Intelligent traffic equipment fault analysis method and device based on generative artificial intelligence, medium and terminal

Abstract

The invention provides a smart traffic equipment fault analysis method and device based on generative artificial intelligence, a medium and a terminal, and has the following beneficial effects: remote inference and evaluation of an equipment fault position are realized in a manner of encoding fault event data and generating hierarchical equipment fault degree description through a generation module, so that the fault analysis efficiency is improved. And the fault position is accurately obtained and the equipment is rapidly maintained. The fault data information contained in the fault event can be detected without labeling the fault event data, so that a large amount of time cost and labor cost required for training the model are saved. And meanwhile, an evaluation model is also arranged for evaluating the authenticity of the generated reasoning result, and the accuracy and reliability of the generated reasoning evaluation data are further improved. Therefore, the problems that the fault position cannot be accurately obtained in the existing operation and management equipment fault management process, rapid maintenance of the equipment is difficult to realize, and the operation and management equipment fault cannot be effectively managed according to the system log are solved.

Description

Technical Field

[0001] The present application relates to the field of generative artificial intelligence, and in particular to a method, device, medium and terminal for analyzing faults of intelligent transportation equipment based on generative artificial intelligence. Background Art

[0002] A large number of transportation management equipment are installed on existing expressways or high-grade roads, and it is quite common for these equipment to fail during use. Due to the large number of equipment and the scattered installation locations, when equipment fails, maintenance personnel need to drive dozens or hundreds of kilometers to the fault point for on-site analysis and determination, or rely solely on experience to determine the equipment failure in the monitoring center. However, this method is both time-consuming and costly, and may delay equipment maintenance time and may also affect the normal operation of the road for a long time. For the possible harm caused by the failure of some equipment, it is necessary to promptly and correctly assess the degree of harm it may cause, so as to avoid the failure of equipment with a high degree of harm causing immeasurable huge losses to the transportation system.

[0003] The various hardware and software components of the transportation management equipment in the existing smart transportation system can operate normally and achieve the expected functions and tasks. When a power failure occurs, the equipment will not be able to start normally or continue to operate. The equipment will automatically shut down or enter standby mode and will not be able to provide normal functions and services. At the same time, the power failure will cause the communication between the equipment and other components or systems to be interrupted. When a failure occurs, the device current sensor will record various events and fault information to the system log. These log files contain detailed information such as current anomalies and timestamps for subsequent fault analysis and troubleshooting. However, the existing technology cannot make timely judgments on equipment failures based on system logs. Summary of the invention

[0004] In view of the shortcomings of the prior art mentioned above, the purpose of this application is to provide a smart transportation equipment fault analysis method, device, medium and terminal based on generative artificial intelligence, which is used to solve the problems of low efficiency and high cost in the existing transportation management equipment fault management process, inability to accurately obtain the fault location, difficulty in rapid maintenance of equipment, and inability to effectively manage transportation management equipment faults based on system logs.

[0005] To achieve the above-mentioned purpose and other related purposes, the first aspect of the present application provides a method for uploading the evaluation results to a smart traffic monitoring center, including: obtaining equipment failure event data; performing event data encoding operations and data correlation analysis operations on the equipment failure event data to pre-process the equipment failure event data and generate data element correlations corresponding to the equipment failure event data; training a fault degree generator based on the data element correlations to generate a hierarchical equipment failure degree description; based on the hierarchical equipment failure degree description, judging the authenticity of the hierarchical equipment failure degree description through adversarial training, and inputting the authenticity of the hierarchical equipment failure degree description into a deep learning model for training to output an evaluation result; uploading the evaluation result to a smart traffic monitoring center.

[0006] In some embodiments of the first aspect of the present application, in performing event data encoding operations and data correlation analysis operations on the equipment failure event data, the event data encoding operation includes the following processes: extracting features from the equipment failure event data to generate a feature graph; encoding and representing the feature graph through an attention mechanism to generate a feature graph encoding representation; and calculating the feature graph encoding representation weights based on the attention mechanism to generate a fault event data encoding vector.

[0007] In some embodiments of the first aspect of the present application, in performing event data encoding operations and data correlation analysis operations on the equipment failure event data, the data correlation analysis operation includes the following processes: obtaining the fault event data encoding vector, and using a dimensionality reduction algorithm to perform dimensionality reduction processing on the fault event data encoding vector; based on the position information contained in the fault event data encoding vector, calculating the weight of the fault event data encoding vector through an attention mechanism to generate a context vector; based on the context vector, calculating the correlation of each vector in the context vector through a similarity algorithm to generate the data element correlation corresponding to the equipment failure event data.

[0008] In some embodiments of the first aspect of the present application, the process of calculating the weight of the fault event data encoding vector through an attention mechanism to generate a context vector includes: based on a multi-head attention mechanism, generating a corresponding projection matrix for each attention head, and performing a linear transformation on the fault event data encoding vector based on the projection matrix; calculating a similarity score based on the fault event data encoding vector after the linear transformation; normalizing the similarity score to generate an attention weight used to characterize the contribution of each encoding vector in the fault event data encoding vector to generating a context vector; and calculating and generating the context vector based on the attention weight of the fault event data encoding vector.

[0009] In some embodiments of the first aspect of the present application, the process of training a fault degree generator based on the data element association to generate a hierarchical equipment fault degree description includes: obtaining the data element association corresponding to the equipment fault event data; inputting the data element association into a neural network for linear mapping to convert the data element association into a vector representation of the data element association; performing a transpose convolution operation on the vector representation of the data element association; and training and optimizing the fault degree generator based on the vector representation of the data element association after the transpose convolution operation to generate a hierarchical equipment fault degree description.

[0010] In some embodiments of the first aspect of the present application, based on the hierarchical device failure degree description, the authenticity of the hierarchical device failure degree description is judged through adversarial training, and the authenticity of the hierarchical device failure degree description is input into a deep learning model for training to output an evaluation result. The process includes: obtaining the hierarchical device failure degree description; performing feature extraction on the hierarchical device failure degree description to obtain a feature vector; performing a spatial mapping operation on the extracted feature vector to generate a matching probability between an input sample and a real sample; based on the generated matching probability between the input sample and the real sample, the authenticity of the hierarchical device failure degree description is judged through adversarial training to generate an authenticity judgment result; and performing inference evaluation on the authenticity judgment result to generate the inference evaluation result.

[0011] In some embodiments of the first aspect of the present application, the authenticity judgment result is subjected to inference evaluation to generate the evaluation result. The process includes: inputting the hierarchical equipment failure degree description into a deep regression model to train the deep regression model; optimizing the deep regression model through an optimization algorithm based on the authenticity judgment result; predicting the authenticity result through the optimized deep regression model to generate an authenticity prediction result; comparing the authenticity prediction result with the authenticity judgment result to perform inference evaluation and generate an inference evaluation result.

