System and method for preventive maintenance of a vehicle

Abstract

The system and method for preventive maintenance of the automobile comprises a vehicle data acquisition module, a communication transmission module, a central processing and prediction module and a user interface module. The vehicle data acquisition module is used for collecting the running data of the vehicle in real time. The communication transmission module is used for transmitting the collected vehicle data to the central processing and prediction unit. The central processing and prediction module is used for processing and analyzing the vehicle data and predicting the maintenance requirement of the vehicle. The user interface module is used for providing a visual interface and vehicle maintenance suggestions for the user, and automatically arranging the maintenance time and service according to the reservation operation of the user interface. The present application proposes an improved time sequence enhanced intelligent maintenance prediction algorithm to accurately predict the vehicle maintenance requirement, proposes an improved deep intelligent vehicle maintenance reservation algorithm to optimize the resource allocation of the vehicle maintenance reservation, determines the best maintenance reservation service time, and provides a better scheme for the system and method for preventive maintenance of the automobile.

Description

Technical Field

[0001] This invention relates to the fields of predictive modeling and deep learning, and specifically to systems and methods for preventive maintenance of automobiles. Background Technology

[0002] Predictive modeling technology is used to predict potential failures of automotive components and perform maintenance before failures occur to avoid more serious problems. It utilizes historical data to train models and predicts future failures based on vehicle usage data (such as mileage, time, and sensor data), using algorithms such as decision trees, random forests, and support vector machines. Deep neural networks (such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and long short-term memory networks (LSTMs)) are used to process complex time-series and multidimensional data for more accurate failure prediction. A vehicle fault tree model is established, connecting the failures of various vehicle components through logical relationships and predicting possible failure paths. By combining these predictive modeling techniques, automotive preventative maintenance systems can significantly improve vehicle operational reliability and reduce failure rates and maintenance costs.

[0003] Deep learning technology utilizes deep learning models to predict and diagnose potential vehicle faults and develop corresponding maintenance plans based on these predictions. It uses recurrent neural networks to process and predict time-series data to capture the time dependencies in vehicle sensor data, and automatic encoders to detect abnormal patterns in vehicle data to help identify potential faults. It uses advanced machine learning and deep learning algorithms and models to analyze vehicle data, perform fault prediction, classification, and regression analysis, optimize vehicle maintenance strategies, and continuously improve the decision-making process through experimentation and feedback. Through the combination of these deep learning technologies, automotive preventive maintenance systems can more accurately predict vehicle faults, optimize maintenance plans, improve vehicle operational reliability and safety, and reduce unexpected breakdowns and downtime. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide a system and method for preventive maintenance of automobiles.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a system and method for preventive maintenance of automobiles, comprising a vehicle data acquisition module, a communication transmission module, a central processing and prediction module, and a user interface module. The vehicle data acquisition module is used to acquire vehicle operating data in real time. The communication transmission module is used to transmit the acquired vehicle data to the central processing and prediction module. The central processing and prediction module includes a data analysis unit and a maintenance demand prediction unit. The data analysis unit is used to process and analyze the vehicle data. The maintenance demand prediction unit proposes an improved time-series enhanced intelligent maintenance prediction algorithm to accurately predict vehicle maintenance needs. The user interface module includes a visualization unit, a maintenance suggestion unit, and a maintenance appointment unit. The visualization unit is used to provide a visual interface for users. The maintenance suggestion unit is used to provide vehicle maintenance suggestions for users. The maintenance appointment unit proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize the resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time.

[0006] Furthermore, the vehicle data acquisition unit collects real-time vehicle operating data through onboard communication sensors, including engine speed, fuel consumption, mileage, temperature, and pressure, ensuring the collection of complete and accurate vehicle data to obtain real-time dynamic vehicle condition information.

[0007] Furthermore, the communication transmission module uses 5G high-speed communication technology to transmit the collected vehicle data from the vehicle to the central processing and prediction module, ensuring the speed and efficiency of data transmission, reducing time delay, and ensuring data synchronization between the vehicle data and the central processing and prediction module.

[0008] Furthermore, the data analysis unit cleans and preprocesses the collected raw data, and uses historical and real-time data to analyze the changing trends of vehicle performance. At the same time, it integrates information from different vehicle communication sensors and data sources to provide comprehensive data analysis.

[0009] Furthermore, the maintenance demand prediction unit proposes an improved time-enhanced intelligent maintenance prediction algorithm to accurately predict vehicle maintenance needs, thereby identifying potential maintenance problems in advance and enabling preventative vehicle maintenance.

