An electrical line overheating anomaly identification system based on deep learning

Abstract

The application provides an electrical line overheating anomaly identification system based on deep learning. The space-time correlation features of multivariate time series data in the operation process of an electrical line are determined, each variable data in the multivariate time series data is taken as a node, and a directed correlation graph representing the electrical line is constructed based on the space-time correlation features. The modal feature vector representing the operation mode of the electrical line is determined based on the historical multivariate time series data under the normal operation state of the electrical line and the directed correlation graph according to the influence features of each node on the overheating anomaly of the electrical line. The overheating anomaly score of the electrical line is determined through the modal feature vector and the multivariate time series data, and the overheating anomaly of the electrical line is identified based on the overheating anomaly score. According to the scheme, the overheating anomaly of the electrical line can be identified based on the complex dynamic correlation among the multivariate in the electrical line.

Description

Technical Field

[0001] This application relates to the field of overheating anomaly identification technology, and more specifically, to an electrical circuit overheating anomaly identification system based on deep learning. Background Technology

[0002] Overheating anomaly identification refers to the technical means of quickly detecting abnormal temperature rises that exceed the normal range by real-time monitoring and analysis of temperature data of equipment or systems, combined with preset thresholds or intelligent algorithms. This technology is widely used in industrial equipment, power systems, electronic components and data centers, and is a key link in realizing predictive maintenance and safety management.

[0003] With the rapid development of industrial automation and smart grids, the power loads borne by electrical lines are becoming increasingly complex, and the operating environments are becoming more diverse. Overheating anomalies have become a core hidden danger leading to line aging, short circuits, and even fires. Traditional electrical line overheating monitoring often relies on threshold judgments of a single parameter or captures local features through simple statistical analysis, which has significant limitations. On the one hand, electrical line overheating is the result of dynamic interactions among multiple variables such as current, voltage, ambient temperature, and equivalent resistance (e.g., a sudden increase in current will lag the temperature rise, and excessively high ambient temperature will weaken heat dissipation, leading to increased resistance). Single-parameter monitoring is difficult to reflect the spatiotemporal correlation between variables (e.g., time lag effect, spatial distribution coordination), and is prone to missing early weak anomalies or false alarms due to environmental fluctuations. On the other hand, the normal operating mode of the line is affected by load changes and environmental disturbances, exhibiting complex nonlinear characteristics. Traditional methods are difficult to construct accurate normal state benchmarks, resulting in insufficient adaptability and accuracy in anomaly identification, making it difficult to meet the safety monitoring needs of complex power systems. Therefore, how to identify electrical line overheating anomalies based on the complex dynamic correlations among multiple variables in electrical lines has become a problem faced by the industry. Summary of the Invention

[0004] This application provides a deep learning-based electrical circuit overheating anomaly identification system, which can identify electrical circuit overheating anomalies based on the complex dynamic correlation between multiple variables in the electrical circuit.

[0005] In a first aspect, this application provides a deep learning-based electrical circuit overheating anomaly identification system, comprising the following steps: The data acquisition module is used to collect multivariable time-series data during the operation of electrical circuits; The graph construction module is used to perform correlation analysis on the spatiotemporal features of each variable data in the multivariate time series data to obtain the spatiotemporal correlation features of the multivariate time series data. Based on the spatiotemporal correlation features, a directed correlation graph representing the correlation between the variable data of the electrical line is constructed using each variable data in the multivariate time series data as a node. The feature fusion module is used to determine the impact characteristics of each node on the overheating anomaly of the electrical line. Based on all the impact characteristics, the historical multivariate time series data of the electrical line under normal operation is fused with the directed correlation graph to obtain the modal feature vector characterizing the operating mode of the electrical line. An anomaly identification module is used to evaluate the overheating anomaly state of the electrical circuit using the modal feature vector and the multivariate time series data, obtain the overheating anomaly score of the electrical circuit, and identify the overheating anomaly of the electrical circuit based on the overheating anomaly score.

[0006] In some embodiments, the multivariate time-series data includes line current data, line voltage data, line surface temperature data, ambient temperature data, and line equivalent resistance data of the electrical circuit.