[0012] To achieve the above-mentioned purpose and other related purposes, the second aspect of the present application provides a smart traffic equipment fault analysis device based on generative artificial intelligence, including: a fault event collection module: used to obtain equipment fault event data; a fault time preprocessing module: used to perform event data encoding operations and data correlation analysis operations on the equipment fault event data, so as to preprocess the equipment fault event data and generate the data element correlation corresponding to the equipment fault event data; a fault description reasoning evaluation module: used to train the fault degree generator based on the data element correlation to generate a hierarchical equipment fault degree description; based on the hierarchical equipment fault degree description, the authenticity of the hierarchical equipment fault degree description is judged through adversarial training, and the authenticity of the hierarchical equipment fault degree description is input into the deep learning model for training to output the evaluation result; an evaluation result uploading module: used to upload the evaluation result to the smart traffic monitoring center.

[0013] To achieve the above-mentioned purpose and other related purposes, the third aspect of the present application provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the intelligent transportation equipment fault analysis method based on generative artificial intelligence.

[0014] To achieve the above-mentioned purpose and other related purposes, the fourth aspect of the present application provides an electronic terminal, including: a processor and a memory; the memory is used to store computer programs, and the processor is used to execute the computer programs stored in the memory, so that the terminal executes the smart transportation equipment fault analysis method based on generative artificial intelligence.

[0015] As described above, the method, device, medium and terminal for analyzing faults of intelligent transportation equipment based on generative artificial intelligence of the present application have the following beneficial effects: by encoding the fault event data and generating a hierarchical description of the degree of equipment fault through a generation module, it is possible to remotely infer and evaluate the equipment fault location, accurately obtain the fault location, and quickly repair the equipment. There is no need to label the fault event data, so the fault data information contained in the fault event can be detected, saving a lot of time and manpower costs required for training the model. At the same time, an evaluation model is also set up to evaluate the authenticity of the generated reasoning results, and further improve the accuracy and reliability of the generated reasoning evaluation data. Thereby solving the problems of inefficiency and high cost in the existing transportation management equipment fault management process, inability to accurately obtain the fault location, difficulty in realizing rapid repair of equipment, and inability to effectively manage transportation management equipment faults based on system logs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1A flow chart of an embodiment of a method for fault analysis of intelligent transportation equipment based on generative artificial intelligence of the present application is shown.

[0017] Figure 2 A schematic diagram of the process of encoding fault event data in an embodiment of a fault analysis method for intelligent transportation equipment based on generative artificial intelligence in the present application is shown.

[0018] Figure 3 A schematic diagram of the process of generating the correlation between various data elements in an embodiment of the intelligent transportation equipment fault analysis method based on generative artificial intelligence of the present application is shown.

[0019] Figure 4 A schematic diagram of the process of generating a hierarchical description of the degree of equipment failure in an embodiment of the intelligent transportation equipment failure analysis method based on generative artificial intelligence in the present application is shown.

[0020] Figure 5 The present application shows that an embodiment of a fault analysis method for intelligent transportation equipment based on generative artificial intelligence evaluates the fault degree description.

[0021] Figure 6 A structural schematic diagram of an embodiment of a smart transportation equipment fault analysis device based on generative artificial intelligence of the present application is shown.

[0022] Figure 7 A structural schematic diagram of another embodiment of the intelligent transportation equipment fault analysis device based on generative artificial intelligence of the present application is shown.

[0023] Figure 8 The schematic diagram of the structure of the electronic terminal for fault analysis of intelligent transportation equipment based on generative artificial intelligence in this application is shown.

[0024] Fig. 9 A schematic diagram of an embodiment of a method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence in the present application is shown.

[0025] Fig.10 A schematic diagram of a ResNet residual network of an embodiment of a method for fault analysis of intelligent transportation equipment based on generative artificial intelligence of the present application is shown.

[0026] Fig.11 A schematic diagram showing a preliminary feature diagram in an embodiment of a method for fault analysis of intelligent transportation equipment based on generative artificial intelligence of the present application.

[0027] Fig.12 A schematic diagram of the fully connected layer in an embodiment of a method for fault analysis of intelligent transportation equipment based on generative artificial intelligence of the present application is shown.

[0028] Fig.13A schematic diagram showing the final characteristic graph in an embodiment of a method for fault analysis of intelligent transportation equipment based on generative artificial intelligence of the present application.

[0029] Fig.14 A schematic diagram of the attention mechanism in an embodiment of a fault analysis method for intelligent transportation equipment based on generative artificial intelligence of the present application is shown.

[0030] Fig.15 A schematic diagram of the nonlinear activation function ReLU in an embodiment of a smart transportation equipment fault analysis method based on generative artificial intelligence in the present application is shown. DETAILED DESCRIPTION

[0031] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0032] It should be noted that in the following description, with reference to the accompanying drawings, several embodiments of the present application are described in the accompanying drawings. It should be understood that other embodiments may also be used, and mechanical composition, structure, electrical and operational changes may be made without departing from the spirit and scope of the present application. The following detailed description should not be considered restrictive, and the scope of the embodiments of the present application is limited only by the claims of the published patents. The terms used here are only for describing specific embodiments and are not intended to limit the present application. Spatially related terms, such as "upper", "lower", "left", "right", "below", "below", "lower", "above", "upper", etc., may be used in the text to facilitate the description of the relationship between an element or feature shown in the figure and another element or feature.

[0033] In this application, unless otherwise clearly specified and limited, the terms "install", "connect", "connect", "fix", "hold" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0034] Furthermore, as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless there is an indication to the contrary in the context. It should be further understood that the terms "comprise", "include" indicate the presence of the described features, operations, elements, components, items, kinds, and / or groups, but do not exclude the presence, occurrence or addition of one or more other features, operations, elements, components, items, kinds, and / or groups. The terms "or" and "and / or" used herein are interpreted as inclusive, or mean any one or any combination. Therefore, "A, B or C" or "A, B and / or C" means "any of the following: A; B; C; A and B; A and C; B and C; A, B and C". Exceptions to this definition will only occur when the combination of elements, functions or operations is inherently mutually exclusive in some way.

[0035] In order to solve the problems in the above-mentioned background technology, the present invention provides a method, device, medium and terminal for analyzing faults of intelligent transportation equipment based on generative artificial intelligence, aiming to solve the problems of low efficiency and high cost in the existing transportation management equipment fault management process, inability to accurately obtain the fault location, difficulty in realizing rapid maintenance of equipment, and inability to effectively manage transportation management equipment faults based on system logs. At the same time, in order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention are further described in detail through the following embodiments and in combination with the accompanying drawings. It should be understood that the specific embodiments described here are only used to explain the present invention and are not used to limit the invention.

[0036] Before further describing the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are applicable to the following interpretations:

[0037] <1> Generative AI: Generative AI refers to a type of AI model that can generate new data based on input data, such as image generation, text generation and other tasks.

[0038] <2> Fault event data: Fault event data refers to data that records relevant information when a device or system failure occurs, including the fault type, occurrence time, fault cause, etc.

[0039] <3> Data encoding operation: Data encoding operation refers to converting raw data into a format that the model can process, usually including data preprocessing, feature extraction, data normalization and other operations.

