A method for predicting global automobile maintenance demand based on correlation perception and fusion

Abstract

The present invention discloses a method for predicting global automobile maintenance demand based on correlation perception and fusion, which relates to the field of automobile maintenance technology, including: constructing an automobile maintenance demand prediction model based on correlation perception and fusion, inputting historical maintenance records of the automobile to be maintained into the automobile maintenance demand prediction model based on correlation perception and fusion for prediction, and obtaining a prediction result; wherein the automobile maintenance demand prediction model based on correlation perception and fusion includes a static correlation perception module for maintenance items, a dynamic correlation perception module for maintenance items, and a correlation fusion prediction module for maintenance items. The present invention provides an accurate demand prediction method for the automobile maintenance field, which helps to reduce use risks and improve automobile use quality.

Description

Technical Field

[0001] The present invention relates to the field of automobile maintenance technology, and in particular to a method for predicting automobile global maintenance demand based on correlation perception and fusion. Background Art

[0002] With the rapid growth of global car ownership, cars are playing an increasingly important role in people's lives. The car maintenance industry, which is closely related to cars, has become an important livelihood service industry related to the quality of life of the people. Regular maintenance of cars is a vital part of its entire life cycle. If a car lacks necessary maintenance operations, it may cause failures, increase the risk of accidents, and bring unnecessary economic losses. Accurately predicting the maintenance needs of cars can reduce the cost and downtime of car use, reduce the risk of car use, ensure safe operation and optimize maintenance plans. It is of great significance to improve the quality of people's car life.

[0003] In actual automobile maintenance practice, different maintenance items are often interdependent and interrelated. For example, in order to ensure the optimal working state of the engine, the oil filter needs to be replaced at the same time when the engine oil is replaced; to extend the life of the tire, the tire needs to be rotated and balanced at the same time; similarly, to maintain the good performance of the braking system, the brake pads and brake fluid need to be replaced. In addition, in the actual use of the car, the multiple maintenance activities performed are also significantly correlated with each other, and the previous maintenance items affect the subsequent maintenance needs. For example, if the brake pads are found to have a certain degree of wear during the initial maintenance but have not reached the replacement standard, the subsequent maintenance may require the replacement of the brake pads and the inspection of the brake discs; the fuel filter was replaced during the previous maintenance, and the fuel injectors need to be checked in the subsequent maintenance activities to prevent potential blockages.

[0004] With the booming development of the automotive industry, the prediction technology for automobile maintenance needs has also made significant progress. Initially, it relied on simple time-based prediction methods, and later developed into more sophisticated prediction models that combined professional knowledge and data statistics. Now, with the rapid growth of automobile condition data, maintenance demand prediction based on machine learning and deep learning has gradually become a hot research topic and has significantly improved the accuracy of predictions. Overall, automobile maintenance demand prediction technology can be roughly divided into the following three categories: prediction methods based on traditional methods, machine learning, and deep learning. Although the above methods have shown certain performance in predicting automobile maintenance needs, they can only predict the maintenance needs of a part of the car, and cannot predict the overall maintenance needs of the car. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a method for predicting automobile global maintenance demand based on correlation perception and fusion, which helps to reduce the risk of use and improve the quality of automobile use.

[0006] To achieve the above objectives, the present invention provides a method for predicting global automobile maintenance demand based on correlation perception and fusion, comprising:

[0007] Constructing an automobile maintenance demand prediction model based on correlation perception and fusion, inputting historical maintenance records of the automobile to be maintained into the automobile maintenance demand prediction model based on correlation perception and fusion for prediction, and obtaining a prediction result;

[0008] Among them, the automobile maintenance demand prediction model based on correlation perception and fusion includes a static correlation perception module of maintenance items, a dynamic correlation perception module of maintenance items, and a correlation fusion prediction module of maintenance items.

