New energy vehicle charging pile fault diagnosis method and system

Abstract

The invention provides a new energy automobile charging pile fault diagnosis method and system. The method comprises the steps of obtaining charging behavior data of a to-be-detected new energy automobile charging pile to construct a feature matrix; when the new energy automobile charging pile breaks down, environment sensing data are collected, and a multi-modal feature matrix is constructed in combination with the abnormal feature vectors and the thermal distribution features; and constructing a fault classification model according to the dynamic fusion features, obtaining a predicted fault classification result of the new energy vehicle charging pile, recording the predicted fault classification result and a risk level for each risk assessment result, synchronizing a real-time response action to an equipment log, and dynamically updating the weight of the fault classification model and a risk assessment threshold for optimization. According to the method, deep fusion of dynamic behaviors and environment perception is realized, higher diagnosis precision, higher adaptability and more intelligent response capability are realized, and the method can be widely applied to the field of operation maintenance and safety guarantee of the new energy automobile charging pile.

Description

Technical Field

[0001] The present invention belongs to the field of new energy vehicle charging piles, and in particular relates to a new energy vehicle charging pile fault diagnosis method and system. Background Art

[0002] The rapid popularization of new energy vehicles has promoted the rapid construction of charging infrastructure. As core facilities, the stability of operation and fault diagnosis capabilities of charging piles directly affect the user's charging experience and equipment safety. However, due to the complex working environment of charging piles and the variety of fault causes, traditional fault diagnosis methods have exposed many problems in practical applications.

[0003] Existing fault diagnosis methods for new energy vehicle charging piles are mainly based on two types of technologies: one is based on electrical parameter analysis, which determines faults by monitoring changes in physical quantities such as current and voltage. However, this method can only identify obvious electrical anomalies, and is not sensitive enough to small and gradual fault characteristics, especially in the dynamic charging process. It cannot provide timely and accurate diagnosis; the other method is based on diagnosis based on fixed threshold rules, which triggers fault alarms by setting fixed operating parameter thresholds (such as temperature, voltage, and current range). The main drawback of this method is that it has poor adaptability to changing actual scenarios, especially the significant differences in thresholds between different models and different charging modes, which is prone to false alarms and missed alarms. In addition, some studies have attempted to introduce artificial intelligence-based diagnostic technologies, such as analyzing device data through machine learning models, but these models usually rely on a single data source (such as electrical signals), ignoring the influence of the external environment (such as temperature and humidity, load fluctuations) and the internal state of the device (such as local temperature rise), and the diagnostic accuracy and reliability are limited.

[0004] The working environment of new energy vehicle charging piles further increases the difficulty of fault diagnosis. Charging piles are placed outdoors all year round and have to face complex situations such as temperature and humidity changes, power grid fluctuations, and high-frequency use. The impact of these environmental factors on equipment operation is difficult to quantify and difficult to evaluate in real time. At the same time, the battery characteristics of different models during the charging process (such as charging curves, battery health status) have different requirements for equipment operation, making it difficult for a unified diagnostic model to fully adapt. In addition, charging pile failures are often hidden, and the precursors to failures are difficult to detect, resulting in problems that affect charging efficiency and even damage equipment when they occur.

[0005] In summary, the current fault diagnosis technology for new energy vehicle charging piles has the following main problems:

[0006] Insufficient dynamic adaptability: Existing diagnostic technologies are difficult to capture abnormal behaviors of different vehicles and charging modes in real time, and the model's generalization ability is weak.

[0007] Environmental impacts have not been fully considered: The failure-inducing mechanism of charging piles caused by environmental factors has not been fully studied, and there is a lack of joint analysis of the external environment (such as high temperature and humidity) and the internal state of the equipment (such as hot spots).

[0008] The response mechanism is not smart enough: Traditional diagnostic methods can only trigger simple alarms and lack hierarchical processing and intelligent adjustment capabilities, making it difficult to ensure the charging safety of equipment and users. Summary of the invention

[0009] The purpose of the present invention is to propose a new energy vehicle charging pile fault diagnosis method and system, which can accurately identify faults in complex and changeable scenarios and adopt intelligent response strategies through a new energy vehicle charging pile fault diagnosis method involving the integration of dynamic charging behavior modeling and environmental perception.

