Charging equipment fault diagnosis method and system combining deep learning and meta learning

Abstract

The invention discloses a deep learning and meta learning combined charging equipment fault diagnosis method and system, and relates to the technical field of electric vehicle charging facility operation and maintenance, and the method comprises the steps: obtaining first parameter data and second parameter data; forming a mapping relation based on the first parameter data and the second parameter data; constructing a fault feature data set based on the mapping relation; constructing a multi-target fault diagnosis model by using the fault feature data set, and optimizing the multi-target fault diagnosis model; and determining a fault diagnosis result of the charging equipment according to the optimized multi-target fault diagnosis model. According to the method, the Transform architecture of the RoBERTa model is utilized, and a self-attention mechanism is used for completing feature extraction, so that a longer-distance context dependency relationship can be captured, the convergence speed of the model is accelerated, and the accuracy of multi-target fault diagnosis of the charging equipment is improved; an MAML meta-learning method is adopted, and the training efficiency of a multi-target fault diagnosis model is improved by learning initialization parameters.

Description

Technical Field

[0001] The present invention relates to the field of electric vehicle charging facility operation and maintenance technology, and in particular to a charging equipment fault diagnosis method and system combining deep learning and meta-learning. Background Art

[0002] With the rapid development of the electric vehicle industry, charging facilities have become a core link in the new energy vehicle industry chain. Their reliability and safety are crucial to user experience and the sustainable development of the industry. However, the issue of charging safety has become increasingly prominent, becoming one of the key bottlenecks restricting the further development of the industry. Under the condition of long-term high-load operation of charging piles, their performance will inevitably decline gradually, and the equipment failure rate will increase significantly. In particular, the situation where multiple faults occur concurrently in a single charging pile not only increases the complexity of maintenance, but also poses great challenges to fault location and troubleshooting for technicians. These problems directly threaten the charging safety of users and pose a hidden danger to the stable operation of the industry. Summary of the Invention

[0003] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and in the abstract and title of the present invention to avoid obscuring the purpose of this section, the abstract and the title of the invention, and such simplifications or omissions should not be used to limit the scope of the present invention.

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the present invention provides a charging equipment fault diagnosis method and system that combines deep learning and meta-learning. By combining a deep neural network model with a meta-learning method, multi-target fault diagnosis of charging equipment is achieved, while enabling the model to achieve rapid learning on different fault diagnosis tasks.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a charging equipment fault diagnosis method that combines deep learning and meta-learning, which includes obtaining first parameter data and second parameter data; forming a mapping relationship based on the first parameter data and the second parameter data; constructing a fault feature data set based on the mapping relationship; using the fault feature data set to construct a multi-objective fault diagnosis model, and optimizing the multi-objective fault diagnosis model; and determining the fault diagnosis result of the charging equipment based on the optimized multi-objective fault diagnosis model.

[0008] As a preferred solution of the charging equipment fault diagnosis method combining deep learning and meta-learning described in the present invention, the obtaining of the first parameter data and the second parameter data includes the following steps: obtaining the first parameter data and the second parameter data through operation and maintenance work order data.

[0009] As a preferred solution of the charging equipment fault diagnosis method combining deep learning and meta-learning described in the present invention, wherein: a mapping relationship is formed based on the first parameter data and the second parameter data, including the following steps: obtaining abnormal status, fault cause and processing process information through operation and maintenance work order data; performing data cleaning on the collected operation and maintenance work order data, first parameter data and second parameter data; establishing a mapping relationship between the operation and maintenance work order data, first parameter data and second parameter data; through the constructed mapping relationship, mapping the charging pile with the abnormal status, fault cause and processing process information to form a comprehensive charging pile data set.

[0010] As a preferred solution of the charging equipment fault diagnosis method combining deep learning and meta-learning described in the present invention, wherein: a fault feature dataset is constructed based on the mapping relationship, including the following steps: analyzing the charging pile comprehensive dataset to extract original features; combining the original features into composite features through feature combination technology; combining time series analysis and statistical feature extraction methods to mine features from the time dimension; integrating the original features and composite features to obtain a fault feature dataset of the charging equipment.

[0011] As an optimal solution of the charging equipment fault diagnosis method combining deep learning and meta-learning described in the present invention, wherein: a multi-objective fault diagnosis model is constructed using the fault feature data set, including the following steps: converting the fault feature data set into charging equipment fault sequence feature data; converting the charging equipment fault sequence feature data into a fault feature vector using the RoBERTa model; providing the position information of the elements in the charging equipment fault sequence feature data by introducing position encoding; inputting the processed fault feature vector into the TRM layer of the RoBERTa model and processing it using the Transformer encoding layer; learning information from different representation subspaces by using multi-head self-attention; connecting the input and output of the attention module using residual connection, and then standardizing the samples using the normalization layer technology; processing the output of the multi-head self-attention layer through a feedforward neural network; training the corresponding classifier based on the fault type, and using the classifier to diagnose the charging equipment fault, and the charging equipment diagnosis result is a set of classifier output results; the TRM layer integrates the information of the input sequence into the CLS vector through the multi-head self-attention mechanism, and passes it through the classifier to obtain a fault feature set of the charging equipment.

