An individualized vehicle maintenance reminding and pushing system based on owner historical data

Abstract

The application relates to the technical field of vehicle maintenance service, in particular to a personalized vehicle maintenance reminding and pushing system based on owner historical data, which obtains a plurality of vehicle use features according to owner historical data through a feature extraction module, obtains a user portrait label according to the vehicle use features through a user analysis module, obtains reference weights of the plurality of vehicle use features according to the user portrait label through a preference analysis module, obtains a maintenance cycle correction amount according to a preset multi-expert hybrid model based on the vehicle use features and the reference weights through a correction and calibration module, and then performs vehicle maintenance reminding and pushing according to the maintenance cycle correction amount through a reminding and pushing module. Compared with the prior art, the application adopts an innovative multi-expert hybrid model architecture, outputs a scientific and reliable maintenance cycle correction amount, and solves the problem that the vehicle maintenance reminding scheme in the prior art lacks personalization.

Description

Technical Field

[0001] This invention relates to the field of vehicle maintenance service technology, and in particular to a personalized vehicle maintenance reminder push system based on the owner's historical data. Background Technology

[0002] Regular vehicle maintenance is crucial for ensuring driving safety, extending vehicle lifespan, and maintaining stable performance. A scientific maintenance plan can promptly identify and resolve potential mechanical faults, reduce the risk of unexpected accidents, and minimize repair costs caused by improper maintenance. Therefore, reminding vehicle owners to perform regular maintenance is an important part of vehicle after-sales service.

[0003] Current mainstream vehicle maintenance reminder schemes mainly rely on linear data accumulated by mileage or time, lacking comprehensive analysis of multi-dimensional variables such as driving behavior (e.g., frequency of rapid acceleration / braking), environmental conditions (e.g., driving in high / low temperature regions), and user habits (e.g., frequent short-distance driving). This simplistic model cannot reflect the true wear and tear of the vehicle, resulting in a disconnect between reminder timing and actual needs.

[0004] Therefore, there is an urgent need for an intelligent push method that integrates historical data and personalized features to achieve precise maintenance reminders tailored to each individual, thereby improving user experience, optimizing after-sales service efficiency, and solving the problem of the lack of personalization in existing vehicle maintenance reminder solutions. Summary of the Invention

[0005] Therefore, the present invention provides a personalized vehicle maintenance reminder push system based on the owner's historical data to solve the problem of lack of personalization in the existing vehicle maintenance reminder schemes.

[0006] This invention provides a personalized vehicle maintenance reminder push system based on vehicle owner historical data, comprising: The feature extraction module is used to acquire vehicle owner historical data and obtain various vehicle usage characteristics based on the vehicle owner historical data; The user analysis module is used to obtain user profile tags based on vehicle usage characteristics; The preference analysis module is used to obtain reference weights for various vehicle usage characteristics based on user profile tags; The correction and calibration module is used to obtain the maintenance cycle correction amount based on vehicle usage characteristics and reference weights, according to a preset multi-expert hybrid model. The preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. The multiple expert models are used to input a vehicle usage characteristic and output the corresponding intermediate feature vector. The feature fusion layer is used to fuse multiple intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model is also used to obtain the output result representing the maintenance cycle correction amount based on the fused feature vector. The reminder push module is used to send vehicle maintenance reminders based on the maintenance cycle adjustment.

[0007] In a preferred implementation: based on user profile tags, reference weights for various vehicle usage characteristics are obtained, including: Based on multiple user profile tags, establish a user profile encoding vector; The user profile encoding vector is input into the preset weight analysis neural network model to obtain the reference weight of each vehicle usage feature output by the preset weight analysis neural network model; The preset weight analysis neural network model is connected to the feature fusion layer and is used to output reference weights to the feature fusion layer. The preset weight analysis neural network and the preset multi-expert hybrid model are trained together on the same training set.

[0008] In a preferred implementation: vehicle usage characteristics include driving behavior characteristics, environmental characteristics, consumption preference characteristics, vehicle status characteristics, and historical maintenance characteristics, wherein the driving behavior characteristics and environmental characteristics are both time-series data; among the multiple expert models, the expert models corresponding to the driving behavior characteristics and environmental characteristics are recurrent neural networks, and the preset multi-expert hybrid model also includes a first fully connected layer, which is connected to the feature fusion layer, and is used to obtain the output result representing the maintenance cycle correction amount based on the fused feature vector; the preset weight analysis neural network model includes an input layer, a second fully connected layer, and an output layer connected in sequence, wherein the input layer is used to input the user profile encoding vector, and the output layer is used to output the reference weight of each vehicle usage characteristic.