[0010] In order to achieve the above object, a first aspect of the present invention provides a new energy vehicle charging pile fault diagnosis method, the method comprising the following steps:

[0011] The charging behavior data of the new energy vehicle charging pile to be detected is obtained to construct a feature matrix, and the historical charging behavior data is clustered in time series to generate a standard charging behavior model library. The real-time charging behavior data is input into the standard charging behavior model library for dynamic time regularization matching, and the matching error of each time step is calculated. The abnormal detection threshold is dynamically adjusted based on the matching error and an abnormal feature vector is constructed; wherein the standard charging behavior model library is optimized by a dynamically weighted objective function, which is expressed as:

[0012]

[0013] Among them, w t is a dynamic weight that gives higher weight to time steps where the power changes significantly, and β is the power change weight, Δp t is the power change; is the i-th charging mode m i The smooth regularization term of limits the mutation of the standard charging behavior model, and is the standard charging behavior model; γ controls the strength of the regularization term, balances the fitting error and model smoothness, and T represents the total time step;

[0014] When a new energy vehicle charging pile fails, the environmental perception data is collected and combined with the abnormal feature vector and thermal distribution features to construct a multimodal feature matrix, and different fusion weights are assigned to the features of different modes in the multimodal feature matrix. A time convolution network is introduced to perform time series processing on the weighted fusion feature matrix to obtain dynamic fusion features.

[0015] A fault classification model is constructed according to dynamic fusion features, and the dynamic fusion features are subjected to predictive fault classification to obtain the predicted fault classification results of the new energy vehicle charging piles. The risk level of the fault is dynamically evaluated according to the predicted fault output results combined with the risk assessment threshold. For each risk assessment result, the predicted fault classification results and risk level are recorded, and the real-time response actions are synchronized to the device log. At the same time, the device log is periodically uploaded to the cloud. Through statistical analysis of historical data, the fault classification model weights and risk assessment thresholds are dynamically updated for optimization; wherein, the risk levels of the faults are divided into low risk, medium risk and high risk; wherein, before constructing the fault classification model according to the dynamic fusion features, the dynamic fusion features are subjected to time smoothing processing based on regularization adjustment to remove deviations caused by acquisition noise or local anomalies;

[0016] The fault classification model adopts a deep learning architecture and introduces a dedicated weighted regularization term to optimize the contribution of features to fault categories. The objective function Defined as:

[0017]

[0018] in, is the dynamic fusion feature after time smoothing; y t,k represents the true fault label at time step t; Represents the category probability predicted by the model; W t A dynamic weight matrix representing features to optimize the adaptability of features to fault categories; is the regularization parameter, which controls the overfitting risk of feature weights; by optimizing Finally, the fault category c of each time step t is output t , K is the total number of nodes k.

[0019] Furthermore, the charging behavior data includes voltage, current and power; the characteristic matrix is ​​represented by X t =[v t ,c t ,p t ,Δv t ,Δc t ,Δp t ], where Δv t =v t -v t-1 Indicates the rate of change of voltage; Δc t =c t -c t-1 Indicates the rate of change of current; Δp t =p t -p t-1 Indicates the rate of change of power; v tIndicates the current voltage; v t-1 Indicates the voltage of the previous step; c t Indicates the current; c t-1 Indicates the current of the previous step; p t Indicates the current power; p t-1 Indicates the power of the previous step;

[0020] And add specific regularization items to the feature matrix to enhance data stability and handle occasional outliers, expressed as:

[0021]

[0022] Among them, λ is the regularization strength control parameter; R t It is a random perturbation matrix used to simulate the influence of external interference during the charging process.

[0023] Furthermore, the real-time charging behavior data is input into the standard charging behavior model library for dynamic time warping matching, and the matching error of each time step is calculated, and the abnormal detection threshold is dynamically adjusted based on the matching error and the abnormal feature vector is constructed, including:

[0024] Real-time dataX real After input, a sliding window mechanism is used, where the window length k is related to the model library Perform dynamic time warping matching on each model in and calculate the matching error at each time step t:

[0025]

[0026] Among them, X real,t+j represents the real-time characteristics of time step t+j; m i (j) represents the j-th step feature of the i-th type model; δ t is the matching error at time step t, which is used to measure the degree of deviation between the current data and the standard model, that is, the abnormal characteristics;

[0027] Based on abnormal feature δ t The mean μ and standard deviation σ of the anomaly detection threshold θ are adjusted dynamically t :

[0028] θ t =μ+α·σ

[0029] Among them, α is the adjustment coefficient, which is used to control the sensitivity of anomaly detection;

[0030] If δ t >θ t , mark time step t as abnormal, and record the abnormal time set {t|δ t >θ t}.

[0031] Furthermore, the environmental perception data is in Indicates temperature T t 、Humidity t and grid load fluctuation L t The thermal distribution characteristic is G = {g t |t=1,2,…,T}, where N is the internal N of the device at time step t h The temperature data of sensor nodes; the abnormal feature vector is = {δ t |t=1,2,…,T}, describing the abnormal degree of charging behavior at each time step t; the multimodal feature matrix is Among them, F t =[δ t ,g t ,T t ,H t ,L t ], where δ t is the abnormal feature at time step t; g t is the thermal distribution characteristic; T t ,H t ,L t They are temperature, humidity and grid load fluctuations.

[0032] Furthermore, a spatial regularization term is introduced into the thermal distribution feature The optimized thermal distribution matrix G is expressed as:

[0033]

[0034] Among them, a ij is the spatial adjacency relationship between nodes i and j, calculated based on the node distance; g t,i is the temperature value of node i, N h is the number of sensors.

