Fault analysis method and device for power distribution equipment, equipment and storage medium

Abstract

The invention discloses a fault analysis method, device and equipment for power distribution equipment and a storage medium, and the method comprises the steps: obtaining historical alarm data and geographical environment data of the power distribution equipment, and determining an environment interference change trend based on the historical alarm data and the geographical environment data; wherein the environment interference change trend is used for representing a dynamic influence trend of each geographical environment factor on power distribution equipment alarm generation; performing feature extraction on the historical alarm data by adopting a convolutional neural network according to the environmental interference change trend to obtain comprehensive feature representation; training the convolutional neural network by adopting comprehensive feature representation to obtain a trained fault prediction model; acquiring real-time alarm data, inputting the real-time alarm data into the fault prediction model, and outputting a preliminary fault reason; performing time sequence analysis on the preliminary fault reason by adopting a long short-term memory network to obtain a fault diagnosis result; according to the invention, the accuracy of fault reason identification and the intelligent level of alarm analysis can be improved, and reliable operation of a power system is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power systems, and in particular, to a method, device, equipment and storage medium for fault analysis of distribution equipment. Background Art

[0002] Alarm analysis of the distribution communication network can timely identify equipment failures and reduce power outage risks, which is crucial for ensuring the stable operation of the power system. With the rapid development of smart grids, alarm analysis methods based on deep learning have gradually become a research hotspot, providing new possibilities for improving the accuracy and efficiency of fault diagnosis.

[0003] However, due to the influence of geographical environmental factors, the alarm characteristics of outdoor equipment show significant dynamics. Existing deep learning-based solutions tend to focus on static feature modeling, resulting in insufficient adaptability of the model in dynamic scenarios. It is difficult to accurately capture the complex relationship between fault causes and environmental factors, and the analysis results often deviate from reality. There is a problem of decreasing accuracy in identifying the causes of outdoor equipment failures, which affects the intelligent level of distribution communication network alarm analysis and the reliable operation of the power system. Summary of the Invention

[0004] Embodiments of the present invention provide a method for fault analysis of distribution equipment, which can improve the accuracy of identifying fault causes, enhance the intelligent level of distribution communication network alarm analysis, and ensure the reliable operation of the power system.

[0005] In a first aspect, embodiments of the present invention provide a method for fault analysis of distribution equipment, including: Obtain historical alarm data and geographical environment data of the distribution equipment, and determine the environmental interference change trend based on the historical alarm data and the geographical environment data; wherein, the environmental interference change trend is used to represent the dynamic influence trend of each geographical environmental factor on the occurrence of alarms of the distribution equipment; According to the environmental interference change trend, use a convolutional neural network to extract features from the historical alarm data to obtain a comprehensive feature representation; Use the comprehensive feature representation to train the convolutional neural network to obtain a trained fault prediction model; Obtain real-time alarm data and input it into the fault prediction model to output preliminary fault causes; Use a long short-term memory network to perform temporal analysis on the preliminary fault causes to obtain a fault diagnosis result.

[0006] Further, the obtaining of the historical alarm data and geographical environment data of the distribution equipment, and determining the environmental interference change trend based on the historical alarm data and the geographical environment data includes: Obtain the historical alarm data of the power distribution equipment, as well as the geographical environment data of the environmental area where the power distribution equipment is located collected by each sensor; wherein, the geographical environment data includes, but is not limited to, environmental temperature data, environmental humidity data, wind speed data, and vegetation coverage rate; According to the geographical environment data, use the random forest algorithm to calculate the equipment exposure distribution of the power distribution equipment in different seasons; Based on the equipment exposure distribution and the historical alarm data, calculate the interference intensity of the equipment exposure of the power distribution equipment on the alarm frequency; Use the clustering algorithm to divide the environmental area according to geographical coordinates and the interference intensity into high-impact areas, medium-impact areas, and low-impact areas; Calculate the environmental interference change trend within a preset time period for each impact area; wherein, the environmental interference change trend is used to represent the dynamic impact trend of geographical environment factors on the occurrence of power distribution equipment alarms.

[0007] Further, according to the environmental interference change trend, use a convolutional neural network to extract features from the historical alarm data to obtain a comprehensive feature representation, including: Use a convolutional neural network to extract features from the historical alarm data to obtain a feature map; Based on the historical alarm data and the feature map, use the attention mechanism to calculate the influence weights of different geographical environment factors on the alarm; According to the influence weights, perform weighted fusion on the feature map to obtain a weighted fusion comprehensive feature representation.

[0008] Further, the use of a convolutional neural network to extract features from the historical alarm data to obtain a feature map includes: Use a convolutional neural network to organize the historical alarm data into a feature expression matrix according to the time dimension and the space dimension; wherein, the matrix element value of the feature expression matrix is the number of alarms at a specific geographical location within a specific time period; Analyze the correlation between the alarm frequency and the geographical environment based on the feature expression matrix and the geographical environment data; Adjust the convolution kernel size of the convolutional neural network according to the correlation, and use the adjusted convolutional neural network to process the feature expression matrix to obtain an adjusted feature map.

[0009] Further, training the convolutional neural network with the comprehensive feature representation to obtain a trained fault prediction model includes: Input the feature representation into a convolutional neural network for training to obtain the first probability distribution of historical alarms; wherein, the first probability distribution is the fault probability of each fault type of the historical alarms. When the fault probability of a certain fault type is greater than a preset fault probability threshold, update the model parameters of the convolutional neural network based on the changing trend of the environmental interference to obtain a fault prediction model.

