Charging pile abnormity identification method based on deep learning

Abstract

The invention discloses a charging pile abnormity identification method based on deep learning. The method comprises the following steps: S1, obtaining operation data of a charging pile for preprocessing; s2, based on the preprocessed data, using a deep learning algorithm to construct a charging pile abnormity identification model; s3, adopting a firefly algorithm to optimize hyper-parameters of the model; s4, adjusting a model structure based on the optimized hyper-parameter; s5, collecting data of the charging pile in real time, inputting the data into the optimized model for abnormal state recognition, and generating abnormal alarm information; s6, according to the abnormal alarm information, automatically triggering an alarm and feeding back the alarm to maintenance personnel for processing; and S7, continuously collecting new charging pile data, and performing incremental learning and updating by using the optimized model. According to the invention, efficient and accurate identification of charging pile anomaly detection is realized, the fault processing speed and the system stability are improved, and the method is widely applied to charging pile management and maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of charging pile identification, and in particular to a method for identifying charging pile anomalies based on deep learning. Background Art

[0002] With the increasing popularity of electric vehicles, charging piles, as key infrastructure for providing charging services, have become an essential component of modern transportation networks. As the number and distribution of charging piles continue to grow, their stable operation is crucial to ensuring the proper use of electric vehicles. However, with this increasing number of charging piles, the incidence of charging pile failures has also increased, posing significant challenges to their maintenance and management. To ensure the proper operation of charging piles, developing effective methods for identifying anomalies and detecting faults has become a crucial task.

[0003] Existing charging pile fault detection methods primarily rely on manual inspections or threshold-based rule-based judgments. While manual inspections can detect certain faults, their reliance on manual operation carries the risk of low inspection efficiency, missed detections, and false detections, and they are unable to monitor the operating status of all charging piles in real time. Threshold-based rule-based judgment methods typically set different thresholds for current, voltage, temperature, and other factors. When the operating parameters of the charging pile exceed the set thresholds, the system will issue an alarm. However, this method has significant limitations: first, the threshold setting relies on experience and cannot adapt to changes in charging piles in different environments and under different usage conditions; second, the threshold method is prone to false positives and missed negatives and cannot accurately identify all types of faults.

[0004] With the rapid development of artificial intelligence (AI), particularly deep learning, fault detection methods based on deep learning have become a research hotspot. Deep learning can effectively identify complex fault patterns by automatically learning features from data, eliminating the manual intervention and reliance on experience required by traditional methods. A deep learning-based charging pile anomaly recognition method, trained on historical charging pile operating data, can accurately identify abnormal charging pile conditions, improving the accuracy and efficiency of fault detection.

[0005] However, existing deep learning-based charging pile anomaly identification methods still have some shortcomings. First, many existing methods rely on large-scale labeled data for model training, but charging pile fault data is often scarce, and the fault types are complex and diverse, resulting in poor representativeness of the training set, which directly affects the generalization ability of the model. Second, most existing methods use traditional deep learning frameworks, which have certain latency issues when processing real-time charging pile data, making it impossible to perform real-time and accurate fault identification. In addition, some existing methods have poor adaptability to abnormal data and cannot effectively respond to the impact of new faults or environmental changes.

[0006] In order to make up for the shortcomings of the existing technology, the present invention proposes a charging pile anomaly identification method based on deep learning. By combining the deep learning model with the optimization algorithm, this method can achieve higher-precision fault detection and abnormal state identification. By introducing the firefly algorithm to optimize the model hyperparameters, the efficiency of hyperparameter selection in the training process can be improved, avoiding falling into the local optimal solution, and further improving the recognition accuracy of the model. At the same time, the convolutional neural network (CNN) is used to extract and classify the charging pile data, which enhances the feature learning ability of the model and enables it to handle complex and diverse charging pile fault types. In addition, through the input of real-time data and the incremental learning mechanism, the present invention can continuously optimize the charging pile anomaly identification model to adapt to new fault modes and environmental changes, thereby improving the adaptability of the system.

[0007] In summary, existing charging pile anomaly identification methods suffer from data scarcity, insufficient model adaptability, and poor real-time performance. This paper, by combining deep learning with optimization algorithms, proposes an efficient and accurate charging pile anomaly identification method. This method improves fault detection accuracy while overcoming the shortcomings of existing technologies, providing strong technical support for intelligent charging pile management and fault warning.

