Vehicle equipment consumption prediction method, model training method, device, computer-readable storage medium and electronic device

Abstract

The present disclosure provides a vehicle equipment consumption prediction method, model training method, device, computer-readable storage medium and electronic device, which relate to the field of vehicle equipment maintenance prediction technology, wherein the method comprises: obtaining vehicle equipment time series data of vehicle equipment to be predicted; according to the vehicle equipment time series data, using the Transformer-KAN model, obtaining the prediction result of vehicle equipment consumption; wherein the Transformer-KAN model is obtained by training an improved Transformer model based on the vehicle equipment time series data set, and the improved Transformer model is obtained by replacing the MLP layer in the Transformer model with the KAN model. The present disclosure achieves the technical effect of quickly and effectively detecting vehicle equipment.

Description

Technical Field

[0001] The present disclosure relates to the technical field of vehicle equipment maintenance prediction, and in particular to a vehicle equipment material consumption prediction method, a model training method, a device, a computer-readable storage medium, and an electronic device. Background Art

[0002] Vehicle equipment maintenance is a general term for a series of technical and management activities such as inspection, commissioning, assembly, lubrication, installation, disassembly, disassembly and fault repair of vehicle equipment. These activities are aimed at maintaining the good technical status of vehicle equipment, extending its service life, and improving operating efficiency and safety.

[0003] At present, vehicle equipment maintenance includes traditional regular maintenance, post-maintenance, etc. However, traditional regular maintenance and post-maintenance often incur high costs.

[0004] Therefore, the related technology is unable to quickly and effectively detect vehicle equipment and equipment, and thus is unable to accurately and effectively perform vehicle equipment maintenance, and is unable to save costs. Summary of the invention

[0005] The main purpose of the present invention is to provide a vehicle equipment consumption prediction method, model training method, device, computer-readable storage medium and electronic device to solve the problem in related technologies that vehicle equipment cannot be detected quickly and effectively.

[0006] In order to achieve the above-mentioned object, the first aspect of the present disclosure provides a vehicle equipment consumption prediction method, comprising:

[0007] Acquire vehicle equipment time series data of vehicle equipment to be predicted, wherein the vehicle equipment time series data contains elements that affect vehicle equipment consumption, including: equipment operation time, equipment failure rate, equipment training environment and equipment management level;

[0008] Generate target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in the Transformer-KAN model;

[0009] According to the target features, a prediction result of vehicle equipment consumption is obtained through a neural network KAN model in the Transformer-KAN model;

[0010] Among them, the Transformer-KAN model is obtained by training an improved Transformer model based on a vehicle equipment time series dataset, and the improved Transformer model is obtained by replacing the multi-layer perceptron MLP layer in the Transformer model with the KAN model; the prediction result is used to indicate the degree of consumption of vehicle equipment.

[0011] Optionally, further, generating target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in a Transformer-KAN model includes:

[0012] Outputting a vector sequence containing the vehicle equipment time series data through an encoder of a Transformer model in a Transformer-KAN model according to the vehicle equipment time series data;

[0013] ‌ Based on the vehicle equipment time series data and the vector sequence, a target sequence is generated through a decoder of the Transformer model, and the target sequence includes target features.

[0014] Optionally, further, obtaining the prediction result of vehicle equipment consumption through the neural network KAN model in the Transformer-KAN model according to the target feature includes:

[0015] Inputting the target features into a KAN model, performing nonlinear transformation through at least one KAN layer, so that each KAN layer outputs a set of new features; wherein each KAN layer is composed of at least one learnable activation function, and the activation function is constructed based on a spline function or other types of functions;

[0016] A set of new features output by each KAN layer is fused to determine the prediction results of vehicle equipment consumption.

[0017] Optionally, further, the method further includes:

[0018] Determining whether fine-grained detection of the vehicle equipment is required according to the prediction result;

[0019] If fine-grained detection is not required, a corresponding prediction time is configured for the vehicle equipment, so as to continue to perform consumption prediction processing on the vehicle equipment based on the prediction time;

[0020] If fine-grained detection is required, the consumption level of the vehicle equipment is determined by performing fine-grained detection on the vehicle equipment according to the prediction result; and the maintenance strategy for the vehicle equipment is determined according to the consumption level.

[0021] Optionally, further, determining the consumption level of the vehicle equipment by performing fine-grained detection on the vehicle equipment according to the prediction result includes:

[0022] Based on the vehicle equipment, determining the detection index of the vehicle equipment by querying the mapping relationship, wherein the mapping relationship is the association relationship between the vehicle equipment, the prediction result, and the detection index, and the detection index includes at least one of the following: material testing, interior and exterior testing, safety-related testing, electronic and electrical equipment testing, and environmental protection and chemical testing;

[0023] According to the detection index of the vehicle equipment, the vehicle equipment is subjected to fine-grained detection to obtain the result of the detection index;

[0024] According to the result of the detection index, the consumption level of the vehicle equipment is determined through the matching relationship between the result of the detection index and the consumption level.

[0025] Optionally, further, the prediction result includes the consumption of the vehicle equipment or the degree of consumption of the vehicle equipment; and determining whether it is necessary to perform fine-grained detection on the vehicle equipment according to the prediction result includes:

[0026] If the consumption of the vehicle equipment is greater than the consumption threshold or the degree of consumption of the vehicle equipment is greater than the consumption threshold, it is determined that the vehicle equipment needs to be subjected to fine-grained detection;

[0027] If the consumption of the vehicle equipment is less than or equal to the consumption threshold or the consumption degree of the vehicle equipment is less than or equal to the consumption degree threshold, it is determined that there is no need to perform fine-grained detection on the vehicle equipment.

[0028] A second aspect of the present disclosure provides a model training method, comprising:

[0029] Dividing the vehicle equipment time series data set into a training set and a test set, wherein the vehicle equipment time series data set contains elements that affect vehicle equipment equipment consumption, including equipment operation time, equipment failure rate, equipment training environment, and equipment management level;

[0030] The KAN model replaces the multi-layer perceptron MLP layer in the Transformer model to obtain an improved Transformer model; wherein the KAN model includes at least one KAN layer, each KAN layer is composed of at least one learnable activation function, and the activation function is constructed based on a spline function or other types of functions;

[0031] According to the training set, the improved Transformer model is trained, and according to the test set, the trained Transformer model is tested to obtain a Transformer-KAN model;

[0032] Among them, the Transformer-KAN model is used to predict the consumption of vehicle equipment.

[0033] A third aspect of the present disclosure provides a vehicle equipment consumption prediction device, comprising:

[0034] An acquisition unit is used to acquire vehicle equipment time series data of vehicle equipment to be predicted, wherein the vehicle equipment time series data contains elements that affect vehicle equipment consumption, and the elements include: equipment operation time, equipment failure rate, equipment training environment and equipment management level;

[0035] A processing unit, configured to generate target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in a Transformer-KAN model;

[0036] The processing unit is further used to obtain a prediction result of vehicle equipment consumption through a neural network KAN model in the Transformer-KAN model according to the target feature;

[0037] Among them, the Transformer-KAN model is obtained by training an improved Transformer model based on a vehicle equipment time series dataset, and the improved Transformer model is obtained by replacing the multi-layer perceptron MLP layer in the Transformer model with the KAN model; the prediction result is used to indicate the degree of consumption of vehicle equipment.

