Power distribution intelligent operation and maintenance management method and system based on data processing

Abstract

The invention discloses a power distribution intelligent operation and maintenance management method and system based on data processing, and the method comprises the steps: carrying out the preprocessing of multi-source power distribution operation and maintenance data, and carrying out the alignment of different sampling frequency data in the preprocessed data through a time sequence alignment algorithm; converting the aligned time sequence data into a two-dimensional feature image through Grubby angle field transformation; electrical quantity time sequence data are processed based on a space-time Transform encoder to establish a Transform power distribution state prediction model, TCN time convolution blocks are introduced into the model to capture long-term dependence, and a space attention mechanism is added to position abnormal equipment nodes; and judging the state of the power distribution equipment according to the power distribution operation and maintenance health index, and performing abnormal reason assessment by using a Bayesian network in combination with historical fault cases and environmental factors. And the possibility of each fault reason can be quickly calculated and sequenced and output. The troubleshooting time is effectively shortened, and the operation and maintenance efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power operation and maintenance, and in particular to a distribution intelligent operation and maintenance management method and system based on data processing. Background Art

[0002] With the deepening of the construction of smart grids, the scale of distribution systems has been continuously expanding, and the types of equipment have become increasingly complex. Many problems have gradually emerged in the traditional operation and maintenance management mode. The operation data of distribution equipment is scattered in multiple systems such as distribution terminals, meters, and inspection records. The data formats and sampling frequencies are different, making it difficult to achieve effective integration and collaborative analysis, resulting in the inability of operation and maintenance personnel to quickly obtain the comprehensive status information of equipment. Existing operation and maintenance means mostly rely on manual inspections and experience judgments, with insufficient fault prediction capabilities. They often respond passively after equipment failures occur, which not only increases the emergency repair cost but also may cause large-scale power outages, affecting power supply reliability and user experience. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems, and a distribution intelligent operation and maintenance management method and system based on data processing are designed.

[0004] Furthermore, in the above-mentioned distribution intelligent operation and maintenance management method based on data processing, the distribution intelligent operation and maintenance management method includes the following steps: Collect current waveforms, voltage waveforms, partial discharge data, temperature and humidity data, geographical information data, and historical fault data in distribution terminals, meters, inspection robots, and user work order systems to obtain multi-source distribution operation and maintenance data; Preprocess the multi-source distribution operation and maintenance data, align data with different sampling frequencies in the preprocessed data using a time series alignment algorithm, and convert the aligned time series data into a two-dimensional feature image through Gramian angular field transformation to obtain fused distribution operation and maintenance data; Based on the spatio-temporal Transformer encoder to process electrical quantity time series data, establish a Transformer distribution state prediction model, introduce a TCN time convolutional block in the model to capture long-term dependencies, and add a spatial attention mechanism to locate abnormal equipment nodes to obtain a Transformer-TCN distribution state prediction model; Input the fused distribution operation and maintenance data into the Transformer-TCN distribution state prediction model for identification, and output a distribution operation and maintenance health index; Judge the state of distribution equipment according to the distribution operation and maintenance health index, and use a Bayesian network to combine historical fault cases and environmental factors to evaluate the cause of anomalies.

[0005] Further, in the above-mentioned power distribution intelligent operation and maintenance management method based on data processing, the preprocessing of the multi-source power distribution operation and maintenance data and the alignment of data with different sampling frequencies in the preprocessed data using the time series alignment algorithm include: Detect outliers in the multi-source power distribution operation and maintenance data based on the Isolation Forest algorithm, delete the outliers and fill them using the linear interpolation method to obtain filled power distribution operation and maintenance data; Perform time series alignment on data with different sampling frequencies through the DTW (Dynamic Time Warping) algorithm, and convert the time series data with different sampling frequencies into a sequence indexed by timestamps; Calculate the distance matrix between two sequences using the Euclidean distance; find the minimum cumulative distance path from the upper left corner to the lower right corner of the matrix through dynamic programming to determine the alignment relationship of the time series; Resample the data according to the aligned timestamps to obtain aligned power distribution operation and maintenance data.

[0006] Further, in the above-mentioned power distribution intelligent operation and maintenance management method based on data processing, the conversion of the aligned time series data into a two-dimensional feature image through Gramian Angular Field (GAF) transformation to obtain fused power distribution operation and maintenance data further includes: Perform normalization processing on the aligned power distribution operation and maintenance data using the min-max normalization method, map the data to the interval [−1, 1] to obtain normalized data; Calculate the angle corresponding to each data point in the normalized data, convert each data point into an angle in polar coordinates, and regard it as a point on the unit circle; Generate two two-dimensional images, including the cosine Gramian Angular Field and the sine Gramian Angular Field, by calculating the cosine and sine combination relationships of angles at different time points to obtain fused power distribution operation and maintenance data.