[0009] In a preferred implementation: based on vehicle usage characteristics and reference weights, the maintenance cycle correction is obtained according to a preset multi-expert hybrid model, including: The vehicle usage characteristics and reference weights are input into a preset multi-expert hybrid model to obtain the first correction value output by the preset multi-expert hybrid model; Based on vehicle usage characteristics and reference weights, and using a pre-defined theoretical calculation model, a second correction amount is obtained. By summing the first and second correction amounts, the maintenance cycle correction amount is obtained.

[0010] In a preferred implementation: vehicle usage characteristics include driving behavior characteristics, environmental characteristics, consumer preference characteristics, vehicle condition characteristics, and historical maintenance characteristics; the calculation process of the preset theoretical calculation model includes: Based on driving behavior characteristics, environmental characteristics, and vehicle state characteristics, the corresponding driving loss factors, environmental loss factors, and vehicle loss factors are obtained respectively. Based on the reference weights of driving behavior characteristics, environmental characteristics, and vehicle state characteristics, the driving loss factor, environmental loss factor, and vehicle loss factor are weighted and summed to obtain the vehicle loss correction amount. Based on consumption preference characteristics and historical maintenance characteristics, the corresponding consumption preference factors and historical preference factors are obtained respectively; Based on the reference weights of consumption preference characteristics and historical maintenance characteristics, the consumption preference factors and historical preference factors are weighted and summed to obtain the user demand adjustment amount; The second correction amount is obtained by summing the vehicle wear correction amount and the user demand adjustment amount.

[0011] In a preferred implementation: the maintenance cycle correction amount is obtained by summing the first correction amount and the second correction amount, including: Based on the similarity between the first correction and the second correction, the weight ratio between the first correction and the second correction is obtained, wherein the weight ratio is proportional to the similarity between the first correction and the second correction. Based on the weight ratio, the first correction amount and the second correction amount are weighted and summed to obtain the maintenance cycle correction amount.

[0012] In a preferred implementation: vehicle maintenance reminders are pushed based on the maintenance cycle adjustment, including: The maintenance items are determined based on the vehicle's usage characteristics and maintenance cycle adjustments. Based on the maintenance items, candidate suppliers are identified; Candidate suppliers are screened based on user profile tags, and maintenance plans are obtained; Reminders will be sent based on the maintenance plan.

[0013] This invention also provides a method for personalized vehicle maintenance reminder push based on vehicle owner historical data, including: Obtain vehicle owner historical data and, based on this data, derive various vehicle usage characteristics; Based on vehicle usage characteristics, user profile tags are obtained; Based on user profile tags, reference weights for various vehicle usage characteristics are obtained; Based on vehicle usage characteristics and reference weights, the maintenance cycle correction amount is obtained according to a preset multi-expert hybrid model. The preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. The multiple expert models are used to input a vehicle usage characteristic and output the corresponding intermediate feature vector. The feature fusion layer is used to fuse multiple intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model is also used to obtain the output result representing the maintenance cycle correction amount based on the fused feature vector. Vehicle maintenance reminders will be sent based on the adjusted maintenance cycle.

[0014] The present invention also provides an electronic device, comprising: Memory and processor; The memory is used to store the program, and the processor is used to run any of the above-mentioned personalized vehicle maintenance reminder push systems based on the owner's historical data when the program is executed.

[0015] The present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can run any of the above-mentioned personalized vehicle maintenance reminder push systems based on the vehicle owner's historical data.

[0016] The beneficial effects of using the above embodiments are: This invention provides a personalized vehicle maintenance reminder push system based on vehicle owner historical data. It acquires vehicle owner historical data through a feature extraction module and obtains various vehicle usage characteristics based on this data. A user analysis module generates user profile tags based on these characteristics. A preference analysis module calculates reference weights for these vehicle usage characteristics based on the user profile tags. A correction and calibration module, based on the vehicle usage characteristics and reference weights, calculates a maintenance cycle correction amount using a preset multi-expert hybrid model. This preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. Each expert model takes one vehicle usage characteristic as input and outputs a corresponding intermediate feature vector. The feature fusion layer fuses these intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model also outputs a result representing the maintenance cycle correction amount based on the fused feature vector. Finally, a reminder push module pushes vehicle maintenance reminders based on the maintenance cycle correction amount. Compared to existing technologies, this invention can comprehensively collect and analyze vehicle owner historical data, extract multi-dimensional vehicle usage characteristics, construct personalized user profile tags based on these characteristics, and further dynamically calculate the reference weights of each characteristic based on the profile tags, ensuring that key influencing factors are reasonably considered. Most importantly, this invention adopts an innovative multi-expert hybrid model architecture, which uses multiple expert models to perform deep modeling on a single feature, and then integrates the intermediate results by weight through a feature fusion layer, finally outputting a scientific and reliable maintenance cycle correction amount. This effectively solves the limitations of the traditional fixed threshold method. Through the deep integration of data-driven and intelligent decision-making, it significantly improves the accuracy of maintenance reminders and user satisfaction, and solves the problem of lack of personalization in existing vehicle maintenance reminder schemes. Attached Figure Description