[0040] <4> Correlation analysis: Correlation analysis refers to discovering the potential relationships and patterns between data by analyzing the correlation and degree of association between data.

[0041] <5> Feature encoding representation: Feature encoding representation refers to the process of converting raw data into a set of feature vectors. Feature extraction algorithms are usually used to extract the key features of the data and encode them into vector form.

[0042] <6> Attention weight: Attention weight refers to the weight value used to measure the contribution of different input information to the output in the attention mechanism.

[0043] <7> Convolutional feedforward network: A convolutional feedforward network is a deep neural network that is usually used to process sequence data or image data. It includes convolutional layers, pooling layers, and fully connected layers.

[0044] <8> Fully connected network: A fully connected network is a deep neural network in which each neuron is connected to all neurons in the previous layer and is usually used to solve classification and regression problems.

[0045] <9> Multi-head attention mechanism: Multi-head attention mechanism refers to applying the attention mechanism to multiple different sets of queries, keys, and value vectors in the self-attention model to improve the performance and generalization ability of the model.

[0046] <10> Cross Entropy: Cross entropy is a measure of the difference between two probability distributions and is often used as a loss function for classification problems.

[0047] <11> Cosine similarity: Cosine similarity is a measure of the similarity between two vectors and is often used in tasks such as text classification and clustering.

[0048] <12> Generator: A generator is a component in a generative model that is used to generate new data, such as images, audio, text, etc.

[0049] <13> Text description: Text description refers to the process of describing an object, scene or event in words.

[0050] <14> Tensor: A tensor is a multidimensional array used to represent vectors, matrices, and higher-dimensional arrays. It is often used in deep learning to represent input data, weight parameters, etc.

[0051] <15> Feature mapping: Feature mapping refers to the process of converting input data into a set of feature maps through convolution or other operations.

[0052] <16> Discriminator: The discriminator is a component in a generative model that is used to determine whether the generated data conforms to the distribution of real data. It is usually used in generative adversarial networks.

[0053] <17> ResNet residual network: ResNet residual network is a deep neural network structure that solves the gradient disappearance and gradient explosion problems in the deep network training process by using residual blocks.

[0054] <18> Skip connection: Skip connection means that in a neural network, the output of a layer is directly connected to the input of a subsequent layer to retain more information and accelerate training.

[0055] The embodiments of the present invention provide a method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence, a system for analyzing faults of intelligent transportation equipment based on generative artificial intelligence, and a storage medium storing an executable program for implementing the method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence. As for the implementation of the method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence, the embodiments of the present invention will illustrate an exemplary implementation scenario of analyzing faults of intelligent transportation equipment based on generative artificial intelligence.

[0056] like Figure 1 As shown, a flow chart of a method for analyzing intelligent transportation equipment failure based on generative artificial intelligence in an embodiment of the present invention is shown. The method for analyzing intelligent transportation equipment failure based on generative artificial intelligence in this embodiment mainly includes the following steps:

[0057] Step S11: Acquire device failure event data.

[0058] In one embodiment of the present invention, the process of obtaining equipment failure event data includes analyzing data collected by a large number of transportation management devices installed in the road space distributed over hundreds of kilometers on the expressway. The collected data includes but is not limited to real-time data, periodically collected and stored data. The collected data is analyzed and preliminarily preprocessed to obtain equipment event data.

[0059] Specifically, under normal power supply, the various hardware and software components of the transportation management equipment under the smart transportation system can operate normally and achieve the expected functions and tasks. Whether it is data processing speed, response time or real-time performance, it meets the system design and operation requirements. Power failure will cause the equipment to fail to start normally or continue to run. The equipment will automatically shut down or enter standby mode, unable to provide normal functions and services. At the same time, power failure will cause communication between the equipment and other components or systems to be interrupted. Due to the variety of transportation management equipment under the smart transportation system, for common street lights, the current under normal circumstances is usually between 0.1-1A, and the current when a fault occurs is 0A, or far more than 1A. When a fault occurs, the device current sensor will record various events and fault information in the system log. These log files contain detailed information such as current anomalies and timestamps for subsequent fault analysis and troubleshooting. The collected fault event data includes but is not limited to the above log files.

[0060] Step S12: performing event data encoding operation and data correlation analysis operation on the equipment failure event data to pre-process the equipment failure event data and generate data element correlation corresponding to the equipment failure event data.

[0061] In one embodiment of the present invention, in performing event data encoding operations and data correlation analysis operations on the equipment failure event data, the event data encoding operation includes the following processes: extracting features from the equipment failure event data to generate a feature graph; encoding and representing the feature graph through an attention mechanism to generate a feature graph encoding representation; and calculating the feature graph encoding representation weight based on the attention mechanism to generate a fault event data encoding vector.

[0062] Figure 2 A schematic diagram of a process for encoding fault event data in one embodiment of the present invention is shown. The process of generating a fault data encoding vector from fault event data includes: acquiring fault event data, extracting features of the input fault event data using a convolutional feedforward network to generate a feature map, encoding and representing the feature map using a multi-head attention mechanism, and calculating the weight of each fault event data to form a fault event data encoding vector.

[0063] In this embodiment, the process of extracting features from input fault event data using a convolutional feedforward network to generate a feature map includes: first subjecting the input fault event data to feature extraction through a convolutional layer to generate a preliminary feature map, then subjecting the input fault event data to dimensionality reduction processing through a pooling layer for subsequent calculations, and finally subjecting the fully connected layer to obtain the final output feature map. The feature map is a high-dimensional feature representation of the fault event data, and the fault event data that has undergone advanced feature extraction operations is more convenient for subsequent encoding representation and inference evaluation generation. Among them, the convolutional feedforward network is a common deep learning model, which is suitable for processing data with a spatial structure, such as data containing time information. Its main feature is that it includes components such as convolutional layers, pooling layers, and fully connected layers.

[0064] In this embodiment, after generating a feature map, the feature map is encoded and represented by a multi-head attention mechanism to generate a fault event data encoding vector. The process includes: using a multi-head attention mechanism to process input data in parallel using multiple attention heads to capture feature representations of different levels and diversity. By fusing the results of multiple attention heads, a richer and more accurate representation is obtained. Specifically, the feature map is first input into a multi-head attention layer, which includes multiple attention heads, each of which is responsible for paying attention to and encoding different parts of the feature map, and each head is assigned a corresponding query, key, and value. In the attention calculation, the similarity between the query and key of each head is calculated, and the value is weighted and summed according to the attention weight. The encoding results of multiple attention heads are then fused. The present application splices the outputs of multiple attention heads to form an encoding representation containing rich features. The present application can also fuse the outputs of multiple heads by linear transformation and weighted summation. Finally, the encoding representation after multi-head fusion is used as output to continue to be used for subsequent tasks, such as fault diagnosis tasks, prediction tasks, or decision-making tasks.

