Distribution line abnormal state identification method and device, electronic equipment and storage medium

Abstract

The invention discloses a distribution line abnormal state identification method and device, electronic equipment and a storage medium, and the method comprises the steps: receiving multi-source heterogeneous data collected by various types of sensors disposed in a distribution line through an edge calculation node, carrying out the preprocessing of the multi-source heterogeneous data, and constructing a feature map network; identifying whether the distribution line has an abnormal target node based on the feature map network; and if yes, extracting a local sub-graph of the target node from the feature graph network, identifying a fault type of the target node based on the local sub-graph, and pushing the fault type to a far-end center. Multi-source heterogeneous data are collected to construct feature map network identification anomalies, fault types are further identified, the method can adapt to dynamic changes of loads and operation modes of the power distribution network, the accuracy rate of anomaly identification is high, node data processing and anomaly identification are calculated at the edge, the load of a far-end center is reduced, and the reliability of the power distribution network is improved. And the delay of uploading the data to a far-end center is avoided, the abnormity identification speed can be improved, and the distribution line abnormity can be rapidly and timely identified.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid abnormality monitoring, and in particular to a method, device, electronic equipment and storage medium for identifying abnormal status of a distribution line. Background Art

[0002] In order to ensure the safe and reliable operation of the distribution network, it is necessary to collect monitoring data on the lines and distribution equipment in the distribution network, and identify whether the distribution lines are abnormal through the monitoring data.

[0003] In the prior art, single monitoring data is mainly collected and uploaded to a remote center, where the monitoring data is compared with a fixed threshold to determine whether the distribution line is abnormal. On the one hand, the single monitoring data is compared with the fixed threshold to identify abnormalities, which makes it difficult to adapt to the dynamic changes in the load and operation mode of the distribution network, and is prone to false alarms or missed alarms. On the other hand, the monitoring data is concentrated in the remote center, which leads to problems such as data transmission delay, bandwidth pressure and high load in the remote center. Summary of the invention

[0004] The present invention provides a method, device, electronic device and storage medium for identifying abnormal status of a distribution line, so as to solve the problems of abnormal false alarm, missed alarm and high remote center load when comparing monitoring data of an existing abnormal identification unit with a fixed threshold.

[0005] In a first aspect, the present invention provides a method for identifying abnormal status of a distribution line, which is applied to an edge computing node, wherein the edge computing node is respectively connected to a sensor and a remote center, and comprises:

[0006] Receive multi-source heterogeneous data collected by various types of sensors deployed in distribution lines;

[0007] Preprocessing the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data;

[0008] Constructing a feature graph network based on the multi-source heterogeneous data;

[0009] Identify whether there is an abnormal target node in the power distribution line based on the characteristic graph network;

[0010] If so, extracting a local subgraph of the target node from the feature graph network;

[0011] Identifying a fault type of the target node based on the local subgraph;

[0012] The fault type of the target node is pushed to the remote center.

[0013] Optionally, before receiving multi-source heterogeneous data collected by various types of sensors deployed in the power distribution line, the method further includes:

[0014] Time synchronization instructions and data acquisition parameters are sent to various types of sensors in the power distribution line, and the sensors are used to perform time synchronization and set data acquisition frequency when receiving the time synchronization instructions and data acquisition parameters.

[0015] Optionally, the multi-source heterogeneous data includes at least an electrical signal, a temperature signal, a vibration signal, and an audio signal, and preprocessing the multi-source heterogeneous data to obtain the preprocessed multi-source heterogeneous data includes:

[0016] Preprocessing the electrical signal, the temperature signal, the vibration signal and the audio signal respectively to obtain preprocessed electrical signal, temperature signal, vibration signal and audio signal, wherein the preprocessing includes at least one of the following:

[0017] Filtering to reduce noise, remove outliers, supplement missing values ​​and perform standardization.

[0018] Optionally, constructing a feature graph network based on the multi-source heterogeneous data includes:

[0019] Extracting features from the preprocessed electrical signal, temperature signal, vibration signal and audio signal respectively to obtain electrical features, temperature features, vibration features and audio features;

[0020] Selecting the electrical characteristics, temperature characteristics, vibration characteristics and audio characteristics to obtain target electrical characteristics, target temperature characteristics, target vibration characteristics and target audio characteristics;

[0021] The target electrical characteristics, target temperature characteristics, target vibration characteristics and target audio characteristics are used as nodes, and the correlation between two nodes is determined based on the graph attention network to form an edge to obtain a feature graph network.

[0022] Optionally, the correlation between two nodes is determined based on the graph attention network to form an edge to obtain a feature graph network, including:

[0023] The attention coefficient is calculated by the following formula:

[0024]

[0025] α is the attention vector, h i 、h j are the features of nodes i and j respectively, W is the weight matrix, N i represents the set of adjacent nodes of a node, k represents the number of nodes in the set of adjacent nodes, the symbol | represents feature concatenation, and LeakyReLU is the activation function;

[0026] The output of the graph attention network is:

[0027]

[0028] Optionally, identifying whether there is an abnormal target node in the power distribution line based on the characteristic graph network includes:

[0029] The reach-dist of the neighboring node o of node p in the feature graph network is calculated by the following preset formula: k (p, o):

[0030] reach-dist k (p, o) = max {k-distance (o), d (p, o)};

[0031] k-distance(o) is the k-neighbor distance of the k-th neighbor node o of node p, and d(p,o) is the Euclidean distance between node P and node o;

[0032] Calculate the local reachability density lrd of node P k (p):

[0033]

[0034] N k (p) represents the incoming set of node p;

[0035] Calculate the anomaly score LOF of node p k (p):

[0036]

[0037] Nodes with anomaly scores greater than a preset threshold are identified as abnormal target nodes.

