Automobile preventive maintenance intelligent optimization system based on random forest algorithm

Abstract

A vehicle preventive maintenance intelligent optimization system based on a random forest algorithm comprises a data acquisition module, a data preprocessing module, a model training and prediction module, a maintenance period optimization module and a user interface module, and the data acquisition module acquires real-time data of a vehicle through a vehicle-mounted sensor and a GPS device; the data preprocessing module is used for cleaning, denoising and standardizing data, the model training and predicting module is used for constructing an integrated model of multiple decision-making trees by utilizing a random forest algorithm based on the preprocessed data, and the maintenance period optimizing module is used for dynamically adjusting a recommendation period in combination with historical maintenance information of a user and a prediction result. And the suggestion period and the related analysis result are displayed to the user through the user interface module, so that the service life of the vehicle is prolonged, the maintenance cost is reduced, and the system has the advantages of being efficient, accurate and capable of achieving real-time feedback and is suitable for the customized maintenance requirement of the intelligent vehicle.

Description

Technical Field

[0001] The invention relates to the field of intelligent services and data processing, and specifically to an automobile preventive maintenance intelligent optimization system based on a random forest algorithm. Background Art

[0002] With the development of the automobile industry, cars have become an indispensable means of transportation in modern people's lives. Daily maintenance and repair of cars are important means to maintain their normal operation, extend their service life and ensure driving safety. However, traditional car maintenance methods usually rely on fixed time or mileage to determine the maintenance cycle. For example, manufacturers or service providers usually recommend maintenance every certain mileage (such as 5,000 kilometers) or after a certain period of time (such as 6 months). Although this method is simple, it does not take into account the actual differences between each car during use, including the driver's driving habits, road conditions, climatic conditions and the condition of the vehicle itself. This "one-size-fits-all" maintenance cycle method often leads to problems of over-maintenance or under-maintenance. Over-maintenance will increase the user's time and money costs, while under-maintenance may lead to a decline in vehicle performance and even safety hazards. Therefore, how to intelligently and individually determine the maintenance cycle of a car has become an important technical issue.

[0003] With the rapid development of information technology and big data analysis, data-driven intelligent prediction technology has been widely used in various fields. Especially driven by the Internet of Things (IoT) and Vehicle-to-Everything (V2X), modern cars can collect and transmit a large amount of sensor data in real time, such as engine temperature, fuel consumption, tire pressure, and driving speed. These real-time data provide technical possibilities for dynamically evaluating vehicle status and predicting vehicle maintenance needs. However, to extract meaningful features from a large amount of data and analyze the specific maintenance needs of the vehicle, it is also necessary to use machine learning intelligent algorithms. In recent years, machine learning algorithms, especially random forests and neural network data-driven models, have shown high accuracy and robustness in the fields of prediction and classification, and have gradually been introduced into intelligent vehicle management and health management systems. As an algorithm based on ensemble learning, the random forest algorithm has good generalization and anti-overfitting capabilities, can effectively process high-dimensional data and complex nonlinear relationships, and is particularly suitable for multi-factor comprehensive evaluation of vehicle status and maintenance cycle prediction.

[0004] The current intelligent maintenance system has gradually acquired certain prediction capabilities, but there are still several key challenges. First, the diversity and real-time nature of data require the system to have strong data processing capabilities. The driving habits and environmental conditions of different models and users vary significantly, resulting in a strong diversity of data sources. The system must be able to quickly analyze and process these heterogeneous data. Secondly, accuracy and robustness are the core requirements for intelligent optimization of maintenance cycles. Traditional simple prediction methods are difficult to meet the needs of practical applications. How to extract key features from complex multidimensional data and continuously optimize the prediction model when the data changes dynamically is the focus of current research. Finally, user experience and acceptance also need to be considered. The user interface of the existing intelligent maintenance system is often complicated and difficult for users to clearly understand the maintenance suggestions given by the system. Therefore, in order to achieve accurate maintenance cycle prediction, personalized maintenance suggestions, and an easy-to-operate user interface, it is necessary to build an intelligent maintenance cycle optimization system that can adapt to multi-dimensional data input, has high prediction performance, and meets user experience. Summary of the invention

[0005] In view of the above problems, the present invention aims to provide an automobile preventive maintenance intelligent optimization system based on random forest algorithm to solve the problems raised in the above background technology.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] The present invention provides an intelligent optimization system for automobile preventive maintenance based on a random forest algorithm, comprising a data acquisition module, a data preprocessing module, a model training and prediction module, a maintenance cycle optimization module and a user interface module. The data acquisition module obtains real-time data of a vehicle through vehicle-mounted sensors, GPS devices, etc. The data preprocessing module cleans, denoises and standardizes the data. The model training and prediction module uses the random forest algorithm to build an integrated model of multiple decision trees based on the preprocessed data, thereby improving prediction accuracy and robustness. The system inputs new data into the model and predicts a suitable maintenance cycle according to the driver's real-time driving behavior and vehicle health status. The maintenance cycle optimization module dynamically adjusts the recommended cycle in combination with the user's historical maintenance information and prediction results, and displays the recommended cycle and related analysis results to the user through the user interface module, so that the user can perform maintenance at an appropriate time, thereby extending the vehicle's service life and reducing maintenance costs.