[0008] Therefore, how to provide a charging pile abnormality identification method based on deep learning is a problem that technicians in this field urgently need to solve. Summary of the Invention

[0009] One purpose of the present invention is to propose a method for identifying abnormalities in charging piles based on deep learning. The present invention makes full use of deep learning, optimization algorithms and real-time data acquisition technology, and describes in detail how to accurately identify abnormal states of charging piles through convolutional neural networks (CNNs), and combines the firefly algorithm to optimize model hyperparameters to further improve the accuracy and robustness of the model. Through the real-time collection of charging pile operation data, the model can continue to learn and optimize, thereby adapting to new failure modes and environmental changes. This method has the advantages of high precision, high efficiency and strong adaptability. It can identify charging pile faults in real time and generate alarm information, significantly improving the operational safety and management efficiency of charging piles.

[0010] A method for identifying abnormalities in a charging pile based on deep learning according to an embodiment of the present invention includes the following steps:

[0011] S1. Obtaining operation data of the charging pile and preprocessing the data;

[0012] S2. Based on the pre-processed operating data, a charging pile anomaly recognition model is constructed using a deep learning algorithm. This model is trained on historical data to generate a preliminary charging pile anomaly recognition model.

[0013] S3. Use the firefly algorithm to optimize the hyperparameters of the charging pile anomaly recognition model, optimize the learning rate, number of network layers, and activation function hyperparameters, perform a global search, and evaluate the optimization effect through the fitness function;

[0014] S4. Based on the optimized hyperparameters, adjust the network structure of the charging pile anomaly recognition model, adjust the number of convolutional layers, fully connected layer structure, and activation function selection;

[0015] S5. Collecting the operating data of the charging pile in real time, and inputting the data into the optimized charging pile abnormality recognition model to identify abnormal conditions and generate abnormal alarm information, which includes the fault type, occurrence time and location, etc.;

[0016] S6. Based on the abnormal alarm information, the alarm mechanism is automatically triggered and the fault information is sent to the maintenance personnel for fault location and repair;

[0017] S7. Continue to collect new charging pile data and input it into the optimized model to update the model.

[0018] Optionally, the S1 specifically includes: obtaining the operating data of the charging pile, including current, voltage, temperature, and power parameters, and preprocessing the data; using the sliding average method or wavelet transform to remove noise, and filling missing values ​​through linear interpolation or K-nearest neighbor algorithm; using minimum-maximum normalization to standardize the data to the range of [0,1]; using the standard deviation method to eliminate outliers that deviate from the mean by more than 3 times the standard deviation; using the Pearson correlation coefficient analysis method to select features related to the abnormal state of the charging pile and remove correlation features; dividing the processed data set into a training set and a test set in a ratio of 7:3; generating a charging pile data feature set.

[0019] Optionally, the S2 specifically includes: selecting a convolutional neural network as the basic model, designing the model, setting the number of network layers and the number of neurons in each layer; when using CNN, selecting a combination of several convolutional layers and pooling layers, and extracting spatial feature information in the input data through the convolution kernel; adding a maximum pooling layer after each convolutional layer to perform feature downsampling; selecting the activation function as ReLU; in the output layer of the model, selecting the softmax function for classification, and outputting the probability of whether the charging pile is in an abnormal state; using the cross-entropy loss function to measure the difference between the model output and the true label, adjusting the weights and biases in the network through the back-propagation algorithm, and minimizing the loss function; using the training set for iterative training to generate a preliminary charging pile anomaly recognition model.

[0020] Optionally, the S3 specifically includes:

[0021] S31. Define the hyperparameter optimization space and select the hyperparameters to be optimized, including the learning rate, the number of network layers, the number of neurons in each layer, and the activation function type. Set the learning rate η within a preset range, and determine the range of the number of network layers L and the number of neurons in each layer N. The activation function type σ includes ReLU, sigmoid, or tanh.

[0022] S32. Initialize a number of fireflies, each firefly represents a set of hyperparameter combinations, and evaluate the fitness of each firefly using a fitness function, where the fitness function is the classification accuracy or loss function value of the model on the validation set;

[0023] S33. Calculate the attraction between fireflies. The attraction is proportional to the fitness. The attraction calculation formula is:

[0024]

[0025] Among them, A ij is the attraction of firefly i to firefly j, β is the maximum attraction, γ is the attraction attenuation coefficient, r ij is the distance between fireflies i and j;

[0026] S34. Update the position of the "firefly" according to the attraction. The position update formula is:

[0027] X i =X i +α(X i -X j )+∈;

[0028] Among them, X i and X j is the current coordinate of fireflies i and j, α is the step coefficient, and ∈ is the random perturbation term;

[0029] S35, repeating the position update process until the optimization stop criterion is met, such as reaching the maximum number of iterations or the fitness change is lower than the set threshold;

[0030] S36. Output the optimal hyperparameter combination and apply it to the charging pile anomaly recognition model.