[0007] In some embodiments, performing correlation analysis on the spatiotemporal features of each variable in the multivariate time series data to obtain the spatiotemporal correlation features of the multivariate time series data specifically includes: The multivariate time series data is segmented along the time dimension to obtain multiple multivariate time series data segments; Extract the spatiotemporal features of each variable from each multivariate time series data segment; The spatiotemporal correlation characteristics of the multivariate time series data are determined based on the correlation between the spatiotemporal characteristics of each variable data.

[0008] In some embodiments, constructing a directed correlation graph representing the correlation between the various variable data of the electrical line based on the spatiotemporal correlation features, using each variable data in the multivariate time series data as a node, specifically includes: Each variable in the multivariate time series data is used as a node; The association status between each node is determined based on the spatiotemporal association characteristics. Based on all the associated states and each node, a directed association graph is constructed to represent the relationship between the variable data of the electrical circuit.

[0009] In some embodiments, determining the impact characteristics of each node on abnormal overheating of electrical lines specifically includes: Obtain historical multivariate time-series data of the normal operating status of the electrical circuit; Based on the historical multivariate time series data, a single variable anomaly analysis is performed on each node to determine the impact characteristics of each node on the electrical circuit overheating anomaly.

[0010] In some embodiments, the historical multivariate time-series data of the electrical line under normal operating conditions and the directed correlation graph are fused based on all the influence features to obtain a modal feature vector characterizing the operating mode of the electrical line, specifically including: Historical time-series feature vectors are extracted from historical multivariate time-series data under normal operating conditions of the electrical lines. Extract the topological feature vector from the directed association graph; Based on all the influence features, the historical time series feature vector and the topology feature vector are fused to obtain the modal feature vector characterizing the operation mode of the electrical line.

[0011] In some embodiments, evaluating the overheating anomaly state of the electrical circuit using the modal feature vector and the multivariate time-series data to obtain the overheating anomaly score of the electrical circuit specifically includes: Extract time-series feature vectors from the multivariate time-series data; Determine the difference features between the modal feature vector and the temporal feature vector; Construct a deep learning evaluation model for the overheating anomaly of the electrical circuit; The differential features are input into the deep learning evaluation model, which outputs the overheating anomaly score of the electrical circuit.

[0012] In some embodiments, the time-series feature vector includes time-domain statistical features and frequency-domain features.

[0013] In some embodiments, identifying overheating anomalies in the electrical circuit based on the overheating anomaly score specifically includes: Determine the overheating anomaly score threshold; The overheating anomaly score is compared with the overheating anomaly score threshold. If the overheating anomaly score is greater than or equal to the overheating anomaly score threshold, it is determined that the electrical circuit has an overheating anomaly, and an anomaly warning message is issued. If the overheating anomaly score is less than the overheating anomaly score threshold, the electrical circuit is determined to be normal, and a continued monitoring message is issued.

[0014] In some embodiments, the abnormal warning information includes the overheating abnormality score that triggered the warning, the corresponding abnormality level, the key abnormal variable that triggered the warning, the time and specific location of the line where the warning occurred, and preliminary handling suggestions.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The deep learning-based electrical circuit overheating anomaly identification system provided in this application collects multivariate time-series data through a data acquisition module, providing a comprehensive data foundation for capturing dynamic correlations between variables and avoiding the defect of missing correlation information in single-variable data. The graph construction module constructs a directed correlation graph through spatiotemporal feature correlation analysis, transforming the abstract multivariate dynamic correlation into a concrete graph structure, accurately depicting the time-varying correlation between variables. The feature fusion module combines node influence features with historical data and the correlation graph to generate modal feature vectors that fully represent the normal operation mode under multivariate dynamic correlation, solving the pain point of difficulty in modeling dynamic correlation modes. The anomaly identification module evaluates anomaly scores based on modal feature vectors and real-time data, realizing the quantitative identification of overheating anomalies under complex correlations, significantly improving the accuracy and reliability of anomaly identification, and effectively avoiding the misjudgment and missed judgment problems caused by neglecting variable correlations in traditional methods. By adopting the above scheme, overheating anomalies in electrical circuits can be identified based on the complex dynamic correlations between multiple variables in the electrical circuit. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a module structure diagram of the electrical circuit overheating anomaly identification system based on deep learning provided in this application; Figure 2 This is an exemplary flowchart for determining spatiotemporal correlation features according to the present application; Figure 3 This is an exemplary flowchart for determining an overheating anomaly score according to the present application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0019] To better understand the above technical solutions, a detailed description of the technical solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. (Refer to...) Figure 1As shown in the figure, this is a module structure diagram of an electrical circuit overheating anomaly identification system based on deep learning according to this embodiment of the application. The system includes: a data acquisition module 100, a graph construction module 200, a feature fusion module 300, and an anomaly identification module 400, which are described below: Data acquisition module 100 is used to acquire multivariable time-series data during the operation of electrical circuits; It should be noted that the multivariate time-series data in this application includes the line current data, line voltage data, line surface temperature data, ambient temperature data, and line equivalent resistance data of the electrical circuit. The multivariate time-series data reflects the dynamic change law of each individual operating parameter over time, and also contains the synergistic correlation characteristics between different parameters. The multivariate time-series data characterizes the comprehensive time-series evolution process of the operating state of the electrical circuit.

