An intelligent management platform for the automotive aftermarket based on big data

Abstract

The present invention discloses an intelligent management platform for the automotive aftermarket based on big data, which relates to the field of automotive data management and includes the following modules: a data acquisition module for real-time collecting vehicle maintenance event data in multiple regions; a spatio-temporal grid division module for dividing the area to be analyzed into multiple spatial grids according to a preset granularity and continuously dividing time into multiple time windows of a fixed length; a data preprocessing module for mapping the vehicle maintenance event data into the corresponding spatial grids and time windows to generate a spatio-temporal event matrix; a maintenance type propagation modeling module for analyzing the spatio-temporal event matrix to simulate the propagation trend of different maintenance types between spatial grids over time; and a maintenance activity prediction module for predicting the maintenance activity of each maintenance type corresponding to a spatial grid at a future moment. The present invention solves the problems of poor prediction accuracy of maintenance services and low resource scheduling efficiency.

Description

Technical Field

[0001] The present invention relates to the field of automotive data management, and more specifically, to an intelligent management platform for the automotive aftermarket based on big data. Background Art

[0002] The automotive aftermarket mainly includes services such as vehicle repair, maintenance, and parts supply, and is an important part of the automotive industry chain. With the rapid growth of the vehicle ownership, the demand for repair services has increased significantly. However, traditional management of the automotive aftermarket mostly relies on manual statistics and experience-based decision-making, making it difficult to efficiently control the real-time changes in the demand for repair services, and prone to problems such as uneven distribution of repair resources and lagging services, especially obvious in large and medium-sized cities.

[0003] In response to the above problems, existing technologies have tried to assist in predicting repair demand through data statistical analysis, such as through conventional time series analysis, hotspot area identification, etc., in order to achieve optimized resource scheduling. However, such methods are usually limited to static or simple statistical models, ignoring the propagation effect of repair activities in space and the dynamic change trend in time, and it is difficult to accurately predict the rapid changes in repair demand in local areas. In addition, the currently widely used prediction technologies also lack in-depth research on the propagation laws between multiple repair types, resulting in the difficulty for existing technologies to accurately respond to repair demand in advance, and the repair resources cannot be refinedly scheduled.

[0004] Therefore, there is an urgent need for a more efficient and accurate repair demand prediction solution in the existing technology to solve the problems of inaccurate prediction of repair services and low efficiency of resource allocation in the automotive aftermarket. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an intelligent management platform for the automotive aftermarket based on big data to solve the problems mentioned in the background art.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] An intelligent management platform for the automotive aftermarket based on big data, comprising:

[0008] A data acquisition module, configured to collect vehicle repair event data in real time from multiple regions, and each repair event data includes the occurrence time, geographical location information, and repair type;

[0009] A spatio-temporal grid division module, configured to divide the area to be analyzed into multiple spatial grids according to a preset granularity, and continuously divide the time into multiple time windows with a fixed length;

[0010] A data preprocessing module, configured to map the vehicle repair event data into the corresponding spatial grids and time windows to generate a spatio-temporal event matrix;

[0011] A maintenance type propagation modeling module, which is used to analyze the spatio-temporal event matrix, construct and train a spatio-temporal graph convolutional network model to simulate the propagation trend of different maintenance types between spatial grids over time;

[0012] A maintenance activity prediction module, which is used to predict the maintenance activity of each maintenance type corresponding to each spatial grid at a future moment based on the trained spatio-temporal graph convolutional network model.

[0013] In some embodiments, the data collected by the data collection module at least includes the longitude and latitude coordinates, timestamp, and classification label of the maintenance type corresponding to the maintenance event.

[0014] In some embodiments, the spatio-temporal grid division module divides the urban area into regular square spatial grids according to the geographical area boundary of the city map and the preset grid division rules, and each spatial grid corresponds to a unique spatial identifier.

[0015] In some embodiments, the area of a single grid of the spatial grid division rules is between 250 meters × 250 meters and 1 kilometer × 1 kilometer.

