Cable status prediction method and system based on smart grid

Abstract

The present invention provides a cable state prediction method and system based on smart grid, the method comprising the following steps: obtaining the cable network topology and cable state data of the target smart grid; constructing the real-time power grid diagram network of the target smart grid in combination with the real-time cable state data and the cable network topology; constructing the initial cable state prediction model in combination with the cable basic state prediction module and the external state prediction module; training the initial cable state prediction model to obtain the target cable state prediction model; obtaining the real-time external environment data of the target smart grid at the same time as the real-time cable state data; synchronously inputting the real-time external environment data and the real-time power grid diagram network into the target cable state prediction model to predict the comprehensive value of the predicted node state; marking the cable diagram node whose predicted node state comprehensive value is lower than the preset comprehensive value threshold as an abnormal cable diagram node. The present invention has the effect of having a high accuracy on the cable state prediction result.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grids, and in particular relates to a cable state prediction method and system based on smart grids. Background Art

[0002] The distribution network is an important part of the power network. Its main function is to receive electric energy from the transmission network or regional power plants, and distribute the electric energy locally or step by step according to the voltage to various users through distribution facilities. This link is crucial to ensure the stability and reliability of power supply, and is the cornerstone of meeting the growing demand for electricity in modern society. However, with the rapid development of society and the continuous growth of electricity demand, traditional distribution network facilities can no longer meet the growing needs of operation and management. Therefore, on the basis of traditional distribution networks, smart grids that combine advanced sensor measurement technology and control methods have emerged.

[0003] In the process of the development of smart grid, the grid topology of smart grid is becoming more and more complex. Cable is the most important component of the grid topology. If a cable fails somewhere in the grid topology, it may affect the normal operation of the entire smart grid, which may cause huge economic losses. Therefore, how to predict the status of cables in smart grid is an urgent problem to be solved in the industry.

[0004] In the prior art, a long short-term memory network model (LSTM) is usually used to predict the cable status. Specifically, the LSTM model is first constructed, and then the historical cable status data is used to train the LSTM model to learn the time series characteristics of the cable status data. Finally, the trained LSTM model is used to predict the future cable status, so as to detect cable failures and damage in advance, and take corresponding measures for maintenance and repair. However, in fact, the cable status data is still interfered by a large amount of external environmental data, and the prior art only considers the time series characteristics of the cable status data in the prediction process, and the feature learning of the prediction model is relatively single, which will lead to the final prediction result of the prediction model being relatively inaccurate. Summary of the invention

[0005] The present invention provides a cable status prediction method and system based on a smart grid to solve the problem that the final prediction result of the prediction model is relatively inaccurate.

[0006] In a first aspect, the present invention provides a cable status prediction method based on a smart grid, the method comprising the following steps:

[0007] Acquire the cable network topology and cable status data of the target smart grid, wherein the cable status data includes real-time cable status data and historical cable status data at multiple historical moments;

[0008] Generate cable graph nodes by combining the real-time cable status data and all target cables in the cable network topology, and generate node edges between the respective cable graph nodes according to the real-time power transmission paths in the cable network topology to obtain a real-time power grid graph network of the target smart grid;

[0009] Constructing a cable basic state prediction module based on a graph convolutional neural network, constructing an external state prediction module based on a long short-term memory network, and combining the cable basic state prediction module and the external state prediction module to construct an initial cable state prediction model;

[0010] Taking all the historical moments as a reference, obtaining historical external environment data of the area where the target smart grid is located corresponding to a plurality of the historical cable status data, training the initial cable status prediction model in combination with all the historical external environment data and all the historical cable status data, and obtaining a target cable status prediction model;

[0011] Acquire real-time external environment data of the target smart grid at the same time as the real-time cable status data;

[0012] The real-time external environment data and the real-time power grid diagram network are synchronously input into the target cable state prediction model to predict the predicted node state comprehensive value of each of the cable diagram nodes in the real-time power grid diagram network;

[0013] The cable diagram node whose predicted node state comprehensive value is lower than a preset comprehensive value threshold is marked as an abnormal cable diagram node, and the target cable corresponding to the abnormal cable diagram node is marked as a predicted faulty cable.

[0014] Optionally, the generating of cable graph nodes by combining the real-time cable status data and all target cables in the cable network topology, and generating node edges between the respective cable graph nodes according to the real-time power transmission paths in the cable network topology, to obtain the real-time power grid graph network of the target smart grid comprises the following steps:

[0015] generating a plurality of initial cable map nodes based on all target cables in the cable network topology;

[0016] According to the real-time cable status data, the initial cable diagram nodes corresponding to the target cables that are not enabled are marked as hidden diagram nodes, and all other retained initial cable diagram nodes are used as cable diagram nodes;

[0017] Determine the node connection relationship between each of the cable diagram nodes based on the real-time transmission path in the cable network topology, and count the node connection number and node transmission length of each of the cable diagram nodes based on the real-time transmission path;

[0018] Node edges between the various cable diagram nodes are generated according to the node connection relationship, and node edge weights are assigned to all the node edges in combination with the node connection quantity and the node transmission length to obtain a real-time power grid diagram network of the target smart grid.