[0009] Preferably, the static correlation perception module of the maintenance items is used to obtain the static correlation representation results of all maintenance items based on the co-occurrence matrix of the maintenance items, and finally obtain the static perception results of all historical maintenance records, wherein the co-occurrence matrix is ​​used to measure the frequency of simultaneous occurrence of global maintenance items;

[0010] The dynamic correlation perception module of the maintenance items is used to obtain the hidden state of the historical maintenance items through the GRU neural network model, obtain the dynamic representation results of the previous maintenance items based on the hidden state, and finally obtain the dynamic perception results of all historical maintenance records;

[0011] The relevance fusion prediction module of the maintenance project is used to use the attention mechanism to fuse the static perception results and the dynamic perception results of the historical maintenance projects, obtain the relevance fusion representation results, and predict the maintenance project required next time based on the fusion representation results.

[0012] Preferably, the method of representing the co-occurrence matrix of the maintenance items by means of a matrix is:

[0013] ;

[0014] In the formula, A ij is the adjacency matrix; t ij is the frequency of co-occurrence; λ is the threshold; Code collection for all automobile maintenance items; The type of maintenance project; ,for The total frequency of meeting the conditions; p i is the i-th type maintenance item; p j It is a maintenance item of Category J.

[0015] Preferably, the impact of the correlation between different maintenance items is considered, and the historical maintenance items of each vehicle are , using static correlation to represent the function The static correlation of all maintenance items is obtained to represent the learning results ;

[0016] Wherein, the static correlation represents the function for:

[0017] ;

[0018] In the formula, , function f means filtering out all zero row vectors, T means the number of maintenance times of the car, is a diagonal matrix and diag() is a function used to construct a diagonal matrix.

[0019] Preferably, after obtaining the learning results of the static correlation representation of the maintenance items, and Input into the attention mechanism to obtain the static perception result O of all maintenance records t , in the static perception results of previous maintenance records O t On this basis, static perception results of all the historical maintenance records are obtained.

[0020] Preferably, the static perception results of the previous maintenance records are obtained. t The method is:

[0021] Specifically:

[0022] ;

[0023] ;

[0024] Where d is the dimension of attention; are the attention weights respectively; Q is the query vector, K is the key vector, V is the value vector; R t M t The static correlation of M t The maintenance items for the tth maintenance of a car.

[0025] Preferably, the working process of the dynamic correlation perception module of the maintenance project includes:

[0026] Use the GRU neural network model to obtain the hidden state of all maintenance records , which will contain the hidden status of the original maintenance record information and the accumulated record information As the attention score of all maintenance records, and then use Softmax as the activation function to calculate the attention weight of all maintenance records ;

[0027] Representing functions using dynamic dependencies The dynamic correlation of each historical maintenance record is obtained to represent the learning result , the dynamic correlation represents the learning result Summing and obtaining dynamic perception results of all the historical maintenance records;

[0028] Wherein, the dynamic correlation represents the learning result Contains all original maintenance record information, all hidden and cumulative maintenance record information, and all weight information of maintenance records.

[0029] Preferably, the dynamic correlation expression function It is expressed as:

[0030] ;

[0031] In the formula, is the Kronecker product operation; is element-by-element multiplication; M t is the maintenance item of the tth maintenance of a car; B t Calculate the process variable, N t is the hidden state of the t-th maintenance record, E t Record the attention weight for the tth maintenance, The type of maintenance project.

[0032] Preferably, the method for fusing the static perception result of the historical maintenance project with the dynamic perception result is:

[0033] ;

[0034] In the formula, Y is the correlation perception fusion result, S is the static perception result of the historical maintenance record, and D is the dynamic perception result of the historical maintenance record;

[0035] The specific method for obtaining the T+1 maintenance items is as follows:

[0036] ;

[0037] In the formula, and are weight and bias respectively, is the dimension of Y, The type of maintenance project; This is the T+1 maintenance project.

[0038] Preferably, the global objective function of the automobile maintenance demand prediction model based on correlation perception and fusion is a binary cross entropy loss function, specifically:

[0039] ;

[0040] In the formula, For maintenance projects i The prediction result, L is the binary cross entropy loss function, is the type of maintenance item, and T is the number of maintenance times of the car.