[0016] In some embodiments, the data preprocessing module maps the maintenance event to the corresponding spatial grid and time window according to the longitude and latitude coordinates and timestamp of each maintenance event data, and accumulatively counts the event frequency of the maintenance type in each grid and each time window to generate a spatio-temporal event matrix with multi-dimensional maintenance type statistical values.

[0017] In some embodiments, the maintenance type propagation modeling module builds a spatio-temporal graph structure, uses the spatial grid as a graph node, and sets edges to connect adjacent grids; the maintenance type statistical value of each node in different time windows is used as a node attribute to construct a spatio-temporal graph network structure.

[0018] In some embodiments, the spatio-temporal graph convolutional network model used by the maintenance type propagation modeling module includes a spatial graph convolutional layer and a temporal convolutional layer;

[0019] Among them, the spatial graph convolutional layer is used to learn the spatial correlation between different spatial grid nodes, and the temporal convolutional layer is used to capture the trend of each node changing over time.

[0020] In some embodiments, the maintenance activity prediction module is based on the trained spatio-temporal graph convolutional network model, and according to the maintenance activity data at the current moment and historical moments, predicts the maintenance activity values of various maintenance types in each spatial grid in the next time window or multiple consecutive future time windows.

[0021] In some embodiments, the maintenance activity prediction value predicted by the maintenance activity prediction module is used to generate a heat map of maintenance types, and the heat map visually shows the activity levels of different maintenance types in future time for different spatial grids.

[0022] In some embodiments, the platform further includes a maintenance resource scheduling module, which is used to allocate in advance the maintenance personnel, maintenance equipment and spare parts inventory of the vehicle maintenance outlets according to the prediction results of the maintenance activity prediction module, so as to cope with the upcoming high-incidence maintenance areas and peak maintenance periods.

[0023] The present invention provides an intelligent management platform for the automotive aftermarket based on big data. By using the vehicle maintenance event data collected in real time and combining spatio-temporal grid division and graph convolutional deep learning models, it realizes the accurate prediction of maintenance service demand and effectively solves the problems of insufficient prediction accuracy and lagging resource allocation in traditional solutions.

[0024] Specifically, the platform of the present invention obtains in real time the occurrence time, geographical location and maintenance type data of maintenance events, constructs a high-precision spatio-temporal event matrix, and effectively depicts the detailed distribution of maintenance events in different regions and times. On this basis, a spatio-temporal graph convolutional network model is introduced, and the propagation effect and time evolution trend of different maintenance types among spatial grids are successfully captured, thus significantly improving the accuracy and reliability of maintenance activity prediction.

[0025] Furthermore, the platform of the present invention enhances the spatial resolution of the prediction results through reasonable division of spatial grids and detailed statistics of maintenance event frequencies. At the same time, the prediction module supports multi-step prediction and independent modeling of multiple types of maintenance events, and visually displays the prediction results with the help of a heat map, enabling managers to clearly grasp the dynamic change trends of maintenance hotspots. In addition, the platform of the present invention also has an intelligent scheduling function for maintenance resources in advance based on the prediction results, thus effectively improving the utilization efficiency of maintenance personnel, maintenance equipment and spare parts inventory, reducing maintenance response delays, and significantly improving the overall maintenance service quality and user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a schematic diagram of the modules of the present invention;

[0027] Figure 2 is a flowchart of data preprocessing and spatio-temporal event matrix generation of the present invention;

[0028] Figure 3 is a flowchart of model construction and training of the present invention;

[0029] Figure 4 is a flowchart of maintenance activity prediction and heat map generation of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The specific implementation manners of the present invention will be described below with reference to the accompanying drawings.

[0031] The present invention provides an intelligent management platform for the automotive aftermarket based on big data. The design concept of the platform of the present invention is to build an appropriate modeling structure by means of the spatial and temporal distribution characteristics of vehicle maintenance big data, so as to predict the evolution trend of maintenance behaviors and provide data support for subsequent resource scheduling.