[0019] Optionally, the calculation formula of the node edge weight is as follows:

[0020]

[0021] Where: i≠j, ω ij represents the node edge weight between the cable graph node i and the cable graph node j, α and β both represent adjustable parameters, exp(·) represents an exponential function, and D i represents the number of node connections of the cable diagram node i, D j represents the number of node connections of the cable diagram node j, D n represents the number of node connections of the nth cable diagram node in the real-time power grid diagram network, N represents the number of nodes of the cable diagram node in the real-time power grid diagram network, l i represents the node transmission length of node i in the cable diagram, l j represents the node transmission length of the cable diagram node j, and ψ represents the standard deviation of the node transmission lengths of all the cable diagram nodes in the real-time power grid diagram network.

[0022] Optionally, the real-time cable status data and the historical cable status data are both multimodal cable status data, the real-time cable status data include real-time cable operation status data and cable joint real-time picture data, and the historical cable status data include historical cable operation status data and cable joint historical picture data.

[0023] Optionally, the step of acquiring historical external environment data of the target smart grid area corresponding to a plurality of historical cable status data based on all the historical moments, training the initial cable status prediction model in combination with all the historical external environment data and all the historical cable status data, and obtaining the target cable status prediction model comprises the following steps:

[0024] Mark the historical moments in which the cable start-stop process exists among the multiple historical moments as target historical moments, retain the target historical cable status data corresponding to all the target historical moments, and sort all the target historical cable status data in reverse order according to the target historical moments to obtain the target historical cable status time series data, wherein the cable start-stop process is the activation process and / or deactivation process of any number of the target cables in the target smart grid;

[0025] Traversing the target historical cable status time series data, reversely updating and saving the real-time power grid diagram network according to each target historical cable status data, and obtaining a historical time series power grid diagram network;

[0026] Taking all the target historical moments as indexes, extracting multiple historical external environment data of the area where the target smart grid is located from a preset external environment database, and sorting all the historical external environment data in reverse order according to the target historical moments to obtain historical external environment time series data;

[0027] The cable basic state prediction module is trained using the historical time-series power grid diagram network, and the external state prediction module is synchronously trained using the historical external environment time-series data based on the target historical moment, all model parameters in the initial cable state prediction model are optimized to obtain the target cable state prediction model.

[0028] Optionally, the real-time power grid diagram network also includes a hidden graph node, the target cable corresponding to the hidden graph node is a cable that is not yet enabled in the target smart grid, and the hidden graph node does not have any node edge. The traversing of the target historical cable status time series data, reversely updating and saving the real-time power grid diagram network according to each target historical cable status data, and obtaining the historical time series power grid diagram network includes the following steps:

[0029] Traversing the target historical cable state time series data, executing a power grid diagram network reverse update step for each target historical cable state data, and obtaining a plurality of historical power grid diagram networks;

[0030] Integrate all the historical power grid diagram networks in reverse order according to the target historical moment to obtain a historical time series power grid diagram network;

[0031] The power grid diagram network reverse updating step specifically includes the following steps:

[0032] Calculate the historical node state comprehensive value of each of the hidden graph nodes and each of the cable graph nodes respectively according to the target historical cable state data;

[0033] Switch the cable diagram node whose historical node status comprehensive value is less than a preset switching threshold to the hidden diagram node, and switch the hidden diagram node whose historical node status comprehensive value is greater than or equal to the switching threshold to the cable diagram node;

[0034] Determine the historical node connection relationship between the cable graph nodes after each node is switched based on the historical power transmission path in the cable network topology when the target historical cable status data is at the target historical moment;

[0035] The historical node edges between the cable diagram nodes after each node switching are generated by combining the historical transmission path and the historical node connection relationship, and historical node edge weights are assigned to the historical node edges to obtain the historical power grid diagram network after reverse update.

[0036] Optionally, the historical cable operation status data includes historical cable temperature data, historical cable current data, historical cable voltage data and historical cable internal resistance data, and the step of calculating the comprehensive historical node status values ​​of each hidden graph node and each cable graph node according to the target historical cable status data includes the following steps:

[0037] The following historical node state comprehensive value calculation steps are performed for each of the hidden graph nodes or the cable graph nodes:

[0038] Normalize and integrate the historical cable operation status data corresponding to the hidden graph node or the cable graph node in the target historical cable status data to obtain a comprehensive value of the historical node operation status of the hidden graph node or the cable graph node;

[0039] Processing the cable joint historical picture data corresponding to the hidden graph node or the cable graph node through a preset image recognition model, and calculating the historical joint breakage rate of the cable joint in the cable joint historical picture data;

[0040] The comprehensive value of the historical node operating status and the historical joint damage rate are weightedly calculated to obtain the comprehensive value of the historical node status of the hidden graph node or the cable graph node.

[0041] Optionally, the method of using the historical time-series power grid diagram network to train the cable basic state prediction module, and using the historical external environment time-series data to synchronously train the external state prediction module based on the target historical moment, optimizing all model parameters in the initial cable state prediction model, and obtaining the target cable state prediction model includes the following steps:

[0042] Executing a matrix construction step on each of the historical power grid diagram networks in the historical time-series power grid diagram network to obtain a historical training matrix for each of the historical power grid diagram networks;

[0043] According to the arrangement order of the historical power grid network in the historical time-series power grid network, the historical power grid network and the historical training matrix are sequentially input as training data into the cable basic state prediction module to obtain a first loss function output by the cable basic state prediction module;

[0044] Respectively extracting historical external features of each of the historical external environment data in the historical external environment time series data;

