A vehicle energy consumption prediction method and device

Abstract

This application provides a vehicle energy consumption prediction method and apparatus, applied in the field of vehicle energy consumption prediction technology. The method may include determining energy consumption impact data related to vehicle energy consumption during the current driving process of the target vehicle, based on vehicle data from the target vehicle's current driving process, under the current energy consumption prediction scenario. The energy consumption impact data is then input into multiple pre-trained primary prediction units to obtain the primary energy consumption prediction results of each primary prediction unit. These multiple primary energy consumption prediction results are then input into a pre-trained fusion prediction unit to obtain the energy consumption prediction result of the target vehicle under the current energy consumption prediction scenario. This method can solve the problems of insufficient adaptability to the current operating conditions and low prediction accuracy in traditional vehicle energy consumption prediction methods.

Description

Technical Field

[0001] This application relates to the field of vehicle energy consumption prediction technology, and in particular to a vehicle energy consumption prediction method and apparatus. Background Technology

[0002] With the rapid development of new energy vehicles, the driving range and energy consumption of electric vehicles have gradually become important indicators of concern for users. Vehicle energy consumption is affected by a variety of factors, such as vehicle operating data, driving environment data, battery status data, and energy-consuming component status data. During actual driving, these factors constantly change with changes in road conditions, external environment, and vehicle usage, making vehicle energy consumption exhibit strong nonlinearity and uncertainty.

[0003] In related technologies, energy consumption prediction models are typically built based on historical vehicle operating data, and these models are used to predict vehicle energy consumption. This method can reflect the relationship between vehicle operating status and energy consumption to some extent. However, in practical applications, because the factors affecting vehicle energy consumption differ under different driving conditions, this method is not adaptable enough to the current operating conditions of the vehicle, and the prediction results usually fail to intuitively reflect the correlation between various types of vehicle data and energy consumption.

[0004] Therefore, there is an urgent need for a vehicle energy consumption prediction method to solve the problems of insufficient adaptability to the current operating conditions of vehicles and low accuracy of prediction results in traditional vehicle energy consumption prediction methods. Summary of the Invention

[0005] The purpose of this application is to provide a vehicle energy consumption prediction method and device to solve the problems of insufficient adaptability to the current operating conditions of vehicles and low accuracy of prediction results in traditional vehicle energy consumption prediction methods.

[0006] In a first aspect, embodiments of this application provide a vehicle energy consumption prediction method. The method includes: determining energy consumption impact data related to vehicle energy consumption during the current driving process of a target vehicle based on vehicle data from the vehicle itself; inputting the energy consumption impact data into multiple pre-trained primary prediction units to obtain primary energy consumption prediction results from each primary prediction unit; and inputting the multiple primary energy consumption prediction results into a pre-trained fusion prediction unit to obtain the energy consumption prediction result of the target vehicle in the current energy consumption prediction scenario.

[0007] The vehicle energy consumption prediction method provided in this application determines the energy consumption impact data related to vehicle energy consumption in the current energy consumption prediction scenario based on vehicle data during the target vehicle's current driving process, so that the data input into the model can match the vehicle's current operating scenario. The energy consumption impact data is then input into multiple primary prediction units for prediction, and the results of these multiple primary energy consumption predictions are fused through a fusion prediction unit. This integrates the representational capabilities of different prediction units on the relationship between vehicle energy consumption changes, reduces the impact of single-model prediction bias on the final prediction result, and thus improves the accuracy and stability of the vehicle energy consumption prediction results.

[0008] One possible implementation involves jointly training multiple primary prediction units and a fusion prediction unit. This joint training is achieved through the following steps: acquiring historical energy consumption impact data related to vehicle energy consumption during vehicle operation under historical energy consumption prediction scenarios, and the corresponding historical energy consumption results; constructing a training dataset based on the historical energy consumption impact data and historical energy consumption results; and training multiple primary prediction units and a fusion prediction unit based on the training dataset, resulting in trained primary prediction units and a trained fusion prediction unit.