[0017] Figure 1 The system architecture diagram of the personalized vehicle maintenance reminder push system based on vehicle owner historical data provided by the present invention; Figure 2The flowchart of the personalized vehicle maintenance reminder push method based on the vehicle owner's historical data provided by the present invention is shown. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Combination Figure 1 As shown, a specific embodiment of the present invention discloses a personalized vehicle maintenance reminder push system based on vehicle owner historical data, comprising: The feature extraction module 110 is used to acquire vehicle owner historical data and obtain various vehicle usage features based on the vehicle owner historical data; User analysis module 120 is used to obtain user profile tags based on vehicle usage characteristics; The preference analysis module 130 is used to obtain reference weights for various vehicle usage characteristics based on user profile tags; The correction calibration module 140 is used to obtain the maintenance cycle correction amount based on vehicle usage characteristics and reference weights, according to a preset multi-expert hybrid model. The preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. The multiple expert models are used to input a vehicle usage characteristic and output the corresponding intermediate feature vector. The feature fusion layer is used to fuse multiple intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model is also used to obtain the output result representing the maintenance cycle correction amount based on the fused feature vector. The reminder push module 150 is used to send vehicle maintenance reminders based on the maintenance cycle adjustment amount.

[0020] In the above process, vehicle owner historical data refers to any data related to vehicle driving, including data stored by the vehicle's sensors and data related to the owner's modifications, refueling, and other consumption activities. The process of extracting vehicle usage features in the feature extraction module and the process of building user profile tags in the user analysis module can be flexibly implemented using any existing method according to actual needs. The maintenance cycle refers to any metric, including time and mileage, that can be used to measure whether vehicle maintenance is necessary.

[0021] The feature extraction module in this embodiment can comprehensively collect and analyze vehicle owner historical data, extracting vehicle usage characteristics covering multiple dimensions such as driving behavior, environmental conditions, and usage habits, laying a data foundation for subsequent analysis. The user analysis module combines these characteristics to construct personalized user profile tags, enabling the system to identify the differentiated needs of different vehicle owners. The preference analysis module further dynamically calculates the reference weights of each feature based on the profile tags, ensuring that key influencing factors are reasonably considered. Particularly noteworthy is the correction and calibration module, which employs an innovative multi-expert hybrid model architecture—multiple expert models perform deep modeling for a single feature, and then the intermediate results are integrated by weight through a feature fusion layer, ultimately outputting a scientifically reliable maintenance cycle correction amount, effectively overcoming the limitations of traditional fixed threshold methods. Finally, the reminder push module intelligently triggers maintenance suggestions based on the corrected cycle, avoiding the waste of resources from over-maintenance and preventing the safety hazards of under-maintenance. This embodiment, through the deep integration of data-driven and intelligent decision-making, achieves a leapfrog upgrade from "unified standards" to "one policy per person," significantly improving the accuracy of maintenance reminders and user satisfaction.

[0022] In the aforementioned preference analysis module 130, the reference weights can be manually set according to the characteristics of different user profile tags. However, the present invention provides a more preferred solution: In a preferred embodiment, the steps performed by the aforementioned preference analysis module 130 include obtaining reference weights for various vehicle usage characteristics based on user profile tags, specifically: Based on multiple user profile tags, establish a user profile encoding vector; The user profile encoding vector is input into the preset weight analysis neural network model to obtain the reference weight of each vehicle usage feature output by the preset weight analysis neural network model; The preset weight analysis neural network model is connected to the feature fusion layer and is used to output reference weights to the feature fusion layer. The preset weight analysis neural network and the preset multi-expert hybrid model are trained together on the same training set.

[0023] This embodiment automatically learns and outputs reference weights for each vehicle usage characteristic through a pre-set weighted analysis neural network model. It leverages the neural network model's ability to autonomously mine the complex nonlinear relationships between different user profile tags and reference weights based on massive historical data, avoiding subjective biases that may arise from manual setting. Furthermore, this network and the multi-expert hybrid model are jointly trained using the same training set, ensuring coordinated optimization of weight calculation and feature fusion. This makes the generated reference weights perfectly adaptable to the model architecture, significantly improving prediction accuracy. Compared to traditional methods with fixed weights, this solution achieves scientific, precise, and automated weight allocation through deep learning technology, providing a more reliable decision-making basis for personalized maintenance cycle adjustments. This eliminates the need for manual setting of reference weights, making them more scientific and reasonable.