[0041] Compared with the prior art, the present invention has the following advantages and technical effects:

[0042] The proposed model takes into account the associations between different maintenance items using a static correlation perception method based on a co-occurrence matrix. In order to capture dependencies on time series, the model combines a dynamic correlation perception method based on attention weights. By combining the two, a comprehensive attention mechanism that integrates static and dynamic information is developed. The prediction model has superior performance over existing technologies, provides an accurate demand prediction method for the automotive maintenance field, and helps reduce usage risks and improve the quality of automotive use. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0044] Figure 1 Schematic diagram of the structure of an automobile maintenance demand prediction model based on correlation perception and fusion according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0046] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0047] The present invention proposes a method for predicting global automobile maintenance demand based on correlation perception and fusion. Figure 1 ,include:

[0048] Constructing an automobile maintenance demand prediction model based on correlation perception and fusion, inputting historical maintenance records of the automobile to be maintained into the automobile maintenance demand prediction model based on correlation perception and fusion for prediction, and obtaining a prediction result;

[0049] Among them, the automobile maintenance demand prediction model based on correlation perception and fusion includes a static correlation perception module of maintenance projects, a dynamic correlation perception module of maintenance projects, and a correlation fusion prediction module of maintenance projects.

[0050] The static correlation perception module of maintenance items is used to obtain the static correlation representation results of previous maintenance items based on the co-occurrence matrix of maintenance items, and finally obtain the static perception results of all historical maintenance records. Among them, the co-occurrence matrix is ​​used to measure the frequency of simultaneous occurrence of global maintenance items.

[0051] Specifically, the method of representing the co-occurrence matrix of maintenance items through the adjacency matrix is:

[0052] ;

[0053] In the formula, A ij is the adjacency matrix; t ij is the frequency of co-occurrence; λ is the threshold; Code collection for all automobile maintenance items; The type of maintenance project; ,for The total frequency of meeting the conditions; p i is the i-th type maintenance item; p j It is a maintenance item of Category J.

[0054] Considering the impact of the correlation between different maintenance items, the historical maintenance items of each vehicle are , using static correlation to represent the function The static correlation of all maintenance items is obtained to represent the learning results ;

[0055] Among them, static correlation represents the function for:

[0056] ;

[0057] In the formula, , function f means filtering out all zero row vectors, T means the number of maintenance times of the car, is a diagonal matrix and diag() is a function used to construct a diagonal matrix.

[0058] After obtaining the learning results of the static correlation representation of the maintenance items, and Input into the attention mechanism to obtain the static perception result O of all maintenance records t , in the static perception results of previous maintenance records O t On this basis, static perception results of all the historical maintenance records are obtained.

[0059] Get the static perception results of all maintenance records O t The method is:

[0060] ;

[0061] ;

[0062] Where d is the dimension of attention; are the attention weights respectively; Q is the query vector, K is the key vector, V is the value vector; R t M t The static correlation of M t The maintenance items for the tth maintenance of a car.

[0063] The dynamic correlation perception module of the maintenance items is used to obtain the hidden state of the historical maintenance items through the GRU neural network model, obtain the dynamic representation results of the previous maintenance items based on the hidden state, and finally obtain the dynamic perception results of all historical maintenance records;

[0064] Specifically, the working process of the dynamic correlation perception module of the maintenance project includes:

[0065] Use the GRU neural network model to obtain the hidden state of all maintenance records , which will contain the hidden status of the original maintenance record information and the accumulated record information As the attention score of all maintenance records, and then use Softmax as the activation function to calculate the attention weight of all maintenance records ;

[0066] Representing functions using dynamic dependencies The dynamic correlation of each historical maintenance record is obtained to represent the learning result , the dynamic correlation represents the learning result The sum is taken to obtain the dynamic perception results of all the historical maintenance records.

[0067] Wherein, the dynamic correlation represents the learning result Contains all original maintenance record information, all hidden and cumulative maintenance record information, and all weight information of maintenance records.

[0068] Dynamic correlation representation function It is expressed as:

[0069] ;

[0070] In the formula, is the Kronecker product operation; is element-by-element multiplication; M t is the maintenance item of the tth maintenance of a car; B t Calculate the process variable, N t is the hidden state of the t-th maintenance record, E t Record the attention weight for the tth maintenance, The type of maintenance project.