[0012] As a preferred solution of the charging equipment fault diagnosis method combining deep learning and meta-learning described in the present invention, the multi-objective fault diagnosis model is optimized, including the following steps: using the MAML meta-learning method to optimize and adjust the parameters of the multi-objective fault diagnosis model of the charging equipment, and initializing the general parameter θ before the model training starts; using the binary cross entropy function as the loss function of the fault task classifier; the core process of the MAML meta-learning method includes inner-loop optimization and outer-loop optimization; in the inner-loop optimization, the model learns the fault type and gradient updates the model parameters to obtain the optimal model parameters for the diagnosed fault type; in the outer-loop optimization, the model summarizes the loss values of the fault task after inner-loop optimization, calculates the gradient of the total loss on the general parameter θ, and updates the general parameter θ; inputs the fault feature data set into the multi-objective fault diagnosis model, iteratively trains the model, and repeatedly updates the model parameters until the maximum number of model iterations T is reached, stops training, and obtains the optimal parameters of the model.

[0013] As a preferred solution of the charging equipment fault diagnosis method combining deep learning and meta-learning described in the present invention, the method of determining the fault diagnosis result of the charging equipment based on the optimized multi-objective fault diagnosis model includes the following steps: processing the operation and maintenance work order data to obtain first parameter data and second parameter data; comparing the operation and maintenance work order data, the first parameter data and the second parameter data with the mapping relationship to obtain a fault feature sequence; and inputting the fault feature sequence into the optimized multi-objective fault diagnosis model to obtain a fault type set.

[0014] On the second aspect, in order to further solve the safety problems existing in the operation and maintenance of electric vehicle charging facilities, the present invention provides a charging equipment fault diagnosis system combining deep learning and meta-learning, which includes: a data acquisition module for acquiring first parameter data and second parameter data; a mapping construction module for forming a mapping relationship based on the first parameter data and the second parameter data; a feature construction module for constructing a fault feature data set based on the mapping relationship; a parameter optimization module for constructing a multi-objective fault diagnosis model using the fault feature data set and optimizing the multi-objective fault diagnosis model; a fault diagnosis module for determining the fault diagnosis result of the charging equipment according to the optimized multi-objective fault diagnosis model.

[0015] Optionally, the data acquisition module is used to: acquire first parameter data and second parameter data through operation and maintenance work order data. The mapping construction module is used to: acquire abnormal status, fault cause and processing process information through operation and maintenance work order data; perform data cleaning on the collected operation and maintenance work order data, first parameter data and second parameter data; establish a mapping relationship between the operation and maintenance work order data, first parameter data and second parameter data; map the charging pile with the abnormal status, fault cause and processing process information through the constructed mapping relationship to form a comprehensive data set of charging piles. The feature construction module is used to: analyze the comprehensive data set of charging piles to extract original features; combine the original features into composite features through feature combination technology; combine time series analysis and statistical feature extraction methods to mine features from the time dimension; integrate the original features and composite features to obtain a fault feature data set of charging equipment.

[0016] Optionally, the parameter optimization module is used to: convert the fault feature data set into charging equipment fault sequence feature data; convert the charging equipment fault sequence feature data into a fault feature vector using the RoBERTa model; provide position information of elements in the charging equipment fault sequence feature data by introducing position encoding; input the processed fault feature vector into the TRM layer of the RoBERTa model, and process it using the Transformer encoding layer; learn information from different representation subspaces by using multi-head self-attention; connect the input and output of the attention module using residual connection, and then standardize the samples using normalization layer technology; process the output of the multi-head self-attention layer through a feedforward neural network; train the corresponding classifier based on the fault type, and use the classifier to diagnose the charging equipment fault, and the charging equipment diagnosis result is a set of classifier output results; the TRM layer integrates the information of the input sequence into the CLS vector through the multi-head self-attention mechanism, and passes it through the classifier to obtain a fault feature set of the charging equipment.

[0017] Optionally, the parameter optimization module is also used to: optimize and adjust the parameters of the multi-objective fault diagnosis model of the charging equipment using the MAML meta-learning method, and initialize the general parameter θ before the model training starts; use the binary cross entropy function as the loss function of the fault task classifier; the core process of the MAML meta-learning method includes inner-loop optimization and outer-loop optimization; in the inner-loop optimization, the model learns the fault type and gradient updates the model parameters to obtain the optimal model parameters for the diagnosed fault type; in the outer-loop optimization, the model summarizes the loss values of the fault task after inner-loop optimization, calculates the gradient of the total loss on the general parameter θ, and updates the general parameter θ; inputs the fault feature data set into the multi-objective fault diagnosis model, iteratively trains the model, and repeatedly updates the model parameters until the maximum number of model iterations T is reached, stops training, and obtains the optimal parameters of the model.

[0018] Optionally, the fault diagnosis module is used to: process the operation and maintenance work order data to obtain first parameter data and second parameter data; compare the operation and maintenance work order data, the first parameter data and the second parameter data with the mapping relationship to obtain a fault feature sequence; input the fault feature sequence into the optimized multi-objective fault diagnosis model to obtain a fault type set.

[0019] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the charging equipment fault diagnosis method combining deep learning and meta-learning as described in the first aspect of the present invention is implemented.

[0020] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the charging equipment fault diagnosis method combining deep learning and meta-learning as described in the first aspect of the present invention is implemented.