[0009] One possible implementation involves, before training multiple primary prediction units and the fused prediction unit based on the training dataset, further comprising: determining the target parameter configuration of the first primary prediction unit based on the training dataset and a preset parameter search strategy; and constructing the first primary prediction unit based on the target parameter configuration.

[0010] One possible implementation involves determining the target parameter configuration of the first primary prediction unit based on a training dataset and a pre-defined parameter search strategy. This includes: determining multiple model parameters to be optimized for the first primary prediction unit, and the corresponding parameter value ranges for each parameter. Based on the values ​​of the multiple model parameters within their respective value ranges, generating multiple parameter configurations for the first primary prediction unit. Constructing candidate first primary prediction units based on each parameter configuration. Training and validating each candidate first primary prediction unit using the training dataset to obtain the prediction error of each candidate first primary prediction unit. Based on the prediction errors of each candidate first primary prediction unit, determining the target parameter configuration from the parameter configurations, and identifying the candidate first primary prediction unit corresponding to the target parameter configuration as the first primary prediction unit.

[0011] One possible implementation involves a set of parameter configurations corresponding to a search individual. Based on the prediction errors of each candidate first primary prediction unit, the target parameter configuration is determined from the parameter configurations, including: determining the fitness of each search individual based on the prediction errors of the candidate first primary prediction units corresponding to each search individual; determining the individual type of each search individual based on its fitness; iteratively updating the parameter configurations corresponding to each search individual based on its individual type; and determining the target parameter configuration based on the fitness of each search individual, provided that the iterative update meets a preset iteration stopping condition.

[0012] One possible implementation involves iteratively updating the parameter configurations for each search individual based on its individual type. This includes: updating the parameter configuration for the first search individual (individual type 1) to expand the search range; updating the parameter configuration for the second search individual (individual type 2) based on the current best search individual; and perturbing the parameter configuration for the third search individual (individual type 3) if a preset warning condition is met. Finally, the fitness of each search individual is re-determined based on the updated parameter configurations.

[0013] One possible implementation involves determining energy consumption impact data related to vehicle energy consumption during the target vehicle's current driving process, based on vehicle data from the current driving process. This includes: determining the target vehicle's current energy consumption prediction scenario based on driving-related data from the vehicle data; and filtering out energy consumption impact data related to vehicle energy consumption from the vehicle data based on the current energy consumption prediction scenario.

[0014] One possible implementation involves determining the current energy consumption prediction scenario for the target vehicle based on vehicle data indicating vehicle operation during its current driving process. This includes: determining the target vehicle's driving condition characteristics based on vehicle operation data in the vehicle data; determining the target vehicle's environmental condition characteristics based on driving environment data in the vehicle data; determining the target vehicle's battery operating characteristics based on battery status data in the vehicle data; determining the target vehicle's additional load characteristics based on energy-consuming component status data in the vehicle data; and finally, determining the target vehicle's current energy consumption prediction scenario based on the driving condition characteristics, environmental condition characteristics, battery operating characteristics, and additional load characteristics.

[0015] One possible implementation involves inputting energy consumption impact data into multiple pre-trained primary prediction units to obtain the primary energy consumption prediction results of each unit. This includes: determining the scenario parameter configurations for the multiple primary prediction units under the current energy consumption prediction scenario; and inputting the energy consumption impact data into the multiple primary prediction units configured with the scenario parameters to obtain the primary energy consumption prediction results output by each unit.

[0016] Secondly, embodiments of this application provide a vehicle energy consumption prediction device, which includes a determination module and an input module.

[0017] The determination module is used to determine the energy consumption impact data related to vehicle energy consumption during the current energy consumption prediction scenario, based on the vehicle data of the target vehicle during its current driving process.

[0018] The input module is used to input energy consumption impact data into multiple pre-trained primary prediction units to obtain the primary energy consumption prediction results of each primary prediction unit.