[0008] Furthermore, the data acquisition module is composed of an on-board sensor integration unit, a global positioning system (GPS) device, an external data interface and a data storage unit. The on-board sensor integration unit is a data acquisition module, which collects the internal operation data of the vehicle in real time through engine sensors, fuel consumption sensors, temperature sensors, tire pressure sensors, brake wear sensors, battery power sensors and speed sensors. Each sensor corresponds to a different collection task. The GPS device collects the vehicle's location information, driving route and driving speed data in real time through the global positioning system. The system can analyze the vehicle's driving conditions. The data collected by the GPS device will be input into the system together with the sensor data to analyze the impact of environmental factors on the vehicle, so as to more accurately predict the maintenance cycle. The data acquisition module also includes an external data interface for receiving information from an external data source. The operation of the vehicle is not only affected by its own state, but also significantly affected by the external environment. Through the external data interface, the system can receive weather data, real-time traffic flow information and road construction road condition information, which are obtained through cooperation with weather data providers, traffic information platforms and road management systems, and updated in real time. The introduction of external data can not only supplement the environmental factors that are difficult to capture by on-board sensors, but also enable the system to understand the operating environment of the vehicle.

[0009] Furthermore, the collected data is stored in a data storage unit for subsequent analysis and prediction. The data storage unit has efficient storage and management functions and can manage real-time data and historical data in layers. Real-time data is used for current maintenance cycle prediction, and historical data is used for model training and long-term trend analysis. During the storage process, the data will be grouped and labeled according to time, data type, and source device dimensions for subsequent rapid retrieval and analysis.

[0010] Furthermore, the data preprocessing module performs cleaning, denoising, and standardization preprocessing operations on the multi-source data collected by the data acquisition module. For numerical extreme values, three times the standard deviation is used for detection; for outliers, they are replaced with the mean value. For missing data, the system fills them through interpolation to avoid the missing data causing deviations in model training. Then the data denoising is completed, and the data is smoothed by optimizing the moving average method and taking the average of continuous data points to reduce abnormal fluctuations. The denoised data is standardized and normalized. Standardization adjusts the data to a standard normal distribution with the same mean and variance, and normalization scales the data proportionally to the [0,1] interval.

[0011] Furthermore, the model training and prediction module completes the model training through the random forest algorithm, as follows:

[0012] Define the training set as x i =[xi1 ,x i2 ,…x im ], where D is the training set, x i is the eigenvector of the i-th group, N represents the number of samples, y i is the target maintenance cycle, x i1 ,x i2 ,…x im are the first feature vector of the i-th sample, the second feature vector of the i-th sample, and the m-th feature vector of the i-th sample, respectively. N` samples are extracted from the training set with replacement, where N`≤N, and K different subsets are generated, each with a length of N`. When constructing each decision tree, FLOOR(sqrtm) features are randomly selected from the m features for splitting, where the FLOOR function is a downward forensic function. For each node of the k-th decision tree, the optimal splitting conditions of all optional splitting points are calculated. For the current sample set of the k-th decision tree Information gain Gain is expressed as follows:

[0013]

[0014] Among them, p n Gini is the proportion of the nth sample in the sample set D, and the Gini index is defined as index ,have:

[0015]

[0016] Define mean square error MSE:

[0017]

[0018] in, for Maintenance cycle, is calculated in the random forest algorithm The greater the information gain, the greater the purity of the feature partition node. If maximization is defined as forward conversion and minimization is defined as reverse conversion, in order to unify the evaluation index, the information gain is reversed as Constructing the indicator structure For K decision trees, there is a K*3-dimensional standardized evaluation matrix A*:

[0019]

[0020] in, is the inverse element of information gain in the standardized evaluation matrix of the first decision tree, is the inverse element of information gain in the standardized evaluation matrix of the second decision tree, Gini is the inverse element of the information gain in the standardized evaluation matrix of the Kth decision tree. index (1) is the Gini index element in the standardized evaluation matrix of the first decision tree, Gini index (2) is the Gini index element in the standardized evaluation matrix of the second decision tree, Gini index (K) is the Gini index element in the standardized evaluation matrix of the Kth decision tree, is the mean square error element in the standardized evaluation matrix of the first decision tree, is the mean square error element in the standardized evaluation matrix of the second decision tree, Define the maximum value vector A* of the standardized matrix as the mean square error element in the standardized evaluation matrix of the Kth decision tree + :

[0021]

[0022] Define the normalized matrix minimum vector A* -

[0023]

[0024] Then the evaluation index SCORE of the kth decision tree is k Calculated as:

[0025]

[0026] in:

[0027]

[0028] Furthermore, the prediction performance is improved by constructing a node loss function, which is expressed as follows:

[0029]

[0030] Among them, Loss(k) is the loss function of the kth decision tree. In order to minimize the weighted loss of the entire decision tree, the loss increment ΔLoss is calculated. k :

[0031]

[0032] Among them, D k< D k The subset of left child nodes under the splitting rule π of the decision tree in D k> D k The right child node subset under the splitting rule π of the decision tree in has {D k< ∩D k> =D k , Loss(k<) is Dk< The loss function, Loss(k>) is D k> The loss function of the regularization term is introduced to prevent the model from overfitting:

[0033]

[0034] Among them, Loss is the loss function of random forest, α is the model complexity penalty coefficient, which constrains the number of samples contained in the node to prevent the tree structure from expanding without limit. If α is large, the complexity penalty of the node is high, and the system tends to remove sub-nodes containing fewer samples and give priority to important nodes; if α is small, the penalty for node complexity is weakened, and the system tends to retain more sub-nodes, allowing the tree structure to be more complex. β is the feature selection penalty coefficient, which reflects the importance distribution of nodes for different features. To measure, increasing β will increase the penalty for feature variance, prompting the system to be more inclined to remove split nodes related to low-importance features, while reducing β will weaken the penalty for feature selection, allowing the random forest to use more features for complex splitting, γ m is the feature weight, which indicates the importance of the feature to the model prediction.

[0035] Furthermore, the maintenance cycle optimization module optimizes the maintenance cycle predicted by the model to provide accurate and personalized maintenance suggestions. By dynamically adjusting and optimizing the initially predicted maintenance cycle, the maintenance suggestions are made to meet actual needs. The maintenance cycle optimization module includes a maintenance cycle optimization rule formulation submodule, a historical data analysis submodule, a dynamic adjustment submodule and a personalized suggestion generation submodule.

[0036] Furthermore, the maintenance cycle optimization rule formulation submodule builds an optimization rule base for the system, and formulates a basic maintenance cycle optimization framework by comprehensively analyzing the wear patterns, service life and failure modes of various vehicle components. The rule base will involve vehicle components and set optimization parameters for the situation. Through the formulation of the rule base, the system makes corrections based on the maintenance cycle predicted by the model, so that the cycle prediction is more in line with actual needs; the historical data analysis submodule identifies factors affecting the maintenance cycle based on the vehicle's historical operating data and maintenance records, and thus makes personalized adjustments to the prediction results; the dynamic adjustment submodule dynamically adjusts the predicted maintenance cycle according to the vehicle's real-time data and current status; while making dynamic adjustments, the personalized suggestion generation submodule generates specific maintenance suggestions based on all information for user reference, integrating the results from the prediction model, rule base, historical analysis and dynamic adjustment to provide users with maintenance suggestions that are more in line with actual needs.

[0037] Furthermore, the user interface module provides users with vehicle status monitoring, maintenance cycle recommendations, and historical data query functions through a visual interface and interactive functions, thereby improving user experience and system usability.

[0038] Furthermore, the user interface module has a data display function, which converts the data analyzed and predicted by the system into visual information that is easy for users to understand. It will integrate the vehicle's real-time operating data, maintenance cycle recommendations, and historical maintenance record information, and display them in the form of charts, graphic dashboards, and diagrams. For maintenance cycle recommendations, the system will display the specific recommended maintenance time or mileage, and list the parts and systems that need maintenance, and attach instructions for each recommendation. If a part is about to reach its maintenance cycle, there will be a prompt on the interface to remind the user to pay attention in time.