[0038] The fourth aspect of the present disclosure provides a computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the vehicle equipment consumption prediction method provided in any one of the first aspects or the model training method provided in any one of the second aspects.

[0039] The fifth aspect of the present disclosure provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor executes the vehicle equipment consumption prediction method provided in any one of the first aspects or the model training method provided in any one of the second aspects.

[0040] The sixth aspect of the present disclosure provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the vehicle equipment consumption prediction method provided in any one of the first aspects or the model training method provided in any one of the second aspects.

[0041] In the vehicle equipment consumption prediction method provided in the embodiment of the present disclosure, the Transformer-KAN model is used to generate target features for the acquired vehicle equipment time series data of the vehicle equipment to be predicted, and then based on the target features, the vehicle equipment consumption degree is predicted, thereby realizing the rapid prediction of vehicle equipment consumption through the Transformer-KAN model, and by using the KAN model to apply the KAN model to the time series prediction model Transformer, and applying it to vehicle equipment maintenance prediction, the purpose of vehicle equipment consumption prediction is achieved, thereby achieving the technical effect of improving the accuracy and efficiency of vehicle maintenance prediction, and then saving costs, thereby solving the technical problem of not being able to quickly and effectively detect vehicle equipment, and then not being able to accurately and effectively perform vehicle equipment maintenance, and at the same time not being able to save costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0043] Figure 1 A schematic diagram of a scenario of a vehicle equipment consumption prediction method provided in an embodiment of the present disclosure;

[0044] Figure 2 A flow chart of a vehicle equipment consumption prediction method provided in an embodiment of the present disclosure;

[0045] Figure 3 A schematic diagram of a shallow model of KAN provided in an embodiment of the present disclosure;

[0046] Figure 4 A flowchart of a model training method provided in an embodiment of the present disclosure;

[0047] Figure 5 A block diagram of a vehicle equipment consumption prediction device provided in an embodiment of the present disclosure;

[0048] Figure 6 A block diagram of an electronic device provided for an embodiment of the present disclosure. DETAILED DESCRIPTION

[0049] In order to enable those skilled in the art to better understand the scheme of the present disclosure, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present disclosure.

[0050] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present disclosure described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0051] In the present disclosure, the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inner", "outer", "middle", "vertical", "horizontal", "lateral", "longitudinal" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings. These terms are mainly used to better describe the present disclosure and its embodiments, and are not used to limit the indicated devices, elements or components to have a specific orientation, or to be constructed and operated in a specific orientation.

[0052] In addition, some of the above terms may be used to express other meanings in addition to indicating orientation or positional relationship. For example, the term "on" may also be used to express a certain dependency or connection relationship in some cases. For those of ordinary skill in the art, the specific meanings of these terms in this disclosure can be understood according to specific circumstances.

[0053] In addition, the terms "installed", "set", "provided with", "connected", "connected", and "socketed" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral structure; it can be a mechanical connection, or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be an internal connection between two devices, elements, or components. For those of ordinary skill in the art, the specific meanings of the above terms in this disclosure can be understood according to specific circumstances.

[0054] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0055] At present, in the related technologies, vehicle equipment maintenance includes traditional regular maintenance, post-maintenance, etc. However, traditional regular maintenance and post-maintenance often incur high costs. Therefore, the related technologies cannot quickly and effectively detect vehicle equipment, and thus cannot accurately and effectively perform vehicle equipment maintenance, and cannot save costs.

[0056] In order to solve the above problems, the technical concept of the present invention is to use a trained model to predict the consumption of vehicle equipment and materials, which can quickly and effectively realize the consumption detection of vehicle equipment and materials, and determine whether the vehicle equipment and materials need to be repaired without regular maintenance or post-repair, so as to avoid wasting unnecessary time, labor and other maintenance costs. In order to ensure the accuracy of the prediction results, the KAN model is applied to the time series prediction model Transformer to replace the MLP layer in the Transformer, which can improve the interpretability of the model and make the prediction more accurate, thereby providing more accurate reference data for whether to perform vehicle equipment maintenance.

[0057] In practical applications, the execution subject of the present disclosure may be a prediction device for vehicle equipment consumption, which may be deployed in electronic devices, such as terminal devices, servers, etc. Users may predict vehicle equipment consumption through electronic devices deployed with the prediction device for vehicle equipment consumption, where the equipment may be various components of vehicle equipment (e.g., maintenance equipment: various components in vehicle equipment that can support maintenance, which are not specifically limited here), and the result of vehicle equipment consumption prediction may be used as reference data for predicting vehicle equipment maintenance, that is, the vehicle equipment consumption prediction may be used as an indicator to determine whether vehicle equipment needs maintenance.

[0058] For example, see Figure 1 As shown, Figure 1A schematic diagram of a scenario of a vehicle equipment consumption prediction method provided in an embodiment of the present disclosure. The scenario includes a vehicle equipment data collection device 101 and a server 102 on which the vehicle equipment consumption prediction device is deployed. The vehicle equipment data collection device is used to collect vehicle equipment data, such as multivariate data of vehicle equipment (vehicles and their various components, such as multiple equipment or each equipment in vehicle equipment) at different collection times: equipment operation time, equipment failure rate, equipment training environment and equipment management level, etc., which constitute vehicle equipment time series data of vehicle equipment, i.e., multivariate time series data. The server 102 is used to obtain the vehicle equipment time series data of vehicle equipment transmitted by the vehicle equipment data collection device, or generate the vehicle equipment time series data of vehicle equipment based on the multivariate data at different collection times transmitted by the vehicle equipment data collection device; and according to the vehicle equipment time series data of vehicle equipment, the vehicle equipment consumption is predicted by the vehicle equipment consumption prediction device to obtain a prediction result to determine the degree of vehicle equipment consumption.

[0059] Specifically, when it is necessary to predict the consumption of any vehicle equipment, an electronic device (e.g., server 102) deployed with a prediction device for vehicle equipment consumption can use the Transformer-KAN model to predict the degree of vehicle equipment consumption, such as the number of vehicle equipment consumed, the proportion of vehicle equipment consumed, etc., based on the acquired vehicle equipment time series data of the vehicle equipment to be predicted (e.g., multivariate time series data: equipment operating time, equipment failure rate, equipment training environment, and equipment management level, etc.).

[0060] Through the above application scenarios, the present invention applies the KAN model to the time series prediction model transformer, and applies it to vehicle equipment maintenance prediction. The use of KAN simplifies the knowledge network and makes it interactive and more interpretable, which greatly improves the accuracy and efficiency of vehicle equipment device consumption prediction, thereby improving the accuracy and efficiency of vehicle equipment device maintenance prediction, while saving costs.

