New energy vehicle charging pile fault monitoring method and system

Abstract

The embodiment of the invention provides a new energy automobile charging pile fault monitoring method and system, and relates to the technical field of charging pile fault detection. The new energy automobile charging pile fault monitoring method comprises the following steps: S1, collecting multi-source data in the operation process of a charging pile; s3, preprocessing the multi-source data to generate a multi-dimensional feature data set; s3, constructing a dynamic fault threshold model; s4, on the basis of the multi-dimensional feature data set and the dynamic fault threshold model, multi-level fault detection is executed; s5, establishing a fault prediction mechanism; and S6, generating a fault monitoring report according to the fault type and the fault severity evaluation index. Through multi-source data acquisition, dynamic threshold construction and a multi-level fault detection and prediction mechanism, real-time accurate monitoring and fault early warning of the operation state of the charging pile can be realized, and compared with a traditional method, the accuracy and the advancement of fault identification are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of charging pile fault detection, and specifically refers to a method and system for monitoring faults of a new energy vehicle charging pile. Background Art

[0002] With the rapid popularization of new energy vehicles, charging piles, as their core infrastructure, play a crucial role in ensuring the energy supply of vehicles. However, during the actual operation of charging piles, due to being in a high-load state for a long time or being affected by environmental factors (such as high temperature and high humidity), various faults are likely to occur, such as electrical short circuits, overloads, communication interruptions, or heat dissipation failures. These faults not only affect the charging efficiency and user experience but also may pose safety hazards, and even lead to equipment damage or fire accidents. Therefore, real-time monitoring of the operating status of charging piles and fault early warning have become technical problems that the industry urgently needs to solve.

[0003] In the prior art, the fault monitoring of charging piles mainly relies on single-parameter threshold detection or manual regular inspections. For example, some methods judge abnormalities by setting fixed current or voltage thresholds, but this method cannot adapt to the dynamic changes of charging piles under different loads and environmental conditions, resulting in a high false alarm or missed alarm rate; other solutions collect data through sensors and then identify faults by combining simple statistical analysis, but lack the ability to predict the development trend of faults and cannot take preventive measures in advance. In addition, the prior art mostly focuses on local monitoring and fails to fully utilize the advantages of cloud computing and multi-device collaboration, restricting the comprehensiveness and intelligence level of monitoring.

[0004] In view of the above problems, there is an urgent need for a method for monitoring faults of a charging pile that can integrate multi-source data, dynamically adjust thresholds, detect faults at multiple levels, and have a prediction function, so as to improve the monitoring accuracy, reduce the operation and maintenance costs, and ensure charging safety. Summary of the Invention

[0005] According to an embodiment of the present invention, there is provided a method and system for monitoring faults of a new energy vehicle charging pile, which are used to solve the problems raised in the above background art.

[0006] In a first aspect of the present invention, there is provided a method and system for monitoring faults of a new energy vehicle charging pile.

[0007] The method for monitoring faults of the new energy vehicle charging pile includes the following steps:

[0008] S1. Collect multi-source data during the operation of the charging pile, where the multi-source data includes the real-time current value, voltage value, temperature value, charging power, environmental humidity, and communication signal integrity parameter between the charging pile and the new energy vehicle;

[0009] S2. Preprocess the multi-source data to generate a multi-dimensional feature dataset;

[0010] S3. Construct a dynamic fault threshold model, which is implemented through the following sub-steps:

[0011] S3a. According to the historical operation data and current environmental parameters, use a long short-term memory network to predict the normal operation range of the charging pile under different load conditions;

[0012] S3b. Combine the normal operation range and adopt an adaptive Kalman filter algorithm to adjust the fault threshold in real time to adapt to the dynamic changes in the operation state of the charging pile;

[0013] S4. Based on the multi-dimensional feature dataset and the dynamic fault threshold model, perform multi-level fault detection, and the multi-level fault detection includes:

[0014] S4a. First-level detection: Use a support vector machine classifier to perform preliminary anomaly classification on the multi-dimensional feature dataset to determine whether there are potential faults;

[0015] S4b. Second-level detection: For the preliminary anomaly classification results, use a deep belief network to deeply mine the abnormal features and extract the fault mode feature vectors;

[0016] S4c. Third-level detection: Input the fault mode feature vectors into a pre-trained Bayesian network, calculate the probability distribution of various faults, and determine the specific fault types;

[0017] S5. Establish a fault prediction mechanism, and the fault prediction mechanism includes:

[0018] S5a. Based on the fault mode feature vectors extracted in step S4, combine with a Markov chain model to predict the time window of fault occurrence;

[0019] S5b. According to the predicted time window and fault types, generate a fault severity evaluation index;

[0020] S6. According to the fault types and the fault severity evaluation index, generate a fault monitoring report, and upload the fault monitoring report to the cloud management system in real time through the communication module of the charging pile, and at the same time push a fault warning message to the user terminal.

[0021] Preferably, the communication signal integrity parameters collected in step S1 include the dynamic combination of the following sub-parameters:

[0022] S1a. The CAN bus data packet loss rate, which is calculated through the analysis of the packet sequence consistency within a sliding time window, and the frequency domain feature extraction based on the Fourier transform is introduced;

[0023] S1b. The signal delay time is measured using a time synchronization algorithm based on phase difference and combined with real-time correction of the ambient electromagnetic interference intensity;

[0024] S1c. The protocol handshake success rate is evaluated by a multi-state hidden Markov model to assess the state transition probability during the handshake process, and weighted feedback adjustment is performed on abnormal states.

