Visual warning analysis method, device, equipment and medium for power equipment damage

Abstract

The present application relates to the field of information technology, and provides a method, device, equipment and medium for visual warning analysis of power equipment damage, including: obtaining power equipment repair order data, and obtaining keywords and semantic information based on the repair order data; classifying the keywords and semantic information to obtain structured repair reason data and generating corresponding tags; integrating the structured repair reason data by time and location according to the regional management division of the power system to obtain a repair data set in the time-space dimension; obtaining meteorological data corresponding to the time period of the time-space dimension repair data set, and processing the meteorological data to obtain meteorological feature data; obtaining the output result of the damage prediction model, judging the probability of equipment failure in each area and performing visual display, and generating a warning information. The present application can realize comprehensive visual analysis and intelligent warning of repair data, and reduce the probability of secondary equipment failure.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and particularly to a method, device, equipment and medium for visual warning analysis of power equipment damage. Background Art

[0002] In modern industrial and infrastructure management, the stable operation of power equipment is crucial for ensuring the continuity of production and life. However, due to the lack of accurate data analysis means and technical support, many power enterprises face the problem of difficulty in detecting potential equipment failures and damages in advance, resulting in sudden equipment shutdowns and production interruptions. Traditional maintenance methods mostly rely on regular inspections and experience judgments, which are not only inefficient but also difficult to accurately predict when equipment failures will occur. In addition, equipment damage is sometimes related to specific time periods or weather conditions, and existing management systems may not fully consider the impact of these external factors, making it difficult to comprehensively evaluate the risk status of equipment. Summary of the Invention

[0003] The present invention provides a method for visual warning analysis of power equipment damage, mainly including:

[0004] Obtain power equipment repair order data, process the text content in the repair order data to obtain keywords and semantic information;

[0005] Construct a classification system for repair reasons, and classify the keywords and semantic information based on the classification system for repair reasons to obtain structured repair reason data and generate corresponding labels;

[0006] According to the regional management division of the power system, integrate the structured repair reason data by time and location to obtain a repair data set in the time-space dimension;

[0007] Obtain meteorological data corresponding to the time period of the repair data set in the time-space dimension, and perform standardization processing on the meteorological data to obtain meteorological feature data;

[0008] Associate the repair data set in the time-space dimension with the meteorological feature data to construct a feature vector;

[0009] Adopt a machine learning algorithm to establish a damage prediction model based on the constructed feature vector;

[0010] Predict the failure probability based on the damage prediction model and output the prediction result;

[0011] Obtain the output result of the damage prediction model, judge the probability of equipment failure in each region and perform visual display;

[0012] Determine the warning levels of each area according to a preset failure probability threshold, and generate warning information.

[0013] The technical solution provided by the embodiment of the present invention may include the following beneficial effects:

[0014] The present invention discloses a method for predicting power equipment maintenance damage based on work order data and meteorological data. Aiming at the problem that the lack of unified and structured data analysis and early warning after power equipment maintenance leads to an increased probability of secondary equipment failures, the present invention combines the text content in the work order and meteorological feature data to construct a damage prediction model, realizing the spatio-temporal dimension analysis and intelligent early warning of power equipment maintenance data. By constructing a classification system for maintenance reasons, the unstructured work order data is transformed into structured maintenance reason data, and meteorological feature data is associated to construct feature vectors, and a prediction model is established using machine learning algorithms. Through the visual display of the prediction results and the generation of warning information, the comprehensive visual analysis and intelligent early warning of maintenance data are realized, thereby reducing the probability of secondary equipment failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flowchart of a method for visual warning analysis of power equipment damage according to the present invention.

[0016] Figure 2 It is a schematic structural diagram of a device for visual warning analysis of power equipment damage according to the present invention.

[0017] Figure 3 It is a schematic structural diagram of a computer device according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To further understand the content of the present invention, the present invention will be described in detail in combination with the drawings and embodiments. The following further describes the present application in detail with reference to the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. In addition, it should be noted that only the parts related to the invention are shown in the drawings for the convenience of description.

[0019] Such as Figure 1 , a method for visual warning analysis of power equipment damage in this embodiment may specifically include:

[0020] Step S101, obtain power equipment work order data, process the text content in the work order data, and obtain keywords and semantic information.

[0021] Firstly, the data of power equipment maintenance orders are collected and preprocessed, and the maintenance order text is segmented and POS tagged using natural language processing technology. For example, the jieba segmentation tool is used to segment the text, and NLTK and other tools are used for POS tagging. Then, the TF-IDF algorithm is used to extract keywords from the preprocessed maintenance order text. By calculating the TF-IDF value of each word, the top 20 words with high TF-IDF values ​​are selected as keywords. These keywords are of great significance to maintenance work. Then, the Word2Vec word embedding model is used to perform semantic analysis on the maintenance order text. The word vector model is constructed by training the maintenance order corpus, and the cosine similarity between words is calculated to mine the implicit information and potential associations in the maintenance work. For example, the similarity between "fault" and "circuit breaker" reaches 85, indicating that they have a strong association in maintenance work. According to the extracted keywords and semantic information, the K-means clustering algorithm is used to classify and cluster the maintenance order data. The number of clusters is set to 5 and the number of iterations is set to 100. Different types of maintenance problems and common failure modes, such as "cable failure" and "transformer failure", are identified to provide a basis for formulating targeted maintenance strategies. Combined with the operating status and historical maintenance records of power equipment, the Apriori association rule mining algorithm is used to set the minimum support to 0.5 and the minimum confidence to 0.8 to find the association pattern between equipment failure and maintenance work. For example, the confidence of the association rule between "transformer overheating" and "insulation aging" reaches 0.9, predicting the potential failure risk of the equipment. Based on the results of keyword and semantic analysis, the Neo4j graph database is used to construct a knowledge graph for power equipment maintenance. Keywords are used as nodes and semantic relationships are used as edges to form a structured maintenance knowledge base, providing intelligent fault diagnosis and maintenance plan recommendations for maintenance personnel. Finally, the results of keyword extraction and semantic analysis are applied to the optimization and management of maintenance work. Through in-depth mining and analysis of maintenance order data, the maintenance process is continuously optimized to improve maintenance efficiency and equipment reliability. For example, preventive maintenance strategies are formulated for equipment with high frequency failures to reduce unplanned equipment downtime and improve the comprehensive utilization rate of equipment.