[0031] Optionally, the S4 specifically includes:

[0032] S41. Based on the optimized hyperparameters, determine the number of network layers and the number of neurons in each layer of the convolutional neural network, and select the appropriate activation function type, including ReLU, sigmoid, or tanh;

[0033] S42. Determine the number of convolution layers and the size of the convolution kernel of each layer. The size of the convolution kernel is set to 3×3 or 5×5, and the step size is set to 1 or 2. After each convolution layer, a pooling layer is connected. The maximum pooling method is used for feature downsampling. The pooling window size is 2×2 and the pooling step size is 2.

[0034] S43. Add a Dropout layer after the convolutional layer and pooling layer, set the dropout rate to a value range of 0.2 to 0.5, and enhance the generalization ability of the model by discarding neuronal connections;

[0035] S44. Select the cross entropy loss function for the classification task. The loss function formula is:

[0036]

[0037] Among them, y i is the actual label, is the model prediction output, N is the number of samples;

[0038] S45. Use the Adam optimizer for training, optimize the network weights and biases through the back-propagation algorithm, minimize the loss function, and finally generate an optimized charging pile anomaly recognition model;

[0039] S46. Verify the performance of the adjusted model on the validation set. If the precision and recall rate meet the requirements, determine the final optimized network structure.

[0040] Optionally, the S42 specifically includes: setting the number of convolution layers of the convolutional neural network, and selecting the convolution kernel size and step size for each convolution layer; the size of the convolution kernel is 3×33 or 5×55, and the step size is set to 1 or 2; after each convolution layer, adding a maximum pooling layer, the pooling window size is 2×22, and the pooling step size is 2; the convolution layer and the pooling layer are stacked in sequence, and the output of each convolution layer is downsampled through the pooling layer.

[0041] Optionally, the S5 specifically includes:

[0042] S51. Collect various operating parameter data of the charging pile in real time through sensors, including current, voltage, temperature, and power, to maintain the real-time and accuracy of the data;

[0043] S52, preprocessing the collected real-time data using the same processing method as the training set data, including denoising, filling missing values, normalization, etc., to maintain a format consistent with the training data;

[0044] S53: Input the pre-processed real-time data into the optimized charging pile abnormality recognition model. The model recognizes the abnormal state according to the input features and outputs the prediction result probability value.

[0045] S54, according to the probability value of the prediction result Compare with the set threshold (θ), if It is considered an abnormal state; if It is judged to be in normal state;

[0046] S55. When an abnormal state is identified, an abnormal alarm message is generated, including the abnormality type, occurrence time, and fault location, and sent to maintenance personnel for processing;

[0047] S56. Maintenance personnel locate faults based on alarm information and adjust and optimize the system through feedback mechanisms.

[0048] Optionally, the S6 specifically includes:

[0049] S61. When the charging pile abnormality recognition model identifies an abnormal state, it generates alarm information including the abnormality type, occurrence time, fault location, and abnormality severity level;

[0050] S62. The system determines the alarm priority based on the severity level of the abnormality. If the abnormality severity level is high, the system triggers an emergency alarm and sends an emergency processing notice to the maintenance personnel first; if the abnormality severity level is low, the system sends the processing information according to the low priority level;

[0051] S63. Alarm information is sent to maintenance personnel in real time via SMS, email or APP notification;

[0052] S64. After receiving the alarm information, the maintenance personnel locate the problem according to the fault type and location and perform corresponding maintenance operations;

[0053] S65. After the maintenance personnel complete the maintenance operation, they feed back the processing results to the system, and the system updates the fault status and records it;

[0054] S66. The system updates the equipment status and optimizes the abnormality recognition model based on the maintenance feedback information.

[0055] Optionally, S7 specifically includes: continuously collecting operating data of the charging pile through sensors, regularly inputting new data into the optimized charging pile abnormality recognition model, updating the model using an incremental learning method, adjusting model parameters, and regularly evaluating the performance of the model on new data through a feedback mechanism, and further optimizing the recognition model based on the evaluation results.

[0056] The beneficial effects of the present invention are:

[0057] (1) By adopting a charging pile anomaly recognition model based on deep learning, the present invention can automatically learn the complex patterns in the charging pile operation data and accurately identify abnormal changes in parameters such as current, voltage, and temperature, avoiding the common false alarm and missed alarm problems in traditional rule-based threshold methods, and significantly improving the accuracy of charging pile fault identification.

[0058] (2) The present invention introduces the firefly algorithm to perform global optimization on the hyperparameters of the deep learning model, avoiding the local optimal problem commonly found in traditional methods. It can find the optimal hyperparameter combination in a larger range, thereby effectively improving the performance of the model and enhancing its adaptability to various charging pile failure modes.