[0010] Furthermore, the improved time-series enhanced intelligent maintenance prediction algorithm is as follows: The vehicle data processed by the data processing unit is converted into a sequence, with each sequence corresponding to all collected data for one vehicle. This sequence is then sorted according to the order of occurrence and mileage. It is assumed that a single vehicle data sequence represents a multivariate fault event, i.e., S = (s...). T s T-1 , ..., s1), where S is the data sequence of a single vehicle, i.e., the reverse-order multivariate fault events from time step T to time step 1, s T For the fault event at time step T, sT-1 Let s1 be the fault event at time step T-1, s2 be the fault event at time step 1, and s3 be the vehicle data sequence at any time step t. Among them, s t This refers to a fault event at time step t. This is the first vehicle data feature at time step t. This is the second vehicle data feature at time step t. Let N be the Nth vehicle data feature at time step t, where N is the total number of vehicle data features at time step t. At time step T+1, the candidate event with the highest probability is selected from all possible fault events as the next predicted event. Simultaneously, a weighting factor is proposed to improve the prediction probability model, adjusting the importance of different candidate events to enable the prediction probability model to more flexibly handle different types of fault events. Where PE represents the predicted next failure event. For all possible candidate events, In order to obtain The function of the maximum value, where ω is the weighting factor for adjusting the probability value, S T＋1 Let θ represent the fault event at time step T+1, and let θ be the probability model parameter. Given a vehicle data sequence S and model parameters θ, the fault event S at the next time step T+1. T+1 Equal to candidate events The probability of events is calculated, and then the order and dependencies between events are captured to predict the next DTC event. A recurrent neural network is introduced to process the event history as a hidden state, and a time dynamic factor between events is proposed to improve the recurrent neural network prediction probability model, so as to better capture the order and dependencies between events, i.e., P(s T＋1 |S,τ,θ)=P(s T＋1 |h T , τ,θ), where τ is the introduced time dynamic factor, P(s T+1 Given a vehicle data sequence s, a time dynamic factor τ, and model parameters θ, predict the fault event s at the next time step T+1. T+1 The probability, h T Let P(s) be the hidden state at time step T. T+1 |h T ,τ,θ) are the values ​​of τ and θ in a given hidden state h T Given the time dynamic factor τ and model parameter θ, predict the fault event s at the next time step T+1. T+1The probability is calculated, and then the hidden state is calculated through an activation function. An improvement to the hidden state calculation formula is proposed, adjusting the learning parameters of the time dynamic factor to automatically adjust the time dynamic factor and improve prediction accuracy. T =F(s) T h T-1 ,τ,α)=F(s T ·ω h +α(h T-1 ·ω hh +τ·ω τ +b h ), where F is the activation function, h T-1 Let α be the hidden state at time step T-1, α be the learning parameter for adjusting the time dynamic factor, and ω be the hidden state. h Let ω be the weight matrix of the hidden state. hh Let ω be the weight matrix from hidden state to hidden state. τ b represents the weight of the influence of the time dynamic factor τ on the hidden state. h As the bias of the hidden layer, the output of the recurrent neural network is Y. T =F(h) T-1 ·ω hh +h T ·ω hq +b q ), where Y T ω is the output of the recurrent neural network at time step T. hq The hidden state h at time step T T Switch to the output weight matrix, b q To bias the output layer and further address the long-term dependency problem, a Long Short-Term Memory (LSTM) network is introduced on top of the recurrent neural network to remember long-term dependencies. This is achieved by adding additional components: a forget gate, an input gate, cell states, and an output gate, to resolve the issue of long-term dependencies in the remembered sequence. in, Let ω represent the candidate unit state at time step T, tanh(·) be the hyperbolic tangent activation function, and ω be the tangent activation function. sc For fault event s T to candidate cell state The weight matrix, ω hc The hidden state h at time step T-1 T-1 to candidate cell state The weight matrix, b c For the bias of the candidate cell state, c T For the cell state at time step T, F T For the output of the forget gate at time step T, I T The input gate is the output of time step T, ⊙ represents the dot product operation, and then the predicted fault event type is classified according to the output. in, For the fault category predicted by the model, Let y be the set of all possible fault categories, and y be the potential fault category. Given the output Y T and classification model parameters The probability of a fault category y is given as a condition to predict and classify vehicle faults. Based on different fault types, maintenance needs can be predicted, allowing for the implementation of different vehicle maintenance strategies. The improved temporal-enhanced intelligent maintenance prediction algorithm first proposes a weighting factor to improve the prediction probability model, adjusting the importance of different candidate events to make the model more flexible in handling different types of fault events. Then, a time-dynamic factor between events is proposed to improve the recurrent neural network prediction probability model, better capturing the order and dependencies between events. Next, the learning parameters of the time-dynamic factor are adjusted to improve the hidden state calculation formula, automatically adjusting the time-dynamic factor to improve prediction accuracy. Then, additional components are introduced to address the problem of long-term dependencies in the memory sequence. Finally, a classification of vehicle fault events is proposed to achieve accurate prediction of vehicle maintenance needs.

[0011] Furthermore, the visualization unit provides users with a user-friendly interface by converting data into an easy-to-understand and manipulate visual format, allowing them to easily obtain the vehicle condition information they need.

[0012] Furthermore, the maintenance recommendation unit provides users with personalized vehicle maintenance recommendations based on the predicted vehicle maintenance needs.

[0013] Furthermore, the maintenance appointment unit proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time.