[0022] Step S102: construct a maintenance reason classification system, and classify keywords and semantic information based on the maintenance reason classification system to obtain structured maintenance reason data and generate corresponding tags.

[0023] Collect historical operation status data related to different device names in the power system, and form an initial data set after integrating device model and voltage level data information. Extract fault location information for each piece of data in the initial data set, and obtain the fault level corresponding to the fault cause information according to the pre-established correspondence table between fault levels and fault causes. Obtain the operation status data of the device name corresponding to the fault phenomenon and the device model information from the patrol inspection record database. When there are multiple data records for this device name, the device models are also information-integrated to obtain a specific data object with device model labels and operation status information. Traverse the specific data object, perform frequency analysis according to the fault level in each object and combine with the service life, and define the fault level with the highest frequency and exceeding the preset frequency threshold as the target fault. Use the random forest algorithm to construct a fault cause analysis model, collect data related to the target fault in the historical database, including the fault cause information, service life and voltage level of the target fault, and input all the collected target fault data into the fault cause analysis model to obtain the output result after model training. Construct a classifier based on the support vector machine. The training input variables of the classifier come from the result after model training. When there are at least two different operation status data outputs for this classifier, judge the distribution of the target fault data corresponding to each operation status data by constructing a clustering algorithm, and determine the one with a large distribution density as the association table between the operation status and the fault cause. Obtain the voltage level data of a specific device name corresponding to a specific time point and the historical patrol inspection records of this specific device, analyze the data and the records, and determine the predicted fault cause of the device according to the association table between the operation status and the fault cause, and generate and provide treatment measures related to the predicted fault cause.

[0024] Exemplarily, historical operation status data of various device names in the power system are collected. For example, 1000 transformer data, including statuses such as "overload", "normal", "slightly overheated", etc., and 800 circuit breaker data, including statuses such as "normal opening position", "abnormal closing position", etc. At the same time, device models are integrated, such as transformer model "S-11-1000 / 10" and voltage level "10kV" and other information to form an initial dataset. For each piece of data in the dataset, fault location information is extracted, such as "low oil level in the transformer body", "fault in the circuit breaker opening coil", etc., and combined with the fault level and fault cause correspondence table. For example, "low oil level" corresponds to the cause "poor sealing" which belongs to the "severe" fault level. Device operation status and model corresponding to the fault phenomenon are extracted from the inspection record database. For example, when the fault phenomenon is "low transformer oil level", 10 relevant records are extracted, including the device model, and the status information may be "normal", "overloaded operation", "high oil temperature". For multiple records of the same device name, model information is integrated, such as "transformer S-11-1000 / 10, overload, high oil temperature" as a specific data object. Suppose after analysis, it is determined that the "severe" fault level has the highest occurrence frequency and exceeds the threshold of 60%, and among the data with a service life exceeding 5 years, the proportion of severe faults reaches 75%. Therefore, the "severe" fault level is defined as the target fault. A random forest model is constructed, and 500 pieces of data related to the "severe" fault level in the historical dataset are collected, including causes such as "poor sealing", "insulation aging", etc., and information such as service life "6 years", voltage level "35kV", etc., and input into the model for training. For example, the number of decision trees in the random forest model is set to 100, and the maximum depth is set to 10. After obtaining the model output, a classifier based on the support vector machine is further constructed. The training results of the model, such as the predicted fault causes, are used as the input of the support vector machine model. For example, "poor sealing" corresponds to "operation status 1", "insulation aging" corresponds to "operation status 2", and training is carried out. The classifier training uses the radial basis kernel function, the penalty parameter is set to 1, and the kernel function parameter is set to 1. When there are at least two outputs of this classifier, a judgment based on the K-means clustering algorithm is constructed. For example, the number of clusters is set to 3, and the operation status data and the corresponding target fault data are input into the clustering model. The clustering result may cluster the "overload" status data together, and it is found that the proportion of data corresponding to the fault cause "poor sealing" is 90%, exceeding the preset threshold of 80%. According to the clustering result, the association table between the operation status and the fault cause is finally determined. For example, overload corresponds to poor sealing, and abnormal heating corresponds to insulation aging. For a specific device such as "transformer S-11-1000 / 10", the voltage level "10kV" at the current time point and the historical inspection records are obtained. For example, the record shows that its status was "overload" during the last inspection. Through the association table, its predicted fault cause is "poor sealing", and then a treatment measure is generated: "It is recommended to check and replace the seals".Obtain the text description data of the maintenance work order, perform preprocessing such as word segmentation and stop word removal on the text to obtain a keyword list. Adopt word embedding models such as Word2Vec to convert the keywords into semantic vectors and obtain the semantic representations of the keywords. Calculate the similarity matrix between the keywords based on the semantic vectors of the keywords. Based on the similarity matrix, use clustering algorithms such as K-Means to cluster the keywords to obtain several maintenance reason categories. For each maintenance reason category, count the word frequencies of the keywords within the category and select the keyword with the highest word frequency as the label of this category. Map the maintenance reason categories and the corresponding labels to obtain structured maintenance reason data. For a new maintenance work order, extract its keywords, and judge its belonging category according to the similarity between the keywords and the existing maintenance reason categories, and assign the corresponding label. First, the jieba library can be used to perform word segmentation on the text description data of the maintenance work order, and the stop word list can be used to remove meaningless words, and finally a keyword list is obtained. Then, the Word2Vec model is used to train the keywords, set the dimension of the word vector to 200, the window size to 5, and the number of iterations to 10, and finally the semantic vector corresponding to each keyword is obtained. Next, calculate the similarity matrix between the keywords based on cosine similarity, and use the K-Means clustering algorithm to cluster the keywords into 10 maintenance reason categories. During the clustering process, continuously optimize the cluster centers until the clustering results converge. For each maintenance reason category, count the word frequencies of the keywords within the category and select the top 3 keywords with the highest word frequencies as the labels of this category to form the mapping relationship between the maintenance reason categories and the labels. Finally, for a new maintenance work order, extract its keywords, calculate the cosine similarity between the keywords and the center points of the existing maintenance reason categories, select the category with the highest similarity as the maintenance reason of this work order, and assign the corresponding label to achieve the automatic classification of the maintenance reason. Through this method, the unstructured text of the maintenance work order can be effectively converted into structured maintenance reason data to provide support for subsequent data analysis and decision-making.