[0059] (3) The present invention collects charging pile operation data in real time and continuously inputs it into the optimized abnormality recognition model, and continuously optimizes the model using incremental learning to respond to new faults or environmental changes, ensuring that the charging pile abnormality recognition system can operate stably and efficiently in the long term and maintain a high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0061] Figure 1 This is an overall flow chart of a charging pile anomaly identification method based on deep learning proposed by the present invention;

[0062] Figure 2 This is a flowchart of the method for identifying abnormalities in charging piles based on deep learning proposed by the present invention: optimizing the hyperparameters of the abnormality identification model of charging piles using the firefly algorithm;

[0063] Figure 3 This is a schematic diagram of the charging pile anomaly recognition model training and optimization of the charging pile anomaly recognition method based on deep learning proposed in the present invention. DETAILED DESCRIPTION

[0064] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0065] refer to Figure 1-3 , a charging pile anomaly recognition method based on deep learning, comprising the following steps:

[0066] S1. Obtaining operation data of the charging pile and preprocessing the data;

[0067] S2. Based on the pre-processed operating data, a charging pile anomaly recognition model is constructed using a deep learning algorithm. This model is trained on historical data to generate a preliminary charging pile anomaly recognition model.

[0068] S3. Use the firefly algorithm to optimize the hyperparameters of the charging pile anomaly recognition model, optimize the learning rate, number of network layers, and activation function hyperparameters, perform a global search, and evaluate the optimization effect through the fitness function;

[0069] S4. Based on the optimized hyperparameters, adjust the network structure of the charging pile anomaly recognition model, adjust the number of convolutional layers, fully connected layer structure, and activation function selection;

[0070] S5. Collecting the operating data of the charging pile in real time, and inputting the data into the optimized charging pile abnormality recognition model to identify abnormal conditions and generate abnormal alarm information, which includes the fault type, occurrence time and location, etc.;

[0071] S6. Based on the abnormal alarm information, the alarm mechanism is automatically triggered and the fault information is sent to the maintenance personnel for fault location and repair;

[0072] S7. Continue to collect new charging pile data and input it into the optimized model to update the model.

[0073] The present invention provides a deep learning-based charging pile anomaly identification method that can monitor charging pile operating data in real time and identify anomalies through a deep learning algorithm, overcoming the false positives and missed negatives of traditional threshold methods. Compared with existing technologies, the present invention has higher fault identification accuracy and real-time response capabilities. By combining convolutional neural networks with optimization algorithms, the present invention can accurately identify complex fault patterns, avoiding the limitations of traditional rule-based methods in various charging pile operating conditions, and improving the intelligence and reliability of charging pile fault detection.

[0074] In this embodiment, S1 specifically includes: obtaining the operating data of the charging pile, including current, voltage, temperature, and power parameters, and preprocessing the data; using the sliding average method or wavelet transform to remove noise, and filling missing values ​​through linear interpolation or K-nearest neighbor algorithm; using minimum-maximum normalization to standardize the data to the range of [0,1]; using the standard deviation method to eliminate outliers that deviate from the mean by more than 3 times the standard deviation; using the Pearson correlation coefficient analysis method to select features related to the abnormal state of the charging pile and remove correlation features; dividing the processed data set into a training set and a test set in a ratio of 7:3; generating a charging pile data feature set.

[0075] This invention ensures the quality and consistency of data input into the deep learning model by systematically preprocessing charging pile operating data. Compared with existing technologies, this invention uses denoising, missing value filling, and normalization to enable the model to operate more stably in different environments and conditions. The efficiency of the preprocessing step ensures high-quality data input to the model, greatly improving the accuracy of subsequent anomaly identification and the adaptability of the model, especially when faced with complex and noisy data, effectively improving the overall performance of the system.

[0076] In this embodiment, S2 specifically includes: selecting a convolutional neural network as the basic model, designing the model, setting the number of network layers and the number of neurons in each layer; when using CNN, selecting a combination of several convolutional layers and pooling layers, and extracting spatial feature information in the input data through the convolution kernel; adding a maximum pooling layer after each convolutional layer to perform feature downsampling; selecting the activation function as ReLU; in the output layer of the model, selecting the softmax function for classification, and outputting the probability of whether the charging pile is in an abnormal state; using the cross-entropy loss function to measure the difference between the model output and the true label, adjusting the weights and biases in the network through the back-propagation algorithm, and minimizing the loss function; using the training set for iterative training to generate a preliminary charging pile anomaly recognition model.

