Method for constructing residual mileage prediction model of electric vehicle
By building a four-dimensional multi-source database and dynamically adjusting the data source weights, combined with long-term and short-term data fusion arbitration output, the accuracy and adaptability problems in the residual mileage prediction of electric vehicles are solved, and high-precision residual mileage prediction is achieved.
Patent Information
- Application Number
- CN202510675419.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-23
- Publication Date
- 2025-08-19
AI Technical Summary
The prior art has problems such as the limitations of a single data source, insufficient adaptability of static models, and long-term and short-term prediction conflicts in the residual mileage prediction of electric vehicles, resulting in insufficient prediction accuracy and poor stability.
Build a four-dimensional multi-source database, collect vehicle status, environment, road topology and driving behavior data, dynamically adjust the data source weight, combine long-term and short-term data fusion arbitration output mechanism, and train the remaining mileage prediction model.
It realizes high-precision prediction of the remaining mileage of electric vehicles, improves prediction accuracy and flexible adaptability in complex driving scenarios, taking into account global trends and instantaneous accuracy.
Smart Images
Figure CN120509315A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of new energy vehicles, and in particular to a method for constructing a remaining mileage prediction model for an electric vehicle. Background Art
[0002] In the electric vehicle sector, remaining range prediction directly impacts user anxiety and vehicle energy management efficiency, making it a key technology for improving user experience and vehicle energy management. Traditional methods rely primarily on battery state of charge (SOC) and historical vehicle energy consumption data, using linear regression or fixed coefficient models for estimation. These methods suffer from limitations of a single data source, the lack of adaptability of static models, and conflicts between long-term and short-term predictions.
[0003] For example, existing technologies mostly use a single data source or simple linear models, which make it difficult to accurately reflect changes in remaining mileage in complex driving scenarios. Even if attempts are made to compensate through environmental data, they still rely on empirical formulas. Although other existing technologies in the industry have proposed improving prediction accuracy through hybrid models, they lack a reliable arbitration mechanism, resulting in the inability to optimize output when prediction results conflict.
[0004] After analyzing the existing technologies, it is found that the current prediction schemes for the remaining range / energy consumption of electric vehicles have at least the following problems:
[0005] (1) Relying solely on battery and vehicle internal data, the synergy of vehicle status, environment, road topology, and driving behavior data is not fully utilized, resulting in insufficient prediction accuracy;
[0006] (2) The algorithms commonly used in the industry rely entirely on offline training and cannot adapt to changes in energy consumption under different and complex driving conditions;
[0007] (3) A single model architecture is difficult to capture the dynamic changes of different time scales at the same time, which not only leads to poor prediction stability, but also makes it impossible to optimize the output results when conflicts occur in prediction results of different time scales. Summary of the Invention
[0008] In view of the above, the present invention aims to provide a method for constructing a remaining range prediction model for an electric vehicle to solve the above-mentioned technical problems.
[0009] The technical solution adopted in the present invention is as follows:
[0010] The present invention provides a method for constructing a remaining range prediction model for an electric vehicle, comprising:
[0011] Collecting four-dimensional multi-source data, the four-dimensional multi-source data including: vehicle status data, environmental data, road topology data, and driving behavior data;
[0012] Dynamically adjust the weight of each data source according to the driving conditions, and obtain the comprehensive energy consumption data corresponding to different driving conditions based on the weighted data sources;
[0013] A remaining mileage prediction model is trained using four-dimensional multi-source data and comprehensive energy consumption data. The architecture of the remaining mileage prediction model includes: a first sub-model that predicts the remaining mileage based on long-term data, a second sub-model that predicts the remaining mileage based on short-term data, and an arbitration output module that dynamically adjusts the sub-model weights based on prediction error feedback and comprehensively calculates the remaining mileage.