[0025] Preferably, the preprocessing process in step S2 includes the following sub-steps:

[0026] S2a. Apply denoising processing based on dual-tree complex wavelet transform to the collected multi-source data, where the number of wavelet decomposition layers is adaptively selected according to the non-stationary characteristics of the signal;

[0027] S2b. Outlier rejection uses the DBSCAN algorithm based on density clustering and combines the local outlier factor of the time series for secondary verification;

[0028] S2c. Data standardization processing introduces a feature distribution reconstruction mechanism based on a generative adversarial network to eliminate the dimensional difference and distribution deviation between different sensor data.

[0029] Preferably, the dynamic fault threshold model in step S3 is further defined by the following enhanced sub-steps:

[0030] S3c. Use a bidirectional long short-term memory network combined with a graph attention network to extract spatio-temporal features from historical operation data and real-time environmental parameters, where the graph structure is dynamically constructed based on the dependence relationship between the charging pile and environmental variables;

[0031] S3d. On the basis of the adaptive Kalman filter in step S3b, introduce a non-linear state transition equation, and perform probability estimation and optimization of the uncertainty during the filtering process through particle filtering;

[0032] S3e. According to the operating conditions of the charging pile and the historical fault frequency, adopt reinforcement learning to adjust the threshold update strategy to maximize the accuracy and recall rate of fault detection.

[0033] Preferably, the multi-level fault detection in step S4 is further refined into the following sub-steps:

[0034] S4d. The first-level detection classifies primary anomalies for the multi-dimensional feature data set through a support vector machine combined with fuzzy C-means clustering;

[0035] S4e. The second-level detection uses a joint model of a deep belief network and a variational autoencoder to extract the latent distribution representation of abnormal features and enhance the feature robustness through adversarial training;

[0036] S4f. The third-level detection inputs the fault mode feature vector into the multi-time-step inference framework based on the dynamic Bayesian network, combines the evidence theory to fuse the multi-source fault probabilities, and outputs the accurate fault type distribution.

[0037] Preferably, the fault prediction mechanism in step S5 is implemented through the following sub-steps:

[0038] S5c. Based on the fault mode feature vector in step S4, construct a multi-dimensional Markov chain model, where the state transition matrix is updated in real time through online learning, and introduce Gaussian process regression to predict the fault evolution trend;

[0039] S5d. Combining the fault evolution trend, use Monte Carlo simulation to generate the probability density function of the fault occurrence time window, and evaluate the uncertainty of the prediction through information entropy;

[0040] S5e. The fault severity evaluation index is calculated through multi-objective optimization, and the formula is as follows: S = α·softmax(P f ) + β·exp(-T p ·σ) + γ·∫I d (t)·w(t)dt.

[0041] Preferably, the generation of the fault monitoring report and warning information in step S6 includes the following sub-steps:

[0042] S6a. The fault monitoring report is automatically generated in multiple languages by the natural language generation model according to the fault type, probability distribution, and severity index, and a visual fault trend graph is embedded;

[0043] S6b. The warning information push adopts a personalized strategy based on user behavior analysis;

[0044] S6c. The cloud management system uses blockchain technology to encrypt and store the fault monitoring report and perform distributed verification to ensure data integrity and traceability.

[0045] Preferably, the method further includes the following enhancement steps:

[0046] S7a. In the cloud management system, construct a knowledge graph-based charging pile fault knowledge base, and use graph convolutional network to mine the causal relationships and propagation paths between faults;

[0047] S7b. Use the federated learning framework to distributively update the dynamic fault threshold model and fault prediction mechanism from multiple charging pile nodes, while protecting data privacy;

[0048] S7c. Introduce an adaptive evolutionary algorithm to dynamically optimize all model parameters according to long-term operation data, and achieve collaborative fault monitoring of the charging pile group through multi-agent collaboration.

[0049] In the second aspect of the present invention, a fault monitoring system for a new energy vehicle charging pile is provided.

[0050] The system includes a data acquisition module, a data preprocessing unit, a dynamic threshold generation module, a fault detection module, a fault prediction module, an information processing and pushing unit, and a cloud management system. The data acquisition module is connected to the data preprocessing unit, the data preprocessing unit is connected to the dynamic threshold generation module, the dynamic threshold generation module is connected to the fault detection module, the fault detection module is connected to the fault prediction module, the fault prediction module is connected to the information processing and pushing unit, the information processing and pushing unit is connected to the cloud management system, and the cloud management system is connected to the data acquisition module.

[0051] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0052] A fault monitoring method for a new energy vehicle charging pile provided by the present invention can realize real-time and accurate monitoring of the operating state of the charging pile and fault warning through multi-source data acquisition, dynamic threshold construction, multi-level fault detection and prediction mechanisms. Compared with traditional methods, it significantly improves the accuracy and advance of fault identification; the intelligent algorithms and adaptive adjustment strategies adopted not only effectively reduce the false alarm rate and missed alarm rate, but also can dynamically optimize the monitoring performance according to environmental changes and operating conditions, thereby extending the service life of the charging pile, improving the charging safety of users and the operation and maintenance efficiency, and having significant technological progress and application value.