[0039] Beneficial effects of the present invention: The present invention integrates multi-dimensional data sources such as vehicle sensor data, historical maintenance records, driving environment data, and driving habit data, and fully explores the factors that affect the maintenance cycle. Traditional maintenance models usually rely on only a single mileage or time indicator, which cannot accurately reflect the actual degree of wear and tear and maintenance needs of the vehicle. This system can more accurately judge the actual health status of each component of the vehicle through comprehensive analysis of multiple data sources, and significantly improve the accuracy of maintenance cycle prediction. The present invention introduces a random forest algorithm model. The random forest algorithm is an algorithm based on ensemble learning. It implements voting decisions on prediction results by constructing multiple decision trees. The present invention uses the random forest algorithm to intelligently predict maintenance cycles, which effectively solves the shortcomings of traditional models in data nonlinearity and high dimensionality. The random forest algorithm has high robustness in dealing with complex and nonlinear data, and can continuously adjust the prediction results according to the dynamic changes of the input data. Unlike the traditional random forest algorithm, the present invention extends three indicators, namely Gain, Gini index , MSE, and these three indicators that can evaluate random forests are processed in the same direction and standardized. Gain, Gini index, MSE, these three indicators have the characteristic information of each decision tree in the random forest and are quantitatively expressed as a matrix of K rows and 3 columns, which greatly reduces the computational load of the system participating in the random forest. The deviation of each indicator is calculated according to the Euclidean distance to integrate the missile score of each decision tree. These score information are associated with the loss function. A regularization term is added to the loss function to prevent the model from overfitting. Unlike the traditional regularization operation, the present invention introduces a model complexity penalty coefficient and a feature selection penalty coefficient, wherein α is the model complexity penalty coefficient, which constrains the number of samples contained in the node to prevent the tree structure from expanding without limit. If α is large, the complexity penalty of the node accounts for a high proportion, and the system tends to remove sub-nodes containing fewer samples and give priority to important nodes; if α is small, the penalty for node complexity is weakened, and the system tends to retain more sub-nodes, allowing the tree structure to be more complex. β is the feature selection penalty coefficient, which reflects the importance distribution of nodes for different features. Through the feature variance To measure, increasing β will increase the penalty for feature variance, prompting the system to be more inclined to remove split nodes related to low-importance features, and reducing β will weaken the penalty for feature selection, and the system allows random forests to use more features for complex splitting. In addition, the random forest algorithm has a strong anti-interference ability for abnormal data, reducing the risk of overfitting of a single model. The introduction of this prediction model enables the system to maintain high prediction accuracy and applicability under different vehicles, different driving environments and different driving habits, and provide users with more accurate maintenance cycle recommendations. The automobile maintenance cycle intelligent optimization system based on the random forest algorithm of the present invention has significant innovation and practicality. Through multi-dimensional data sources, intelligent prediction models, dynamic optimization, personalized recommendations and remote monitoring, multiple technical innovations, it provides vehicles with accurate maintenance cycle recommendations, optimizes user experience and maintenance costs, meets the intelligent and personalized needs of modern vehicle management, and has broad application prospects in the maintenance management of future smart cars. It not only improves the safety and reliability of vehicles, but also provides car owners with more convenient and economical maintenance solutions, with important market value and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The invention is further illustrated by the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0041] Figure 1 It is a schematic diagram of the structure of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] See also Figure 1 The present invention will be further described in conjunction with the following examples.

[0044] See also Figure 1 The present invention aims to provide an intelligent optimization system for automobile preventive maintenance based on random forest algorithm, including a data acquisition module, a data preprocessing module, a model training and prediction module, a maintenance cycle optimization module and a user interface module. The data acquisition module obtains real-time data of the vehicle through vehicle-mounted sensors, GPS devices, etc. The data preprocessing module cleans, denoises and standardizes the data. The model training and prediction module uses the random forest algorithm to build an integrated model of multiple decision trees based on the preprocessed data, thereby improving the prediction accuracy and robustness. The system will input new data into the model and predict the appropriate maintenance cycle according to the driver's real-time driving behavior and vehicle health status. The maintenance cycle optimization module will dynamically adjust the recommended cycle based on the user's historical maintenance information and prediction results, and display the recommended cycle and related analysis results to the user through the user interface module, so that the user can perform maintenance at an appropriate time, thereby extending the service life of the vehicle and reducing maintenance costs.

[0045] Specifically, the data acquisition module is composed of a vehicle-mounted sensor integration unit, a global positioning system (GPS) device, an external data interface and a data storage unit. The vehicle-mounted sensor integration unit is a data acquisition module, which collects the vehicle's internal operating data in real time through engine sensors, fuel consumption sensors, temperature sensors, tire pressure sensors, brake wear sensors, battery charge sensors and speed sensors. Each sensor corresponds to different collection tasks.

[0046] Specifically, the engine sensor collects the engine's temperature, speed and workload parameters to help monitor the engine's operating status; the fuel consumption sensor records the fuel consumption per unit time to provide a basis for analyzing the vehicle's energy efficiency; the temperature sensor is used to measure the temperature of the vehicle and its key components to promptly identify temperature abnormalities; the tire pressure sensor collects pressure data from each tire to systematically analyze the tire wear condition and safety; the brake wear sensor monitors the degree of wear of the brake system to ensure the effectiveness and safety of the brake system; the battery charge sensor and speed sensor respectively record the vehicle's battery charge and speed changes. Through the coordinated work of these multiple sensors, the on-board sensor integrated unit can comprehensively and meticulously reflect the vehicle's health and operating status.

[0047] Specifically, the GPS device collects the vehicle's location information, driving route and driving speed data in real time through the global positioning system, and the system can analyze the vehicle's driving conditions.

[0048] Specifically, the vehicle's driving conditions include long-term driving on congested roads, highways or rugged mountain roads. Different road conditions will have different effects on the vehicle's consumption and wear, especially in frequent starts and stops, long-term high-speed driving or harsh road environments. The vehicle's maintenance needs will change accordingly. Therefore, the data collected by the GPS device will be input into the system together with the sensor data to comprehensively analyze the impact of environmental factors on the vehicle, thereby more accurately predicting the maintenance cycle.