[0025] In step S103, according to the regional management division of the power system, integrate the structured maintenance reason data by time and location to obtain a maintenance data set in the spatio-temporal dimension.

[0026] Obtain the regional management division information of the power system, and determine the geographical scope and division granularity of data integration. For the structured maintenance reason data, extract its time attribute and location attribute to form a time dimension and a location dimension. According to the time dimension and the location dimension, group and aggregate the maintenance reason data to obtain a subset of maintenance data indexed by time and location. Adopt data fusion technology to merge the subsets of maintenance data at different times and locations to generate a maintenance data set in the spatio-temporal dimension covering the entire power system.

[0027] Furthermore, the generated maintenance dataset in the spatio-temporal dimension can be subjected to data cleaning and preprocessing to eliminate abnormal data and complete missing values, ensuring data quality. Through data visualization techniques, the maintenance dataset in the spatio-temporal dimension is presented in multiple dimensions to mine the spatio-temporal features and patterns contained in the data. The maintenance dataset in the spatio-temporal dimension is stored in a database, and indexes for time and location are established to facilitate subsequent data retrieval and analysis.

[0028] Exemplarily, first, obtain the regional management division information of the power system from the power company, determine that the geographical scope of data integration is the power system of a certain province, and the division granularity is at the prefecture-level city. For the structured maintenance reason data, use regular expressions to extract the time attribute and location attribute therein to form a time dimension (accurate to days) and a location dimension (accurate to prefecture-level cities). According to the time dimension and location dimension, use the MapReduce algorithm to group and aggregate the maintenance reason data to obtain a subset of maintenance data indexed by time and location. Adopt data fusion technology and use the DataFrame API of Apache Spark to merge subsets of maintenance data at different times and locations to generate a maintenance dataset in the spatio-temporal dimension covering the entire power system. Perform data cleaning and preprocessing on the generated maintenance dataset in the spatio-temporal dimension, use data cleaning functions in the Pandas library of Python to eliminate abnormal data, such as removing data with incorrect time formats and removing data with missing location attributes, and use the KNN algorithm to complete missing values to ensure data quality. Through data visualization techniques, use Tableau to present the maintenance dataset in the spatio-temporal dimension in multiple dimensions, such as showing the change in the number of maintenance times in each prefecture-level city according to the time trend and showing the maintenance heat map of each prefecture-level city according to the geographical location, to mine the spatio-temporal features and patterns contained in the data. Finally, store the maintenance dataset in the spatio-temporal dimension in a PostgreSQL database, and use a B+ tree index to establish indexes for time and location to facilitate subsequent data retrieval and analysis. Through the above steps, the scattered maintenance reason data can be integrated into a unified maintenance dataset in the spatio-temporal dimension, laying a data foundation for subsequent applications such as maintenance reason analysis and fault prediction.

[0029] Step S104: Obtain meteorological data corresponding to the time period of the maintenance dataset in the spatio-temporal dimension, and perform standardization processing on the meteorological data to obtain meteorological feature data.

[0030] According to the start and end times of the maintenance dataset in the space-time dimension, retrieve the meteorological data records for the same time period, and judge the integrity of the meteorological data records. If the meteorological data is missing by more than the preset threshold of 10% within the time period, delete the maintenance data at the corresponding time. Obtain the original meteorological data, including multi-dimensional data such as weather type, wind force level, precipitation, temperature value, humidity value, air pressure value, and visibility. Standardize the original meteorological data to obtain the normalized meteorological data. According to the meteorological data recorded every hour, extract the maximum temperature and minimum temperature in the meteorological data to obtain the daily temperature difference data. Establish a decision tree model through the daily temperature difference data, weather type, precipitation, and visibility data to obtain the time distribution characteristics of the maintenance data corresponding to each region, that is, meteorological characteristic data; at this time, according to the time distribution characteristics of the maintenance data corresponding to each region obtained, use the wind force level and humidity value, and adopt the random forest algorithm to establish a prediction model, and train a model with a prediction accuracy higher than 85%. After obtaining the trained model, input the wind force level and humidity value in the real-time meteorological data to determine the possible maintenance data in each region in the next three days at the corresponding time. For the possible maintenance data in each region in the next three days obtained, make task assignments according to the pre-arranged personnel. Judge that if the number of repairs exceeds 50 units, add task leaders.