[0010] Further, obtaining real-time alarm data and inputting it into the fault prediction model to output a preliminary fault cause, including: Obtain the real-time alarm data and geographical environment data of the power distribution equipment, and input them into the fault prediction model to output the second probability distribution corresponding to the alarm; wherein, the second probability distribution is the fault probability of each fault type of the real-time alarm predicted by the fault prediction model. Determine the maintenance difficulty index and fault driving factors corresponding to each fault type in the second probability distribution; wherein, the fault driving factors are the main driving factors causing the alarm of the power distribution equipment. Calculate the contribution degree of the fault driving factors to the fault type, and adjust the second probability distribution according to the contribution degree. Obtain the preliminary fault cause according to the adjusted second probability distribution.

[0011] Further, perform temporal analysis on the preliminary fault cause using a long short-term memory network to obtain a fault diagnosis result, including: Obtain the geographical environment data within a preset time period before the occurrence of the real-time alarm, and analyze the geographical environment data using a long short-term memory network to output environmental temporal features. Obtain the fault diagnosis result according to the environmental temporal features and the preliminary fault cause.

[0012] In a second aspect, an embodiment of the present invention provides a fault analysis device for power distribution equipment, including: A data acquisition module, configured to acquire historical alarm data and geographical environment data of power distribution equipment, and determine the changing trend of environmental interference based on the historical alarm data and the geographical environment data; wherein, the changing trend of environmental interference is used to represent the dynamic influence trend of each geographical environment factor on the occurrence of power distribution equipment alarms. A feature extraction module, configured to extract features from the historical alarm data using a convolutional neural network according to the changing trend of environmental interference to obtain a comprehensive feature representation. A model training module, configured to train the convolutional neural network using the comprehensive feature representation to obtain a trained fault prediction model. A fault prediction module, configured to obtain real-time alarm data and input the data into the fault prediction model to output a preliminary fault cause. A fault diagnosis module, configured to perform time series analysis on the preliminary fault cause by using a long short-term memory network to obtain a fault diagnosis result.

[0013] In a third aspect, an embodiment of the present invention provides an electronic device, including: A memory, configured to store a computer program; A processor, configured to execute the computer program; Wherein, when the processor executes the computer program, the fault analysis method of the power distribution device described in any item of the first aspect above is implemented.

[0014] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed, the fault analysis method of the power distribution device described in any item of the first aspect above is implemented.

[0015] Compared with the prior art, the beneficial effects of the fault analysis method of the power distribution device provided by the embodiment of the present invention are as follows: obtaining historical alarm data and geographical environment data of the power distribution device, and determining an environmental interference change trend based on the historical alarm data and the geographical environment data; wherein, the environmental interference change trend is used to represent the dynamic influence trend of each geographical environment factor on the occurrence of alarms of the power distribution device; according to the environmental interference change trend, using a convolutional neural network to extract features from the historical alarm data to obtain a comprehensive feature representation; using the comprehensive feature representation to train the convolutional neural network to obtain a trained fault prediction model; obtaining real-time alarm data and inputting the data into the fault prediction model to output a preliminary fault cause; performing time series analysis on the preliminary fault cause by using a long short-term memory network to obtain a fault diagnosis result; the present invention can improve the accuracy of fault cause identification, improve the intelligent level of alarm analysis of the power distribution communication network, and ensure the reliable operation of the power system. Description of the Drawings

[0016] In order to more clearly illustrate the technical features of the embodiments of the present invention, the following will briefly introduce the drawings required to be used in the embodiments of the present invention. Obviously, the following described drawings are only some embodiments of the present invention, and those skilled in the art can obtain other drawings according to these drawings without creative efforts.

[0017] Figure 1 It is a schematic flowchart of an embodiment of a fault analysis method of a power distribution device provided by the present invention; Figure 2It is a schematic structural diagram of an embodiment of a fault analysis device for a power distribution device provided by the present invention; Figure 3 It is a schematic structural diagram of an embodiment of an electronic device provided by the present invention. Detailed implementation manners

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] It should be noted that although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from the module division in the device or the order in the flowchart. Terms such as "first" and "second" in the description, claims, and the above-mentioned drawings are used to distinguish similar objects and do not necessarily need to describe a specific order or sequence.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs. The terms used herein are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0021] In a first aspect, an embodiment of the present invention provides a fault analysis method for a power distribution device. Refer to Figure 1 , which is a schematic flowchart of an embodiment of a fault analysis method for a power distribution device provided by the present invention.

[0022] As Figure 1 shown, the method includes the following steps: S1: Obtain historical alarm data and geographical environment data of the power distribution device, and determine the environmental interference change trend based on the historical alarm data and the geographical environment data; wherein, the environmental interference change trend is used to represent the dynamic influence trend of each geographical environment factor on the occurrence of alarms of the power distribution device; S2: According to the environmental interference change trend, use a convolutional neural network to extract features from the historical alarm data to obtain a comprehensive feature representation; S3: Use the comprehensive feature representation to train the convolutional neural network to obtain a trained fault prediction model; S4: Obtain real-time alarm data and input it into the fault prediction model to output a preliminary fault cause; S5: Perform temporal analysis on the preliminary fault causes using a long short-term memory network to obtain a fault diagnosis result.