[0061] It should be noted that in the technical solution of the present invention, the collection, storage, use, processing, transmission, provision and disclosure of information such as vehicle equipment data or user data involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0062] The technical solution of the present disclosure is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0063] The present disclosure provides a method for predicting vehicle equipment consumption. Figure 2 As shown, the method includes the following steps S201 to S203:

[0064] Step S201: Acquire vehicle equipment time series data of vehicle equipment to be predicted, wherein the vehicle equipment time series data contains elements that affect vehicle equipment consumption, including equipment operation time, equipment failure rate, equipment training environment and equipment management level.

[0065] Step S202: Generate target features based on the vehicle equipment time series data through the neural network Transformer model based on the self-attention mechanism in the Transformer-KAN model.

[0066] Step S203: According to the target features, the prediction result of vehicle equipment consumption is obtained through the neural network KAN model in the Transformer-KAN model.

[0067] Among them, the Transformer-KAN model is obtained by training an improved Transformer model based on a vehicle equipment time series dataset, and the improved Transformer model is obtained by replacing the multi-layer perceptron MLP layer in the Transformer model with the KAN model; the prediction result is used to indicate the degree of consumption of vehicle equipment.

[0068] In the disclosed embodiment, based on the acquired vehicle equipment time series data (multivariate time series data such as equipment operation time, equipment failure rate, equipment training environment and equipment management level), target features are generated through Transformer as input data of the KAN model, and then the vehicle equipment consumption is predicted through the KAN model. By applying the KAN model to the time series prediction model Transformer, it is applied to vehicle equipment maintenance prediction. Since the use of KAN simplifies the knowledge network and makes it interactive and more interpretable, the accuracy and efficiency of vehicle equipment device consumption prediction are greatly improved, thereby improving the accuracy and efficiency of vehicle equipment device maintenance prediction, while saving costs.

[0069] Among them, vehicle equipment device consumption prediction (or vehicle equipment maintenance prediction) refers to the use of advanced data analysis and prediction technology, combined with sensors, the Internet of Things, big data, artificial intelligence and other technical means to monitor and analyze the status and performance of the vehicle and its various components in real time, so as to predict possible failures or maintenance needs in advance. Its purpose is to take preventive measures before failure occurs, ensure the normal operation and safety of the vehicle, reduce unplanned downtime, improve maintenance efficiency, and reduce maintenance costs.

[0070] Transformer: is a deep learning model architecture based on the attention mechanism, which can be used for time series prediction. The Transformer model performs particularly well in sequence-to-sequence tasks such as machine translation, text generation, and summary generation. Key components: 1. Attention mechanism: Self-attention, which calculates the relationship between each position in the input sequence and other positions to capture long-distance dependencies; multi-head attention, which parallelizes the input sequence through multiple attention heads, and then concatenates and linearly transforms the results to enhance the expressiveness of the model. 2. Encoder-decoder structure: Encoder, which is composed of multiple identical layers stacked together, each layer contains a multi-head self-attention sublayer and a feedforward neural network sublayer. The encoder receives the input sequence and generates a hidden representation; the decoder, which is composed of multiple identical layers stacked together, each layer contains a multi-head self-attention sublayer, an encoder-decoder attention sublayer, and a feedforward neural network sublayer. The decoder receives the output of the encoder and the previously generated target sequence to generate the final output sequence.

[0071] MLP: Multilayer Perceptron is a classic feedforward artificial neural network widely used in various machine learning tasks. MLP consists of multiple layers, including an input layer, one or more hidden layers, and an output layer. Each layer consists of several neurons (also called nodes or units), which are connected by weights to transform the input data and extract features. Key components: 1. Input layer, which accepts input data, and each neuron corresponds to an input feature. The input layer does not perform any calculations, but simply passes the data to the hidden layer. 2. Hidden layer, which consists of one or more layers, each containing several neurons. The hidden layer performs nonlinear transformation and feature extraction on the input data through weight matrices and activation functions; commonly used activation functions include linear rectification function or rectified linear unit (Rectified Linear Unit, ReLU), S-shaped growth curve function (Sigmoid), Hyperbolic Tangent Function (HyperbolicTangent Function, Tanh), etc. 3. Output layer, which generates the final output result. The number of neurons in the output layer depends on the specific task (such as the number of categories in classification tasks or the number of targets in regression tasks).

[0072] Optionally, a vehicle equipment consumption prediction method includes:

[0073] Acquire vehicle equipment time series data of vehicle equipment to be predicted; wherein the vehicle equipment time series data contains elements that affect vehicle equipment consumption, and the elements include: equipment operation time, equipment failure rate, equipment training environment and equipment management level;

[0074] Based on the vehicle equipment time series data, the prediction results of vehicle equipment consumption are obtained through the Transformer-KAN model;

[0075] Among them, the Transformer-KAN model is obtained by training an improved Transformer model based on a vehicle equipment time series dataset, and the improved Transformer model is obtained by replacing the MLP layer in the Transformer model with the KAN model; the prediction result is used to indicate the degree of consumption of vehicle equipment.

[0076] Optionally, the Transformer-KAN model includes an encoder, a decoder, and a KAN model of a Transformer model; prediction based on the encoder, the decoder, and the KAN model of the Transformer model can be performed in at least the following two ways:

[0077] Method 1: The encoder and decoder of the Transformer model obtain the target features, and then based on the target features, the weights are nonlinearly transformed through the KAN mechanism through a learnable activation function to enhance the expression ability of the model and obtain accurate prediction results.

[0078] Optionally, generating target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in a Transformer-KAN model includes:

[0079] Outputting a vector sequence containing the vehicle equipment time series data through an encoder of a Transformer model in a Transformer-KAN model according to the vehicle equipment time series data;

[0080] ‌ Based on the vehicle equipment time series data and the vector sequence, a target sequence is generated through a decoder of the Transformer model, and the target sequence includes target features.

[0081] In the disclosed embodiment, the vehicle equipment time series data is input into the Transformer-KAN model. The vehicle equipment time series data is first converted into a continuous vector representation with a fixed dimension, namely, word embedding, through an embedding layer, and then positionally encoded and added to the word embedding as the input of the encoder of the Transformer model.

[0082] The self-attention mechanism in the encoder calculates the correlation score (attention weight) between each word and other words to perform weighted averaging on the input sequence to obtain a new representation for each word. If the encoder contains a multi-head attention mechanism, the encoder divides the input sequence into multiple heads (usually 8 or 16). Each head performs self-attention calculations independently and learns different attention modes. The outputs of each head are concatenated and the final self-attention layer output is obtained through linear transformation.

[0083] Likewise, a masked multi-head attention mechanism is used through the decoder’s self-attention layer to ensure that the model can only see the words at and before the current position when generating a sequence, allowing the decoder to see the encoder’s output in order to better generate the target sequence.

[0084] Optionally, obtaining the prediction result of vehicle equipment consumption by using the neural network KAN model in the Transformer-KAN model according to the target feature includes:

[0085] Inputting the target features into a KAN model, performing nonlinear transformation through at least one KAN layer, so that each KAN layer outputs a set of new features; wherein each KAN layer is composed of at least one learnable activation function, and the activation function is constructed based on a spline function or other types of functions;

[0086] A set of new features output by each KAN layer is fused to determine the prediction results of vehicle equipment consumption.