[0021] Beneficial effects of the present invention: The present invention proposes a charging equipment fault diagnosis method that combines deep learning and meta-learning. By acquiring operation and maintenance work order data, the mapping relationship between operation and maintenance work orders, charging messages and charging orders and the fault feature data set of charging equipment are constructed, and based on the RoBERTa deep learning model optimized by the MAML (Model-Agnostic Meta-Learning) meta-learning method, a multi-objective fault diagnosis model for charging equipment is constructed. The model obtains the multi-objective fault diagnosis results of the charging equipment by analyzing the fault feature sequence of the charging equipment; by utilizing the Transformer architecture of the RoBERTa model and using the self-attention mechanism to complete the feature extraction of the charging equipment, the model can capture longer-range context dependencies, while accelerating the convergence speed of the model, improving the accuracy of the multi-objective fault diagnosis of the charging equipment, ensuring the charging safety of electric vehicles, and providing effective services for the operation and maintenance of charging stations; MAML (Model-Agnostic Meta-Learning) is adopted. Meta-Learning) method, by learning a good initialization parameter, enables the model to quickly adapt and perform efficient training when training different types of fault diagnosis tasks. It significantly improves the training efficiency of multi-objective fault diagnosis models, reduces the time and computational overhead of training new tasks from scratch, and accelerates model deployment. In actual fault diagnosis, this method can respond to new fault scenarios more quickly and improve the stability and security of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0023] Figure 1 This is an overall flow chart of the charging equipment fault diagnosis method combining deep learning and meta-learning in Example 1.

[0024] Figure 2 Schematic diagram of the charging equipment fault diagnosis device combining deep learning and meta-learning in Example 1.

[0025] Figure 3 This is a structural diagram of the computer equipment in Example 3. DETAILED DESCRIPTION

[0026] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0027] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0028] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0029] Example 1

[0030] Reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a charging equipment fault diagnosis method that combines deep learning and meta-learning.

[0031] Existing charging equipment fault diagnosis methods have the following main problems: when charging piles operate at high loads for a long time, their performance inevitably declines gradually, and the equipment failure rate increases significantly. In particular, when a single charging pile has multiple faults occurring simultaneously, it not only increases the complexity of maintenance, but also poses great challenges to technicians in fault location and troubleshooting. These problems directly threaten users' charging safety and pose a hidden danger to the stable operation of the industry.

[0032] To address these issues, accurate and efficient multi-objective fault diagnosis of the operating status of charging equipment has become a core issue in ensuring the safe charging of electric vehicles. The introduction of multi-objective fault diagnosis technology for charging equipment provides a new solution for real-time status monitoring and multi-objective fault identification of charging equipment. By analyzing equipment operating data and quickly locating the source of problems, this technology can significantly improve the reliability and availability of charging facilities, reduce maintenance costs, and improve operational efficiency. In addition, multi-objective fault diagnosis can also effectively reduce charging interruptions caused by faults, ensure the stability and safety of the charging process, and enhance the user experience.

[0033] The present invention provides a method that can effectively solve the above-mentioned problems. Next, we will use multiple embodiments to elaborate on how to implement the charging equipment fault diagnosis method that combines deep learning and meta-learning.

[0034] Figure 1 The overall flow chart of the charging equipment fault diagnosis method combining deep learning and meta-learning is shown, including:

[0035] S1: Acquire first parameter data and second parameter data.

[0036] In an optional embodiment, the first parameter data may be charging pile message data such as the working status of the charging pile, charging start and end time, charging voltage, and charging current; the second parameter data may be charging order data such as the vehicle frame number, user ID, charging stop reason identifier, transaction power, and charging time.

[0037] In an embodiment of the present invention, obtaining the first parameter data and the second parameter data includes the following steps:

[0038] The first parameter data and the second parameter data are obtained through the operation and maintenance work order data of the charging pile.

[0039] S2: forming a mapping relationship based on the first parameter data and the second parameter data.

[0040] In an embodiment of the present invention, a mapping relationship is formed based on the first parameter data and the second parameter data, including the following steps: obtaining information related to charging pile failure and maintenance through the operation and maintenance work order data of the charging pile, including abnormal status of the charging equipment, cause of the failure and processing process information.

[0041] The collected operation and maintenance work order data, first parameter data and second parameter data are cleaned to fill in missing values in the data and remove outliers and duplicate data.

[0042] According to the unique number and timestamp keyword of the charging pile, a mapping relationship between the operation and maintenance work order data, the first parameter data and the second parameter data is established, and each operation and maintenance work order data is associated with the corresponding first parameter data and the second parameter data.

[0043] Through the constructed mapping relationship, the charging piles are mapped with abnormal status, fault causes and processing process information to form a comprehensive charging pile data set that includes the real-time status, fault information and processing operations of the charging piles.

[0044] In an optional embodiment, the abnormal status, fault cause and handling process information of the charging device can be charging device related data such as the charging device fault start time, fault end time, fault time, alarm level, charging gun number and fault code.

[0045] In an optional embodiment, data cleaning can adopt a variety of methods and techniques to improve data quality. For example, when filling missing values, machine learning algorithms or statistical methods such as K-nearest neighbor algorithm, regression model-based method or time series interpolation method are used to fill missing values in the data; when removing outliers, statistical methods can be used to detect and remove outliers. For complex scenarios, machine learning-based anomaly detection algorithms such as isolation forest, local anomaly factor or deep learning-based autoencoder can also be introduced to identify anomalies in the data; when removing duplicate data, deduplication can be performed by setting a unique identifier (such as a combination of charging pile number and timestamp), or clustering algorithm can be combined to group similar data to further identify potential duplicate records.