[0035] Furthermore, the fusion weight is calculated by the modal attention mechanism:

[0036]

[0037] Among them, w mod,t is the weight distribution of different modes at time step t; ∥δ t ∥ indicates the importance of the abnormality to the mode; β is a regulation parameter that controls the dynamics of the modal weight distribution;

[0038] Perform weighted fusion on the feature matrix F according to the modal weights to generate the fused feature matrix F fusion :

[0039]

[0040] where F t,mod is the modal characteristic at time step t, w mod,t is the modal weight, mod is the corresponding mode;

[0041] The method of introducing a time convolutional network to perform time series processing on the weighted fusion feature matrix to obtain dynamic fusion features includes:

[0042] The convolution kernel size of the temporal convolutional network is k, the convolution step size is s, and the output sequence dimension is consistent with the input; the time series feature F seq By formula:

[0043] F seq =TCN(F fusion )

[0044] in, It represents the dynamic fusion features of time step t, and TCN is the temporal convolutional network.

[0045] Furthermore, the time smoothing process based on regularization adjustment is expressed as:

[0046]

[0047] in, represents the smoothed feature; λ is the regularization coefficient, which balances the consistency between the current feature and the features of the adjacent time steps; w is the sliding window length, which controls the smoothing range; Indicates that in the dynamic fusion feature F t Find the set, j represents node j.

[0048] Furthermore, the risk level R of the fault is dynamically evaluated based on the predicted fault output result combined with the risk assessment threshold. t , expressed as:

[0049]

[0050] in, is the corresponding prediction confidence, It represents the risk assessment function of the feature, which integrates the significant contribution of multimodal features; α and β are weight parameters, which measure the relative importance of classification confidence and feature risk respectively.

[0051] Furthermore, the low risk: R t >τ low , continue charging and monitor the status; the medium risk: τ mid ≤R t ≤τ low , limit charging power and prompt the user; the high risk: Rt <τ mid , immediately stop charging, trigger power-off protection, and generate maintenance suggestions; where τ low is the low risk threshold, τ mid The medium risk threshold.

[0052] In a second aspect of the present invention, a new energy vehicle charging pile fault diagnosis system is provided, the system comprising:

[0053] The new energy vehicle charging pile data acquisition unit is used to acquire the charging behavior data of the new energy vehicle charging pile to be detected to construct a feature matrix, and to perform time series clustering on the historical charging behavior data to generate a standard charging behavior model library, input the real-time charging behavior data into the standard charging behavior model library for dynamic time regularization matching, and calculate the matching error of each time step, dynamically adjust the anomaly detection threshold based on the matching error and construct an anomaly feature vector; wherein the standard charging behavior model library is optimized by a dynamically weighted objective function, which is expressed as:

[0054]

[0055] Among them, w t is a dynamic weight that gives higher weight to time steps where the power changes significantly, and β is the power change weight, Δp t is the power change; is the standard charging behavior model m i The smooth regularization term of limits the mutation of the standard charging behavior model, and is the standard charging behavior model; γ controls the strength of the regularization term, balancing the fitting error and model smoothness;

[0056] A new energy vehicle charging pile fault analysis unit is used to collect environmental perception data and combine it with abnormal feature vectors and thermal distribution features to construct a multi-modal feature matrix when a new energy vehicle charging pile fails, and assign different fusion weights to features of different modes in the multi-modal feature matrix, and introduce a time convolution network to perform time series processing on the weighted fusion fusion feature matrix to obtain dynamic fusion features;

[0057] A new energy vehicle charging pile fault diagnosis unit is used to construct a fault classification model based on dynamic fusion features, predict fault classification for the dynamic fusion features, obtain the predicted fault classification results of the new energy vehicle charging pile, dynamically evaluate the risk level of the fault based on the predicted fault output results combined with the risk assessment threshold, record the predicted fault classification results and risk level for each risk assessment result, and synchronize the real-time response action to the device log, and periodically upload the device log to the cloud, dynamically update the fault classification model weights and risk assessment thresholds for optimization through statistical analysis of historical data; wherein the risk level of the fault is divided into low risk, medium risk and high risk; wherein, before constructing the fault classification model based on the dynamic fusion features, the dynamic fusion features are subjected to time smoothing based on regularization adjustment to remove deviations caused by acquisition noise or local anomalies;

[0058] The fault classification model adopts a deep learning architecture and introduces a dedicated weighted regularization term to optimize the contribution of features to fault categories. The objective function Defined as:

[0059]

[0060] in, is the dynamic fusion feature after time smoothing; y t,k represents the true fault label at time step t; Represents the category probability predicted by the model; W t A dynamic weight matrix representing features to optimize the adaptability of features to fault categories; is the regularization parameter, which controls the overfitting risk of feature weights; by optimizing Finally, the fault category c of each time step t is output t .