[0024] Specifically, in a preferred embodiment, vehicle usage characteristics include driving behavior characteristics, environmental characteristics, consumer preference characteristics, vehicle status characteristics, and historical maintenance characteristics, wherein the driving behavior characteristics and environmental characteristics are both time-series data. Based on this, among multiple expert models, the expert models corresponding to the driving behavior characteristics and environmental characteristics are recurrent neural networks. The preset multi-expert hybrid model also includes a first fully connected layer, which is connected to the feature fusion layer, and is used to obtain an output representing the maintenance cycle correction amount based on the fused feature vector. The preset weight analysis neural network model includes an input layer, a second fully connected layer, and an output layer connected in sequence, wherein the input layer is used to input the user profile encoding vector, and the output layer is used to output the reference weights for each type of vehicle usage characteristic.

[0025] In this embodiment, time-series data is specifically constructed based on driving behavior features and environmental features. A recurrent neural network (in this invention, a recurrent neural network refers to any existing neural network model capable of processing time-series data, including RNNs and LSTMs) is selected as the expert model. This effectively captures the temporal dependencies and dynamic change patterns in the data, better reflecting the complexity of real-world driving scenarios. Simultaneously, the feature fusion layer effectively ensures the integration of multi-source heterogeneous features, simplifies the model structure, improves computational efficiency, fully leverages the advantages of various models, and significantly enhances prediction accuracy.

[0026] The present invention also provides a more specific embodiment to explain the above content: Firstly, based on the vehicle owner's historical data, the vehicle usage characteristics that can be extracted include: Driving behavior characteristics: These reflect the owner's daily driving habits and directly affect the vehicle's wear and tear rate. Data can be collected in real-time via the onboard OBD module and supplemented by GPS tracking data from a mobile phone. Specific characteristics may include the frequency of rapid acceleration (times / 100 km / h), the frequency of emergency braking (times / 100 km / h), average vehicle speed (km / h), and the percentage of idling time over a period of time. Average data or data from multiple time periods can be used to create time-series data.

[0027] Environmental characteristics: These reflect the impact of vehicle usage environment on maintenance needs (e.g., climate and terrain accelerating fluid degradation). They can be obtained through GPS positioning (recording the vehicle's permanent location), meteorological API (synchronizing climate data), and historical trajectory clustering (identifying congested road sections). Specific characteristic data include the climate zone of the permanent location (tropical / temperate / arctic), annual average temperature difference (°C), commuting route altitude (m), and percentage of time spent in congested areas (%). Similarly, environmental characteristics can be presented as averaged numerical data or as time-series data from multiple time periods.

[0028] Consumer preference characteristics: These reflect car owners' preferences for maintenance services (e.g., price sensitivity, quality priority). This can be obtained through methods such as connecting to gas station / repair shop POS systems (spending amount, parts brand), manual uploads via car owner apps (brand preference), and coupon usage records (discount sensitivity). Specific data includes average maintenance expenditure (RMB / service), parts brand preference (original equipment manufacturer / Mann / Mahle), service channel preference (4S store / chain store), and discount sensitivity (frequency of historical coupon usage).

[0029] Vehicle status characteristics: Features reflecting the vehicle's current actual needs (such as aging level, impact of modifications). These can be obtained through the onboard OBD interface (reading mileage, fluid data, fault codes), sensors (such as fluid sensors), and maintenance records (parts replacement). Specific characteristic data include vehicle age (years), total mileage (ten thousand kilometers), component aging index (0-10 points), fluid health (%), and modification intensity (0-1 point).

[0030] Historical maintenance characteristics: Features reflecting a vehicle owner's past maintenance habits and results (such as cycle deviation and supplier selection). These can be obtained through vehicle manufacturer databases (original factory maintenance records), vehicle owner APP history records (service provider selection), and fault records (post-maintenance faults). Specific characteristic data include past maintenance cycle deviation (days / km), maintenance response rate (%), supplier preference (XX chain store), and post-maintenance fault interval (days).

[0031] Based on this, various user profile tags can be obtained by setting relevant thresholds. In this embodiment, the user profile tags include: Driving style tags, such as aggressive (acceleration frequency > 5 times / 100 km / h) and mild (acceleration frequency < 2 times / 100 km / h).

[0032] Environmentally sensitive labels, such as cold-climate sensitive type (annual average temperature <0℃) and tropical adapted type (annual average temperature >25℃).

[0033] Consumer preference labels, such as price-sensitive (maintenance expenditure < 80% of industry average) and quality-priority (parts brands are original / high-end).

[0034] Vehicle status labels, such as aging vehicle (vehicle age > 5 years) or active modification vehicle (number of modification parts > 3).

[0035] Maintenance habit tags: High responsiveness (maintenance response rate > 70%), procrastination (maintenance response rate < 30%).