[0007] Further, in the above-mentioned power distribution intelligent operation and maintenance management method based on data processing, the establishment of a Transformer power distribution state prediction model based on a spatio-temporal Transformer encoder for processing electrical quantity time series data includes: The spatio-temporal Transformer encoder consists of multiple layers of Transformer encoding layers, each layer contains a multi-head self-attention mechanism and a feed-forward neural network, and positional encoding is introduced; Use multi-head self-attention to map the input data to multiple subspaces, independently calculate the dependency relationships between different time steps and device nodes within each subspace, and then splice the outputs to capture the spatio-temporal correlations of the data; The input is the aligned electrical quantity time series data, and the interaction features of different time steps and device nodes are extracted through multi-head self-attention.

[0008] Further, in the above-mentioned power distribution intelligent operation and maintenance management method based on data processing, introducing a TCN temporal convolutional block in the model to capture long-term dependencies, adding a spatial attention mechanism to locate abnormal device nodes, and obtaining a Transformer-TCN power distribution status prediction model, further includes: The TCN temporal convolutional block includes dilated causal convolution and a residual connection structure. The receptive field of the convolutional kernel is enlarged through dilated causality to capture long-term dependencies without increasing parameters. The input is directly superimposed on the output of the convolutional layer using the residual connection. Each block contains three layers of dilated convolution, and a deep network is obtained after stacking. The features output by the Transformer are averaged and max-pooled in the time dimension to generate two spatial feature descriptors, and the original features are weighted using attention weights to highlight the features of abnormal nodes.

[0009] Further, in the above-mentioned power distribution intelligent operation and maintenance management method based on data processing, inputting the fused power distribution operation and maintenance data into the Transformer-TCN power distribution status prediction model for identification and outputting a power distribution operation and maintenance health index, includes: Inputting the fused power distribution operation and maintenance data into the Transformer-TCN power distribution status prediction model for identification, and extracting local features of image edges and textures through a convolutional layer. Integrating spatio-temporal and long-term time features through the Transformer-TCN module, capturing complex patterns of device operation, and mapping to an evaluation value from 0 to 100 through a fully connected layer to obtain a power distribution operation and maintenance health index.

[0010] Further, in the above-mentioned power distribution intelligent operation and maintenance management method based on data processing, judging the status of power distribution equipment according to the power distribution operation and maintenance health index, and using a Bayesian network to combine historical fault cases and environmental factors for abnormal cause assessment, includes: When it is judged that the device is in a warning state, collect the current operation data, environmental parameters, and historical fault information as evidence. Use the Bayesian inference algorithm to calculate the posterior probabilities of each abnormal cause, and output the abnormal causes sorted by the posterior probabilities.

[0011] Further, in a power distribution intelligent operation and maintenance management system based on data processing, the power distribution intelligent operation and maintenance management system includes the following modules: A power distribution data acquisition module, used to collect current waveforms, voltage waveforms, partial discharge data, temperature and humidity data, geographic information data, and historical fault data in power distribution terminals, electric meters, inspection robots, and user work order systems to obtain multi-source power distribution operation and maintenance data. The power distribution data processing module is used to preprocess the multi-source power distribution operation and maintenance data, align the data with different sampling frequencies in the preprocessed data using the time series alignment algorithm, convert the aligned time series data into a two-dimensional feature image through the Gramian angular field transformation, and obtain the fused power distribution operation and maintenance data; The prediction model establishment module is used to establish a Transformer power distribution state prediction model based on the spatio-temporal Transformer encoder to process the electrical quantity time series data. In the model, a TCN time convolutional block is introduced to capture long-term dependencies, and a spatial attention mechanism is added to locate abnormal device nodes, resulting in a Transformer-TCN power distribution state prediction model; The power distribution health identification module is used to input the fused power distribution operation and maintenance data into the Transformer-TCN power distribution state prediction model for identification, and output the power distribution operation and maintenance health index; The power distribution operation and maintenance management module is used to judge the state of power distribution equipment according to the power distribution operation and maintenance health index, and use the Bayesian network to combine historical fault cases and environmental factors to evaluate the cause of anomalies.

[0012] Furthermore, in a power distribution intelligent operation and maintenance management system based on data processing, the power distribution data processing module includes the following sub-modules: The processing sub-module is used to normalize the power distribution operation and maintenance data after alignment using the min-max normalization method, map the data to the interval [−1,1], and obtain the normalized data; The calculation sub-module is used to calculate the angle corresponding to each data point in the normalized data, convert each data point into an angle in polar coordinates, and regard it as a point on the unit circle; The obtaining sub-module is used to generate two two-dimensional images, including the cosine Gramian angular field and the sine Gramian angular field, by calculating the cosine and sine combination relationships of angles at different time points, and obtain the fused power distribution operation and maintenance data.