[0019] The input module is also used to input multiple primary energy consumption prediction results into a pre-trained fusion prediction unit to obtain the energy consumption prediction result of the target vehicle in the current energy consumption prediction scenario.

[0020] One possible implementation involves jointly training multiple primary prediction units and a fusion prediction unit. This joint training is achieved through the following steps: acquiring historical energy consumption impact data related to vehicle energy consumption during vehicle operation under historical energy consumption prediction scenarios, and the corresponding historical energy consumption results; constructing a training dataset based on the historical energy consumption impact data and historical energy consumption results; and training multiple primary prediction units and a fusion prediction unit based on the training dataset, resulting in trained primary prediction units and a trained fusion prediction unit.

[0021] In one possible implementation, before training multiple primary prediction units and the fusion prediction unit based on the training dataset, the device is further configured to: determine the target parameter configuration of the first primary prediction unit based on the training dataset and a preset parameter search strategy; and construct the first primary prediction unit based on the target parameter configuration.

[0022] One possible implementation involves a determination module, used to determine the target parameter configuration of the first primary prediction unit based on the training dataset and a preset parameter search strategy. Specifically, this module: determines multiple model parameters to be optimized for the first primary prediction unit, and the corresponding parameter value ranges for each parameter. Based on the values ​​of the multiple model parameters within their respective value ranges, it generates multiple parameter configurations for the first primary prediction unit. Based on each parameter configuration, it constructs candidate first primary prediction units. It trains and validates each candidate first primary prediction unit using the training dataset, obtaining the prediction error of each candidate first primary prediction unit. Based on the prediction errors of each candidate first primary prediction unit, it determines the target parameter configuration from the parameter configurations and identifies the candidate first primary prediction unit corresponding to the target parameter configuration as the first primary prediction unit.

[0023] One possible implementation is that a set of parameter configurations corresponds to one search individual. The determination module, used to determine the target parameter configuration from the parameter configurations based on the prediction errors of each candidate first primary prediction unit, specifically performs the following steps: determining the fitness of each search individual based on the prediction errors of the candidate first primary prediction units corresponding to each search individual; determining the individual type of each search individual based on its fitness; iteratively updating the parameter configurations corresponding to each search individual based on its individual type; and determining the target parameter configuration based on the fitness of each search individual, provided that the iterative update meets a preset iteration stopping condition.

[0024] One possible implementation involves a module that, when iteratively updating the parameter configurations for each search individual based on its individual type, specifically: updating the parameter configurations for the first search individual (individual type 1) to expand the search range; updating the parameter configurations for the second search individual (individual type 2) based on the current optimal search individual; and, under preset warning conditions, perturbing the parameter configurations for the third search individual (individual type 3) with a perturbation update. Finally, the fitness of each search individual is re-determined based on the updated parameter configurations.

[0025] One possible implementation involves a determination module that, based on vehicle data during the current driving process of the target vehicle, determines energy consumption impact data related to vehicle energy consumption during the current energy consumption prediction scenario. Specifically, this module is used to: determine the current energy consumption prediction scenario for the target vehicle based on vehicle data indicating vehicle driving-related data during the current driving process; and, based on the current energy consumption prediction scenario, filter out energy consumption impact data related to vehicle energy consumption from the vehicle data.

[0026] One possible implementation involves a determination module that, based on vehicle data indicating vehicle driving-related data during the current driving process, determines the current energy consumption prediction scenario for the target vehicle. Specifically, this module: determines the target vehicle's driving condition characteristics based on vehicle operation data within the vehicle data; determines the target vehicle's environmental condition characteristics based on driving environment data within the vehicle data; determines the target vehicle's battery operating characteristics based on battery status data within the vehicle data; determines the target vehicle's additional load characteristics based on energy-consuming component status data within the vehicle data; and finally, based on the driving condition characteristics, environmental condition characteristics, battery operating characteristics, and additional load characteristics, determines the target vehicle's current energy consumption prediction scenario.