[0087] In the disclosed embodiment, the KAN model may be composed of at least one KAN layer. Based on the target features, the KAN mechanism uses a learnable activation function to perform nonlinear transformation on the weights to enhance the expression ability of the model, thereby obtaining a prediction result.

[0088] Method 2: In the improved Transformer model, the KAN model or KAN layer can be used as a sublayer of the encoder and / or decoder in the Transformer model.

[0089] Optionally, according to the vehicle equipment time series data, a prediction result is obtained by using a Transformer-KAN model, including:

[0090] Generate a first feature through an encoder of a Transformer model in a Transformer-KAN model according to the vehicle equipment time series data;

[0091] According to the KAN layer in the encoder, a nonlinear transformation is performed to obtain new features;

[0092] According to the new feature and the first feature, a prediction result is determined through a decoder of the Transformer model in the Transformer-KAN model.

[0093] Optionally, according to the vehicle equipment time series data, a prediction result is obtained by using a Transformer-KAN model, including:

[0094] Generate a first feature through an encoder of a Transformer model in a Transformer-KAN model according to the vehicle equipment time series data;

[0095] Determine a second feature based on the first feature through a decoder of a Transformer model in the Transformer-KAN model;

[0096] According to the KAN layer in the decoder, a nonlinear transformation is performed to obtain a prediction result.

[0097] Optionally, according to the vehicle equipment time series data, a prediction result is obtained by using a Transformer-KAN model, including:

[0098] Generate a first feature through an encoder of a Transformer model in a Transformer-KAN model according to the vehicle equipment time series data;

[0099] According to the first feature, a nonlinear transformation is performed through a KAN layer in the encoder to obtain a new feature;

[0100] Obtain a second feature through a decoder of a Transformer model in a Transformer-KAN model according to the new feature and the first feature;

[0101] According to the second feature, a nonlinear transformation is performed through the KAN layer in the decoder to obtain a prediction result.

[0102] In the disclosed embodiment, the self-attention mechanism in the encoder calculates the correlation score (attention weight) between each word and other words to perform weighted averaging on the input sequence, thereby obtaining a new representation of each word. If the encoder includes a multi-head attention mechanism, the encoder divides the input sequence into multiple heads (usually 8 or 16 heads), each head performs self-attention calculations independently, and learns different attention modes, concatenates the outputs of each head, and obtains the final self-attention layer output through linear transformation.

[0103] If the KAN layer replaces the feedforward neural network of the encoder, the weights can be nonlinearly transformed through the KAN mechanism through a learnable activation function to enhance the expressiveness of the model. The feature vector processed by the encoder is passed to the decoder to obtain the prediction result.

[0104] Similarly, if the KAN layer replaces the feedforward neural network of the decoder, the processing process is similar to the above-mentioned KAN layer replacing the feedforward neural network of the encoder, and will not be repeated here.

[0105] Optionally, the method further comprises:

[0106] Determining whether fine-grained detection of the vehicle equipment is required according to the prediction result;

[0107] If fine-grained detection is not required, a corresponding prediction time is configured for the vehicle equipment, so as to continue to perform consumption prediction processing on the vehicle equipment based on the prediction time;

[0108] If fine-grained detection is required, the consumption level of the vehicle equipment is determined by performing fine-grained detection on the vehicle equipment according to the prediction result; and the maintenance strategy for the vehicle equipment is determined according to the consumption level.

[0109] In the disclosed embodiment, the purpose of the prediction can be to roughly calculate the degree of consumption of vehicle equipment and equipment, and then determine the availability of the vehicle equipment. If the value of the prediction result is considerable, that is, it reaches the expected value, there is no need for further testing, and only the time for the next prediction needs to be set, thereby saving testing or maintenance costs; if the value of the prediction result is not considerable, that is, it does not reach the expected value, then in order to clearly analyze which components have problems or where the problems occur, and thereby ensure the safe use of vehicle equipment and reduce the occurrence of safety hazards, it is necessary to further test the vehicle equipment components. Through further testing, the consumption level of the vehicle equipment and equipment can be determined, and then the maintenance strategy for the vehicle equipment and equipment can be determined.

[0110] Optionally, determining the consumption level of the vehicle equipment by performing fine-grained detection on the vehicle equipment according to the prediction result includes:

[0111] Based on the vehicle equipment, determining the detection index of the vehicle equipment by querying the mapping relationship, wherein the mapping relationship is the association relationship between the vehicle equipment, the prediction result, and the detection index, and the detection index includes at least one of the following: material testing, interior and exterior testing, safety-related testing, electronic and electrical equipment testing, and environmental protection and chemical testing;

[0112] According to the detection index of the vehicle equipment, the vehicle equipment is subjected to fine-grained detection to obtain the result of the detection index;

[0113] According to the result of the detection index, the consumption level of the vehicle equipment is determined through the matching relationship between the result of the detection index and the consumption level.

[0114] In the disclosed embodiment, before determining to perform fine-grained detection, it is first determined what needs to be detected, that is, the detection index or detection item; wherein different prediction results correspond to different detection indicators, and based on the results detected by the detection indicators, the consumption level of the vehicle equipment is determined by finding the relationship between the results of the detection indicators and the consumption level.

[0115] In addition, there is another way to determine the consumption level: there is an association relationship or mapping table between the prediction result and the consumption level, and the consumption level is determined directly based on the prediction result, and then the processing strategy is determined.

[0116] Optionally, the prediction result includes the consumption of the vehicle equipment or the degree of consumption of the vehicle equipment; and determining whether it is necessary to perform fine-grained detection on the vehicle equipment according to the prediction result includes:

[0117] If the consumption of the vehicle equipment is greater than the consumption threshold or the degree of consumption of the vehicle equipment is greater than the consumption threshold, it is determined that the vehicle equipment needs to be subjected to fine-grained detection;

[0118] If the consumption of the vehicle equipment is less than or equal to the consumption threshold or the consumption degree of the vehicle equipment is less than or equal to the consumption degree threshold, it is determined that there is no need to perform fine-grained detection on the vehicle equipment.

[0119] The vehicle equipment consumption (or consumption value, the higher the predicted consumption value, the greater the need for maintenance or the greater the possibility of maintenance) here can refer to the number of vehicle equipment consumed; the degree of vehicle equipment consumption can be determined by the vehicle equipment consumption. The degree of vehicle equipment consumption can be represented by the ratio of the number of vehicle equipment consumed to the total number of vehicle equipment. No specific limitation is made here.

[0120] In the disclosed embodiment, whether fine-grained detection of vehicle equipment is required is determined by comparing the detection result with a set threshold. If the detection result is greater than the threshold, fine-grained detection is not required; if the detection result is less than the threshold, fine-grained detection is required.

[0121] From the above description, it can be seen that the present disclosure achieves the following technical effects: the KAN model is applied to the time series prediction model transformer, and applied to the vehicle equipment maintenance prediction. The use of KAN simplifies the knowledge network and makes it interactive and more interpretable, which greatly improves the accuracy and efficiency of the vehicle equipment device consumption prediction, thereby improving the accuracy and efficiency of the vehicle equipment device maintenance prediction, while saving costs.