[0046] In an optional embodiment, when establishing a mapping relationship between operation and maintenance work order data, first parameter data, and second parameter data, each operation and maintenance work order data can be associated with the corresponding first parameter data and second parameter data through simple rule matching based on the unique number and timestamp keyword of the charging pile. This is suitable for scenarios with clear data structure and clear rules. In addition, in complex scenarios, machine learning algorithms such as support vector machines, random forests, or neural networks can be used to intelligently match data. For example, a classifier can be trained using a supervised learning model to automatically determine the correlation between the operation and maintenance work order and the first parameter data and the second parameter data based on the charging pile number, timestamp, and other relevant features, thereby being able to handle noise and uncertainty in the data and improve the robustness of the mapping. Secondly, a graph model can also be constructed, with the charging pile number and timestamp as nodes, and the operation and maintenance work order data, first parameter data, and second parameter data as edge attributes. The implicit relationship between the data can be mined through graph embedding or graph neural network technology to achieve more accurate mapping. This embodiment does not make specific limitations on this.

[0047] In an optional embodiment, a deep learning-based mapping method can also be used to establish a mapping relationship between the operation and maintenance work order data, the first parameter data, and the second parameter data, such as using algorithms such as convolutional neural networks or recurrent neural networks to automatically extract features from the data; specifically, the charging pile number, timestamp, and related parameter data can be used as input, and local features can be extracted through a convolutional neural network, or dependencies in the time series can be captured through a recurrent neural network, thereby realizing intelligent association between the operation and maintenance work order and the first parameter data and the second parameter data.

[0048] It should be noted that since charging pile fault data is scattered in operation and maintenance work orders, charging messages and order data, there is a lack of correlation between the data, making it difficult to restore the complete fault scenario. The present invention establishes a mapping relationship through unique numbers and timestamps, which can associate scattered fault data into a unified time series data set. This mapping enables the system to track abnormal signs before the fault occurs and the impact after the fault occurs, thereby providing a more complete basis for fault diagnosis; compared with the method of using only fault alarm data in the prior art, the mapping mechanism of the present invention can provide more comprehensive fault information.

[0049] S3: Construct a fault feature dataset based on the mapping relationship.

[0050] In an embodiment of the present invention, constructing a fault feature data set based on the mapping relationship includes the following steps: analyzing the acquired charging pile comprehensive data set to extract original features.

[0051] Through feature combination technology, the original features are combined into composite features.

[0052] Combining time series analysis and statistical feature extraction methods, features are mined from the time dimension. That is, by counting the charging orders of charging equipment within a week or other time periods, the charging frequency, failure frequency and failure duration characteristics of the charging equipment are obtained.

[0053] The original features and composite features are integrated to obtain the fault feature dataset of the charging equipment.

[0054] In an optional embodiment, the original features may be the charging voltage, charging current, temperature, fault cause, charging time, alarm level, transaction power, and charging power outage original identification of the charging pile, etc., which are relevant data that can be directly extracted from the charging pile comprehensive data set.

[0055] In an embodiment of the present invention, combining original features into composite features can be achieved in a variety of ways, including generating a charging power feature by multiplying the charging current and the charging voltage to reflect the charging capacity of the charging pile; generating a temperature change rate feature by the ratio of the temperature change and the charging time to evaluate the thermal management of the charging equipment; generating a charging efficiency feature by the ratio of the transaction power and the charging time to evaluate the charging capacity of the charging equipment.

[0056] In an optional embodiment, the original features can be combined into composite features by adopting different strategies such as feature combination based on machine learning or feature combination based on deep learning. For example, in complex scenarios, the nonlinear relationship between the original features can be automatically discovered through machine learning algorithms (such as decision trees, random forests or gradient boosting trees, etc.), and composite features can be generated; secondly, the deep learning model can automatically learn high-order combinations of original features through its hidden layer, such as mapping the original features to a higher-dimensional space through the nonlinear transformation of the neural network, thereby capturing complex feature interactions; regardless of the strategy adopted, in the feature combination process, it is necessary to ensure that the characteristics of the original features are retained, and at the same time, the composite features can provide sufficient flexibility to adapt to the needs of subsequent models. For example, when generating charging power features, it is necessary to ensure that the physical meaning of the charging current and charging voltage are not destroyed, and at the same time, the composite features should be able to effectively reflect the actual operating status of the charging pile. This embodiment does not make specific limitations on this.

[0057] It should be noted that charging pile failures are often the result of the combined effects of multiple parameter anomalies, and a single parameter is difficult to reflect the essential characteristics of the failure. Therefore, the composite features constructed by the feature combination technology in the present invention can more accurately reflect the operating status of the charging equipment. For example, the temperature change rate feature not only reflects the temperature status of the equipment, but also reveals the heat dissipation performance, which is of great significance for preventing overheating failures. The design of this composite feature enables the system to discover potential failures that are difficult to identify by single parameter detection.

[0058] S4: Use the fault feature dataset to build a multi-objective fault diagnosis model and optimize the multi-objective fault diagnosis model.

[0059] In an optional embodiment, the construction of a multi-objective fault diagnosis model can use traditional machine learning algorithms to generate multiple independent binary classifiers by training the fault feature data set, each classifier corresponding to a fault type; a deep learning model can be used to automatically extract high-order features in the fault feature data set and capture complex nonlinear relationships. For example, a multi-task learning framework can be used to share the diagnostic tasks of multiple fault types with the underlying feature extraction module of a deep neural network, and an independent output layer can be designed for each fault type, thereby achieving multi-objective fault diagnosis; graph neural networks can also be introduced in complex scenarios to model the fault feature data set, for example, charging piles and their fault types can be used as nodes of the graph, and fault features can be used as attributes of the edges. The implicit relationship between charging piles and fault types can be mined through graph embedding technology, thereby achieving more accurate multi-objective fault diagnosis.