[0013] Furthermore, in a power distribution intelligent operation and maintenance management system based on data processing, the prediction model establishment module includes the following sub-modules: The introduction sub-module is used for the spatio-temporal Transformer encoder to consist of multiple Transformer encoding layers, each layer containing a multi-head self-attention mechanism and a feed-forward neural network, and introducing position encoding; The mapping sub-module is used to map the input data to multiple sub-spaces using multi-head self-attention, independently calculate the dependencies of different time steps and device nodes within each sub-space, and then splice the outputs to capture the spatio-temporal correlations of the data; The input sub-module is used for the input to be the aligned electrical quantity time series data, and the multi-head self-attention is used to extract the interaction features of different time steps and device nodes.

[0014] Its beneficial effects are as follows: 1. Compared with the traditional qualitative judgment method, it greatly improves the accuracy and objectivity of equipment status evaluation, realizing the transformation from passive maintenance to active prevention. 2. By establishing the probability relationship between input variables and output variables, when the equipment shows abnormalities, it can quickly calculate the probabilities of various fault causes and sort and output them. This method makes full use of historical experience data, effectively shortens the fault troubleshooting time, improves the operation and maintenance efficiency, and reduces the economic losses caused by faults. 3. It realizes the intelligence and automation of the operation and maintenance of distribution equipment, reduces the frequency and intensity of manual inspections, and reduces the labor cost. At the same time, by predicting equipment faults in advance and giving timely warnings, it avoids the expansion of faults, reduces the equipment repair and replacement costs, and prolongs the service life of the equipment. In addition, accurate fault diagnosis can shorten the power outage time and improve user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention.

[0016] Figure 1 Schematic diagram of the first embodiment of a distribution intelligent operation and maintenance management method based on data processing in an embodiment of the present invention; Figure 2 Schematic diagram of the second embodiment of a distribution intelligent operation and maintenance management method based on data processing in an embodiment of the present invention; Figure 3 Schematic diagram of the first embodiment of a distribution intelligent operation and maintenance management system based on data processing in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Those skilled in the art of the present technology can understand that unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0019] The present invention will be specifically described below with reference to the drawings, Figure 1As shown in the figure, a distribution intelligent operation and maintenance management method based on data processing, the distribution intelligent operation and maintenance management method includes the following steps: Step 101, collect current waveforms, voltage waveforms, partial discharge data, temperature and humidity data, geographic information data, and historical fault data in distribution terminals, electric meters, inspection robots, and user work order systems to obtain multi-source distribution operation and maintenance data; Specifically, I. Data collection objects and sources The data collected in this solution covers various types of data in distribution terminals, electric meters, inspection robots, and user work order systems, specifically including current waveforms, voltage waveforms, partial discharge data, temperature and humidity data, geographic information data, and historical fault data, to obtain multi-source equipment operation and maintenance data. II. Collection methods for each data source Distribution terminal data collection Collection equipment: Deploy high-precision current sensors, voltage sensors, and partial discharge sensors on key distribution terminal equipment such as distribution transformers, switch cabinets, and cable joints. The current sensor selects a Rogowski coil sensor, which has a wide frequency band response (0.1Hz - 1MHz) and can accurately capture transient and steady-state current waveforms; the voltage sensor uses a capacitive voltage divider sensor with a measurement accuracy of up to 0.2 level; the partial discharge sensor uses an ultra-high frequency sensor and can detect discharge amounts above 1pC. Collection frequency: For current waveforms and voltage waveforms, in order to accurately reflect the dynamic characteristics of the distribution system, the collection frequency is set to 10kHz; due to the intermittent and random occurrence of partial discharge data, a real-time trigger collection method is adopted. When the detected discharge signal exceeds the set threshold (10pC), continuous collection is started, and the collection duration is 100ms to ensure complete recording of the discharge pulse waveform. Electric meter data collection Collection scope: Covers various types of electric energy meters in the distribution network, including high-voltage electric energy meters, low-voltage electric energy meters, and smart electric energy meters on the user side. Communication method: Realize real-time transmission of electric meter data through various communication technologies such as power line carrier (PLC) and wireless communication (4G, NB-IoT). For high-voltage electric energy meters, optical fiber communication is preferred to ensure the stability and real-time nature of data transmission, and the communication rate is not less than 10Mbps; low-voltage electric energy meters and smart electric energy meters on the user side can select NB-IoT or 4G communication according to actual situations. The NB-IoT communication module supports low-power wide area network (LPWAN) technology and meets the requirements of massive connections and long-distance transmission. Acquisition content: In addition to collecting basic electrical energy data such as active power, reactive power, voltage, and current, the operating status parameters of the electricity meter are also collected, such as the clock error of the electricity meter and the communication signal strength. The acquisition frequency is once every 15 minutes, and real-time acquisition is triggered for abnormal data (such as voltage limit violation and current mutation). Data Acquisition of Patrol Inspection Robot Robot types: For different power distribution scenarios, indoor patrol inspection robots and outdoor patrol inspection robots are adopted. Indoor robots are mainly used in enclosed spaces such as substations and distribution rooms, and have functions of autonomous navigation and obstacle avoidance; outdoor robots are suitable for inspecting outdoor distribution lines, poles and other equipment, and have properties such as waterproof, dustproof and wind resistance. Sensors carried: Patrol inspection robots carry a variety of sensors, including an infrared thermal imager (temperature measurement accuracy ±2°C, temperature measurement range -20°C - 500°C) for collecting surface temperature and humidity data of equipment, a visible light camera (resolution not less than 20 million pixels) for obtaining equipment appearance images and geographic information data (realized through a GPS / Beidou dual-mode positioning system, positioning accuracy ≤10m), and an ultrasonic sensor for detecting ultrasonic signals generated by partial discharge of equipment (detection frequency range 20kHz - 100kHz).