[0053] It should be understood that the content described in the summary of the invention is not intended to limit the key or important features of the embodiments of the present invention, nor to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. Brief Description of the Drawings

[0054] Combined with the drawings and referring to the following detailed description, the above and other features, advantages and aspects of the embodiments of the present invention will become more obvious. In the drawings, the same or similar reference numerals represent the same or similar elements, where:

[0055] Figure 1 Shows a flowchart of a fault monitoring method for a new energy vehicle charging pile according to an embodiment of the present invention;

[0056] Figure 2 Shows a system block diagram of a fault monitoring system for a new energy vehicle charging pile according to an embodiment of the present invention; Detailed Embodiments

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0058] In addition, the term "and / or" in this article is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article generally represents an "or" relationship between the associated objects before and after.

[0059] The present invention provides a method for monitoring faults in a new energy vehicle charging pile, aiming to achieve precise monitoring and fault prediction of the operating state of the charging pile through multi-source data collection, dynamic threshold construction, and multi-level fault detection. The following details this method in combination with specific implementation steps:

[0060] Step S1: Multi-source data collection

[0061] During the operation of the charging pile, multi-source data is collected in real time through sensors and communication modules installed inside the charging pile. These data include, but are not limited to:

[0062] The real-time current value output by the charging pile, measured by a high-precision current sensor, reflecting the current stability during the charging process;

[0063] The real-time voltage value of the charging pile, collected by a voltage sensor, for monitoring voltage fluctuations;

[0064] The temperature values of key components inside the charging pile (such as power modules and heat dissipation devices), obtained through thermistors or infrared thermometers, to evaluate the overheating risk;

[0065] The charging power, calculated in real time by a power calculation module based on the product of current and voltage, reflecting the charging efficiency;

[0066] The ambient humidity, measured by a humidity sensor to obtain the humidity value of the environment where the charging pile is located, and analyze the impact of moisture on the operation of the equipment;

[0067] The communication signal integrity parameters between the charging pile and the new energy vehicle, including the integrity of data packet transmission, signal delay, and the response status of the communication protocol, monitored through a communication interface module, for judging the reliability of the communication link.

[0068] In actual implementation, the data acquisition frequency can be set to once per second and recorded through timestamps to ensure the timing accuracy of subsequent analysis.

[0069] Step S2: Multi-source data preprocessing

[0070] Preprocess the multi-source data sampled in Step S1 to eliminate noise, outliers, and data distribution differences, and generate a multi-dimensional feature dataset suitable for fault detection. The specific preprocessing process includes the following sub-steps:

[0071] Time series denoising: Use a wavelet transform-based method to process the collected raw data. By selecting an appropriate wavelet basis function (such as the Daubechies wavelet), decompose the signal into multiple scales, remove the high-frequency noise components, and retain the low-frequency trend information. The denoised data can better reflect the true operating state of the charging pile.

[0072] Outlier removal: Detect outliers in the denoised data using a statistical rule-based method (such as the three-standard-deviation method), combined with the continuity characteristics of the time series, and remove the outliers caused by sensor failures or external interferences. For example, if the current value at a certain moment suddenly exceeds three standard deviations of the historical mean and there is no obvious trend support in the surrounding data, it is determined as an outlier and removed.

[0073] Data standardization processing: Convert data with different dimensions (such as current, voltage, temperature) into a dimensionless form using the min-max normalization method, and map the data to the [0,1] interval to ensure the equal weight of each dimension feature during subsequent model processing.

[0074] After the preprocessing is completed, the generated multi-dimensional feature dataset contains time-series parameters such as current, voltage, and temperature, providing reliable input for subsequent fault detection.

[0075] Step S3: Construction of a dynamic fault threshold model

[0076] Based on the multi-dimensional feature dataset generated in Step S2, construct a dynamic fault threshold model to determine whether the operating state of the charging pile is abnormal. This model is implemented through the following sub-steps:

[0077] S3a. Prediction of the normal operating range: Use historical operating data (such as the current, voltage, and temperature records in the past 30 days) and current environmental parameters (such as environmental temperature and humidity), and use a long short-term memory network to predict the normal operating range of the charging pile under different load conditions. The long short-term memory network analyzes the long-term dependence relationship of the time series, outputs the normal range of each parameter (such as the current is between 50A and 100A), and dynamically adjusts the prediction result according to the load change.

[0078] S3b. Real-time adjustment of fault threshold: Combining the normal operation range predicted in step S3a, the adaptive Kalman filter algorithm is used to update the fault threshold in real time. The Kalman filter dynamically adjusts the threshold by fusing historical data and real-time collected data through state estimation and observation update. For example, when the ambient temperature rises and the normal current range moves up, the threshold is adaptively adjusted accordingly to avoid misjudgment.

[0079] In actual implementation, the dynamic threshold model can be deployed in the embedded controller of the charging pile and updated once a minute to ensure a quick response to changes in the operating state.

[0080] Step S4: Multi-level fault detection

[0081] Based on the multi-dimensional feature dataset in step S2 and the dynamic fault threshold model in step S3, multi-level fault detection is performed to accurately identify the fault type. It specifically includes the following sub-steps:

[0082] S4a. First-level detection - preliminary anomaly classification: Analyze the multi-dimensional feature dataset through a support vector machine classifier to determine whether there are potential faults. The support vector machine trains a model based on preset normal and abnormal samples, maps real-time data to the feature space, and outputs the anomaly probability. If the probability exceeds 50%, it is determined as a potential fault and proceeds to the next level of detection.

