A charging pile safety detection method based on artificial intelligence

Abstract

The present application provides a charging pile safety detection method based on artificial intelligence, including: according to the results of high-voltage dielectric testing and partial discharge detection, using a support vector machine algorithm to establish a mathematical model between insulation performance and environmental parameters, and predicting the insulation margin of the charging pile under different environmental conditions through the model; using virtual reality technology to visualize the insulation performance test results of the charging pile under different environmental conditions, and marking the insulation status of the charging pile with different colors in a virtual scene according to the level of insulation margin; using a deep learning algorithm to extract and classify features of the partial discharge signal of the charging pile, determine the type and severity of the partial discharge, and optimize the insulation structure design of the charging pile according to the classification results.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a charging pile safety detection method based on artificial intelligence. Background Art

[0002] Charging piles work in outdoor environments for a long time and face the test of various harsh environments such as high and low temperature, humidity, salt spray, and dust. These environmental factors may cause the insulation performance of charging piles to deteriorate, increase the risk of leakage and short circuit, and threaten the safety of use. At present, the test methods for the insulation performance of charging piles are mainly limited to tests under single environmental conditions, such as high temperature, low temperature or humidity test, and cannot fully evaluate the insulation performance of charging piles in actual use environments. In addition, the existing test methods lack consideration of the use scenarios of charging piles, there are differences between the test conditions and the actual conditions, and the test results are difficult to reflect the true insulation level of the charging piles. At the same time, the analysis and visualization methods of insulation performance test data are relatively single, which is not conducive to intuitive and comprehensive evaluation of the environmental adaptability of charging piles. Therefore, it is urgent to establish a set of intelligent charging pile insulation performance environmental adaptability test system that comprehensively considers multiple environmental factors, is close to actual use scenarios, and improves the accuracy, comprehensiveness and interpretability of the test results, and provides a reliable basis for the safe and reliable operation of charging piles. Summary of the invention

[0003] The present invention provides a charging pile safety detection method based on artificial intelligence, which mainly includes:

[0004] According to the usage scenarios of charging piles, machine learning algorithms are used to automatically generate test conditions in various extreme environments, obtain environmental parameters such as high and low temperature, humidity and heat, salt spray, dust, vibration, etc., and use them as control parameters of the composite environmental test chamber;

[0005] The working state of the charging pile under different environmental conditions is simulated by a composite environmental test chamber. During the environmental cycle, the insulation strength of the charging pile is tested by a high-voltage dielectric tester, and the partial discharge signal caused by insulation defects is detected by a partial discharge detector.

[0006] Based on the results of high-voltage dielectric tests and partial discharge detection, a mathematical model between insulation performance and environmental parameters was established using the support vector machine algorithm. The insulation margin of the charging pile under different environmental conditions was predicted through the model.

[0007] Using virtual reality technology, the insulation performance test results of charging piles under different environmental conditions are visualized. The insulation status of the charging piles is marked with different colors in the virtual scene according to the insulation margin.

[0008] Through big data analysis technology, the test data of charging piles under different environmental conditions are mined to obtain the key environmental factors that affect the insulation performance of charging piles, and the control strategy of the composite environmental test chamber is optimized according to the key factors;

[0009] A deep learning algorithm is used to extract and classify the features of the partial discharge signals of the charging pile, determine the type and severity of the partial discharge, and optimize the insulation structure design of the charging pile based on the classification results.

[0010] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0011] The present invention discloses a charging pile safety detection method based on artificial intelligence. The method uses a machine learning algorithm to generate extreme environmental test conditions, and simulates the working state of the charging pile in different environments through a composite environmental test chamber. During the environmental cycle process, high-voltage dielectric testing and partial discharge detection technology are used to evaluate the insulation performance of the charging pile. Based on the test results, a support vector machine algorithm is used to establish a mathematical model of insulation performance and environmental parameters to predict the insulation margin under different environments. The present invention also uses virtual reality technology to visualize the test results, and mines key environmental factors through big data analysis. In addition, a deep learning algorithm is used to extract and classify the features of the partial discharge signal to optimize the insulation structure design of the charging pile. The method realizes the intelligent testing and evaluation of the insulation performance of the charging pile, and improves the reliability and safety of the charging pile. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 The present invention is a flowchart of a charging pile safety detection method based on artificial intelligence.

[0013] Figure 2 It is a schematic diagram of a charging pile safety detection method based on artificial intelligence of the present invention.

[0014] Figure 3 This is another schematic diagram of a charging pile safety detection method based on artificial intelligence according to the present invention. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this specification.

[0016] like Figure 1-3In this embodiment, a charging pile safety detection method based on artificial intelligence may specifically include:

[0017] S101. Based on the usage scenarios of charging piles, a machine learning algorithm is used to automatically generate test conditions under various extreme environments, obtain environmental parameters such as high and low temperature, humidity, salt spray, dust, vibration, etc., and use them as control parameters of the composite environmental test chamber.