[0023] In specific implementation, first obtain the historical alarm data and geographical environment data of the power distribution equipment. The historical alarm data may include information such as the time, type, and severity of equipment failures. The geographical environment data may include factors such as temperature, humidity, air pressure, and altitude. By analyzing the historical alarm data and geographical environment data, the dynamic influence trend of each geographical environment factor on the occurrence of power distribution equipment alarms can be understood, thereby determining the environmental interference change trend. For example, an increase in temperature may lead to an increase in the probability of certain equipment failures, and changes in humidity may affect the insulation performance of the equipment, etc.

[0024] According to the environmental interference change trend, use a convolutional neural network to extract features from the historical alarm data, and more representative comprehensive feature representations can be obtained. These comprehensive feature representations can reflect the potential patterns and rules in the historical alarm data, providing strong support for subsequent fault prediction and diagnosis. Use these feature representations to train the convolutional neural network to obtain a trained fault prediction model. During the training process, the historical alarm data can be used as the input, and the corresponding fault type probability can be used as the output. Continuously adjust the parameters of the convolutional neural network through an optimization algorithm to enable the model to accurately predict the fault type probability.

[0025] After obtaining the trained fault prediction model, obtain the real-time alarm data and input it into the fault prediction model to obtain a preliminary prediction of the fault cause. For example, when the real-time alarm data shows that the equipment has an overcurrent alarm, the fault prediction model can predict the possible fault causes, such as equipment overload, short circuit, etc., based on the feature representations of the historical alarm data and environmental data. Perform temporal analysis on the preliminary fault causes using a long short-term memory network. Use the preliminary fault causes as the input, and after the processing of the long short-term memory network, obtain a more accurate fault diagnosis result.

[0026] In summary, the present invention obtains historical alarm data and geographical environment data of a power distribution device, and determines an environmental interference change trend based on the historical alarm data and the geographical environment data; wherein, the environmental interference change trend is used to represent the dynamic influence trend of each geographical environment factor on the occurrence of alarms of the power distribution device; according to the environmental interference change trend, a convolutional neural network is used to extract features from the historical alarm data to obtain a comprehensive feature representation; the feature representation is used to train the convolutional neural network to obtain a trained fault prediction model; real-time alarm data is obtained and input into the fault prediction model to output a preliminary fault cause; a long short-term memory network is used to perform temporal analysis on the preliminary fault cause to obtain a fault diagnosis result; the present invention considers the real-time influence of outdoor environmental factors, can improve the accuracy of fault cause identification, has high accuracy and reliability, can improve the intelligent level of alarm analysis of a power distribution communication network, provides strong support for the operation and maintenance of a smart grid, and thus ensures the reliable operation of a power system.

[0027] In an alternative embodiment, the obtaining of the historical alarm data and geographical environment data of the power distribution device and determining the environmental interference change trend based on the real-time alarm data and environmental data includes: Obtain the historical alarm data of the power distribution device and the geographical environment data of the environment area where the power distribution device is located collected by each sensor; wherein, the geographical environment data includes but is not limited to environmental temperature data, environmental humidity data, wind speed data and vegetation coverage rate. According to the geographical environment data, use the random forest algorithm to calculate the device exposure distribution of the power distribution device in different seasons. Based on the device exposure distribution and the historical alarm data, calculate the interference intensity of the device exposure of the power distribution device on the alarm frequency. Use a clustering algorithm to divide the environment area according to geographical coordinates and the interference intensity into a high-impact area, a medium-impact area and a low-impact area. Calculate the environmental interference change trend within a preset time period for each impact area; wherein, the environmental interference change trend is used to represent the dynamic influence trend of geographical environment factors on the occurrence of alarms of the power distribution device.

[0028] Specifically, obtain the historical alarm data of outdoor equipment in the distribution communication network within a preset time period, such as alarm data for alarms like voltage anomalies or communication interruptions, as well as the geographical environment data of the environmental area where the distribution equipment is located collected by each sensor. The sensors can be deployed on distribution poles or communication base stations to monitor environmental parameters such as temperature, humidity, and wind speed in real time. By combining remote sensing images with ground sensor data, the vegetation coverage rate around the equipment can be judged. For example, in summer, the vegetation is lush and the coverage rate reaches 70%, which may block the ventilation of the equipment and cause poor heat dissipation, while in winter, the coverage rate drops to 30%.

[0029] Adopt the random forest algorithm to calculate the equipment exposure distribution of distribution equipment in different seasons. Equipment exposure can be understood as the degree to which distribution equipment is affected by various factors in a specific geographical environment. For example, in an environment with a high vegetation coverage rate, the equipment exposure may be lower, while in an environment with a low vegetation coverage rate, the equipment exposure may be higher. Since the geographical environment factors change in different seasons, the equipment exposure distribution of distribution equipment will also be different. The random forest is an ensemble learning algorithm that improves the accuracy and stability of prediction by constructing multiple decision trees and combining their prediction results. Exemplarily, the algorithm votes through a large number of decision trees to obtain the rule that the equipment exposure decreases by 20% in summer and increases by 15% in winter.

[0030] Based on the equipment exposure distribution and historical alarm data, determine the interference intensity of equipment exposure on the alarm frequency. The interference intensity reflects the degree of influence of equipment exposure on the alarm frequency of distribution equipment. If the interference intensity is high, it means that the equipment exposure has a greater impact on the alarm frequency, and corresponding measures need to be taken to reduce the equipment exposure to reduce the occurrence of alarms. If the interference intensity is low, it means that the equipment exposure has a smaller impact on the alarm frequency, and the control of equipment exposure can be appropriately relaxed. Exemplarily, when it is found that the exposure is higher than 0.6, the alarm frequency increases by 30%, indicating that the high-exposure area is more strongly interfered by sand and dust or low temperature. It can be understood that when the interference intensity exceeds the preset interference intensity threshold, the support vector machine algorithm is used to judge the dynamic change trend of the interference intensity. The support vector machine takes time series data as input and divides the trend characteristics of the rising or falling interference intensity. For example, through the support vector machine algorithm, it can be predicted that the interference intensity in a certain area will increase by 10% in the next 3 months. The advantage of this method is that it can quickly classify complex data and is suitable for dynamic environment analysis.