[0071] The correlation fusion prediction module of the maintenance project is used to use the attention mechanism to fuse the static perception results and the dynamic perception results of the historical maintenance projects, obtain the correlation fusion representation results, and predict the maintenance project required next time based on the fusion representation results.

[0072] Specifically, the method of fusing the static perception result of the historical maintenance project with the dynamic perception result is:

[0073] ;

[0074] In the formula, Y is the correlation perception fusion result, S is the static perception result of the historical maintenance record, and D is the dynamic perception result of the historical maintenance record;

[0075] The specific method for obtaining the T+1 maintenance items is as follows:

[0076] ;

[0077] In the formula, and are weight and bias respectively, is the dimension of Y, The type of maintenance project; This is the T+1 maintenance project.

[0078] The global objective function of the automobile maintenance demand prediction model based on correlation perception and fusion is a binary cross entropy loss function, specifically:

[0079] ;

[0080] In the formula, For maintenance projects i The prediction result, L is the binary cross entropy loss function, is the type of maintenance item, and T is the number of maintenance times of the car.

[0081] In order to more clearly express the technical solution of the present invention, the following specific embodiments are provided to introduce the solution:

[0082] During the maintenance of a car, multiple maintenance items are required for one maintenance. For example, replacing the oil filter and the air filter will appear in the maintenance record of the car at the same time, which means that different maintenance items are correlated during the maintenance process. In order to make full use of the correlation between different maintenance items, a static correlation perception module for car maintenance items based on the co-occurrence matrix is ​​proposed.

[0083] All maintenance items that occur simultaneously during the automobile maintenance process are combined to form a global item co-occurrence graph G. This structure reflects the correlation between different maintenance items. The nodes of G are the maintenance items in P. If a maintenance item is coded If both of them appear in the maintenance record of a vehicle, two equal weights will be added to G. and , then calculate The frequency of occurrence in the maintenance records of all cars is t ij It is used to further calculate the edge weights. In addition, it is necessary to detect important and common item pairs, for p i , define a threshold λ to filter out low-frequency combinations and obtain a qualified combination .make For p i The total probability of meeting the conditions, using the adjacency matrix represents G, The definition of is shown in formula (1).

[0084] ;

[0085] In this embodiment, It is designed to be symmetrical to show the impact between the two projects. As a static diagram, It measures the frequency of global maintenance items appearing at the same time. However, different maintenance items appear and disappear at different stages. Even if a certain maintenance item does not appear in the current maintenance record, due to the correlation between maintenance items, related maintenance items may appear in the future. Therefore, in order to consider the impact of the correlation between different maintenance items, the historical maintenance items of each vehicle are , using static correlation to represent the function Get the static correlation of each historical maintenance project to represent the learning result . The definition of is shown in formula (2).

[0086] ;

[0087] In the formula, , function f means filtering out all row vectors that are all zero. After obtaining the static correlation representation results of all maintenance records, and Input into the attention mechanism to obtain the static perception result O of all maintenance records t , the specific implementation is shown in formula (3).

[0088] ;

[0089] The attention mechanism in the above formula is defined as shown in formula (4);

[0090] ;

[0091] Where d is the dimension of attention; are the weights of attention respectively; t Based on the historical maintenance records, the static perception results are obtained. .

[0092] In the maintenance process of the entire life cycle of a car, there is not only correlation between different maintenance items, but also correlation between maintenance items carried out at different stages. The early maintenance items will affect the development process of the car's faults, thereby affecting the subsequent maintenance items. In order to make full use of the dynamic correlation between maintenance items at different stages, a dynamic correlation perception module for automobile maintenance items with attention weights is proposed.