[0020] In practice, appropriate sensing devices are selected based on the laying path of the electrical lines, the distribution of key nodes, and the monitoring requirements of operating parameters. Line current data is collected using high-precision current transformers, line voltage data is obtained through voltage sensors, line surface temperature data is collected using contact temperature sensors or infrared temperature sensors, and ambient temperature data is collected by temperature and humidity sensors placed at key locations around the electrical cabinet and lines. The equivalent resistance of the lines is calculated using the collected current and voltage data combined with Ohm's law. Subsequently, each sensor is activated to synchronously collect data according to a preset sampling frequency. Simultaneously, the collected raw data undergoes real-time preprocessing, including removing outliers caused by sensor malfunctions, supplementing missing data through interpolation, and standardizing and unifying the dimensions of data from different dimensions. Ultimately, multivariate time-series data of the electrical lines during operation are obtained.

[0021] The graph construction module 200 is used to perform spatiotemporal feature correlation analysis on the spatiotemporal features of each variable data in the multivariate time series data to obtain the spatiotemporal correlation features of the multivariate time series data. Based on the spatiotemporal correlation features, a directed correlation graph representing the correlation relationship between the variable data of the electrical circuit is constructed using each variable data in the multivariate time series data as a node.

[0022] In some embodiments, reference Figure 2 As shown, this figure is an exemplary flowchart for determining spatiotemporal correlation features in some embodiments of this application. In this embodiment, the spatiotemporal correlation features of each variable in the multivariate time series data are analyzed for correlation to obtain the spatiotemporal correlation features of the multivariate time series data. This can be achieved by the following steps: In step S21, the multivariate time series data is segmented along the time dimension to obtain multiple multivariate time series data segments; In step S22, the spatiotemporal features of each variable data are extracted from each multivariate time series data segment; In step S23, the spatiotemporal correlation characteristics of the multivariate time series data are determined based on the correlation between the spatiotemporal characteristics of each variable data.

[0023] In specific implementation, the multivariate time series data can be segmented along the time dimension to obtain multiple multivariate time series data segments in the following way: using the sliding window method, combined with the actual change cycle of electrical circuit operating parameters, such as current and voltage usually fluctuating at the second level, and temperature gradually changing at the minute level, a fixed window duration (e.g., 5-10 seconds) and a sliding step size (e.g., 2-3 seconds) are set, and the continuous multivariate time series data is truncated according to the sliding step size in units of window duration, to ensure that the data within each window duration can completely reflect the parameter change trend in a short period of time, and finally obtain multiple multivariate time series data segments that do not overlap or have a small overlap. Other segmentation methods can also be used in other embodiments, which are not limited here.

[0024] Furthermore, in practical implementation, extracting the spatiotemporal features of each variable from various multivariate time series data segments can be achieved in the following way: Spatiotemporal features include temporal and spatial features. Temporal feature extraction employs basic statistical analysis and Fourier transform. For the single variable data within each multivariate time series data segment, time-domain statistics are calculated: the mean reflects the average operating level, the variance reflects the fluctuation amplitude, and the peak reflects extreme values. The time-domain data of the multivariate time series data segment is converted to frequency-domain data using a fast Fourier transform, extracting frequency-domain features such as the dominant frequency and the proportion of spectral energy to reflect the periodic variation of parameters. The obtained values ​​are used as temporal features; spatial feature extraction is based on the collection location information of each parameter, using the Pearson correlation coefficient method to calculate the correlation of the same variable data at different collection points, judge the consistency of the spatial distribution of variable data, and at the same time calculate the synchronicity of the changes of different parameters at the same collection point, initially capturing the correlation clues in the spatial dimension, and using the above-mentioned correlation results as spatial features, finally obtaining the spatiotemporal features of each variable data. Among them, the spatiotemporal features represent the characteristics of each variable data in the electrical circuit in time and space. Other methods can be used for extraction in other embodiments, which are not limited here.