[0038] Optionally, identifying the fault type of the target node based on the local subgraph includes:

[0039] Constructing a fault feature vector based on the local subgraph;

[0040] Calculating the cosine similarity between the fault feature vector and the feature vector in the fault type template library;

[0041] The N fault types with the highest cosine similarity are determined as the fault types of the target node.

[0042] In a second aspect, the present invention provides a distribution line abnormal state identification device, which is applied to an edge computing node, wherein the edge computing node is respectively connected to a sensor and a remote center, and includes:

[0043] A data acquisition module, used to receive multi-source heterogeneous data collected by various types of sensors deployed in the power distribution line;

[0044] A data preprocessing module, used to preprocess the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data;

[0045] A feature graph network construction module, used to construct a feature graph network based on the multi-source heterogeneous data;

[0046] An abnormal node identification module is used to identify whether there is an abnormal target node in the distribution line based on the characteristic graph network, and if so, execute a local subgraph extraction module;

[0047] A local subgraph extraction module, used to extract the local subgraph of the target node from the feature graph network;

[0048] A fault type identification module, used to identify the fault type of the target node based on the local subgraph;

[0049] A push module is used to push the fault type of the target node to the remote center.

[0050] In a third aspect, the present invention provides an electronic device, the electronic device comprising:

[0051] at least one processor; and

[0052] a memory communicatively connected to the at least one processor; wherein,

[0053] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for identifying abnormal status of a distribution line as described in any one of the first aspects of the present invention.

[0054] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a processor to implement the method for identifying abnormal status of a distribution line as described in any one of the first aspect of the present invention when executed.

[0055] In the embodiment of the present invention, an edge computing node is set to receive multi-source heterogeneous data collected by various types of sensors deployed in the distribution line, and a feature graph network is constructed based on the preprocessed multi-source heterogeneous data. After the abnormal target node is identified based on the feature graph network, a local subgraph of the target node is extracted from the feature graph network, and the fault type of the target node is identified based on the local subgraph and pushed to the remote center. On the one hand, the fault type is further identified by collecting multi-source heterogeneous data to construct a feature graph network to identify the anomaly. Compared with using a single monitoring data to compare with a fixed threshold to identify the anomaly, it can adapt to the dynamic changes of the load and operation mode of the distribution network, and has a high accuracy in identifying the anomaly. On the other hand, the multi-source heterogeneous data is preprocessed and anomaly identified at the edge computing node without uploading to the remote center, which not only reduces the load of the remote center, but also avoids the delay in uploading data to the remote center, which can improve the speed of anomaly identification and can quickly and timely identify the anomaly of the distribution line.

[0056] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0058] Figure 1 This is a flow chart of a method for identifying abnormal status of a power distribution line provided in Embodiment 1 of the present invention;

[0059] Figure 2 This is a flow chart of a method for identifying abnormal status of a power distribution line provided in Embodiment 2 of the present invention;

[0060] Figure 3 It is a structural schematic diagram of a device for identifying abnormal state of a power distribution line provided in Embodiment 3 of the present invention;

[0061] Figure 4 It is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0063] Embodiment 1

[0064] Figure 1 This is a flow chart of a method for identifying abnormal state of a distribution line provided in the first embodiment of the present invention. This embodiment is applicable to the case of identifying abnormal state of a distribution line. The method can be executed by a device for identifying abnormal state of a distribution line. The device for identifying abnormal state of a distribution line can be implemented in the form of hardware and / or software and can be configured in an electronic device, such as an edge computing node. Figure 1 As shown, the method for identifying abnormal state of distribution line includes:

[0065] S101. Receive multi-source heterogeneous data collected by various types of sensors deployed in a power distribution line.

[0066] The system architecture of this embodiment may include sensors, edge computing nodes, and remote centers. The edge computing nodes, sensors, and remote centers may be connected through the Internet of Things. Exemplarily, the network topology of the Internet of Things adopts a hybrid type, including star and tree structures, to improve the reliability and coverage of the network. The sensors around the substation use a star structure to communicate directly with the data aggregation node, while the sensors far away from the substation form a tree structure to transmit data through a multi-hop method. Factors such as communication distance, power consumption, and anti-interference ability need to be considered during network construction to ensure the stability and real-time performance of data transmission. Network simulation software such as NS-3 or OPNET can be used to optimize network performance, such as by adjusting transmission power, routing protocols, and network topology to improve the overall performance of the network.

[0067] For network transmission, ZigBee technology can be selected for short-distance (within 100 meters) sensor communication, using the 2.4GHz frequency band and a data transmission rate of 250kbps. For long-distance communication, LoRa technology is adopted, using the 433MHz frequency band, which can achieve a communication distance of several kilometers.