[0018] According to the usage scenarios and environmental characteristics of charging piles, machine learning algorithms are used to automatically generate test conditions in various extreme environments. By analyzing historical data and expert knowledge, the range and distribution characteristics of environmental parameters such as high and low temperature, humidity, salt spray, dust, and vibration are obtained, and used as constraints for generating test conditions. Data enhancement techniques, such as adding random noise, rotation, and translation, are used to generate more test condition samples based on the existing environmental parameter data, expand the training data set, and improve the generalization ability of the machine learning model. Through feature selection algorithms such as recursive feature elimination (RFE) and feature importance evaluation, the environmental parameter features that have the greatest impact on the test conditions are selected, the representation of the test conditions is optimized, and redundancy and noise are reduced. Using the decision tree algorithm, the generated test condition samples are mapped to the control parameters of the composite environmental test chamber, including temperature, humidity, salt spray concentration, dust concentration, vibration frequency and amplitude, etc., to achieve automated testing. During the test, the performance data of the charging pile, such as charging efficiency and failure rate, are collected in real time through sensors. Correlation analysis and regression analysis are used to study the relationship between performance data and environmental parameters, and the reliability and adaptability of the charging pile in different environments are evaluated. According to the test results, grid search and cross-validation methods are used to optimize the hyperparameters of the machine learning algorithm and improve the accuracy of test condition generation. At the same time, by increasing the diversity and extreme degree of test conditions, the coverage of the test is improved, providing data support and decision-making basis for the design and improvement of charging piles.

[0019] Specifically, in order to obtain the range and distribution characteristics of environmental parameters such as high and low temperature, humidity, salt fog, dust, vibration, etc., historical data can be statistically analyzed. For example, by analyzing the meteorological data of a certain area in the past 10 years, it is found that the temperature range is -20℃ to 45℃, the humidity range is 30% to 95%, and the average wind speed is 5m / s. These environmental parameters are modeled using probability distribution models such as normal distribution and exponential distribution to generate test condition samples that conform to the actual situation. On this basis, data enhancement techniques can be applied, such as adding random noise from -2℃ to 2℃ to the temperature data and performing random offsets from -5% to 5% in the humidity data to expand the training data set. Through the recursive feature elimination algorithm, the degree of influence of each environmental parameter on the performance of the charging pile is evaluated, and temperature, humidity, and salt fog concentration are selected as key features. Using the decision tree algorithm, the generated test condition samples are mapped to the control parameters of the composite environmental test chamber, such as the temperature range of -25℃ to 50℃, the humidity range of 20% to 98%, and the salt fog concentration of 5% to 15%. During the test, the charging efficiency and failure rate data of the charging pile were collected every 10 minutes through sensors. The relationship between environmental parameters and performance data was analyzed using the Pearson correlation coefficient. It was found that temperature and failure rate showed a significant positive correlation (r=0.85), and humidity and charging efficiency showed a significant negative correlation (r=-0.76). According to the results of the grid search, the maximum depth parameter of the decision tree algorithm was adjusted to 8, and the minimum number of leaf node samples was adjusted to 20, which improved the accuracy of the test condition generation. In order to comprehensively evaluate the reliability of the charging pile, some extreme test conditions can also be generated, such as temperature -30℃ and humidity 95%, temperature 50℃ and salt spray concentration 20%, etc., to further improve the coverage of the test.

[0020] S102. The working state of the charging pile under different environmental conditions is simulated by a composite environmental test chamber. During the environmental cycle, the insulation strength of the charging pile is tested by a high-voltage dielectric tester, and the partial discharge signal caused by insulation defects is detected by a partial discharge detector.

[0021] Obtaining the working status data of the charging pile under various temperature, humidity, and salt spray environmental conditions, wherein the working status data includes the key parameters of the voltage, current, and temperature of the charging pile;

[0022] Obtaining dielectric strength data of the charging pile insulation material under different environmental conditions, wherein the dielectric strength data is a curve showing the change of dielectric strength with environmental conditions;

[0023] When the dielectric strength is lower than a preset threshold, it is determined that the insulating material has a risk of aging or damage;

[0024] Acquire partial discharge signal data of the charging pile during operation, wherein the partial discharge signal data includes amplitude, frequency, and phase characteristic parameters of the discharge signal;

[0025] When the discharge amount exceeds a preset threshold, it is determined that there is an insulation defect, and the type and severity of the insulation defect are obtained;

[0026] Optimizing the insulation structure design parameters of the charging pile according to the dielectric strength data and the partial discharge signal data, wherein the insulation structure design parameters include the type, thickness, and process of the insulation material;

[0027] Using a support vector machine algorithm to comprehensively analyze and extract features from the working state data, the dielectric strength data, and the partial discharge signal data, and to train an insulation failure prediction model;

[0028] When the insulation failure prediction model predicts the risk of insulation failure, it is determined that the charging pile needs to be shut down for maintenance;

[0029] The working status data, the dielectric strength data and the partial discharge signal data are comprehensively analyzed to output a charging pile environmental adaptability evaluation report, wherein the evaluation report includes key performance parameters of the charging pile under various temperature, humidity and salt spray environmental conditions, potential risk points of insulation materials and structures, improved insulation design solutions and performance indicators of insulation failure prediction models.