[0122] The present disclosure also provides a model training method, the model training method comprising:

[0123] Dividing the vehicle equipment time series data set into a training set and a test set, wherein the vehicle equipment time series data set contains elements that affect vehicle equipment equipment consumption, including equipment operation time, equipment failure rate, equipment training environment, and equipment management level;

[0124] The KAN model replaces the multi-layer perceptron MLP layer in the Transformer model to obtain an improved Transformer model; wherein the KAN model includes at least one KAN layer, each KAN layer is composed of at least one learnable activation function, and the activation function is constructed based on a spline function or other types of functions;

[0125] According to the training set, the improved Transformer model is trained, and according to the test set, the trained Transformer model is tested to obtain a Transformer-KAN model;

[0126] Among them, the Transformer-KAN model is used for vehicle equipment consumption prediction.

[0127] In the disclosed embodiment, due to the shortcomings of MLP in Transformer (i.e., in Transformer, MLPs consume almost all non-embedded parameters, and in the absence of post-analysis tools, their interpretability (relative to the attention layer) is generally poor. MLP modules generally contain a large number of fully connected layers and nonlinear activation functions, and have high computational complexity, especially when processing large-scale data. The amount of computation in each layer is proportional to the dimensions of the input and output, which results in consumption of computing resources when processing high-dimensional data. The fully connected layers in the MLP module introduce a large number of parameters. For large Transformer models, the number of these parameters is very large, requiring a large amount of storage space and computing resources for training and inference), KAN is used to replace MLP, which simplifies the knowledge network and makes it interactive, thereby improving the interpretability of the model and making predictions more accurate.

[0128] Specifically, the Transformer model first converts the input data (such as training sets, test sets, etc.) into word embedding vectors, and then the encoder of the Transformer model is responsible for mapping the input sequence (here refers to the word embedding vector based on the time series) to a continuous representation space. The KAN model may perform further nonlinear transformation and feature extraction on this continuous representation space. The decoder of the Transformer model generates the target sequence based on the output of the encoder (which may also include the output of the KAN model) to obtain the prediction result. Therefore, the Transformer-KAN model achieves efficient processing of input data and high-quality output by combining the powerful representation ability of the Transformer model and the nonlinear transformation ability of the KAN model.

[0129] For example, see Figure 4 As shown, the vehicle equipment time series data set is divided into a training set and a prediction set (or a validation set or a test set, which is not specifically limited here) according to a certain ratio. For example, if there are 240 multivariate time series, the first 180 are taken as the training set and the last 60 are taken as the prediction set.

[0130] The vehicle equipment time series is a multivariate time series, which mainly includes factors affecting the consumption of vehicle equipment, including operating time, equipment (parts) failure rate, training environment (for example, average temperature) and equipment (parts) management level. The expression form is shown in Table 1 below.

[0131] Table 1 Equipment consumption data table

[0132]

[0133] Based on the training set and the prediction set, the model that is the fusion of the Transformer model and the KAN model (for example, the KAN model replaces the MLP layer in the Transformer model), that is, the improved Transformer model, is trained and evaluated to obtain a trained model, that is, the Transformer-KAN model.

[0134] Specifically, taking the training set as an example, the input data is first preprocessed, such as word segmentation, word embedding, etc., to obtain a series of vector representations. These vector representations are input into the encoder of the Transformer model for self-attention calculation and feature extraction. The KAN mechanism is introduced at a certain stage of the Transformer model (such as a layer of the encoder or decoder). The KAN mechanism performs a nonlinear transformation on the weights through a learnable activation function to enhance the expressive power of the model. The feature vector processed by the encoder of the Transformer model is passed to the decoder of the Transformer model. The decoder predicts the next word based on the output of the encoder and the prefix of the target sequence. Finally, the decoder of the Transformer model generates a sequence of representation vectors, each of which corresponds to a position in the target sequence. These vectors are processed through a linear layer and a softmax function to generate a probability distribution of the vocabulary size. Words are selected from the probability distribution using strategies such as greedy search or beam search to generate the final output sequence. Then the predicted value is obtained to represent the prediction result. The training process for the KAN model is as follows: first, the time series is decomposed based on the preprocessed features to determine whether it is a combination of a single variable and a two-step addition; if so, the residual activation function is determined to obtain the activation value, and then each grid is updated in time according to the input activation value to update the spline network, and then the prediction value is output through network expansion and simplification.

[0135] Among them, the architecture of a neural network model KAN (Kolmogorov-Arnold) is:

[0136] First, let me introduce the Kolmogorov-Arnold representation theorem:

[0137] if is a multivariate continuous function on a bounded domain, then can be written as a finite combination of continuous functions of a single variable and binary addition operations. More specifically, for smooth ,have

[0138] (2.1)

[0139] in, and .

[0140] Representation vector No. p elements, so p The range is from 1 to n (n is the dimension of the input vector).

[0141] q This index is used to traverse the external function Each component of (or univariate function) processes the input vector No. p components, and for q The sum of the external functions contributes one term.

[0142] The theorem states that we can use 2n+1 such external functions—each external function is a unary function (it acts on the inner unary function ), to represent any multivariable function .

[0143] Suppose there is an input-output pair The supervised learning task consists of finding f so that all data points Formula (2.1) means that if we can find a suitable single variable function and , then the task is completed. Therefore, this study designs a neural network and explicitly parameterizes formula (2.1). Since all functions to be learned are single-variable functions, each one-dimensional function can be parameterized as a B-spline curve, and its learnable coefficients are local B-spline basis functions. Now, with the prototype of KAN, its computational graph fully meets the requirements of formula (2.1), as shown in Figure 3 As shown in Figure 1 (input dimension n=2), it is a two-layer neural network with activation functions located on the edges instead of nodes (simple summation is performed on the nodes), and the width of the middle layer is 2n+1. Figure 3 The shallow model of KAN is shown.

[0144] In order to make it possible to arbitrarily approximate any function with smoothing splines in practice, KANs are generalized to make them wider and deeper.

[0145] One with Dimensional input and The KAN layer with a dimensional output can be defined as a one-dimensional function matrix

[0146] (2.2)

[0147] The function With trainable parameters. In the Kolmogov-Arnold theory, the inner functions form a KAN layer, where , ; The outer functions form a KAN layer, where , . Therefore, the Kolmogov-Arnold representation in (2.1) is just a combination of two KAN layers. Now the meaning of a deeper Kolmogov-Arnold representation is clear: just stack more KAN layers.