[0020] Patrol inspection path planning: Through a pre-set patrol inspection route or an autonomous path planning algorithm based on the real-time environment, ensure that the robot conducts a comprehensive and non-blind-spot inspection of power distribution equipment. Indoor robots use SLAM (Simultaneous Localization and Mapping) technology to build an environmental map and achieve autonomous navigation; outdoor robots combine GIS (Geographic Information System) data and real-time sensor data to plan the optimal patrol inspection path. The patrol inspection period is determined according to the importance of the equipment and the operating environment. Generally, important equipment is inspected once a week, and ordinary equipment is inspected once every two weeks. Data Acquisition of User Work Order System Data interface: Establish a standardized data interface with the user work order system, support multiple interface protocols such as RESTful API and SOAP, and ensure real-time synchronization and sharing of data. Acquisition content: Information such as fault repair work orders and equipment abnormality complaint work orders submitted by users is collected, including work order numbers, repair times, fault descriptions, fault locations (positioned by combining the address information provided by users with the GIS system), processing status, etc. At the same time, semantic analysis is performed on the work order data to extract key fault features, such as fault types (short circuit, grounding, equipment damage, etc.) and fault phenomena (power outage, voltage fluctuation, equipment abnormal noise, etc.).

[0021] Step 102: Preprocess the multi-source power distribution operation and maintenance data, align data with different sampling frequencies in the preprocessed data using a time series alignment algorithm, and convert the aligned time series data into a two-dimensional feature image through Gramian Angular Field transformation to obtain fused power distribution operation and maintenance data; Specifically, in this embodiment, the isolation forest algorithm is used to detect outliers in multi-source power distribution operation and maintenance data. After deleting the outliers, linear interpolation is used for filling to obtain filled power distribution operation and maintenance data; The DTW (Dynamic Time Warping) algorithm is used to align time series of data with different sampling frequencies, and the time series data with different sampling frequencies is converted into a sequence indexed by timestamps; The Euclidean distance is used to calculate the distance matrix between two sequences; the minimum cumulative distance path from the upper left corner to the lower right corner of the matrix is found through dynamic programming to determine the alignment relationship of the time series; The data is resampled according to the aligned timestamps to obtain aligned power distribution operation and maintenance data.

[0022] The minimum-maximum normalization method is used to normalize the aligned power distribution operation and maintenance data, and the data is mapped to the interval [−1, 1] to obtain normalized data; The angle corresponding to each data point in the normalized data is calculated, and each data point is converted into an angle in polar coordinates and regarded as a point on the unit circle; By calculating the cosine and sine combination relationships of angles at different time points, two two-dimensional images are generated, including a cosine Gram angle field and a sine Gram angle field, to obtain fused power distribution operation and maintenance data.

[0023] Specifically, I. Data Preprocessing Data Cleaning Outlier Detection: For continuous data such as current and voltage, first, suspected outliers are screened through statistical methods (standard deviation method, interquartile range method), and then machine learning algorithms (isolation forest, one-class support vector machine) are used for further verification; for work order text data, after processes such as word segmentation and stop word removal, text similarity detection is used to detect duplicate or contradictory information.

[0024] Missing Value Handling: In time series data, short-term missing values (less than 1 sampling period) are filled using linear interpolation; long-term missing values are predicted and filled using the periodic law of historical data. If the fault location is missing in the work order data, it is supplemented through user communication or geographical information is extracted from the description.

[0025] Dynamic Time Warping (DTW) algorithm is used for time series alignment to process data with different sampling frequencies: first, the data is converted into a timestamp-indexed sequence, the distance matrix between sequences is calculated, the best alignment path is found through dynamic programming, and finally, resampling is performed at a unified frequency (1Hz) to ensure the consistency of the time dimension of multi-source data.

[0026] II. Gram Angle Field Transformation Convert the one-dimensional time series into a two-dimensional feature image. The specific steps are as follows: Data normalization: Map the aligned time-series data to the interval [-1, 1] to highlight the data fluctuation characteristics; Angle mapping: Convert each data point to an angle in polar coordinates (based on the arccosine function), regarded as a point on the unit circle; Image construction: Generate two two-dimensional images (cosine Gram angle field and sine Gram angle field) by calculating the cosine and sine combination relationships of angles at different time points, and the fused image is used as the feature image for model input, retaining the trend and periodic information of the time series.