[0032] As shown in FIG. 1, the platform of the present invention includes functional components such as a data acquisition module, a spatio-temporal grid division module, a data preprocessing module, a maintenance type propagation modeling module, and a maintenance activity prediction module.

[0033] Among them, the data acquisition module is responsible for obtaining maintenance event data in real time from vehicle maintenance outlets or relevant information systems in multiple regions. The collected data includes, but is not limited to, the occurrence time of each maintenance event, the longitude and latitude coordinates of the maintenance location, and the corresponding maintenance type label. The classification of maintenance types can be divided according to industry habits, such as categories like engine failure, brake system repair, body repair, and electrical system maintenance. By continuously running this module, a large-scale and continuously updated maintenance event database can be constructed.

[0034] To facilitate the modeling and analysis of the regional distribution and time evolution of maintenance events, it is necessary to perform standardization processing on the time and space dimensions in the original data. In this embodiment, the spatio-temporal grid division module divides the urban area into spatial grids according to the actual boundary of the city map according to preset rules. The shape of the spatial grid is usually selected as a square to simplify subsequent calculation and processing. Each grid has a unique identifier for subsequent positioning and reference in the model.

[0035] More specifically, an appropriate grid granularity can be set according to the city scale, road density, and spatial distribution density of maintenance events. In practice, the area of a single grid is usually set between 250 meters × 250 meters and 1 kilometer × 1 kilometer. Smaller-area grids are helpful for capturing finer-grained maintenance event distribution characteristics and are suitable for the core areas of large cities with high event density and rapid spatial changes. In suburban areas or areas with relatively sparse maintenance event distributions, larger-area grids can be used to reduce computing resource consumption.

[0036] In terms of the processing of the time dimension, the entire observation period can be continuously divided into multiple time windows of a fixed length. The length of the time window can be set according to the occurrence frequency of maintenance events, such as set to 30 minutes, 1 hour, or 1 day. Each maintenance event is classified into the corresponding time window according to its timestamp, forming a data structure available for time series modeling.

[0037] After the division of space and time is completed, it is necessary to accurately map each piece of maintenance event data to the corresponding spatial grid and time window. As shown in Figure 2, this process is completed by the data preprocessing module. The specific method is as follows:

[0038] First, read the longitude and latitude information in each maintenance event, and combine it with the geographical boundary information of the spatial grid to determine the spatial grid number where the event is located;

[0039] Secondly, read the timestamp information of the event and perform a time windowing operation on it to determine the corresponding time window number.

[0040] Subsequently, in three-dimensional space, a statistical matrix is established with "spatial grid × time window × maintenance type" as the coordinate axes.

[0041] Every time a maintenance event occurs, the system increments the frequency of that type of event in the corresponding cell. By processing all historical maintenance events, a spatio-temporal event matrix with the dimension of maintenance type can be obtained. This matrix not only reflects the maintenance activity level in a certain area during a certain period of time, but also reveals the evolution trend of different types of maintenance events, providing rich spatio-temporal feature information for subsequent modeling.

[0042] The corresponding formula can be expressed as:

[0043] ;

[0044] Among them, represents the frequency of maintenance events on the spatial grid number s, time window number t, and maintenance type number k;

[0045] N is the total number of maintenance events;

[0046] is the spatial grid number to which the th maintenance event belongs, and the time window number to which the

[0047] th

[0048] To further explore the propagation law of maintenance behavior in urban space, as shown in Figure 3, the present invention establishes a data representation method based on a graph structure, abstracting each spatial grid as a node in the graph. The connection relationship between nodes is determined according to spatial adjacency, that is, if two spatial grids are geographically adjacent, a connection edge is set for them in the graph structure. This way of building a graph preserves the local adjacency relationship in the urban spatial structure, enabling the subsequent graph neural network to learn the influence of maintenance behavior between adjacent regions. The attributes of each node within different time windows are composed of the statistical values of its corresponding maintenance types, that is, the frequencies of various maintenance events at this node within this time window. Through this way of building a graph, the originally isolated spatio-temporal event matrix can be transformed into a dynamic graph network with structural information, providing a structured input for modeling the propagation of maintenance behavior.