[0027] One possible implementation involves an input module that, when inputting energy consumption impact data into multiple pre-trained primary prediction units to obtain the primary energy consumption prediction results of each unit, specifically: determining the scenario parameter configurations corresponding to the multiple primary prediction units under the current energy consumption prediction scenario; inputting the energy consumption impact data into the multiple primary prediction units configured with scenario parameters to obtain the primary energy consumption prediction results output by each primary prediction unit.

[0028] Thirdly, embodiments of this application provide a vehicle energy consumption prediction device, which has the function of implementing the vehicle energy consumption prediction method of the first aspect or any possible implementation of the first aspect. This function can be implemented by hardware or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above-described function.

[0029] Fourthly, embodiments of this application provide a computer-readable storage medium storing instructions that, when executed on a computer, enable the computer to perform the vehicle energy consumption prediction method of the first aspect or any possible implementation thereof.

[0030] Fifthly, embodiments of this application provide a computer program product containing instructions that, when run on a computer, enable the computer to execute the vehicle energy consumption prediction method of the first aspect or any possible implementation thereof.

[0031] The technical effects of any of the design methods in aspects two through five can be found in aspect one or in different possible implementations of aspect one, and will not be repeated here. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the specific embodiments of this application or the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0033] Figure 1 A system structure diagram of a vehicle energy consumption prediction system provided in this application embodiment; Figure 2 A flowchart of a vehicle energy consumption prediction method provided in this application embodiment; Figure 3 This is another method flowchart of a vehicle energy consumption prediction method provided in an embodiment of this application; Figure 4 A schematic diagram of a vehicle energy consumption prediction device provided in an embodiment of this application; Figure 5 Another system architecture diagram of a vehicle energy consumption prediction system provided in this application embodiment. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0035] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0036] In related technologies, energy consumption prediction models are typically built based on historical vehicle driving data, and these models are used to predict the energy consumption of vehicles in the future or under current operating conditions. While this method can reflect the relationship between vehicle operating data and energy consumption to some extent, the actual operating scenarios of vehicles vary significantly, and the main influencing factors on vehicle energy consumption are not entirely the same. If multiple types of collected vehicle data are used as a unified input for prediction, the lack of differentiation between the current energy consumption prediction scenarios and the difficulty in extracting corresponding key energy consumption influencing data for different scenarios lead to redundant or weakly correlated factors in the model input, thus affecting the accuracy and stability of the energy consumption prediction results.

[0037] Based on this, embodiments of this application provide a vehicle energy consumption prediction method and apparatus. The method includes determining energy consumption impact data related to vehicle energy consumption during the current driving process of the target vehicle, based on vehicle data from the target vehicle's current driving process, under the current energy consumption prediction scenario. The energy consumption impact data is then input into multiple pre-trained primary prediction units to obtain primary energy consumption prediction results from each primary prediction unit. Finally, the multiple primary energy consumption prediction results are input into a pre-trained fusion prediction unit to obtain the energy consumption prediction result of the target vehicle under the current energy consumption prediction scenario.

[0038] The vehicle energy consumption prediction method provided in this application determines the energy consumption impact data related to vehicle energy consumption in the current energy consumption prediction scenario based on vehicle data during the target vehicle's current driving process, so that the data input into the model can match the vehicle's current operating scenario. The energy consumption impact data is then input into multiple primary prediction units for prediction, and the results of these multiple primary energy consumption predictions are fused through a fusion prediction unit. This integrates the representational capabilities of different prediction units on the relationship between vehicle energy consumption changes, reduces the impact of single-model prediction bias on the final prediction result, and thus improves the accuracy and stability of the vehicle energy consumption prediction results.

[0039] The methods provided in the embodiments of this application will now be described in conjunction with the specific accompanying drawings.

[0040] On the one hand, embodiments of this application provide a vehicle energy consumption prediction system. For example... Figure 1 As shown, the vehicle energy consumption prediction system 100 may include: multiple sensors 101, a vehicle controller 102, a battery management system 103, an energy consumption component controller 104, an on-board computing device 105, a memory 106, and a display device 107.