[0061] The beneficial technical effects of the present invention are at least as follows:

[0062] In view of the shortcomings of the existing technology, the present invention can accurately identify faults in complex and changing scenarios and adopt intelligent response strategies. Its main innovations are as follows:

[0063] The present invention designs a charging behavior modeling algorithm based on time series clustering and constructs a normal charging mode library based on the dynamic behavior characteristics of different vehicle models and different charging modes. During the charging process, the abnormal changes can be accurately captured by comparing the charging curve with the standard behavior model in real time, solving the problem of insufficient adaptability of the existing technology to the charging behavior of different vehicles.

[0064] The environmental perception module is introduced to monitor external temperature, humidity, power grid fluctuations and other environmental parameters in real time, and combined with the thermal distribution data inside the equipment, the multi-dimensional feature joint analysis is used to accurately identify the risk of environmental-induced failures. This method effectively makes up for the lack of analysis of environmental factors in the existing technology.

[0065] According to the severity of the fault (such as slight fluctuation, moderate abnormality, severe fault), a graded response mechanism is designed: minor faults are mitigated by optimizing the charging curve, moderate faults prompt the user and limit power output, and severe faults trigger power-off protection and upload diagnostic data. This mechanism improves the security and user experience of the system and solves the problems of single response and low intelligence in traditional methods.

[0066] Through the above-mentioned technology, the present invention realizes the deep integration of dynamic behavior and environmental perception, has higher diagnostic accuracy, stronger adaptability and more intelligent response capability, and can be widely used in the operation, maintenance and safety assurance of new energy vehicle charging piles. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0068] Figure 1 This is a flow chart of a new energy vehicle charging pile fault diagnosis method of the present invention. DETAILED DESCRIPTION

[0069] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0070] like Figure 1 As shown, an embodiment of the present invention provides a new energy vehicle charging pile fault diagnosis method, the method comprising:

[0071] 101. Obtain the charging behavior data of the new energy vehicle charging pile to be tested to construct a feature matrix, and perform time series clustering on the historical charging behavior data to generate a standard charging behavior model library. Input the real-time charging behavior data into the standard charging behavior model library for dynamic time regularization matching, and calculate the matching error of each time step. Based on the matching error, dynamically adjust the anomaly detection threshold and construct an anomaly feature vector.

[0072] Specifically, the input data of this step is charging behavior data D={d t|t=1,2,…,T}, where d t Including voltage v t (Unit: Volt), current c t (unit: ampere) and power p t =v t ·c t (Unit: Watt).

[0073] Further, construct the feature matrix X∈R T×6 , each row X t The feature vector representing time step t:

[0074] X t =[v t ,c t ,p t ,Δv t ,Δc t ,Δp t ]

[0075] Where Δv t =v t -v t-1 Indicates the rate of change of voltage; Δc t =c t -c t-1 Indicates the rate of change of current; Δp t =p t -p t-1 Indicates the rate of change of power.

[0076] Additional specific regularization terms are added to enhance data stability and handle occasional outliers:

[0077]

[0078] Among them, λ is the regularization strength control parameter; R t It is a random perturbation matrix used to simulate the impact of external interference (such as grid fluctuations) during the charging process.

[0079] Furthermore, for the historical data X hist Perform time series clustering to generate a standard charging behavior model library Where m i Indicates the i-th charging mode (such as fast charging, slow charging).

[0080] The optimization model uses a dynamically weighted objective function:

[0081]

[0082] in, is a dynamic weight that gives higher weight to time steps with significant power changes; is the smooth regularization term of the model, which limits the mutation of the model; γ controls the strength of the regularization term, balancing the fitting error and model smoothness.

[0083] Optimized model library It can adapt to the charging behaviors of different models and provide a basis for subsequent real-time matching.

[0084] Furthermore, real-time data X real After input, a sliding window mechanism (window length k) is used to compare with the model library Perform dynamic time warping (DTW) matching on each model in and calculate the matching error at each time step t:

[0085]

[0086] Among them, X real,t+j represents the real-time characteristics of time step t+j; m i (j) represents the j-th step feature of the i-th type model; δ t is the matching error at time step t, which is used to measure the degree of deviation between the current data and the standard model.

[0087] Furthermore, based on the historical matching error δ t The mean μ and standard deviation σ of the anomaly detection threshold θ are adjusted dynamically t :

[0088] θ t =μ+α·σ

[0089] Among them, α is the adjustment coefficient, which is used to control the sensitivity of anomaly detection. t >θ t , mark time step t as abnormal, and record the abnormal time set {t|δ t >θ t}.

[0090] Finally, the abnormal feature vector E = {δ t |t=1,2,…,T}, used to describe the degree of abnormality; the abnormal time set {t|δ t >θ t}, providing basic data for subsequent multimodal feature fusion.