[0030] Specifically, in order to obtain the working status data of the charging pile under various environmental conditions, 27 groups of test conditions can be generated according to the principle of orthogonal experimental design within the range of temperature -30℃ to 50℃, humidity 30% to 95%, and salt spray concentration 0 to 20%. Each test lasts for 24 hours, and the key parameters such as voltage, current, and temperature of the charging pile are collected every hour. For the dielectric strength test of insulating materials, five commonly used materials such as epoxy resin and silicone rubber can be selected to test their dielectric strength change curves under different temperature, humidity, and salt spray conditions. Each test point is repeated 3 times and the average value is taken. When the dielectric strength is found to be lower than 50kV / mm, it is judged that the insulating material is at risk of aging or damage. During the operation of the charging pile, an ultra-high frequency partial discharge tester is used to collect partial discharge signals in real time at a sampling frequency of 1MHz, and the characteristic parameters such as amplitude, frequency, and phase of the discharge signal are extracted. When the discharge amount exceeds 500pC, it is judged that there is an insulation defect, and the type (such as air gap discharge, surface discharge) and severity of the insulation defect are identified based on the community characteristics and phase distribution of the discharge. Comprehensively analyze the dielectric strength and partial discharge data to optimize the insulation structure design of the charging pile, such as replacing the insulation material with the American UL 94V-0 grade epoxy resin with a higher temperature resistance, increasing the insulation thickness by 20%, and coating the insulation surface with nano-modified silicone rubber to improve the anti-aging performance. Finally, the working status data, dielectric strength data, and partial discharge signal data are used as input features of the support vector machine algorithm to train the insulation failure prediction model. After 5-fold cross-validation, the accuracy of the model reached more than 95%. When the model predicts the risk of insulation failure, it immediately notifies the shutdown for maintenance. The generated charging pile environmental adaptability assessment report shows that in a high temperature, high humidity, and salt spray environment, the maximum charging current of the charging pile needs to be reduced by 20%, and the service life of the insulation material is shortened by 30%. It is necessary to focus on the risk of insulation defects at the connectors and cable joints. It is recommended to conduct preventive insulation testing and maintenance every six months to ensure the safe and reliable operation of the charging pile.

[0031] S103. Based on the results of high-voltage dielectric testing and partial discharge detection, a support vector machine algorithm is used to establish a mathematical model between insulation performance and environmental parameters, and the insulation margin of the charging pile under different environmental conditions is predicted through the model.

[0032] The high-voltage dielectric test and partial discharge detection data of the charging pile under different environmental conditions are obtained as the input of the support vector machine algorithm. The obtained test data is preprocessed to remove outliers and noise, extract key feature parameters, and obtain a cleaned data set. According to the processed test data and environmental parameters, the kernel function type and model hyperparameters of the support vector machine algorithm are optimized by grid search and k-fold cross validation methods. Using the optimized hyperparameter configuration, the support vector machine model is trained based on the cleaned data set to establish a nonlinear mapping relationship between the insulation performance of the charging pile and the environmental parameters. The environmental parameters under different environmental conditions are input into the trained support vector machine model to obtain the corresponding insulation performance prediction values. According to the insulation performance prediction value and the rated insulation level of the charging pile, the insulation margin under the environmental condition is calculated, that is, the difference between the actual insulation performance and the rated insulation level. A prediction report containing different environmental conditions, the corresponding insulation performance prediction values ​​and insulation margin is generated and presented in the form of tables and charts to provide a decision-making basis for the safe operation of the charging pile.

[0033] Specifically, in order to obtain high-voltage dielectric test and partial discharge detection data of charging piles under different environmental conditions, an orthogonal experiment can be designed, considering three factors: temperature (-30℃~50℃), humidity (30%~95%), and salt spray concentration (0~20%). Three levels are selected for each factor, for a total of 27 groups of experiments. The breakdown voltage of the insulation material of the charging pile under different environmental conditions is tested using a high-voltage dielectric tester. Each group of tests is repeated 3 times and the average value is taken. A partial discharge detector is used to collect discharge signals at a sampling frequency of 1MHz to extract characteristic parameters such as discharge amount and frequency. The acquired test data is preprocessed, the 3σ criterion is used to remove outliers, and the wavelet transform is used to reduce noise on the discharge signal. Ten key characteristic parameters such as breakdown voltage, mean discharge amount, and standard deviation of discharge frequency are extracted to obtain a cleaned data set. The kernel function type (RBF kernel, polynomial kernel) and hyperparameters (penalty coefficient C, kernel function parameter γ) of the support vector machine model are optimized by grid search and 5-fold cross validation. The search range is C between 1 and 100 and γ between 0.1 and 1. Using the optimized hyperparameter configuration (C=10, γ=1, RBF kernel), the support vector machine regression model is trained based on the cleaned data set to establish a nonlinear mapping relationship between the insulation performance of the charging pile and the environmental parameters. The temperature, humidity, and salt spray concentration under different environmental conditions are input into the trained support vector machine model to obtain the corresponding insulation performance prediction value, such as the breakdown voltage of 50kV. According to the rated insulation level of the charging pile (such as 60kV), the insulation margin under the environmental conditions is calculated to be 10kV, that is, the actual insulation performance is 10kV higher than the rated level. Generate a prediction report, presenting the predicted insulation performance values ​​and insulation margins for different environmental conditions. For example, when the temperature is 40°C, the humidity is 80%, and the salt spray concentration is 10%, the breakdown voltage is 45kV and the insulation margin is -15kV, indicating that the insulation performance does not meet the requirements in this environment and there are safety hazards. Based on the prediction results, provide a decision-making basis for the safe operation of the charging pile. For example, in a high temperature, high humidity and salt spray environment, it is recommended to reduce the charging current, shorten the maintenance cycle, and focus on weak parts that are prone to insulation failure.