[0065] Furthermore, after obtaining the encoded representation after multi-head fusion, in order to convert the attention score into a probability distribution, maintain continuity and differentiability, suppress non-critical information, and use it in conjunction with the cross-entropy loss function to improve the performance of the model, this application applies the SOFTMAX function, which is usually used to calculate the attention weights of the encoded representation, and converts the original attention weights into a probability distribution to ensure that the sum of the attention weights is 1, thereby forming the final fault event data encoding vector.

[0066]

[0067] Formula 1 shows the Softmax function, where x i represents the i-th original attention weight.

[0068] like Figure 3 As shown, in one embodiment of the present invention, in performing event data encoding operations and data association analysis operations on the equipment failure event data, the data association analysis operation includes the following processes: obtaining the fault event data encoding vector, and using a dimensionality reduction algorithm to perform dimensionality reduction processing on the fault event data encoding vector; based on the position information contained in the fault event data encoding vector, calculating the weight of the fault event data encoding vector through an attention mechanism to generate a context vector; based on the context vector, calculating the association of each vector in the context vector through a similarity algorithm to generate the data element association corresponding to the equipment failure event data.

[0069] In one embodiment of the present invention, the process of using a dimensionality reduction algorithm to perform dimensionality reduction processing on the fault event data encoding vector includes: using a convolutional feedforward network to perform dimensionality reduction processing on the fault event data encoding vector. Specifically, the number and size of convolution kernels used to control the dimension of the output fault event data encoding vector are set. Exemplarily, the size of the convolution kernel is set to 1×1. At the same time, a pooling layer is set to set the size of the convolution kernel to 1×1, while reducing the size while retaining the main features of the original fault event data. The types of pooling layers used include but are not limited to maximum pooling (Max Pooling) and average pooling (Average Pooling). Finally, the pooling window size and stride are set to balance the number of parameters, the amount of calculation and the information loss of the dimensionality reduction model, thereby effectively reducing the dimension of the feature map.

[0070] In one embodiment of the present invention, the process of calculating the weight of the fault event data encoding vector through the attention mechanism to generate a context vector includes: based on the multi-head attention mechanism, generating a corresponding projection matrix for each attention head, and linearly transforming the fault event data encoding vector based on the projection matrix; calculating the similarity score based on the fault event data encoding vector after the linear change; normalizing the similarity score to generate an attention weight used to characterize the contribution of each encoding vector in the fault event data encoding vector to generating the context vector; and calculating and generating the context vector based on the attention weight of the fault event data encoding vector.

[0071] Specifically, a multi-head attention mechanism is used to help the model focus on the relevant information at different positions in the input fault event data encoding vector, and integrate this information to generate a context vector. Through the projection matrix, each attention head assigns attention weights to different positions in the input encoding vector, and then the similarity scores are normalized through the SOFTMAX function to obtain the attention weights.

[0072] Further, the context vector is calculated by attention weights, where the weights represent the contribution of each encoding vector to the generation of the context vector. The attention weights are weighted and summed to obtain a weighted encoding representation, where the attention weights are used as weights and the encoding representations are used as weighted vectors. The weighted summed encoding representation is the context vector, which takes all encoding representations into account and is weighted according to the attention weights.

[0073] In one embodiment of the present invention, the process of calculating the relevance of each vector in the context vector by a similarity algorithm to generate the data element relevance corresponding to the equipment failure event data includes: calculating the similarity between the context vectors by a similarity measurement method to generate the data element relevance. The similarity measurement method includes but is not limited to: cosine similarity, Euclidean distance, Manhattan distance, etc. Exemplarily, the cosine angle between the context vectors is calculated using cosine similarity, and the closer the value is to 1, the more similar it is, and the closer the value is to -1, the less similar it is.

[0074]

[0075] Among them, A and B represent two context vectors respectively, and |A||B| represents the modulus of the vector.

[0076] Step S13: training a fault degree generator based on the data element association to generate a hierarchical device fault degree description.

[0077] like Figure 4 As shown, in one embodiment of the present invention, the process of training a fault degree generator based on the data element association to generate a hierarchical device fault degree description includes: obtaining the data element association corresponding to the device fault event data; inputting the data element association into a neural network for linear mapping to convert the data element association into a vector representation of the data element association; performing a transpose convolution operation on the vector representation of the data element association; and training and optimizing the fault degree generator based on the vector representation of the data element association after the transpose convolution operation to generate a hierarchical device fault degree description.

[0078] In one embodiment of the present invention, based on the hierarchical device failure degree description, the authenticity of the hierarchical device failure degree description is judged through adversarial training, and the authenticity of the hierarchical device failure degree description is input into a deep learning model for training to output an evaluation result. The process includes: obtaining the hierarchical device failure degree description; performing feature extraction on the hierarchical device failure degree description to obtain a feature vector; performing a spatial mapping operation on the extracted feature vector to generate a matching probability between an input sample and a real sample; based on the generated matching probability between the input sample and the real sample, the authenticity of the hierarchical device failure degree description is judged through adversarial training to generate an authenticity judgment result; and performing inference evaluation on the authenticity judgment result to generate the inference evaluation result.

[0079] Exemplarily, the process of inputting the data element association into a neural network for linear mapping to convert the data element association into a vector representation of the data element association includes: inputting the data element association into a fully connected layer for linear mapping to convert it into a vector representation of the data element association. Among them, the fully connected layer (FullyConnected Layer) is a commonly used type in artificial neural networks. Each neuron in the fully connected layer is connected to all neurons in the previous layer. The characteristic is that each neuron receives all outputs of the previous layer and generates a new output. The function of the fully connected layer is to linearly map the input data element association and weight, and introduce nonlinear characteristics through the activation function to convert it into a vector representation of the data association.

[0080] Output = f(w*input + b) (Formula 3)

[0081] Among them, output represents the output correlation vector representation, input represents the input data element correlation, w represents the weight matrix, b is the bias vector, and f is the activation function.

[0082] In this embodiment, the process of performing a transposed convolution operation on the vector representation of the data element association includes: setting the convolution kernel size, stride, padding and number of output channels in the transposed convolution layer to determine the operation mode of the transposed convolution and the shape of the output. Subsequently, the data association vector representation is input into the transposed convolution layer, and the output tensor of the same dimension as the real sample is obtained through transposed convolution calculation. Specifically, the calculation process of the transposed convolution is to obtain the corresponding position tensor value of the output tensor of the same dimension as the real sample by performing weighted product and addition of the convolution kernel for each position of the input data association vector.

[0083] In one embodiment of the present invention, the process of training and optimizing the fault degree generator to generate a hierarchical device fault degree description includes: inputting the corresponding position tensor value of the tensor with the same dimension as the real sample and the randomly generated noise into the fault degree generator for iterative training, and adjusting the parameters of the fault degree generator according to the generated tensor representation, so that the fault degree generator gradually learns the correct fault degree description. Finally, a generated tensor representation is obtained, which is the fault degree description of the device.

[0084] Step S14: Based on the hierarchical device failure degree description, the authenticity of the hierarchical device failure degree description is judged through adversarial training, and the authenticity of the hierarchical device failure degree description is input into the deep learning model for training to output an evaluation result.