[0060] It should be noted that in the process of building a multi-objective fault diagnosis model, it is necessary to ensure that the characteristics of each fault type are retained, and at the same time the model can provide sufficient flexibility to adapt to the diagnosis requirements of different fault types. This embodiment does not make specific limitations on this.

[0061] Exemplarily, in an embodiment of the present invention, a multi-objective fault diagnosis model is constructed using a fault feature dataset, including the following steps: by using a segmented description method, the numerical features and categorical features in the fault feature dataset are converted into charging equipment fault sequence feature data in a text-like format as input to the RoBERTa model.

[0062] By using BPE (Byte-Pair Encoding) word segmentation technology, the RoBERTa model converts the input charging equipment fault sequence feature data in text-like format into a fault feature vector.

[0063] By introducing positional encoding to represent the position of features in the input sequence, the RoBERTa model is provided with the position information of each element in the input sequence, which helps the model understand the order and temporal relationship of the input data. The specific formula is as follows:

[0064]

[0065] Where pos is the position encoding vector; j is the dimension index in the position encoding; d model is the total number of dimensions of the position encoding vector; PE(pos,j) is the value of the jth element in the position encoding vector.

[0066] The processed fault feature vector is input into the TRM layer of the RoBERTa model, where the RoBERTa model uses multiple stacked Transformer encoding layers to process it. Each layer contains two main sublayers, namely the self-attention sublayer and the feedforward neural network sublayer. The specific formula of the self-attention layer is as follows:

[0067]

[0068] Q=W q ·X

[0069] K=W k ·X

[0070] V=W v ·X

[0071] Among them, Attention(Q,K,V) is the attention weight; Q is the query vector; K is the key vector; V is the value vector; W q 、W k and W v are the weight matrices of query, key, and value corresponding to the attention head respectively; X is the input fault feature sequence; d k is the dimension of query vector, key vector and value vector; softmax(·) is the activation function.

[0072] The RoBERTa model uses multi-head self-attention to enable the model to learn information from different representation subspaces. The specific formula of multi-head self-attention is as follows:

[0073] MultiHead(Q,K,V)=Concat(head1,...,head h )W O

[0074] head h =Attention(Q,K,V)

[0075] Among them, MultiHead(Q,K,V) is the output of the multi-head self-attention mechanism; head h is the output of the h-th head; Q is the query vector; K is the key vector; V is the value vector; W O is the output weight matrix; Attention(Q,K,V) is the attention weight; Concat(·) is the concatenation operation.

[0076] In each Transformer sublayer, a residual connection is used to connect the input and output of the attention module. The residual network enables more efficient training of the deep network. Then, a normalization layer technology is used to normalize each sample in the feature dimension to reduce the dependency between different features. The specific formula is as follows:

[0077] LayerNorm(X+Attention(X))

[0078] Where LayerNorm(·) is the layer normalization function; X is the input fault feature sequence; Attention(X) is the result of the fault feature sequence processed by the self-attention layer.

[0079] The output of the multi-head self-attention layer is processed by a feedforward neural network. The feedforward neural network consists of two linear layers and a nonlinear activation function ReLU. The specific formula is as follows:

[0080] FFN(Z)=ReLU(ZW1+B1)W2+B2

[0081] Among them, FFN(Z) is the output of the Transformer sublayer; Z is the output of the multi-head self-attention layer; W1, W2, B1 and B2 are the learnable parameters in the neural network.

[0082] For multi-objective fault diagnosis of charging equipment, a classifier is trained for each fault type. Each classifier is a binary classification, indicating whether the charging equipment has the fault. All classifiers are used to diagnose the fault of the charging equipment, and the charging equipment diagnosis result is the set of the output results of all classifiers.

[0083] The TRM layer integrates the information of the input sequence into the CLS vector through the multi-head self-attention mechanism, passes it through all classifiers, and obtains the fault feature set of the charging equipment. The specific formula is as follows:

[0084]

[0085] Where Y is the fault set of charging equipment; To diagnose whether the charging equipment has the i-th fault, 0 means no fault, 1 means fault; s is the CLS vector; w i and b i are the parameters of the neural network in the i-th fault classifier; N is the total number of fault types; softmax(·), argmax(·) and ReLU(·) are activation functions.

[0086] It should be noted that, since the fault characteristics of charging piles are temporal and correlated, traditional machine learning methods are difficult to fully utilize the relationship between these features. The present invention adopts a segmented description method to convert numerical features into a text-like format, and combines the self-attention mechanism of the RoBERTa model to learn the temporal correlation between features. This method enables the model to understand the parameter change rules during the charging process, improves the accuracy of fault diagnosis, and has better diagnostic effects than traditional classification methods when dealing with complex faults such as charging interruptions and communication failures.

[0087] In an optional embodiment, a variety of different strategies can be used to optimize the multi-objective fault diagnosis model. For example, traditional optimization algorithms (such as gradient descent, stochastic gradient descent SGD, or Adam optimizer) can be used to optimize the model parameters. These methods gradually adjust the model parameters by calculating the gradient of the loss function to optimize the performance of the model on the training data. In addition, a learning rate scheduling strategy (such as cosine annealing or exponential decay) can be introduced to dynamically adjust the learning rate to accelerate model convergence; or in complex scenarios, meta-learning methods can be used to optimize the model parameters. These methods simulate the training process of multiple tasks so that the model can quickly adapt to new tasks with a small number of samples. For example, in the multi-objective fault diagnosis of charging equipment, the initialization parameters of the model can be optimized by meta-learning methods so that it can converge quickly when facing new fault types.