[0045] According to the arrangement order of the historical external environment data in the historical external environment time series data, the historical external features are sequentially input as training data into the external state prediction module to obtain a second loss function output by the external state prediction module, and the training data input process of the external state prediction module and the training data input process of the cable foundation state prediction module are synchronized with each other based on the target historical moment;

[0046] Back-propagating the first loss function to the cable basic state prediction module to optimize the model parameters in the cable basic state prediction module, and synchronously back-propagating the second loss function to the external state prediction module based on the target historical moment to optimize the model parameters in the external state prediction module;

[0047] When the model parameters in the cable basic state prediction module and the external state prediction module are optimized, a target cable state prediction model is obtained;

[0048] The matrix construction step comprises the following steps:

[0049] Constructing a historical node state feature matrix based on the historical node state comprehensive value of each of the cable diagram nodes in the historical power grid diagram network;

[0050] Constructing a historical adjacency matrix based on the historical node edge weights of each of the historical node edges in the historical power grid graph network;

[0051] The historical node state feature matrix and the historical adjacency matrix are combined into a historical training matrix.

[0052] Optionally, the historical external environment data includes historical external humidity data and historical external temperature data, and the real-time external environment data includes real-time external humidity data and real-time external temperature data.

[0053] In a second aspect, the present invention also provides a cable status prediction system based on a smart grid, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the cable status prediction method based on the smart grid as described in the first aspect is implemented.

[0054] The beneficial effects of the present invention are:

[0055] The present invention constructs a cable state prediction model that comprehensively utilizes graph convolutional neural networks and long short-term memory networks. This model can monitor and predict the cable state in the smart grid in real time by integrating real-time and historical cable state data, cable network topology information, and external environmental data. The prediction model deeply considers the impact of external environmental data on the cable state and extracts external environmental features during the prediction process. Therefore, compared with the prior art, the present invention has a more accurate prediction result of the future state of the cable. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a schematic flow chart of the steps of a cable status prediction method based on a smart grid in one embodiment of the present application.

[0057] Figure 2 This is a schematic diagram of the framework flow of a cable status prediction method based on a smart grid in one embodiment of the present application. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present application to clearly describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments in the present application belong to the scope of protection of this application.

[0059] The terms "first", "second", etc. in the specification and claims of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally of one type, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification and claims represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0060] Figure 1 FIG. 1 is a flow chart of a method for predicting cable status based on a smart grid in one embodiment. It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1At least part of the steps in the above method may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps. Figure 1 As shown, a cable status prediction method based on smart grid disclosed in the present invention specifically includes the following steps:

[0061] S101. Obtain cable network topology and cable status data of the target smart grid.

[0062] Among them, the target smart grid refers to the smart grid that needs to predict the cable status. The cable network topology and cable status data can be obtained through the background management system of the target smart grid. The cable network topology data provides the physical and logical connections between cables. The cable status data includes real-time cable status data and historical cable status data at multiple historical moments. The real-time cable status data and historical cable status data are both multimodal cable status data. The real-time cable status data includes real-time cable operation status data and real-time image data of cable joints. The historical cable status data includes historical cable operation status data and historical image data of cable joints. The historical cable operation status data includes historical cable temperature data, historical cable current data, historical cable voltage data and historical cable internal resistance data. The real-time cable operation status data includes real-time cable temperature data, real-time cable current data, real-time cable voltage data and real-time cable internal resistance data. Among them, the cable joint picture can be collected by an image acquisition device pre-set at the cable connection in the smart grid.

[0063] S102. Generate cable diagram nodes by combining the real-time cable status data and all target cables in the cable network topology, and generate node edges between each cable diagram node according to the real-time transmission path in the cable network topology to obtain the real-time grid diagram network of the target smart grid.

[0064] S103. Construct a cable basic state prediction module based on the graph convolutional neural network, construct an external state prediction module based on the long short-term memory network, and construct an initial cable state prediction model by combining the cable basic state prediction module and the external state prediction module.

[0065] Among them, graph convolutional neural networks (GCNs) and long short-term memory networks (LSTMs) are two different types of deep learning models, which are used to process graph structure data and time series data respectively. Graph convolutional neural networks (GCNs) are a type of deep learning model specifically designed to process graph data. It can learn directly on the graph structure and capture the complex relationships and patterns between nodes. GCN applies convolution operations on the graph, allowing each node to aggregate and learn the features of neighboring nodes to obtain global information. The core idea of ​​GCN is to update the representation of nodes through graph convolution layers. In each convolution operation, a node aggregates the information (including feature information) of its neighboring nodes, usually by weighted averaging. Then, this aggregated information is transformed through a nonlinear activation function to generate a new node feature representation. By stacking multiple graph convolution layers, the model can capture a wider range of neighbor information and achieve deep learning.

[0066] Long short-term memory networks (LSTMs) are a special type of recurrent neural network (RNN) designed to solve the vanishing or exploding gradient problem faced by standard RNNs when processing long sequences of data. By introducing three gates (input gate, forget gate, and output gate) and a cell state, LSTMs can effectively control the storage, update, and forgetting of information, making it perform well in learning time dependencies of long sequences. Among them, the input gate determines which information will be added to the cell state. The forget gate determines which information will be deleted from the cell state. The output gate determines the output value based on the current cell state and input. The cell state is the core part of the LSTM network, which can flow and save important information in the network.