[0031] Exemplarily, assume that there are 1000 maintenance data sets in a certain area from January 1, 2023 to December 31, 2023, and each data set contains the start and end times of maintenance. For example, from 9:00 to 12:00 on January 5, 2023. Using this start and end time information, retrieve the corresponding meteorological records from the meteorological database. Taking each hour as a minimum time unit, for example, the maintenance starts at 9:00 and ends at 12:00, which lasts for 3 hours in total. If there is one meteorological record per hour within a certain time period, then there will be 3 records in 3 hours. If the total time has N time units, and there are n records in total, if n is less than 9N, it means there are missing records. If the number of meteorological data records in a certain time period is less than 90% of the expected number of records, for example, a 4-hour time period should have 4 meteorological records but actually only has 3, then the corresponding maintenance data for that time period will be deleted from the data set. Assume that 900 maintenance data sets remain after screening, then the corresponding number of meteorological data records will be obtained. Then, for the obtained meteorological data, such as the weather type is classified as 1 - sunny, 2 - cloudy, 3 - overcast, 4 - rainy; the wind force level is quantified as an integer from 1 to 10; the precipitation is converted to millimeters; the temperature range is set from -20 to 40 degrees Celsius; the humidity range is from 0% to 100%; the air pressure range is from 800 to 1100 hPa; the visibility range is from 0 to 30 kilometers. Use the linear normalization formula. For example, for the temperature value, assume the original data is 30 degrees Celsius, and the normalized value is (30 - (-20)) / (40 - (-20)) = 0.83. Similarly, the normalized values of the other data can be obtained. Then extract the daily maximum and minimum temperature values by date. For example, the maximum temperature on a certain day is 25 degrees Celsius and the minimum temperature is 15 degrees Celsius, so the daily temperature difference is 10 degrees Celsius. Based on this daily temperature difference data, combined with the weather type, precipitation, and visibility data, establish a decision tree model. For example, if the temperature difference is greater than 10 degrees Celsius, the weather type is sunny, the precipitation is less than 1 millimeter, and the visibility is higher than 20 kilometers, then mark the time distribution of the maintenance data for that day as 0. Another example is that if the temperature difference is less than 5 degrees Celsius, the weather type is rainy, the precipitation is greater than 10 millimeters, and the visibility is lower than 5 kilometers, then mark the time distribution of the maintenance data for that day as 1. A classification model based on the Gini index calculation nodes can be trained with the 900 processed data sets to calculate the characteristics of the maintenance time distribution in the area.

[0032] Evaluate with another 100 sets of test data. If the eigenvalue of the regional maintenance time distribution obtained from the decision tree model in Region A is 0, then continue to input the wind level and humidity values of these 800 sets of data into a random forest model with 100 trees. After internal voting in the random forest algorithm model, finally output the predicted number of repairs in this region as 30 units. For the trained model, if the prediction accuracy obtained after testing with 100 sets of data is higher than 85%, then standardize the wind and humidity values obtained in the next period. Input the real-time wind level and humidity values into the model in the same way to determine the possible number of repairs in each region in the next three days. Suppose it is predicted that there will be 60 repairs in a certain region in the next three days, which exceeds the preset threshold of 50, then automatically adjust the electronic workflow system, add a maintenance task leader for this region, and intelligently allocate the tasks according to the current load of the personnel.

[0033] Step S105: Associate the spatio-temporal dimension maintenance data set with the meteorological feature data to construct a feature vector.

[0034] According to the failure time and failure location, obtain the meteorological data information at the failure time to get the original meteorological data set. According to the original meteorological data set, calculate the magnitude of each data value and compare it with a preset first threshold. If it is greater than the first threshold, extract the data to obtain the filtered meteorological data set. Based on the three data dimensions of meteorological temperature, meteorological humidity, and meteorological wind force in the filtered meteorological data set, establish a three-dimensional rectangular coordinate system and obtain the three-dimensional vector values composed of the three data to get the data vector set. According to the data vector set, calculate the Euclidean distance between each vector and the vectors. If the distance is greater than the second threshold, then this vector is a noise point and delete this vector to obtain the denoised data vector set.

[0035] Furthermore, according to the equipment type, query the maintenance record database corresponding to each equipment. If the equipment has no maintenance record, mark it and obtain the list of unmaintained equipment, otherwise proceed to the next step. According to the denoised data vector set and the maintenance records, calculate the time difference between the occurrence time of each maintenance record and the acquisition time of all meteorological data. If the absolute value of the time difference is less than 30 minutes, associate the data with this maintenance record and use it as a positive sample to obtain a maintenance data set containing positive samples. According to the maintenance data set containing positive samples, extract the attributes of the faulty equipment and associate the list of unmaintained equipment. Extract the equipment in the list with the same attributes as the maintained equipment and construct a negative sample set. Input the positive samples and negative samples into the Isolation Forest algorithm and the Local Outlier Factor algorithm to train the model. When the error is less than the preset value, obtain the classification model.

[0036] Exemplarily, assume that a certain device fails from 9:00 to 12:00 on January 5, 2023. Using the geographical location information (such as longitude and latitude or municipal division) within this time period, retrieve the meteorological data records for the same time period from the meteorological database. For example, if the device is located in Haidian District, Beijing, obtain meteorological data such as weather type, wind force level, precipitation, etc. in this area during the above time period to form an original meteorological data set. Set the temperature threshold to -20 to 40 degrees Celsius, the humidity threshold to 0% to 100%, and the air pressure threshold to 800 to 1100 hPa. Mark and exclude data points that exceed these thresholds to ensure that the meteorological data used is within a reasonable range. For example, if the temperature record at a certain moment is 45 degrees Celsius, then this data point will be excluded and not used for subsequent analysis. Next, construct a three-dimensional rectangular coordinate system with temperature, humidity, and wind force in the filtered meteorological data as the three axes. For example, if the temperature at a certain moment is 25 degrees Celsius, the humidity is 60%, and the wind force is level 3, then the meteorological data at this moment can be represented as a three-dimensional vector (25, 60, 3). In this way, all the filtered meteorological data is converted into three-dimensional vectors to form a data vector set. Set a Euclidean distance threshold, for example, 10 units. For each pair of three-dimensional vectors, calculate the Euclidean distance between them. If the distance between two vectors exceeds 10 units, then one of the vectors is considered an outlier (noise point) and is deleted from the data set. After this step, finally obtain a denoised feature vector set, ensuring the quality and reliability of the data. Finally, fuse the denoised feature vectors with other attributes (such as device type, fault location, fault cause, etc.) in the spatio-temporal dimension maintenance data set to form a complete feature vector. For example, a certain fault event occurs on a specific type of transformer, the fault location is the high-voltage side, and the fault cause is overheating. Combine these attributes with the denoised meteorological feature vectors to form a comprehensive feature vector, such as (transformer, high-voltage side, overheating, 25, 60, 3), where the first three elements represent the device type, fault location, and fault cause, and the last three elements represent the meteorological feature data, thus providing high-quality input data for subsequent machine learning models.