[0077] In this embodiment, S3 specifically includes:

[0078] S31. Define the hyperparameter optimization space and select the hyperparameters to be optimized, including the learning rate, the number of network layers, the number of neurons in each layer, and the activation function type. Set the learning rate η within a preset range, and determine the range of the number of network layers L and the number of neurons in each layer N. The activation function type σ includes ReLU, sigmoid, or tanh.

[0079] S32. Initialize a number of fireflies, each firefly represents a set of hyperparameter combinations, and evaluate the fitness of each firefly using a fitness function, where the fitness function is the classification accuracy or loss function value of the model on the validation set;

[0080] S33. Calculate the attraction between fireflies. The attraction is proportional to the fitness. The attraction calculation formula is:

[0081]

[0082] Among them, A ij is the attraction of firefly i to firefly j, β is the maximum attraction, γ is the attraction attenuation coefficient, r ij is the distance between fireflies i and j;

[0083] S34. Update the position of the "firefly" according to the attraction. The position update formula is:

[0084] X i =X i +α(X i -X j )+∈

[0085] Among them, X i and X j is the current coordinate of fireflies i and j, α is the step coefficient, and ∈ is the random perturbation term;

[0086] S35, repeating the position update process until the optimization stop criterion is met, such as reaching the maximum number of iterations or the fitness change is lower than the set threshold;

[0087] S36. Output the optimal hyperparameter combination and apply it to the charging pile anomaly recognition model.

[0088] This paper introduces the Firefly algorithm to optimize the hyperparameters of deep learning models. Compared to traditional hyperparameter adjustment methods, the Firefly algorithm can more effectively avoid local optimal solutions and achieve a global search for optimal hyperparameters. This optimization process not only improves model performance but also reduces the manual parameter adjustment work during model training, greatly enhancing the model's automation and intelligence. Through global search and fitness evaluation, this paper improves the accuracy and robustness of deep learning models in practical applications, demonstrating its strong innovation.

[0089] In this embodiment, the S4 specifically includes:

[0090] S41. Based on the optimized hyperparameters, determine the number of network layers and the number of neurons in each layer of the convolutional neural network, and select the appropriate activation function type, including ReLU, sigmoid, or tanh;

[0091] S42. Determine the number of convolution layers and the size of the convolution kernel of each layer. The size of the convolution kernel is set to 3×3 or 5×5, and the step size is set to 1 or 2. After each convolution layer, a pooling layer is connected. The maximum pooling method is used for feature downsampling. The pooling window size is 2×2 and the pooling step size is 2.

[0092] S43. Add a Dropout layer after the convolutional layer and pooling layer, set the dropout rate to a value range of 0.2 to 0.5, and enhance the generalization ability of the model by discarding neuronal connections;

[0093] S44. Select the cross entropy loss function for the classification task. The loss function formula is:

[0094]

[0095] Among them, y i is the actual label, is the model prediction output, N is the number of samples;

[0096] S45. Use the Adam optimizer for training, optimize the network weights and biases through the back-propagation algorithm, minimize the loss function, and finally generate an optimized charging pile anomaly recognition model;

[0097] S46. Verify the performance of the adjusted model on the validation set. If the precision and recall rate meet the requirements, determine the final optimized network structure.

[0098] The present invention refines the structure of the charging pile anomaly recognition model, specifically optimizing the configuration of the convolutional layer, pooling layer, activation function, and dropout layer in the design of the convolutional neural network. Compared with the prior art, the structural optimization of the present invention is more targeted and can efficiently extract features from charging pile operation data. By adjusting the convolution kernel size, step size, and pooling layer settings, the present invention further enhances the model's sensitivity and recognition ability to abnormal charging pile conditions, particularly with diverse and complex data inputs.

[0099] In this embodiment, the S42 specifically includes: setting the number of convolutional layers of the convolutional neural network, and selecting the convolution kernel size and step size for each convolution layer; the size of the convolution kernel is 3×33 or 5×55, and the step size is set to 1 or 2; after each convolution layer, adding a maximum pooling layer, the pooling window size is 2×22, and the pooling step size is 2; the convolution layer and the pooling layer are stacked in sequence, and the output of each convolution layer is downsampled through the pooling layer.

[0100] In this embodiment, the S5 specifically includes:

[0101] S51. Collect various operating parameter data of the charging pile in real time through sensors, including current, voltage, temperature, and power, to maintain the real-time and accuracy of the data;

[0102] S52, preprocessing the collected real-time data using the same processing method as the training set data, including denoising, filling missing values, normalization, etc., to maintain a format consistent with the training data;

[0103] S53: Input the pre-processed real-time data into the optimized charging pile abnormality recognition model. The model recognizes the abnormal state according to the input features and outputs the prediction result probability value.