[0031] The clustering algorithm is used to divide the environmental area according to geographical coordinates and interference intensity, which are divided into high-impact areas, medium-impact areas, and low-impact areas. The interference intensity in high-impact areas is relatively high, which may be concentrated on the mountaintops with sparse vegetation. Power distribution equipment is more vulnerable to geographical environmental factors in these areas, and the alarm frequency is also relatively high. The interference intensity in medium-impact areas is moderate, and the alarm frequency of power distribution equipment in these areas is also at a medium level. The interference intensity in low-impact areas is relatively low, and power distribution equipment is less affected by geographical environmental factors in these areas, and the alarm frequency is also relatively low.

[0032] Calculate the environmental interference change trend within a preset time period for each impact area. Exemplarily, if the recent alarms in a high-impact area suddenly increase by 20%, it is concluded that the environmental interference change trend is deteriorating. Thus, the high-interference area can be pre-warned according to the environmental interference change trend, reducing the equipment failure rate and optimizing the allocation of maintenance resources.

[0033] In an optional implementation manner, according to the environmental interference change trend, a convolutional neural network is used to extract features from the historical alarm data to obtain a comprehensive feature representation, including: Using a convolutional neural network to extract features from the historical alarm data to obtain a feature map; Based on the historical alarm data and the feature map, an attention mechanism is used to calculate the influence weights of different geographical environmental factors on the alarm; According to the influence weights, the feature map is weighted and fused to obtain a weighted and fused comprehensive feature representation.

[0034] Specifically, using a convolutional neural network to extract features from historical alarm data to obtain a feature map, combining the historical alarm data with the feature map, performing attention mechanism processing, and analyzing the influence weights of seasonal vegetation coverage, equipment exposure fluctuation, and environmental interference distribution on the alarm characteristic change pattern. Exemplarily, assume that when the vegetation coverage rate is high in summer, the alarm frequency decreases by 15%, while when the equipment exposure is high, the alarm frequency increases by 20%. The attention mechanism will assign a weight of 0.6 to the exposure and a weight of 0.3 to the vegetation. This weight distribution intuitively reflects the influence intensity of each factor. If the influence weight exceeds the preset influence weight threshold, the weight distribution is adjusted through the attention mechanism to obtain an updated weight distribution.

[0035] According to the updated weight distribution, the feature maps of seasonal vegetation, equipment exposure, and environmental interference are weighted and fused to obtain a comprehensive feature representation. Exemplarily, if the weights of seasonal vegetation, equipment exposure, and environmental interference are 0.25, 0.65, and 0.1 respectively, the vegetation feature map can be multiplied by 0.25, the exposure feature map by 0.65, and the interference feature map by 0.1 to obtain the comprehensive feature representation. This weighting method highlights the main influencing factors and facilitates the subsequent analysis of the change patterns of alarm characteristics.

[0036] In an alternative embodiment, the use of a convolutional neural network to extract features from the historical alarm data to obtain a feature map includes: Using a convolutional neural network, organizing the historical alarm data into a feature expression matrix according to the time dimension and the spatial dimension; wherein, the matrix element value of the feature expression matrix is the number of alarms at a specific geographical location within a specific time period; Analyzing the correlation between the alarm frequency and the geographical environment based on the feature expression matrix and the geographical environment data; Adjusting the convolution kernel size of the convolutional neural network according to the correlation, and using the adjusted convolutional neural network to process the feature expression matrix to obtain an adjusted feature map.

[0037] Specifically, historical alarm data may contain time series. The convolutional neural network analyzes these series through a sliding window to extract the periodic features of abnormal occurrences. For example, a certain device records 10 alarms per day in summer. The network may identify the characteristic patterns during high-temperature periods, such as the concentration of alarms from 12:00 noon to 3:00 pm. This method can effectively mine hidden change rules. By combining the change patterns in the alarm data with geographical environment factors, a more comprehensive understanding of the relationship between device alarms and the environment can be obtained. For example, when the temperature in a certain area reaches 35 degrees and the humidity is 80% in summer, the alarm pattern shows frequent overheating problems, while the alarms decrease when the temperature is 5 degrees in winter. After integration, a dynamic trend that high temperature and high humidity have a greater impact on the device can be obtained. By taking environmental factors such as temperature, vegetation coverage rate, and wind speed as inputs, the environmental data and alarm data are integrated into a multi-dimensional vector to generate a feature expression. By combining the feature expression with dynamic changes, an expression matrix can be generated, which can intuitively reflect dynamic changes such as seasonal fluctuations in the device state. The support vector machine takes the feature matrix as input and classifies the intensity levels, including high intensity and low intensity. Exemplarily, when the exposure is 0.6, the alarm frequency increases by 30% and is classified as high intensity. When the alarm frequency exceeds a preset threshold, such as more than 50 times per month, the support vector machine can quickly quantify the impact of the change. By combining the intensity distribution with the geographical environment, areas with a greater environmental impact can be accurately located. For example, the mountaintop area with high exposure and strong wind speed in hilly areas is classified as a high-dynamic-change area with frequent alarms. This classification can provide specific target areas for optimizing maintenance strategies and improve maintenance efficiency. When the recent alarms in the high-dynamic-change area suddenly increase, the corresponding eigenvalue in the matrix is up-regulated. This update mechanism can reflect the change in the device state in real time and ensure the accuracy of the analysis. By continuously updating the expression matrix, the change in the device state can be timely grasped.