[0067] S104. Taking all historical moments as a benchmark, obtain historical external environment data of the target smart grid area corresponding to multiple historical cable status data, train an initial cable status prediction model by combining all historical external environment data and all historical cable status data, and obtain a target cable status prediction model.

[0068] The historical external environment data includes historical external humidity data and historical external temperature data.

[0069] S105. Acquire the real-time external environment data of the target smart grid at the same time as the real-time cable status data.

[0070] The real-time external environment data includes real-time external humidity data and real-time external temperature data.

[0071] S106. The real-time external environment data and the real-time power grid diagram network are synchronously input into the target cable state prediction model, and the predicted node state comprehensive value of each cable diagram node in the real-time power grid diagram network is predicted.

[0072] Among them, a real-time node state feature matrix is ​​constructed based on the real-time node state comprehensive value of each cable diagram node in the real-time power grid diagram network. A real-time adjacency matrix is ​​constructed based on the node edge weights of each node edge in the real-time power grid diagram network. The real-time node state feature matrix and the real-time adjacency matrix are combined into a real-time training matrix. Among them, the real-time node state comprehensive value can be calculated by the real-time cable state data, specifically:

[0073] The real-time cable operation status data includes real-time cable temperature data, real-time cable current data, real-time cable voltage data and real-time cable internal resistance data. For each cable diagram node, the following real-time node status comprehensive value calculation steps are performed:

[0074] The real-time cable operation status data corresponding to the cable diagram node in the real-time cable status data is normalized and integrated to obtain the comprehensive value of the real-time node operation status of the cable diagram node. Specifically, the difference ratios of the real-time cable temperature data, the real-time cable current data, the real-time cable voltage data and the real-time cable internal resistance data in their respective preset normal data intervals are calculated respectively. Taking the real-time cable temperature data as an example, assuming that the real-time cable temperature data is 8, and the preset normal data interval of the real-time cable temperature data is [0,10], first determine that the middle value of the normal data interval is (0+10) / 2=5, and then use the absolute value of the difference between the middle value and the upper limit or lower limit of the normal data interval as the denominator of the difference ratio, and then use the absolute value of the difference between the real-time cable temperature data and the middle value as the numerator of the difference ratio, so as to obtain the difference ratio of the real-time cable temperature data in the normal data interval. Then, the difference ratios of the real-time cable temperature data, the real-time cable current data, the real-time cable voltage data and the real-time cable internal resistance data are weighted averaged to obtain the comprehensive value of the real-time node operation status.

[0075] The real-time image data of the cable joints corresponding to the cable diagram nodes are processed by a preset image recognition model, and the real-time joint damage rate of the cable joints in the real-time image data of the cable joints is calculated. Specifically, before the image data are input into the image recognition model, they need to be preprocessed to ensure that they meet the input requirements of the model. The preprocessing includes adjusting the image to the input size required by the model, normalizing the pixel values, and converting the color space of the image. The preprocessed image data is then input into the trained image recognition model. The model is usually a deep learning network, such as a convolutional neural network (CNN), which can extract features from the image and perform classification or detection tasks. The final image recognition model outputs the location and size of the damaged area of ​​the cable joint. According to the output results of the image recognition model, the pixels of all damaged areas in the image are identified, and then the number of pixels occupied by the damaged area is calculated, and then the number of pixels of the cable joint as a whole is calculated. Finally, the damaged area is divided by the total area to obtain the real-time joint damage rate. After obtaining the comprehensive value of the real-time node operation status and the real-time joint damage rate, the comprehensive value of the real-time node operation status and the real-time joint damage rate are weightedly calculated to obtain the comprehensive value of the real-time node status of the cable diagram node.

[0076] Next, the real-time external data features of the real-time external environment data are extracted, and then the real-time external data features and the real-time training matrix are input into the target cable state prediction model. The output results of the real-time external data features and the real-time training matrix are respectively combined with the external state prediction module and the cable basic state prediction module to finally obtain the predicted node state comprehensive value of each cable diagram node in the real-time power grid diagram network.

[0077] S107. Mark the cable diagram nodes whose predicted node status comprehensive values ​​are lower than the preset comprehensive value threshold as abnormal cable diagram nodes, and mark the target cables corresponding to the abnormal cable diagram nodes as predicted faulty cables.

[0078] In one implementation manner, step S102 specifically includes the following steps:

[0079] generating a plurality of initial cable map nodes based on all target cables in the cable network topology;

[0080] According to the real-time cable status data, the initial cable diagram nodes corresponding to the unactivated target cables are marked as hidden diagram nodes, and all other retained initial cable diagram nodes are used as cable diagram nodes;

[0081] Determine the node connection relationship between each cable diagram node based on the real-time transmission path in the cable network topology, and count the node connection number and node transmission length of each cable diagram node based on the real-time transmission path;

[0082] The node edges between the nodes of each cable diagram are generated according to the node connection relationship, and the node edge weights are assigned to all node edges in combination with the number of node connections and the node transmission length to obtain the real-time power grid diagram network of the target smart grid.