[0091] It can be understood that this step starts from the modeling and anomaly detection of dynamic charging behavior, and uses regularization and dynamic weight improvement to achieve adaptive processing of complex data fluctuations. The output abnormal feature vector and abnormal moment set are the key inputs for subsequent environmental perception and multimodal fusion of the patent, and also reflect the effective adaptation to multi-model and multi-mode problems in charging scenarios.

[0092] 102. When a new energy vehicle charging pile fails, environmental perception data is collected and combined with abnormal feature vectors and thermal distribution features to construct a multimodal feature matrix. Different fusion weights are assigned to features of different modes in the multimodal feature matrix, and a time convolution network is introduced to perform time series processing on the weighted fusion feature matrix to obtain dynamic fusion features.

[0093] Specifically, multimodal feature construction:

[0094] The abnormal feature vector E={δ t |t=1,2,…,T}, describing the abnormality of charging behavior at each time step t; environmental perception data in Indicates temperature T t 、Humidity t and grid load fluctuation L t ; Thermal distribution characteristics G = {g t |t=1,2,…,T}, where N is the internal N of the device at time step t h The temperature data of each sensor node.

[0095] Furthermore, all input data are combined to construct a multimodal feature matrix

[0096] F t =[δ t ,g t ,T t ,H t ,L t ]

[0097] Among them, δ t is the abnormal feature at time step t; g t is the thermal distribution characteristic; T t ,H t ,L t They are temperature, humidity and grid load fluctuations.

[0098] Furthermore, in order to improve the contribution of thermal characteristics to fault location, a spatial regularization term is introduced to optimize the thermal distribution matrix G:

[0099]

[0100] Among them, a ij is the spatial adjacency relationship between nodes i and j, calculated based on the node distance; g t,i is the temperature value of node i.

[0101] This regularization term makes the spatial variation of thermal distribution smoother and reduces the error caused by sensor noise.

[0102] Furthermore, different fusion weights are assigned to features of different modalities (abnormal features, thermal features, and environmental data) to capture the impact of key modalities. The fusion weights are calculated by the modality attention mechanism:

[0103]

[0104] Among them, w mod,t is the weight distribution of different modes at time step t; ∥δ t ∥ indicates the importance of the abnormality degree to the mode;

[0105] β is a tuning parameter that controls the dynamics of the modal weight distribution.

[0106] Furthermore, the feature matrix F is weighted and fused according to the modal weights to generate a fused feature matrix F fusion :

[0107]

[0108] where F t,mod is the modal feature at time step t.

[0109] Furthermore, in order to capture the temporal dependency of multimodal features, a temporal convolutional network (TCN) is introduced to perform time series processing on the fused features:

[0110] The convolution kernel size is k, the convolution step size is s, and the output sequence dimension is consistent with the input.

[0111] Time series feature F seq By formula:

[0112] F seq =TCN(F fusion )

[0113] Represents the dynamic fusion features at time step t.

[0114] Finally, the dynamic fusion feature is output Provide high-dimensional feature input for the next step of fault classification.

[0115] This step extracts the multimodal features of the fault through joint modeling of environmental perception data, thermal distribution, and abnormal charging behavior. The introduced modal attention mechanism and spatial regularization enhance the contribution of key modalities to fault features, while using temporal dynamic modeling to capture feature changes, providing complete data support for the next step of fault classification.

[0116] 103. Construct a fault classification model based on dynamic fusion features, perform predictive fault classification on the dynamic fusion features, and obtain the predicted fault classification results of new energy vehicle charging piles. Dynamically evaluate the risk level of the fault based on the predicted fault output results combined with the risk assessment threshold. For each risk assessment result, record the predicted fault classification results and risk level, synchronize the real-time response actions to the device log, and periodically upload the device log to the cloud. Through statistical analysis of historical data, dynamically update the fault classification model weights and risk assessment thresholds for optimization.

[0117] Specifically, considering that there may be deviations in the features due to acquisition noise or local anomalies, a time smoothing mechanism based on regularization adjustment is introduced:

[0118]

[0119] in, represents the smoothed feature; λ is the regularization coefficient, which balances the consistency between the current feature and the features of the adjacent time steps; w is the sliding window length, which controls the smoothing range.

[0120] Furthermore, the fault classification model is constructed:

[0121] Using the smoothed feature F sm , the time step t is mapped to the fault category C = {c1,c2,…,c k}, represents k possible fault types. The model adopts a deep learning architecture and introduces a dedicated weighted regularization term to optimize the contribution of features to fault categories. The objective function is defined as:

[0122]

[0123] Among them, y t,k represents the true fault label at time step t; Represents the category probability predicted by the model; W t A dynamic weight matrix representing features to optimize the adaptability of features to fault categories; is a regularization parameter that controls the overfitting risk of feature weights. Finally, the fault category c of each time step t is output t .