[0068] In this embodiment, sensors can be deployed at key points of the distribution line to form a monitoring network. The sensors may include voltage transformers, current transformers, temperature sensors, vibration sensors, and noise sensors. By way of example, current transformers and voltage transformers can be arranged at locations such as transformers, switchgear, and line connection points of the distribution line to measure electrical data of current and voltage; temperature sensors can be installed on transformer casings and line towers to monitor temperature changes; vibration sensors and noise sensors can be installed around transformers and switchgear to detect abnormal vibration and noise of the transformers and switchgear.

[0069] In one embodiment, before receiving multi-source heterogeneous data, the edge computing node can also send time synchronization instructions and data acquisition parameters to various types of sensors in the distribution line, and the sensors are used to perform time synchronization and set the data acquisition frequency when receiving the time synchronization instructions and data acquisition parameters. Exemplarily, the network time protocol (NTP) or the precision time protocol (PTP) can be used to achieve time synchronization between sensors to ensure that data from different sensors have consistent timestamps. For data acquisition frequency, different types of sensors can be set with different acquisition frequencies. Exemplarily, the sampling frequency of current transformers and voltage transformers can be set to a higher acquisition frequency to capture rapidly changing electrical parameters. The sampling frequency of the temperature sensor can be set to once per minute, the sampling frequency of the vibration signal is 1kHz, and the sampling frequency of the sound signal is 44.1kHz.

[0070] S102: Preprocess the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data.

[0071] After receiving the multi-source heterogeneous data, the edge computing node can perform preprocessing to obtain preprocessed multi-source heterogeneous data. Specifically, the preprocessing may include at least one of filtering and denoising, removing outliers, supplementing missing values, and standardization.

[0072] S103. Construct a feature graph network based on the preprocessed multi-source heterogeneous data.

[0073] In this embodiment, the multi-source heterogeneous data may include electrical signals such as voltage and current, as well as temperature signals, vibration signals and audio signals. Feature extraction may be performed on signals of different modes first, and then the features of each mode may be selected to obtain the final features of each mode data. Then, a fusion model based on a graph attention network is used to fuse the final features of the multi-modal data to obtain a feature graph network. Each node in the feature graph network is a feature of a different mode, and the edges between the nodes represent the correlation between data of different modes. For example, the edge between the current feature node and the temperature feature node represents the correlation between current anomaly and temperature rise.

[0074] S104. Identify whether there are abnormal target nodes in the distribution line based on the feature graph network.

[0075] In one embodiment, a local outlier factor (LOF) algorithm may be used to calculate the score of each node in the feature graph network, and nodes with scores greater than a preset threshold may be determined as abnormal target nodes, such as target nodes with abnormal current or temperature.

[0076] In another embodiment, the feature graph network may also be input into a pre-trained anomaly recognition model, and the abnormal target nodes in the feature graph network may be identified by the anomaly recognition model.

[0077] If an abnormal target node is identified from the feature graph network, S105 may be executed. If no abnormal target node is identified from the feature graph network, the process may return to S101.

[0078] S105. Extract a local subgraph of the target node from the feature graph network.

[0079] Specifically, after an abnormal target node is identified in the feature graph network, several neighboring nodes of the target node or neighboring nodes within a preset range, or all nodes connected to the target node can be determined from the feature graph network to obtain a local subgraph of the target node.

[0080] S106: Identify the fault type of the target node based on the local subgraph.

[0081] In one embodiment, fault features may be extracted from the local subgraph, and the fault features may be matched with features in the fault type template to obtain the fault type of the target node as the fault type of the distribution line.

[0082] S107: Push the fault type of the target node to the remote center.

[0083] After the edge computing node determines that the distribution line is abnormal and determines the fault type, the fault type can be pushed to the remote center for alarm.

[0084] In the embodiment of the present invention, an edge computing node is set to receive multi-source heterogeneous data collected by various types of sensors deployed in the distribution line, and a feature graph network is constructed based on the preprocessed multi-source heterogeneous data. After the abnormal target node is identified based on the feature graph network, a local subgraph of the target node is extracted from the feature graph network, and the fault type of the target node is identified based on the local subgraph and pushed to the remote center. On the one hand, the fault type is further identified by collecting multi-source heterogeneous data to construct a feature graph network to identify the anomaly. Compared with using a single monitoring data to compare with a fixed threshold to identify the anomaly, it can adapt to the dynamic changes of the load and operation mode of the distribution network, and has a high accuracy in identifying the anomaly. On the other hand, the multi-source heterogeneous data is preprocessed and anomaly identified at the edge computing node without uploading to the remote center, which not only reduces the load of the remote center, but also avoids the delay in uploading data to the remote center, which can improve the speed of anomaly identification and can quickly and timely identify the anomaly of the distribution line.

[0085] Embodiment 2

[0086] Figure 2 This is a flow chart of a method for identifying abnormal status of a distribution line provided in the second embodiment of the present invention. The embodiment of the present invention is optimized on the basis of the above-mentioned first embodiment. Figure 2 As shown, the method for identifying abnormal state of distribution line includes:

[0087] S201. Receive multi-source heterogeneous data collected by various types of sensors deployed in a power distribution line, where the multi-source heterogeneous data includes electrical signals, temperature signals, vibration signals, and audio signals.