[0034] S104. Use virtual reality technology to visualize the insulation performance test results of the charging pile under different environmental conditions, and mark the insulation status of the charging pile with different colors in the virtual scene according to the insulation margin.

[0035] The insulation performance test data of the charging pile under different environmental conditions are obtained, and the insulation margin value of the charging pile under each test environment is calculated based on the test data. A virtual reality scene model is pre-built, the charging pile model is integrated into the virtual reality scene, and the insulation margin value is associated with the charging pile model. The insulation margin value is compared with the preset threshold. If the insulation margin value is greater than the threshold, the charging pile model under the test environment is rendered green, indicating that the insulation state is good; if the insulation margin value is less than the threshold, the charging pile model is rendered red, indicating that the insulation state is poor. The user's interactive operation is obtained through the virtual reality device, and the user's area is determined according to the virtual scene position selected by the user, using methods such as ray detection, and the charging pile model at the corresponding position is presented, and the insulation margin value and insulation state of the charging pile are displayed. The K-means clustering algorithm is used to group the insulation margin values ​​under different environments to obtain multiple insulation margin intervals, and the charging pile models in different intervals are rendered in different colors to achieve a hierarchical display of the insulation state. Based on the Apriori association rule mining algorithm, the correlation between different environmental factors and insulation margin values ​​is analyzed, the key environmental factors affecting insulation performance are determined, and the analysis results are presented in the form of text or charts in a virtual scene.

[0036] Specifically, in order to obtain the insulation performance data of the charging pile under different environmental conditions, an orthogonal test was designed, considering three factors: temperature (-20℃~40℃), humidity (40%~90%), and dust concentration (0~10g / m³). Three levels were selected for each factor, with a total of 27 groups of tests. The breakdown voltage of the charging pile insulation material under different environmental conditions was tested using a high-voltage dielectric tester. Each group of tests was repeated 3 times and the average value was taken. According to the rated insulation level of 60kV of the charging pile, the insulation margin value under each test environment was calculated. Unity3D was used to build a virtual reality scene, the 3D model of the charging pile was imported into the scene, and the insulation margin value was associated with the model. The insulation margin threshold was set to 10kV, and the charging pile model with an insulation margin greater than 10kV was rendered green and that with an insulation margin less than 10kV was rendered red through script programming. The HTC Vive headset and handle were used to realize the user's interactive operation in the virtual scene. The user's area was determined by ray detection, and the charging pile model at the corresponding position was dynamically loaded, and the insulation margin value and status were displayed. The insulation margin values ​​obtained from 27 groups of tests were clustered using K-means, and the number of clusters k was set to 3. The insulation margin intervals were [0,5], [5,15], and [15,30]. The charging pile models in different intervals were rendered in red, yellow, and green, respectively, to achieve hierarchical display. The Apriori algorithm was used to set the minimum support to 20% and the minimum confidence to 60% to mine the association rules between environmental factors and insulation margin. It was found that when "temperature ≥ 30°C and humidity ≥ 70%", the insulation margin value was low, and the analysis results were presented in text form in the virtual scene.

[0037] S105. Through big data analysis technology, the test data of charging piles under different environmental conditions are mined to obtain the key environmental factors affecting the insulation performance of charging piles, and the control strategy of the composite environmental test chamber is optimized based on the key factors.

[0038] The test data of charging piles under various environmental conditions are obtained, and the data are cleaned, integrated, transformed and other preprocessing operations are performed to obtain a high-quality data set. The Apriori algorithm is used to mine the association rules of the data set, analyze the correlation between the insulation performance of the charging pile and the environmental conditions, and find out the key environmental factors that have a greater impact on the insulation performance, such as temperature, humidity, and salt spray concentration. According to the key environmental factors mined, the support vector machine algorithm is selected to establish a prediction model for the insulation performance of the charging pile. The data set is divided into a training set and a test set, and the model hyperparameters are optimized using grid search and cross-validation methods. The model performance is evaluated on the test set to obtain the prediction model with the best performance. The key environmental factors are used as control variables of the composite environmental test chamber. By setting different combinations of environmental parameters, the actual working environment of the charging pile is simulated, and accelerated life tests and reliability tests are carried out. During the test, sensors are used to collect environmental parameters and charging pile performance data in real time, and the data is transmitted to the industrial computer through a data acquisition card, and then uploaded to the Hadoop cluster of the big data analysis platform through Ethernet for online analysis. When the support vector machine prediction model determines that the insulation performance is decreasing significantly, the PID control parameters of the test chamber are dynamically adjusted through the industrial computer. For example, when the insulation performance decreases faster, the temperature and humidity setting values ​​are automatically lowered or the salt spray cycle is shortened to delay the aging process of the insulation material. By continuously optimizing the control strategy of the composite environmental test chamber, it can simulate the actual use conditions of the charging pile and comprehensively evaluate the environmental adaptability of the product. The environmental parameters and performance data collected during the test are stored in the Hadoop distributed file system, and MapReduce is used for data processing and analysis. The Spark MLlib machine learning library is used to train and optimize the prediction model to provide data support for improving the reliability of the charging pile.