[0041] The multiple sensors 101 are used to collect vehicle data of the target vehicle during its current driving process. The multiple sensors 101 may include at least one of the following: vehicle speed sensor, acceleration sensor, slope sensor, ambient temperature sensor, wind speed sensor, wind direction sensor, humidity sensor, tire pressure sensor, and battery temperature sensor.

[0042] One possible implementation involves using a vehicle speed sensor to collect the target vehicle's speed data; an acceleration sensor to collect the target vehicle's acceleration data; a gradient sensor to collect the road gradient data of the road where the target vehicle is located; ambient temperature, wind speed, wind direction, and humidity sensors to collect driving environment data of the target vehicle's surroundings; a tire pressure sensor to collect the target vehicle's tire pressure data; and a battery temperature sensor to collect the target vehicle's battery temperature data.

[0043] The vehicle controller 102 can be used to acquire data such as the target vehicle's driving mode, vehicle mass, vehicle load, regenerative braking status, and vehicle operating status.

[0044] The battery management system 103 can be used to acquire battery status information such as the battery state of charge, battery voltage, battery current, and battery temperature of the target vehicle.

[0045] The energy consumption component controller 104 can be used to obtain the status of air conditioning, cabin heating, defrosting, and the operating status of other electrical components.

[0046] The on-board computing device 105 can communicate with multiple sensors 101, vehicle controller 102, battery management system 103 and energy consumption component controller 104 respectively, to determine energy consumption impact data based on the collected vehicle data, and input the energy consumption impact data into multiple pre-trained primary prediction units and fusion prediction units to obtain the energy consumption prediction result of the target vehicle in the current energy consumption prediction scenario.

[0047] The memory 106 can be used to store vehicle historical data, historical energy consumption results, multiple trained primary prediction units, trained fusion prediction units, and data configuration relationships corresponding to different energy consumption prediction scenarios.

[0048] The display device 107 can be installed in the vehicle cabin to display energy consumption prediction results, range estimation results, or energy consumption analysis results.

[0049] It should be noted that the above Figure 1 The vehicle energy consumption prediction system 100 shown is an example of an application scenario of the embodiments of this application, and is not intended to limit the application scenarios of the embodiments of this application.

[0050] On one hand, embodiments of this application provide a vehicle energy consumption prediction method. To facilitate the explanation of the vehicle energy consumption prediction method provided in this application, the joint training process of multiple primary prediction units and a fusion prediction unit provided in this application is first explained. The multiple primary prediction units and the fusion prediction unit are jointly trained. Specifically, the multiple primary prediction units are used to perform primary predictions of vehicle energy consumption respectively, and the fusion prediction unit is used to fuse the primary energy consumption prediction results output by the multiple primary prediction units to obtain the final energy consumption prediction result. The multiple primary prediction units and the fusion prediction unit are jointly trained through the following steps: [Example...] Figure 2 As shown, the method may include the following steps.

[0051] S201, acquire historical energy consumption impact data related to vehicle energy consumption during the vehicle's operation in the historical energy consumption prediction scenario, as well as historical energy consumption results corresponding to the historical energy consumption impact data.

[0052] The historical energy consumption prediction scenario can be the driving scenario corresponding to the vehicle's historical driving process. This historical energy consumption prediction scenario can be determined based on the vehicle's operating status, driving environment status, battery operating status, and energy consumption component operating status during the historical driving process.

[0053] For example, historical energy consumption prediction scenarios may include high-speed driving scenarios, urban road driving scenarios, slope driving scenarios, low-temperature driving scenarios, air conditioning-on scenarios, regenerative braking scenarios, and other driving scenarios that affect vehicle energy consumption.

[0054] Historical energy consumption impact data can include various types of data related to vehicle energy consumption during vehicle operation under historical energy consumption prediction scenarios. This data can include vehicle operating information, driving environment information, battery status information, and energy-consuming component status information.