[0083] In this embodiment, the cable graph nodes represent cables in the power grid. When the power grid topology is represented in the graph, each cable corresponds to a node. Based on the data of the cable network topology, a graph node is created for each cable in the power grid. In the real-time power grid graph network, unenabled cables should not affect the real-time analysis of the power grid, so they can be marked as hidden. Check the real-time cable status data and identify the cables that are currently unenabled. Mark the graph nodes corresponding to these unenabled cables as hidden. These nodes can be ignored or specially treated in subsequent graph network analysis. The connection relationship between the cable graph nodes reflects the transmission path of the power grid. Changes in the real-time transmission path will affect the operating status of the power grid. Based on the real-time transmission path data, determine which cable graph nodes have direct connections. For each node, count the number of other nodes it connects to and the transmission length between these nodes.

[0084] In a graph network, the connections between nodes are represented by edges. The weight of an edge can represent the importance of a connection or the efficiency of transmission, which is very important for analyzing load distribution and fault propagation in a power grid. Based on the node connection relationship, an edge is created for each pair of directly connected nodes in the graph. A weight is assigned to each edge. The weight can be calculated based on the number of connections and the transmission length. The calculation formula for the node edge weight is as follows:

[0085]

[0086] Where: i≠j, ω ij represents the node edge weight between cable graph node i and cable graph node j, α and β are both adjustable parameters, exp(·) represents the exponential function, and D i represents the number of node connections of node i in the cable diagram, D j represents the number of node connections of node j in the cable diagram, D n Indicates the number of node connections of the nth cable diagram node in the real-time power grid network. N indicates the number of nodes of the cable diagram node in the real-time power grid network. l i represents the node transmission length of node i in the cable diagram, L j represents the node transmission length of the cable diagram node j, and ψ represents the standard deviation of the node transmission lengths of all cable diagram nodes in the real-time power grid diagram network.

[0087] In another embodiment, the node edge weight can be calculated by setting a control threshold ε, so that the matrix distribution and sparsity can be controlled when the adjacency weight matrix is ​​constructed by the node edge weight. In this case, the calculation formula of the node edge weight is as follows:

[0088]

[0089] In one implementation manner, step S104 specifically includes the following steps:

[0090] Mark the historical moments in which the cable start-stop process exists among the multiple historical moments as the target historical moments, retain the target historical cable status data corresponding to all the target historical moments, and sort all the target historical cable status data in reverse order according to the target historical moments to obtain the target historical cable status time series data. The cable start-stop process is the activation process and / or deactivation process of any number of target cables in the target smart grid;

[0091] Traversing the target historical cable status time series data, reversely updating and saving the real-time power grid diagram network according to each target historical cable status data, and obtaining the historical time series power grid diagram network;

[0092] Taking all target historical moments as indexes, extracting historical external environment data of the areas where multiple target smart grids are located from a preset external environment database, and sorting all historical external environment data in reverse order according to the target historical moments to obtain historical external environment time series data;

[0093] The cable basic state prediction module is trained using the historical time-series power grid diagram network, and the external state prediction module is trained synchronously using the historical external environment time-series data based on the target historical moment, so as to optimize all model parameters in the initial cable state prediction model and obtain the target cable state prediction model.

[0094] In this implementation, the activation and deactivation process of the cable may be an important indicator of the change in the cable state, because these events may have an impact on the health of the cable. Therefore, selecting these specific moments as target historical moments can help the model better learn the change pattern of the cable state. All moments involving the start and stop of the cable in the historical data are marked, and only the cable state data corresponding to these target historical moments are retained. These data are sorted in reverse chronological order to form a time series data set so that the model can capture the most recent state changes.

[0095] Historical cable status data reflects the working status of the cable at different points in time, and the status of the cable may be affected by the grid topology. By constructing a historical time-series grid graph network, the model can understand how the cable status changes over time and network topology. For each target historical moment, the graph network is updated according to the cable status at that time to ensure that the network status at each moment accurately reflects the historical data. Since external environmental factors can affect the performance and life of the cable, incorporating this data into the model can improve the accuracy of the prediction.

[0096] In one embodiment, the real-time power grid diagram network also includes hidden graph nodes, the target cables corresponding to the hidden graph nodes are cables that are not yet enabled in the target smart grid, and the hidden graph nodes do not have any node edges. Traversing the target historical cable status time series data, reversely updating and saving the real-time power grid diagram network according to each target historical cable status data, and obtaining the historical time series power grid diagram network specifically includes the following steps:

[0097] Traversing the target historical cable state time series data, performing a reverse update step of the power grid diagram network for each target historical cable state data, and obtaining multiple historical power grid diagram networks;

[0098] All historical power grid diagram networks are integrated in reverse order according to the target historical moment to obtain the historical time series power grid diagram network;

[0099] The reverse updating step of the power grid diagram network specifically includes the following steps:

[0100] According to the target historical cable status data, the historical node status comprehensive values ​​of each hidden graph node and each cable graph node are calculated respectively;

[0101] Switch the cable graph nodes whose historical node status comprehensive value is less than the preset switching threshold to hidden graph nodes, and switch the hidden graph nodes whose historical node status comprehensive value is greater than or equal to the switching threshold to cable graph nodes;

[0102] Based on the historical power transmission path in the cable network topology at the target historical moment of the target historical cable status data, determine the historical node connection relationship between the cable diagram nodes after each node is switched;

[0103] Combining the historical transmission path and the historical node connection relationship, the historical node edges between the cable graph nodes after each node switching are generated, and the historical node edge weights are assigned to the historical node edges to obtain the historical power grid graph network after reverse update.