[0014] Furthermore, the improved deep intelligent vehicle maintenance appointment algorithm is as follows: First, a reward function is defined to evaluate the effectiveness of the maintenance appointment strategy. Then, the reward function is improved by considering the timeliness of maintenance, customer satisfaction, and resource utilization to adapt to the resource allocation for vehicle maintenance appointments. Where R(π) is the improved reward function, π is the maintenance appointment strategy, and r i Let be the immediate reward for the i-th maintenance task, n be the total number of maintenance tasks, D be the set of completion times for all maintenance tasks, max(D) be the maximum completion time of D, min(D) be the minimum completion time of D, and the cumulative reward be... Among them, G tt For cumulative rewards, r tt The instant reward for time tt, r tt+1 The instant reward at time tt+1, rtt+2 The instant reward at time tt+2, r tt+K Let γ be the immediate reward at time t+K, γ be a discount factor to reduce the impact of future rewards, and k be a time integer between [0, K]. To optimize the vehicle maintenance appointment strategy, a deep Q-network is proposed to learn to take actions under different states to maximize long-term cumulative rewards. The update process of the maintenance appointment Q-value in the deep Q-network is Q... tt+1 (e tt a tt )=Q tt (e tt a tt )+β[r tt+1 +γmax a Q tt (e tt+1 ,a)-Q tt (e tt a tt )], where Q tt+1 (e tt a tt Let e ​​be the state at the next time step tt+1. tt Take action a tt Q value, Q tt (e tt a tt () represents the state e at the current time step tt. tt Take action a tt The Q-value is given by β, where β is the learning rate controlling the step size of Q-value updates in each iteration, and r is the learning rate. tt+1 For state e tt Take action a tt The immediate reward obtained afterward, where 'a' represents all possible actions, and 'max' represents the maximum reward. a Q tt (e tt+1 a) For the new state e tt+1 To find the maximum Q-value among all possible actions 'a', and to address the overestimation problem in deep Q-networks, Q is... tt+1 (e tt a tt It was improved into two independent Q-value functions, namely in, The Q-value of the first independent Q-value function at time step t+1. The Q-value of the second independent Q-value function at time step t+1. Let be the Q-value of the first independent Q-value function at time step t. The Q-value of the second independent Q-value function at time step tt. In the new state e tt+1Find all possible actions a. The action when the value is at its maximum. In the new state e tt+1 Find all possible actions a. The action when the value is at its maximum will be Used to evaluate the effectiveness of maintenance appointment strategies, To update the Q-value, two independent Q-value functions are used interactively to address the overestimation problem in deep Q-networks. To improve the algorithm's adaptability and convergence during maintenance appointments, the learning rate β is improved to β′ through learning rate decay, i.e., β′=β·exp(-δ·u), where β′ is the improved learning rate, δ is the decay factor, and u is the number of iteration steps. The maintenance appointment unit needs to consider factors such as vehicle maintenance needs, resource availability, and customer time preferences; therefore, the resource allocation optimization formula for vehicle maintenance appointments is: Where M represents the efficiency of resource allocation optimization, and V i For customer satisfaction, L i For resource utilization, D is the set of completion times for all maintenance tasks, max(D) is the maximum completion time of D, min(D) is the minimum completion time of D, and the optimal maintenance appointment time is TT. best =argmax tt (G tt ), where TT best For the best maintenance appointment time, G tt For cumulative rewards, argmax tt (G tt To accumulate the reward G tt The time step tt when the value is maximized is used to optimize the resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time. The improved deep intelligent vehicle maintenance appointment algorithm first decomposes the original deep Q-value function into two independent deep Q-value functions to solve the overestimation problem of deep Q-networks. Then, it proposes learning rate decay to improve the learning rate, making the algorithm more adaptive and convergent during the maintenance appointment process. Finally, it optimizes the resource allocation for vehicle maintenance appointments and obtains the optimal maintenance appointment time, thereby maximizing the utility of resources and improving the efficiency of maintenance appointment services.

[0015] Compared with the prior art, the beneficial effects of the present invention are:

[0016] 1. This invention proposes an improved temporal-enhanced intelligent maintenance prediction algorithm to accurately predict vehicle maintenance needs, enabling preventative vehicle maintenance. The innovation lies in the following: First, the improved temporal-enhanced intelligent maintenance prediction algorithm improves the prediction probability model by proposing a weighting factor to adjust the importance of different candidate events, allowing the prediction probability model to more flexibly handle different types of fault events. Then, it proposes a time-dynamic factor between events to improve the recurrent neural network prediction probability model, better capturing the order and dependencies between events. Next, it proposes adjusting the learning parameters of the time-dynamic factor to improve the hidden state calculation formula, automatically adjusting the time-dynamic factor to improve prediction accuracy. Then, it introduces additional components to address the problem of long-term dependencies in the memory sequence. Finally, it proposes a classification of vehicle fault events to achieve accurate prediction of vehicle maintenance needs.

[0017] 2. This invention proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time. The innovation of this invention lies in the following: First, the improved deep intelligent vehicle maintenance appointment algorithm decomposes the original deep Q-value function into two independent deep Q-value functions to solve the overestimation problem of deep Q-networks. Then, it proposes a learning rate decay method to improve the learning rate, making the algorithm more adaptive and convergent during the maintenance appointment process. Finally, it optimizes the resource allocation for vehicle maintenance appointments and obtains the optimal maintenance appointment time, thereby maximizing resource utilization and improving the efficiency of maintenance appointment services. Attached Figure Description

[0018] The invention will be further illustrated with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the invention. For those skilled in the art, other drawings can be obtained based on the following drawings without any creative effort.