[0104] S54, according to the probability value of the prediction result Compare with the set threshold (θ), if It is considered an abnormal state; if It is judged to be in normal state;

[0105] S55. When an abnormal state is identified, an abnormal alarm message is generated, including the abnormality type, occurrence time, and fault location, and sent to maintenance personnel for processing;

[0106] S56. Maintenance personnel locate faults based on alarm information and adjust and optimize the system through feedback mechanisms.

[0107] In this embodiment, S6 specifically includes:

[0108] S61. When the charging pile abnormality recognition model identifies an abnormal state, it generates alarm information including the abnormality type, occurrence time, fault location, and abnormality severity level;

[0109] S62. The system determines the alarm priority based on the severity level of the abnormality. If the abnormality severity level is high, the system triggers an emergency alarm and sends an emergency processing notice to the maintenance personnel first; if the abnormality severity level is low, the system sends the processing information according to the low priority level;

[0110] S63. Alarm information is sent to maintenance personnel in real time via SMS, email or APP notification;

[0111] S64. After receiving the alarm information, the maintenance personnel locate the problem according to the fault type and location and perform corresponding maintenance operations;

[0112] S65. After the maintenance personnel complete the maintenance operation, they feed back the processing results to the system, and the system updates the fault status and records it;

[0113] S66. The system updates the equipment status and optimizes the abnormality recognition model based on the maintenance feedback information.

[0114] In this embodiment, S7 specifically includes: continuously collecting operating data of the charging pile through sensors, regularly inputting new data into the optimized charging pile abnormality recognition model, updating the model using an incremental learning method, adjusting model parameters, and regularly evaluating the performance of the model on new data through a feedback mechanism, and further optimizing the recognition model based on the evaluation results.

[0115] Example 1:

[0116] In order to verify the feasibility of the present invention in implementation, the present invention was applied to a certain electric vehicle charging pile management system, which is deployed at multiple charging stations to monitor and manage the operating status of the charging piles. Charging piles often face problems such as voltage fluctuations, current overloads, and temperature anomalies in daily use. Traditional monitoring methods often rely on manual inspections, which have a high probability of false alarms and missed alarms, and are unable to discover and handle problems in real time. In order to solve these problems, the present invention proposes a charging pile anomaly identification method based on deep learning. Through the deep learning algorithm, the data of the charging pile is monitored in real time, anomaly identification and alarm are performed, ensuring that the charging pile can promptly detect problems and take countermeasures during operation.

[0117] In this charging pile management system, sensors first collect real-time operating data from charging piles, including parameters such as current, voltage, temperature, and power. This data is transmitted to a central monitoring system, where it is analyzed using a pre-trained deep learning model. By training on historical data, the model can distinguish between normal operation and abnormal conditions and provide real-time fault warnings based on real-time data. During model training, we employed a deep learning method based on a convolutional neural network (CNN), combined with the Firefly algorithm to optimize hyperparameters, ensuring the model's efficiency and accuracy.

[0118] In practical applications, charging pile operating data first undergoes preprocessing, including denoising, filling in missing values, and normalization. After processing, the data is fed into an anomaly recognition model. The system analyzes the data in real time, outputting predictions and identifying any abnormal conditions. For example, if the charging pile current exceeds a set safety range, the model quickly detects this anomaly and generates an alarm, prompting maintenance personnel to address it promptly. This process eliminates the manual intervention required by traditional methods, improving the speed and accuracy of detecting charging pile faults.

[0119] In actual testing, the charging pile system successfully identified various types of faults by collecting and analyzing data in real time. For example, a charging pile overheated due to current overload, which traditional monitoring methods were unable to detect in time, leading to damage to the equipment. Using the charging pile anomaly identification method of the present invention, the system can immediately trigger an alarm when the current exceeds a preset threshold and send a fault message to maintenance personnel, successfully preventing further damage to the equipment. The system is also capable of self-learning and optimizing based on the operating status data of the charging pile, improving the accuracy and robustness of future anomaly identification.

[0120] To verify the effectiveness of this invention, we collected historical operating data of a charging station in a real environment. The following is a partial data display:

[0121] Table 1: Charging pile historical operation data and abnormality identification results

[0122]

[0123] Based on the historical operation data of the charging pile and the abnormality identification results in the table, we can make the following analysis:

[0124] This data set contains the historical operating conditions and fault identification results for five charging piles. The table includes information such as the charging pile number, current, voltage, temperature, power, anomaly type, identification result, handling time, and handling personnel. By analyzing this data, we can understand the operating status of the charging piles and how the system responds and handles faults in abnormal situations.