[0049] In practical applications, this graph structure not only supports static modeling but also can be dynamically updated over time. With the continuous collection of new maintenance events, the node attributes will be continuously updated, and the graph structure can also adjust the edge weights or perform local reconstruction according to the transfer of maintenance event hotspots. The establishment of this graph network provides a natural structural basis for subsequent deep learning models based on spatio-temporal graph convolution, enabling the model to simultaneously capture the mutual influence between spatial neighborhoods and the time evolution trend, which helps to achieve high-precision prediction of maintenance activity and analysis of behavior evolution.

[0050] More specifically, by combining the advantages of graph neural networks and time series modeling, through the collaborative action of the spatial graph convolution layer and the time convolution layer, joint learning of maintenance behavior in both spatial and temporal dimensions can be achieved.

[0051] The spatial graph convolution layer is mainly used to capture the correlation between adjacent spatial grid nodes. In the urban environment, the maintenance behaviors in different regions are not completely independent. The occurrence of high-frequency maintenance events in a certain region often has an impact on the surrounding regions. For example, when frequent brake system repairs occur along a traffic artery, it may mean abnormal road conditions or that there are structural faults in a certain type of vehicle in a specific area, and this pattern will show a certain spatial continuity. Through graph convolution operations, the information of adjacent nodes can be effectively fused, and the propagation path of maintenance behavior across regions can be learned.

[0052] The time convolution layer is used to extract the change trend of each spatial node in the time dimension. The activity level of maintenance events often has obvious periodicity and suddenness. It may change with the daily rhythm, the difference between weekdays and weekends, or may also be affected by external factors such as weather, holidays, and promotional activities. The time convolution layer takes time windows as units and performs convolution operations on the time series of each node, which helps the model identify dynamic features such as upward trends, sharp increases, or continuous low frequencies.

[0053] After the model training is completed, it can be applied to the prediction task of maintenance activity.

[0054] As shown in Figure 4, the platform uses this model to input the statistical data of maintenance events within the historical time window and combines the current status to predict the activity values of various maintenance types in different spatial grids within a future time window or multiple consecutive time windows. These predicted values essentially reflect the possible density of different types of maintenance activities at different times and locations, with strong spatio-temporal directivity and business guiding significance.

[0055] In some embodiments, to present the prediction results more intuitively to the management personnel and the operation and maintenance team, the predicted maintenance activity values can be converted into a heat map of maintenance types. In the heat map, different colors or brightness levels represent different activity intensities, enabling users to clearly identify high-risk areas, potential maintenance hotspots, and activity change trends at a glance.

[0056] Furthermore, to convert the prediction results into actual management actions, the platform also sets up a maintenance resource scheduling module. Based on the prediction results of maintenance activity, this module combines the geographical locations, personnel reserves, and spare part inventories of each maintenance point to allocate resources in advance. For example, before a certain area is about to enter a high-incidence period of maintenance, the platform can notify the corresponding maintenance points in advance to arrange more technicians to be stationed, allocate common spare parts, or inspect and maintain the diagnostic equipment for key vehicle models, thereby effectively improving the maintenance response ability, reducing customer waiting time, and reducing service congestion.

[0057] The specific algorithms of the graph neural network and the convolutional neural network are introduced as follows:

[0058] In the system of the present invention, the graph neural network adopts the form of a graph convolutional network (GCN mentioned later) to process the relationships between spatial grids and capture the propagation law of maintenance events in space. The system first divides the urban area into multiple spatial grids, which are regarded as nodes in the graph. If two grids are adjacent geographically, an undirected edge is established between them, and the adjacency relationship of the entire graph is constructed in this way. Each node has its own attributes, which are derived from the frequencies of various maintenance events in this grid within a specific time window. In other words, the occurrence frequency of maintenance events is organized into a feature vector and passed as input data of the node to the GCN model.