[0014] In at least one possible implementation, the training of the remaining mileage prediction model specifically includes:
[0015] According to the established time standards, long-term data and short-term data are divided and selected from the four-dimensional multi-source data and comprehensive energy consumption data;
[0016] Input the long-term data into the first sub-model for processing to obtain the first remaining mileage prediction result;
[0017] Input the short-term data into the second sub-model for processing to obtain the second remaining mileage prediction result;
[0018] Based on the prediction error, the confidence of the first sub-model and the second sub-model are calculated respectively;
[0019] Dynamically setting a first weight and a second weight corresponding to the first sub-model and the second sub-model according to the confidence level;
[0020] A fusion calculation is performed based on the first remaining mileage prediction result and the first weight, the second remaining mileage prediction result and the second weight to obtain a comprehensive remaining mileage as the final output of the remaining mileage prediction model.
[0021] In at least one possible implementation, obtaining comprehensive energy consumption data corresponding to different driving conditions includes:
[0022] Based on the acceleration data in the driving behavior data, identifying the driving condition type and marking it in the corresponding four-dimensional multi-source data;
[0023] Based on the marked driving condition types, the data sources are weighted and the comprehensive energy consumption value of each driving condition is calculated.
[0024] In at least one possible implementation, the marking in the corresponding four-dimensional multi-source data includes: based on the time stamp alignment processing of the data source, associating the acceleration data with other data collected during the same period.
[0025] In at least one possible implementation, the identification of the driving condition type includes: based on the comparison relationship between the acceleration data and the preset acceleration and deceleration thresholds, and combined with the preset acceleration and deceleration state duration, at least the following driving condition types are obtained: acceleration stage, cruising stage, and braking stage.
[0026] In at least one possible implementation, dynamically adjusting the weight of each data source according to the driving condition includes:
[0027] For the acceleration phase, at least some of the driving behavior data weights are increased;
[0028] For the cruising phase, at least the weights of some data in the vehicle status data and road topology data are increased;
[0029] During the braking phase, at least some of the driving behavior data weights are increased.
[0030] Compared to existing technologies, the main design concept of this invention lies in achieving high-precision prediction of the remaining range of electric vehicles by constructing a four-dimensional multi-source database, dynamically assigning data source weights based on driving condition identification and calculating corresponding energy consumption, and integrating an arbitration output mechanism that fuses long-term and short-term data. The construction of the four-dimensional database enables the fusion of multi-source data to improve prediction accuracy; the weight configuration based on driving conditions significantly improves flexible adaptability to complex driving scenarios; and the use of long-term and short-term fusion arbitration enables comprehensive prediction of energy consumption trends across different time scales. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described below with reference to the accompanying drawings, in which:
[0032] Figure 1 A schematic diagram of a method for constructing a remaining range prediction model for an electric vehicle provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0033] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.
[0034] The present invention proposes an embodiment of a method for constructing a remaining range prediction model for an electric vehicle. Specifically, Figure 1 shown, including:
[0035] Step S1: collecting four-dimensional multi-source data, wherein the four-dimensional multi-source data includes: vehicle status data, environmental data, road topology data, and driving behavior data;
[0036] Specifically, the following data sources can be collected in different vehicle operating scenarios through onboard sensors and external data sources (such as high-precision maps, weather APIs, etc.):
[0037] (1) Vehicle status data, such as battery SOC, voltage, current, temperature, vehicle speed, etc.
[0038] (2) Environmental data, such as temperature, wind speed, rainfall, etc.
[0039] (3) Road topology data, such as slope, curves, etc.
[0040] (4) Driving behavior data, such as acceleration, braking frequency, steering angle, energy recovery efficiency, etc.
[0041] Furthermore, timestamp alignment, preprocessing and normalization operations are performed on the above different data sources, thereby obtaining a four-dimensional multi-source database.
[0042] Step S2: dynamically adjusting the weight of each data source according to the driving condition, and obtaining the comprehensive energy consumption data corresponding to different driving conditions based on the data sources with updated weights;
[0043] In actual operation, this step can specifically refer to:
[0044] Based on the acceleration in the driving behavior data, the driving condition type is identified and marked in the corresponding four-dimensional multi-source data (the correspondence mentioned here means that certain acceleration data can be associated and marked with other data collected during the same period in combination with timestamp alignment); based on the marked driving conditions, the data sources are weighted (which can be learned online and dynamically adjusted) and the comprehensive energy consumption value of each driving condition is calculated (the method of calculating energy consumption values based on vehicle-related multi-dimensional data is not the focus of the present invention, and mature technologies can be used for reference during implementation).