[0037] Next, query the device maintenance record database according to the device type. For example, if the device type is "Device A", 1000 maintenance records of this type of device are queried. At the same time, 200 devices of the same type have no maintenance records, then these 200 devices are marked and recorded in the unmaintained device list. Then calculate the difference between the occurrence time of the maintenance record and the acquisition time of the meteorological data. For example, the occurrence time of a certain maintenance record is 15:45 on November 11, 2023, and the acquisition time of the meteorological data is 15:30 on November 11, 2023. The absolute value of the time difference between the two is 15 minutes, which is less than 30 minutes. At this time, the meteorological data such as temperature 25 degrees Celsius, humidity 60%, and wind force level 3 is associated with this maintenance record and marked as a positive sample. Traverse all 1000 maintenance records and 92 meteorological data, associate all maintenance records that meet the time difference requirements with the meteorological data, and mark them as positive samples. Suppose there are 150 that meet the conditions, and a maintenance data set containing 150 positive samples is obtained. Further, extract the attributes of the faulty devices in these 150 maintenance records, such as manufacturer, model, etc., and match these attributes with the unmaintained device list. From the 200 devices in the unmaintained device list, extract the devices that have the same attributes as these 150 faulty devices. For example, a total of 50 devices have the same manufacturer and model attributes as the faulty devices, then these 50 devices are added to the negative sample set. Finally, use the 150 positive samples and 50 negative samples as training data, input them into the Isolation Forest algorithm and the Local Outlier Factor algorithm for training. Set the number of trees in the Isolation Forest algorithm to 100, set the number of neighbors in the Local Outlier Factor algorithm to 20, continuously adjust the parameters and train until the model error is less than 0.5 to obtain the final classification model.

[0038] Step S106, adopt a machine learning algorithm to establish a damage prediction model based on the constructed feature vectors.

[0039] Obtain the historical operation data and fault records of the devices in each region, preprocess the historical operation data and fault records of the devices in each region, and extract the features related to device faults. According to the extracted features, use the random forest algorithm to establish a device damage prediction model, and through training, enable the model to predict the probability of a device failing in a future period of time based on the device operation data. Divide each region into several sub-regions, and perform fault prediction on the devices in each sub-region respectively to obtain the probability distribution of device failures in each sub-region in a future period of time. Analyze the fault probability distributions of each sub-region, identify the high-fault regions, and determine the regions that need to be monitored and maintained with emphasis. According to the importance of the device and the impact of the fault on the business, issue a warning for devices whose fault probability exceeds the preset probability threshold, and generate a device maintenance plan. During the actual operation of the device, continuously collect the device operation data, regularly update the fault prediction model, and dynamically adjust the device maintenance plan. According to the fault prediction results and the actual fault situation, continuously optimize the feature selection and algorithm parameters of the damage prediction model to improve the accuracy of fault prediction.

[0040] Exemplarily, assume that a power company owns grid devices covering multiple regions, including substations, transmission lines, etc. First, obtain the historical operation data and fault records of the devices in each region from the database. These data include information such as device type, fault location, fault cause, repair time, etc. Next, preprocess these data, such as removing duplicate data, filling in missing values, and extracting features related to device faults, such as device service life, historical fault frequency, etc. Then, combine the above-extracted features with the previously constructed feature vectors to form a complete input data set. Use the random forest algorithm to train a device fault prediction model, which can predict the probability of a device failing in a future period of time (such as one week or one month) based on the input feature vectors. To improve the prediction accuracy, the entire region can be divided into several sub-regions, for example, divided by city or county, and perform fault prediction on the devices in each sub-region respectively to obtain the probability distribution of device failures in each sub-region in a future period of time. By analyzing the fault probability distributions of each sub-region, high-fault regions can be identified. For example, if the fault probability of a certain section in a certain city is significantly higher than other regions, it is determined as the region that needs to be monitored and maintained with emphasis. For those devices whose fault probability exceeds the preset threshold (such as 5%), the system will automatically generate a warning message and put forward corresponding maintenance suggestions, such as arranging an emergency inspection or preventive maintenance. In addition, the system will continuously collect the device operation data and regularly update the fault prediction model to adapt to the changes in the device status. Finally, continuously optimize the feature selection and algorithm parameters of the model according to the actual fault situation to improve the accuracy and reliability of fault prediction. For example, if it is found that certain specific types of devices are more likely to fail in high-temperature weather, the weight of temperature-related features can be increased in the model to further improve the prediction effect.

[0041] Step S107, perform failure probability prediction based on the damage prediction model and output the prediction result.

[0042] Perform failure probability prediction based on the damage prediction model and output the prediction result. Specifically, obtain the latest operation data of the equipment in each area from the real-time monitoring system or the historical database, including the current meteorological conditions (such as temperature, humidity, wind force level) and the operation parameters of the equipment (such as load conditions, operation time, etc.). Then, combine the newly obtained data with the previously constructed feature vectors to form a new input feature vector. Use the trained damage prediction model to perform failure probability prediction on the newly input feature vector. The model will calculate and output the probability values of the equipment in each area failing within a certain period in the future, and store these prediction results in a structured format. Finally, parse the output result data to extract the failure probability values of the equipment in each area for subsequent analysis and decision-making. For example, the probability of a transformer in a certain substation failing within the next week is 8%, while the probability of another circuit breaker in the same substation failing is 2%. These prediction results not only help the management understand the equipment health status in the future period, but also provide a scientific basis for formulating a reasonable maintenance plan.