[0049] Specifically, the data acquisition module also includes an external data interface for receiving information from external data sources. The operation of a vehicle is not only affected by its own state, but also significantly affected by the external environment, including climate conditions, traffic flow, and road conditions. Through the external data interface, the system can receive weather data, real-time traffic flow information, and road construction conditions information, which are obtained through cooperation with weather data providers, traffic information platforms, and road management systems, and updated in real time. The introduction of external data can not only supplement environmental factors that are difficult to capture by on-board sensors, but also enable the system to understand the operating environment of the vehicle. For vehicles that have been driving in an environment of high humidity, low temperature, or frequent rainfall for a long time, their internal parts are more likely to age or rust, resulting in the need to appropriately shorten the maintenance cycle.

[0050] Specifically, the collected data is stored in a data storage unit for subsequent analysis and prediction. The data storage unit has efficient storage and management functions and can manage real-time data and historical data in layers. Real-time data is used for current maintenance cycle prediction, and historical data is used for model training and long-term trend analysis. During the storage process, the data will be grouped and labeled according to time, data type, and source device dimensions for subsequent rapid retrieval and analysis.

[0051] Specifically, the data preprocessing module performs cleaning, denoising, and standardization preprocessing operations on the multi-source data collected by the data acquisition module. For numerical extreme values, three times the standard deviation is used for detection; for outliers, they are replaced with the mean value. For missing data, the system fills them through interpolation to avoid the missing data causing deviations in model training. Then the data denoising is completed, and the data is smoothed by optimizing the moving average method and taking the average of continuous data points to reduce abnormal fluctuations. The denoised data is standardized and normalized. Standardization adjusts the data to a standard normal distribution with the same mean and variance, and normalization scales the data proportionally to the [0,1] interval.

[0052] Specifically, the model training and prediction module completes the model training through the random forest algorithm, as follows:

[0053] Define the training set as x i =[x i1 ,x i2 ,…x im ], where D is the training set, x i is the eigenvector of the i-th group, N represents the number of samples, y i is the target maintenance cycle, x i1 ,x i2 ,…x im are the first feature vector of the i-th sample, the second feature vector of the i-th sample, and the m-th feature vector of the i-th sample, respectively. N` samples are extracted from the training set with replacement, where N`≤N, and K different subsets are generated, each with a length of N`. When constructing each decision tree, FLOOR(sqrtm) features are randomly selected from the m features for splitting, where the FLOOR function is a downward forensic function. For each node of the k-th decision tree, the optimal splitting conditions of all optional splitting points are calculated. For the current sample set of the k-th decision tree Information gain Gain is expressed as follows:

[0054]

[0055] Among them, p n Gini is the proportion of the nth sample in the sample set D, and the Gini index is defined as index ,have:

[0056]

[0057] Define mean square error MSE:

[0058]

[0059] in, for Maintenance cycle, is calculated in the random forest algorithm The greater the information gain, the greater the purity of the feature partition node. If maximization is defined as forward conversion and minimization is defined as reverse conversion, in order to unify the evaluation index, the information gain is reversed as Constructing the indicator structure For K decision trees, there is a K*3-dimensional standardized evaluation matrix A*:

[0060]

[0061] in, is the inverse element of information gain in the standardized evaluation matrix of the first decision tree, is the inverse element of information gain in the standardized evaluation matrix of the second decision tree, Gini is the inverse element of the information gain in the standardized evaluation matrix of the Kth decision tree. index (1) is the Gini index element in the standardized evaluation matrix of the first decision tree, Gini index (2) is the Gini index element in the standardized evaluation matrix of the second decision tree, Gini index (K) is the Gini index element in the standardized evaluation matrix of the Kth decision tree, is the mean square error element in the standardized evaluation matrix of the first decision tree, is the mean square error element in the standardized evaluation matrix of the second decision tree, Define the maximum value vector A* of the standardized matrix as the mean square error element in the standardized evaluation matrix of the Kth decision tree + :

[0062]

[0063] Define the normalized matrix minimum vector A* -

[0064]

[0065] Then the evaluation index SCORE of the kth decision tree is k Calculated as:

[0066]

[0067] in:

[0068]

[0069] Specifically, the prediction performance is improved by constructing a node loss function, which is expressed as follows:

[0070]

[0071] Among them, Loss(k) is the loss function of the kth decision tree. In order to minimize the weighted loss of the entire decision tree, the loss increment ΔLoss is calculated. k :

[0072]

[0073] Among them, D k< D k The subset of left child nodes under the splitting rule π of the decision tree in D k> D k The right child node subset under the splitting rule π of the decision tree in has {D k< ∩D k> =D k , Loss(k<) is D k< The loss function, Loss(k>) is D k> The loss function of the regularization term is introduced to prevent the model from overfitting:

[0074]

[0075] Among them, Loss is the loss function of random forest, α is the model complexity penalty coefficient, which constrains the number of samples contained in the node to prevent the tree structure from expanding without limit. If α is large, the complexity penalty of the node is high, and the system tends to remove sub-nodes containing fewer samples and give priority to important nodes; if α is small, the penalty for node complexity is weakened, and the system tends to retain more sub-nodes, allowing the tree structure to be more complex. β is the feature selection penalty coefficient, which reflects the importance distribution of nodes for different features. To measure, increasing β will increase the penalty for feature variance, prompting the system to be more inclined to remove split nodes related to low-importance features, while reducing β will weaken the penalty for feature selection, allowing the random forest to use more features for complex splitting, γ m is the feature weight, which indicates the importance of the feature to the model prediction.