[0124] Furthermore, according to the fault output c of the classification model t and the corresponding prediction confidence Dynamically evaluate the risk level R of the fault t :

[0125]

[0126] in, represents the risk assessment function of the feature, integrating the significant contribution of multimodal features; α and β are weight parameters, which measure the relative importance of classification confidence and feature risk respectively. t The value divides the failure risk into three categories:

[0127] Low risk: R t >τ low , continue charging and monitor the status;

[0128] Medium risk: τ mid ≤R t ≤τ low , limit the charging power and prompt the user;

[0129] High risk: R t <τ mid , immediately stop charging, trigger power-off protection, and generate maintenance recommendations.

[0130] Furthermore, for each risk assessment result, the classification result C and risk level R are recorded, and the real-time response action (such as power limitation or power-off protection) is synchronized to the device log.

[0131] Periodically upload device logs to the cloud, and dynamically update the model weight W through statistical analysis of historical data t and the risk assessment threshold τ low ,τ mid , achieving continuous optimization.

[0132] Finally, the fault category sequence C is outputted as t |t=1,2,…,T}, used to describe the fault type during charging; risk level sequence R={R t |t=1,2,…,T}, reflecting the real-time risk of failure; triggering and logging of real-time response actions provide data support for subsequent optimization.

[0133] This step dynamically fuses the features F seq Combined with the fault classification model, it outputs accurate fault types and risk assessment results, providing a reliable basis for the intelligent response of charging piles. At the same time, the optimization and feedback mechanism of response actions ensure the continuous adaptability and efficiency of the system in practical applications.

[0134] The embodiment of the present invention further provides a new energy vehicle charging pile fault diagnosis system, the system comprising:

[0135] The new energy vehicle charging pile data acquisition unit 1011 is used to acquire the charging behavior data of the new energy vehicle charging pile to be detected to construct a feature matrix, and to perform time series clustering on the historical charging behavior data to generate a standard charging behavior model library, input the real-time charging behavior data into the standard charging behavior model library for dynamic time regularization matching, and calculate the matching error of each time step, dynamically adjust the anomaly detection threshold based on the matching error and construct an anomaly feature vector; wherein the standard charging behavior model library is optimized by a dynamically weighted objective function, which is expressed as:

[0136]

[0137] Among them, w t is a dynamic weight that gives higher weight to time steps where the power changes significantly, and β is the power change weight, Δp t is the power change; is the standard charging behavior model m i The smooth regularization term of limits the mutation of the standard charging behavior model, and is the standard charging behavior model; γ controls the strength of the regularization term, balancing the fitting error and model smoothness;

[0138] The new energy vehicle charging pile fault analysis unit 1012 is used to collect environmental perception data and combine abnormal feature vectors and thermal distribution features to construct a multi-modal feature matrix when a new energy vehicle charging pile fails, and assign different fusion weights to features of different modes in the multi-modal feature matrix, and introduce a time convolution network to perform time series processing on the weighted fusion fusion feature matrix to obtain dynamic fusion features;

[0139] The new energy vehicle charging pile fault diagnosis unit 1013 is used to build a fault classification model according to the dynamic fusion features, predict fault classification for the dynamic fusion features, obtain the predicted fault classification results of the new energy vehicle charging pile, dynamically evaluate the risk level of the fault according to the predicted fault output results combined with the risk assessment threshold, record the predicted fault classification results and risk level for each risk assessment result, and synchronize the real-time response action to the device log, and periodically upload the device log to the cloud, and dynamically update the fault classification model weight and risk assessment threshold for optimization through statistical analysis of historical data; wherein the risk level of the fault is divided into low risk, medium risk and high risk; wherein, before building the fault classification model according to the dynamic fusion features, the dynamic fusion features are subjected to time smoothing based on regularization adjustment to remove deviations caused by acquisition noise or local anomalies;

[0140] The fault classification model adopts a deep learning architecture and introduces a dedicated weighted regularization term to optimize the contribution of features to fault categories. The objective function Defined as:

[0141]

[0142] in, is the dynamic fusion feature after time smoothing; y t,k represents the true fault label at time step t; Represents the category probability predicted by the model; W t A dynamic weight matrix representing features to optimize the adaptability of features to fault categories; is the regularization parameter, which controls the overfitting risk of feature weights; by optimizing Finally, the fault category c of each time step t is output t .

[0143] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0144] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a division of logical functions. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be electrical, mechanical or other forms.