[0039] Specifically, in order to obtain the test data of charging piles under various environmental conditions, an orthogonal test was designed, considering three factors: temperature (-40℃~60℃), humidity (30%~95%), and salt spray concentration (0~10mg / m³). Five levels were selected for each factor, with a total of 125 groups of tests. The breakdown voltage of the charging pile insulation material under different environmental conditions was tested using a high-voltage dielectric tester. Each group of tests was repeated 3 times and the average value was taken. The obtained data was cleaned, missing values ​​and outliers were removed, and data integration was performed to integrate data from different sources. Then the data was normalized and mapped to the [0,1] interval to eliminate the dimension effect. The Apriori algorithm was used to mine association rules for the data set, setting the minimum support to 10% and the minimum confidence to 80%. The results showed that temperature and humidity were the key environmental factors affecting insulation performance. The support vector machine algorithm was selected to establish a prediction model, and the data set was divided into a training set and a test set in a ratio of 8:2. The model hyperparameters were optimized by grid search and 5-fold cross validation methods, and the optimal parameter combination was obtained: the kernel function was RBF, the penalty coefficient C=10, and the kernel function parameter γ=1. The model performance was evaluated on the test set, and the root mean square error RMSE was 2 and the determination coefficient R² was 95, indicating that the model has good prediction performance. Temperature and humidity were used as control variables of the composite environmental test chamber, and the temperature range was set to -20℃~50℃ and the humidity range was set to 40%~90%. The actual working environment of the charging pile was simulated and a 6-month accelerated life test was carried out. During the test, the environmental parameters were collected in real time using temperature and humidity sensors, and the data was transmitted to the industrial computer through the data acquisition card, and then uploaded to the Hadoop cluster of the big data analysis platform through Ethernet. When the support vector machine prediction model judged that the insulation performance degradation rate exceeded 5kV / day, the temperature setting value was reduced by 5℃ and the humidity setting value was reduced by 10% through the industrial computer to delay the aging process of the insulation material. After the test, the collected data was analyzed, and the mean insulation resistance was calculated to be 50MΩ and the standard deviation was 5MΩ using MapReduce. The SparkMLlib machine learning library was used to optimize the prediction model. By increasing the number of training samples and feature engineering, the root mean square error (RMSE) was reduced to 8 and the coefficient of determination (R²) was increased to 98, further improving the prediction performance of the model and providing data support for the reliability design of charging piles.

[0040] S106. Use a deep learning algorithm to extract and classify features of the local discharge signal of the charging pile, determine the type and severity of the local discharge, and optimize the insulation structure design of the charging pile based on the classification results.

[0041] The partial discharge signal data during the operation of the charging pile is obtained. The data should include partial discharge signals under different charging power, charging time, ambient temperature and humidity, etc. The collected partial discharge signal data is preprocessed, and the median filter algorithm is used to remove random noise, and the wavelet threshold denoising algorithm is used to remove background noise to improve the signal-to-noise ratio. The preprocessed partial discharge signal is analyzed in the time and frequency domains by wavelet transform to extract the time domain characteristic parameters of the signal, including the number of discharges, discharge amplitude, pulse interval, etc., and the frequency domain characteristic parameters of the signal are extracted, including spectrum peak, spectrum center of gravity, spectrum peak ratio, etc. A one-dimensional convolutional neural network model is constructed, and the input layer is the extracted partial discharge signal feature vector. The convolution layer uses 3 convolution kernels with sizes of 3, 4, and 5, respectively. The pooling layer uses maximum pooling, and the fully connected layer uses 2 hidden layers with 128 and 64 neurons, respectively. The number of neurons in the output layer is the number of partial discharge types. The data set is divided into training set, validation set and test set with a ratio of 6:2:2. The Adam optimizer is used for training for 100 epochs with a batch size of 32. The early stopping method is used to prevent overfitting. The generalization ability of the model is judged according to the classification accuracy of the convolutional neural network model on the test set. If the classification accuracy reaches more than 95%, it is considered that the model can effectively identify the type and severity of partial discharge. Partial discharge is divided into three types: pinpoint discharge, floating potential discharge and surface discharge. According to the repetition rate and amplitude of the discharge, the severity of partial discharge is divided into mild, moderate and severe. According to the different types and severity of partial discharge, the corresponding insulation fault warning level is formulated. When severe surface discharge is detected, the charging pile should be stopped immediately, the insulation material and structure should be checked, and the aging or damaged insulation parts should be replaced in time; when moderate pinpoint discharge or floating potential discharge is detected, the insulation status of the charging pile should be monitored more closely and the maintenance cycle should be shortened; when mild partial discharge is detected, it can continue to operate, but insulation resistance test and infrared thermal imaging detection should be carried out regularly. For weak spots where partial discharge is prone to occur, such as cable terminals and connectors, highly flame-retardant, high-strength, and aging-resistant epoxy resin insulation materials are used to increase insulation thickness and shielding layers, optimize insulation structure design, and improve insulation performance. Through multi-physics field simulation analysis, the insulation performance of the optimized charging pile insulation structure under different temperature, humidity, vibration, and voltage is evaluated, and partial discharge tests are conducted to test the partial discharge resistance of insulation materials and structures under high voltage, high frequency, and long-term working conditions, verifying the effectiveness of the optimized design.