[0148] The shape of KAN is represented by an array of integers

[0149] (2.3)

[0150] in, is the number of nodes in the i-th layer of the computational graph. We use Indicates The first neurons, with express The activation value of the neuron. Layer and Between the layers, there are Activation function: concatenation The activation function is expressed as

[0151] (2.4)

[0152] The previous activation is simply expressed as , Post-activation in express. The activation value of a neuron is simply the activation after all inputs: (2.5)

[0153] It is expressed in matrix form as

[0154] (2.6)

[0155] in, It is with The function matrix corresponding to each KAN layer. The general KAN network consists of L layers: given an input vector , the output of KAN is:

[0156] (2.7)

[0157] The above formula can also be rewritten to make it more similar to formula (2.1), assuming that the output dimension is , and define :

[0158] (2.8)

[0159] Specifically:

[0160] (1) Residual activation function. It contains a basis function b(x) (similar to residual connection), so the activation function It is the sum of the basis function b(x) and the spline function:

[0161] (2.9)

[0162] set up

[0163] (2.10)

[0164] In most cases, spline(x) is parameterized as a linear combination of B-spline curves, i.e.

[0165] (2.11)

[0166] in It is trainable.

[0167] (2) Initialization scale. The initialization value of each activation function is , . It is initialized according to the Xavier initialization method.

[0168] (3) Update the spline grid. Update each grid in time according to the input activation value to solve the problem that the spline is defined on a bounded region, but the activation value may leave the fixed region during training.

[0169] Parameter count. For example, suppose a network

[0170] (1) Depth L

[0171] (2) Each layer has equal width

[0172] (3) Each spline has an order of k (usually k=3) and runs on G intervals (G+1 grid points). So the total number of parameters. For one-dimensional problems, we can take N=L=1, and the implemented KAN network is just a spline approximation. For higher dimensions, the following theorem will be used to describe the generalization behavior of KAN.

[0173] 2.2 Approximation capability and scaling laws of KAN

[0174] Recalling formula (2.1), a 2-layer width (2n+1) representation may not be smooth. However, deeper representations may have the advantage of smoother activations. For example, the 4-variable function

[0175] (2.12)

[0176] can be represented smoothly by a 3-layer [4,2,1,1] KAN, but may not be represented by a 2-layer KAN with smooth activations. For ease of approximation, we still assume smoothness of activations, but allow representation of arbitrary width and depth, as shown in Equation (2.7). To emphasize the dependence of KAN on a finite grid of points, we use and To replace the symbols used in formulas (2.5) and (2.6) and .

[0177] Approximate theory:

[0178] set up Assume that the function f(x) is represented as:

[0179] (2.13)

[0180] As shown in formula (2.7), each are all (k+1) times continuously differentiable. Then there exists a constant C that depends on f and its representation, which gives the following approximation constraints on the grid size G: There exists a k-order B-spline function Make for , all have boundaries

[0181] (2.14)

[0182] Here, the The sign of (normal) is used to measure the size of derivatives below the mth order.

[0183]

[0184] As long as the assumptions of the approximation theory hold, KANs with finite grid size can approximate the function well with a residual rate that is independent of the dimension, thus defeating the dimensionality limitation. This is natural, since only one-dimensional functions are approximated using splines. In particular, for m = 0, we recover The precision of the norm, which in turn provides a bound on the RMSE over a finite field, gives the scaling exponent k + 1. Of course, the constant C depends on the representation and therefore on the dimensionality.

[0185] Although the Kolmogorov-Arnold theorem in formula (2.1) corresponds to a KAN representation of shape [d,2d+1,1], its function is not necessarily smooth. On the other hand, if a smooth representation can be found (or the KAN can be made wider than the theory dictates), then approximation theory shows that the curse of dimensionality (COD) can be overcome.

[0186] 2.3 Accuracy: Grid Extension

[0187] In principle, splines can be arbitrarily accurate to the target function, because the grid can be arbitrarily fine. KANs inherit this advantage. One can first train a KAN with fewer parameters, and then scale it up to a KAN with more parameters simply by making its spline grid finer, without having to retrain the larger model from scratch.

[0188] Next, we will introduce how to perform grid expansion. The basic principle is to fit a new fine-grained spline to the old coarse-grained spline. Suppose we want to fit a one-dimensional function f in the bounded region [a, b] using a k-order B-spline. The coarse-grained grid with a spacing of , which is expanded to .have B-spline basis functions, the i-th B-spline Only in Then f on the coarse grid can be expressed as a linear combination of these B-spline basis functions Given a The finer grid of the interval, f on the fine grid is related .parameter From the parameters Initialize in, by minimizing and The distance between (in some distribution X):

[0189] (2.15)

[0190] This can be achieved by the least squares method. All splines in KAN are meshed independently.

[0191] 2.4 For interpretability: simplifying knowledge networks and making them interactive

[0192] If we know that the data set is represented by a symbolic formula If we generate a [2, 1, 1] KAN, we know that it can express this function. However, in practice, this information is not known a priori, so it would be nice to have a way to automatically determine this shape. We start with a large enough KAN, train it with sparse regularization, and then prune it. These pruned KANs are more interpretable than unpruned ones. To maximize the interpretability of the KAN, we propose some simplification techniques in Section 2.4.1, and give examples of how to interact with the KAN to improve its interpretability in Section 2.4.2.

[0193] 2.4.1 Simplification technology

[0194] 1. Sparsification. For MLP, L1 regularization of linear weights is beneficial for sparsity. KANS can adopt this high-level idea, but two modifications are required:

[0195] (1) There are no linear “weights” in KAN. Linear weights are replaced by learnable activation functions, so we should define the L1 norm of these activation functions.

[0196] (2) We found that L1 is not enough for the sparsification of KANs and additional entropy regularization is required. We use the activation function The L1 norm of The average value of the inputs,

[0197] (2.16)

[0198] So, for Input and KAN layer with output , we define The L1 criterion of is the sum of the L1 criteria of all activation functions, that is,

[0199] (2.17)

[0200] In addition, we will The entropy is defined as

[0201] (2.18)

[0202] Overall training goal is the predicted loss Plus L1 and entropy regularization for all KAN layers:

[0203] (2.19)

[0204] in , is the relative magnitude, usually set to , Controls the overall regularization magnitude.

[0205] 2. Visualization. When visualizing KAN, in order to get a sense of magnitude, we use The transparency is set to Proportional, where Therefore, functions with smaller amplitudes are faded out so that the important functions can be focused on.

[0206] 3. Pruning. After training with a sparsification penalty, you may also want to prune the network into smaller sub-networks. KAN is sparsified at the node level (not at the edge level). For each node (e.g., the i-th neuron in the i-th layer), its incoming and outgoing scores are defined as

[0207] (2.20)

[0208] By default, if both the incoming and outgoing scores are greater than the threshold hyperparameter , then the node is considered important. All unimportant neurons will be pruned.

[0209] 4. Symbolization. When it is suspected that some activation functions (such as cos or log) are actually in symbolic form, an interface is provided to set them to the specified symbolic form. You can The activation function is set to f. However, the activation function cannot be simply set to an exact symbolic formula because its input and output may be offset and scaled. Therefore, the pre-activation x and post-activation y are taken from the sample and the affine parameters (a, b, c, d) are fitted so that y ≈ cf(ax+b)+d.

[0210] The present disclosure overcomes the problem that models of different sizes need to be trained independently. For KAN, one can first train a model with fewer parameters and then expand it to a model with more parameters by simply making its spline grid finer, without having to retrain a larger model from scratch. Using KAN simplifies the knowledge network and makes it interactive and more interpretable. Therefore, the KAN model is applied to the time series prediction model transformer and applied to vehicle equipment maintenance prediction, which greatly improves the accuracy and efficiency of vehicle maintenance prediction.