[0088] In this embodiment, current transformers and voltage transformers can be arranged at locations such as transformers, switchgear and line connection points of the distribution line to measure electrical data of current and voltage; temperature sensors can be installed on the transformer casing and line towers to monitor temperature changes; vibration sensors and noise sensors can be installed around the transformer and switchgear to detect abnormal vibration and noise of the transformer and switchgear.

[0089] The edge computing node can receive electrical signals, temperature signals, vibration signals, and audio signals from sensors in the power distribution lines, where the electrical signals may include voltage signals and current signals.

[0090] S202 , preprocessing the electrical signal, the temperature signal, the vibration signal and the audio signal respectively to obtain preprocessed electrical signal, temperature signal, vibration signal and audio signal.

[0091] In one embodiment, the preprocessing may include filtering and noise reduction, removing outliers, supplementing missing values, and standardization.

[0092] Specifically, for voltage and current signals, a Butterworth low-pass filter can be used to remove high-frequency noise, with the cutoff frequency set to 5kHz and the filter order set to 4. Then, the FFT algorithm can be used for harmonic analysis to calculate the amplitude of the fundamental wave and each harmonic. For temperature signals, a moving average filter can be used with a window length of 5 to smooth the temperature curve. For vibration signals, wavelet transform denoising can be used, with the db4 wavelet selected and the number of decomposition layers set to 3. For audio signals, a pre-emphasis filter can be used to enhance the high-frequency components. Then, a short-time Fourier transform can be performed with a frame length of 256 points and a frame shift of 128 points to obtain a time-frequency diagram for subsequent analysis. All collected data can be attached with a timestamp and sensor ID for subsequent processing and analysis.

[0093] Furthermore, statistical methods (such as the 3σ criterion) can be used to identify outliers, and data outside the normal range can be marked or deleted. Missing data can be filled by interpolation methods (such as linear interpolation or spline interpolation) and denoised by median filtering or wavelet transform.

[0094] Exemplarily, the 3σ criterion can be used to identify outliers for voltage and current signals, that is, data outside the range of ±3 times the standard deviation of the mean is marked as abnormal. For temperature signals, reasonable upper and lower limits (such as -20°C to 80°C) are set, and data outside this range is considered abnormal. For outlier detection of vibration signals and audio signals, the interquartile range (IQR) method can be used to mark data less than Q1-1.5IQR or greater than Q3+1.5IQR as abnormal, where Q1 and Q3 are the first and third quartiles, respectively. For missing data, linear interpolation is used to fill in voltage and current, forward filling is used (filling with the nearest valid value) for temperature, and spline interpolation is used for vibration and audio.

[0095] When performing standardization, you can use minimum-maximum standardization or Z-score standardization. The minimum-maximum standardization formula is:

[0096]

[0097] The Z-score normalization formula is:

[0098]

[0099] Where μ is the mean and σ is the standard deviation.

[0100] For example, parameters such as current and voltage can be standardized using the minimum-maximum method, and the voltage and current can be mapped to the [0,1] interval. Temperature, vibration RMS value, etc. can be standardized using the Z-score method to make their mean 0 and standard deviation 1. For categorized data collected from the SCADA system, such as switch status, one-hot encoding is used to convert it into a numerical type. During data cleaning and standardization, a backup of the original data is retained and all processing steps are recorded so that they can be traced back or reprocessed when necessary.

[0101] S203 , extracting features from the preprocessed electrical signal, temperature signal, vibration signal and audio signal respectively to obtain electrical features, temperature features, vibration features and audio features.

[0102] Specifically, the current, voltage, temperature, vibration, and audio signals are first synchronized in time. For example, a sliding time window method is used, with the window size set to 1 minute and the step length set to 10 seconds. For data with different sampling rates, such as high-frequency sampled voltage and current data and low-frequency sampled temperature, a linear interpolation method is used for alignment. Then different features can be extracted from data of different modes, as follows:

[0103] The following features are extracted from the voltage and current signals:

[0104] Statistical characteristics: mean, standard deviation, skewness, kurtosis, quantiles (25%, 50%, 75%);

[0105] Frequency domain characteristics: fundamental wave amplitude, total harmonic distortion (THD), amplitude of each harmonic (up to the 13th order), frequency band energy ratio (such as 0-100Hz, 100-1000Hz, and above 1000Hz);

[0106] Time-frequency characteristics: wavelet energy entropy, wavelet packet energy, intrinsic mode function (IMF) energy of empirical mode decomposition (EMD).

[0107] The features extracted from the temperature signal include: average temperature, maximum temperature, minimum temperature, temperature change rate (per hour), and temperature periodicity (calculated by autocorrelation function).

[0108] The features extracted from the vibration signal include: RMS value, peak factor, crest factor, margin factor, pulse factor, frequency band energy ratio (such as 0-10Hz, 10-100Hz, 100-1000Hz).

[0109] The features extracted from the audio signal include: A-weighted equivalent continuous sound level, maximum sound level, and duration ratio when the sound level exceeds the background by 10dB.

[0110] For SCADA system data, extract features such as switch operation frequency, load rate, power factor, etc.

[0111] S204, selecting electrical characteristics, temperature characteristics, vibration characteristics, and audio characteristics to obtain target electrical characteristics, target temperature characteristics, target vibration characteristics, and target audio characteristics.