[0038] It can be understood that the feature expression matrix can reflect the spatio-temporal distribution law of alarms. By extracting data related to the geographical environment from the feature expression matrix and determining the change trend, the impact of the geographical environment on alarms can be captured. Specifically, column data related to rainfall and terrain height can be extracted from the matrix. For example, when the rainfall in a certain area increases by 20 millimeters, the alarm frequency increases by 30%. By comparing the data in different time periods, the change trend of the alarm frequency is obtained, and the convolutional kernel size is adjusted according to the change trend to obtain the depth adjustment parameter. Exemplarily, if the change trend shows that the alarms are concentrated in a local area, the convolutional kernel size can be adjusted from 5×5 to 3×3 to focus on the detailed features. If the trend is relatively flat, a larger size such as 7×7 is maintained, and the depth adjustment parameter is set accordingly to the pixel range covered by the convolutional kernel. The hierarchical structure is updated using the depth adjustment parameter to obtain an adaptive feature extraction model, which can improve the adaptability of the model to dynamic data. Exemplarily, small convolutional kernels are used in the shallow network to extract local features, and large convolutional kernels are fused in the deep network to capture the global trend, forming a model containing three layers of convolution. The parameters of each layer are dynamically updated according to the adjustment results. By processing the matrix structure with the adaptive feature extraction model, a feature map after breadth adjustment can be obtained, which can further expand the expression ability of the features. If the change trend exceeds the preset threshold, the feature map is subjected to environmental modulation to obtain the final feature map. Exemplarily, when the growth rate of the alarm frequency exceeds 50%, the feature map can be modulated by weighting the rainfall data, such as increasing the feature value of the area with larger rainfall by 20%, so as to highlight the environmental impact. By adjusting steps such as the convolutional kernel size and environmental modulation, the model can be made more flexible to adapt to alarm changes in different geographical environments.

[0039] In an alternative embodiment, training the convolutional neural network using the comprehensive feature representation to obtain a trained fault prediction model includes: Inputting the feature representation into the convolutional neural network for training to obtain the first probability distribution of historical alarms; wherein, the first probability distribution is the fault probability of each fault type of historical alarms; When the fault probability of a certain fault type is greater than the preset fault probability threshold, updating the model parameters of the convolutional neural network based on the change trend of the environmental interference to obtain the fault prediction model.

[0040] Specifically, the feature representation is input into a convolutional neural network for training. The convolutional neural network can automatically learn the complex relationships between different features, thereby better extracting the probability distribution of historical faults and obtaining the first probability distribution of historical alarms, that is, the fault probability of each fault type in historical alarms. Exemplarily, there are 1000 device fault records in the historical alarm data. After training, the network outputs the probability of each fault type. For example, the probability of a circuit short circuit is 0.6, and the probability of mechanical wear is 0.3. This probability distribution reflects the potential laws of faults and helps to identify risk points in advance.

[0041] Furthermore, assume that the preset circuit short circuit probability threshold is 0.4, and the actual probability reaches 0.6, indicating an abnormal situation. In this case, geographical data is introduced, such as the rainfall or altitude of the area where the device is located, which can further reveal the impact of environmental changes on device faults. For example, if the rainfall increases from 50 millimeters per month to 200 millimeters, it may cause abnormal humidity, which in turn affects the fault probability. Through time series analysis, it can be determined whether the fault probability increases with the increase of geographical environmental factors such as rainfall. For example, the data for three consecutive months shows that the probability increases from 0.4 to 0.7, and the trend is clearly upward. The model parameters such as the convolutional kernel size or learning rate can be dynamically adjusted according to the environmental change trend. When the trend is upward, the learning rate is adjusted from 0.01 to 0.005 to more precisely adapt to the new data. This adjustment can enhance the model's response ability to the dynamic environment and make it better adapt to the changing actual situation. The convolutional neural network after updating the model parameters is the fault prediction model, and the fault prediction model can be used for fault prediction.

[0042] In an alternative embodiment, the obtaining of real-time alarm data and inputting it into the fault prediction model to output the preliminary fault cause includes: Obtain the real-time alarm data and geographical environment data of the power distribution equipment and input them into the fault prediction model to output the second probability distribution corresponding to the alarm; wherein, the second probability distribution is the fault probability of each fault type of the real-time alarm predicted by the fault prediction model; Determine the maintenance difficulty index and fault driving factors corresponding to each fault type in the second probability distribution; wherein, the fault driving factor is the main driving factor that causes the power distribution equipment to issue an alarm; Calculate the contribution degree of the fault driving factor to the fault type, and adjust the second probability distribution according to the contribution degree; Obtain the preliminary fault cause according to the adjusted second probability distribution.

[0043] Specifically, by processing real-time alarm data through a fault prediction model, when a certain power equipment is operating, the real-time alarm data may include information such as abnormal voltage and current fluctuations. The fault prediction model analyzes these data, extracts the deviation characteristics of equipment exposure and environmental interference, and generates a second probability distribution, that is, the fault probability of each fault type in the real-time alarm. For example, assume that a certain transformer is operating in a high-temperature environment and its real-time alarm data is input into the fault prediction model. The model identifies that the probability of overheating fault is 70%, while the probability of short-circuit fault is 20%.