[0085] like Figure 5 As shown, in one embodiment of the present invention, the process of performing reasoning evaluation on the authenticity judgment result to generate the evaluation result includes: inputting the hierarchical equipment failure degree description into the deep regression model to train the deep regression model; optimizing the deep regression model through an optimization algorithm based on the authenticity judgment result; predicting the authenticity result through the optimized deep regression model to generate an authenticity prediction result; comparing the authenticity prediction result with the authenticity judgment result to perform reasoning evaluation and generate a reasoning evaluation result.

[0086] In one embodiment of the present invention, the hierarchical equipment failure degree description is input into a deep regression model, and the process of training the deep regression model includes: extracting features of the failure degree description through a convolutional layer and a fully connected layer, mapping the extracted features to a judgment result space to generate a matching probability between an input sample and a real sample, and performing authenticity judgment on the equipment failure degree description generated by the generator through an adversarial deep regression model.

[0087] In this embodiment, the process of extracting features from the fault degree description through a convolutional layer and a fully connected layer includes: using a convolutional layer to extract local features of the fault degree description. Specifically, the convolution kernel size is set to extract features of different sizes from the fault degree description by sliding on sliding windows of different sizes. Each convolution kernel generates a feature map to indicate the presence or absence of different features. Optionally, one or more fully connected layers are set at the output end of the pooling layer to output features through the fully connected layer to express a richer feature representation. Finally, a weight matrix is ​​set at the eigenvalue output end to connect all inputs and outputs to extract features from the fault degree description.

[0088] In this embodiment, the process of mapping the extracted features to the judgment result space to generate the matching probability between the input sample and the real sample refers to converting the feature representation of the input data into a form that can be used for judgment or classification. The specific process includes: mapping the extracted features to the probability distribution of different categories through the fully connected layer, where each output category corresponds to a node in the output layer of the fully connected layer, and the value of the node is used to represent the probability that the current feature value belongs to the category. In order to further use the node value to represent the discrimination result, the result after feature mapping is input into the SOFTMAX function to convert the output into a probability distribution.

[0089]

[0090] where x i Represents the result after feature mapping.

[0091] In this embodiment, the process of using an adversarial deep regression model to judge the authenticity of the equipment fault degree description generated by the generator includes: by constructing an adversarial model, during the training process, alternating adversarial training is performed using the generator and the discriminator, so that the goal of the generator is to generate a fault degree description with high authenticity, and the discriminator improves the accuracy of distinguishing real samples from generated samples.

[0092] In the above adversarial training process, based on the authenticity discrimination result, the process of optimizing the deep regression model through the optimization algorithm includes: optimizing by calculating the loss function of the discriminator and the generator. For the discriminator, the loss function aims to minimize its classification error of the real sample and the generated sample. For the generator, the loss function aims to maximize the probability of misclassification of the generated sample by the discriminator, and the loss function used includes but is not limited to: cross entropy loss function.

[0093]

[0094] where x i Represents the general distribution of the input.

[0095] In one embodiment of the present invention, the authenticity prediction result is compared with the authenticity discrimination result to perform reasoning evaluation and generate a reasoning evaluation result, including: using a deep regression model to evaluate the nonlinear mapping relationship between input features and fault degree, updating the parameters of the deep regression model by minimizing the mean square error, and performing regression prediction for a given input to obtain a description or evaluation of the fault degree. When performing regression prediction for a given input, the prediction result output by the deep regression model is compared with the actual fault degree, and the mean square error is calculated to evaluate the performance and accuracy of the deep regression model.

[0096] Step S15: Upload the evaluation results to the smart traffic monitoring center.

[0097] In one embodiment of the present invention, the evaluation results are uploaded to the smart traffic monitoring center to assist the smart traffic monitoring center in timely maintenance of faulty equipment, and efficient equipment maintenance routes can be specified based on the geographical locations and priorities of multiple faulty equipment to reduce the time and labor costs of equipment maintenance.

[0098] like Figure 6 and Figure 7 As shown, a schematic diagram of the structure of a smart transportation equipment fault analysis device based on generative artificial intelligence in an embodiment of the present invention is shown. In this embodiment, the smart transportation equipment fault analysis device 600 based on generative artificial intelligence includes:

[0099] Fault event collection unit 601: used to obtain device fault event data.

[0100] The fault time preprocessing unit 602 is used to perform event data encoding operations and data correlation analysis operations on the equipment fault event data, so as to preprocess the equipment fault event data and generate data element correlations corresponding to the equipment fault event data.

[0101] Fault description reasoning evaluation unit 603: used to train the fault degree generator based on the data element association to generate a hierarchical device fault degree description. Based on the hierarchical device fault degree description, the authenticity of the hierarchical device fault degree description is judged through adversarial training, and the authenticity of the hierarchical device fault degree description is input into the deep learning model for training to output an evaluation result.

[0102] Evaluation result uploading unit 604: used to upload the evaluation result to the intelligent traffic monitoring center.

[0103] It should be noted that: the intelligent transportation equipment fault analysis device based on generative artificial intelligence provided in the above embodiment only uses the division of the above program modules as an example when performing intelligent transportation equipment fault analysis based on generative artificial intelligence. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device is divided into different program modules to complete all or part of the processing described above. In addition, the intelligent transportation equipment fault analysis device based on generative artificial intelligence provided in the above embodiment and the intelligent transportation equipment fault analysis method embodiment based on generative artificial intelligence belong to the same concept. The specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0104] Among them, Figure 7As shown, the intelligent transportation equipment fault analysis device based on generative artificial intelligence includes a fault collection unit, a fault event preprocessing unit, a fault description reasoning evaluation unit and an evaluation result uploading unit. The fault event preprocessing unit includes an event data encoding module and a data correlation analysis module. The fault description reasoning evaluation unit includes a fault degree description generation module and a fault degree reasoning evaluation module. The event data encoding module in the fault event preprocessing unit is used to encode the input fault event data to form a corresponding encoding vector, which is used in the data correlation analysis module to infer the correlation of each data element in the fault event, and drive the subsequent generation of accurate fault degree description and reasoning evaluation. The fault degree description generation module in the fault description reasoning evaluation unit is used to generate a hierarchical equipment fault degree description, so as to induce the fault degree reasoning evaluation module to correctly distinguish between the real description and the generated hierarchical equipment fault degree description through adversarial training, thereby realizing accurate equipment fault degree reasoning evaluation.

[0105] The method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence provided by the embodiment of the present invention can be implemented on the terminal side or the server side. As for the hardware structure of the terminal for analyzing faults of intelligent transportation equipment based on generative artificial intelligence, please refer to Figure 8 , is an optional hardware structure diagram of a smart transportation equipment fault analysis terminal 800 based on generative artificial intelligence provided in an embodiment of the present invention. The terminal 800 may be a mobile phone, a computer device, a tablet device, a personal digital processing device, a factory background processing device, etc. The smart transportation equipment fault analysis terminal 800 based on generative artificial intelligence includes: at least one processor 801, a memory 802, at least one network interface 804 and a user interface 806. The various components in the device are coupled together through a bus system 805. It can be understood that the bus system 805 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 805 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, in Figure 8 In the specification, various buses are labeled as bus systems.