[0019] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] A system and method for preventive vehicle maintenance includes a vehicle data acquisition module, a communication transmission module, a central processing and prediction module, and a user interface module. The vehicle data acquisition module is used to collect vehicle operating data in real time. The communication transmission module is used to transmit the collected vehicle data to the central processing and prediction unit. The central processing and prediction module includes a data analysis unit and a maintenance demand prediction unit. The data analysis unit is used to process and analyze vehicle data. The maintenance demand prediction unit proposes an improved time-series enhanced intelligent maintenance prediction algorithm to accurately predict vehicle maintenance needs. The user interface module includes a visualization unit, a maintenance suggestion unit, and a maintenance appointment unit. The visualization unit provides a visual interface for users. The maintenance suggestion unit provides users with vehicle maintenance suggestions. The maintenance appointment unit proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize the resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time.

[0022] Preferably, the vehicle data acquisition unit collects vehicle operating data in real time through onboard communication sensors, including engine speed, fuel consumption, mileage, temperature, and pressure, to ensure that complete and accurate vehicle data is collected in order to obtain real-time dynamic vehicle condition information.

[0023] Preferably, the communication transmission module uses 5G high-speed communication technology to transmit the collected vehicle data from the vehicle to the central processing and prediction module, ensuring the speed and efficiency of data transmission, reducing time delay, and ensuring data synchronization between the vehicle data and the central processing and prediction module.

[0024] Preferably, the data analysis unit cleans and preprocesses the collected raw data to remove noise and outliers, extract useful information, and analyze the changing trends of vehicle performance using historical and real-time data. At the same time, it integrates information from different vehicle communication sensors and data sources to provide comprehensive data analysis.

[0025] Preferably, the maintenance demand prediction unit proposes an improved time-enhanced intelligent maintenance prediction algorithm to accurately predict vehicle maintenance needs, thereby identifying potential maintenance problems in advance, enabling preventive maintenance, improving vehicle reliability, and extending vehicle lifespan.

[0026] Specifically, the improved time-series enhanced intelligent maintenance prediction algorithm is as follows: The vehicle data processed by the data processing unit is converted into a sequence, with each sequence corresponding to all collected data for one vehicle. This sequence is then sorted according to the order of occurrence and mileage. It is assumed that a single vehicle data sequence represents a multivariate fault event, i.e., S = (s...). T s T-1 , ..., s1), where S is a data sequence for a single vehicle, i.e., the reverse-order multivariate fault events from time step T to time step 1 (the latest event is at the beginning of the sequence), sT For the fault event at time step T, s T-1 Let s1 be the fault event at time step T-1, s2 be the fault event at time step 1, and s3 be the vehicle data sequence at any time step t. Among them, s t This refers to a fault event at time step t. This is the first vehicle data feature at time step t. This is the second vehicle data feature at time step t. Let N be the Nth vehicle data feature at time step t, where N is the total number of vehicle data features at time step t. At time step T+1, the candidate event with the highest probability is selected from all possible fault events as the next predicted event. Simultaneously, a weighting factor is proposed to improve the prediction probability model, adjusting the importance of different candidate events to enable the prediction probability model to more flexibly handle different types of fault events. Where PE represents the predicted next failure event. For all possible candidate events, In order to obtain The function of the maximum value, where ω is the weighting factor for adjusting the probability value, s T+1 Let θ represent the fault event at time step T+1, and let θ be the probability model parameter. Given a vehicle data sequence S and model parameters θ, the fault event s at the next time step T+1. T+1 Equal to candidate events The probability of predicting the next DTC event is determined by capturing the order and dependencies between events to predict the next DTC event. This is achieved because the data sequence consists of multiple DTC events (DTC events are standard codes used in the automotive industry to identify potential vehicle malfunctions) before time step T. A recurrent neural network is introduced to process the event history as hidden states, and a time dynamic factor between events is proposed to improve the recurrent neural network prediction probability model, thereby better capturing the order and dependencies between events, i.e., P(s). T＋1 |S,τ,θ)=P(s T＋1 |h T , τ, θ), where τ is the introduced time dynamic factor, P(s T+1 Given a vehicle data sequence S, a time dynamic factor τ, and model parameters θ, predict the fault event s at the next time step T+1. T+1 The probability, h T Let P(s) be the hidden state at time step T. T+1 |h T ,τ,θ) are the values ​​of τ and θ in a given hidden state h TGiven the time dynamic factor τ and model parameter θ, predict the fault event s at the next time step T+1. T+1 The probability is calculated, and then the hidden state is calculated through an activation function. An improvement to the hidden state calculation formula is proposed, adjusting the learning parameters of the time dynamic factor to automatically adjust the time dynamic factor and improve prediction accuracy. T =F(s) T h T-1 ,τ,α)=F(s T ·ω h +α(h T-1 ·ω hh +τ·ω τ +b h ), where F is the activation function, h T-1 Let α be the hidden state at time step T-1, α be the learning parameter for adjusting the time dynamic factor, and ω be the hidden state. h Let ω be the weight matrix of the hidden state. hh Let ω be the weight matrix from hidden state to hidden state. τ b represents the weight of the influence of the time dynamic factor τ on the hidden state. h As the bias of the hidden layer, the output of the recurrent neural network is Y. T =F(h) T-1 ·ω hh +h T ·ω hq +b q ), where Y T ω is the output of the recurrent neural network at time step T. hq The hidden state h at time step T T Switch to the output weight matrix, b q To bias the output layer and further address the long-term dependency problem, a Long Short-Term Memory (LSTM) network is introduced on top of the recurrent neural network to remember long-term dependencies. This is achieved by adding additional components: a forget gate, an input gate, cell states, and an output gate, to resolve the issue of long-term dependencies in the remembered sequence. in, Let ω represent the candidate unit state at time step T, tanh(·) be the hyperbolic tangent activation function, and ω be the tangent activation function. sc For fault event s T to candidate cell state The weight matrix, ω hc The hidden state h at time step T-1 T-1 to candidate cell state The weight matrix, b c For the bias of the candidate cell state, c T For the cell state at time step T, F T For the output of the forget gate at time step T, I TThe input gate is the output of time step T, ⊙ represents the dot product operation, and then the predicted fault event type is classified according to the output. in, For the fault category predicted by the model, Let y be the set of all possible fault categories, and y be the potential fault category. Given the output Y T and classification model parameters The probability of a fault category y is given as a condition to predict and classify vehicle faults. Based on different fault types, maintenance needs can be predicted, allowing for the implementation of different vehicle maintenance strategies. The improved temporal-enhanced intelligent maintenance prediction algorithm first proposes a weighting factor to improve the prediction probability model, adjusting the importance of different candidate events to make the model more flexible in handling different types of fault events. Then, a time-dynamic factor between events is proposed to improve the recurrent neural network prediction probability model, better capturing the order and dependencies between events. Next, the learning parameters of the time-dynamic factor are adjusted to improve the hidden state calculation formula, automatically adjusting the time-dynamic factor to improve prediction accuracy. Then, additional components are introduced to address the problem of long-term dependencies in the memory sequence. Finally, a classification of vehicle fault events is proposed to achieve accurate prediction of vehicle maintenance needs.