[0112] Feature selection can first use ANOVA (Analysis of Variance) for preliminary screening, select features with p-values ​​less than 0.05, then use recursive feature elimination (RFE) combined with random forest to select the optimal feature subset, and finally apply PCA dimensionality reduction to retain the principal components required to explain 95% of the variance.

[0113] Specifically, firstly, a one-way analysis of variance (ANOVA) is performed on all extracted features. For example, each feature is used as an independent variable, and the system state (normal / abnormal) is used as a dependent variable, and the F value and the corresponding p value are calculated. The significance level α is set to 0.05, and features with a p value less than 0.05 are selected to obtain features that are significantly correlated with the system state.

[0114] Secondly, recursive feature elimination (RFE) is used in combination with random forest for further feature selection. For example, a random forest model is first constructed, and appropriate parameters such as the number of trees (such as 500 trees) and the maximum depth are set. In the RFE process, the least important features are removed in each iteration until the preset number of features is reached. The optimal number of features can be determined by cross-validation, such as using the cv parameter of RFE to determine the number of features, and finally the feature subset that achieves the highest classification accuracy is selected.

[0115] Finally, principal component analysis (PCA) is applied to the selected feature subset for dimensionality reduction. For example, the PCA class of sklearn can be used for implementation. Specifically, the n_components parameter is set to 'mle' (maximum likelihood estimation) or a floating point number less than 1 (such as 0.95). PCA will calculate the covariance matrix of the feature, perform eigenvalue decomposition, and then select the principal component with a cumulative explained variance of 95%, and project the original features onto the principal component to obtain the feature representation after dimensionality reduction. The final feature subset has both strong distinguishing ability and reduced redundancy, providing high-quality input for subsequent anomaly detection and fault diagnosis.

[0116] S205. Taking the target electrical characteristics, target temperature characteristics, target vibration characteristics and target audio characteristics as nodes, the correlation between two nodes is determined based on the graph attention network to form an edge to obtain a feature graph network.

[0117] Specifically, feature fusion adopts a multi-view learning method. First, the covariance matrix is ​​calculated for the features of any two modes. For example, for the current and voltage features, the covariance matrix is ​​calculated:

[0118]

[0119] Solve the eigenvalue equation ∑ν=λν, where ∑ is the covariance matrix mentioned above, and ν is the eigenvector. After the matrix ∑ acts on the vector ν, the direction of ν remains unchanged, so that the eigenvector ν satisfies the condition ∑ν=λν, and λ is the eigenvalue. Thus, the first k eigenvectors with a cumulative contribution rate of 95% are selected, and then DCCA (Dynamic Coefficient of Corelation) is used to learn the nonlinear correlation of different modal features, such as the relationship between voltage, current and temperature. The objective function of DCCA is to maximize the correlation after projection of different views. The specific objective function is as follows:

[0120]

[0121] Among them, ∑ xy is the covariance matrix of the two view features, W x and W y is the projection matrix, and the optimal projection matrix is ​​obtained by optimizing the parameters through the back propagation algorithm.

[0122] The correlation representation learned by DCCA is concatenated with the original PCA features to form a fused feature vector.

[0123] The features of different modes are regarded as nodes of the graph, such as current feature nodes, temperature feature nodes, etc., and a fusion model based on the graph attention network (GAT) is constructed. The relationship between nodes is learned through the attention mechanism, where the attention coefficient is calculated as follows:

[0124]

[0125] α is the attention vector, h i 、h j are the features of nodes i and j respectively, W is the weight matrix, N i represents the set of adjacent nodes of a node, k represents the number of nodes in the set of adjacent nodes, the symbol | represents feature concatenation, and LeakyReLU is the activation function;

[0126] The output of the GAT layer of the graph attention network is:

[0127]

[0128] σ is a nonlinear activation function.

[0129] By stacking multiple GAT layers, we can capture high-order interactions between different modalities, such as the correlation between current anomalies and temperature rise, and finally obtain a feature graph network after data feature fusion.

[0130] S206. Calculate the reachable distance of the neighboring node o of the node p in the feature graph network using a preset formula.

[0131] The characteristic graph network of the distribution line includes multiple nodes, each node has characteristics for a modal data, such as temperature node, voltage node, current node, etc. The reachable distance of the neighboring node o of node p in the characteristic graph network can be calculated, where node p can be any node, and the reachable distance of the neighboring node o of node p is reach-dist k (p, o):

[0132] reach-dist k (p, o) = max {k-distance (o), d (p, o)};

[0133] k-distance(o) is the k-neighbor distance of the k-th neighbor node o of node p, and d(p,o) is the Euclidean distance between node P and node o.

[0134] S207. Calculate the local reachability density of node p.

[0135] Specifically, the local reachable density calculation formula is as follows:

[0136]

[0137] N k (p) represents the incoming set of node p.

[0138] S208: Calculate the abnormality score of node p.

[0139] Specifically, the anomaly score calculation formula is as follows:

[0140]

[0141] S209: Determine the nodes whose abnormality scores are greater than a preset threshold as abnormal target nodes.