[0036] Mapping each of the above user profile tags to numerical values, and then combining these values ​​into a vector, yields the user profile encoding vector. Inputting this user profile encoding vector into a pre-defined weighted analysis neural network model provides reference weights for each vehicle usage feature, for example: If a car owner's user profile is tagged as Aggressive Driving + Tropical Adaptability + Quality Priority + Aging Vehicle + High Responsiveness, the reference weight for each vehicle usage characteristic can be: Driving behavior characteristic = 0.4 (aggressive driving has a higher impact on vehicle wear and tear); Environmental characteristics = 0.15 (the environment has a relatively small impact on vehicle wear and tear); Consumer preference characteristic = 0.1 (quality first, routine repairs and maintenance have little impact on vehicle wear and tear); Vehicle condition characteristic = 0.3 (vehicles are old and worn out; their condition should be carefully observed). Historical maintenance characteristics = 0.05 (good daily maintenance habits).

[0037] Understandably, the aforementioned reference weights can also be set manually.

[0038] The preset multi-expert hybrid model in this embodiment includes expert models corresponding to each vehicle usage characteristic, namely driving behavior expert, environmental expert, consumer preference expert, vehicle status expert, and historical maintenance expert. The driving behavior expert and environmental expert are recurrent neural network models such as RNN, LSTM, and Transformer used to process time-series data, while the other experts are ordinary feedforward neural networks. The preset multi-expert hybrid model also includes a fully connected layer connected after the feature fusion layer, used to output the maintenance cycle correction amount based on the fused vector.

[0039] It is worth noting that in this embodiment, the preset weight analysis neural network and the preset multi-expert hybrid model are connected together, which means that the two can be trained together. For example, known maintenance records can be collected, and driving behavior features and user profile labels can be established based on the historical vehicle owner data corresponding to these maintenance records as input data. Output data can be obtained based on the corresponding maintenance time. A loss function can be established based on the input and output data. When the loss is calculated to the feature fusion layer based on the loss function, the deviation of the reference weights can be obtained, thereby optimizing the preset weight analysis neural network through backpropagation.

[0040] Furthermore, in a preferred embodiment, the steps performed by the aforementioned correction calibration module 140, specifically including: obtaining the maintenance cycle correction amount based on vehicle usage characteristics and reference weights according to a preset multi-expert hybrid model, include: The vehicle usage characteristics and reference weights are input into a preset multi-expert hybrid model to obtain the first correction value output by the preset multi-expert hybrid model; Based on vehicle usage characteristics and reference weights, and using a pre-defined theoretical calculation model, a second correction amount is obtained. By summing the first and second correction amounts, the maintenance cycle correction amount is obtained.

[0041] The maintenance cycle correction can be directly obtained through a pre-set multi-expert hybrid model. However, this embodiment further introduces a pre-set theoretical calculation model to improve accuracy through redundant planning. Specifically, this embodiment uses a pre-set multi-expert hybrid model to perform deep learning analysis on vehicle usage characteristics and reference weights, capturing complex nonlinear relationships and potential patterns, and outputting a highly adaptable first correction. Simultaneously, by establishing a calculation model based on validated vehicle engineering theory, the second correction ensures a solid theoretical foundation and interpretability. This avoids the overfitting risk that may occur with purely data-driven methods and overcomes the shortcomings of traditional theoretical models in adapting to complex real-world scenarios. By dynamically weighting and fusing the two corrections, the final maintenance cycle correction output by the system conforms to both the actual operating state of the vehicle and meets engineering safety standards, achieving a perfect balance between personalization and reliability, and significantly improving the scientific rigor and practicality of maintenance reminders.

[0042] Specifically, in a preferred embodiment, vehicle usage characteristics include driving behavior characteristics, environmental characteristics, consumer preference characteristics, vehicle condition characteristics, and historical maintenance characteristics. Based on these, the calculation process of a pre-defined theoretical calculation model includes: Based on driving behavior characteristics, environmental characteristics, and vehicle state characteristics, the corresponding driving loss factors, environmental loss factors, and vehicle loss factors are obtained respectively. Based on the reference weights of driving behavior characteristics, environmental characteristics, and vehicle state characteristics, the driving loss factor, environmental loss factor, and vehicle loss factor are weighted and summed to obtain the vehicle loss correction amount. Based on consumption preference characteristics and historical maintenance characteristics, the corresponding consumption preference factors and historical preference factors are obtained respectively; Based on the reference weights of consumption preference characteristics and historical maintenance characteristics, the consumption preference factors and historical preference factors are weighted and summed to obtain the user demand adjustment amount; The second correction amount is obtained by summing the vehicle wear correction amount and the user demand adjustment amount.