[0027] Preferably, the visualization unit provides users with a user-friendly interface by converting data into an easy-to-understand and manipulate visual format, allowing them to easily obtain the required vehicle condition information.

[0028] Preferably, the maintenance recommendation unit provides users with personalized vehicle maintenance recommendations based on the predicted vehicle maintenance needs, helping users to develop and optimize vehicle maintenance plans and ensure that the vehicle operates in optimal condition.

[0029] Preferably, the maintenance appointment unit proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize the resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time.

[0030] Specifically, the improved deep intelligent vehicle maintenance appointment algorithm is as follows: First, a reward function is defined to evaluate the effectiveness of the maintenance appointment strategy. Then, the reward function is improved by considering the timeliness of maintenance, customer satisfaction, and resource utilization to adapt to the resource allocation for vehicle maintenance appointments. Where R(π) is the improved reward function, π is the maintenance appointment strategy, and r i Let be the immediate reward for the i-th maintenance task, n be the total number of maintenance tasks, D be the set of completion times for all maintenance tasks, max(D) be the maximum completion time of D, min(D) be the minimum completion time of D, and the cumulative reward be... Among them, Gtt For cumulative rewards, r tt The instant reward for time tt, r tt+1 The instant reward at time tt+1, r tt+2 The instant reward at time tt+2, r tt+K Let γ be the immediate reward at time t+K, γ be the discount factor to reduce the impact of future rewards, and k be a time integer between [0, K]. To optimize the vehicle maintenance appointment strategy, a deep Q-network is proposed to learn to take actions (i.e., schedule maintenance) under different states (vehicle maintenance needs, resource availability, customer time preferences, etc.) to maximize long-term cumulative rewards. The update process of the maintenance appointment Q-value in the deep Q-network is Q... tt+1 (e tt a tt )=Q tt (e tt a tt )+β[r tt+1 +γmax a Q tt (e tt+1 ,a)-Q tt (e tt a tt )], where Q tt+1 (e tt a tt Let e ​​be the state at the next time step tt+1. tt Take action a tt Q value, Q tt (e tt a tt () represents the state e at the current time step tt. tt Take action a tt The Q-value is given by β, where β is the learning rate controlling the step size of Q-value updates in each iteration, and r is the learning rate. tt+1 For state e tt Take action a tt The immediate reward obtained afterward, where 'a' represents all possible actions, and 'max' represents the maximum reward. a Q tt (e tt+1 a) For the new state e tt+1 To find the maximum Q-value among all possible actions 'a', and to address the overestimation problem in deep Q-networks, Q is... tt+1 (e tt a tt It was improved into two independent Q-value functions, namely in, The Q-value of the first independent Q-value function at time step t+1. The Q-value of the second independent Q-value function at time step t+1. Let be the Q-value of the first independent Q-value function at time step t. The Q-value of the second independent Q-value function at time step tt. In the new state e tt+1 Find all possible actions a. The action when the value is at its maximum. In the new state e tt+1 Find all possible actions a. The action when the value is at its maximum will be Used to evaluate the effectiveness of maintenance appointment strategies, To update the Q-value, two independent Q-value functions are used interactively to address the overestimation problem in deep Q-networks. To improve the algorithm's adaptability and convergence during maintenance appointments, the learning rate β is improved to β′ through learning rate decay, i.e., β′=β·exp(-δ·u), where β′ is the improved learning rate, δ is the decay factor, and u is the number of iteration steps. The maintenance appointment unit needs to consider factors such as vehicle maintenance needs, resource availability, and customer time preferences; therefore, the resource allocation optimization formula for vehicle maintenance appointments is: Where M represents the efficiency of resource allocation optimization, and V i For customer satisfaction, L i For resource utilization, D is the set of completion times for all maintenance tasks, max(D) is the maximum completion time of D, min(D) is the minimum completion time of D, and the optimal maintenance appointment time is TT. best =argmax tt (G tt ), where TT best For the best maintenance appointment time, G tt For cumulative rewards, argmax tt (G tt To accumulate the reward G tt The time step tt when the value is maximized is used to optimize the resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time. The improved deep intelligent vehicle maintenance appointment algorithm first decomposes the original deep Q-value function into two independent deep Q-value functions to solve the overestimation problem of deep Q-networks. Then, it proposes learning rate decay to improve the learning rate, making the algorithm more adaptive and convergent during the maintenance appointment process. Finally, it optimizes the resource allocation for vehicle maintenance appointments and obtains the optimal maintenance appointment time, thereby maximizing the utility of resources and improving the efficiency of maintenance appointment services.