[0025] In addition, in specific implementation, the spatiotemporal correlation characteristics of the multivariate time series data can be determined based on the correlation between the spatiotemporal characteristics of each variable data in the following way: Calculate the linear correlation strength between different spatiotemporal characteristics. For example, first use cross-correlation analysis to calculate the cross-correlation coefficient between the peak current and the average temperature. If the absolute value of the coefficient is greater than 0.7, it is determined to be a strong correlation, and the time lag of the correlation is recorded. The average temperature reaches its maximum value 3 seconds after the peak current appears. Then, further verify the causal relationship of the correlation. For example, verify it through Granger causality test to check whether the increase in current is a Granger cause of the increase in temperature. If the p-value of the test result is less than 0.05, it is determined that there is a one-way causal correlation between the two. Finally, quantify and integrate the correlation strength, correlation direction, and time lag information to form the spatiotemporal correlation characteristics of the multivariate time series data. For example, the spatiotemporal correlation characteristics are that the peak current of the line and the average surface temperature are strongly positively correlated, with a correlation strength of 0.82. The change in the peak current leads the change in the average temperature by 4 seconds, and the increase in current is a Granger cause of the increase in temperature. Other methods can be used to determine this in other embodiments, which are not limited here.

[0026] It should be noted that the spatiotemporal correlation features in this application represent the correlation features between the spatiotemporal features of individual variable data on electrical lines, which can be used to analyze the overheating abnormal state of electrical lines.

[0027] In some embodiments, constructing a directed correlation graph representing the correlation between the various variable data of the electrical circuit based on the spatiotemporal correlation features, using each variable data in the multivariate time series data as a node, can be achieved by the following steps: Each variable in the multivariate time series data is used as a node; The association status between each node is determined based on the spatiotemporal association characteristics. Based on all the associated states and each node, a directed association graph is constructed to represent the relationship between the variable data of the electrical circuit.

[0028] In specific implementation, the determination of the association state between each node based on the spatiotemporal association characteristics can be achieved in the following way: Based on the spatiotemporal association characteristics, first set an association strength threshold, such as a cross-correlation coefficient ≥ 0.6 or a Granger causality test p-value < 0.05, and filter out variable data pairs that meet the threshold conditions as potential association node pairs; for potential association node pairs, determine the direction of the directed edge between each node (i.e., from the current node to the temperature node) according to the causal direction in the spatiotemporal association characteristics; at the same time, use the association strength and time lag as the attribute parameters of the edge, and use the above-mentioned direction and attribute parameters between the nodes as the association state between the corresponding nodes. Here, the association state represents the association state between each node. In other embodiments, other methods can be used to determine it, which are not limited here.

[0029] In addition, in specific implementation, the directed association graph representing the relationship between the variables of the electrical circuit can be constructed based on all the associated states and each node in the following way: the adjacency matrix commonly used in graph theory is used as the storage structure, with the matrix rows and columns corresponding to each node, and the matrix element values ​​representing the association state between nodes. If two nodes are associated, the element value is set as the association strength, with positive numbers indicating positive association and negative numbers indicating negative association. The direction is reflected by the row and column index order of the elements. For example, a non-zero element in the i-th row and j-th column indicates the existence of a directed edge from node i to node j. At the same time, the time lag of each edge is recorded through an additional attribute table, and finally a complete directed association graph containing the set of nodes, the set of directed edges, and edge attributes is formed. Other methods can be used to construct it in other embodiments, which are not limited here.

[0030] It should be noted that the directed correlation graph in this application represents the correlation direction, strength and dynamic lag relationship between the data of each variable, and can be used to perform dynamic correlation analysis on the overheating state of electrical circuits by each variable data.