[0043] The theoretical calculation model in this embodiment first establishes multiple wear factors determined by three objective factors: driving behavior, environmental conditions, and vehicle condition. These factors are then scientifically weighted using reference weights to accurately quantify the actual wear and tear on the vehicle. Simultaneously, a user behavior dimension is creatively introduced, establishing user demand adjustment parameters that reflect personalized needs and service preferences through consumption preference factors and historical preference factors. This two-layer calculation architecture of "objective wear and tear + subjective needs" ensures that the calculation results conform to vehicle engineering principles while adapting to the differentiated needs of different users. Particularly noteworthy is that the model dynamically adjusts the contribution of each factor through reference weights, enabling the calculation results to reflect both the actual condition of the vehicle and match user preferences, achieving an organic unity between objective laws and subjective needs.

[0044] The present invention also provides a more specific solution, wherein the second correction amount is calculated as follows: ; in, This is the second correction amount. This is the correction amount for vehicle wear and tear. Adjust the quantity according to user needs.

[0045] ; Specifically, The three terms in the formula are driving loss factor, environmental loss factor, and vehicle loss factor, respectively. To rapidly increase the frequency, This refers to the frequency of emergency braking. The average annual temperature difference For the percentage of time spent in congestion, The component aging index. For the health of the oil, The preset lower limit for oil health. , , , and These are the sensitivity coefficients for each feature (preset hyperparameters that can be obtained through regression of historical data). , and These are the reference weights for driving behavior characteristics, environmental characteristics, and vehicle state characteristics, respectively.

[0046] ; Specifically, The two terms in the formula are the consumption preference factor and the historical preference factor, respectively. The initial base period (kilometers or time) to be adjusted. This is the average maintenance expenditure. This represents the industry average maintenance expenditure. This is the driving style coefficient (aggressive = 1, mild = 0). The industry average driving style coefficient (e.g., 0.5). and These are the reference weights corresponding to consumption preference characteristics and historical maintenance characteristics, respectively.

[0047] Furthermore, in a preferred embodiment, the above step of summing the first correction amount and the second correction amount to obtain the maintenance cycle correction amount specifically includes: Based on the similarity between the first correction and the second correction, the weight ratio between the first correction and the second correction is obtained, wherein the weight ratio is proportional to the similarity between the first correction and the second correction. Based on the weight ratio, the first correction amount and the second correction amount are weighted and summed to obtain the maintenance cycle correction amount.

[0048] From the perspective of the inherent characteristics of the models, theoretical calculation models are built upon long-term validated vehicle engineering principles and physical laws. Their calculation results have clear causal relationships and interpretability, essentially providing a "deterministic" assessment of vehicle wear and tear. In contrast, the pre-defined multi-expert hybrid model is a neural network model. As a data-driven model, its predictive ability relies on statistical correlations in historical data, and it may exhibit "black box" misjudgments when facing special operating conditions not covered by the training set. When significant differences arise between the two models, it is more likely that the neural network has encountered situations outside the data distribution. In such cases, increasing the weights of the theoretical model essentially guides the system back to the engineering safety baseline.

[0049] Therefore, this embodiment quantitatively evaluates the similarity between the data-driven model (first correction) and the theoretical calculation model (second correction). When the two calculation results are highly consistent, it indicates that the judgment of the preset multi-expert hybrid model is accurate and reliable, and the system will automatically increase its weight ratio. Conversely, when a significant deviation occurs, the weight of the theoretical calculation model is intelligently increased to ensure the reliability of the final correction. This embodiment constructs a model credibility feedback mechanism through similarity evaluation, enabling the system to have self-verification capabilities and effectively reducing the risk of systematic bias that may exist in a single model. Finally, the weighted fusion process realizes dynamic optimization of calculation accuracy, maximizing the predictive advantages of the data-driven model while ensuring a safety baseline.

[0050] Furthermore, in a preferred embodiment, the steps performed by the reminder push module 150, namely: pushing vehicle maintenance reminders based on the maintenance cycle adjustment, specifically include: The maintenance items are determined based on the vehicle's usage characteristics and maintenance cycle adjustments. Based on the maintenance items, candidate suppliers are identified; Candidate suppliers are screened based on user profile tags, and maintenance plans are obtained; Reminders will be sent based on the maintenance plan.

[0051] Existing maintenance reminders only state "maintenance is required" without providing specific items, costs, service providers, or other crucial information, requiring users to search manually, resulting in a poor user experience. Therefore, this implementation, based on the maintenance cycle adjustment, further incorporates vehicle usage characteristics and user profile tags to provide a refined recommendation solution. The final recommended maintenance plan not only meets the vehicle's actual needs but also precisely matches the user's personalized requirements, effectively guiding users to choose highly compatible, high-quality service providers. This optimizes the user experience and improves service conversion rates.