[0211] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0212] The present disclosure also provides a vehicle equipment consumption prediction device for implementing the above-mentioned vehicle equipment consumption prediction method embodiment, such as Figure 5 As shown, the vehicle equipment consumption prediction device 50 includes:

[0213] The acquisition unit 501 is used to acquire vehicle equipment time series data of vehicle equipment to be predicted, wherein the vehicle equipment time series data contains elements that affect vehicle equipment consumption, including equipment operation time, equipment failure rate, equipment training environment and equipment management level;

[0214] A processing unit 502 is used to generate target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in a Transformer-KAN model;

[0215] The processing unit 502 is further used to obtain a prediction result of vehicle equipment consumption through a neural network KAN model in the Transformer-KAN model according to the target feature;

[0216] Among them, the Transformer-KAN model is obtained by training an improved Transformer model based on a vehicle equipment time series dataset, and the improved Transformer model is obtained by replacing the multi-layer perceptron MLP layer in the Transformer model with the KAN model; the prediction result is used to indicate the degree of consumption of vehicle equipment.

[0217] Optionally, the processing unit 502, when executing the process of generating target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in a Transformer-KAN model, specifically includes:

[0218] Outputting a vector sequence containing the vehicle equipment time series data through an encoder of a Transformer model in a Transformer-KAN model according to the vehicle equipment time series data;

[0219] ‌ Based on the vehicle equipment time series data and the vector sequence, a target sequence is generated through a decoder of the Transformer model, and the target sequence includes target features.

[0220] Optionally, the processing unit 502, when executing the neural network KAN model in the Transformer-KAN model according to the target feature to obtain the prediction result of vehicle equipment consumption, specifically includes:

[0221] Inputting the target features into a KAN model, performing nonlinear transformation through at least one KAN layer, so that each KAN layer outputs a set of new features; wherein each KAN layer is composed of at least one learnable activation function, and the activation function is constructed based on a spline function or other types of functions;

[0222] A set of new features output by each KAN layer is fused to determine the prediction results of vehicle equipment consumption.

[0223] Optionally, the vehicle equipment consumption prediction device is further used to perform the following operations:

[0224] Determining whether fine-grained detection of the vehicle equipment is required according to the prediction result;

[0225] If fine-grained detection is not required, a corresponding prediction time is configured for the vehicle equipment, so as to continue to perform consumption prediction processing on the vehicle equipment based on the prediction time;

[0226] If fine-grained detection is required, the consumption level of the vehicle equipment is determined by performing fine-grained detection on the vehicle equipment according to the prediction result; and the maintenance strategy for the vehicle equipment is determined according to the consumption level.

[0227] Optionally, the vehicle equipment consumption prediction device, when performing fine-grained detection on the vehicle equipment according to the prediction result to determine the consumption level of the vehicle equipment, specifically includes:

[0228] Based on the vehicle equipment, determining the detection index of the vehicle equipment by querying the mapping relationship, wherein the mapping relationship is the association relationship between the vehicle equipment, the prediction result, and the detection index, and the detection index includes at least one of the following: material testing, interior and exterior testing, safety-related testing, electronic and electrical equipment testing, and environmental protection and chemical testing;

[0229] According to the detection index of the vehicle equipment, the vehicle equipment is subjected to fine-grained detection to obtain the result of the detection index;

[0230] According to the result of the detection index, the consumption level of the vehicle equipment is determined through the matching relationship between the result of the detection index and the consumption level.

[0231] Optionally, the prediction result includes the consumption of the vehicle equipment or the degree of consumption of the vehicle equipment; when the vehicle equipment consumption prediction device determines whether it is necessary to perform fine-grained detection on the vehicle equipment according to the prediction result, it specifically includes:

[0232] If the consumption of the vehicle equipment is greater than the consumption threshold or the degree of consumption of the vehicle equipment is greater than the consumption threshold, it is determined that the vehicle equipment needs to be subjected to fine-grained detection;

[0233] If the consumption of the vehicle equipment is less than or equal to the consumption threshold or the consumption degree of the vehicle equipment is less than or equal to the consumption degree threshold, it is determined that there is no need to perform fine-grained detection on the vehicle equipment.

[0234] The present disclosure also provides a model training device for implementing the above-mentioned model training method embodiment, including:

[0235] A division unit is used to divide the vehicle equipment time series data set into a training set and a test set, wherein the vehicle equipment time series data set contains elements that affect vehicle equipment equipment consumption, and the elements include: equipment operation time, equipment failure rate, equipment training environment and equipment management level;

[0236] A model building unit, used for replacing the multi-layer perceptron MLP layer in the Transformer model with a KAN model to obtain an improved Transformer model; wherein the KAN model includes at least one KAN layer, each KAN layer is composed of at least one learnable activation function, and the activation function is built based on a spline function or other types of functions;

[0237] A model training unit, used to train the improved Transformer model according to the training set, and test the trained Transformer model according to the test set to obtain a Transformer-KAN model;

[0238] Among them, the Transformer-KAN model is used for vehicle equipment consumption prediction.

[0239] The specific manner in which each unit in the above device embodiment performs operations has been described in detail in the embodiment of the method, and will not be elaborated on here.

[0240] The present disclosure also provides an electronic device, such as Figure 6 As shown, the electronic device includes one or more processors 61 and a memory 62. Figure 6 A processor 61 is taken as an example.

[0241] The controller may further include: an input device 63 and an output device 64 .

[0242] The processor 61, the memory 62, the input device 63 and the output device 64 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0243] The processor 61 may be a central processing unit (CPU), or other general-purpose processors, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips. The general-purpose processor may be a microprocessor or any conventional processor.

[0244] The memory 62 is a non-transient computer-readable storage medium that can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the control method in the embodiment of the present disclosure. The processor 61 executes various functional applications and data processing of the server by running the non-transient software programs, instructions and modules stored in the memory 62, that is, the vehicle equipment consumption prediction method of the above method embodiment is realized.

[0245] The memory 62 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of a processing device operated by the server, etc. In addition, the memory 62 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 62 may optionally include a memory remotely arranged relative to the processor 61, and these remote memories may be connected to a network connection device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0246] The input device 63 can receive input digital or character information and generate key signal input related to user settings and function control of the processing device of the server. The output device 64 can include display devices such as a display screen.

[0247] One or more modules are stored in the memory 62 , and when executed by one or more processors 61 , the method as described above is performed.

[0248] The embodiment of the present disclosure also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the vehicle equipment consumption prediction method or model training method as described above.

[0249] The disclosed embodiments also provide a computer program product, including a computer program, which, when executed by a processor, implements the vehicle equipment consumption prediction method or model training method as described above.

[0250] Those skilled in the art can understand that the implementation of all or part of the processes in the above method embodiments can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes in the above method embodiments. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (FM), a hard disk (HDD), or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above types of memory.