[0027] Step 103: Establish a Transformer distribution state prediction model based on processing electrical quantity time-series data by a spatio-temporal Transformer encoder. Introduce a TCN time convolution block in the model to capture long-term dependencies, and add a spatial attention mechanism to locate abnormal device nodes, obtaining a Transformer-TCN distribution state prediction model; Specifically, in this embodiment, the spatio-temporal Transformer encoder consists of multiple layers of Transformer encoding layers. Each layer contains a multi-head self-attention mechanism and a feed-forward neural network, and position encoding is introduced; Use multi-head self-attention to map the input data to multiple subspaces, independently calculate the dependencies between different time steps and device nodes in each subspace, and then splice the outputs to capture the spatio-temporal correlations of the data; The input is the aligned electrical quantity time-series data, and the multi-head self-attention extracts the interaction features of different time steps and device nodes.

[0028] The TCN time convolution block includes dilated causal convolution and residual connection structures. Expand the receptive field of the convolution kernel through dilated causal convolution to capture long-term dependencies without increasing parameters; Use residual connection to directly stack the input onto the output of the convolution layer. Each block contains 3 layers of dilated convolution, and after stacking, a deep network is obtained; Perform average and max pooling on the features output by the Transformer in the time dimension to generate two spatial feature descriptors, and use attention weights to weight the original features to highlight the abnormal node features.

[0029] Specifically, I. Spatio-temporal Transformer encoder Consists of multiple layers of Transformer encoding layers. Each layer contains a multi-head self-attention mechanism and a feed-forward neural network, and position encoding is introduced: Multi-head self-attention: Map the input data to multiple subspaces, independently calculate the dependencies between different time steps and device nodes in each subspace, and then splice the outputs to capture the spatio-temporal correlations of the data; Position Encoding: Generate position information through sine and cosine functions and embed it into the input data so that the model can perceive the sequentiality of the time series. The input is the aligned time series data of electrical quantities (current, voltage). After being processed by the multi-head self-attention mechanism, the interaction features of different time steps and device nodes are extracted.

[0030] II. TCN Time Convolution Block Adopt dilated causal convolution and residual connection structure: Dilated causal convolution: Expand the receptive field of the convolution kernel through exponentially increasing dilation rates (1, 2, 4, 8) to capture long-term dependencies (from hours to days) without increasing the number of parameters; Residual connection: Directly stack the input on the output of the convolutional layer to avoid the problem of gradient vanishing in deep networks. Each block contains three layers of dilated convolution (kernel size 3, activation function ReLU). After stacking, a deep network is formed to enhance the long-term feature extraction ability of the time series.

[0031] III. Spatial Attention Mechanism Locate abnormal device nodes, and the steps are as follows: Feature fusion: Average and max-pool the features output by the Transformer in the time dimension to generate two spatial feature descriptors; Weight calculation: Merge the two descriptors and pass them through a 1×1 convolutional layer. After passing through the activation function, generate spatial attention weights to represent the importance of each device node; Feature weighting: Weight the original features with the attention weights to highlight the features of abnormal nodes and suppress irrelevant information.

[0032] IV. Model Integration Data processing flow: The time series data of electrical quantities first extracts spatio-temporal features through the spatio-temporal Transformer, then captures long-term time dependencies through the TCN block, and finally locates abnormal nodes through the spatial attention mechanism. Finally, the device status feature vector is output through the fully connected layer.

[0033] Step 104: Input the fused distribution operation and maintenance data into the Transformer-TCN distribution state prediction model for identification, and output the distribution operation and maintenance health index; Specifically, in this embodiment, the fused distribution operation and maintenance data is input into the Transformer-TCN distribution state prediction model for identification, and the local features of the image edges and textures are extracted through the convolutional layer; Integrate spatio-temporal and long-term time features through the Transformer-TCN module, capture the complex patterns of device operation, and map them to an evaluation value of 0-100 through the fully connected layer to obtain the distribution operation and maintenance health index.

[0034] Specifically, I. Data Input Input the preprocessed two-dimensional feature image (224×224 pixels) into the model. The image is generated by the Gram angular field transformation and integrates the spatio-temporal features of multi-source time-series data.

[0035] II. Model Recognition Process Shallow Feature Extraction: Extract local features such as image edges and textures through the convolutional layer; Deep Feature Extraction: Integrate spatio-temporal and long-term time features through the Transformer-TCN module to capture the complex patterns of device operation; Health Index Output: Map to a health index of 0-100 through the fully connected layer. The higher the value, the better the state. When it is below 60, an abnormal warning is triggered.

[0036] III. Output Explanation The health index synthesizes multi-source data and model analysis results to quantify the device operation state: Normal (≥80): Conduct regular inspections; Warning (60-80): Shorten the inspection cycle and strengthen monitoring; Fault (<60): Issue an emergency alarm and stop the machine for maintenance.