[0088] In an optional embodiment, an evolutionary algorithm (such as a genetic algorithm (GA), differential evolution (DE), or particle swarm optimization (PSO)) may be used to optimize the multi-objective fault diagnosis model. The model parameters may be optimized by simulating the biological evolution process. For example, the model parameters may be encoded as individual genes, and new parameter combinations may be generated through selection, crossover, and mutation operations. The fitness of the individuals may be evaluated based on the model's performance on the validation set, thereby gradually optimizing the model performance. Secondly, Bayesian optimization is also an efficient global optimization method, particularly suitable for optimization problems in high-dimensional parameter spaces. By constructing a surrogate model to model the objective function, and using an acquisition function to select the most promising parameter combination for evaluation, efficient optimization of the model parameters may be achieved.

[0089] It should be noted that during the model parameter optimization process, it is necessary to ensure that the optimization method can balance the training efficiency and generalization ability of the model, while meeting the requirements of different fault type diagnosis tasks. For example, the weights of different fault types can be dynamically adjusted through a weighted loss function, or regularization technology can be used to prevent model overfitting. This embodiment does not make specific limitations on this.

[0090] Exemplarily, in an embodiment of the present invention, the multi-objective fault diagnosis model is optimized, including the following steps: selecting the MAML (Model-Agnostic Meta-Learning) meta-learning method to optimize and adjust the parameters of the multi-objective fault diagnosis model of the charging equipment; before the model training begins, initializing the general parameter θ, which represents the initialization parameter in all fault tasks.

[0091] In multi-fault diagnosis, the binary cross entropy function is used as the loss function of each fault task classifier. The specific formula is as follows:

[0092] L i =-(y i log(p(y i ))+(1-y i )log(1-p(y i )))

[0093] Among them, L i is the loss value of fault type i; y i ∈{0,1} is the true label value of the i-th fault, 0 means no fault occurs, 1 means a fault occurs; p(y i ) is the probability of the i-th classifier predicting the i-th fault.

[0094] The core process of the MAML meta-learning method includes inner-loop optimization and outer-loop optimization. In the inner-loop optimization, the model learns each fault type separately and performs multiple gradient updates on the model parameters to obtain the optimal model parameters for diagnosing the fault type. The specific formula is as follows:

[0095]

[0096] Among them, θ i ′ is the optimal parameter of the model on fault task i; θ is the general parameter; α is the inner loop learning rate; is the gradient calculation; L i (f θ ) is the loss value of faulty task i.

[0097] In the outer loop optimization, the model calculates the loss value of each fault task after the inner loop optimization. Summarize and calculate the gradient of the total loss on the universal parameter θ, and update the universal parameter θ so that the model can quickly adapt to new tasks with a small amount of updates when training different fault type tasks. The specific formula is as follows:

[0098]

[0099] Among them, θ is the general parameter of the model; is the gradient calculation; is the loss value of fault task i after inner loop optimization; N is the total number of fault types; β is the outer loop learning rate.

[0100] The fault feature dataset of the charging equipment is input into the multi-objective fault diagnosis model, the model is iteratively trained, and the model parameters are repeatedly updated until the maximum number of model iterations T is reached. The training is stopped and the optimal parameters of the model are obtained.

[0101] It should be noted that, given the diverse fault types of charging equipment and the continuous emergence of new fault modes, traditional models require a large number of samples to adapt to the new fault types. The present invention adopts a two-layer optimization structure of the MAML meta-learning method to enable the model to quickly adapt to new fault types. In practical applications, when a new fault type appears, only a small number of samples are needed to achieve better diagnostic effects, which greatly reduces the amount of data and training time required for the model to adapt to new fault types.

[0102] S5: Determine the fault diagnosis result of the charging equipment according to the optimized multi-objective fault diagnosis model.

[0103] In an embodiment of the present invention, determining the fault diagnosis result of the charging device according to the optimized multi-objective fault diagnosis model includes the following steps: processing the operation and maintenance work order data of a single charging device to obtain the first parameter data and the second parameter data of the charging device.

[0104] The operation and maintenance work order data, the first parameter data, and the second parameter data are compared with the constructed mapping relationship, and then the data is processed to obtain a fault feature sequence of the charging device.

[0105] The constructed fault feature sequence is input into the optimized multi-objective fault diagnosis model to obtain the fault type set of the charging equipment. The specific formula is as follows:

[0106] Y=Diagnostic(x;θ;θ′)

[0107] θ′={θ1′,θ2′,...θ′ N}

[0108] Where Y is the fault set of the charging equipment; Diagnostic(x; θ; θ i ) is the optimized multi-objective fault diagnosis model; θ is the general parameter of the model after training optimization; θ i ′ is the optimal parameter of fault task i after training optimization; N is the total number of fault types; θ′ is the parameter set of all fault tasks; x is the fault feature sequence of a single charging device.

[0109] In summary, the present invention proposes a charging equipment fault diagnosis method that combines deep learning and meta-learning. By acquiring operation and maintenance work order data, the mapping relationship between operation and maintenance work orders, charging messages and charging orders and the fault feature data set of charging equipment are constructed, and based on the RoBERTa deep learning model optimized by the MAML (Model-Agnostic Meta-Learning) meta-learning method, a multi-objective fault diagnosis model for charging equipment is constructed. The model obtains the multi-objective fault diagnosis results of the charging equipment by analyzing the fault feature sequence of the charging equipment; by utilizing the Transformer architecture of the RoBERTa model and using the self-attention mechanism to complete the feature extraction of the charging equipment, the model can capture longer-range context dependencies, while accelerating the convergence speed of the model, improving the accuracy of the multi-objective fault diagnosis of the charging equipment, ensuring the charging safety of electric vehicles, and providing effective services for the operation and maintenance of charging stations; the MAML (Model-Agnostic Meta-Learning) meta-learning method is adopted. Meta-Learning) method, by learning a good initialization parameter, enables the model to quickly adapt and perform efficient training when training different types of fault diagnosis tasks. It significantly improves the training efficiency of multi-objective fault diagnosis models, reduces the time and computational overhead of training new tasks from scratch, and accelerates model deployment. In actual fault diagnosis, this method can respond to new fault scenarios more quickly and improve the stability and security of the system.