[0142] After obtaining the anomaly score of each node, the node with an anomaly score greater than the preset threshold is determined as the target node. Compared with the prior art that compares a single detection data with a threshold, the target node anomaly identification of this embodiment integrates multi-source heterogeneous data. For example, for temperature anomaly judgment, the prior art usually compares the detected temperature with a temperature threshold, while in this embodiment, for the temperature node, the multi-dimensional features of the temperature node and the voltage node, current node, vibration node, and audio node are calculated, and then the anomaly score of the temperature node is calculated to perform temperature anomaly judgment. This can adapt to dynamic changes in line voltage, current, etc. in the distribution network, has a high accuracy in identifying temperature anomalies, and avoids false alarms and missed alarms.

[0143] S210, extracting a local subgraph of the target node from the feature graph network.

[0144] The feature vectors of abnormal nodes are extracted from the feature graph network output by the Graph Attention Network (GAT) fusion model. For each node marked as abnormal, the information of its adjacent nodes is collected to construct a local subgraph. For example, several neighboring nodes of the target node or neighboring nodes within a preset range, or all nodes connected to the target node can be determined from the feature graph network to obtain a local subgraph of the target node.

[0145] S211. Construct a fault feature vector based on the local subgraph.

[0146] In an optional embodiment, the attention weights between nodes in the subgraph can be calculated to identify other features that are highly correlated with abnormal nodes. For example, if the attention weight between the current abnormal node and the temperature node is high, it means that there may be a fault caused by equipment overheating. The time series data of the abnormal node can be fine-grained analyzed, such as using a sliding window method with a window size of 1 minute and a step size of 1 second to recalculate the features in each time window. For example:

[0147] For current and voltage abnormal nodes, basic electrical parameters such as instantaneous value, effective value, phase angle, frequency, etc. are calculated, and voltage and current mutation amounts are extracted.

[0148] For the temperature nodes, the rate of change dT / dt is calculated, and the autocorrelation function is used to analyze the periodicity of the temperature change.

[0149] For vibration nodes, calculate the RMS value, peak factor, crest factor, clearance factor, and impulse factor.

[0150] For audio nodes, calculate the A-weighted equivalent continuous sound level LAeq and the maximum sound level LAmax.

[0151] Combined with SCADA system data, analyze changes in switch operation frequency, load rate and power factor.

[0152] The fault feature vector is constructed using the features of each node mentioned above.

[0153] S212. Calculate the cosine similarity between the fault feature vector and the feature vector in the fault type template library.

[0154] In this embodiment, feature vector samples corresponding to different fault types may be configured in advance in the fault type template library, and the cosine similarity between the fault feature vector and each feature vector sample may be calculated. The cosine similarity calculation formula is as follows:

[0155] S(A,B)=(A·B) / (||A|||B||);

[0156] Among them, A is the fault feature vector, and B is the feature vector sample in the fault type template library.

[0157] S213: Determine the N fault types with the highest cosine similarity as the fault types of the target node.

[0158] After obtaining the cosine similarity between the fault feature vector and each feature vector sample in the fault type template library, the cosine similarity can be sorted from large to small, and the fault type corresponding to the feature vector samples ranked in the top three positions is determined as the fault type of the target node, that is, the fault type of the distribution line.

[0159] S214: Push the fault type of the target node to the remote center.

[0160] After the edge computing node determines that the distribution line is abnormal and determines the fault type, the fault type can be pushed to the remote center for alarm.

[0161] The embodiment of the present invention receives electrical signals, temperature signals, vibration signals and audio signals collected by various types of sensors deployed in the distribution line and performs preprocessing, extracts features from the preprocessed electrical signals, temperature signals, vibration signals and audio signals respectively to obtain electrical features, temperature features, vibration features and audio features, performs feature selection to obtain target electrical features, target temperature features, target vibration features and target audio features, takes the target electrical features, target temperature features, target vibration features and target audio features as nodes, determines the correlation between two nodes based on the graph attention network to form an edge to obtain a feature graph network, calculates the reachable distance of the neighboring node o of the node p in the feature graph network by a preset formula, further calculates the local reachable density and then calculates the abnormality score of the node p, and selects the node with an abnormality score greater than a preset threshold. The target node is determined to be abnormal, the local subgraph of the target node is extracted from the feature graph network and the fault feature vector of the local subgraph is constructed. The fault feature vector is matched with the feature vector in the fault type template library to obtain the fault type of the distribution line, and the fault type is pushed to the remote center. On the one hand, by collecting multi-source heterogeneous data to construct a feature graph network to identify anomalies and further identify the fault type, compared with using a single monitoring data and a fixed threshold to compare and identify anomalies, it can adapt to the dynamic changes of the distribution network load and operation mode, and has a high accuracy in identifying anomalies. On the other hand, multi-source heterogeneous data is preprocessed and anomaly identified at the edge computing node without uploading to the remote center, which not only reduces the load of the remote center, but also avoids the delay in uploading data to the remote center, which can improve the speed of anomaly identification and can quickly and timely identify distribution line anomalies.