[0042] Specifically, in order to obtain the partial discharge signal data during the operation of the charging pile, a partial discharge online monitoring system was designed. The system uses high-frequency current sensors and ultra-high frequency antenna arrays with a frequency range of 3MHz~300MHz and a sensitivity better than 2pC. Sensors are arranged at key parts of the charging pile, such as high-voltage conductors, insulators, and cable terminals, to collect partial discharge signals in real time. Considering that the charging power of the charging pile is 7kW~120kW, the charging time is 5h~8h, the ambient temperature is -30℃~55℃, and the relative humidity is 30%~95%, the layout position and number of sensors need to meet the detection requirements under different working conditions. The collected partial discharge signal is first preprocessed, and the median filter algorithm is used to remove random noise, and the filter window size is 5. Then the wavelet threshold denoising algorithm is used to remove the background noise, and the db4 wavelet basis function is selected. The number of decomposition layers is 5, and the threshold is determined by the 3σ criterion. After preprocessing, the signal-to-noise ratio of the signal is improved by 20dB. Wavelet transform was used to analyze the preprocessed signal in the time and frequency domains, and the time domain characteristic parameters, including the number of discharges, average discharge amplitude, pulse interval, etc., were extracted. The frequency domain characteristic parameters, including spectrum peak, spectrum center of gravity, spectrum peak ratio, etc., were extracted. A one-dimensional convolutional neural network model was constructed. The input layer was a feature vector composed of 12 extracted characteristic parameters. The convolution layer used 3 convolution kernels with sizes of 3, 4, and 5, respectively. The activation function was ReLU. The pooling layer used maximum pooling with a pooling size of 2. The fully connected layer used 2 hidden layers with 128 and 64 neurons, respectively. The activation function was ReLU. The output layer used the Softmax activation function with 3 neurons, representing 3 types of partial discharge. The 1000 sets of data collected were divided into training set, validation set, and test set in a ratio of 6:2:2. The Adam optimizer was used with a learning rate of 0.01, 100 epochs of training, and a batch size of 32. The early stopping method monitored the validation set loss function. The training was stopped when the loss function did not decrease for 5 consecutive epochs. The model performance was evaluated on the test set, and the classification accuracy reached 98%, indicating that the model can effectively identify the type of partial discharge. According to the discharge repetition rate and amplitude, the severity of partial discharge is divided into three types: mild (repetition rate <10 times / s, amplitude <100pC), moderate (repetition rate 10~100 times / s, amplitude 100~1000pC) and severe (repetition rate >100 times / s, amplitude >1000pC). When severe surface discharge is detected, an alarm signal and shutdown command are immediately sent to the charging pile controller. When moderate needle tip discharge or suspended potential discharge is detected, a warning signal is sent to the controller to adjust the charging parameters and reduce the charging power and voltage. When mild partial discharge is detected, a prompt signal is sent to the controller to continue running, but the next maintenance cycle is shortened. For weak parts where partial discharge is prone to occur, epoxy resin casting insulation is used to increase the insulation thickness to 5mm, and a 2mm thick semi-conductive shielding layer is coated on the surface to optimize the insulation structure design.Through multi-physics field simulation analysis, the maximum electric field strength of the insulating material is 5kV / mm at -40℃~60℃, 30%~95%RH, 0~5g vibration, and 90kV voltage, which is less than its breakdown field strength of 8kV / mm, meeting the safety margin requirements. The optimized insulation structure was subjected to a partial discharge test. Under the conditions of 90kV, 50Hz, and 1000h, the partial discharge amount was always less than 10pC, indicating that the optimized design can effectively suppress partial discharge.

[0043] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A charging pile safety detection method based on artificial intelligence, characterized in that: The method comprises: According to the usage scenarios of charging piles, machine learning algorithms are used to automatically generate test conditions under various extreme environments, obtain environmental parameters including at least one of high and low temperature, humidity, salt spray, dust, and vibration, and use them as control parameters of the composite environmental test chamber; The working state of the charging pile under different environmental conditions is simulated by a composite environmental test chamber. During the environmental cycle, the insulation strength of the charging pile is tested by a high-voltage dielectric tester, and the partial discharge signal caused by insulation defects is detected by a partial discharge detector. Based on the results of high-voltage dielectric testing and partial discharge detection, a mathematical model between insulation performance and environmental parameters is established using a support vector machine algorithm, and the insulation margin of the charging pile under different environmental conditions is predicted through the model; wherein, the environmental parameters under different environmental conditions are input into the trained support vector machine model to obtain the corresponding insulation performance prediction value; based on the insulation performance prediction value and the rated insulation level of the charging pile, the insulation margin under the environmental condition is calculated, that is, the difference between the insulation performance prediction value and the rated insulation level; Using virtual reality technology, the insulation status of the charging pile is marked with different colors in the virtual scene according to the insulation margin. Through big data analysis technology, the test data of the charging pile under different environmental conditions is mined to obtain the key environmental factors that affect the insulation performance of the charging pile, and the control strategy of the composite environmental test chamber is optimized based on the key factors. A deep learning algorithm is used to extract and classify the features of the partial discharge signals of the charging pile, determine the type and severity of the partial discharge, and optimize the insulation structure design of the charging pile based on the classification results.