[0125] First, charging piles 001 and 005 both experienced an "overload current" anomaly, identified as "abnormal" during anomaly identification. The system generated alarms for these two charging piles, which were addressed by maintenance personnel Zhang and Wang within 15 and 12 minutes, respectively. In both cases, the current anomaly could have caused overheating or damage, but prompt troubleshooting prevented further damage.

[0126] Charging station number 003 experienced an "overtemperature" issue and was identified as an anomaly. Li resolved the issue within 10 minutes, demonstrating the system's ability to promptly identify temperature anomalies and respond quickly, preventing more serious consequences from overheating.

[0127] Charging pile numbers 002 and 004 were both identified as "normal" and no faults occurred, which shows that the system can accurately distinguish between normal and abnormal conditions, further improving the efficiency of charging pile management.

[0128] In summary, the data in the table demonstrates the effectiveness of this invention in charging pile fault detection. By using a deep learning model, abnormal charging pile conditions are promptly and accurately identified, and the system ensures safe operation of the equipment through real-time alarms and rapid processing. This not only improves fault handling efficiency but also reduces the workload of manual inspections, significantly enhancing the intelligent management of charging piles.

[0129] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A charging pile abnormality identification method based on deep learning, characterized in that: The steps include: S1. Obtaining operation data of the charging pile and preprocessing the data; S2. Based on the pre-processed operating data, a charging pile anomaly recognition model is constructed using a deep learning algorithm. This model is trained on historical data to generate a preliminary charging pile anomaly recognition model. S3. Use the firefly algorithm to optimize the hyperparameters of the charging pile anomaly recognition model, optimize the learning rate, number of network layers, and activation function hyperparameters, perform a global search, and evaluate the optimization effect through the fitness function; S4. Based on the optimized hyperparameters, adjust the network structure of the charging pile anomaly recognition model, adjust the number of convolutional layers, fully connected layer structure, and activation function selection; S5. Collecting the operating data of the charging pile in real time, and inputting the data into the optimized charging pile abnormality recognition model to identify abnormal conditions and generate abnormal alarm information, which includes the fault type, occurrence time and location, etc.; S6. Based on the abnormal alarm information, the alarm mechanism is automatically triggered and the fault information is sent to the maintenance personnel for fault location and repair; S7. Continue to collect new charging pile data and input it into the optimized model to update the model.

2. The method for identifying abnormalities in charging piles based on deep learning according to claim 1, characterized in that: The S1 specifically includes: obtaining the operating data of the charging pile, including current, voltage, temperature, and power parameters, and preprocessing the data; using the sliding average method or wavelet transform to remove noise, and filling missing values ​​through the linear interpolation method or the K-nearest neighbor algorithm; using the minimum-maximum normalization to standardize the data to the range of [0,1]; using the standard deviation method to eliminate outliers that deviate from the mean by more than 3 times the standard deviation; using the Pearson correlation coefficient analysis method to select features related to the abnormal state of the charging pile and remove correlated features; dividing the processed data set into a training set and a test set in a ratio of 7:3; generating a charging pile data feature set.

3. The method for identifying abnormalities in charging piles based on deep learning according to claim 1, characterized in that: The S2 specifically includes: selecting a convolutional neural network as the basic model, designing the model, setting the number of network layers and the number of neurons in each layer; when using CNN, selecting a combination of several convolutional layers and pooling layers, and extracting spatial feature information in the input data through the convolution kernel; adding a maximum pooling layer after each convolutional layer to perform feature downsampling; selecting the activation function as ReLU; in the output layer of the model, selecting the softmax function for classification, and outputting the probability of whether the charging pile is in an abnormal state; using the cross-entropy loss function to measure the difference between the model output and the true label, adjusting the weights and biases in the network through the backpropagation algorithm, and minimizing the loss function; using the training set for iterative training to generate a preliminary charging pile anomaly recognition model.

4. The method for identifying abnormalities in charging piles based on deep learning according to claim 1, characterized in that: The S3 specifically includes: S31. Define the hyperparameter optimization space and select the hyperparameters to be optimized, including the learning rate, the number of network layers, the number of neurons in each layer, and the activation function type. Set the learning rate η within a preset range, and determine the range of the number of network layers L and the number of neurons in each layer N. The activation function type σ includes ReLU, sigmoid, or tanh. S32. Initialize a number of fireflies, each firefly represents a set of hyperparameter combinations, and evaluate the fitness of each firefly using a fitness function, where the fitness function is the classification accuracy or loss function value of the model on the validation set; S33. Calculate the attraction between fireflies. The attraction is proportional to the fitness. The attraction calculation formula is: Among them, A ij is the attraction of firefly i to firefly j, β is the maximum attraction, γ is the attraction attenuation coefficient, r ij is the distance between fireflies i and j; S34. Update the position of the "firefly" according to the attraction. The position update formula is: X i =X i +α(X i -X j )+∈; Among them, X i and X j is the current coordinate of fireflies i and j, α is the step coefficient, and ∈ is the random perturbation term; S35, repeating the position update process until the optimization stop criterion is met, such as reaching the maximum number of iterations or the fitness change is lower than the set threshold; S36. Output the optimal hyperparameter combination and apply it to the charging pile anomaly recognition model.