[0162] Embodiment 3

[0163] Figure 3 This is a schematic diagram of the structure of a device for identifying abnormal status of a power distribution line provided in Embodiment 3 of the present invention. Figure 3 As shown, the distribution line abnormal state identification device includes:

[0164] The data acquisition module 301 is used to receive multi-source heterogeneous data collected by various types of sensors deployed in the power distribution line;

[0165] A data preprocessing module 302 is used to preprocess the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data;

[0166] A feature graph network construction module 303 is used to construct a feature graph network based on the multi-source heterogeneous data;

[0167] An abnormal node identification module 304 is used to identify whether there is an abnormal target node in the distribution line based on the characteristic graph network, and if so, execute the local subgraph extraction module 305;

[0168] A local subgraph extraction module 305 is used to extract the local subgraph of the target node from the feature graph network;

[0169] A fault type identification module 306, configured to identify the fault type of the target node based on the local subgraph;

[0170] The push module 307 is used to push the fault type of the target node to the remote center.

[0171] Optionally, also include:

[0172] The time synchronization and acquisition parameter setting module is used to send time synchronization instructions and data acquisition parameters to various types of sensors in the distribution line. The sensors are used to perform time synchronization and set data acquisition frequency when receiving the time synchronization instructions and data acquisition parameters.

[0173] Optionally, the multi-source heterogeneous data includes at least an electrical signal, a temperature signal, a vibration signal, and an audio signal, and the data preprocessing module 302 includes:

[0174] A preprocessing unit is used to preprocess the electrical signal, the temperature signal, the vibration signal and the audio signal respectively to obtain the preprocessed electrical signal, the temperature signal, the vibration signal and the audio signal, wherein the preprocessing includes at least one of the following:

[0175] Filtering to reduce noise, remove outliers, supplement missing values ​​and perform standardization.

[0176] Optionally, the feature graph network construction module 303 includes:

[0177] A feature extraction unit, used to extract features from the preprocessed electrical signal, temperature signal, vibration signal and audio signal to obtain electrical features, temperature features, vibration features and audio features;

[0178] A feature selection unit, configured to select the electrical feature, the temperature feature, the vibration feature, and the audio feature to obtain a target electrical feature, a target temperature feature, a target vibration feature, and a target audio feature;

[0179] A graph construction unit is used to use the target electrical characteristics, target temperature characteristics, target vibration characteristics and target audio characteristics as nodes, determine the correlation between two nodes based on the graph attention network to form edges to obtain a feature graph network.

[0180] Optional, graph construction unit, including:

[0181] The attention coefficient calculation subunit is used to calculate the attention coefficient using the following formula:

[0182]

[0183] α is the attention vector, h i 、h j are the features of nodes i and j respectively, W is the weight matrix, N i represents the set of adjacent nodes of a node, k represents the number of nodes in the set of adjacent nodes, the symbol | represents feature concatenation, and LeakyReLU is the activation function;

[0184] The output of the graph attention network is:

[0185]

[0186] σ is a nonlinear activation function.

[0187] Optionally, the abnormal node identification module 304 includes:

[0188] The distance calculation unit is used to calculate the reach-dist of the neighboring node o of node p in the feature graph network using the following preset formula: k (p, o):

[0189] reach-dist k (p, o) = max {k-distance (o), d (p, o)};

[0190] k-distance(o) is the k-neighbor distance of the k-th neighbor node o of node p, and d(p,o) is the Euclidean distance between node P and node o;

[0191] Local reachability density calculation unit, used to calculate the local reachability density lrd of node p k (p):

[0192]

[0193] N k (p) represents the incoming set of node p;

[0194] Anomaly score calculation unit, used to calculate the anomaly score LOF of node p k (p):

[0195]

[0196] The abnormal node determination unit is used to determine the node whose abnormal score is greater than a preset threshold as an abnormal target node.

[0197] Optionally, the fault type identification module includes:

[0198] A fault feature vector construction unit, used to construct a fault feature vector based on the local subgraph;

[0199] A feature similarity calculation unit, used to calculate the cosine similarity between the fault feature vector and the feature vector in the fault type template library;

[0200] The fault type determination unit is used to determine the N fault types with the highest cosine similarity as the fault types of the target node.

[0201] The device for identifying abnormal state of a distribution line provided in the embodiment of the present invention can execute the method for identifying abnormal state of a distribution line provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0202] Embodiment 4

[0203] Figure 4 A schematic diagram of the structure of an electronic device 40 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0204] like Figure 4 As shown, the electronic device 40 includes at least one processor 41, and a memory connected to the at least one processor 41, such as a read-only memory (ROM) 42, a random access memory (RAM) 43, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 41 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 42 or the computer program loaded from the storage unit 48 to the random access memory (RAM) 43. In the RAM 43, various programs and data required for the operation of the electronic device 40 can also be stored. The processor 41, the ROM 42, and the RAM 43 are connected to each other through a bus 44. An input / output (I / O) interface 45 is also connected to the bus 44.

[0205] A number of components in the electronic device 40 are connected to the I / O interface 45, including: an input unit 46, such as a keyboard, a mouse, etc.; an output unit 47, such as various types of displays, speakers, etc.; a storage unit 48, such as a disk, an optical disk, etc.; and a communication unit 49, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 49 allows the electronic device 40 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0206] The processor 41 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 41 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 41 executes the various methods and processes described above, such as a distribution line abnormal state identification method.