[0044] It should be noted that after obtaining the second probability distribution, it is necessary to determine the maintenance difficulty index and fault driving factors corresponding to each fault type. The clustering analysis method can be used to group the second probability distribution of fault types. The second probability distribution of fault types can be divided into two categories: high risk and low risk. The trend of the change in the maintenance difficulty index can be obtained from the grouping results. For example, for an overheating fault caused by an increase in temperature, the maintenance difficulty index may increase from 3 to 5 due to the deterioration of the equipment's heat dissipation conditions. The fault driving factor is the main driving factor that causes the distribution equipment to issue an alarm. For example, an increase in temperature may be the main driving factor for an overheating fault.

[0045] The contribution degree of the fault driving factor to the fault type is calculated through regression analysis. For example, it is calculated that for every 1-degree increase in temperature, the probability of overheating fault increases by 5%. This can accurately reflect the influence degree of the fault driving factor on the fault type in the current scenario. The second probability distribution is adjusted according to the calculated contribution degree to make the probability distribution more accurately reflect the current scenario. The adjusted probability distribution can improve the pertinence of the prediction and provide a more reliable basis for subsequent fault cause analysis.

[0046] According to the adjusted second probability distribution, the mapping relationship between the fluctuation of equipment exposure and the cause distribution is analyzed to determine the preliminary fault cause. For example, when the equipment exposure of a certain equipment increases from 60% to 80% due to sandy weather, the main fault cause changes from overheating to insulation aging. This mapping relationship helps to trace the root cause of the fault.

[0047] In an alternative implementation manner, the long short-term memory network is used to perform temporal analysis on the preliminary fault cause to obtain a fault diagnosis result, including: Obtain the geographical environment data within a preset time period before the occurrence of the real-time alarm, and use the long short-term memory network to analyze the geographical environment data to output environmental temporal features; According to the environmental temporal features and the preliminary fault cause, obtain the fault diagnosis result.

[0048] Specifically, obtain the geographical environment data within a preset time period before the occurrence of real-time alarms, and use a long short-term memory network to analyze the data and output environmental time series features. The long short-term memory network can analyze the time series patterns of geographical environment data. For example, in the device operation scenario, assume that a certain device records data continuously for 30 days, and the daily light angle offset range is from 5 degrees to 15 degrees, and the humidity level fluctuates between 60% and 80%. The long short-term memory network can analyze these time series patterns of the data and output initial time series features, such as the periodic peak of the light angle offset appears at 10 am, and the humidity level reaches the maximum value at night. This method can effectively identify the dynamic rules of the data and provide a basis for subsequent fault diagnosis.

[0049] Extract factors with long-term change trends such as the dynamic offset of the light angle and the periodic change of the humidity level from the environmental time series features. For the analysis of the light angle offset, the frequency and amplitude of the offset can be concerned. For example, assume that the light angle offsets 10 degrees every day, and there are 2 days in a week with an offset exceeding 12 degrees, which can be judged as a dynamic offset caused by external occlusion. For the analysis of the periodic change of the humidity level, the change trend of the humidity level and the impact of the humidity level on the device can be understood. For example, the humidity rises sharply from 70% to 85% during the rainy season and then drops after 3 days.

[0050] It can be understood that the vegetation density will change the light and humidity environment of the device with the seasons. For example, if the vegetation density increases by 20% in summer, it may reduce the light angle offset by 5 degrees and increase the humidity by 10%. By analyzing the seasonal modulation effect of the vegetation density change, the impact of the vegetation density change on the light and humidity environment of the device and the potential impact on the device failure can be understood. If the modulation influence intensity exceeds the standard, the long short-term memory network will adjust the time series data processing flow, allocate more weights to the short-term fluctuations of the humidity change, and reduce the long-term dependence on the light offset, resulting in the corrected time series correlation.

[0051] According to the corrected time series correlation and combined with the fault cause distribution, obtain the fault diagnosis result. Exemplarily, if the humidity periodically exceeds 80% and lasts for 5 days, and the fault probability rises from 10% to 30%, the fault diagnosis result is a short circuit of the circuit caused by excessive humidity.

[0052] In a second aspect, an embodiment of the present invention provides a fault analysis device for a power distribution device. Refer to Figure 2 , which is a schematic structural diagram of an embodiment of a fault analysis device for a power distribution device provided by the present invention.

[0053] As Figure 2 shown, the device includes: A data acquisition module 21, configured to acquire historical alarm data and geographical environment data of power distribution equipment, and determine an environmental interference change trend based on the historical alarm data and the environmental data; wherein, the environmental interference change trend is used to represent the dynamic influence trend of each geographical environment factor on the occurrence of alarms of power distribution equipment. A feature extraction module 22, configured to perform feature extraction on the historical alarm data by using a convolutional neural network according to the environmental interference change trend, so as to obtain a comprehensive feature representation. A model training module 23, configured to train the convolutional neural network by using the comprehensive feature representation to obtain a trained fault prediction model. A fault prediction module 24, configured to acquire real-time alarm data and input the real-time alarm data into the fault prediction model, and output a preliminary fault cause. A fault diagnosis module 25, configured to perform time series analysis on the preliminary fault cause by using a long short-term memory network to obtain a fault diagnosis result.