[0045] It is necessary to expand on the method of identifying driving conditions based on acceleration, which specifically includes: based on the comparison relationship between acceleration data and preset acceleration and deceleration thresholds, and combined with the duration of acceleration and deceleration, at least the following driving condition types are obtained: acceleration stage, cruising stage, and braking stage.
[0046] Furthermore, the weight distribution mechanism mentioned above, that is, the dynamic impact factor distribution strategy for different driving conditions, can be referred to as follows:
[0047] (1) During the acceleration phase, the weight of driving behavior data can be increased (e.g., highlighting the impact of rapid acceleration on energy consumption).
[0048] (2) During the cruising phase, the weight of vehicle status and road topology data can be increased (e.g., highlighting the impact of vehicle speed and slope on energy consumption).
[0049] (3) During the braking phase, the energy recovery efficiency weight can be increased (e.g., highlighting the impact of braking energy recovery on the remaining mileage).
[0050] Step S3: Using the four-dimensional multi-source data and the comprehensive energy consumption data, a remaining mileage prediction model is trained; wherein the architecture of the remaining mileage prediction model training includes: a first sub-model for predicting the remaining mileage based on long-term data, a second sub-model for predicting the remaining mileage based on short-term data, and an arbitration output module for dynamically adjusting the sub-model weights based on prediction error feedback and comprehensively calculating the remaining mileage.
[0051] The training of the remaining mileage prediction model specifically includes:
[0052] According to the established time standards, long-term data and short-term data are divided and selected from the four-dimensional multi-source data and comprehensive energy consumption data;
[0053] Input the long-term data into the first sub-model for processing and obtain the first prediction result of the remaining mileage P XGB ;
[0054] The short-term data is input into the second sub-model for processing to obtain the second prediction result of the remaining mileage P LSMT ;
[0055] Based on the prediction error, calculate the confidence of the first sub-model and the second sub-model respectively; (for example, confidence = 1 / prediction variance)
[0056] Dynamically set the first weight W corresponding to the first sub-model and the second sub-model according to the confidence XGB With the second weight W LSTM ;
[0057] Based on the first remaining mileage prediction result and the first weight, the second remaining mileage prediction result and the second weight, a fusion calculation is performed to obtain the comprehensive remaining mileage (for example, W XGB P XGB +W LSTM P LSMT ), as the final output of the remaining mileage prediction model.
[0058] Specifically, in some embodiments, the prediction model can be constructed using a hybrid prediction architecture combining an XGBoost regression model (for long-term data) and an LSTM real-time prediction network (for short-term data) with an arbitration mechanism. Those skilled in the art will appreciate that the XGBoost model is merely illustrative and, in practice, can be replaced with models such as LightGBM or CatBoost; and the LSTM network can be replaced with models such as GRU (Gated Recurrent Unit) or Transformer.
[0059] The XGBoost regression model, based on the Gradient Boosted Decision Tree (GBDT), is suitable for regression and classification tasks on structured data. It features strong interpretability, the ability to analyze feature importance, robustness to outliers, adaptability to noise in historical data, fast training speed, and suitability for offline training of long-term data. For example, based on seven days of historical data (seven days of four-dimensional data divided by time and corresponding comprehensive energy consumption data), it can predict the remaining mileage for a period of time, thereby reflecting the long-term impact of a driver's driving habits and fixed routes on energy consumption.
[0060] The LSTM real-time prediction network excels at processing time series data, capturing instantaneous driving behavior and adapting to dynamic environmental changes. For example, based on 15 minutes of real-time data, it can output the current dynamic SOC mileage to respond to unexpected road conditions.
[0061] Afterwards, through the error feedback loop, when long-term and short-term prediction results conflict, the model output with higher confidence can be given priority, thus taking into account both the global trend and instantaneous accuracy.