[0145] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0146] Although embodiments of the present invention have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A new energy vehicle charging pile fault diagnosis method, characterized in that: The method comprises the following steps: The charging behavior data of the new energy vehicle charging pile to be detected is obtained to construct a feature matrix, and the historical charging behavior data is clustered in time series to generate a standard charging behavior model library. The real-time charging behavior data is input into the standard charging behavior model library for dynamic time regularization matching, and the matching error of each time step is calculated. The abnormal detection threshold is dynamically adjusted based on the matching error and an abnormal feature vector is constructed; wherein the standard charging behavior model library is optimized by a dynamically weighted objective function, which is expressed as: Among them, w t is a dynamic weight that gives higher weight to time steps where the power changes significantly, and β is the power change weight, Δp t is the power change; is the i-th charging mode m i The smooth regularization term of limits the mutation of the standard charging behavior model, and is the standard charging behavior model; γ controls the strength of the regularization term, balances the fitting error and model smoothness, and T represents the total time step; When a new energy vehicle charging pile fails, the environmental perception data is collected and combined with the abnormal feature vector and thermal distribution features to construct a multimodal feature matrix, and different fusion weights are assigned to the features of different modes in the multimodal feature matrix. A time convolution network is introduced to perform time series processing on the weighted fusion feature matrix to obtain dynamic fusion features. A fault classification model is constructed according to dynamic fusion features, and the dynamic fusion features are used to predict fault classification to obtain the predicted fault classification results of new energy vehicle charging piles. The risk level of the fault is dynamically evaluated according to the predicted fault output results combined with the risk assessment threshold. For each risk assessment result, the predicted fault classification results and risk level are recorded, and the real-time response actions are synchronized to the device log. At the same time, the device log is periodically uploaded to the cloud. Through statistical analysis of historical data, the fault classification model weights and risk assessment thresholds are dynamically updated for optimization; wherein, the risk levels of the faults are divided into low risk, medium risk and high risk; wherein, before constructing the fault classification model according to the dynamic fusion features, the dynamic fusion features are subjected to time smoothing processing based on regularization adjustment to remove deviations caused by acquisition noise or local anomalies; The fault classification model adopts a deep learning architecture and introduces a dedicated weighted regularization term to optimize the contribution of features to fault categories. The objective function Defined as: in, is the dynamic fusion feature after time smoothing; y t,k Indicates the real fault label; Represents the category probability predicted by the model; W t A dynamic weight matrix representing features to optimize the adaptability of features to fault categories; is the regularization parameter, which controls the overfitting risk of feature weights; by optimizing Finally, the fault category c of each time step t is output t , K is the total number of nodes k.

2. A new energy vehicle charging pile fault diagnosis method according to claim 1, characterized in that: The charging behavior data includes voltage, current and power; the characteristic matrix is ​​represented by X t =[v t ,c t ,p t ,Δv t ,Δc t ,Δp t ], where Δv t =v t -v t-1 Indicates the rate of change of voltage; Δc t =c t -c t-1 Indicates the rate of change of current; Δp t =p t -p t-1 Indicates the rate of change of power; v t Indicates the current voltage; v t-1 Indicates the voltage of the previous step; c t Indicates the current; c t-1 Indicates the current of the previous step; p t Indicates the current power; p t-1 Indicates the power of the previous step; And add specific regularization items to the feature matrix to enhance data stability and handle occasional outliers, expressed as: Among them, λ is the regularization strength control parameter; R t It is a random perturbation matrix used to simulate the influence of external interference during the charging process.

3. A new energy vehicle charging pile fault diagnosis method according to claim 1, characterized in that: The real-time charging behavior data is input into the standard charging behavior model library for dynamic time warping matching, and the matching error of each time step is calculated. Based on the matching error, the abnormal detection threshold is dynamically adjusted and the abnormal feature vector is constructed, including: Real-time dataX real After input, a sliding window mechanism is used, where the window length k is related to the model library Perform dynamic time warping matching on each model in and calculate the matching error at each time step t: Among them, X real,t+j represents the real-time characteristics of time step t+j; m i (j) represents the j-th step feature of the i-th type model; δ t is the matching error at time step t, which is used to measure the degree of deviation between the current data and the standard model, that is, the abnormal characteristics; Based on abnormal feature δ t The mean μ and standard deviation σ of the anomaly detection threshold θ are adjusted dynamically t : i t =μ+a·s Among them, α is the adjustment coefficient, which is used to control the sensitivity of anomaly detection; If δ t >θ t , mark time step t as abnormal, and record the abnormal time set {t|δ t >θ t }.

4. A new energy vehicle charging pile fault diagnosis method according to claim 3, characterized in that: The environmental perception data is in Indicates temperature T t 、Humidity t and grid load fluctuation L t The thermal distribution characteristic is G = {g t |t=1,2,…,T}, where N is the internal N of the device at time step t h The temperature data of sensor nodes; the abnormal feature vector is E = {δ t |t=1,2,…,T}, describing the abnormal degree of charging behavior at each time step t; the multimodal feature matrix is Among them, F t =[δ t ,g t ,T t ,H t ,L t ], where δ t is the abnormal feature at time step t; g t is the thermal distribution characteristic; T t ,H t ,L t They are temperature, humidity and grid load fluctuations.

5. A new energy vehicle charging pile fault diagnosis method according to claim 4, characterized in that: Introducing a spatial regularization term into the thermal distribution feature The optimized thermal distribution matrix G is expressed as: Among them, a ij is the spatial adjacency relationship between nodes i and j, calculated based on the node distance; g t,i is the temperature value of node i, N h is the number of sensors.