[0106] The user interface 806 may include a display, a keyboard, a mouse, a trackball, a click gun, keys, buttons, a touch pad or a touch screen.

[0107] It is understood that the memory 802 can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM). The memory described in the embodiments of the present invention is intended to include but is not limited to these and any other suitable categories of memory.

[0108] The memory 802 in the embodiment of the present invention is used to store various categories of data to support the operation of the intelligent transportation equipment fault analysis terminal 800 based on generative artificial intelligence. Examples of these data include: any executable program used to operate on the intelligent transportation equipment fault analysis terminal 800 based on generative artificial intelligence, such as an operating system 8021 and an application 8022; the operating system 8021 includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks. The application 8022 may include various applications, such as a media player (Media Player), a browser (Browser), etc., for implementing various application services. The intelligent transportation equipment fault analysis method based on generative artificial intelligence provided in the embodiment of the present invention may be included in the application 8022.

[0109] The method disclosed in the above embodiment of the present invention can be applied to the processor 801, or implemented by the processor 801. The processor 801 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit in the processor 801 or the instruction in the form of software. The above processor 801 may be a general processor, a digital signal processor (DSP, Digital Signal Processor), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 801 can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiment of the present invention. The general processor 801 can be a microprocessor or any conventional processor, etc. In combination with the steps of the accessory optimization method provided in the embodiment of the present invention, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0110] In an exemplary embodiment, the generative artificial intelligence-based intelligent transportation equipment fault analysis terminal 800 can be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), and complex programmable logic devices (CPLDs) to execute the aforementioned generative artificial intelligence-based intelligent transportation equipment fault analysis method.

[0111] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to the computer program. The aforementioned computer program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other media that can store program codes.

[0112] In the embodiments provided in the present application, the computer readable and writable storage medium may include a read-only memory, a random access memory, an EEPROM, a CD-ROM or other optical disk storage device, a disk storage device or other magnetic storage device, a flash memory, a USB flash drive, a mobile hard disk, or any other medium that can be used to store a desired program code in the form of an instruction or data structure and can be accessed by a computer. In addition, any connection can be appropriately referred to as a computer-readable medium. For example, if the instruction is sent from a website, a server or other remote source using a coaxial cable, an optical fiber cable, a twisted pair, a digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, optical fiber cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. However, it should be understood that computer readable and writable storage media and data storage media do not include connections, carriers, signals, or other temporary media, but are intended to be non-temporary, tangible storage media. Disk and disc, as used in this application, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers.

[0113] The above describes the method, device, terminal and medium for analyzing faults of intelligent transportation equipment based on generative artificial intelligence provided by the present application. Figures 9 to 15 The fault analysis method and device of intelligent transportation equipment based on generative artificial intelligence are further described.

[0114] like Fig. 9 An example of a current fault event log file obtained by a fault event acquisition unit in an embodiment of the present application is shown. The fault event acquisition unit (such as a current sensor as an embodiment) acquires fault event data into a system log. These log files contain detailed information such as current anomalies and timestamps for subsequent fault analysis and troubleshooting.

[0115] In this embodiment, the process of extracting features from the collected fault event data includes: inputting the collected fault event data into a ResNet model to generate a preliminary extraction result; and fusing and outputting the preliminary extraction result and the fault event data through a skip-layer connection to generate a feature map.

[0116] Specifically, the residual block of this embodiment is as follows: Fig.10 As shown in , it consists of two 3×3 convolutional layers, and the original input and final output are fused together through skip layer connections to improve the representation ability of the feature map. Fig.11The schematic diagram of the structure of generating a preliminary feature map by first extracting the input fault event data through a convolutional layer in one embodiment of the present invention is shown. In this embodiment, the dimension of the preliminary feature map is set to 512 dimensions, and the size of the feature map is set to 14*14*512. The preliminary feature map is input to the pooling layer for dimensionality reduction to facilitate subsequent calculations. Finally, the final output features are obtained through a fully connected layer. Fig.12 The structural diagram of the fully connected layer is shown. Both ends of the fully connected layer are composed of neurons, and the neurons at both ends are connected to each other. The number of neurons at the input end is set to 512 dimensions, and the number of neurons at the output end is set to 1024 dimensions. The feature map finally generated is the high-dimensional feature representation of the fault event data. Fig.13 FIG. 2 shows a schematic diagram of the structure of the deep feature information of the log file in this embodiment, wherein the dimension of the deep feature information of the log file is 1024 dimensions and the size is 7*7*1024.

[0117] In one embodiment of the present invention, the process of using multi-head attention to infer the context vector of fault event data includes: initializing each attention head, and performing a convolution operation on the weight matrix of the initialized attention head and the above-mentioned feature map to generate a fault event data vector encoding.

[0118] It is worth noting that the process of initializing each attention head includes: using a multi-head attention mechanism containing 3 scaled dot product attention heads. For each attention head, there are three weight matrices, namely the query matrix (W Q ), key matrix (W K ) and the numerical matrix (W V ). Where W Q , W K , W V Represent the first query weight matrix, the first key weight matrix and the first numerical weight matrix respectively. The above weight matrix is ​​a parameter obtained through training and learning, which is used to linearly transform the input feature map. The dimensions of the weight matrix are represented by d and c, where d represents the dimension of the input or the dimension of the attention head, and c represents the dimension of the output.

[0119] Specifically, Fig.14The figure shows the process of using multi-head attention to infer the context vector of fault event data in one embodiment of the present invention. First, initialize each attention head. Exemplarily, three weight matrices are created for each attention head, which includes determining the dimension of the weight matrix and the initialization value. Exemplarily, d is 1024 dimensions and c is 512 dimensions. Secondly, for each attention head, a convolution operation is performed on the input feature map and the initialized weight matrix to spatially map the fault data encoding vector, and finally generate a fault event data vector encoding. Exemplarily, the fault event data vector encoding dimension obtained after the convolution operation is 7*7*512.

[0120] In one embodiment of the present invention, the process of generating the correlation of individual data elements of a fault event based on the fault event data vector encoding includes: inputting the fault event data vector encoding into a convolutional feedforward network for vector dimensionality reduction processing, using a multi-head attention mechanism to infer the context vector of the fault event data, and calculating the similarity between each fault event data element and the context vector to obtain the correlation of each data element of the fault event.

[0121] Specifically, the process of inputting the fault event data vector code into the convolutional feedforward network for vector dimensionality reduction processing includes: using a 1*1*256 convolution kernel to perform a convolution operation on the fault event data vector code with a coding dimension of 7*7*512 to achieve dimensionality reduction of the fault event data vector code. After the dimensionality reduction operation, the fault event data vector code dimension is 7*7*256.