[0031] This paper proposes a system and method for preventive vehicle maintenance. It utilizes predictive modeling and deep learning techniques to monitor vehicle operating status in real time, predicting and preventing potential faults to optimize maintenance strategies and improve vehicle reliability and safety. By integrating a vehicle data acquisition module, communication transmission module, central processing and prediction module, and user interface module, a system and method for preventive vehicle maintenance are provided. An improved temporal-enhanced intelligent maintenance prediction algorithm is proposed to accurately predict vehicle maintenance needs, enabling preventive maintenance. The innovation of this invention lies in the following: First, the improved temporal-enhanced intelligent maintenance prediction algorithm improves the prediction probability model by proposing a weighting factor to adjust the importance of different candidate events, allowing the prediction probability model to more flexibly handle different types of fault events. Then, a time dynamic factor between events is proposed to improve the recurrent neural network prediction probability model to better capture the order and dependencies between events. Next, the learning parameters of the time dynamic factor are adjusted to improve the hidden state calculation formula, automatically adjusting the time dynamic factor to improve prediction accuracy. Finally, additional components are introduced to address long-term dependencies in the memory sequence. This invention addresses the issue of dependency relationships and proposes a classification system for vehicle malfunction events to accurately predict vehicle maintenance needs. It also proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize resource allocation for vehicle maintenance appointments and determine the optimal appointment time. The innovation of this invention lies in the following: First, the improved deep intelligent vehicle maintenance appointment algorithm decomposes the original deep Q-value function into two independent deep Q-value functions to solve the overestimation problem of deep Q-networks. Then, it proposes a learning rate decay method to improve the learning rate, giving the algorithm better adaptability and convergence during the maintenance appointment process. Finally, it optimizes the resource allocation for vehicle maintenance appointments and obtains the optimal appointment time, maximizing resource utilization and improving the efficiency of maintenance appointment services. This effectively enhances the working effect of automotive preventive maintenance systems and methods, providing more comprehensive and accurate technical support for automotive preventive maintenance systems and better decision support for scientific and efficient automotive preventive maintenance systems. Furthermore, this invention relates to predictive modeling and deep learning technologies, providing users with an efficient and convenient automotive preventive maintenance system and method, contributing significant application value to the field of automotive preventive maintenance.

[0032] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A system and method for preventive maintenance of automobiles, characterized in that, The system includes a vehicle data acquisition module, a communication transmission module, a central processing and prediction module, and a user interface module. The vehicle data acquisition module collects real-time vehicle operating data. The communication transmission module transmits the collected vehicle data to the central processing and prediction module. The central processing and prediction module includes a data analysis unit and a maintenance demand prediction unit. The data analysis unit processes and analyzes the vehicle data. The maintenance demand prediction unit proposes an improved time-series enhanced intelligent maintenance prediction algorithm to accurately predict vehicle maintenance needs. The user interface module includes a visualization unit, a maintenance suggestion unit, and a maintenance appointment unit. The visualization unit provides a visual interface for users. The maintenance suggestion unit provides vehicle maintenance suggestions for users. The maintenance appointment unit proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time.

2. The system and method for preventive maintenance of automobiles according to claim 1, characterized in that, The vehicle data acquisition unit collects real-time vehicle operating data through onboard communication sensors, including engine speed, fuel consumption, mileage, temperature, and pressure, ensuring the collection of complete and accurate vehicle data to obtain real-time dynamic vehicle condition information.

3. The system and method for preventive maintenance of automobiles according to claim 1, characterized in that, The communication transmission module uses 5G high-speed communication technology to transmit the collected vehicle data from the vehicle to the central processing and prediction module, ensuring the speed and efficiency of data transmission, reducing time delay, and ensuring data synchronization between the vehicle data and the central processing and prediction module.