[0093] First, use the GRU neural network to obtain the hidden state of all maintenance records , Each element in not only obtains the information of the original corresponding maintenance record, but also contains the accumulated information of the historical maintenance records. The specific implementation process is shown in formula (5);

[0094] ;

[0095] The calculation process of the GRU neural network in the above formula is shown in formula (6);

[0096] ;

[0097] in, To reset the activation result of the gate, Calculate the weights for the first intermediate process, Calculate weights for the second intermediate process, It is the hidden status of the previous maintenance record. Calculate the bias for the first intermediate process, is the hidden status of the current maintenance record. Calculate weights for the third intermediate process, Calculate weights for the fourth intermediate process, Calculate the bias for the second intermediate process, To update the gate activation result, Calculate weights for the fourth intermediate process, Calculate the weights for the fifth intermediate process, Calculate the bias for the third intermediate process.

[0098] During the car maintenance process, the maintenance personnel will pay attention to the car's previous maintenance records and decide the maintenance items that the car needs to undergo based on the historical maintenance information. To simulate the above process, the hidden state of the original maintenance record information and the accumulated record information will be included. As the attention score of all maintenance records, and then use Softmax as the activation function to calculate the attention weight of all maintenance records , the specific implementation process is shown in formula (7);

[0099] ;

[0100] After calculating the attention weights of all maintenance records, the dynamic correlation representation function is used The dynamic correlation of each historical maintenance record is obtained to represent the learning result .

[0101] The definition of is shown in formula (8).

[0102] ;

[0103] In the formula, is the Kronecker product operation; is element-wise multiplication.

[0104] Dynamic correlation represents learning results It not only contains the original maintenance records, but also contains the hidden and accumulated maintenance records, as well as the weight information of the maintenance records. The dynamic perception results of historical maintenance records are obtained by summing up .

[0105] After the static perception and dynamic perception processes, the static correlation perception results S and dynamic correlation perception results D of the historical maintenance records are obtained respectively. In order to fully fuse the two correlation perception results, the attention mechanism shown in formula (4) is used to obtain the correlation perception fusion result Y. The implementation process is shown in formula (9);

[0106] ;

[0107] The T+1th maintenance item is obtained based on the correlation perception fusion result Y. The specific implementation process is shown in formula (10).

[0108] ;

[0109] In the formula, and are weight and bias respectively, is the dimension of Y. In order to maximize the robustness of the model, a dropout operation is performed before Y is predicted.

[0110] The automobile maintenance demand prediction model based on correlation perception and fusion is trained to predict the T+1th maintenance item of each vehicle. The global objective function is the binary cross entropy loss function, as shown in Equation (11).

[0111] (11);

[0112] In the formula, For maintenance projects i The prediction result, L is the binary cross entropy loss function, is the type of maintenance item, and T is the number of maintenance times of the car.

[0113] The model proposed in this paper takes into account the association between different maintenance items by using a static correlation perception method based on a co-occurrence matrix. Furthermore, in order to capture the dependencies on the time series, the model combines a dynamic correlation perception method based on attention weights. Through the combination of the two, a comprehensive attention mechanism that integrates static and dynamic information is developed. The prediction model has superior performance over the existing technology, provides an accurate demand prediction method for the automotive maintenance field, and helps to reduce the use risk and improve the quality of automotive use.