[0052] For example, based on real-time vehicle data (such as oil degradation rate and brake pad thickness read by OBD) and historical maintenance records, it can automatically generate "must-do items" (such as oil must be changed) and "optional items" (such as air filter replacement depending on air quality). Then, it can adjust the push notification focus through user profile tags, for example, prioritizing original equipment manufacturer (OEM) certified / high-end brand parts for quality-conscious users.

[0053] Combination Figure 2 As shown, the present invention also provides a method for personalized vehicle maintenance reminder push based on vehicle owner historical data, including: S201. Obtain vehicle owner's historical data and, based on the vehicle owner's historical data, obtain various vehicle usage characteristics; S202. Based on vehicle usage characteristics, obtain user profile tags; S203. Based on user profile tags, obtain reference weights for various vehicle usage characteristics; S204. Based on vehicle usage characteristics and reference weights, the maintenance cycle correction amount is obtained according to a preset multi-expert hybrid model. The preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. The multiple expert models are used to input a vehicle usage characteristic and output the corresponding intermediate feature vector. The feature fusion layer is used to fuse multiple intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model is also used to obtain the output result representing the maintenance cycle correction amount according to the fused feature vector. S205. Based on the maintenance cycle adjustment, vehicle maintenance reminders will be sent.

[0054] The present invention also provides an electronic device, comprising: Memory and processor; The memory is used to store the program, and the processor is used to run any of the above-mentioned personalized vehicle maintenance reminder push systems based on the owner's historical data when the program is executed.

[0055] The present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can run any of the above-mentioned personalized vehicle maintenance reminder push systems based on the vehicle owner's historical data.

[0056] This invention provides a personalized vehicle maintenance reminder push system based on vehicle owner historical data. It acquires vehicle owner historical data through a feature extraction module and obtains various vehicle usage characteristics based on this data. A user analysis module generates user profile tags based on these characteristics. A preference analysis module calculates reference weights for these vehicle usage characteristics based on the user profile tags. A correction and calibration module, based on the vehicle usage characteristics and reference weights, calculates a maintenance cycle correction amount using a preset multi-expert hybrid model. This preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. Each expert model takes one vehicle usage characteristic as input and outputs a corresponding intermediate feature vector. The feature fusion layer fuses these intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model also outputs a result representing the maintenance cycle correction amount based on the fused feature vector. Finally, a reminder push module pushes vehicle maintenance reminders based on the maintenance cycle correction amount. Compared to existing technologies, this invention can comprehensively collect and analyze vehicle owner historical data, extract multi-dimensional vehicle usage characteristics, construct personalized user profile tags based on these characteristics, and further dynamically calculate the reference weights of each characteristic based on the profile tags, ensuring that key influencing factors are reasonably considered. Most importantly, this invention adopts an innovative multi-expert hybrid model architecture, which uses multiple expert models to perform deep modeling on a single feature, and then integrates the intermediate results by weight through a feature fusion layer, finally outputting a scientific and reliable maintenance cycle correction amount. This effectively solves the limitations of the traditional fixed threshold method. Through the deep integration of data-driven and intelligent decision-making, it significantly improves the accuracy of maintenance reminders and user satisfaction, and solves the problem of lack of personalization in existing vehicle maintenance reminder schemes.

[0057] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0058] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A personalized vehicle maintenance reminder push system based on vehicle owner's historical data, characterized in that, include: The feature extraction module is used to acquire vehicle owner historical data and obtain various vehicle usage characteristics based on the vehicle owner historical data; The user analysis module is used to obtain user profile tags based on vehicle usage characteristics; The preference analysis module is used to obtain reference weights for various vehicle usage characteristics based on user profile tags; The correction and calibration module is used to obtain the maintenance cycle correction amount based on vehicle usage characteristics and reference weights, according to a preset multi-expert hybrid model. The preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. The multiple expert models are used to input a vehicle usage characteristic and output the corresponding intermediate feature vector. The feature fusion layer is used to fuse multiple intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model is also used to obtain the output result representing the maintenance cycle correction amount based on the fused feature vector. The reminder push module is used to send vehicle maintenance reminders based on the maintenance cycle adjustment.

2. The personalized vehicle maintenance reminder push system based on vehicle owner historical data according to claim 1, characterized in that, Based on user profile tags, reference weights for various vehicle usage characteristics are obtained, including: Based on multiple user profile tags, establish a user profile encoding vector; The user profile encoding vector is input into the preset weight analysis neural network model to obtain the reference weight of each vehicle usage feature output by the preset weight analysis neural network model; The preset weight analysis neural network model is connected to the feature fusion layer and is used to output reference weights to the feature fusion layer. The preset weight analysis neural network and the preset multi-expert hybrid model are trained together on the same training set.