[0104] In this embodiment, the historical node state comprehensive value reflects the operating state of the cable at a certain moment in the past. Based on the historical state comprehensive value and the preset switching threshold, it can be determined which cables should be considered as active (i.e., visible graph nodes) at the historical moment and which should be hidden. The historical state comprehensive value of each cable graph node is compared with the preset switching threshold. If the historical state comprehensive value of the cable graph node is less than the threshold, the node is marked as a hidden graph node at the historical moment; if it is greater than or equal to the threshold, it is marked as an active cable graph node.

[0105] For the historical transmission path in the cable network topology at the target historical moment based on the target historical cable status data, the step of determining the historical node connection relationship between the cable diagram nodes after each node is switched, and combining the historical transmission path and the historical node connection relationship to generate the historical node edges between the cable diagram nodes after each node is switched, and assigning historical node edge weights to the historical node edges to obtain the historical power grid diagram network after reverse update, please refer to the detailed steps of step S102 for details.

[0106] In one embodiment, the historical cable operation status data includes historical cable temperature data, historical cable current data, historical cable voltage data, and historical cable internal resistance data. The step of calculating the historical node status comprehensive value of each hidden graph node and each cable graph node according to the target historical cable status data specifically includes the following steps:

[0107] For each hidden graph node or cable graph node, the following historical node status comprehensive value calculation steps are performed:

[0108] Normalize and integrate the historical cable operation status data corresponding to the hidden graph node or the cable graph node in the target historical cable status data to obtain the historical node operation status comprehensive value of the hidden graph node or the cable graph node;

[0109] The cable joint historical image data corresponding to the hidden graph node or the cable graph node is processed by a preset image recognition model to calculate the historical joint breakage rate of the cable joint in the cable joint historical image data;

[0110] The comprehensive value of the historical node operating status and the historical joint damage rate are weighted and calculated to obtain the comprehensive value of the historical node status of the hidden graph node or the cable graph node.

[0111] In this embodiment, the calculation of the historical node operation status comprehensive value, the historical joint damage rate and the historical node status comprehensive value may specifically refer to the detailed description in step S106.

[0112] In one embodiment, the cable basic state prediction module is trained using the historical time-series power grid diagram network, and the external state prediction module is trained using the historical external environment time-series data synchronously based on the target historical moment, and all model parameters in the initial cable state prediction model are optimized to obtain the target cable state prediction model. This step specifically includes the following steps:

[0113] Performing a matrix construction step on each historical power grid network in the historical time-series power grid network to obtain a historical training matrix for each historical power grid network;

[0114] According to the arrangement order of the historical power grid network in the historical time-series power grid network, the historical power grid network and the historical training matrix are sequentially input into the cable basic state prediction module as training data to obtain a first loss function output by the cable basic state prediction module;

[0115] Extracting historical external features of each historical external environment data in the historical external environment time series data respectively;

[0116] According to the arrangement order of the historical external environment data in the historical external environment time series data, the historical external features are sequentially input as training data into the external state prediction module to obtain the second loss function output by the external state prediction module. The training data input process of the external state prediction module and the training data input process of the cable foundation state prediction module are synchronized with each other based on the target historical moment;

[0117] Back-propagating the first loss function to the cable basic state prediction module to optimize the model parameters in the cable basic state prediction module, and synchronously back-propagating the second loss function to the external state prediction module based on the target historical moment to optimize the model parameters in the external state prediction module;

[0118] When the model parameters in the cable basic state prediction module and the external state prediction module are optimized, the target cable state prediction model is obtained;

[0119] The matrix construction step includes the following steps:

[0120] Construct a historical node state feature matrix based on the historical node state comprehensive value of each cable diagram node in the historical power grid diagram network;

[0121] Construct a historical adjacency matrix based on the historical node edge weights of each historical node edge in the historical power grid graph network;

[0122] The historical node state feature matrix and the historical adjacency matrix are combined into a historical training matrix.

[0123] In this embodiment, in order to train the model, the historical power grid diagram network data needs to be converted into a format that the model can process. The historical node state feature matrix and the historical adjacency matrix provide a mathematical expression of the power grid state and structure. Based on the historical node state comprehensive value of each cable diagram node, a feature matrix W1 is constructed, in which each row represents a node and each column represents the historical node state comprehensive value of the node. The historical adjacency matrix W2 represents the connection relationship between nodes in the graph network. If there is a connection between two nodes, the corresponding matrix element is assigned according to the historical node edge weight; otherwise, it is assigned to 0. The feature matrix and the adjacency matrix are juxtaposed to form a historical training matrix W O .

[0124] Reference Figure 2 , the historical power grid graph network G k =(V k , E k ) and the historical training matrix W O The signals are input into the cable basic state prediction module in the order of arrangement. The historical power grid graph network with time sequence is transformed into a power grid spectrum matrix through the sequence transformation layer in the cable basic state prediction module, and then converted to the frequency domain through discrete Fourier transform. The matrix features of the power grid spectrum matrix are obtained in the frequency domain through the convolution layer and gated linear unit. The frequency domain signal is then converted into a time domain signal through the inverse discrete Fourier transform. At this time, the power grid spectrum matrix is ​​input into the graph convolution layer of the cable basic state prediction module. According to the historical training matrix W O The graph convolution layer is used to perform graph convolution on the power grid spectrum matrix. The calculation formula is as follows:

[0125]

[0126] Where: θ represents the convolution kernel of the graph convolution layer, L represents the normalized graph Laplacian operator, U represents the eigenvector matrix of L, and Λ represents the eigenvalue diagonal matrix of L.