[0031] The feature fusion module 300 is used to determine the influence characteristics of each node on the overheating anomaly of the electrical line. Based on all the influence characteristics, the historical multivariate time series data of the electrical line under normal operation is fused with the directed correlation graph to obtain a modal feature vector characterizing the operating mode of the electrical line.

[0032] In some embodiments, determining the impact characteristics of each node on electrical circuit overheating anomalies can be achieved using the following steps: Obtain historical multivariate time-series data of the normal operating status of the electrical circuit; Based on the historical multivariate time series data, a single variable anomaly analysis is performed on each node to determine the impact characteristics of each node on the electrical circuit overheating anomaly.

[0033] In specific implementation, firstly, historical multivariate time-series data of electrical lines marked as operating normally are extracted from the database. Then, the controlled variable method is used: for each node, only the variable is abnormally disturbed in the historical normal data to adjust its value to outside the normal range, while keeping other variables within the normal range. Subsequently, the characteristics of the line surface temperature change under the abnormal state of the variable are calculated, including the difference between the temperature peak and the normal state (ΔT), the duration of the temperature exceeding the safety threshold (t), the rate of temperature rise (ΔT / Δt), and other indicators. The significance of these temperature change indicators compared with the normal state is verified by t-test, and the influence degree of the node is quantified based on the weighted sum of ΔT, t, and Δt, with the weights set according to the degree of harm caused by temperature exceeding the standard in the project. Finally, the influence degree of each node is standardized to the influence coefficient in the range [0,1] as the influence characteristic of the node on the overheating anomaly. Other methods can be used to determine this in other embodiments, which are not limited here.

[0034] It should be noted that the influence characteristics in this application represent the degree of influence of each node's corresponding variable on the electrical circuit overheating anomaly, and can be used to analyze the abnormal state of the electrical circuit.

[0035] In some embodiments, the modal feature vector characterizing the operating mode of the electrical line can be obtained by fusing historical multivariate time-series data under normal operating conditions with the directed correlation graph based on all influence features using the following steps: Historical time-series feature vectors are extracted from historical multivariate time-series data under normal operating conditions of the electrical lines. Extract the topological feature vector from the directed association graph; Based on all the influence features, the historical time series feature vector and the topology feature vector are fused to obtain the modal feature vector characterizing the operation mode of the electrical line.

[0036] In specific implementation, the historical time-series feature vector extracted from the historical multivariate time-series data under normal operating conditions of the electrical line can be achieved in the following way: the historical multivariate time-series data under normal operating conditions of the electrical line is divided into fixed time windows, and the time-domain statistical features (i.e., mean, variance, maximum, minimum, and rate of change) and frequency-domain features (i.e., the dominant frequency and spectral energy distribution obtained through Fourier transform) of each variable data in each window are calculated. The features of all variables in the same window are concatenated in the order of variables to form an initial time-series feature vector. Then, principal component analysis is used to reduce the dimensionality of the initial time-series feature vector, and the principal components with a cumulative variance contribution rate ≥95% are retained to obtain a historical time-series feature vector with uniform dimensionality.

[0037] In addition, in specific implementation, the topological feature vector extracted from the directed association graph can be achieved in the following way: Based on the directed association graph, calculate the basic topological indicators of each node. The number of edges pointing to the node and the number of edges originating from the node reflect the association activity of the node. The percentage of times the node is located on the shortest path of other nodes reflects the node's hub role in the association network. The average association strength of the edges connected to the node reflects the degree of association of the node. Concatenate the above indicators of all nodes in node order to form the initial topological feature vector. Similarly, perform dimensionality reduction through principal component analysis and retain the cumulative variance contribution rate ≥90% to obtain the topological feature vector.

[0038] In addition, in specific implementation, the fusion of the historical time-series feature vector and the topology feature vector based on all the influence features to obtain the modal feature vector representing the operation mode of the electrical line can be achieved in the following way: using the influence features of each node as weights, the feature components of each corresponding node in the historical time-series feature vector are multiplied by the weight of that node; the feature components of each corresponding node in the topology feature vector are also multiplied by the weight of that node; the two weighted vectors are added element by element to obtain the fused feature vector; and then the elements of the fused feature vector are mapped to the [0,1] interval through maximum-minimum standardization, finally forming the modal feature vector that can represent the normal operation mode of the electrical line.