[0076] Specifically, the maintenance cycle optimization module optimizes the maintenance cycle predicted by the model to provide accurate and personalized maintenance recommendations. It dynamically adjusts and optimizes the initially predicted maintenance cycle to make the maintenance recommendations meet actual needs. The maintenance cycle optimization module includes a maintenance cycle optimization rule formulation sub-module, a historical data analysis sub-module, a dynamic adjustment sub-module, and a personalized recommendation generation sub-module.

[0077] Specifically, the maintenance cycle optimization rule formulation submodule builds the system's optimization rule base, and formulates a basic maintenance cycle optimization framework by comprehensively analyzing the wear patterns, service life and failure modes of various vehicle components. The rule base will involve vehicle components and set optimization parameters according to the situation. Through the formulation of the rule base, the system makes corrections based on the maintenance cycle predicted by the model, so that the cycle prediction is more in line with actual needs; the historical data analysis submodule identifies factors affecting the maintenance cycle based on the vehicle's historical operating data and maintenance records, and thus makes personalized adjustments to the prediction results; the dynamic adjustment submodule dynamically adjusts the predicted maintenance cycle according to the vehicle's real-time data and current status; while making dynamic adjustments, the personalized suggestion generation submodule generates specific maintenance suggestions based on all information for user reference, integrating the results from the prediction model, rule base, historical analysis and dynamic adjustment to provide users with maintenance suggestions that are more in line with actual needs.

[0078] Specifically, the user interface module provides users with vehicle status monitoring, maintenance cycle recommendations, and historical data query functions through a visual interface and interactive functions, thereby improving user experience and system usability.

[0079] Specifically, the user interface module has a data display function, which converts the data analyzed and predicted by the system into visual information that is easy for users to understand. It will integrate the vehicle's real-time operating data, maintenance cycle recommendations, and historical maintenance record information, and display them in the form of charts, graphic dashboards, and diagrams. For maintenance cycle recommendations, the system will display the specific recommended maintenance time or mileage, and list the components and systems that require maintenance, and attach instructions for each recommendation. If a component is about to reach its maintenance cycle, there will be a prompt on the interface to remind the user to pay attention in time.

[0080] Beneficial effects of the present invention: The present invention integrates multi-dimensional data sources such as vehicle sensor data, historical maintenance records, driving environment data, and driving habit data, and fully explores the factors that affect the maintenance cycle. Traditional maintenance models usually rely on only a single mileage or time indicator, which cannot accurately reflect the actual degree of wear and tear and maintenance needs of the vehicle. This system can more accurately judge the actual health status of each component of the vehicle through comprehensive analysis of multiple data sources, and significantly improve the accuracy of maintenance cycle prediction. The present invention introduces a random forest algorithm model. The random forest algorithm is an algorithm based on ensemble learning. It implements voting decisions on prediction results by constructing multiple decision trees. The present invention uses the random forest algorithm to intelligently predict maintenance cycles, which effectively solves the shortcomings of traditional models in data nonlinearity and high dimensionality. The random forest algorithm has high robustness in dealing with complex and nonlinear data, and can continuously adjust the prediction results according to the dynamic changes of the input data. Unlike the traditional random forest algorithm, the present invention extends three indicators, namely Gain, Gini index, MSE, and these three indicators that can evaluate random forests are processed in the same direction and standardized. Gain, Gini index , MSE, these three indicators have the characteristic information of each decision tree in the random forest and are quantitatively expressed as a matrix of K rows and 3 columns, which greatly reduces the computational load of the system participating in the random forest. The deviation of each indicator is calculated according to the Euclidean distance to integrate the missile score of each decision tree. These score information are associated with the loss function. A regularization term is added to the loss function to prevent the model from overfitting. Unlike the traditional regularization operation, the present invention introduces a model complexity penalty coefficient and a feature selection penalty coefficient, wherein α is the model complexity penalty coefficient, which constrains the number of samples contained in the node to prevent the tree structure from expanding without limit. If α is large, the complexity penalty of the node accounts for a high proportion, and the system tends to remove sub-nodes containing fewer samples and give priority to important nodes; if α is small, the penalty for node complexity is weakened, and the system tends to retain more sub-nodes, allowing the tree structure to be more complex. β is the feature selection penalty coefficient, which reflects the importance distribution of nodes for different features. Through the feature variance To measure, increasing β will increase the penalty for feature variance, prompting the system to be more inclined to remove split nodes related to low-importance features, and reducing β will weaken the penalty for feature selection, and the system allows random forests to use more features for complex splitting. In addition, the random forest algorithm has a strong anti-interference ability for abnormal data, reducing the risk of overfitting of a single model. The introduction of this prediction model enables the system to maintain high prediction accuracy and applicability under different vehicles, different driving environments and different driving habits, and provide users with more accurate maintenance cycle recommendations. The automobile maintenance cycle intelligent optimization system based on the random forest algorithm of the present invention has significant innovation and practicality. Through multi-dimensional data sources, intelligent prediction models, dynamic optimization, personalized recommendations and remote monitoring, multiple technical innovations, it provides vehicles with accurate maintenance cycle recommendations, optimizes user experience and maintenance costs, meets the intelligent and personalized needs of modern vehicle management, and has broad application prospects in the maintenance management of future smart cars. It not only improves the safety and reliability of vehicles, but also provides car owners with more convenient and economical maintenance solutions, with important market value and social benefits.