[0083] S4b. Second-level detection - in-depth mining of abnormal features: For the potential fault data identified in step S4a, use a deep belief network to deeply analyze the abnormal features. The deep belief network extracts non-linear patterns in the data through a multi-layer neural network structure and generates a fault pattern feature vector, such as the temperature anomaly feature caused by overheating or the signal feature caused by communication interruption.

[0084] S4c. Third-level detection - determination of fault type: Input the fault pattern feature vector extracted in step S4b into a pre-trained Bayesian network to calculate the probability distribution of various faults. The Bayesian network constructs a conditional probability table based on historical fault data and combines it with the real-time feature vector to output the probability of specific fault types (such as short circuit, overload, communication failure), and finally determines the most likely fault type.

[0085] In practical applications, the three-level detection can be completed collaboratively by the cloud server. The first level is executed locally, and the latter two levels are uploaded to the cloud for analysis to improve the computing efficiency.

[0086] Step S5: Establishment of a fault prediction mechanism

[0087] Based on the detection results in step S4, a fault prediction mechanism is established to early warn of potential faults. It specifically includes the following sub-steps:

[0088] S5a. Fault occurrence time prediction: Using the fault mode feature vectors extracted in step S4b, combined with the Markov chain model to predict the time window of fault occurrence. The Markov chain calculates the probability of fault occurrence within a future period (such as within 24 hours) according to the state transition law of the feature vectors, and outputs the time range (such as "It is predicted that an overload will occur in 6 hours").

[0089] S5b. Fault severity assessment: Generate a fault severity assessment index according to the time window predicted in step S5a and the fault type determined in step S4c. The assessment index comprehensively considers the fault probability, the urgency of occurrence time, and the degree of impact on the operation of the charging pile, and is obtained through weighted calculation. For example, a short-circuit fault has a high severity score because of its large impact and high urgency.

[0090] In implementation, the prediction results can be displayed through a visualization interface to facilitate maintenance personnel to take measures in a timely manner.

[0091] Step S6: Fault monitoring report generation and information push;

[0092] Generate a detailed fault monitoring report according to the fault type in step S4c and the severity assessment index in step S5b. The report content includes the fault type, occurrence probability, predicted time, and recommended maintenance measures (such as "Replace the cooling fan"). Through the communication module of the charging pile (such as a 4G or Wi-Fi module), the report is uploaded to the cloud management system in real time. At the same time, use text messages or mobile applications to push fault warning information to the user terminal to prompt the user to suspend use or contact the maintenance personnel.

[0093] In an actual scenario, the cloud system can store historical reports for subsequent fault trend analysis and equipment optimization.

[0094] The present invention provides a method for monitoring faults in a new energy vehicle charging pile, which realizes accurate detection and prediction of faults through multi-level data processing and intelligent analysis. The implementation process is described in detail below in combination with the sub-steps of each step:

[0095] In this embodiment, the communication signal integrity parameters collected in step S1 include the dynamic combination of the following sub-parameters to ensure the reliability of the communication link:

[0096] S1a. Calculation of CAN bus data packet loss rate: Analyze the consistency of the CAN bus data packet sequence by setting a sliding time window (such as 10 seconds), and calculate the proportion of lost data packets per unit time. To improve the accuracy, a frequency domain feature extraction method based on Fourier transform is introduced to convert the time series into a frequency domain signal to identify periodic packet loss patterns. For example, if a packet loss peak at a specific frequency is detected, it can be judged as an abnormality caused by communication interference.

[0097] S1b. Signal Delay Time Measurement: Using a time synchronization algorithm based on phase difference, measure the signal transmission delay time between the charging pile and the new energy vehicle. Specifically, during implementation, send test data packets and record their arrival times, calculate the phase difference, and perform real-time correction in combination with the ambient electromagnetic interference intensity (measured by an electromagnetic sensor, for example). If the interference intensity exceeds a preset threshold (such as 50 μT), the delay time will be dynamically weighted and adjusted.

[0098] S1c. Protocol Handshake Success Rate Evaluation: Analyze the handshake process of the communication protocol between the charging pile and the vehicle through a multi-state hidden Markov model to evaluate the state transition probability. For example, divide the handshake into three states: "initiation", "response", and "confirmation", and calculate the success probability of each state. For abnormal states (such as response timeout), adopt a weighted feedback mechanism to adjust the model parameters to improve the robustness of the evaluation.

[0099] In this embodiment, the preprocessing process in step S2 includes the following sub-steps to ensure that the data quality meets the requirements of subsequent analysis:

[0100] S2a. Denoising by Dual-Tree Complex Wavelet Transform: Apply the dual-tree complex wavelet transform to the collected multi-source data (such as current and voltage) for denoising. Compared with the traditional wavelet transform, this method provides better directionality and shift invariance through a dual-tree structure. The number of wavelet decomposition layers is adaptively selected according to the non-stationary characteristics of the signal. For example, if the signal fluctuates violently, the decomposition layer can be increased to 5 layers, and the signal is reconstructed after removing high-frequency noise.