[0039] It should be noted that the historical time-series feature vector in this application represents the vector of time-domain statistical characteristics and frequency-domain characteristics of each historical variable data, which reflects the dynamic change law of each single variable over time when the line is operating normally; the topology feature vector represents the vector of node topology attributes and edge attributes, which reflects the correlation structure characteristics between each variable data when the line is operating normally; the modal feature vector represents the vector of the coordinated characteristics of time dynamic changes and variable correlation structure when the line is operating normally, which can reflect both the time-series fluctuations of each variable and the correlation and interaction between variables, and characterize the typical mode of stable operation of the electrical line.

[0040] The anomaly identification module 400 is used to evaluate the overheating anomaly state of the electrical circuit through the modal feature vector and the multivariate time series data, obtain the overheating anomaly score of the electrical circuit, and identify the overheating anomaly of the electrical circuit based on the overheating anomaly score.

[0041] In some embodiments, reference Figure 3 As shown, this figure is an exemplary flowchart for determining the overheating anomaly score in some embodiments of this application. In this embodiment, the overheating anomaly state of the electrical circuit is evaluated using the modal feature vector and the multivariate time series data. The overheating anomaly score of the electrical circuit can be obtained by the following steps: In step S41, a time series feature vector is extracted from the multivariate time series data; In step S42, the difference features between the modal feature vector and the temporal feature vector are determined; In step S43, a deep learning evaluation model for the overheating anomaly of the electrical circuit is constructed; In step S44, the difference features are input into the deep learning evaluation model, and the overheating anomaly score of the electrical circuit is output.

[0042] In specific implementation, the extraction of time-series feature vectors from the multivariate time-series data can be achieved in the following way: adopting the same process as historical time-series feature extraction, the real-time acquired multivariate time-series data is divided into a fixed window of 5-10 seconds and a sliding step of 2-3 seconds. For each variable in each window, the mean, variance, peak value and other time-domain statistical features are calculated. Frequency-domain features such as the main frequency and the proportion of spectral energy are extracted by fast Fourier transform. After concatenating all features in the order of variables into an initial vector, principal component analysis is used to retain the principal components with a cumulative variance contribution rate of ≥95%. Then, the elements of the initial vector are mapped to the [0,1] interval by maximum-minimum standardization to obtain a time-series feature vector with the same dimension as the modal feature vector.

[0043] In addition, in specific implementation, the difference between the modal feature vector and the temporal feature vector can be determined in the following way: calculate the three core differences between the modal feature vector and the temporal feature vector: use Euclidean distance to quantify the overall numerical deviation of the vector, use cosine similarity to measure the fit of the vector direction, and use Kullback-Leibler divergence to evaluate the difference in the feature distribution of each dimension of the vector; map the three types of values ​​to the [0,1] interval through min-max normalization, and calculate the difference according to Euclidean distance (weight 0.4) + cosine similarity (weight 0.3) + K The L-divergence (weight 0.3) is proportionally weighted and concatenated to form a difference feature vector with a fixed dimension. From the perspective of domain characteristics, Euclidean distance directly quantifies the degree of numerical deviation between feature vectors. Overheating of electrical circuits is often directly related to the absolute value of key parameters exceeding the standard, and its degree of deviation is more indicative of anomalies. Therefore, it is initially assigned a higher weight based on engineering experience. Cosine similarity reflects the consistency of vector direction, and KL divergence measures the difference in feature distribution. Both reflect anomalies from the perspective of pattern matching. However, for faults such as overheating that are strongly correlated with numerical values, their influence is less than that of Euclidean distance. Therefore, they are initially assigned a lower weight.