[0037] Step 105: Judge the state of the power distribution equipment according to the health index of the power distribution operation, and use the Bayesian network to combine historical fault cases and environmental factors to evaluate the cause of the anomaly.

[0038] Specifically, in this embodiment, when it is judged that the device is in a warning state, collect the current operation data, environmental parameters, and historical fault information as evidence; Use the Bayesian inference algorithm to calculate the posterior probability of each cause of the anomaly, and output the cause of the anomaly sorted by the posterior probability.

[0039] Specifically, I. State Judgment Rules Judge the state of the device according to the health index of the power distribution equipment, and set three state levels: Normal State: The health index ≥80, the device runs stably, without obvious abnormal signs, and can be monitored according to the regular inspection cycle. Warning State: 60 ≤ health index < 80, the device has potential fault risks, it is necessary to shorten the inspection cycle, strengthen real-time monitoring, and arrange professional personnel for on-site inspection. Fault State: The health index < 60, the device may have failed or is about to fail, immediately trigger an emergency alarm, stop the device operation, and start the emergency plan for maintenance. II. Bayesian Network Construction Node Definition: The nodes of a Bayesian network include input variable nodes and output variable nodes. The input variable nodes include historical fault data (fault type, fault occurrence frequency), environmental factors (temperature and humidity, geographical information, weather conditions), equipment operation data (current, voltage, partial discharge), etc.; the output variable node is the cause of equipment anomalies (insulation aging, overloading operation, poor contact, etc.). Conditional Probability Table: Through the statistical analysis of historical fault cases, the conditional probability relationship between the input variable nodes and the output variable nodes is established. For example, when the environmental temperature is too high (>40°C) and the equipment current exceeds 120% of the rated current, the probability of the equipment having an overload fault is 70%. The conditional probability table can be established using the maximum likelihood estimation method or the Bayesian parameter learning method to estimate the probability parameters using historical data. Network Structure Learning: Use a scoring search-based method (K2 algorithm, hill climbing algorithm) to learn the structure of the Bayesian network from historical data and determine the causal relationship between nodes. III. Abnormality Cause Evaluation Process Evidence Input: When the equipment status is judged to be in a warning or fault state, collect the current equipment operation data, environmental data, and historical fault data as evidence and input them into the Bayesian network. Probability Inference: Use the inference algorithms of the Bayesian network (joint tree algorithm, variable elimination method) for probability inference, calculate the posterior probabilities of each abnormality cause node, and determine the most likely abnormality cause. Result Output: Output the probability ranking of the abnormality causes to provide the direction for maintenance personnel to troubleshoot faults. For example, output "Insulation Aging (probability 65%) > Overloading Operation (probability 25%) > Poor Contact (probability 10%)" to guide the maintenance personnel to check the equipment insulation condition first.

[0040] Its beneficial effects are as follows: 1. Compared with the traditional qualitative judgment method, it greatly improves the accuracy and objectivity of equipment status evaluation, realizing the transformation from passive maintenance to active prevention. 2. By establishing the probability relationship between input variables and output variables, when the equipment shows abnormalities, it can quickly calculate the possibilities of various fault causes and sort them for output. This method makes full use of historical experience data, effectively shortens the fault troubleshooting time, improves the maintenance efficiency, and reduces the economic losses caused by faults. 3. It realizes the intelligence and automation of distribution equipment maintenance, reduces the frequency and intensity of manual inspections, and reduces the labor cost. At the same time, by predicting equipment faults in advance and giving timely warnings, it avoids the expansion of faults, reduces the equipment repair and replacement costs, and extends the service life of the equipment. In addition, accurate fault diagnosis can shorten the power outage time and improve user satisfaction.

[0041] Please refer to Figure 2, in a power distribution intelligent operation and maintenance management method based on data processing, preprocessing multi-source power distribution operation and maintenance data, and aligning data with different sampling frequencies in the preprocessed data using a time series alignment algorithm includes the following steps: Step 201: Detect outliers in the multi-source power distribution operation and maintenance data based on the Isolation Forest algorithm, delete the outliers, and fill them using linear interpolation to obtain filled power distribution operation and maintenance data; Step 202: Align the time series of data with different sampling frequencies through the DTW (Dynamic Time Warping) algorithm, and convert the time series data with different sampling frequencies into a sequence indexed by timestamps; Step 203: Calculate the distance matrix between the two sequences using the Euclidean distance; find the minimum cumulative distance path from the upper left corner to the lower right corner of the matrix through dynamic programming to determine the alignment relationship of the time series; Step 204: Resample the data according to the aligned timestamps to obtain aligned power distribution operation and maintenance data.