[0101] S2b. Outlier Removal and Verification: Use an algorithm based on density clustering (such as DBSCAN) to identify outliers in the data, and set parameters such as the neighborhood radius (such as 0.5) and the minimum number of samples (such as 5). For the preliminarily removed outliers, perform secondary verification in combination with the local outlier factor method of the time series to ensure that only the truly abnormal points are removed. For example, if a certain temperature value deviates from the local density of the surrounding data, it is confirmed as abnormal.

[0102] S2c. Data Standardization and Distribution Reconstruction: Standardize the multi-source data through a generative adversarial network. The generator learns the latent distribution of the data and generates standardized samples consistent with the real data, while the discriminator optimizes the distribution deviation, ultimately eliminating the dimensional differences of data such as current and voltage. For example, the current value is mapped from 50 A - 100 A to the interval [0, 1], and the temperature value is also mapped from 20 °C - 80 °C to [0, 1].

[0103] In this embodiment, the dynamic fault threshold model in step S3 is further defined by the following enhanced sub-steps:

[0104] S3c. Spatiotemporal feature extraction: The bidirectional long short-term memory network is combined with the graph attention network to analyze the historical operation data and real-time environmental parameters. The bidirectional long short-term memory network captures the dependencies between the time series, and the graph attention network builds a graph structure based on the dynamic dependencies between the charging pile and the environmental variables (such as temperature and humidity), and assigns feature weights through the attention mechanism. For example, the weight of the temperature feature will increase dynamically in a high temperature environment.

[0105] S3d. Nonlinear filtering optimization: Based on the adaptive Kalman filter in step S3b, a nonlinear state transfer equation is introduced to describe the complex changes in the operating state of the charging pile. The uncertainty in the filtering process is estimated by the particle filtering method, and 1000 particles are used to sample the state space to optimize the accuracy of threshold adjustment. For example, when the voltage fluctuation is nonlinear, the particle filter can effectively track the real state.

[0106] S3e. Threshold update strategy adjustment: According to the operating conditions of the charging pile (such as high load, low load) and the historical fault frequency, the threshold update strategy is optimized by reinforcement learning. With fault detection accuracy and recall rate as the reward function, the threshold update frequency and amplitude are adjusted through trial and error learning. For example, the threshold update frequency can be increased to once every 30 seconds under high load.

[0107] In this embodiment, the multi-level fault detection in step S4 is further refined into the following sub-steps:

[0108] S4d. First-level detection optimization: The first-level detection uses support vector machine combined with fuzzy C-means clustering to perform anomaly classification. The kernel function of the support vector machine uses a hybrid kernel function based on the wavelet kernel, and the kernel parameters and penalty factors are optimized by genetic algorithm (for example, the population size is set to 50 and iterated for 20 generations) to improve the classification accuracy. Fuzzy C-means clustering provides a fuzzy division basis for the anomaly boundary.

[0109] S4e. Second-level feature extraction: The second-level detection uses a joint model of deep belief network and variational autoencoder to extract the potential distribution of abnormal features. The deep belief network learns nonlinear features through multi-layer restricted Boltzmann machines, and the variational autoencoder generates the probability distribution of features and enhances the robustness of features to noise through adversarial training. For example, it extracts temperature anomaly patterns caused by overheating.

[0110] S4f. Third-level fault reasoning: The third-level detection inputs the fault mode feature vector into the dynamic Bayesian network and combines the evidence theory to fuse the multi-source fault probability. The dynamic Bayesian network analyzes the trend of fault evolution over time through multi-time step reasoning, and the evidence theory performs weighted fusion of evidence such as current anomaly and temperature anomaly, and outputs the precise fault type (such as "overload probability 80%)".

[0111] In this embodiment, the fault prediction mechanism in step S5 is implemented through the following sub-steps:

[0112] S5c. Fault evolution prediction: Based on the fault mode feature vector in step S4, a multi-dimensional Markov chain model is constructed, and the state transition matrix is updated in real time through online learning. Gaussian process regression is used to predict the fault evolution trend. For example, the overheat time point is predicted according to the temperature rise rate, and a continuous trend curve is output. Specifically, when implemented, the characteristic data of the past 24 hours is selected as the training sample to predict the fault trend within the next 12 hours.

[0113] S5d. Time window probability estimation: Combining the fault evolution trend, the probability density function of the fault occurrence time window is generated using Monte Carlo simulation. The simulation runs 10,000 times, calculates the confidence interval of the fault occurrence (such as within 6 hours with 95% confidence), and quantifies the uncertainty of the prediction through information entropy. For example, if the information entropy value is less than 0.5, the prediction result has a high credibility.

[0114] S5e. Severity assessment calculation: The fault severity assessment index is calculated through a multi-objective optimization method, comprehensively considering the fault probability, time urgency, and impact degree. The specific formula is as follows:

[0115] S = α·softmax(P f ) + β·exp(-T p ·σ) + γ·∫I d (t)·w(t)dt

[0116] where P f is the fault probability, obtained from the fault type probability distribution in step S4f; T p is the time window, representing the time length when the fault is expected to occur (such as hours), calculated in step S5d; I d (t) is the time-dependent impact degree, statistically analyzing the change of the impact of the fault on the operation of the charging pile over time through historical data; w(t) is the time weight function, adopting an exponential decay form (such as w(t) = e -t ) to emphasize the importance of recent impacts; α, β, γ are dynamically adjusted coefficients, representing the weights of probability, time, and impact respectively, and are solved through the Pareto optimization method (such as setting the objective function to balance accuracy and response speed) to ensure an equilibrium evaluation result. For example, if P f = 0.85, T p = 4 hours, σ = 2, and the integral value of I d (t) is 10, then the severity score S can be calculated to guide the maintenance priority.