[0043] Suppose a power company has established the damage prediction model in the foregoing steps. Now it is necessary to use this model to predict the failure risks of equipment in each region within the next week. First, obtain the latest operation data of equipment in each region from the real-time monitoring system, such as meteorological conditions like current temperature, humidity, wind force level, as well as operation parameters like the load condition and operation time of the equipment. Next, combine the newly obtained data with the previously constructed feature vectors to form new input feature vectors. For example, for a transformer in a substation, its feature vectors may include equipment type (transformer), failure location (high voltage side), failure cause (overheating), and current temperature (25 degrees Celsius), humidity (60%), wind force (level 3), etc. Then, input these feature vectors into the trained damage prediction model to perform failure probability prediction. The model will calculate and output the probability values of equipment failures in each region within the next week according to the input feature vectors. For example, the probability of a transformer in a substation failing within the next week is 8%, while the probability of another circuit breaker in the same substation failing is 2%. These prediction results will be stored in a database in a structured table form for subsequent query and analysis. Finally, parse the output result data, extract the failure probability values of equipment in each region, generate a report listing the failure probabilities of each piece of equipment in each region, and mark high-risk regions according to a preset threshold (such as 5%). By continuously updating the equipment operation data and retraining the model regularly, the accuracy and reliability of the prediction can be continuously improved to ensure the safe and stable operation of the power system.

[0044] Step S108, obtain the output result of the damage prediction model, judge the probability of equipment failure in each region, and perform visual display.

[0045] Obtain the output result data generated by the damage prediction model, which contains the failure probability information of equipment in each region. Parse the output result data to extract the failure probability values of equipment in each region. According to the preset failure probability threshold, judge the possibility of equipment failure in each region, and mark the regions with failure probabilities higher than the threshold as high-risk regions. Use data visualization technology to map the failure probability values of equipment in each region into different colors or patterns to generate an intuitive visualization chart. In the visualization chart, identify the high-risk regions with different colors or patterns to facilitate maintenance personnel to quickly locate the problem areas. Overlay the visualization chart with the regional map so that maintenance personnel can intuitively understand the relationship between equipment failure probability and geographical location. According to the visualization results, determine the high-risk regions that need to be focused on and repaired, reasonably allocate maintenance resources, and improve the maintenance efficiency and accuracy.

[0046] Exemplarily, the output result data generated by the damage prediction model contains the failure probability information of the devices in each region. The data is parsed through the pandas library of Python to extract the failure probability values of the devices in each region. According to the preset failure probability threshold 7, the possibility of device failure in each region is judged, and the regions with a probability higher than the threshold are marked as high-risk regions. Using data visualization libraries such as Matplotlib and Seaborn, the failure probability values of the devices in each region are mapped to different colors or patterns. For example, the regions with a failure probability above 7 are represented in red, the regions between 5 and 7 are represented in orange, the regions between 3 and 5 are represented in yellow, and the regions below 3 are represented in green, generating an intuitive visualization chart. In the visualization chart, the high-risk regions are identified with different colors or patterns, facilitating the maintenance personnel to quickly locate the problem areas. The visualization chart is overlaid and displayed with the regional map. The Basemap library of Python is used to combine the failure probability data with the geographical location information to generate a regional device failure probability distribution map, enabling the maintenance personnel to intuitively understand the relationship between the device failure probability and the geographical location. According to the visualization results, the high-risk regions that need to be focused on and maintained are determined. By analyzing the characteristics of the device types, usage time, environmental factors, etc. in the high-risk regions, machine learning algorithms such as decision trees and random forests are used to establish a maintenance priority model to rank the devices in the high-risk regions, rationally allocate maintenance resources, and improve the maintenance efficiency and accuracy. By real-time monitoring the device operation data, the failure probability prediction results are dynamically updated to realize the continuous evaluation of the device health status and risk warning, providing data support for preventive maintenance.

[0047] Step S109, according to the preset failure probability threshold, determine the warning levels of each region and generate warning information.

[0048] The warning information includes warning content, relevant repair methods, and precautions. Specifically, first, according to the probability values of equipment failures in each area, and referring to the preset warning level classification criteria (such as low risk, medium risk, high risk), the warning level of each area is determined. For example, three warning levels are set: low risk (failure probability below 3%), medium risk (failure probability between 3% - 5%), and high risk (failure probability above 5%). Then, for different levels of warnings, detailed warning information is automatically generated, including specific warning content (such as which type of equipment may fail in the future), recommended relevant repair methods (such as emergency inspections, preventive maintenance, etc.), and precautions (such as operation safety tips, environmental condition requirements, etc.). These warning information can be conveyed to relevant departments and personnel through various channels to ensure that necessary measures are taken in a timely manner to prevent failures or mitigate their impacts. For example, assume that the probability of a transformer in a substation failing within the next week is 8%, so this substation is marked as a high-risk area. The system will automatically generate a detailed warning message as follows: The warning content is that the probability of the transformer in Substation A failing within the next week is 8%, belonging to a high-risk area; the relevant repair methods include immediately arranging technicians to conduct an emergency inspection of the transformer, focusing on checking whether there is overheating on the high-voltage side. If any abnormalities are found, preventive maintenance measures should be taken immediately, such as replacing aging components or upgrading the cooling system, and arranging regular monitoring to record the temperature change every 4 hours; the precautions remind technicians to wear protective equipment during the inspection to ensure personal safety, pay attention to weather conditions, and if there is a high-temperature warning in the next few days, this problem should be given priority. During the repair period, maintain close communication with the dispatching center to ensure that the normal operation of the power grid is not affected. The warning information will be sent to relevant responsible persons and technical teams through the internal management system, and at the same time, a paper document will be generated for archival and reference. For other areas with medium and low risks, the system will also generate corresponding warning information, but the content and urgency are different. For example, the medium-risk area may recommend increasing the inspection frequency, while the low-risk area only needs to maintain regular maintenance. In addition, all warning information will be updated regularly to reflect the latest equipment status and meteorological condition changes, ensuring the real-time and effectiveness of the warning system.