[0059] The model structure design of the GCN consists of two graph convolutional layers. The role of each layer is to update the feature representation of the current node by aggregating the information of neighboring nodes. Specifically, the first graph convolution receives the initial features of the nodes, that is, the feature vectors generated based on the maintenance event frequencies, and then generates a set of intermediate feature representations through convolution operations. The second graph convolution then processes these intermediate features, further refining the information, and generating the final node representations for subsequent prediction tasks. The output dimension of each graph convolution layer is set to 64, which means that the features of each node will be transformed into a 64-dimensional vector. To enable the model to capture non-linear relationships, a ReLU activation function is connected after each convolution layer, enhancing the expressive power of the model through this non-linear transformation. The design of multiple graph convolutional layers enables the model to not only focus on the maintenance situation of a single grid but also understand the propagation law of maintenance behaviors between adjacent grids, thereby revealing the dynamic trends of maintenance events in space.

[0060] Meanwhile, the system also uses a one-dimensional convolutional neural network (the 1D CNN mentioned later) to analyze time series data, aiming to capture the change patterns of maintenance events in the time dimension. The model structure of the 1D CNN includes two convolutional layers and a fully connected layer. The first convolutional layer uses a convolutional kernel of size 3, with a stride of 1 and 32 output channels, which means it will extract 32 different feature patterns from the input time series. The second convolutional layer also uses a convolutional kernel of size 3, with a stride of 1, but the number of output channels is increased to 64, further enriching the expressive power of the features. A ReLU activation function is connected after each convolutional layer to introduce non-linearity, and a max pooling layer with a pooling window size of 2 is also connected. This pooling operation can extract the local key features in the time series while reducing the dimension of the features and the computational burden. After two layers of convolution and pooling, the extracted features are fed into a fully connected layer, and finally mapped to the predicted maintenance activity value, that is, a quantitative estimate of future maintenance needs.

[0061] To train these two models, the historical maintenance event data of the past year can be used. After these data are sorted, they are divided into a training set, a validation set, and a test set according to the ratio of 8:1:1. The training set is used for learning the model parameters, the validation set is used to adjust the hyperparameters and prevent overfitting, and the test set is used to evaluate the final performance of the model. During training, the batch size is set to 32, which means that 32 samples will be processed simultaneously in each iteration. The entire training process can last for 100 rounds to ensure that the model has enough opportunities to converge to a better solution.

[0062] The loss function of the model can be selected as the mean squared error, which is used to measure the gap between the predicted value and the actual value. The Adam algorithm is adopted in the optimization process, which can dynamically adjust the step size according to the change of the gradient and accelerate the convergence. The learning rate is specifically set to 0.001. The performance of the model is evaluated by the root mean squared error, and the RMSE on the test set reaches 0.15. This result indicates that the average error between the predicted value and the actual value of the model is small, and it has high prediction accuracy, which can provide a reliable basis for subsequent resource allocation.

[0063] The running process of the entire algorithm combines the analysis of two dimensions, space and time, to form a complete prediction system. First, the system maps the original maintenance event data into a spatial grid and a time window to generate a spatio-temporal event matrix. This matrix contains both the frequency of maintenance events in each grid and the sequence of these events over time. Then, based on the geographical adjacency relationship of the spatial grid, the system constructs a graph structure and uses GCN to extract spatial features from it. Through the multi-layer convolution operation of GCN, the model can identify how maintenance events spread from one grid to adjacent grids, forming a dynamic pattern in space.

[0064] At the same time, 1D CNN processes the time series data of each grid. The time series reflects the fluctuations of maintenance events in different time windows. CNN extracts the key trends in these fluctuations, such as peak periods or periodic patterns, through convolution and pooling operations. After that, the outputs of GCN and CNN are fused together to generate the prediction results of the maintenance activity in the future time window. This fusion makes full use of the information on both the spatial propagation law and the time change trend, making the prediction more comprehensive and accurate.

[0065] Finally, the system will optimize the allocation of personnel, equipment and spare parts according to the predicted maintenance activity and the actual resource situation of the maintenance outlets. For example, if a certain grid is predicted as a high-activity area, the system will preferentially allocate more maintenance personnel and spare parts to this area. This prediction-based resource allocation method can significantly improve the response efficiency of maintenance services, reduce waiting time, and improve customer satisfaction.