[0054] In an alternative embodiment, the data acquisition module 21 is further configured to: Acquire historical alarm data of power distribution equipment, and geographical environment data of the environmental area where the power distribution equipment is located collected by each sensor; wherein, the geographical environment data includes but is not limited to environmental temperature data, environmental humidity data, wind speed data, and vegetation coverage rate. According to the geographical environment data, use a random forest algorithm to calculate the equipment exposure distribution of the power distribution equipment in different seasons. Based on the equipment exposure distribution and the historical alarm data, calculate the interference intensity of the equipment exposure of the power distribution equipment on the alarm frequency. Use a clustering algorithm to divide the environmental area according to geographical coordinates and the interference intensity into a high-impact area, a medium-impact area, and a low-impact area. Calculate the environmental interference change trend within a preset time period for each impact area; wherein, the environmental interference change trend is used to represent the dynamic influence trend of geographical environment factors on the occurrence of alarms of power distribution equipment.

[0055] In an alternative embodiment, the feature extraction module 22 is further configured to: Use a convolutional neural network to perform feature extraction on the historical alarm data to obtain a feature map. Based on the historical alarm data and the feature map, use an attention mechanism to calculate the influence weights of different geographical environment factors on alarms. Perform weighted fusion on the feature map according to the influence weights to obtain a weighted fusion comprehensive feature representation.

[0056] In an alternative embodiment, the feature extraction module 22 is further configured to: Organize the historical alarm data into a feature expression matrix according to the time dimension and the space dimension by using a convolutional neural network; wherein, the matrix element value of the feature expression matrix is the number of alarms at a specific geographical location within a specific time period; Analyze the correlation between the alarm frequency and the geographical environment according to the feature expression matrix and the geographical environment data; Adjust the convolutional kernel size of the convolutional neural network according to the correlation, and process the feature expression matrix by using the adjusted convolutional neural network to obtain an adjusted feature map.

[0057] In an alternative embodiment, the model training module 23 is further configured to: Input the feature representation into a convolutional neural network for training to obtain a first probability distribution of historical alarms; wherein, the first probability distribution is the failure probability of each failure type of the historical alarms; When the failure probability of a certain failure type is greater than a preset failure probability threshold, update the model parameters of the convolutional neural network based on the change trend of the environmental interference to obtain a failure prediction model.

[0058] In an alternative embodiment, the failure prediction module 24 is further configured to: Obtain the real-time alarm data and geographical environment data of the power distribution equipment, and input them into the failure prediction model to output a second probability distribution corresponding to the alarm; wherein, the second probability distribution is the failure probability of each failure type of the real-time alarm predicted by the failure prediction model; Determine the maintenance difficulty index and failure driving factors corresponding to each failure type in the second probability distribution; wherein, the failure driving factor is the main driving factor that causes the power distribution equipment to generate an alarm; Calculate the contribution degree of the failure driving factor to the failure type, and adjust the second probability distribution according to the contribution degree; Obtain a preliminary failure cause according to the adjusted second probability distribution.

[0059] In an alternative embodiment, the failure diagnosis module 25 is further configured to: Obtain the geographical environment data within a preset time period before the occurrence of the real-time alarm, analyze the geographical environment data by using a long short-term memory network, and output environmental time series features; Obtain a failure diagnosis result according to the environmental time series features and the preliminary failure cause.

[0060] In a third aspect, an embodiment of the present invention provides an electronic device, see Figure 3As shown in the figure, it is a schematic structural diagram of an electronic device provided by an embodiment of the present invention.

[0061] As Figure 3 shown, the device includes: A memory 31 for storing a computer program; A processor 32 for executing the computer program; Wherein, when the processor 32 executes the computer program, it implements the fault analysis method of the power distribution device as described in any of the above embodiments.

[0062] Exemplarily, the computer program can be divided into one or more modules / units. The one or more modules / units are stored in the memory 31 and executed by the processor 32 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program in the electronic device.

[0063] The processor 32 can be a central processing unit (CPU), or can also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0064] The memory 31 can be used to store the computer programs and / or modules. By running or executing the computer programs and / or modules stored in the memory 31, and invoking the data stored in the memory 31, the processor 32 realizes various functions of the electronic device. The memory 31 mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function (such as a sound playback function, an image playback function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory 31 may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0065] It should be noted that the above-mentioned electronic device includes, but is not limited to, a processor and a memory. Those skilled in the art can understand that Figure 3 The structural schematic diagram is only an example of the above-mentioned electronic device, and does not constitute a limitation on the electronic device. It may include more components than shown in the figure, or combine some components, or different components.

[0066] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed, it implements the fault analysis method of the power distribution device described in any of the above embodiments.

[0067] It should be understood that all or part of the processes in implementing the fault analysis method of the above-mentioned power distribution device of the present invention can also be completed by instructing relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps of the fault analysis method of the above-mentioned power distribution device can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0068] The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. It should be pointed out that for those skilled in the art, several equivalent obvious variations and / or equivalent substitutions can be made without departing from the technical principles of the present invention. These obvious variations and / or equivalent substitutions should also be regarded as the protection scope of the present invention.

Claims

1. A fault analysis method for a power distribution device, characterized in that Including: Obtain the historical alarm data and geographical environment data of the power distribution equipment, and determine the environmental interference change trend based on the historical alarm data and the geographical environment data; wherein, the environmental interference change trend is used to represent the dynamic influence trend of each geographical environment factor on the occurrence of alarms of the power distribution equipment; According to the environmental interference change trend, use a convolutional neural network to extract features from the historical alarm data to obtain a comprehensive feature representation; Use the comprehensive feature representation to train the convolutional neural network to obtain a trained fault prediction model; Obtain real-time alarm data and input it into the fault prediction model to output a preliminary fault cause; Use a long short-term memory network to perform temporal analysis on the preliminary fault cause to obtain a fault diagnosis result.