3. The personalized vehicle maintenance reminder push system based on vehicle owner historical data according to claim 2, characterized in that, Vehicle usage characteristics include driving behavior characteristics, environmental characteristics, consumer preference characteristics, vehicle status characteristics, and historical maintenance characteristics. Among these, driving behavior characteristics and environmental characteristics are time-series data. In the multiple expert models, the expert models corresponding to driving behavior characteristics and environmental characteristics are recurrent neural networks. The preset multi-expert hybrid model also includes a first fully connected layer, which is connected to the feature fusion layer. This first fully connected layer is used to obtain the output representing the maintenance cycle correction amount based on the fused feature vector. The preset weight analysis neural network model includes an input layer, a second fully connected layer, and an output layer connected in sequence. The input layer is used to input the user profile encoding vector, and the output layer is used to output the reference weights for each vehicle usage characteristic.

4. The personalized vehicle maintenance reminder push system based on vehicle owner historical data according to claim 1, characterized in that, Based on vehicle usage characteristics and reference weights, and according to a pre-set multi-expert hybrid model, the maintenance cycle correction is obtained, including: The vehicle usage characteristics and reference weights are input into a preset multi-expert hybrid model to obtain the first correction value output by the preset multi-expert hybrid model; Based on vehicle usage characteristics and reference weights, and using a pre-defined theoretical calculation model, a second correction amount is obtained. By summing the first and second correction amounts, the maintenance cycle correction amount is obtained.

5. The personalized vehicle maintenance reminder push system based on vehicle owner historical data according to claim 4, characterized in that, Vehicle usage characteristics include driving behavior characteristics, environmental characteristics, consumer preference characteristics, vehicle condition characteristics, and historical maintenance characteristics; The calculation process of the pre-defined theoretical calculation model includes: Based on driving behavior characteristics, environmental characteristics, and vehicle state characteristics, the corresponding driving loss factors, environmental loss factors, and vehicle loss factors are obtained respectively. Based on the reference weights of driving behavior characteristics, environmental characteristics, and vehicle state characteristics, the driving loss factor, environmental loss factor, and vehicle loss factor are weighted and summed to obtain the vehicle loss correction amount. Based on consumption preference characteristics and historical maintenance characteristics, the corresponding consumption preference factors and historical preference factors are obtained respectively; Based on the reference weights of consumption preference characteristics and historical maintenance characteristics, the consumption preference factors and historical preference factors are weighted and summed to obtain the user demand adjustment amount; The second correction amount is obtained by summing the vehicle wear correction amount and the user demand adjustment amount.

6. The personalized vehicle maintenance reminder push system based on vehicle owner historical data according to claim 4, characterized in that, Summarizing the first and second correction amounts, we obtain the maintenance cycle correction amount, including: Based on the similarity between the first correction and the second correction, the weight ratio between the first correction and the second correction is obtained, wherein the weight ratio is proportional to the similarity between the first correction and the second correction. Based on the weight ratio, the first correction amount and the second correction amount are weighted and summed to obtain the maintenance cycle correction amount.

7. The personalized vehicle maintenance reminder push system based on vehicle owner historical data according to claim 1, characterized in that, Based on the maintenance cycle adjustment, vehicle maintenance reminders will be sent, including: The maintenance items are determined based on the vehicle's usage characteristics and maintenance cycle adjustments. Based on the maintenance items, candidate suppliers are identified; Candidate suppliers are screened based on user profile tags, and maintenance plans are obtained; Reminders will be sent based on the maintenance plan.

8. A method for personalized vehicle maintenance reminders based on vehicle owner historical data, characterized in that, include: Obtain vehicle owner historical data and, based on this data, derive various vehicle usage characteristics; Based on vehicle usage characteristics, user profile tags are obtained; Based on user profile tags, reference weights for various vehicle usage characteristics are obtained; Based on vehicle usage characteristics and reference weights, the maintenance cycle correction amount is obtained according to a preset multi-expert hybrid model. The preset multi-expert hybrid model includes multiple expert models and a feature fusion layer. The multiple expert models are used to input a vehicle usage characteristic and output the corresponding intermediate feature vector. The feature fusion layer is used to fuse multiple intermediate feature vectors according to the reference weights to obtain a fused feature vector. The preset multi-expert hybrid model is also used to obtain the output result representing the maintenance cycle correction amount based on the fused feature vector. Vehicle maintenance reminders will be sent based on the adjusted maintenance cycle.

9. An electronic device, characterized in that, include: Memory and processor; The memory is used to store the program, and the processor is used to run any one of the personalized vehicle maintenance reminder push systems based on the vehicle owner's historical data as described in claims 1-7 when the program is executed.

10. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, enable the operation of any one of the personalized vehicle maintenance reminder push systems based on the vehicle owner's historical data as claimed in claims 1-7.

Images