4. The system and method for preventive maintenance of automobiles according to claim 1, characterized in that, The data analysis unit cleans and preprocesses the collected raw data, and uses historical and real-time data to analyze the changing trends of vehicle performance. At the same time, it integrates information from different vehicle communication sensors and data sources to provide comprehensive data analysis.

5. The system and method for preventive maintenance of automobiles according to claim 1, characterized in that, The maintenance demand prediction unit proposes an improved time-enhanced intelligent maintenance prediction algorithm to accurately predict vehicle maintenance needs, thereby identifying potential maintenance problems in advance and enabling preventative vehicle maintenance.

6. The system and method for preventive maintenance of automobiles according to claim 5, characterized in that, The improved time-series enhanced intelligent maintenance prediction algorithm is as follows: The vehicle data processed by the data processing unit is converted into a sequence, with each sequence corresponding to all collected data for one vehicle. These sequences are then sorted according to the order of occurrence and mileage. It is assumed that a single vehicle data sequence represents a multivariate fault event, i.e., S = (s...). T s T-1 , ..., s1), where S is the data sequence of a single vehicle, i.e., the reverse-order multivariate fault events from time step T to time step 1, s T For the fault event at time step T, s T-1 Let s1 be the fault event at time step T-1, s2 be the fault event at time step 1, and s3 be the vehicle data sequence at any time step t. Among them, s t This refers to a fault event at time step t. This is the first vehicle data feature at time step t. This is the second vehicle data feature at time step t. Let N be the Nth vehicle data feature at time step t, where N is the total number of vehicle data features at time step t. At time step T+1, the candidate event with the highest probability is selected from all possible fault events as the next predicted event. Simultaneously, a weighting factor is proposed to improve the prediction probability model, adjusting the importance of different candidate events to enable the prediction probability model to more flexibly handle different types of fault events. Where PE represents the predicted next failure event. For all possible candidate events, In order to obtain The function of the maximum value, where ω is the weighting factor for adjusting the probability value, S T+1 Let θ represent the fault event at time step T+1, and let θ be the probability model parameter. Given a vehicle data sequence S and model parameters θ, the fault event S at the next time step T+1. T+1 Equal to candidate events The probability of events is calculated, and then the order and dependencies between events are captured to predict the next DTC event. A recurrent neural network is introduced to process the event history as a hidden state, and a time dynamic factor between events is proposed to improve the recurrent neural network prediction probability model, so as to better capture the order and dependencies between events, i.e., P(s T+1 |S,τ,θ)=P(S T+1 |h T , τ, θ), where τ is the introduced time dynamic factor, P(s T+1 Given a vehicle data sequence S, a time dynamic factor τ, and model parameters θ, predict the fault event S at the next time step T+1. T+1 The probability, h T Let P(s) be the hidden state at time step T. T+1 |h T ,τ,θ) are the values ​​of τ and θ in a given hidden state h T Given the time dynamic factor τ and model parameter θ, predict the fault event s at the next time step T+1. T+1 The probability is calculated, and then the hidden state is calculated through an activation function. An improvement to the hidden state calculation formula is proposed, adjusting the learning parameters of the time dynamic factor to automatically adjust the time dynamic factor and improve prediction accuracy. T =F(s) T h T-1 ,τ,α)=F(s T ·ω h +α(h T-1 ·ω hh +τ·ω τ +b h ), where F is the activation function, h T-1 Let α be the hidden state at time step T-1, α be the learning parameter for adjusting the time dynamic factor, and ω be the hidden state. h Let ω be the weight matrix of the hidden state. hh Let ω be the weight matrix from hidden state to hidden state. τ b represents the weight of the influence of the time dynamic factor τ on the hidden state. h As the bias of the hidden layer, the output of the recurrent neural network is Y. T =F(h) T-1 ·ω hh +h T ·ω hq +b q ), where Y T ω is the output of the recurrent neural network at time step T. hq The hidden state h at time step T T Switch to the output weight matrix, b q To bias the output layer and further address the long-term dependency problem, a Long Short-Term Memory (LSTM) network is introduced on top of the recurrent neural network to remember long-term dependencies. This is achieved by adding additional components: a forget gate, an input gate, cell states, and an output gate, to resolve the issue of long-term dependencies in the remembered sequence. in, Let ω represent the candidate unit state at time step T, tanh(·) be the hyperbolic tangent activation function, and ω be the tangent activation function. sc For fault event s T to candidate cell state The weight matrix, ω hc The hidden state h at time step T-1 T-1 to candidate cell state The weight matrix, b c For the bias of the candidate cell state, c T For the cell state at time step T, F T For the output of the forget gate at time step T, I T The input gate is the output of time step T, ⊙ represents the dot product operation, and then the predicted fault event type is classified according to the output. in, For the fault category predicted by the model, Let p(y|Y) be the set of all possible fault categories, y be the potential fault category, and p(y|Y) be the set of all possible fault categories. T ,θ) is the given output Y T The probability of a fault category being y given the classification model parameters θ is used to predict and classify vehicle faults. Based on different fault types, maintenance needs can be predicted, allowing for the adoption of different vehicle maintenance strategies. The improved temporal-enhanced intelligent maintenance prediction algorithm first proposes a weighting factor to improve the prediction probability model, adjusting the importance of different candidate events to make the model more flexible in handling different types of fault events. Then, a time-dynamic factor between events is proposed to improve the recurrent neural network prediction probability model, better capturing the order and dependencies between events. Next, the learning parameters of the time-dynamic factor are adjusted to improve the hidden state calculation formula, automatically adjusting the time-dynamic factor to improve prediction accuracy. Then, additional components are introduced to address the problem of long-term dependencies in the memory sequence. Finally, a classification of vehicle fault events is proposed to achieve accurate prediction of vehicle maintenance needs.