5. The method for identifying abnormalities in charging piles based on deep learning according to claim 1, characterized in that: The S4 specifically includes: S41. Based on the optimized hyperparameters, determine the number of network layers and the number of neurons in each layer of the convolutional neural network, and select the appropriate activation function type, including ReLU, sigmoid, or tanh; S42. Determine the number of convolution layers and the size of the convolution kernel of each layer. The size of the convolution kernel is set to 3×3 or 5×5, and the step size is set to 1 or 2. After each convolution layer, a pooling layer is connected. The maximum pooling method is used for feature downsampling. The pooling window size is 2×2 and the pooling step size is 2. S43. Add a Dropout layer after the convolutional layer and pooling layer, set the dropout rate to a value range of 0.2 to 0.5, and enhance the generalization ability of the model by discarding neuronal connections; S44. Select the cross entropy loss function for the classification task. The loss function formula is: Among them, y i is the actual label, is the model prediction output, N is the number of samples; S45. Use the Adam optimizer for training, optimize the network weights and biases through the back-propagation algorithm, minimize the loss function, and finally generate an optimized charging pile anomaly recognition model; S46. Verify the performance of the adjusted model on the validation set. If the precision and recall rate meet the requirements, determine the final optimized network structure.

6. The method for identifying abnormalities in charging piles based on deep learning according to claim 5, characterized in that: The S42 specifically includes: setting the number of convolutional layers of the convolutional neural network, and selecting the convolution kernel size and step size for each convolution layer; the size of the convolution kernel is 3×33 or 5×55, and the step size is set to 1 or 2; after each convolution layer, adding a maximum pooling layer, the pooling window size is 2×22, and the pooling step size is 2; the settings of the convolution layer and the pooling layer are stacked in sequence, and the output of each convolution layer is downsampled through the pooling layer.

7. The method for identifying abnormalities in charging piles based on deep learning according to claim 1, characterized in that: The S5 specifically includes: S51. Collect various operating parameter data of the charging pile in real time through sensors, including current, voltage, temperature, and power, to maintain the real-time and accuracy of the data; S52, preprocessing the collected real-time data using the same processing method as the training set data, including denoising, filling missing values, normalization, etc., to maintain a format consistent with the training data; S53: Input the pre-processed real-time data into the optimized charging pile abnormality recognition model. The model recognizes the abnormal state according to the input features and outputs the prediction result probability value. S54, according to the probability value of the prediction result Compare with the set threshold (θ), if It is considered an abnormal state; if It is judged to be in normal state; S55. When an abnormal state is identified, an abnormal alarm message is generated, including the abnormality type, occurrence time, and fault location, and sent to maintenance personnel for processing; S56. Maintenance personnel locate faults based on alarm information and adjust and optimize the system through feedback mechanisms.

8. The method for identifying abnormalities in charging piles based on deep learning according to claim 1, characterized in that: The S6 specifically includes: S61. When the charging pile abnormality recognition model identifies an abnormal state, it generates alarm information including the abnormality type, occurrence time, fault location, and abnormality severity level; S62. The system determines the alarm priority based on the severity level of the abnormality. If the abnormality severity level is high, the system triggers an emergency alarm and sends an emergency processing notice to the maintenance personnel first; if the abnormality severity level is low, the system sends the processing information according to the low priority level; S63. Alarm information is sent to maintenance personnel in real time via SMS, email or APP notification; S64. After receiving the alarm information, the maintenance personnel locate the problem according to the fault type and location and perform corresponding maintenance operations; S65. After the maintenance personnel complete the maintenance operation, they feed back the processing results to the system, and the system updates the fault status and records it; S66. The system updates the equipment status and optimizes the abnormality recognition model based on the maintenance feedback information.

9. The method for identifying abnormalities in charging piles based on deep learning according to claim 1, characterized in that: The S7 specifically includes: continuously collecting the operating data of the charging pile through sensors, regularly inputting new data into the optimized charging pile abnormality recognition model, updating the model using the incremental learning method, adjusting the model parameters, and regularly evaluating the performance of the model on new data through the feedback mechanism, and further optimizing the recognition model based on the evaluation results.

Images