[0055] The vehicle operation information may include vehicle speed, acceleration, road gradient, driving mode, vehicle load, vehicle mass, cargo mass, tire pressure, tire type, and air drag coefficient; the driving environment information may include ambient temperature, wind speed, wind direction, and air humidity; the battery status information may include battery temperature, battery state of charge, battery voltage, and battery current; and the energy consumption component status information may include air conditioning status, regenerative braking status, cabin heating status, and the status of other electrical components.

[0056] Historical energy consumption results are the actual energy consumption data corresponding to the historical energy consumption impact data.

[0057] For example, historical energy consumption results may include the actual power consumption corresponding to a historical driving segment, energy consumption per unit mileage, power consumption per 100 kilometers, or energy consumption value within a preset time window. Historical energy consumption results can be determined based on data such as battery charge changes, vehicle mileage, and vehicle driving time collected by the battery management system. For instance, the energy consumption per unit mileage corresponding to a historical driving segment can be determined based on the change in battery state of charge, battery capacity, and mileage of the historical driving segment.

[0058] One possible implementation involves acquiring raw vehicle data from the vehicle's historical driving process according to a preset sampling period. Based on the raw vehicle data and actual energy consumption data corresponding to the same sampling time or the same driving segment, a set of historical samples is formed. Each set of historical samples may include historical energy consumption impact data and the historical energy consumption result corresponding to that historical energy consumption impact data.

[0059] S202, construct a training dataset based on historical energy consumption impact data and historical energy consumption results.

[0060] Specifically, after obtaining historical energy consumption impact data and results, the historical energy consumption impact data can be preprocessed. This preprocessing process can include one or more of the following: data cleaning, data transformation, standardization, and feature selection.

[0061] Data cleaning can include processing outliers, missing data, and duplicate data in historical energy consumption impact data. For example, data that clearly does not conform to vehicle operating patterns can be removed, or missing data can be filled in. Data transformation can include standardizing vehicle data from different sources and in different formats. Standardization can be used to eliminate the impact of differences in feature units on the model training process, ensuring that different types of data, such as vehicle speed, acceleration, ambient temperature, battery state of charge, and air conditioning status, are on the same or similar data scale.

[0062] For example, this standardization process can be implemented using the following formula.

[0063]

[0064] in, The standardized results for the features The mean of the features, The standard deviation of the feature The original value of the feature.

[0065] Furthermore, after preprocessing the historical energy consumption impact data, feature filtering is performed on the preprocessed historical energy consumption impact data to remove redundant or weakly correlated features.

[0066] One possible approach is to use vehicle energy consumption per unit mile as the target variable, establish a regularized regression model based on standardized historical energy consumption impact data, and determine the correlation between each historical energy consumption impact factor and vehicle energy consumption based on the regression coefficients corresponding to each historical energy consumption impact factor in the regularized regression model. Historical energy consumption impact factors whose regression coefficients meet preset screening criteria are identified as target historical energy consumption impact factors, and training input data is constructed based on these target historical energy consumption impact factors.

[0067] For example, a regularized regression model can be a LASSO regression model. By using a LASSO regression model to select features from historical energy consumption data, the impact of redundant inputs on the model training and prediction process can be reduced.

[0068] When constructing the training dataset, preprocessed and feature-filtered historical energy consumption impact data can be used as training samples, and the historical energy consumption results corresponding to these historical energy consumption impact data can be used as sample labels to obtain the training dataset. The training dataset is then divided into training, validation, and test sets according to a preset ratio.

[0069] The training set is used to train the primary prediction units, the validation set is used to determine the model parameter configuration and train the fusion prediction units, and the test set is used to test and evaluate the trained primary prediction units and fusion prediction units.

[0070] S203, train multiple primary prediction units and fusion prediction units based on the training dataset to obtain multiple trained primary prediction units and fusion prediction units.