[0081] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to replace some of the technical features therein. Any modification, replacement, and improvement made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent optimization system for automobile preventive maintenance based on random forest algorithm, including a data acquisition module, a data preprocessing module, a model training and prediction module, a maintenance cycle optimization module and a user interface module. The data acquisition module obtains real-time data of the vehicle through vehicle-mounted sensors, GPS devices, etc. The data preprocessing module cleans, denoises and standardizes the data. The model training and prediction module uses the random forest algorithm to build an integrated model of multiple decision trees based on the preprocessed data, thereby improving the prediction accuracy and robustness. The system will input new data into the model and predict the appropriate maintenance cycle according to the driver's real-time driving behavior and vehicle health status. The maintenance cycle optimization module will dynamically adjust the recommended cycle based on the user's historical maintenance information and prediction results, and display the recommended cycle and related analysis results to the user through the user interface module, so that the user can perform maintenance at an appropriate time, thereby extending the service life of the vehicle and reducing maintenance costs.

2. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 1 is characterized in that: The data acquisition module is composed of an on-board sensor integration unit, a global positioning system (GPS) device, an external data interface and a data storage unit. The on-board sensor integration unit is a data acquisition module. It collects the internal operation data of the vehicle in real time through engine sensors, fuel consumption sensors, temperature sensors, tire pressure sensors, brake wear sensors, battery power sensors and speed sensors. Each sensor corresponds to a different collection task. The GPS device collects the vehicle's location information, driving route and driving speed data in real time through the global positioning system. The system can analyze the vehicle's driving conditions. The data collected by the GPS device will be input into the system together with the sensor data to analyze the impact of environmental factors on the vehicle, so as to more accurately predict the maintenance cycle. The data acquisition module also includes an external data interface for receiving information from an external data source. The operation of the vehicle is not only affected by its own state, but also significantly affected by the external environment. Through the external data interface, the system can receive weather data, real-time traffic flow information and road construction road condition information, which are obtained through cooperation with weather data providers, traffic information platforms and road management systems, and updated in real time. The introduction of external data can not only supplement the environmental factors that are difficult to capture by on-board sensors, but also enable the system to understand the operating environment of the vehicle.

3. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 2 is characterized in that: The collected data is stored in the data storage unit for subsequent analysis and prediction. The data storage unit has efficient storage and management functions, and can manage real-time data and historical data in layers. Real-time data is used for current maintenance cycle prediction, and historical data is used for model training and long-term trend analysis. During the storage process, the data will be grouped and labeled according to time, data type, and source device dimensions for subsequent rapid retrieval and analysis.

4. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 1 is characterized in that: The data preprocessing module performs cleaning, denoising, and standardization preprocessing operations on the multi-source data collected by the data acquisition module. For numerical extreme values, it is detected by three times the standard deviation; for outliers, they are replaced with the mean value. For missing data, the system fills them through interpolation to avoid the missing data causing deviations in model training. Then the data denoising is completed, and the data is smoothed by optimizing the moving average method and taking the average of continuous data points to reduce abnormal fluctuations. The denoised data is standardized and normalized. Standardization adjusts the data to a standard normal distribution with the same mean and variance, and normalization scales the data proportionally to the [0, 1] interval.

5. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 1 is characterized in that: The model training and prediction module completes the model training through the random forest algorithm, as follows: Define the training set as x i =[x i1 ,x i2 , …x im ], where D is the training set, x i is the eigenvector of the i-th group, N represents the number of samples, y i is the target maintenance cycle, x i1 ,x i2 ,...x im are the first feature vector of the i-th sample, the second feature vector of the i-th sample, and the m-th feature vector of the i-th sample, respectively. N` samples are extracted from the training set with replacement, where N`≤N, and K different subsets are generated, each with a length of N`. When constructing each decision tree, FLOOR(sqrtm) features are randomly selected from the m features for splitting, where the FLOOR function is a downward forensic function. For each node of the k-th decision tree, the optimal splitting conditions of all optional splitting points are calculated. For the current sample set of the k-th decision tree Information gain Gain is expressed as follows: Among them, p n Gini is the proportion of the nth sample in the sample set D, and the Gini index is defined as index ,have: Define mean square error MSE: in, for Maintenance cycle, is calculated in the random forest algorithm The greater the information gain, the greater the purity of the feature partition node. If maximization is defined as forward conversion and minimization is defined as reverse conversion, in order to unify the evaluation index, the information gain is reversed as Constructing the indicator structure For K decision trees, there is a K*3-dimensional standardized evaluation matrix A*: in, is the inverse element of information gain in the standardized evaluation matrix of the first decision tree, is the inverse element of information gain in the standardized evaluation matrix of the second decision tree, Gini is the inverse element of the information gain in the standardized evaluation matrix of the Kth decision tree. index (1) is the Gini index element in the standardized evaluation matrix of the first decision tree, Gini index (2) is the Gini index element in the standardized evaluation matrix of the second decision tree, Gini index (K) is the Gini index element in the standardized evaluation matrix of the Kth decision tree, is the mean square error element in the standardized evaluation matrix of the first decision tree, is the mean square error element in the standardized evaluation matrix of the second decision tree, Define the maximum value vector A* of the standardized matrix as the mean square error element in the standardized evaluation matrix of the Kth decision tree + : Define the normalized matrix minimum vector A* - Then the evaluation index SCORE of the kth decision tree is k Calculated as: in:

6. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 5 is characterized in that: The prediction performance is improved by constructing a node loss function, which is expressed as follows: Among them, Loss(k) is the loss function of the kth decision tree. In order to minimize the weighted loss of the entire decision tree, the loss increment ΔLoss is calculated. k : Among them, D k＜ D k The subset of left child nodes under the splitting rule π of the decision tree in D k＞ D k The right child node subset under the splitting rule π of the decision tree in has {D k＜ ∩D k＞ =D k , Loss(k<) is D k＜ The loss function of Loss(k>) is D k＞ The loss function of the regularization term is introduced to prevent the model from overfitting: Among them, Loss is the loss function of random forest, α is the model complexity penalty coefficient, which constrains the number of samples contained in the node to prevent the tree structure from expanding without limit. If α is large, the complexity penalty of the node is high, and the system tends to remove sub-nodes containing fewer samples and give priority to important nodes; if α is small, the penalty for node complexity is weakened, and the system tends to retain more sub-nodes, allowing the tree structure to be more complex. β is the feature selection penalty coefficient, which reflects the importance distribution of nodes for different features. To measure, increasing β will increase the penalty for feature variance, prompting the system to be more inclined to remove split nodes related to low-importance features, while reducing β will weaken the penalty for feature selection, allowing the random forest to use more features for complex splitting, γ m is the feature weight, which indicates the importance of the feature to the model prediction.

7. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 1 is characterized in that: The maintenance cycle optimization module optimizes the maintenance cycle predicted by the model to provide accurate and personalized maintenance suggestions. It dynamically adjusts and optimizes the initially predicted maintenance cycle to make the maintenance suggestions meet actual needs. The maintenance cycle optimization module includes a maintenance cycle optimization rule formulation submodule, a historical data analysis submodule, a dynamic adjustment submodule and a personalized suggestion generation submodule.

8. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 7 is characterized in that: The maintenance cycle optimization rule formulation submodule builds the system's optimization rule base, and formulates a basic maintenance cycle optimization framework by comprehensively analyzing the wear patterns, service life and failure modes of various vehicle components. The rule base will involve vehicle components and set optimization parameters according to the situation. Through the formulation of the rule base, the system makes corrections based on the maintenance cycle predicted by the model, so that the cycle prediction is more in line with actual needs; the historical data analysis submodule identifies factors affecting the maintenance cycle based on the vehicle's historical operating data and maintenance records, and thus makes personalized adjustments to the prediction results; the dynamic adjustment submodule dynamically adjusts the predicted maintenance cycle according to the vehicle's real-time data and current status; while making dynamic adjustments, the personalized suggestion generation submodule generates specific maintenance suggestions based on all information for user reference, integrating the results from the prediction model, rule base, historical analysis and dynamic adjustment to provide users with maintenance suggestions that are more in line with actual needs.

9. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 1 is characterized in that: The user interface module provides users with vehicle status monitoring, maintenance cycle recommendations, and historical data query functions through a visual interface and interactive functions, improving user experience and system usability.

10. The automobile preventive maintenance intelligent optimization system based on random forest algorithm according to claim 1, characterized in that: The user interface module has a data display function, which converts the data analyzed and predicted by the system into visual information that is easy for users to understand. It will integrate the vehicle's real-time operating data, maintenance cycle recommendations, and historical maintenance record information, and display them in the form of charts, graphic dashboards, and diagrams. For maintenance cycle recommendations, the system will display the specific recommended maintenance time or mileage, and list the components and systems that require maintenance, and attach instructions for each recommendation. If a component is about to reach its maintenance cycle, there will be a prompt on the interface to remind the user to pay attention in time.

Images