[0042] The above introduces the embodiments of a power distribution intelligent operation and maintenance management method based on data processing of the present invention. Please refer to Figure 3 , in a power distribution intelligent operation and maintenance management system based on data processing, the power distribution intelligent operation and maintenance management system includes the following modules: A power distribution data acquisition module, used to collect current waveforms, voltage waveforms, partial discharge data, temperature and humidity data, geographic information data, and historical fault data in power distribution terminals, electric meters, inspection robots, and user work order systems to obtain multi-source power distribution operation and maintenance data; A power distribution data processing module, used to preprocess multi-source power distribution operation and maintenance data, align data with different sampling frequencies in the preprocessed data using a time series alignment algorithm, and convert the aligned time series data into a two-dimensional feature image through Gramian Angular Field transformation to obtain fused power distribution operation and maintenance data; A prediction model establishment module, used to establish a Transformer power distribution state prediction model based on processing electrical quantity time series data by a spatio-temporal Transformer encoder, introduce a TCN (Time Convolutional Network) time convolution block in the model to capture long-term dependencies, and add a spatial attention mechanism to locate abnormal device nodes to obtain a Transformer-TCN power distribution state prediction model; A power distribution health identification module, used to input the fused power distribution operation and maintenance data into the Transformer-TCN power distribution state prediction model for identification, and output a power distribution operation and maintenance health index; A power distribution operation and maintenance management module, used to judge the state of power distribution equipment according to the power distribution operation and maintenance health index, and use a Bayesian network to combine historical fault cases and environmental factors to evaluate the cause of anomalies.

[0043] The basic principles, main features and advantages of the present invention have been shown and described above. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A power distribution intelligent operation and maintenance management method based on data processing, characterized in that, The described intelligent distribution operation and maintenance management method includes the following steps: Collect current waveforms, voltage waveforms, partial discharge data, temperature and humidity data, geographical information data, and historical fault data from distribution terminals, electricity meters, inspection robots, and user work order systems to obtain multi-source distribution operation and maintenance data; Preprocess the multi-source distribution operation and maintenance data. Use the time series alignment algorithm to align data with different sampling frequencies in the preprocessed data, and convert the aligned time series data into two-dimensional feature images through Gramian angular field transformation to obtain fused distribution operation and maintenance data; Based on the spatio-temporal Transformer encoder to process electrical quantity time series data, establish a Transformer distribution state prediction model. Introduce a TCN time convolutional block in the model to capture long-term dependencies, and add a spatial attention mechanism to locate abnormal device nodes to obtain a Transformer-TCN distribution state prediction model; Input the fused distribution operation and maintenance data into the Transformer-TCN distribution state prediction model for identification, and output the distribution operation and maintenance health index; Judge the state of distribution equipment according to the distribution operation and maintenance health index, and use the Bayesian network to combine historical fault cases and environmental factors to evaluate the cause of anomalies.

2. The power distribution intelligent operation and maintenance management method based on data processing according to claim 1, characterized in that, The preprocessing of the multi-source distribution operation and maintenance data, and the use of the time series alignment algorithm to align data with different sampling frequencies in the preprocessed data, include: Detect outliers in the multi-source distribution operation and maintenance data based on the Isolation Forest algorithm, delete the outliers, and then fill them using linear interpolation to obtain filled distribution operation and maintenance data; Perform time series alignment on data with different sampling frequencies through the DTW (Dynamic Time Warping) algorithm, and convert time series data with different sampling frequencies into sequences indexed by timestamps; Calculate the distance matrix between two sequences using the Euclidean distance; find the minimum cumulative distance path from the upper left corner to the lower right corner of the matrix through dynamic programming to determine the alignment relationship of the time series; Resample the data according to the aligned timestamps to obtain aligned distribution operation and maintenance data.

3. The distribution intelligent operation and maintenance management method based on data processing according to claim 2, characterized in that, The conversion of the aligned time series data into two-dimensional feature images through Gramian angular field transformation to obtain fused distribution operation and maintenance data further includes: Normalize the aligned distribution operation and maintenance data using the min-max normalization method, and map the data to the interval [−1, 1] to obtain normalized data; Calculate the angle corresponding to each data point in the normalized data, convert each data point into an angle in polar coordinates, and regard it as a point on the unit circle; Generate two two-dimensional images, including a cosine Gramian angular field and a sine Gramian angular field, by calculating the cosine and sine combination relationships of angles at different time points to obtain fused distribution operation and maintenance data.

4. The power distribution intelligent operation and maintenance management method based on data processing according to claim 1, wherein, The establishment of a Transformer distribution state prediction model based on the spatio-temporal Transformer encoder to process electrical quantity time series data includes: The spatio-temporal Transformer encoder consists of multiple Transformer encoding layers. Each layer contains a multi-head self-attention mechanism and a feed-forward neural network, and position encoding is introduced; The input data is mapped to multiple subspaces using multi-head self-attention, the dependencies of different time steps and device nodes within each subspace are calculated independently, and then the outputs are concatenated to capture the spatio-temporal correlations of the data; The input is the aligned electrical quantity time series data, and the interaction features of different time steps and device nodes are extracted by multi-head self-attention.