6. A new energy vehicle charging pile fault diagnosis method according to claim 4, characterized in that: The fusion weights are calculated by the modal attention mechanism: Among them, w mod,t is the weight distribution of different modes at time step t; ∥δ t ∥ indicates the importance of the abnormality to the mode; β is a regulation parameter that controls the dynamics of the modal weight distribution; Perform weighted fusion on the feature matrix F according to the modal weights to generate the fused feature matrix F fusion : Among them, F t,mod is the modal characteristic at time step t, w mod,t is the modal weight, mod is the corresponding mode; The method of introducing a time convolutional network to perform time series processing on the weighted fusion feature matrix to obtain dynamic fusion features includes: The convolution kernel size of the temporal convolutional network is k, the convolution step size is s, and the output sequence dimension is consistent with the input; the time series feature F seq By formula: F seq =TCN(F fusion in, It represents the dynamic fusion features of time step t, and TCN is the temporal convolutional network.

7. A new energy vehicle charging pile fault diagnosis method according to claim 1, characterized in that: The time smoothing process based on regularization adjustment is expressed as: in, represents the smoothed feature; λ is the regularization coefficient, which balances the consistency between the current feature and the features of the adjacent time steps; w is the sliding window length, which controls the smoothing range; Indicates that in the dynamic fusion feature F t Find the set, j represents node j.

8. A new energy vehicle charging pile fault diagnosis method according to claim 1, characterized in that: The risk level R of the fault is dynamically evaluated according to the predicted fault classification result combined with the risk assessment threshold t , expressed as: in, is the corresponding prediction confidence, It represents the risk assessment function of the feature, which integrates the significant contribution of multimodal features; α and β are weight parameters, which measure the relative importance of classification confidence and feature risk respectively.

9. A new energy vehicle charging pile fault diagnosis method according to claim 8, characterized in that: Low risk: R t >τ low , continue charging and monitor the status; the medium risk: τ mid ≤R t ≤τ low , limit charging power and prompt the user; the high risk: R t <τ mid , immediately stop charging, trigger power-off protection, and generate maintenance suggestions; where τ low is the low risk threshold, τ mid The medium risk threshold.

10. A new energy vehicle charging pile fault diagnosis system, characterized in that: The system comprises: The new energy vehicle charging pile data acquisition unit is used to acquire the charging behavior data of the new energy vehicle charging pile to be detected to construct a feature matrix, and to perform time series clustering on the historical charging behavior data to generate a standard charging behavior model library, input the real-time charging behavior data into the standard charging behavior model library for dynamic time regularization matching, and calculate the matching error of each time step, dynamically adjust the anomaly detection threshold based on the matching error and construct an anomaly feature vector; wherein the standard charging behavior model library is optimized by a dynamically weighted objective function, which is expressed as: Among them, w t is a dynamic weight that gives higher weight to time steps where the power changes significantly, and β is the power change weight, Δp t is the power change; It is the i-th charging mode m i The smooth regularization term of limits the mutation of the standard charging behavior model, and is the standard charging behavior model; γ controls the strength of the regularization term, balances the fitting error and model smoothness, and T represents the total time step; A new energy vehicle charging pile fault analysis unit is used to collect environmental perception data and combine it with abnormal feature vectors and thermal distribution features to construct a multi-modal feature matrix when a new energy vehicle charging pile fails, and assign different fusion weights to features of different modes in the multi-modal feature matrix, and introduce a time convolution network to perform time series processing on the weighted fusion fusion feature matrix to obtain dynamic fusion features; A new energy vehicle charging pile fault diagnosis unit is used to construct a fault classification model based on dynamic fusion features, predict fault classification for the dynamic fusion features, obtain the predicted fault classification results of the new energy vehicle charging pile, dynamically evaluate the risk level of the fault based on the predicted fault output results combined with the risk assessment threshold, record the predicted fault classification results and risk level for each risk assessment result, and synchronize the real-time response action to the device log, and periodically upload the device log to the cloud, dynamically update the fault classification model weights and risk assessment thresholds for optimization through statistical analysis of historical data; wherein the risk level of the fault is divided into low risk, medium risk and high risk; wherein, before constructing the fault classification model based on the dynamic fusion features, the dynamic fusion features are subjected to time smoothing based on regularization adjustment to remove deviations caused by acquisition noise or local anomalies; The fault classification model adopts a deep learning architecture and introduces a dedicated weighted regularization term to optimize the contribution of features to fault categories. The objective function Defined as: in, is the dynamic fusion feature after time smoothing; y t,k Indicates the real fault label; Represents the category probability predicted by the model; W t A dynamic weight matrix representing features to optimize the adaptability of features to fault categories; is the regularization parameter, which controls the overfitting risk of feature weights; by optimizing Finally, the fault category c of each time step t is output t .

Images