[0127] Among them, the normalized graph Laplacian operator is defined as follows:

[0128]

[0129] Where: I N represents the identity matrix, represents a diagonal matrix.

[0130] Finally, the inverse graph Fourier transform is performed on the result of the graph convolution and the training result is output through the fully connected layer to obtain the first loss function, which can be defined by the L2 loss. The first loss function is back-propagated to the cable basic state prediction module to optimize the model parameters in the cable basic state prediction module.

[0131] Considering the influence of external features such as temperature and humidity on the cable state, an external state prediction module is constructed by combining LSTM and a fully connected network. LSTM has a long-term memory function, which can solve the problems of gradient disappearance and gradient explosion in the long sequence training process, and is suitable for processing features that are highly correlated with time series. Since the external factors in this embodiment are highly correlated with time series, LSTM is selected to extract historical external features, and then a fully connected layer is connected for mapping features from low dimensions to high dimensions. Finally, the external feature prediction results are output, and the second loss function is obtained. The second loss function can be defined by the cross entropy loss function. The second loss function is synchronously back-propagated to the external state prediction module based on the target historical moment to optimize the model parameters in the external state prediction module.

[0132] The present invention also discloses a cable status prediction system based on a smart grid, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the cable status prediction method based on the smart grid as described in any one of the above-mentioned embodiments is implemented.

[0133] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0134] Among them, the memory can be an internal storage unit of a computer device, such as a hard disk or memory of a computer device, or an external storage device of a computer device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD) or a flash memory card (FC) equipped on the computer device, etc., and the memory can also be a combination of an internal storage unit and an external storage device of a computer device. The memory is used to store computer programs and other programs and data required by the computer device. The memory can also be used to temporarily store data that has been output or is to be output, and this application does not impose any restrictions on this.

[0135] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as above, which are not provided in detail for the sake of simplicity.

[0136] One or more embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the present application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of the present application should be included in the protection scope of the present application.

Claims

1. A cable status prediction method based on smart grid, characterized in that: The steps include: Acquire the cable network topology and cable status data of the target smart grid, wherein the cable status data includes real-time cable status data and historical cable status data at multiple historical moments; generating a plurality of initial cable map nodes based on all target cables in the cable network topology; According to the real-time cable status data, the initial cable diagram nodes corresponding to the target cables that are not enabled are marked as hidden diagram nodes, and all other retained initial cable diagram nodes are used as cable diagram nodes; Determine the node connection relationship between each of the cable diagram nodes based on the real-time transmission path in the cable network topology, and count the node connection number and node transmission length of each of the cable diagram nodes based on the real-time transmission path; Generate node edges between the nodes of the cable diagram according to the node connection relationship, assign node edge weights to all the node edges in combination with the node connection quantity and the node transmission length, and obtain a real-time power grid diagram network of the target smart grid; Constructing a cable basic state prediction module based on a graph convolutional neural network, constructing an external state prediction module based on a long short-term memory network, and combining the cable basic state prediction module and the external state prediction module to construct an initial cable state prediction model; Taking all the historical moments as a reference, obtaining historical external environment data of the area where the target smart grid is located corresponding to a plurality of the historical cable status data, training the initial cable status prediction model in combination with all the historical external environment data and all the historical cable status data, and obtaining a target cable status prediction model; Acquire real-time external environment data of the target smart grid at the same time as the real-time cable status data; The real-time external environment data and the real-time power grid diagram network are synchronously input into the target cable state prediction model to predict the predicted node state comprehensive value of each of the cable diagram nodes in the real-time power grid diagram network; The cable diagram node whose predicted node state comprehensive value is lower than a preset comprehensive value threshold is marked as an abnormal cable diagram node, and the target cable corresponding to the abnormal cable diagram node is marked as a predicted faulty cable.

2. The cable status prediction method based on smart grid according to claim 1 is characterized in that: The calculation formula of the node edge weight is as follows: ； Where: , Represents the cable diagram node and the cable diagram nodes The edge weights between the nodes, and All represent adjustable parameters. represents the exponential function, Represents the cable diagram node The number of node connections, Represents the cable diagram node The number of node connections, Indicates the real-time power grid diagram network The number of node connections of the cable diagram node, represents the number of nodes of the cable diagram node in the real-time power grid diagram network, Represents the cable diagram node The node transmission length, Represents the cable diagram node The node transmission length, Represents the standard deviation of the node transmission lengths of all the cable diagram nodes in the real-time power grid diagram network.

3. The cable status prediction method based on smart grid according to claim 1 is characterized in that: The real-time cable status data and the historical cable status data are both multimodal cable status data. The real-time cable status data includes real-time cable operation status data and real-time picture data of cable joints. The historical cable status data includes historical cable operation status data and historical picture data of cable joints.