[0110] Example 2 is an embodiment of the present invention, which provides a charging equipment fault diagnosis system that combines deep learning and meta-learning, including: a data acquisition module for acquiring first parameter data and second parameter data; a mapping construction module for forming a mapping relationship based on the first parameter data and the second parameter data; a feature construction module for constructing a fault feature data set based on the mapping relationship; a parameter optimization module for constructing a multi-objective fault diagnosis model using the fault feature data set and optimizing the multi-objective fault diagnosis model; and a fault diagnosis module for determining the fault diagnosis result of the charging equipment based on the optimized multi-objective fault diagnosis model.

[0111] In an embodiment of the present invention, the data acquisition module is used to:

[0112] The first parameter data and the second parameter data are obtained through the operation and maintenance work order data.

[0113] In an embodiment of the present invention, the mapping construction module is used to:

[0114] Obtain abnormal status, fault causes, and handling process information through operation and maintenance work order data;

[0115] Clean up the collected operation and maintenance work order data, first parameter data, and second parameter data;

[0116] Establishing a mapping relationship between operation and maintenance work order data, first parameter data, and second parameter data;

[0117] Through the constructed mapping relationship, the charging piles are mapped with abnormal status, fault causes and handling process information to form a comprehensive data set of charging piles.

[0118] In an embodiment of the present invention, the feature construction module is used to:

[0119] Analyze the comprehensive data set of charging piles and extract original features;

[0120] Through feature combination technology, the original features are combined into composite features;

[0121] Combine time series analysis and statistical feature extraction methods to mine features from the time dimension;

[0122] The original features and composite features are integrated to obtain the fault feature dataset of the charging equipment.

[0123] In an embodiment of the present invention, the parameter optimization module is used to:

[0124] Converting the fault feature data set into charging equipment fault sequence feature data;

[0125] The RoBERTa model is used to convert the charging equipment fault sequence feature data into a fault feature vector;

[0126] By introducing position coding, the position information of elements in the characteristic data of the charging equipment fault sequence is provided;

[0127] The processed fault feature vector is input into the TRM layer of the RoBERTa model and processed using the Transformer encoding layer;

[0128] Learning information from different representation subspaces by using multi-head self-attention;

[0129] The input and output of the attention module are connected using residual connections, and the normalization layer technology is used to normalize the samples;

[0130] Process the output of the multi-head self-attention layer through a feedforward neural network;

[0131] Based on the fault type, a corresponding classifier is trained and used to diagnose the fault of the charging equipment. The charging equipment diagnosis result is a collection of the classifier output results.

[0132] The TRM layer integrates the information of the input sequence into the CLS vector through the multi-head self-attention mechanism, and passes it through the classifier to obtain the fault feature set of the charging equipment.

[0133] In an embodiment of the present invention, the parameter optimization module is further configured to:

[0134] The MAML meta-learning method is used to optimize and adjust the parameters of the multi-objective fault diagnosis model of charging equipment. Before the model training begins, the universal parameter θ is initialized;

[0135] Use binary cross entropy function as the loss function of the fault task classifier;

[0136] The core process of the MAML meta-learning method includes inner-loop optimization and outer-loop optimization. In the inner-loop optimization, the model learns the fault type and performs gradient updates on the model parameters to obtain the optimal model parameters for the fault type.

[0137] In the outer loop optimization, the model aggregates the loss values of the faulty tasks after inner loop optimization, calculates the gradient of the total loss on the universal parameter θ, and updates the universal parameter θ.

[0138] The fault feature data set is input into the multi-objective fault diagnosis model, the model is iteratively trained, and the model parameters are repeatedly updated until the maximum number of model iterations T is reached. The training is stopped and the optimal parameters of the model are obtained.

[0139] In an embodiment of the present invention, the fault diagnosis module is used to:

[0140] Processing the operation and maintenance work order data to obtain first parameter data and second parameter data;

[0141] Comparing the operation and maintenance work order data, the first parameter data, and the second parameter data with the mapping relationship to obtain a fault feature sequence;

[0142] The fault feature sequence is input into the optimized multi-objective fault diagnosis model to obtain the fault type set.

[0143] Example 3 is an embodiment of the present invention, which is different from the previous embodiment in that:

[0144] like Figure 3As shown, 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 invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0145] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0146] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0147] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0148] Example 4 is an embodiment of the present invention, which provides a charging equipment fault diagnosis method that combines deep learning and meta-learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0149] In this embodiment, by acquiring the operation and maintenance work order data of the charging pile and performing data processing, a multi-objective fault diagnosis model is constructed to diagnose the fault of the charging equipment. Table 1 shows the abnormal status, fault cause and processing process information table of the charging equipment.