2. The fault analysis method of the power distribution equipment according to claim 1, characterized in that, The obtaining of the historical alarm data and geographical environment data of the power distribution equipment, and determining the environmental interference change trend based on the historical alarm data and the geographical environment data includes: Obtain the historical alarm data of the power distribution equipment and the geographical environment data of the environment area where the power distribution equipment is located collected by each sensor; wherein, the geographical environment data includes but is not limited to environmental temperature data, environmental humidity data, wind speed data, and vegetation coverage rate; According to the geographical environment data, use the random forest algorithm to calculate the equipment exposure distribution of the power distribution equipment in different seasons; Based on the equipment exposure distribution and the historical alarm data, calculate the interference intensity of the equipment exposure of the power distribution equipment on the alarm frequency; Use a clustering algorithm to divide the environment area according to geographical coordinates and the interference intensity into a high-impact area, a medium-impact area, and a low-impact area; Calculate the environmental interference change trend within a preset time period for each impact area; wherein, the environmental interference change trend is used to represent the dynamic influence trend of geographical environment factors on the occurrence of alarms of the power distribution equipment.

3. The fault analysis method of the power distribution equipment according to claim 1, characterized in that, The according to the environmental interference change trend, using a convolutional neural network to extract features from the historical alarm data to obtain a comprehensive feature representation includes: Use a convolutional neural network to extract features from the historical alarm data to obtain a feature map; Based on the historical alarm data and the feature map, use an attention mechanism to calculate the influence weights of different geographical environment factors on alarms; According to the influence weights, perform weighted fusion on the feature map to obtain a weighted fusion comprehensive feature representation.

4. The fault analysis method of the power distribution equipment according to claim 3, characterized in that, The using a convolutional neural network to extract features from the historical alarm data to obtain a feature map includes: Use a convolutional neural network to organize the historical alarm data into a feature expression matrix according to the time dimension and the space dimension; wherein, the matrix element value of the feature expression matrix is the number of alarms at a specific geographical location within a specific time period; According to the feature expression matrix and the geographical environment data, analyze and obtain the correlation between the alarm frequency and the geographical environment; According to the correlation, adjust the convolution kernel size of the convolutional neural network, and use the adjusted convolutional neural network to process the feature expression matrix to obtain an adjusted feature map.

5. The fault analysis method of the power distribution equipment according to claim 1, characterized in that Training the convolutional neural network with the comprehensive feature representation to obtain a trained fault prediction model includes: Inputting the feature representation into the convolutional neural network for training to obtain a first probability distribution of historical alarms; wherein, the first probability distribution is the fault probability of each fault type in the historical alarms. When the fault probability of a certain fault type is greater than a preset fault probability threshold, updating the model parameters of the convolutional neural network based on the change trend of the environmental interference to obtain a fault prediction model.

6. The fault analysis method of the power distribution equipment according to claim 1, characterized in that Obtaining real-time alarm data and inputting it into the fault prediction model to output a preliminary fault cause, including: Obtaining real-time alarm data and geographical environment data of the power distribution equipment and inputting them into the fault prediction model to output a second probability distribution corresponding to the alarm; wherein, the second probability distribution is the fault probability of each fault type in the real-time alarm predicted by the fault prediction model. Determining the maintenance difficulty index and fault driving factors corresponding to each fault type in the second probability distribution; wherein, the fault driving factors are the main driving factors causing the alarm of the power distribution equipment. Calculating the contribution degree of the fault driving factors to the fault type and adjusting the second probability distribution according to the contribution degree. Obtaining a preliminary fault cause according to the adjusted second probability distribution.

7. The fault analysis method of the power distribution equipment according to claim 1, characterized in that Performing a time series analysis on the preliminary fault cause using a long short-term memory network to obtain a fault diagnosis result, including: Obtaining geographical environment data within a preset time period before the occurrence of the real-time alarm, analyzing the geographical environment data using a long short-term memory network, and outputting environmental time series features. Obtaining a fault diagnosis result according to the environmental time series features and the preliminary fault cause.

8. A fault analysis device for a power distribution device, characterized in that, Including: A data acquisition module for obtaining historical alarm data and geographical environment data of the power distribution equipment and determining the change trend of environmental interference based on the historical alarm data and the geographical environment data; wherein, the change trend of environmental interference is used to represent the dynamic influence trend of each geographical environment factor on the occurrence of the power distribution equipment alarm. A feature extraction module for extracting features from the historical alarm data using a convolutional neural network according to the change trend of environmental interference to obtain a comprehensive feature representation. A model training module for training the convolutional neural network with the comprehensive feature representation to obtain a trained fault prediction model. A fault prediction module for obtaining real-time alarm data and inputting it into the fault prediction model to output a preliminary fault cause. A fault diagnosis module for performing a time series analysis on the preliminary fault cause using a long short-term memory network to obtain a fault diagnosis result.

9. An electronic device, characterized in that, Including: A memory for storing a computer program. A processor for executing the computer program. Wherein, when the processor executes the computer program, it implements the fault analysis method of the power distribution equipment according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed, it implements the fault analysis method of the power distribution equipment according to any one of claims 1 to 7.

Images