[0122] Specifically, Fig.14 The process of calculating the context vector of fault event data using a multi-head attention mechanism as shown includes: In this embodiment, the attention mechanism is adopted and three new weight matrices are generated for each attention head, wherein each weight matrix contains a linear layer. The linear layers of the second query weight matrix, the second key weight matrix and the second numerical weight matrix are connected to the scaled dot product attention mechanism layer. Exemplarily, the dimensions d of the generated query weight matrix, the key weight matrix and the numerical weight matrix are 265 dimensions and c is 128 dimensions. The fault event data vector encoding of the dimension 7*7*256 after the dimensionality reduction operation is convolved with the second query weight matrix, the second key weight matrix and the second numerical weight matrix to spatially map the fault data encoding vector, and generate three sets of three-dimensional tensors mapped to the query weight matrix, the key weight matrix and the numerical weight matrix space.

[0123] Specifically, the process of generating the context vector of the fault event data includes: calculating the attention score matrix based on three sets of three-dimensional tensors mapped to the query weight matrix, key weight matrix and value weight matrix space; and calculating the attention weight based on the attention score data to generate the context vector of the fault event data. Specifically, Fig.14 As shown, three sets of three-dimensional tensors with dimensions of 7*7*128 mapped to the query weight matrix, key weight matrix and numerical weight matrix space are respectively passed to the scaled dot product attention mechanism, and the following operations are performed: First, based on the dot product of the key weight matrix and the query weight matrix, a scaling operation is performed to generate an attention score matrix. Exemplarily, the scaling operation is performed by dividing by the square root of 128. Secondly, the attention score data frame is input into the SOFTMAX function for a soft maximum operation to generate the attention weight of each point in the mapping space. Finally, the attention weight is weighted and summed with the numerical weight matrix to generate a context vector of the final fault event data with a dimension of 7*7*128.

[0124] Specifically, the process of calculating the similarity between each fault event data element and the context vector to obtain the correlation between each data element of the fault event includes using cosine similarity to calculate the similarity between each fault event data element and the context vector. Exemplarily, the fault event data set includes the following data elements: data element A: [0.2, 0.4, 0.6, 0.8] data element B: [0.1, 0.3, 0.5, 0.7] data element C: [0.3, 0.5, 0.7, 0.9]. The context vector generated by the above-mentioned multi-head attention mechanism is [0.5, 0.4, 0.3, 0.6]. The correlation of the context vector is calculated using the cosine similarity calculation method shown in Formula 2 to generate the correlation between each data element of the fault event.

[0125] In one embodiment of the present invention, based on the correlation of each data element that generates a fault event, linear mapping is performed through a fully connected layer and nonlinear characteristics are introduced through an activation function to convert the correlation of each data element that generates the fault event into a vector representation of the data element correlation. Subsequently, a transposed convolution is used to map the vector representation of the data element correlation into a sample space and convert it into a tensor of the same dimensionality as the real sample. Finally, the dimensional tensor and randomly generated noise are input into a generator, and the generator is iteratively optimized using a back-propagation algorithm to generate a description of the degree of equipment fault.

[0126] Specifically, based on the correlation of each data element generating the fault event, a linear mapping is performed through a fully connected layer and a nonlinear characteristic is introduced through an activation function to convert the correlation of each data element generating the fault event into a vector representation of the data element correlation, including: using a fully connected layer to perform a linear mapping on the input data element correlation, and introducing a nonlinear characteristic through an activation function to convert it into a vector representation of the data correlation. Exemplarily, a fully connected layer including 128 input neurons and 512 output neurons is used for linear mapping, wherein the activation function is as follows: Fig.15 The ReLU activation function shown is used to introduce nonlinear characteristics, that is, if the correlation of each data element of the input fault event is greater than zero, the original value is retained, otherwise the correlation of each data element of the fault event is set to 0. After the fully connected layer and the activation function, a vector representation of the data correlation with a dimension of 7*7*512 is generated.

[0127] Specifically, the transposed convolution is used to map the vector representation of the association degree of data elements into the sample space and convert it into a tensor of the same dimension as the real sample. Among them, the transposed convolution layer, also known as the deconvolution layer, is usually used for upsampling or deconvolution operations to convert low-resolution input into high-resolution output. The transposed convolution performs weighted product and addition of the convolution kernel at each position of the input data association vector. For each position, the weight of the convolution kernel will be multiplied by the channel value of the association vector and then added to generate the value of the corresponding position of the output tensor, and the output tensor value of the same dimension as the real sample is obtained.

[0128] Specifically, the dimensional tensor and the randomly generated noise are input into the generator, and the generator is iteratively optimized using the back propagation algorithm to generate a description of the degree of equipment failure. The process includes: inputting the dimensional tensor and the randomly generated noise into the generator, setting the number of iterations of the generator to 100, and optimizing the generator through the back propagation algorithm and the gradient descent algorithm to generate a hierarchical description of the degree of equipment failure with high accuracy.

[0129] In one embodiment of the present invention, the process of performing reasoning and evaluation on the generated hierarchical equipment failure degree description includes: obtaining the hierarchical equipment failure degree description input by the failure degree risk description generation module, performing feature extraction on the failure degree description through a convolutional layer and a fully connected layer, mapping the extracted features to the space of the judgment results and generating a matching probability between the input sample and the real sample, accurately judging the authenticity of the equipment failure degree description generated by the generator through adversarial training, and performing reasoning and evaluation on the failure degree description using a deep regression model and uploading the evaluation result output.

[0130] Specifically, the process of extracting features from the fault degree description through the convolution layer and the fully connected layer includes: first, extracting features from the input hierarchical device fault degree description using the convolution layer and the fully connected layer. Exemplarily, the number of convolution kernels of the convolution layer is set to 64, and the size of the hierarchical device fault degree description feature map output is 28*28*64. The output neurons of the fully connected layer are set to 512, and the size of the hierarchical device fault degree description feature map output by the fully connected layer is 14*14*512.

[0131] Specifically, the process of mapping the extracted features to the space of the judgment results and generating the matching probability between the input sample and the real sample includes: inputting the hierarchical equipment fault degree description feature map with a dimension of 14*14*512 into the fully connected layer to convert the hierarchical equipment fault degree description feature map into a form that can be used for judgment or classification, and generating the matching probability between the input sample and the real sample. Exemplarily, the number of output neurons of the fully connected layer used to generate the matching probability is set to 1024, and the dimension of the output feature map after the fully connected layer is 14*14*1024. The output 14*14*1024 is input into the SOFTMAX function for a soft maximum operation to generate the matching probability between the input sample and the real sample. The value output by the SOFTMAX function is between 0 and 1, and the closer to 1, the higher the matching degree with the real sample.

[0132] Specifically, the process of accurately judging the authenticity of the description of the degree of equipment failure generated by the generator through adversarial training includes: based on the matching probability between the input sample and the real sample and the description of the degree of equipment failure, the discriminator is trained, and the loss function of the discriminator and the generator is calculated, and the discriminator is optimized and trained based on the loss function, the back propagation algorithm and the gradient descent algorithm, so that the optimized discriminator is used to generate the authenticity of the description of the degree of equipment failure. Wherein, the loss function includes the cross entropy loss function. Exemplarily, the number of training iterations is set to 100, and the partial derivative of the objective function (loss function) to each neuron weight is obtained layer by layer through the back propagation algorithm to form the gradient of the objective function to the weight vector as the basis for modifying the weight, wherein the training process of the discriminator is improved during the weight modification process.