2. The method according to claim 1, characterized in that According to the usage scenarios of the charging piles, a machine learning algorithm is used to automatically generate test conditions under various extreme environments, obtain environmental parameters including at least one of high and low temperature, humidity, salt spray, dust, and vibration, and use them as control parameters of the composite environmental test chamber, including: According to the usage scenarios and environmental characteristics of the charging piles, machine learning algorithms are used to automatically generate test conditions in a variety of extreme environments; By analyzing historical data, the range and distribution characteristics of high and low temperature, humidity, heat, salt fog, dust, and vibration environment parameters are obtained and used as constraints for generating test conditions; Using data enhancement technology, including adding random noise, rotation and translation, to generate more test condition samples based on the existing environmental parameter data and expand the training data set; By using a feature selection algorithm, including one of recursive feature elimination and feature importance evaluation, the environmental parameter features that have the greatest impact on the test conditions are selected to optimize the representation of the test conditions; Using the decision tree algorithm, the generated test condition samples are mapped to the control parameters of the composite environmental test chamber, including temperature, humidity, salt spray concentration, dust concentration, vibration frequency and amplitude, to achieve automated testing; During the test, sensors are used to collect real-time performance data of the charging pile, including charging efficiency and failure rate; Use correlation analysis and regression analysis to study the relationship between performance data and environmental parameters, and evaluate the reliability and adaptability of charging piles in different environments; Based on the test results, use either grid search or cross validation to optimize the hyperparameters of the machine learning algorithm.

3. The method according to claim 1, characterized in that The method uses a composite environmental test chamber to simulate the working state of the charging pile under different environmental conditions, uses a high-voltage dielectric tester to test the insulation strength of the charging pile during the environmental cycle, and uses a partial discharge detector to detect partial discharge signals caused by insulation defects, including: Obtaining the working status data of the charging pile under various temperature, humidity and salt spray environmental conditions, wherein the working status data includes the key parameters of the voltage, current and temperature of the charging pile; Obtaining dielectric strength data of the charging pile insulation material under different environmental conditions, wherein the dielectric strength data is a curve showing the change of dielectric strength with environmental conditions; When the dielectric strength is lower than a preset threshold, it is determined that the insulating material has a risk of aging or damage; Acquire partial discharge signal data of the charging pile during operation, wherein the partial discharge signal data includes amplitude, frequency and phase characteristic parameters of the discharge signal; When the discharge amount exceeds a preset threshold, it is determined that there is an insulation defect, and the type and severity of the insulation defect are obtained; Optimizing the insulation structure design parameters of the charging pile according to the dielectric strength data and the partial discharge signal data, wherein the insulation structure design parameters include the type, thickness and process of the insulation material; Using a support vector machine algorithm to comprehensively analyze and extract features from the working state data, the dielectric strength data, and the partial discharge signal data, and to train an insulation failure prediction model; When the insulation failure prediction model predicts the risk of insulation failure, it is determined that the charging pile needs to be shut down for maintenance; The working status data, the dielectric strength data and the partial discharge signal data are comprehensively analyzed to output a charging pile environmental adaptability evaluation report, wherein the evaluation report includes key performance parameters of the charging pile under various temperature, humidity and salt spray environmental conditions, potential risk points of insulation materials and structures, improved insulation design solutions and performance indicators of insulation failure prediction models.

4. The method according to claim 1, characterized in that: According to the results of high-voltage dielectric testing and partial discharge detection, a mathematical model between insulation performance and environmental parameters is established using a support vector machine algorithm, and the insulation margin of the charging pile under different environmental conditions is predicted by the model, including: Obtain high-voltage dielectric test and partial discharge detection data of charging piles under different environmental conditions as input to the support vector machine algorithm; Preprocess the acquired test data, remove outliers and noise, extract key feature parameters, and obtain a cleaned data set; According to the processed test data and environmental parameters, the kernel function type and model hyperparameters of the support vector machine algorithm are optimized using grid search and k-fold cross validation methods; Using the optimized hyperparameter configuration, the support vector machine model is trained based on the cleaned data set to establish a nonlinear mapping relationship between the insulation performance of the charging pile and the environmental parameters; Generate a prediction report containing different environmental conditions, corresponding insulation performance prediction values ​​and insulation margins, and present it in the form of tables and charts to provide a decision-making basis for the safe operation of charging piles.

5. The method according to claim 1, characterized in that The virtual reality technology is used to visualize the insulation performance test results of the charging pile under different environmental conditions. According to the insulation margin, the insulation status of the charging pile is marked with different colors in the virtual scene, including: Pre-build a virtual reality scene model, integrate the charging pile (3D) model into the virtual reality scene, and associate the insulation margin value with the charging pile model; The insulation margin value is compared with the preset threshold value. If the insulation margin value is greater than the threshold value, the charging pile model under the test environment is rendered green, indicating that the insulation state is good; If the insulation margin value is less than the threshold, the charging pile model is rendered in red, indicating that the insulation status is poor; The user's interactive operation is obtained through the virtual reality device, and the area where the user is located is determined by the ray detection method according to the virtual scene position selected by the user, and the charging pile model at the corresponding position is presented, and the insulation margin value and insulation status of the charging pile are displayed; The K-means clustering algorithm is used to group the insulation margin values ​​under different environments to obtain multiple insulation margin intervals. The charging pile models in different intervals are rendered in different colors to achieve hierarchical display of insulation status. Based on the Apriori association rule mining algorithm, the correlation between different environmental factors and insulation margin values ​​is analyzed, the key environmental factors affecting insulation performance are determined, and the analysis results are presented in the form of text or charts in a virtual scene.