[0049] Referring to Figure 2 , this application also proposes a visual warning analysis device for power equipment damage, and the device includes:

[0050] Data acquisition and processing module 100: used to obtain power equipment repair order data, process the text content in the repair order data, and obtain keywords and semantic information;

[0051] Classification system construction module 200: used to construct a classification system for maintenance reasons, classify keywords and semantic information based on the classification system for maintenance reasons, obtain structured maintenance reason data, and generate corresponding labels;

[0052] Spatio-temporal data integration module 300: used to integrate the structured maintenance reason data by time and location according to the regional management division of the power system, and obtain a spatio-temporal dimensional maintenance data set;

[0053] Meteorological data processing module 400: used to obtain meteorological data corresponding to the time period of the spatio-temporal dimensional maintenance data set, and perform standardization processing on the meteorological data to obtain meteorological feature data;

[0054] Feature vector construction module 500: used to associate the spatio-temporal dimensional maintenance data set with the meteorological feature data to construct a feature vector;

[0055] Prediction model establishment module 600: used to establish a damage prediction model based on the constructed feature vector by using a machine learning algorithm;

[0056] Fault probability prediction module 700: used to predict the fault probability based on the damage prediction model and output the prediction result;

[0057] Visualization display module 800: used to obtain the output result of the damage prediction model, judge the probability of equipment failure in each area and perform visualization display;

[0058] Early warning information generation module 900: used to determine the early warning level of the area according to a preset fault probability threshold and generate early warning information.

[0059] Refer to Figure 3 , in the embodiment of the present application, a computer device is further provided. The computer device may be a server, and its internal structure may be as Figure 3As shown in the figure. The computer device includes a processor, an internal memory, a storage medium (non-volatile storage medium), and a network interface connected through a system bus. Among them, the processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes the above storage medium (non-volatile storage medium) and the internal memory. The storage medium (non-volatile storage medium) stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the storage medium (non-volatile storage medium). The database of the computer device is used to store usage data, etc. The network interface of the computer device is used to communicate with external terminals through a network connection. Further, the above computer device may also be provided with an input device, a display screen, etc. When the above computer program is executed by the processor, it realizes a method for visual warning analysis of power equipment damage, including the following steps: obtaining power equipment repair order data, processing the text content in the repair order data to obtain keywords and semantic information; constructing a repair reason classification system, and classifying the keywords and semantic information based on the repair reason classification system to obtain structured repair reason data and generating corresponding labels; according to the regional management division of the power system, integrating the structured repair reason data by time and location to obtain a repair data set in the time-space dimension; obtaining meteorological data corresponding to the time period of the time-space dimension repair data set, and performing standardization processing on the meteorological data to obtain meteorological feature data; associating the time-space dimension repair data set with the meteorological feature data to construct a feature vector; using a machine learning algorithm to establish a damage prediction model based on the constructed feature vector; predicting the failure probability based on the damage prediction model and outputting the prediction result; obtaining the output result of the damage prediction model, judging the failure probability of equipment in each region and performing visual display; determining the warning level of each region according to a preset failure probability threshold and generating a warning message. Figure 3 The structure shown in the figure is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied.

[0060] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, a visual warning analysis method for power equipment damage is implemented, including the following steps: constructing a classification system for maintenance reasons, classifying keywords and semantic information based on the classification system for maintenance reasons to obtain structured maintenance reason data, and generating corresponding tags; integrating the structured maintenance reason data by time and location according to the regional management division of the power system to obtain a maintenance data set in the time-space dimension; obtaining meteorological data corresponding to the time period of the maintenance data set in the time-space dimension, and performing standardization processing on the meteorological data to obtain meteorological feature data; associating the maintenance data set in the time-space dimension with the meteorological feature data to construct a feature vector; using a machine learning algorithm to establish a damage prediction model based on the constructed feature vector; predicting the failure probability based on the damage prediction model and outputting a prediction result; obtaining the output result of the damage prediction model, judging the failure probability of equipment in each region and performing visual display; determining the warning level of each region according to a preset failure probability threshold and generating a warning message. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0061] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above method embodiments. Among them, any reference to a memory, storage, database, or other medium provided in the present application and used in the embodiments can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or an external cache. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0062] It should be noted that in this document, the term "comprising", "including" or any other variation thereof is intended to cover non-exclusive inclusion, such that a process, apparatus, article or method comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article or method. Without further limitation, an element defined by the phrase "comprising a..." does not exclude the presence of additional identical elements in the process, apparatus, article or method comprising such element.

[0063] The above are only the preferred embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, is equally included in the patent protection scope of the present application.

[0064] The above description is only the preferred embodiment of the present application and the explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the technical solution formed by mutually replacing the above features with the (but not limited to) technical features having similar functions disclosed in the present application.

Claims

1. A visual early warning analysis method for power equipment damage, characterized in that: The method comprises: Acquire power equipment maintenance order data, process text content in the maintenance order data, and obtain keywords and semantic information; Constructing a maintenance reason classification system, and classifying keywords and semantic information based on the maintenance reason classification system to obtain structured maintenance reason data and generate corresponding labels; According to the regional management division of the power system, the structured maintenance reason data is integrated according to time and place to obtain a maintenance data set in time and space dimensions; Acquire meteorological data of a time period corresponding to the spatiotemporal maintenance data set, and perform standardization processing on the meteorological data to obtain meteorological characteristic data; Associating the spatiotemporal maintenance data set with the meteorological characteristic data to construct a characteristic vector; Using machine learning algorithms, a maintenance damage prediction model is established based on the constructed feature vectors; Performing failure probability prediction based on the maintenance damage prediction model and outputting the prediction result; Obtaining the output results of the maintenance damage prediction model, determining the probability of equipment failure in each area and displaying it visually; According to the preset fault probability threshold, determine the warning level of each area and generate warning information; The construction of the maintenance reason classification system includes: Collect historical operating status data related to different equipment names in the power system, and form an initial data set by integrating equipment model and voltage level data information; Extract the fault location information of each data in the initial data set, and combine it with the pre-established fault level and fault cause correspondence table to obtain the fault level corresponding to the fault cause information; Obtain the device name, operating status data and device model information corresponding to the fault phenomenon in the inspection record database. If there are multiple data records for this device name, the device model is also merged to obtain a specific data object; Traversing the specific data objects, performing frequency analysis based on the fault level in each object and combining the service life, determining the fault level with the highest frequency and exceeding a preset frequency threshold as a target fault; Use the random forest algorithm to build a fault cause analysis model, collect relevant data of the target fault in the historical database, input all collected target fault data into the fault cause analysis model, and obtain the output result after training; Construct a classifier based on support vector machine. When the classifier has at least two different operating status data outputs, use clustering algorithm to determine the target fault data distribution corresponding to each operating status data, and determine the association table between operating status and fault cause with the largest density; Analyze the voltage level data and historical inspection records of the specific equipment name corresponding to the specific time point, determine the predicted fault cause of the corresponding equipment according to the association table between the operating status and the fault cause, and generate treatment measures related to the predicted fault cause.