[0150] Table 1 Abnormal status, fault causes and handling process of charging equipment

[0151]

[0152] Secondly, in order to build a multi-objective fault diagnosis model, this embodiment obtains the first parameter data and the second parameter data of the charging pile, extracts the feature data therein, and uses it in subsequent model construction and training. Table 2 shows the charging order data table of the charging pile, where 2 indicates remote termination by the user (mobile phone, website); 173 indicates 022BH: BMS abnormally terminates charging; 108 indicates 0126H: BRO message (0x00) reception timeout; 149 indicates 0213H: DC output contactor fault / malfunction.

[0153] Table 2 Charging order data table of charging piles

[0154]

[0155] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A charging equipment fault diagnosis method combining deep learning and meta-learning, characterized by: include: Acquire first parameter data and second parameter data; forming a mapping relationship based on the first parameter data and the second parameter data; Constructing a fault feature data set based on the mapping relationship; constructing a multi-objective fault diagnosis model using the fault feature data set, and optimizing the multi-objective fault diagnosis model; The fault diagnosis results of the charging equipment are determined based on the optimized multi-objective fault diagnosis model.

2. The charging equipment fault diagnosis method combining deep learning and meta-learning according to claim 1, characterized in that: The obtaining of the first parameter data and the second parameter data comprises the following steps: The first parameter data and the second parameter data are obtained through the operation and maintenance work order data.

3. The charging equipment fault diagnosis method combining deep learning and meta-learning according to claim 2, characterized in that: Forming a mapping relationship based on the first parameter data and the second parameter data includes the following steps: Obtain abnormal status, fault causes, and handling process information through operation and maintenance work order data; Clean up the collected operation and maintenance work order data, first parameter data, and second parameter data; Establishing a mapping relationship between operation and maintenance work order data, first parameter data, and second parameter data; Through the constructed mapping relationship, the charging pile is mapped with the abnormal state, fault cause and processing process information to form a comprehensive data set of the charging pile.

4. The charging equipment fault diagnosis method combining deep learning and meta-learning according to claim 3, characterized in that: Constructing a fault feature dataset based on the mapping relationship includes the following steps: Analyzing the charging pile comprehensive data set and extracting original features; Through feature combination technology, the original features are combined into composite features; Combine time series analysis and statistical feature extraction methods to mine features from the time dimension; The original features and composite features are integrated to obtain the fault feature dataset of the charging equipment.

5. The charging equipment fault diagnosis method combining deep learning and meta-learning according to claim 4, characterized in that: Constructing a multi-objective fault diagnosis model using the fault feature data set includes the following steps: Converting the fault characteristic data set into charging equipment fault sequence characteristic data; The RoBERTa model is used to convert the charging equipment fault sequence feature data into a fault feature vector; By introducing position coding, the position information of elements in the characteristic data of the charging equipment fault sequence is provided; The processed fault feature vector is input into the TRM layer of the RoBERTa model and processed using the Transformer encoding layer; Learning information from different representation subspaces by using multi-head self-attention; The input and output of the attention module are connected using residual connections, and the normalization layer technology is used to normalize the samples; Process the output of the multi-head self-attention layer through a feedforward neural network; Based on the fault type, a corresponding classifier is trained and used to diagnose the fault of the charging equipment. The charging equipment diagnosis result is a collection of the classifier output results. The TRM layer integrates the information of the input sequence into the CLS vector through the multi-head self-attention mechanism, and passes it through the classifier to obtain the fault feature set of the charging equipment.

6. The charging equipment fault diagnosis method combining deep learning and meta-learning according to claim 5, characterized in that: Optimizing the multi-objective fault diagnosis model includes the following steps: The MAML meta-learning method is used to optimize and adjust the parameters of the multi-objective fault diagnosis model of charging equipment. Before the model training begins, the universal parameter θ is initialized; Use binary cross entropy function as the loss function of the fault task classifier; The core process of the MAML meta-learning method includes inner-loop optimization and outer-loop optimization. In the inner-loop optimization, the model learns the fault type and performs gradient updates on the model parameters to obtain the optimal model parameters for the diagnosed fault type. In the outer loop optimization, the model aggregates the loss values of the faulty tasks after inner loop optimization, calculates the gradient of the total loss on the universal parameter θ, and updates the universal parameter θ. The fault feature data set is input into the multi-objective fault diagnosis model, the model is iteratively trained, and the model parameters are repeatedly updated until the maximum number of model iterations T is reached. The training is stopped and the optimal parameters of the model are obtained.

7. The charging equipment fault diagnosis method combining deep learning and meta-learning according to claim 6, characterized in that: Determining the fault diagnosis result of the charging equipment according to the optimized multi-objective fault diagnosis model includes the following steps: Processing the operation and maintenance work order data to obtain first parameter data and second parameter data; Comparing the operation and maintenance work order data, the first parameter data, and the second parameter data with the mapping relationship to obtain a fault feature sequence; The fault feature sequence is input into the optimized multi-objective fault diagnosis model to obtain the fault type set.

8. A charging equipment fault diagnosis system combining deep learning and meta-learning, characterized by: include, A data acquisition module, used to acquire first parameter data and second parameter data; A mapping construction module, configured to form a mapping relationship based on the first parameter data and the second parameter data; Feature construction module, used to construct fault feature dataset based on mapping relationship; Parameter optimization module, used to build a multi-objective fault diagnosis model using the fault feature data set and optimize the multi-objective fault diagnosis model; The fault diagnosis module is used to determine the fault diagnosis result of the charging equipment according to the optimized multi-objective fault diagnosis model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the charging equipment fault diagnosis method combining deep learning and meta-learning described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the charging equipment fault diagnosis method combining deep learning and meta-learning described in any one of claims 1 to 7 are implemented.

Images