6. The method according to claim 1, characterized in that The test data of charging piles under different environmental conditions are mined through big data analysis technology to obtain key environmental factors affecting the insulation performance of charging piles, and the control strategy of the composite environmental test chamber is optimized according to the key factors, including: Obtain test data of charging piles under various environmental conditions and perform preprocessing operations on the data, including cleaning, integration, and transformation operations, to obtain high-quality data sets; The Apriori algorithm is used to mine association rules on the data set, analyze the correlation between the insulation performance of the charging pile and the environmental conditions, and find out the key environmental factors that have a greater impact on the insulation performance; According to the key environmental factors discovered, the support vector machine algorithm is used to establish a prediction model for the insulation performance of charging piles. The data set is divided into a training set and a test set, and the model hyperparameters are optimized using grid search and cross-validation methods. The model performance is evaluated on the test set to obtain the prediction model with the best performance. Use key environmental factors as control variables of the composite environmental test chamber, set different environmental parameter combinations, simulate the actual working environment of the charging pile, and carry out accelerated life test and reliability test; During the test, sensors were used to collect environmental parameters and charging pile performance data in real time, and the data was transmitted to the industrial computer through a data acquisition card. The data was then uploaded to the Hadoop cluster of the big data analysis platform via Ethernet for online analysis. When the support vector machine prediction model determines that the insulation performance is decreasing significantly, the PID control parameters of the test chamber are dynamically adjusted through the industrial computer. When the insulation performance decreases faster, the temperature and humidity setting values ​​are automatically lowered or the salt spray cycle is shortened to delay the aging process of the insulation material. By continuously optimizing the control strategy of the composite environmental test chamber, it can simulate the actual use conditions of the charging pile and comprehensively evaluate the environmental adaptability of the product; The environmental parameters and performance data collected during the experiment are stored in the Hadoop distributed file system, MapReduce is used for data processing and analysis, and the SparkMLlib machine learning library is used to train and optimize the prediction model.

7. The method according to claim 1, characterized in that The method uses a deep learning algorithm to extract and classify the local discharge signal of the charging pile, judge the type and severity of the local discharge, and optimize the insulation structure design of the charging pile according to the classification result, including: Obtain partial discharge signal data during the operation of the charging pile. The data should include partial discharge signals under different charging power, charging time, and ambient temperature and humidity conditions; Preprocess the collected partial discharge signal data, use the median filter algorithm to remove random noise, and use the wavelet threshold denoising algorithm to remove background noise; The pre-processed partial discharge signal is analyzed in the time and frequency domain by using wavelet transform to extract the time domain characteristic parameters of the signal, including the number of discharges, discharge amplitude and pulse interval, and the frequency domain characteristic parameters of the signal, including the spectrum peak, spectrum center of gravity and spectrum peak ratio; A one-dimensional convolutional neural network model was constructed. The input layer was the extracted PD signal feature vector. The convolution layer used three convolution kernels with sizes of 3, 4, and 5, respectively. The pooling layer used maximum pooling. The fully connected layer used two hidden layers with 128 and 64 neurons, respectively. The number of neurons in the output layer was the number of PD types. The dataset is divided into training set, validation set and test set with a ratio of 6:2:

2. The Adam optimizer is used for training for 100 epochs with a batch size of 32. The early stopping method is used to prevent overfitting. According to the classification accuracy of the convolutional neural network model on the test set, the generalization ability of the model is judged; If the classification accuracy reaches more than 95%, the model is considered to be able to effectively identify the type and severity of partial discharge; The partial discharge is divided into three types: pinpoint discharge, floating potential discharge and surface discharge. According to the repetition rate and amplitude of the discharge, the severity of partial discharge is divided into mild, moderate and severe. According to the different types and severity of partial discharge, formulate corresponding insulation fault warning levels; When severe surface discharge is detected, the charging pile should be stopped immediately, the insulation material and structure should be checked, and aged or damaged insulation parts should be replaced in time; When moderate pinpoint discharge or suspended potential discharge is detected, the insulation status of the charging pile should be monitored more closely to shorten the maintenance cycle; When mild partial discharge is detected, operation can continue, but insulation resistance testing and infrared thermal imaging testing should be performed regularly; Aiming at the weak spots where partial discharge is prone to occur, including cable terminals and connectors, we use highly flame-retardant, high-strength and aging-resistant epoxy resin insulation materials, increase insulation thickness and shielding layer, optimize insulation structure design, and improve insulation performance; Through multi-physics field simulation analysis, the insulation performance of the optimized charging pile insulation structure under different temperatures, humidity, vibrations and voltages is evaluated. Partial discharge tests are also carried out to test the ability of the insulation materials and structures to resist partial discharge under high voltage, high frequency and long-term working conditions, thereby verifying the effectiveness of the optimized design.

Images