[0251] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A vehicle equipment consumption prediction method, characterized in that: include: Acquire vehicle equipment time series data of vehicle equipment to be predicted, wherein the vehicle equipment time series data contains elements that affect vehicle equipment consumption, including equipment operation time, equipment failure rate, equipment training environment and equipment management level; the vehicle equipment refers to various components in the vehicle equipment; Generate target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in the Transformer-KAN model; According to the target features, a prediction result of vehicle equipment consumption is obtained through a neural network KAN model in the Transformer-KAN model; The Transformer-KAN model is obtained by training an improved Transformer model based on a vehicle equipment time series dataset, and the Transformer-KAN model includes an encoder, a decoder and a KAN model of the Transformer model; or, the Transformer-KAN model includes an encoder and a decoder of the Transformer model, and the KAN model is used as a sublayer of the encoder and / or decoder in the Transformer model; the prediction result is used to indicate the degree of consumption of vehicle equipment; The method further comprises: Determining whether fine-grained detection of the vehicle equipment is required according to the prediction result; If fine-grained detection is not required, a corresponding prediction time is configured for the vehicle equipment, so as to continue to perform consumption prediction processing on the vehicle equipment based on the prediction time; If fine-grained detection is required, the consumption level of the vehicle equipment is determined by performing fine-grained detection on the vehicle equipment according to the prediction result; and the maintenance strategy for the vehicle equipment is determined according to the consumption level.

2. The method according to claim 1, characterized in that The target features are generated according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in the Transformer-KAN model, including: Outputting a vector sequence containing the vehicle equipment time series data through an encoder of a Transformer model in a Transformer-KAN model according to the vehicle equipment time series data; ‌ Based on the vehicle equipment time series data and the vector sequence, a target sequence is generated through a decoder of the Transformer model, and the target sequence includes target features.

3. The method according to claim 1, characterized in that The method of obtaining the prediction result of vehicle equipment consumption according to the target feature through the neural network KAN model in the Transformer-KAN model includes: Inputting the target features into a KAN model, performing nonlinear transformation through at least one KAN layer, so that each KAN layer outputs a set of new features; wherein each KAN layer is composed of at least one learnable activation function, and the activation function is constructed based on a spline function or other types of functions; A set of new features output by each KAN layer is fused to determine the prediction results of vehicle equipment consumption.

4. The method according to any one of claims 1 to 3, characterized in that: The step of determining the consumption level of the vehicle equipment by performing fine-grained detection on the vehicle equipment according to the prediction result includes: Based on the vehicle equipment, determining the detection index of the vehicle equipment by querying the mapping relationship, wherein the mapping relationship is the association relationship between the vehicle equipment, the prediction result, and the detection index, and the detection index includes at least one of the following: material testing, interior and exterior testing, safety-related testing, electronic and electrical equipment testing, and environmental protection and chemical testing; According to the detection index of the vehicle equipment, the vehicle equipment is subjected to fine-grained detection to obtain the result of the detection index; According to the result of the detection index, the consumption level of the vehicle equipment is determined through the matching relationship between the result of the detection index and the consumption level.

5. The method according to any one of claims 1 to 3, characterized in that: The prediction result includes the consumption of the vehicle equipment or the degree of consumption of the vehicle equipment; and determining whether it is necessary to perform fine-grained detection on the vehicle equipment according to the prediction result includes: If the consumption of the vehicle equipment is greater than the consumption threshold or the degree of consumption of the vehicle equipment is greater than the consumption threshold, it is determined that the vehicle equipment needs to be subjected to fine-grained detection; If the consumption of the vehicle equipment is less than or equal to the consumption threshold or the consumption degree of the vehicle equipment is less than or equal to the consumption degree threshold, it is determined that there is no need to perform fine-grained detection on the vehicle equipment.

6. A model training method, characterized in that: The method further comprises: The vehicle equipment time series data set is divided into a training set and a test set. The vehicle equipment time series data set contains elements that affect vehicle equipment consumption, including equipment operation time, equipment failure rate, equipment training environment and equipment management level; the vehicle equipment refers to various components in the vehicle equipment; The KAN model replaces the multi-layer perceptron MLP layer in the Transformer model to obtain an improved Transformer model; wherein the KAN model includes at least one KAN layer, each KAN layer is composed of at least one learnable activation function, and the activation function is constructed based on a spline function or other types of functions; According to the training set, the improved Transformer model is trained, and according to the test set, the trained Transformer model is tested to obtain a Transformer-KAN model, wherein the Transformer-KAN model includes an encoder, a decoder and a KAN model of the Transformer model; or, the Transformer-KAN model includes an encoder and a decoder of the Transformer model, and the KAN model is used as a sublayer of the encoder and / or decoder in the Transformer model; The Transformer-KAN model is used for vehicle equipment consumption prediction; After obtaining the prediction result, the method further includes: Determining whether fine-grained detection of the vehicle equipment is required according to the prediction result; If fine-grained detection is not required, a corresponding prediction time is configured for the vehicle equipment, so as to continue to perform consumption prediction processing on the vehicle equipment based on the prediction time; If fine-grained detection is required, the consumption level of the vehicle equipment is determined by performing fine-grained detection on the vehicle equipment according to the prediction result; and the maintenance strategy for the vehicle equipment is determined according to the consumption level.

7. A vehicle equipment consumption prediction device, characterized in that: include: An acquisition unit is used to acquire vehicle equipment time series data of vehicle equipment to be predicted, wherein the vehicle equipment time series data contains elements that affect vehicle equipment consumption, including equipment operation time, equipment failure rate, equipment training environment and equipment management level; the vehicle equipment refers to various components in the vehicle equipment; A processing unit, configured to generate target features according to the vehicle equipment time series data through a neural network Transformer model based on a self-attention mechanism in a Transformer-KAN model; The processing unit is further used to obtain a prediction result of vehicle equipment consumption through a neural network KAN model in the Transformer-KAN model according to the target feature; The Transformer-KAN model is obtained by training an improved Transformer model based on a vehicle equipment time series dataset, and the Transformer-KAN model includes an encoder, a decoder and a KAN model of the Transformer model; or, the Transformer-KAN model includes an encoder and a decoder of the Transformer model, and the KAN model is used as a sublayer of the encoder and / or decoder in the Transformer model; the prediction result is used to indicate the degree of consumption of vehicle equipment; The vehicle equipment consumption prediction device is also used to perform the following operations: Determining whether fine-grained detection of the vehicle equipment is required according to the prediction result; If fine-grained detection is not required, a corresponding prediction time is configured for the vehicle equipment, so as to continue to perform consumption prediction processing on the vehicle equipment based on the prediction time; If fine-grained detection is required, the consumption level of the vehicle equipment is determined by performing fine-grained detection on the vehicle equipment according to the prediction result; and the maintenance strategy for the vehicle equipment is determined according to the consumption level.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the vehicle equipment consumption prediction method described in any one of claims 1 to 5 or the model training method described in claim 6.

9. An electronic device, characterized in that: The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor executes the vehicle equipment consumption prediction method described in any one of claims 1 to 5 or the model training method described in claim 6.

Images