[0071] Specifically, a Stacking ensemble learning framework can be used to predict vehicle energy consumption. This Stacking ensemble learning framework can include a first-layer prediction structure and a second-layer prediction structure. The first-layer prediction structure includes multiple primary prediction units, each outputting a primary energy consumption prediction result based on historical energy consumption impact data. The second-layer prediction structure includes a fusion prediction unit, which uses the primary energy consumption prediction results output by the multiple primary prediction units as fusion input data and outputs the final energy consumption prediction result based on the fusion input data.

[0072] For example, the multiple primary prediction units may include a first primary prediction unit, a second primary prediction unit, and a third primary prediction unit. The first primary prediction unit may be a random forest (RF) prediction unit, the second primary prediction unit may be a lightweight gradient boosting machine (LightGBM) prediction unit, and the third primary prediction unit may be a support vector regression (SVR) prediction unit. The fusion prediction unit may be a linear regression prediction unit.

[0073] Furthermore, a five-fold cross-validation method can be used to divide the training dataset into five training subsets and validation subsets. In each round of cross-validation, one set of data is used as the validation subset, and the remaining data is used as the training subset. The first primary prediction unit, the second primary prediction unit, and the third primary prediction unit are then trained based on this training subset.

[0074] When training the first primary prediction unit, a random forest prediction unit can be constructed based on a predetermined target parameter configuration, and this random forest prediction unit can be trained using a training subset. This target parameter configuration can include the number of trees in the random forest prediction unit, the minimum number of leaf samples, the number of features involved in each split, the tree depth, and the sampling parameters. The second primary prediction unit can be trained using a lightweight gradient boosting method, and the third primary prediction unit can be trained using a support vector regression method.

[0075] The validation subsets corresponding to each round of cross-validation are then input into the trained first, second, and third primary prediction units to obtain the first, second, and third validation prediction results. For the same validation sample, the first, second, and third validation prediction results can be combined in a preset order to form the fused training features corresponding to that validation sample. By summarizing the fused training features obtained from each round of cross-validation, the fused training data corresponding to the training dataset can be obtained.

[0076] After obtaining the fusion training data, the fusion training data is used as the input to the fusion prediction unit, and the historical energy consumption results corresponding to the fusion training data are used as the output labels of the fusion prediction unit to train the fusion prediction unit.

[0077] After the fusion prediction unit is trained, the first primary prediction unit, the second primary prediction unit, and the third primary prediction unit can be retrained separately based on the complete training dataset to obtain multiple primary prediction units for final prediction.

[0078] Furthermore, before training multiple primary prediction units, a benchmark energy consumption model with physical guidance can be constructed based on the vehicle's energy consumption mechanism. This benchmark energy consumption model is used to characterize the correspondence between vehicle energy consumption per unit mile and characteristic variables. Vehicle energy consumption per unit mile can be expressed in Wh / km, and characteristic variables can include vehicle speed, acceleration, road gradient, vehicle mass, rolling resistance-related parameters, air resistance coefficient, wind speed, wind direction, battery status, air conditioning status, and regenerative braking status.

[0079] For example, the baseline energy consumption model can be shown below.

[0080]

[0081] in, For vehicle energy consumption, As a baseline energy consumption, These are the model parameters corresponding to the respective features. For vehicle speed, For acceleration, For battery temperature, For ambient temperature, Air conditioning is in operation. For road slope, For cargo capacity, For tire pressure, Driving mode In regenerative braking state, The battery is in its state of charge. For vehicle quality, For tire type, The air drag coefficient, For wind speed, For air humidity, This is the error term.

[0082] Furthermore, before training multiple primary prediction units and the fused prediction unit based on the training dataset, the process may further include: determining the target parameter configuration of the first primary prediction unit based on the training dataset and a preset parameter search strategy; and constructing the first primary prediction unit based on the target parameter configuration.

[0083] Specifically, the preset parameter search strategy can be used to search for target parameter configurations that enable the first primary prediction unit to have better prediction performance within the parameter value range corresponding to multiple model parameters to be optimized in the first primary prediction unit. The first primary prediction unit can be a random forest prediction unit, and the corresponding model parameters to be optimized may include the number of trees, the minimum number of leaf samples, the number of features involved in each split, the tree depth, and sampling parameters.