[0207] In some embodiments, the distribution line abnormal state identification method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 48. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 40 via the ROM 42 and / or the communication unit 49. When the computer program is loaded into the RAM 43 and executed by the processor 41, one or more steps of the distribution line abnormal state identification method described above can be performed. Alternatively, in other embodiments, the processor 41 can be configured to execute the distribution line abnormal state identification method in any other appropriate manner (for example, by means of firmware).

[0208] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0209] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0210] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0211] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0212] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0213] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0214] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0215] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying abnormal status of a distribution line, characterized in that: Applied to edge computing nodes, the edge computing nodes are respectively connected to sensors and remote centers, including: Receive multi-source heterogeneous data collected by various types of sensors deployed in distribution lines; Preprocessing the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; Construct a feature graph network based on preprocessed multi-source heterogeneous data; Identify whether there is an abnormal target node in the power distribution line based on the characteristic graph network; If so, extracting a local subgraph of the target node from the feature graph network; Identifying a fault type of the target node based on the local subgraph; The fault type of the target node is pushed to the remote center.

2. The method for identifying abnormal state of a power distribution line according to claim 1, characterized in that: Before receiving multi-source heterogeneous data collected by various types of sensors deployed in the distribution lines, it also includes: Time synchronization instructions and data acquisition parameters are sent to various types of sensors in the power distribution line, and the sensors are used to perform time synchronization and set data acquisition frequency when receiving the time synchronization instructions and data acquisition parameters.

3. The method for identifying abnormal state of a power distribution line according to claim 1, characterized in that: The multi-source heterogeneous data at least includes an electrical signal, a temperature signal, a vibration signal, and an audio signal. The multi-source heterogeneous data is preprocessed to obtain the preprocessed multi-source heterogeneous data, including: Preprocessing the electrical signal, the temperature signal, the vibration signal and the audio signal respectively to obtain preprocessed electrical signal, temperature signal, vibration signal and audio signal, wherein the preprocessing includes at least one of the following: Filtering to reduce noise, remove outliers, supplement missing values ​​and perform standardization.

4. The method for identifying abnormal state of a power distribution line according to claim 3, characterized in that: Construct a feature graph network based on preprocessed multi-source heterogeneous data, including: Extracting features from the preprocessed electrical signal, temperature signal, vibration signal and audio signal respectively to obtain electrical features, temperature features, vibration features and audio features; Selecting the electrical characteristics, temperature characteristics, vibration characteristics and audio characteristics to obtain target electrical characteristics, target temperature characteristics, target vibration characteristics and target audio characteristics; The target electrical characteristics, target temperature characteristics, target vibration characteristics and target audio characteristics are used as nodes, and the correlation between two nodes is determined based on the graph attention network to form an edge to obtain a feature graph network.

5. The method for identifying abnormal state of a power distribution line according to claim 4, characterized in that: Based on the graph attention network, the correlation between two nodes is determined to form an edge to obtain a feature graph network, including: The attention coefficient is calculated by the following formula: α is the attention vector, h i 、h j are the features of nodes i and j respectively, W is the weight matrix, N i represents the set of adjacent nodes of a node, k represents the number of nodes in the set of adjacent nodes, the symbol | represents feature concatenation, and LeakyReLU is the activation function; The output of the graph attention network is: σ is a nonlinear activation function.

6. The method for identifying abnormal status of a power distribution line according to any one of claims 1 to 5, characterized in that: Identifying whether there is an abnormal target node in the power distribution line based on the characteristic graph network includes: The reach-dist of the neighboring node o of node p in the feature graph network is calculated by the following preset formula: k (p, o): reach-dist k (p,o)=max{k-distance(o),d(p,o)}; k-distance(o) is the k-neighbor distance of the k-th neighbor node o of node p, and d(p,o) is the Euclidean distance between node P and node o; Calculate the local reachability density lrd of node p k (p): N k (p) represents the incoming set of node p; Calculate the anomaly score LOF of node p k (p): Nodes with anomaly scores greater than a preset threshold are identified as abnormal target nodes.

7. The method for identifying abnormal status of a power distribution line according to any one of claims 1 to 5, characterized in that: Identifying the fault type of the target node based on the local subgraph includes: Constructing a fault feature vector based on the local subgraph; Calculating the cosine similarity between the fault feature vector and the feature vector in the fault type template library; The N fault types with the highest cosine similarity are determined as the fault types of the target node.

8. A device for identifying abnormal status of a power distribution line, characterized in that: Applied to edge computing nodes, the edge computing nodes are respectively connected to sensors and remote centers, including: A data acquisition module, used to receive multi-source heterogeneous data collected by various types of sensors deployed in the power distribution line; A data preprocessing module, used to preprocess the multi-source heterogeneous data to obtain preprocessed multi-source heterogeneous data; A feature graph network construction module, used to construct a feature graph network based on the multi-source heterogeneous data; An abnormal node identification module is used to identify whether there is an abnormal target node in the distribution line based on the characteristic graph network, and if so, execute a local subgraph extraction module; A local subgraph extraction module, used to extract the local subgraph of the target node from the feature graph network; A fault type identification module, used to identify the fault type of the target node based on the local subgraph; A push module is used to push the fault type of the target node to the remote center.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for identifying abnormal status of a distribution line according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for identifying abnormal status of a distribution line according to any one of claims 1 to 7 when executed.

Images