4. The cable status prediction method based on smart grid according to claim 3 is characterized in that: The step of obtaining the historical external environment data of the target smart grid area corresponding to the multiple historical cable status data based on all the historical moments, training the initial cable status prediction model in combination with all the historical external environment data and all the historical cable status data, and obtaining the target cable status prediction model comprises the following steps: Mark the historical moments in which the cable start-stop process exists among the multiple historical moments as target historical moments, retain the target historical cable status data corresponding to all the target historical moments, and sort all the target historical cable status data in reverse order according to the target historical moments to obtain the target historical cable status time series data, wherein the cable start-stop process is the activation process and / or deactivation process of any number of the target cables in the target smart grid; Traversing the target historical cable status time series data, reversely updating and saving the real-time power grid diagram network according to each target historical cable status data, and obtaining a historical time series power grid diagram network; Taking all the target historical moments as indexes, extracting multiple historical external environment data of the area where the target smart grid is located from a preset external environment database, and sorting all the historical external environment data in reverse order according to the target historical moments to obtain historical external environment time series data; The cable basic state prediction module is trained using the historical time-series power grid diagram network, and the external state prediction module is synchronously trained using the historical external environment time-series data based on the target historical moment, all model parameters in the initial cable state prediction model are optimized to obtain the target cable state prediction model.

5. The cable status prediction method based on smart grid according to claim 4 is characterized in that: The real-time power grid diagram network also includes a hidden graph node, the target cable corresponding to the hidden graph node is a cable that is not yet enabled in the target smart grid, and the hidden graph node does not have any node edge. The traversing of the target historical cable state time series data, reversely updating and saving the real-time power grid diagram network according to each target historical cable state data, and obtaining the historical time series power grid diagram network includes the following steps: Traversing the target historical cable state time series data, executing a power grid diagram network reverse update step for each target historical cable state data, and obtaining a plurality of historical power grid diagram networks; Integrate all the historical power grid diagram networks in reverse order according to the target historical moment to obtain a historical time series power grid diagram network; The power grid diagram network reverse updating step specifically includes the following steps: Calculate the historical node state comprehensive value of each of the hidden graph nodes and each of the cable graph nodes respectively according to the target historical cable state data; Switch the cable diagram node whose historical node status comprehensive value is less than a preset switching threshold to the hidden diagram node, and switch the hidden diagram node whose historical node status comprehensive value is greater than or equal to the switching threshold to the cable diagram node; Determine the historical node connection relationship between the cable graph nodes after each node is switched based on the historical power transmission path in the cable network topology when the target historical cable status data is at the target historical moment; The historical node edges between the cable diagram nodes after each node switching are generated by combining the historical transmission path and the historical node connection relationship, and historical node edge weights are assigned to the historical node edges to obtain the historical power grid diagram network after reverse update.

6. The cable status prediction method based on smart grid according to claim 5 is characterized in that: The historical cable operation status data includes historical cable temperature data, historical cable current data, historical cable voltage data and historical cable internal resistance data, and the historical node status comprehensive value of each hidden graph node and each cable graph node is calculated according to the target historical cable status data, including the following steps: For each of the hidden graph nodes or the cable graph nodes, the following historical node state comprehensive value calculation steps are performed: Normalize and integrate the historical cable operation status data corresponding to the hidden graph node or the cable graph node in the target historical cable status data to obtain a comprehensive value of the historical node operation status of the hidden graph node or the cable graph node; Processing the cable joint historical picture data corresponding to the hidden graph node or the cable graph node through a preset image recognition model, and calculating the historical joint breakage rate of the cable joint in the cable joint historical picture data; The comprehensive value of the historical node operating status and the historical joint damage rate are weightedly calculated to obtain the comprehensive value of the historical node status of the hidden graph node or the cable graph node.

7. The cable status prediction method based on smart grid according to claim 5 is characterized in that: The method of using the historical time-series power grid diagram network to train the cable basic state prediction module, and using the historical external environment time-series data to synchronously train the external state prediction module based on the target historical moment, optimizing all model parameters in the initial cable state prediction model, and obtaining the target cable state prediction model comprises the following steps: Executing a matrix construction step on each of the historical power grid diagram networks in the historical time-series power grid diagram network to obtain a historical training matrix for each of the historical power grid diagram networks; According to the arrangement order of the historical power grid network in the historical time-series power grid network, the historical power grid network and the historical training matrix are sequentially input as training data into the cable basic state prediction module to obtain a first loss function output by the cable basic state prediction module; Respectively extracting historical external features of each of the historical external environment data in the historical external environment time series data; According to the arrangement order of the historical external environment data in the historical external environment time series data, the historical external features are sequentially input as training data into the external state prediction module to obtain a second loss function output by the external state prediction module, and the training data input process of the external state prediction module and the training data input process of the cable foundation state prediction module are synchronized with each other based on the target historical moment; Back-propagating the first loss function to the cable basic state prediction module to optimize the model parameters in the cable basic state prediction module, and synchronously back-propagating the second loss function to the external state prediction module based on the target historical moment to optimize the model parameters in the external state prediction module; When the model parameters in the cable basic state prediction module and the external state prediction module are optimized, a target cable state prediction model is obtained; The matrix construction step comprises the following steps: Constructing a historical node state feature matrix based on the historical node state comprehensive value of each of the cable diagram nodes in the historical power grid diagram network; Constructing a historical adjacency matrix based on the historical node edge weights of each of the historical node edges in the historical power grid graph network; The historical node state feature matrix and the historical adjacency matrix are combined into a historical training matrix.

8. The cable status prediction method based on smart grid according to claim 1 is characterized in that: The historical external environment data includes historical external humidity data and historical external temperature data, and the real-time external environment data includes real-time external humidity data and real-time external temperature data.

9. A cable status prediction system based on a smart grid, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the cable status prediction method based on the smart grid is implemented as described in any one of claims 1 to 8.

Images