[0044] In addition, in specific implementation, the deep learning evaluation model for electrical circuit overheating anomaly can be constructed in the following way: A mature convolutional neural network architecture used in engineering is selected, the input layer dimension matches the dimension of the differential feature vector, three convolutional layers are set sequentially to extract local correlation information of the differential features, a max pooling layer is connected after each convolutional layer to reduce the dimension, and two fully connected layers are set for feature mapping. The output layer uses the Sigmoid activation function to output 0-1 probability values. The training data is constructed using normal historical data and simulated abnormal data. Normal data consists of differential features corresponding to labeled normal time-series data, and abnormal data is generated by adding different degrees of perturbation to the normal data. The data is divided into training set, validation set, and test set in a 7:2:1 ratio. The cross-entropy loss function and Adam optimizer are used for training with mini-batch gradient descent. The iteration stops when the validation set loss does not decrease for five consecutive rounds. The model is optimized by adjusting parameters such as the number of convolutional kernels and the number of neurons in the fully connected layer using the test set to obtain the deep learning evaluation model for electrical circuit overheating anomaly.

[0045] In addition, in specific implementation, the differential feature vector is input into the deep learning evaluation model, and the output of the overheating anomaly score of the electrical circuit can be achieved in the following way: the differential feature vector is input into the trained model, the output 0-1 probability value is obtained, and then mapped to the overheating anomaly score of 0-100 according to the rule that score = probability value × 100. The higher the score, the more serious the deviation of the current operating state from the normal mode, and the higher the risk of overheating anomaly.

[0046] It should be noted that the time-series feature vector in this application represents the vector of time-domain statistical characteristics and frequency-domain characteristics of each variable in multivariate time-series data, reflecting the dynamic change law of each variable over time under the current operating state of the electrical line; the difference feature represents the overall deviation between the current time-series feature vector and the modal feature vector representing the normal operating mode; the deep learning evaluation model represents the model of quantitative evaluation capability of the degree of deviation of the line operating state from the normal mode; the overheating anomaly score represents the risk level of overheating anomaly in the electrical line, which can be used to judge the overheating anomaly of the electrical line.

[0047] In some embodiments, identifying overheating anomalies in the electrical circuit based on the overheating anomaly score can be achieved using the following steps: Determine the overheating anomaly score threshold; The overheating anomaly score is compared with the overheating anomaly score threshold. If the overheating anomaly score is greater than or equal to the overheating anomaly score threshold, it is determined that the electrical circuit has an overheating anomaly, and an anomaly warning message is issued. If the overheating anomaly score is less than the overheating anomaly score threshold, the electrical circuit is determined to be normal, and a continued monitoring message is issued.

[0048] It should be noted that the abnormal warning information in this application includes the overheating anomaly score that triggered the warning, the corresponding anomaly level, the key anomaly variables that triggered the warning, the time and specific location of the line where the warning occurred, and preliminary handling suggestions. It reflects that the current operating status of the electrical line has significantly deviated from the normal mode, there is a clear risk of overheating and it may further develop into a fault, indicating that immediate intervention measures are required to avoid safety problems such as line burnout and short circuits. The continued monitoring information includes the current overheating anomaly score, the real-time values ​​of each key operating variable, the recent fluctuation trend of the score, and whether there are variables approaching the threshold. It reflects that the current operating status of the electrical line is in line with the normal mode, but there are potential minor fluctuations or trend changes, which do not require emergency intervention at present. It indicates that continuous tracking and monitoring are required to prevent subsequent fluctuations from accumulating and causing anomalies, and to ensure the long-term stable operation of the line.

[0049] Additionally, it should be noted that the overheating anomaly scoring threshold can be set according to the requirements of electrical circuit overheating anomalies. For example, if the electrical circuit is a general civil lighting circuit, the overheating anomaly scoring threshold can be set in a high range. Such circuits have small load fluctuations and low overheating risk, and a high threshold can reduce unnecessary false alarms. If the water body to be tested is a power supply line for critical industrial production equipment, the overheating anomaly scoring threshold can be set in a low range. Once such a line overheats, it will directly lead to production interruption and equipment damage. A low threshold can more sensitively capture early anomalies. In other embodiments, for example, when the water body to be tested is a fire emergency power supply line for a specific high-rise building, the overheating anomaly scoring threshold can be set in a high range. Because fire lines usually have extremely low loads and are in standby mode for a long time, a high threshold can avoid false alarms caused by slight parameter fluctuations due to long-term idleness of the line, while ensuring accurate identification when there is real overheating, thereby improving the detection accuracy of electrical circuit overheating anomalies.