[0117] In this embodiment, the generation of the fault monitoring report and warning information in step S6 includes the following sub-steps:

[0118] S6a. Automatic report generation: Through a natural language generation model, generate multilingual reports (such as Chinese, English) based on fault types, probability distributions, and severity indicators, and embed visual charts of fault trends. For example, the report content includes "overload probability 85%, expected to occur within 4 hours".

[0119] S6b. Personalized warning push: Based on user behavior analysis (such as usage frequency), optimize the push time and method through the Q-learning algorithm. The Q-table is updated according to the user response time. For example, select the user active period to push warnings via text message or APP.

[0120] S6c. Secure data storage: The cloud management system uses blockchain technology to encrypt and store fault reports. Each report generates a unique hash value and is verified through distributed nodes to ensure data immutability and traceability.

[0121] In this embodiment, the method further includes the following enhancement steps:

[0122] S7a. Fault knowledge base construction: Build a fault knowledge base based on a knowledge graph in the cloud, and mine the causal relationships between faults through a graph convolutional network. For example, the correlation between overheating and short circuit is explicitly represented by the graph.

[0123] S7b. Distributed model update: Utilize the federated learning framework to distributively update the threshold model and prediction mechanism from multiple charging pile nodes. Each node only shares the model gradient and does not transmit the original data to protect privacy.

[0124] S7c. Parameter optimization and collaborative monitoring: Introduce an adaptive evolutionary algorithm to optimize the model parameters, and achieve collaborative fault monitoring of a charging pile group through multi-agent cooperation. For example, adjacent charging piles share fault patterns to improve the group detection efficiency.

[0125] The present invention also provides a fault monitoring system for a new energy vehicle charging pile, including a data acquisition module, a data preprocessing unit, a dynamic threshold generation module, a fault detection module, a fault prediction module, an information processing and push unit, and a cloud management system;

[0126] The data acquisition module is connected to the data preprocessing unit and is used to transmit the collected original multi-source data to the preprocessing unit for denoising and standardization processing. Specifically, it is connected through the GPIO, I2C interface, and ADC channel of the STM32F4 microcontroller, and transmits data such as current, voltage, and temperature to the data preprocessing unit in the format of 32-bit floating-point numbers once per second.

[0127] The data preprocessing unit is connected to the dynamic threshold generation module, and is used to provide the preprocessed multi-dimensional feature data set to the dynamic threshold generation module to construct the fault threshold. Specifically, it is connected through the internal data bus and shared memory of the STM32F4 microcontroller, and the standardized time series matrix (such as 10×6 dimensions) is transmitted to the dynamic threshold generation module every minute.

[0128] The dynamic threshold generation module is connected to the fault detection module, and is used to provide the real-time updated fault threshold to the fault detection module to judge the abnormal state. Specifically, it is connected through the shared memory locally, and the threshold range (such as 50A - 100A for current) is directly transmitted; through the 4G communication module (model: SIM7600) in the cloud, it is uploaded once per minute using the TCP / IP protocol.

[0129] The fault detection module is connected to the fault prediction module, and is used to transmit the detected fault type and feature vector to the fault prediction module for time prediction and severity assessment. Specifically, it is connected to the cloud server (equipped with NVIDIA GTX 1080 GPU) through the 4G communication module, and the fault probability distribution and feature vector (in JSON format) are uploaded to the fault prediction module every 10 seconds.

[0130] The fault prediction module is connected to the information processing and push unit, and is used to transmit the fault prediction result to the information processing and push unit to generate a report and push a warning. Specifically, it is connected through the high-speed data bus (such as PCIe) inside the cloud server, and the time window and severity assessment metrics (in the form of a structure) are transmitted to the information processing and push unit every hour.

[0131] The information processing and push unit is connected to the cloud management system, and is used to upload the generated fault monitoring report to the cloud management system for storage and optimization. Specifically, it is connected through the gigabit Ethernet interface of the cloud server, and the encrypted report file (using SHA-256 hash) is uploaded to the cloud management system every 5 seconds.

[0132] The cloud management system is connected to the data acquisition module, and is used to transmit the optimized model parameters back to the data acquisition module to update the local system configuration. Specifically, it is connected through the 4G communication module using the bidirectional TCP / IP protocol, and the parameter file (in binary format) is transmitted back to the data acquisition module every 24 hours.

[0133] Furthermore, in actual use, the user connects the new energy vehicle to the charging pile and the system starts automatically. The data acquisition module starts working, and collects multi-source data during the operation of the charging pile through the current sensor (ACS758), voltage sensor (LV25-P), temperature sensor (DS18B20), humidity sensor (DHT22) and CAN bus interface, including real-time current value (such as 80A), voltage value (such as 400V), temperature value (such as 45°C), charging power (32kW), ambient humidity (60%RH) and communication signal integrity parameters (packet loss rate 0.5%, delay 20ms, handshake success rate 98%). The acquisition frequency is once per second, and the data is transmitted to the data preprocessing unit through the GPIO and I2C interface of the STM32F4 microcontroller. After receiving the raw data, the data preprocessing unit immediately performs preprocessing operations. First, high-frequency noise is removed by dual-tree complex wavelet transform (the number of decomposition layers is adaptively adjusted to 4 layers according to signal fluctuations), such as filtering out instantaneous interference peaks in the current; then, an algorithm based on density clustering (neighborhood radius 0.5, minimum number of samples 5) is used to remove outliers, such as the temperature suddenly changing to 100°C at a certain moment is removed; finally, the data is standardized to the [0,1] interval using a generative adversarial network to form a multidimensional feature data set (such as a matrix of 6 parameters within 10 seconds). The preprocessing results are transmitted to the dynamic threshold generation module every minute through the internal data bus.