5. The power distribution intelligent operation and maintenance management method based on data processing according to claim 1, characterized in that Introducing a TCN (Temporal Convolutional Network) time convolutional block in the model to capture long-term dependencies, and adding a spatial attention mechanism to locate abnormal device nodes, resulting in a Transformer-TCN power distribution status prediction model, further including: The TCN time convolutional block includes dilated causal convolution and residual connection structures. The receptive field of the convolutional kernel is enlarged through dilated causality to capture long-term dependencies without increasing the number of parameters; The input is directly stacked on the output of the convolutional layer using residual connection. Each block contains three layers of dilated convolution, and after stacking, a deep network is obtained; The features output by the Transformer are averaged and max-pooled in the time dimension to generate two spatial feature descriptors, and the original features are weighted using attention weights to highlight the features of abnormal nodes.

6. A power distribution intelligent operation and maintenance management method based on data processing according to claim 1, characterized in that Inputting the fused power distribution operation and maintenance data into the Transformer-TCN power distribution status prediction model for identification, and outputting a power distribution operation and maintenance health index, including: Inputting the fused power distribution operation and maintenance data into the Transformer-TCN power distribution status prediction model for identification, and extracting local features of image edges and textures through a convolutional layer; Integrating spatio-temporal and long-term time features through the Transformer-TCN module to capture the complex patterns of device operation, and mapping to an evaluation value between 0 and 100 through a fully connected layer to obtain the power distribution operation and maintenance health index.

7. A power distribution intelligent operation and maintenance management method based on data processing according to claim 1, characterized in that, Judging the status of power distribution equipment according to the power distribution operation and maintenance health index, and using a Bayesian network to combine historical fault cases and environmental factors to evaluate the causes of anomalies, including: When it is judged that the device is in a warning state, collect the current operating data, environmental parameters, and historical fault information as evidence; Use the Bayesian inference algorithm to calculate the posterior probabilities of each abnormal cause, and output the abnormal causes sorted by the posterior probabilities.

8. A distribution intelligent operation and maintenance management system based on data processing, characterized in that, The power distribution intelligent operation and maintenance management system includes the following modules: A power distribution data acquisition module, which is used to collect current waveforms, voltage waveforms, partial discharge data, temperature and humidity data, geographical information data, and historical fault data in power distribution terminals, electric meters, inspection robots, and user work order systems to obtain multi-source power distribution operation and maintenance data; A power distribution data processing module, which is used to preprocess the multi-source power distribution operation and maintenance data, align data with different sampling frequencies in the preprocessed data using a time series alignment algorithm, and convert the aligned time series data into a two-dimensional feature image through Gramian Angular Field transformation to obtain fused power distribution operation and maintenance data; A prediction model establishment module, which is used to establish a Transformer power distribution status prediction model based on processing electrical quantity time series data by a spatio-temporal Transformer encoder, introduce a TCN time convolutional block in the model to capture long-term dependencies, and add a spatial attention mechanism to locate abnormal device nodes to obtain a Transformer-TCN power distribution status prediction model; The power distribution health identification module is used to input the fused power distribution operation and maintenance data into the Transformer-TCN power distribution status prediction model for identification, and output the power distribution operation and maintenance health index; The power distribution operation and maintenance management module is used to judge the status of power distribution equipment according to the power distribution operation and maintenance health index, and use the Bayesian network to combine historical fault cases and environmental factors to evaluate the abnormal reasons.

9. A power distribution intelligent operation and maintenance management system based on data processing according to claim 8, characterized in that, The power distribution data processing module includes the following sub-modules: The processing sub-module is used to perform normalization processing on the power distribution operation and maintenance data aligned by using the min-max normalization method, map the data to the interval [−1, 1], and obtain the normalized data; The calculation sub-module is used to calculate the angle corresponding to each data point in the normalized data, convert each data point into an angle in polar coordinates, and regard it as a point on the unit circle; The obtaining sub-module is used to generate two two-dimensional images, including the cosine Gram angle field and the sine Gram angle field, by calculating the cosine and sine combination relationships of angles at different time points, and obtain the fused power distribution operation and maintenance data.

10. A power distribution intelligent operation and maintenance management system based on data processing as claimed in claim 8, characterized in that, The prediction model establishment module includes the following sub-modules: The introduction sub-module is used for the spatio-temporal Transformer encoder, which consists of multiple Transformer encoding layers, each layer contains a multi-head self-attention mechanism and a feed-forward neural network, and position encoding is introduced; The mapping sub-module is used to map the input data to multiple sub-spaces by using multi-head self-attention, independently calculate the dependency relationships of different time steps and device nodes in each sub-space, and then splice the outputs to capture the spatio-temporal correlation of the data; The input sub-module is used for the input to be the aligned electrical quantity time series data, and the interaction features of different time steps and device nodes are extracted by multi-head self-attention.

Images