[0062] In summary, the main design concept of this invention is to achieve high-precision prediction of the remaining range of electric vehicles by constructing a four-dimensional multi-source database, dynamically assigning data source weights based on driving condition identification and calculating the corresponding energy consumption, and combining an arbitration output mechanism that integrates long-term and short-term data. The construction of the four-dimensional database enables the integration of multi-source data to improve prediction accuracy; the weight configuration based on driving conditions significantly improves flexible adaptability to complex driving scenarios; and the use of long-term and short-term fusion arbitration enables comprehensive prediction of energy consumption trends across different time scales.
[0063] If the expressions expressing directions are mentioned in the embodiments of the present invention, they are relative concepts based on the embodiments. In addition, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can represent: a, b, c, a and b, a and c, b and c or a, b and c, where a, b, c can be single or multiple.
[0064] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings, but the above is only a preferred embodiment of the present invention. It should be noted that the technical features involved in the above embodiments and their preferred modes can be reasonably combined and matched into a variety of equivalent schemes by those skilled in the art without departing from or changing the design ideas and technical effects of the present invention; therefore, the scope of implementation of the present invention is not limited to what is shown in the drawings. Any changes made in accordance with the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the scope of protection of the present invention.
Claims
1. A method for constructing a remaining range prediction model for an electric vehicle, characterized in that: include: Collecting four-dimensional multi-source data, the four-dimensional multi-source data including: vehicle status data, environmental data, road topology data, and driving behavior data; Dynamically adjust the weight of each data source according to the driving conditions, and obtain the comprehensive energy consumption data corresponding to different driving conditions based on the weighted data sources; A remaining mileage prediction model is trained using four-dimensional multi-source data and comprehensive energy consumption data. The architecture of the remaining mileage prediction model includes: a first sub-model that predicts the remaining mileage based on long-term data, a second sub-model that predicts the remaining mileage based on short-term data, and an arbitration output module that dynamically adjusts the sub-model weights based on prediction error feedback and comprehensively calculates the remaining mileage.
2. The method for constructing a remaining range prediction model for an electric vehicle according to claim 1, wherein: The training of the remaining mileage prediction model specifically includes: According to the established time standards, long-term data and short-term data are divided and selected from the four-dimensional multi-source data and comprehensive energy consumption data; Input the long-term data into the first sub-model for processing to obtain the first remaining mileage prediction result; Input the short-term data into the second sub-model for processing to obtain the second remaining mileage prediction result; Based on the prediction error, the confidence of the first sub-model and the second sub-model are calculated respectively; Dynamically setting a first weight and a second weight corresponding to the first sub-model and the second sub-model according to the confidence level; A fusion calculation is performed based on the first remaining mileage prediction result and the first weight, the second remaining mileage prediction result and the second weight to obtain a comprehensive remaining mileage as the final output of the remaining mileage prediction model.
3. The method for constructing a remaining range prediction model for an electric vehicle according to claim 1 or 2, wherein: The obtaining of comprehensive energy consumption data corresponding to different driving conditions includes: Based on the acceleration data in the driving behavior data, identifying the driving condition type and marking it in the corresponding four-dimensional multi-source data; Based on the marked driving condition types, the data sources are weighted and the comprehensive energy consumption value of each driving condition is calculated.
4. The method for constructing a remaining range prediction model for an electric vehicle according to claim 3, wherein: The marking includes, in the corresponding four-dimensional multi-source data: based on the time stamp alignment processing of the data source, associating the acceleration data with other data collected during the same period.
5. The method for constructing a remaining range prediction model for an electric vehicle according to claim 3, wherein: The identification of the driving condition type includes: based on the comparison relationship between the acceleration data and the preset acceleration and deceleration thresholds, and in combination with the preset acceleration and deceleration state duration, obtaining at least the following driving condition types: acceleration stage, cruising stage, and braking stage.
6. The method for constructing a remaining range prediction model for an electric vehicle according to claim 5, characterized in that: The dynamically adjusting the weight of each data source according to the driving condition includes: For the acceleration phase, at least some of the driving behavior data weights are increased; For the cruising phase, at least the weights of some data in the vehicle status data and road topology data are increased; During the braking phase, at least some of the driving behavior data weights are increased.