[0066] Example 1 is as follows:

[0067] Taking City A as an example, assume that the city has a permanent population of about 5 million and a vehicle ownership of 2 million, which are distributed in the main urban area, surrounding urban areas and suburbs. There are about 3,000 various types of vehicle maintenance outlets in the city, covering quick repair chains, 4S stores and individual repair shops.

[0068] After adopting the solution of the present invention, the platform first deploys a data collection module, accesses the system interfaces of maintenance points throughout the city, and shares the historical traffic event database with the Traffic Management Bureau. It collects in real time every day the maintenance event data reported by each maintenance point, including the event occurrence time, the longitude and latitude of the vehicle's location, maintenance type tags (such as engine maintenance, tire replacement, brake failure, electrical system, etc.), and some basic vehicle condition information. The collection frequency is in minutes, and tens of thousands of data are accumulated every day.

[0069] For the convenience of analysis, the platform of the present invention divides the urban area of the entire City A into square grids of 500 meters × 500 meters, forming a total of about 1,800 spatial grids. During the division process, the system refers to the city boundary, road density, and the distribution of maintenance event heat, and tries to reduce the invalid areas while maintaining the same grid shape. In the time dimension, the platform uses 1 hour as a unit for windowing, so that there are 24 time windows every day.

[0070] When the system receives a new piece of maintenance event data, the data preprocessing module will automatically identify the spatial grid number to which the event belongs according to the longitude and latitude of the event, and judge the time window in which it is located based on the timestamp. Subsequently, the system records the frequency of this type of maintenance event at the corresponding position in the three-dimensional matrix. With the continuous accumulation of data, a stable and dynamically updated spatio-temporal event matrix is gradually formed, covering the distribution statistics of multiple maintenance types in the city's spatial grids and time windows.

[0071] Next, the platform constructs these spatial grids into a graph structure, where each grid is a node, and edge connections are established between geographically adjacent grids. The system further uses the well-developed spatio-temporal graph convolutional network model in graph neural networks for modeling, which is jointly composed of a spatial graph convolutional layer and a temporal convolutional layer.

[0072] During the training process, the spatial graph convolutional layer identifies the conduction trend of maintenance events between different regions through multi-hop information fusion, while the temporal convolutional layer identifies the temporal patterns such as the growth, decay, or periodic fluctuations of maintenance events on the nodes.

[0073] After being trained and verified with a large amount of historical data, this model has strong generalization ability and can make relatively accurate predictions for future maintenance behaviors.

[0074] For example, if the platform discovers through model operation that the event frequencies of "tire replacement" and "chassis damage" have increased significantly in several spatial grids in the southwestern area of the main urban area in the past three hours, the system predicts that there may be a peak in maintenance demand in this area in the next two hours, especially concentrated in several areas near the intersections of highway entrances and exits and old roads. At the same time, the platform generates a heat map of the predicted maintenance activity data and presents different levels of activity through a visualization interface with color scales such as red, orange, and yellow. Managers can directly see the specific scope and intensity of the hot spot concentration area.

[0075] Based on this heat map and the model prediction results, the platform automatically triggers the maintenance resource scheduling module and recommends that four maintenance outlets around the hot spot area make preparations in advance: two of the outlets are recommended to increase tire inventory, and the other two are recommended to allocate chassis inspection tools and repair technicians. If the user authorizes, the platform can also prompt the driver to avoid areas with a high probability of maintenance or plan a conservative route through the in-vehicle navigation application.

[0076] Throughout the process, from the occurrence of maintenance events to data collection, preprocessing, propagation modeling, activity prediction, heat map display to resource scheduling, the platform is based on data-driven and establishes a closed-loop intelligent management mechanism to help managers achieve active perception, early intervention, and precise allocation, empower the urban maintenance system, and also lay a foundation for building a multi-city collaborative maintenance network in the future.