[0134] The dynamic threshold generation module builds the fault threshold based on the preprocessed data set. The long short-term memory network predicts the normal operating range (such as current 50A-100A) based on the historical data of the past 30 days. The adaptive Kalman filter algorithm combines real-time data (such as current load 70%) to update the threshold every minute, for example, adjusting the current upper limit to 105A. If the ambient temperature rises (such as from 25°C to 35°C), the enhancement function extracts spatiotemporal features through the bidirectional long short-term memory network and the graph attention network, optimizes nonlinear estimation through particle filtering, and adjusts the threshold to a more precise range (such as 103A) through reinforcement learning. The threshold is transmitted to the fault detection module through shared memory.

[0135] The fault detection module receives thresholds and datasets and performs multi-level detection. At the first level, local detection uses a support vector machine classifier to determine anomalies. For example, when the current exceeds 103A, it is marked as a potential fault (probability 60%); if an anomaly is detected, the data is uploaded to the cloud through a 4G module (SIM7600). At the second level, a deep belief network and a variational autoencoder are used to extract feature vectors (such as abnormal temperature patterns); at the third level, the dynamic Bayesian network and the theory of evidence are used in the cloud to fuse probabilities and confirm the fault type (such as "overload probability 85%"). The detection results are transmitted to the fault prediction module every 10 seconds through the TCP / IP protocol. The fault prediction module analyzes the detection results on the cloud server. The multi-dimensional Markov chain model and Gaussian process regression predict the fault trend. For example, if the temperature continues to rise, it may overheat in 6 hours; Monte Carlo simulation (10,000 iterations) generates time window probabilities (such as 4 - 8 hours with 95% confidence); the severity assessment is calculated by the following formula:

[0136] S = α·softmax(P f ) + β·exp(-T p / σ) + γ·∫I d (t)·w(t)dt

[0137] The prediction results are transmitted to the information processing and pushing unit through the internal bus of the server every hour. The information processing and pushing unit receives the prediction results, generates a fault monitoring report (such as "overload probability 85%, expected to occur within 6 hours, it is recommended to check the heat dissipation"), converts it into Chinese and English versions through a natural language generation model, and embeds a trend graph. The Q-learning algorithm optimizes the pushing time according to user habits (such as being active at 8 pm) and sends a warning to the user in the form of a text message or APP through the 4G module (such as "Please suspend charging and contact the maintenance"). The report is encrypted (SHA-256 hash) and uploaded to the cloud management system every 5 seconds. The cloud management system receives the report, constructs a knowledge graph to analyze the fault causality (such as overload is related to high temperature), updates the model parameters from multiple charging piles through federated learning (only sharing gradients), and an adaptive evolutionary algorithm optimizes the threshold strategy. The optimization results (such as the new threshold range of 55A - 110A) are transmitted back to the data collection module through the 4G module every 24 hours to update the local system and achieve collaborative monitoring of the charging pile group.

[0138] The above specific implementation manners do not constitute a limitation to the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for monitoring faults of a new energy vehicle charging pile, characterized in that: The following steps are involved: S1. Collect multi-source data during the operation of the charging pile, the multi-source data including the real-time current value, voltage value, temperature value, charging power, ambient humidity and communication signal integrity parameters between the charging pile and the new energy vehicle; S2. Preprocessing the multi-source data to generate a multi-dimensional feature data set; S3. Construct a dynamic fault threshold model, which is implemented by the following sub-steps: S3a. Based on historical operation data and current environmental parameters, the long short-term memory network is used to predict the normal operating range of the charging pile under different load conditions; S3b. Combined with the normal operating range, an adaptive Kalman filter algorithm is used to adjust the fault threshold in real time to adapt to the dynamic changes in the operating state of the charging pile; S4. Based on the multidimensional feature data set and the dynamic fault threshold model, perform multi-level fault detection, the multi-level fault detection comprising: S4a. First level detection: preliminary abnormal classification of the multidimensional feature data set by a support vector machine classifier to determine whether there is a potential fault; S4b. Second level detection: For the preliminary abnormal classification results, use the deep belief network to deeply mine the abnormal features and extract the fault mode feature vector; S4c. The third level detection: the fault mode feature vector is input into the pre-trained Bayesian network, the probability distribution of various types of faults is calculated, and the specific fault type is determined; S5. Establish a fault prediction mechanism, the fault prediction mechanism comprising: S5a. Based on the fault mode feature vector extracted in step S4, the Markov chain model is combined to predict the time window of the fault occurrence; S5b. Generate fault severity assessment indicators based on the predicted time window and fault type; S6. Generate a fault monitoring report based on the fault type and fault severity assessment index, and upload the fault monitoring report to the cloud management system in real time through the communication module of the charging pile, and push fault warning information to the user terminal at the same time.

