribution in the equipment aging model by combining historical data, and obtaining a calibration model; The aging impact is quantified according to the calibration model, and the aging-power supply data set is obtained by simulating the time series prediction in the aging model and quantifying the aging rate of each key component and the preset power supply demand of the customer.

4. The power supply prediction method based on the service life of power transmission and transformation equipment according to claim 1 is characterized in that: Performing time series analysis and processing on the aging-power supply data set and the historical power supply data to obtain power supply cost data includes: Performing time series feature extraction processing according to the aging-power supply data set, analyzing the time dependence of the aging rate data by using an autoregressive integrated moving average model, extracting trend, seasonality and residual time series features, and obtaining time series data; Performing classification processing according to the time series data, and obtaining classification results by dividing the time series data with similar aging patterns into at least two groups; A model is constructed based on the classification results, and a power supply cost model is obtained by quantifying the contribution of different features to the power supply cost by using a Bayesian linear regression algorithm and generating an estimated value of the power supply cost under each category; A standardized power supply cost data set is output based on the power supply cost model.

5. A power supply prediction system based on the service life of power transmission and transformation equipment, characterized in that: include: An acquisition module is used to acquire a basic information data set and historical power supply data for each power transmission and transformation equipment, wherein the basic information includes the type, rated capacity, technical specifications and expected service life of the equipment, and the historical power supply data includes power supply demand data for the operating area of ​​each power transmission and transformation equipment; A construction module, used for performing a graph construction process according to all the basic information data sets, wherein a multi-level characteristic graph is constructed by combining the properties of the insulating material, the configuration of the cooling system, the type of the switch device and the corresponding operating frequency; An aggregation module is used to process the connection relationship between different levels in the multi-level characteristic map based on a preset hybrid graph convolutional network, and extract implicit features in the map in combination with a random walk algorithm to obtain a map data set for each power transmission and transformation equipment; A generation module is used to perform model building processing according to all the atlas data sets, analyze the aging path of the equipment by using a dynamic Bayesian network, quantify the comprehensive impact of material aging, structural fatigue and operating frequency changes on the life of the equipment, establish an equipment aging model, and generate an aging-power supply data set based on the equipment aging model, wherein the aging-power supply data set includes aging rate data of each key component and power supply demand data of the customer; An analysis module, configured to perform time series analysis and processing based on the aging-power supply data set and the historical power supply data to obtain power supply cost data; A prediction module, used to make predictions based on the power supply cost data, combined with the historical power supply data and the customer's power supply demand data, and generate a power supply prediction report; Wherein, the building blocks include: A first analysis unit is used to perform material property cluster analysis based on the basic information data set, cluster the chemical properties, electrical properties, withstand voltage grades and thermal stability of different insulating materials by using a K-means algorithm to obtain material clusters with similar properties, and generate a material property cluster data set; A second analysis unit is used to perform cooling system configuration mode analysis based on the basic information data set, extract medium type, flow control mode and heat exchanger type configuration features of the cooling system by using principal component analysis, and perform dimension reduction and feature selection processing to obtain a cooling system feature vector set; A third analysis unit is used to perform association analysis based on the basic information data set, mine potential patterns of high-frequency operations and specific types of switch devices using a frequent pattern mining algorithm, and analyze the matching degree between the design load and actual usage of the switch device to obtain a switch device characteristic data set; The first construction unit is used to perform graph construction processing based on the material property cluster data set, the cooling system feature vector set and the switch device characteristic data set, and to construct a multi-level characteristic graph reflecting the comprehensive characteristics of the power transmission and transformation equipment by using a graph neural network to perform node feature aggregation and multi-relationship edge weight learning, and quantify the relationship strength and influence between different characteristics.

6. The power supply prediction system based on the service life of power transmission and transformation equipment according to claim 5 is characterized in that: The aggregation module includes: A first processing unit is used to perform graph convolution processing according to the node features and edge connection relationships of the multi-level feature map, by applying a hybrid graph convolution network to transfer and aggregate feature information between different layers, taking the neighbor features of the node as input, and performing convolution operations in combination with a specific weight matrix to capture the interactive relationship between different features, thereby obtaining a node feature matrix after convolution; A first fusion unit is used to perform multi-scale feature fusion processing according to the node feature matrix, by introducing a multi-scale feature fusion mechanism in the hybrid graph convolutional network, weightedly combining the features output by different layers to integrate information of different scales, and obtaining a fused first feature matrix; A second processing unit is used to perform random walk processing according to the fused first feature matrix, and obtain a second feature matrix containing random walk features by simulating multiple random paths on the graph and capturing long-distance dependencies and structural information in the graph; The first integration unit is used to perform feature integration processing according to the second feature matrix, and perform dimension reduction and standardization processing on the integrated features to obtain a graph data set for each power transmission and transformation equipment.

7. The power supply prediction system based on the service life of power transmission and transformation equipment according to claim 5 is characterized in that: The generation module comprises: A first extraction unit, configured to perform feature extraction processing according to the atlas data set, and obtain an aging feature set by screening out representative features that affect device aging from the atlas data set; A third processing unit is used to perform structural learning processing according to the aging feature set, and to mine the conditional dependency relationship between features in the time series data by applying a dynamic Bayesian network and to establish a model structure of the equipment aging path, so as to obtain an aging model structure diagram; A first calibration unit, configured to perform parameter learning and model calibration processing according to the aging model structure diagram, and obtain a calibration model by estimating parameter values ​​of each conditional probability distribution in the device aging model in combination with historical data; The fourth processing unit is used to quantify the aging impact according to the calibration model, by simulating the time series prediction in the aging model, and quantifying the aging rate of each key component and the preset power supply demand of the customer, and integrating to obtain the aging-power supply data set.

8. The power supply prediction system based on the service life of power transmission and transformation equipment according to claim 5 is characterized in that: The analysis module comprises: A second extraction unit is used to perform time series feature extraction processing according to the aging-power supply data set, and extract trend, seasonality and residual time series features by using an autoregressive integrated moving average model to analyze the time dependence of the aging rate data, so as to obtain time series data; A first classification unit is used to perform classification processing according to the time series data, and obtain a classification result by dividing the time series data with similar aging patterns into at least two groups; The second construction unit is used to perform model construction processing according to the classification result, quantify the contribution of different features to the power supply cost by using a Bayesian linear regression algorithm, and generate an estimated value of the power supply cost under each category to obtain a power supply cost model; the first output unit outputs a standardized power supply cost data set based on the power supply cost model.

Images