7. The system and method for preventive maintenance of automobiles according to claim 1, characterized in that, The visualization unit provides users with a user-friendly interface by converting data into an easy-to-understand and manipulate visual format, allowing them to easily obtain the vehicle condition information they need.

8. The system and method for preventive maintenance of automobiles according to claim 1, characterized in that, The maintenance recommendation unit provides users with personalized vehicle maintenance recommendations based on the predicted vehicle maintenance needs.

9. The system and method for preventive maintenance of automobiles according to claim 1, characterized in that, The maintenance appointment unit proposes an improved deep intelligent vehicle maintenance appointment algorithm to optimize resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time.

10. The system and method for preventive maintenance of automobiles according to claim 9, characterized in that, The improved deep intelligent vehicle maintenance appointment algorithm is as follows: First, a reward function is defined to evaluate the effectiveness of the maintenance appointment strategy. Then, the reward function is improved by considering the timeliness of maintenance, customer satisfaction, and resource utilization to adapt to the resource allocation for vehicle maintenance appointments. Where R(π) is the improved reward function, π is the maintenance appointment strategy, and r i Let be the immediate reward for the i-th maintenance task, n be the total number of maintenance tasks, D be the set of completion times for all maintenance tasks, max(D) be the maximum completion time of D, min(D) be the minimum completion time of D, and the cumulative reward be... Among them, G tt For cumulative rewards, r tt The instant reward for time tt, r tt+1 The instant reward at time tt+1, r tt+2 The instant reward at time tt+2, r tt+K Let γ be the immediate reward at time t+K, γ be a discount factor to reduce the impact of future rewards, and k be a time integer between [0, K]. To optimize the vehicle maintenance appointment strategy, a deep Q-network is proposed to learn to take actions under different states to maximize long-term cumulative rewards. The update process of the maintenance appointment Q-value in the deep Q-network is Q... tt+1 (e tt a tt )=Q tt (e tt a tt )+β[r tt+1 +γmax a Q tt (e tt+1 ,a)-Q tt (e tt a tt )], where Q tt+1 (e tt a tt Let e ​​be the state at the next time step tt+1. tt Take action a tt Q value, Q tt (e tt a tt () represents the state e at the current time step tt. tt Take action a tt The Q-value is given by β, where β is the learning rate controlling the step size of Q-value updates in each iteration, and r is the learning rate. tt+1 For state e tt Take action a tt The immediate reward obtained afterward, where 'a' represents all possible actions, and 'max' represents the maximum reward. a Q tt (e tt+1 a) For the new state e tt+1 To find the maximum Q-value among all possible actions 'a', and to address the overestimation problem in deep Q-networks, Q is... tt+1 (e tt a tt It was improved into two independent Q-value functions, namely in, The Q-value of the first independent Q-value function at time step t+1. The Q-value of the second independent Q-value function at time step t+1. Let be the Q-value of the first independent Q-value function at time step t. The Q-value of the second independent Q-value function at time step tt. In the new state e tt+1 Find all possible actions a. The action when the value is at its maximum. In the new state e tt+1 Find all possible actions a. The action when the value is at its maximum will be Used to evaluate the effectiveness of maintenance appointment strategies, To update the Q-value, two independent Q-value functions are used interactively to address the overestimation problem in deep Q-networks. To improve the algorithm's adaptability and convergence during maintenance appointments, the learning rate β is improved to β′ through learning rate decay, i.e., β′=β·exp(-δ·u), where β′ is the improved learning rate, δ is the decay factor, and u is the number of iteration steps. The maintenance appointment unit needs to consider factors such as vehicle maintenance needs, resource availability, and customer time preferences; therefore, the resource allocation optimization formula for vehicle maintenance appointments is: Where M represents the efficiency of resource allocation optimization, and V i For customer satisfaction, L i For resource utilization, D is the set of completion times for all maintenance tasks, max(D) is the maximum completion time of D, min(D) is the minimum completion time of D, and the optimal maintenance appointment time is TT. best =argmax tt (G tt ), where TT best For the best maintenance appointment time, G tt For cumulative rewards, argmax tt (G tt To accumulate the reward G tt The time step tt when the value is maximized is used to optimize the resource allocation for vehicle maintenance appointments and determine the optimal maintenance appointment service time. The improved deep intelligent vehicle maintenance appointment algorithm first decomposes the original deep Q-value function into two independent deep Q-value functions to solve the overestimation problem of deep Q-networks. Then, it proposes learning rate decay to improve the learning rate, making the algorithm more adaptive and convergent during the maintenance appointment process. Finally, it optimizes the resource allocation for vehicle maintenance appointments and obtains the optimal maintenance appointment time, thereby maximizing the utility of resources and improving the efficiency of maintenance appointment services.

Images