2. The new energy vehicle charging pile fault monitoring method according to claim 1, characterized in that: The communication signal integrity parameters collected in step S1 include a dynamic combination of the following sub-parameters: S1a. CAN bus data packet loss rate is calculated by analyzing the consistency of packet sequences within a sliding time window, and frequency domain feature extraction based on Fourier transform is introduced; S1b. Signal delay time, measured using a time synchronization algorithm based on phase difference, combined with real-time correction of the environmental electromagnetic interference intensity; S1c. Protocol handshake success rate, the state transition probability during the handshake process is evaluated through a multi-state hidden Markov model, and weighted feedback adjustment is performed on abnormal states.

3. The new energy vehicle charging pile fault monitoring method according to claim 1 is characterized in that: The preprocessing process in step S2 includes the following sub-steps: S2a. Apply denoising processing based on dual-tree complex wavelet transform to the collected multi-source data, where the number of wavelet decomposition layers is adaptively selected according to the non-stationary characteristics of the signal; S2b. The DBSCAN algorithm based on density clustering is used to remove outliers, and the local outlier factor of the time series is combined for secondary verification; S2c. Data standardization processing introduces a feature distribution reconstruction mechanism based on a generative adversarial network to eliminate the dimensional differences and distribution biases between different sensor data.

4. The new energy vehicle charging pile fault monitoring method according to claim 1, characterized in that: The dynamic fault threshold model in step S3 is further defined by the following enhancement sub-steps: S3c. Use a bidirectional long short-term memory network combined with a graph attention network to extract spatiotemporal features of historical operation data and real-time environmental parameters, where the graph structure is dynamically constructed by the dependency relationship between charging piles and environmental variables; S3d. Based on the adaptive Kalman filter in step S3b, a nonlinear state transfer equation is introduced, and the probability estimation and optimization of the uncertainty in the filtering process are performed by particle filtering; S3e. According to the operating conditions and historical fault frequencies of the charging pile, reinforcement learning is used to adjust the threshold update strategy to maximize the accuracy and recall rate of fault detection.

5. The new energy vehicle charging pile fault monitoring method according to claim 1, characterized in that: The multi-level fault detection in step S4 is further refined into the following sub-steps: S4d. The first level detection uses support vector machine combined with fuzzy C-means clustering to perform primary anomaly classification on multidimensional feature data sets; S4e. The second level detection uses a joint model of deep belief network and variational autoencoder to extract the potential distribution representation of abnormal features and enhance feature robustness through adversarial training; S4f. The third level of detection inputs the fault mode feature vector into a multi-time-step reasoning framework based on a dynamic Bayesian network, combines it with evidence theory to fuse multi-source fault probabilities, and outputs an accurate fault type distribution.

6. The new energy vehicle charging pile fault monitoring method according to claim 1, characterized in that: The fault prediction mechanism in step S5 is implemented by the following sub-steps: S5c. Based on the fault mode feature vector of step S4, a multidimensional Markov chain model is constructed, in which the state transfer matrix is ​​updated in real time through online learning, and Gaussian process regression is introduced to predict the fault evolution trend; S5d. Combined with the fault evolution trend, the probability density function of the fault occurrence time window is generated by Monte Carlo simulation, and the uncertainty of the prediction is evaluated by information entropy; S5e. The fault severity evaluation index is calculated through multi-objective optimization, and the formula is as follows: S=α·softmax(P f )+β·exp(-T p / σ)+γ·∫I d (t)·w(t)dt.

7. The new energy vehicle charging pile fault monitoring method according to claim 1, characterized in that: The generation of fault monitoring report and warning information in step S6 includes the following sub-steps: S6a. The fault monitoring report is automatically generated in multiple languages ​​based on the fault type, probability distribution and severity index through the natural language generation model, and a visual fault trend chart is embedded; S6b. Early warning information push adopts a personalized strategy based on user behavior analysis; S6c. The cloud management system uses blockchain technology to encrypt and store fault monitoring reports and perform distributed verification to ensure data integrity and traceability.

8. The new energy vehicle charging pile fault monitoring method according to claim 1, characterized in that: The method further comprises the following enhancement steps: S7a. In the cloud management system, a knowledge base of charging pile faults based on knowledge graph is constructed, and the causal relationship and propagation path between faults are mined through graph convolutional network; S7b. Using the federated learning framework, the dynamic fault threshold model and fault prediction mechanism are distributedly updated from multiple charging pile nodes while protecting data privacy; S7c. Introduce an adaptive evolutionary algorithm to dynamically optimize all model parameters based on long-term operation data, and realize collaborative fault monitoring of charging pile groups through multi-agent collaboration.

9. A new energy vehicle charging pile fault monitoring system, characterized in that: The system is used to implement the new energy vehicle charging pile fault monitoring method described in any one of claims 1 to 8, and the system includes a data acquisition module, a data preprocessing unit, a dynamic threshold generation module, a fault detection module, a fault prediction module, an information processing and push unit, and a cloud management system, wherein the data acquisition module is connected to the data preprocessing unit, the data preprocessing unit is connected to the dynamic threshold generation module, the dynamic threshold generation module is connected to the fault detection module, the fault detection module is connected to the fault prediction module, the fault prediction module is connected to the information processing and push unit, the information processing and push unit is connected to the cloud management system, and the cloud management system is connected to the data acquisition module.

Images