[0114] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting global automobile maintenance demand based on correlation perception and fusion, characterized in that: include: Constructing an automobile maintenance demand prediction model based on correlation perception and fusion, inputting historical maintenance records of the automobile to be maintained into the automobile maintenance demand prediction model based on correlation perception and fusion for prediction, and obtaining a prediction result; The automobile maintenance demand prediction model based on correlation perception and fusion includes a static correlation perception module of maintenance items, a dynamic correlation perception module of maintenance items, and a correlation fusion prediction module of maintenance items; The static correlation perception module of the maintenance items is used to obtain the static correlation representation results of all maintenance items based on the co-occurrence matrix of the maintenance items, and finally obtain the static perception results of all historical maintenance records, wherein the co-occurrence matrix is ​​used to measure the frequency of simultaneous occurrence of global maintenance items; The dynamic correlation perception module of the maintenance items is used to obtain the hidden state of the historical maintenance items through the GRU neural network model, obtain the dynamic representation results of the previous maintenance items based on the hidden state, and finally obtain the dynamic perception results of all historical maintenance records; The relevance fusion prediction module of the maintenance project is used to fuse the static perception results and the dynamic perception results of the historical maintenance projects by using the attention mechanism to obtain a relevance fusion representation result, and predict the maintenance project required next time based on the fusion representation result; The method of expressing the co-occurrence matrix of the maintenance items by means of a matrix is: ； In the formula, A ij is the adjacency matrix; t ij is the frequency of co-occurrence; λ is the threshold; Code collection for all automobile maintenance items; The type of maintenance project; ,for The total frequency of meeting the conditions; p i is the i-th type maintenance item; p j It is a maintenance project of category j; Considering the impact of the correlation between different maintenance items, the historical maintenance items of each vehicle are , using static correlation to represent the function The static correlation of all maintenance items is obtained to represent the learning results ; Wherein, the static correlation represents the function for: ； In the formula, , function f means filtering out all zero row vectors, T means the number of maintenance times of the car, is a diagonal matrix, diag() is a function used to construct a diagonal matrix; After obtaining the learning results of the static correlation representation of the maintenance items, and Input into the attention mechanism to obtain the static perception result O of all maintenance records t , in the static perception results of previous maintenance records O t On the basis of, obtaining static perception results of all the historical maintenance records; Obtain the static perception results of the previous maintenance records t The method is: Specifically: ； ； Where d is the dimension of attention; are the attention weights respectively; Q is the query vector, K is the key vector, V is the value vector; R t M t The static correlation of M t The maintenance items for the tth maintenance of a car; The working process of the dynamic correlation perception module of the maintenance project includes: Use the GRU neural network model to obtain the hidden state of all maintenance records , which will contain the hidden status of the original maintenance record information and the accumulated record information As the attention score of all maintenance records, and then use Softmax as the activation function to calculate the attention weight of all maintenance records ; Said Each element in not only obtains the original corresponding maintenance record information, but also contains the cumulative information of historical maintenance records. The specific implementation process is as follows: ； Among them, the GRU neural network calculation process is as follows: ； In the formula, To reset the activation result of the gate, Calculate the weights for the first intermediate process, Calculate weights for the second intermediate process, It is the hidden status of the previous maintenance record. Calculate the bias for the first intermediate process, is the hidden status of the current maintenance record. Calculate weights for the third intermediate process, Calculate weights for the fourth intermediate process, Calculate the bias for the second intermediate process, To update the gate activation result, Calculate weights for the fourth intermediate process, Calculate the weights for the fifth intermediate process, Calculate bias for the third intermediate process; Representing functions using dynamic dependencies The dynamic correlation of each historical maintenance record is obtained to represent the learning result , the dynamic correlation represents the learning result Summing and obtaining dynamic perception results of all the historical maintenance records; Wherein, the dynamic correlation represents the learning result Contains all original maintenance record information, all hidden and accumulated maintenance record information, and all maintenance record weight information; The method for fusing the static perception results of the historical maintenance items with the dynamic perception results is: ； In the formula, Y is the correlation perception fusion result, S is the static perception result of the historical maintenance record, and D is the dynamic perception result of the historical maintenance record; The specific method for obtaining the T+1 maintenance items is as follows: ； In the formula, and are weight and bias respectively, is the dimension of Y, The type of maintenance project; This is the T+1 maintenance project.

2. The method for predicting global automobile maintenance demand based on correlation perception and fusion according to claim 1 is characterized in that: The dynamic correlation expression function It is expressed as: ； In the formula, is the Kronecker product operation; is element-by-element multiplication; M t is the maintenance item of the tth maintenance of a car; B t Calculate the process variable, N t is the hidden state of the t-th maintenance record, E t Record the attention weight for the tth maintenance, The type of maintenance project.

3. The method for predicting global automobile maintenance demand based on correlation perception and fusion according to claim 1 is characterized in that: The global objective function of the automobile maintenance demand prediction model based on correlation perception and fusion is a binary cross entropy loss function, specifically: ； In the formula, For maintenance projects i The prediction result, L is the binary cross entropy loss function, is the type of maintenance item, and T is the number of maintenance times of the car.

Images