[0077] In summary, through the integration of the above modeling, prediction, visualization, and scheduling mechanisms, the present invention realizes the full-process closed-loop management of the automotive aftermarket maintenance behavior from data perception, trend analysis to resource optimization, and greatly improves the intelligent level and response efficiency of the urban-level maintenance service system.

[0078] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An intelligent management system for the automotive aftermarket based on big data, characterized in that: include: The data collection module is used to collect vehicle maintenance event data in multiple areas in real time. Each maintenance event data includes the occurrence time, geographic location information and maintenance type; The space-time grid division module is used to divide the area to be analyzed into multiple space grids according to the preset granularity, and to divide the time continuously into multiple time windows of fixed length; A data preprocessing module, used for mapping the vehicle maintenance event data into corresponding spatial grids and time windows to generate a spatiotemporal event matrix; A maintenance type propagation modeling module is used to analyze the spatiotemporal event matrix, build and train a spatiotemporal graph convolutional network model to simulate the propagation trend of different maintenance types between spatial grids over time; The maintenance activity prediction module is used to predict the maintenance activity of the corresponding maintenance type of each spatial grid at future moments based on the trained spatiotemporal graph convolutional network model.

2. According to the big data-based intelligent management system for the automotive aftermarket of claim 1, it is characterized in that: The data collected by the data collection module specifically includes at least the latitude and longitude coordinates corresponding to the maintenance event, a timestamp, and a classification label of the maintenance type.

3. The big data-based intelligent management system for the automotive aftermarket according to claim 1 is characterized in that: The space-time grid division module divides the city area into regular square space grids according to the geographical area boundaries of the city map and the preset grid division rules, and each space grid corresponds to a unique space identifier.

4. The big data-based intelligent management system for the automotive aftermarket according to claim 3 is characterized in that: The area of ​​a single grid of the spatial grid division rule is between 250 meters × 250 meters and 1 kilometer × 1 kilometer.

5. The intelligent management system for the automotive aftermarket based on big data according to claim 1 is characterized in that: The data preprocessing module maps the maintenance event to the corresponding spatial grid and time window according to the latitude and longitude coordinates and timestamp of each maintenance event data, and accumulates and counts the event frequency of the maintenance type in each grid and each time window to generate a spatiotemporal event matrix with multi-dimensional maintenance type statistical values.

6. The big data-based intelligent management system for the automotive aftermarket according to claim 1 is characterized in that: The maintenance type propagation modeling module establishes a spatiotemporal graph structure, takes spatial grids as graph nodes, and sets edges between adjacent grids for connection; the maintenance type statistics of each node in different time windows are used as node attributes to construct a spatiotemporal graph network structure.

7. The big data-based intelligent management system for the automotive aftermarket according to claim 6 is characterized in that: The spatiotemporal graph convolutional network model used by the maintenance type propagation modeling module includes a spatial graph convolutional layer and a temporal convolutional layer; The spatial graph convolution layer is used to learn the spatial correlation between nodes in different spatial grids, and the temporal convolution layer is used to capture the trend of each node changing over time.

8. The big data-based intelligent management system for the automotive aftermarket according to claim 1 is characterized in that: The maintenance activity prediction module is based on the trained spatiotemporal graph convolutional network model and predicts the maintenance activity values ​​of various maintenance types in each spatial grid in the next time window or multiple consecutive time windows in the future according to the maintenance activity data at the current time and historical time.

9. The big data-based intelligent management system for the automotive aftermarket according to claim 8 is characterized in that: The maintenance activity value predicted by the maintenance activity prediction module is used to generate a maintenance type heat map, and the heat map intuitively displays the activity level of maintenance types at different spatial grids at future moments.

10. The big data-based intelligent management system for the automotive aftermarket according to claim 1, characterized in that: The system also includes a maintenance resource scheduling module for pre-allocating maintenance personnel, maintenance equipment and spare parts inventory of automobile maintenance outlets based on the prediction results of the maintenance activity prediction module to cope with the upcoming high-incidence maintenance areas and maintenance peak periods.

Images