[0050] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0052] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A deep learning-based electrical circuit overheating anomaly identification system, characterized in that, The system includes: The data acquisition module is used to collect multivariable time-series data during the operation of electrical circuits; The graph construction module is used to perform correlation analysis on the spatiotemporal features of each variable data in the multivariate time series data to obtain the spatiotemporal correlation features of the multivariate time series data. Based on the spatiotemporal correlation features, a directed correlation graph representing the correlation between the variable data of the electrical line is constructed using each variable data in the multivariate time series data as a node. The feature fusion module is used to determine the impact characteristics of each node on the overheating anomaly of the electrical line. Based on all the impact characteristics, the historical multivariate time series data of the electrical line under normal operation is fused with the directed correlation graph to obtain the modal feature vector characterizing the operating mode of the electrical line. An anomaly identification module is used to evaluate the overheating anomaly state of the electrical circuit using the modal feature vector and the multivariate time series data, obtain the overheating anomaly score of the electrical circuit, and identify the overheating anomaly of the electrical circuit based on the overheating anomaly score.

2. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 1, characterized in that, The multivariate time-series data includes the line current data, line voltage data, line surface temperature data, ambient temperature data, and line equivalent resistance data of the electrical circuit.

3. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 1, characterized in that, The spatiotemporal correlation characteristics of the multivariate time series data are analyzed to obtain the following specific spatiotemporal correlation characteristics: The multivariate time series data is segmented along the time dimension to obtain multiple multivariate time series data segments; Extract the spatiotemporal features of each variable from each multivariate time series data segment; The spatiotemporal correlation characteristics of the multivariate time series data are determined based on the correlation between the spatiotemporal characteristics of each variable data.

4. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 1, characterized in that, Using each variable in the multivariate time-series data as a node, constructing a directed correlation graph representing the correlation between the various variable data of the electrical line based on the spatiotemporal correlation features specifically includes: Each variable in the multivariate time series data is used as a node; The association status between each node is determined based on the spatiotemporal association characteristics. Based on all the associated states and each node, a directed association graph is constructed to represent the relationship between the variable data of the electrical circuit.

5. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 1, characterized in that, Determining the impact characteristics of each node on abnormal overheating of electrical circuits specifically includes: Obtain historical multivariate time-series data of the normal operating status of the electrical circuit; Based on the historical multivariate time series data, a single variable anomaly analysis is performed on each node to determine the impact characteristics of each node on the electrical circuit overheating anomaly.

6. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 1, characterized in that, Based on all the influencing features, the historical multivariate time-series data of the electrical line under normal operating conditions are fused with the directed correlation graph to obtain the modal feature vector characterizing the operating mode of the electrical line, specifically including: Historical time-series feature vectors are extracted from historical multivariate time-series data under normal operating conditions of the electrical lines. Extract the topological feature vector from the directed association graph; Based on all the influence features, the historical time series feature vector and the topology feature vector are fused to obtain the modal feature vector characterizing the operation mode of the electrical line.

7. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 1, characterized in that, The overheating anomaly state of the electrical circuit is evaluated using the modal feature vector and the multivariate time-series data to obtain the overheating anomaly score of the electrical circuit, specifically including: Extract time-series feature vectors from the multivariate time-series data; Determine the difference features between the modal feature vector and the temporal feature vector; Construct a deep learning evaluation model for the overheating anomaly of the electrical circuit; The differential features are input into the deep learning evaluation model, which outputs the overheating anomaly score of the electrical circuit.

8. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 7, characterized in that, The time-series feature vector includes time-domain statistical features and frequency-domain features.

9. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 1, characterized in that, The identification of overheating anomalies in the electrical circuit based on the overheating anomaly score specifically includes: Determine the overheating anomaly score threshold; The overheating anomaly score is compared with the overheating anomaly score threshold. If the overheating anomaly score is greater than or equal to the overheating anomaly score threshold, it is determined that the electrical circuit has an overheating anomaly, and an anomaly warning message is issued. If the overheating anomaly score is less than the overheating anomaly score threshold, the electrical circuit is determined to be normal, and a continued monitoring message is issued.

10. The electrical circuit overheating anomaly identification system based on deep learning as described in claim 9, characterized in that, The abnormal warning information includes the overheating abnormality score that triggered the warning, the corresponding abnormality level, the key abnormal variable that triggered the warning, the time and specific location of the line where the warning occurred, and preliminary handling suggestions.

Images