2. The method according to claim 1, characterized in that According to the regional management division of the power system, the structured maintenance reason data is integrated according to time and location to obtain a maintenance data set in the time and space dimensions, including: Obtain regional management division information of the power system and determine the geographical scope and division granularity of data integration; Extract the time attribute and location attribute of the structured maintenance reason data to form the time dimension and location dimension; According to the time dimension and location dimension, the maintenance reason data is grouped and aggregated to obtain a maintenance data subset indexed by time and location; Data fusion technology is used to merge maintenance data subsets at different times and locations to generate a maintenance data set covering the spatiotemporal dimensions of the entire power system.

3. The method according to claim 1, characterized in that The step of acquiring meteorological data of a time period corresponding to the spatiotemporal maintenance data set and performing standardization processing on the meteorological data to obtain meteorological characteristic data includes: According to the start and end time of the maintenance data set in the spatiotemporal dimension, the meteorological data records of the same time period are retrieved. If the missing meteorological data in the time period exceeds a first preset ratio, the maintenance data at the corresponding time is deleted; Acquire meteorological data, perform standardization processing on the meteorological data, and obtain normalized meteorological data; Extract the maximum and minimum temperature values ​​from the normalized meteorological data and calculate the daily temperature difference data; The normalized meteorological data and daily temperature difference data are combined into meteorological characteristic data.

4. The method according to claim 1, characterized in that: Associating the spatiotemporal maintenance data set with the meteorological characteristic data to construct a characteristic vector includes: According to the fault time and fault location, the meteorological data information of the fault time is obtained to obtain the original meteorological data set; According to the original meteorological data set, the value of each data is calculated and compared with the preset threshold range; If it is greater than the threshold range, the data is extracted to obtain a filtered meteorological data set; According to the three data dimensions of meteorological temperature, meteorological humidity and meteorological wind force in the filtered meteorological data set, a spatial rectangular coordinate system is established to obtain the three-dimensional vector values ​​composed of the three data to obtain a data vector set; According to the data vector set, the Euclidean distance between each vector is calculated. If the distance is greater than the threshold, the corresponding vector is a noise point. The vector is deleted to obtain the feature vector of the denoised data.

5. The method according to claim 1, characterized in that The machine learning algorithm is used to establish a maintenance damage prediction model based on the constructed feature vector, including: Obtain and pre-process the historical operation data and fault records of equipment in each area, and extract features related to equipment failures; Using the feature vector as input, a maintenance damage prediction model is established using a random forest algorithm, so that the model can predict the probability of failure within a future period of time based on equipment operation data; Each area is divided into several sub-areas, and fault prediction is performed on the equipment in each sub-area to obtain the probability distribution of failure of the equipment in each sub-area in the future; Analyze the fault probability distribution of each sub-area, identify the areas with high fault incidence, and determine the areas that need key monitoring and maintenance; Issue early warnings for equipment whose failure probability exceeds a preset probability threshold and generate equipment maintenance plans; Continuously collect equipment operation data, regularly update fault prediction models, and dynamically adjust equipment maintenance plans; According to the fault prediction results and actual fault conditions, the feature selection and algorithm parameters of the maintenance damage prediction model are optimized to improve the accuracy of fault prediction.

6. The method according to claim 1, characterized in that The step of obtaining the output result of the maintenance damage prediction model, determining the probability of equipment failure in each area and performing visual display includes: Obtain output result data generated by the maintenance damage prediction model, which includes failure probability information of equipment in each area; Analyze the output result data and extract the failure probability value of the equipment in each area; According to the preset failure probability threshold, the possibility of equipment failure in each area is determined, and areas above the threshold are marked as high-risk areas; Using data visualization technology, the failure probability values ​​of equipment in each area are mapped into different colors or patterns to generate intuitive visualization charts; Among them, in the visual chart, high-risk areas are marked with different colors or patterns; Overlay visualization charts with regional maps to show the relationship between equipment failure probability and geographic location, identify high-risk areas that require special attention and maintenance, and allocate maintenance resources.

7. A visual early warning analysis device for power equipment damage, used to execute the method according to any one of claims 1 to 6, characterized in that: The device comprises: Data acquisition and processing module: used to obtain power equipment maintenance order data, process the text content in the maintenance order data, and obtain keywords and semantic information; Classification system construction module: used to construct a maintenance reason classification system, and classify keywords and semantic information based on the maintenance reason classification system to obtain structured maintenance reason data and generate corresponding labels; Spatiotemporal data integration module: used to integrate the structured maintenance reason data by time and location according to the regional management division of the power system to obtain a spatiotemporal maintenance data set; Meteorological data processing module: used to obtain meteorological data of a time period corresponding to the time-space dimension maintenance data set, perform standardization processing on the meteorological data, and obtain meteorological characteristic data; Feature vector construction module: used to associate the spatiotemporal maintenance data set with the meteorological feature data to construct a feature vector; Prediction model building module: used to establish a maintenance damage prediction model based on the constructed feature vector using a machine learning algorithm; Fault probability prediction module: used to predict the fault probability based on the maintenance damage prediction model and output the prediction result; Visualization display module: used to obtain the output results of the maintenance damage prediction model, determine the probability of equipment failure in each area and perform visual display; Warning information generation module: used to determine the warning level of the area according to a preset fault probability threshold and generate warning information.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images