[0133] Specifically, the process of using a deep regression model to infer and evaluate the fault degree description and output and upload the evaluation results includes: constructing a deep regression model to learn the nonlinear mapping relationship between the input features and the equipment fault degree description, so as to perform regression prediction of the equipment fault degree description, and generate an evaluation result of the fault degree. Exemplarily, the deep regression model in this embodiment includes five hidden layers, and ReLU is used as the activation function of the deep regression model.

[0134] In summary, the present application provides a method, device, terminal and medium for fault analysis of intelligent transportation equipment based on generative artificial intelligence. The present invention provides a method for improving the efficiency of fault analysis of intelligent transportation equipment based on generative artificial intelligence. By encoding the fault event data and generating a hierarchical description of the degree of equipment fault through a generation module, it is possible to remotely infer and evaluate the equipment fault location, accurately obtain the fault location, and quickly repair the equipment. There is no need to mark the fault event data, and the fault data information contained in the fault event can be detected, saving a lot of time and manpower costs required for training the model. At the same time, an evaluation model is also set to evaluate the authenticity of the generated reasoning results, and further improve the accuracy and reliability of the generated reasoning evaluation data. Therefore, the present application effectively overcomes the various shortcomings in the prior art and has a high industrial utilization value.

[0135] The above embodiments are merely illustrative of the principles and effects of the present application and are not intended to limit the present application. Anyone familiar with the technology may modify or change the above embodiments without violating the spirit and scope of the present application. Therefore, all equivalent modifications or changes made by a person of ordinary skill in the art without departing from the spirit and technical ideas disclosed in the present application shall still be covered by the claims of the present application.

Claims

1. A method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence, characterized in that: include: Obtain equipment failure event data; Performing an event data encoding operation and a data relevance analysis operation on the equipment failure event data to pre-process the equipment failure event data and generate a data element relevance corresponding to the equipment failure event data; Training a fault severity generator based on the data element association to generate a hierarchical device fault severity description; Based on the hierarchical device failure degree description, the authenticity of the hierarchical device failure degree description is judged through adversarial training, and the authenticity of the hierarchical device failure degree description is input into a deep learning model for training to output an evaluation result; The evaluation results are uploaded to the smart traffic monitoring center.

2. The method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence according to claim 1 is characterized in that: In performing an event data encoding operation and a data correlation analysis operation on the equipment failure event data, the event data encoding operation includes the following process: Extracting features from the equipment failure event data to generate a feature graph; Encoding the feature map through an attention mechanism to generate a feature map encoding representation; Based on the attention mechanism, the feature map encoding representation weight is calculated to generate a fault event data encoding vector.

3. The method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence according to claim 2 is characterized in that: In performing the event data encoding operation and the data correlation analysis operation on the equipment failure event data, the data correlation analysis operation includes the following process: Acquire the fault event data encoding vector, and perform dimensionality reduction processing on the fault event data encoding vector using a dimensionality reduction algorithm; Based on the position information contained in the fault event data encoding vector, a weight of the fault event data encoding vector is calculated through an attention mechanism to generate a context vector; Based on the context vector, the relevance of each vector in the context vector is calculated by a similarity algorithm to generate a data element relevance corresponding to the equipment failure event data.

4. The method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence according to claim 3 is characterized in that: The process of calculating the weight of the fault event data encoding vector through the attention mechanism to generate a context vector includes: Based on the multi-head attention mechanism, a corresponding projection matrix is ​​generated for each attention head, and a linear transformation is performed on the fault event data encoding vector based on the projection matrix; Calculate the similarity score based on the linearly transformed fault event data encoding vector; Normalizing the similarity scores to generate an attention weight for characterizing the contribution of each encoding vector in the fault event data encoding vector to generating a context vector; The context vector is calculated and generated based on the attention weight of the fault event data encoding vector.

5. The method for analyzing intelligent transportation equipment failure based on generative artificial intelligence according to claim 1 is characterized in that: The process of training the fault severity generator based on the data element association to generate a hierarchical device fault severity description includes: Obtaining the data element correlation degree corresponding to the equipment failure event data; Inputting the data element association degree into a neural network for linear mapping to convert the data element association degree into a vector representation of the data element association degree; Performing a transposed convolution operation on the vector representation of the association degree of the data elements; Based on the vector representation of the correlation degree of data elements after the transposed convolution operation, the fault severity generator is trained and optimized to generate a hierarchical description of the device fault severity.

6. The method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence according to claim 1 is characterized in that: Based on the hierarchical device failure degree description, judging the authenticity of the hierarchical device failure degree description through adversarial training, and inputting the authenticity of the hierarchical device failure degree description into a deep learning model for training to output an evaluation result includes: Obtaining a description of the graded device failure degree; Extracting features from the hierarchical device fault degree description to obtain a feature vector; Performing a spatial mapping operation on the extracted feature vector to generate a matching probability between the input sample and the real sample; Based on the matching probability between the generated input sample and the real sample, the authenticity of the hierarchical device fault degree description is judged through adversarial training to generate an authenticity judgment result; Performing reasoning evaluation on the authenticity determination result to generate the reasoning evaluation result.

7. The method for analyzing faults of intelligent transportation equipment based on generative artificial intelligence according to claim 6 is characterized in that: The process of performing reasoning evaluation on the authenticity judgment result to generate the evaluation result includes: Inputting the hierarchical device failure degree description into a deep regression model to train the deep regression model; Based on the authenticity determination result, optimizing the deep regression model through an optimization algorithm; Performing authenticity result prediction through the optimized deep regression model to generate authenticity prediction results; The authenticity prediction result is compared with the authenticity determination result to perform reasoning evaluation and generate a reasoning evaluation result.

8. A fault analysis device for intelligent transportation equipment based on generative artificial intelligence, characterized in that: include: Fault event collection unit: used to obtain equipment fault event data; A fault time preprocessing unit is used to perform an event data encoding operation and a data correlation degree analysis operation on the equipment fault event data, so as to preprocess the equipment fault event data and generate a data element correlation degree corresponding to the equipment fault event data; A fault description reasoning and evaluation unit is used to train a fault degree generator based on the data element association to generate a hierarchical device fault degree description; based on the hierarchical device fault degree description, determine the authenticity of the hierarchical device fault degree description through adversarial training, and input the authenticity of the hierarchical device fault degree description into a deep learning model for training to output an evaluation result; Evaluation result uploading unit: used to upload the evaluation result to the smart traffic monitoring center.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the intelligent transportation equipment fault analysis method based on generative artificial intelligence as described in any one of claims 1 to 7.

10. An electronic terminal, characterized in that: include: Processor and memory; The memory is used to store computer programs; The processor is used to execute the computer program stored in the memory so that the terminal executes the intelligent transportation equipment fault analysis method based on generative artificial intelligence as described in any one of claims 1 to 7.

Images