[0084] One possible implementation involves determining the target parameter configuration of the first primary prediction unit based on the training dataset and a pre-defined parameter search strategy. This may include: determining multiple model parameters to be optimized for the first primary prediction unit, and the corresponding parameter value ranges for each model parameter. Based on the values ​​of the multiple model parameters within their respective parameter value ranges, generating multiple parameter configurations for the first primary prediction unit. Constructing candidate first primary prediction units based on each parameter configuration. Training and validating each candidate first primary prediction unit using the training dataset to obtain the prediction error of each candidate first primary prediction unit. Based on the prediction errors of each candidate first primary prediction unit, determining the target parameter configuration from the parameter configurations, and identifying the candidate first primary prediction unit corresponding to the target parameter configuration as the first primary prediction unit.

[0085] The prediction error can be expressed as root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination.

[0086] For example, when the first primary prediction unit is a random forest prediction unit, a set of parameter configurations may include the number of trees, the minimum number of leaf samples, the number of features involved in each split, and the tree depth. After generating multiple parameter configurations for the first primary prediction unit based on the values ​​of multiple model parameters to be optimized within their corresponding parameter ranges, a candidate first primary prediction unit is constructed based on each set of parameter configurations. Using historical energy consumption impact data and historical energy consumption results from the training dataset, each candidate first primary prediction unit is trained and validated to obtain the prediction error corresponding to each candidate first primary prediction unit. Then, based on the prediction errors of each candidate first primary prediction unit, a target parameter configuration is determined from the parameter configurations, and the candidate first primary prediction unit corresponding to the target parameter configuration is determined as the first primary prediction unit.

[0087] Furthermore, a set of parameter configurations corresponds to one search individual. Based on the prediction errors of each candidate first primary prediction unit, the target parameter configuration is determined from the parameter configurations, including: determining the fitness of each search individual based on the prediction errors of the candidate first primary prediction units corresponding to each search individual; determining the individual type of each search individual based on its fitness; iteratively updating the parameter configuration corresponding to each search individual based on its individual type; and determining the target parameter configuration based on the fitness of each search individual, provided that the iterative update meets a preset iteration stopping condition.

[0088] The preset iteration stopping conditions may include the number of iterations reaching a preset number, the current optimal fitness meeting preset requirements, the current optimal prediction error being less than a preset error threshold, or the change in the current optimal fitness being less than a preset change threshold in multiple consecutive iterations.

[0089] Specifically, a set of parameter configurations can be considered as a search individual. The position of a search individual in the parameter search space can be used to characterize the set of parameter configurations corresponding to that search individual. Multiple search individuals together constitute a parameter search population. For each search individual, its fitness can be determined based on the prediction error of its corresponding candidate first primary prediction unit. During the parameter search process, the current optimal search individual is determined based on the fitness of each search individual. The current optimal search individual can be the search individual whose fitness meets the preset optimization conditions in the current iteration round, such as the search individual with the smallest prediction error. Then, based on the fitness of each search individual, the search individuals are divided into different individual types.

[0090] Based on the individual type of each search individual, the parameter configurations corresponding to each search individual are iteratively updated. In each iteration, the parameter configurations for each search individual are updated according to the update methods corresponding to the first, second, and third types of search individuals, respectively. Then, candidate first primary prediction units are reconstructed based on the updated parameter configurations, and the prediction error and fitness for each search individual are redefined.

[0091] If the updated parameter configuration exceeds the parameter value range of the corresponding model to be optimized, boundary correction can be performed on the updated parameter configuration to bring it within the corresponding parameter value range. Then, based on the redefined fitness, the next iteration begins, until the preset iteration stopping condition is met.

[0092] If the preset iteration stopping condition is met, the target search individual can be determined based on the fitness of each search individual, and the parameter configuration corresponding to the target search individual is determined as the target parameter configuration of the first primary prediction unit. The first primary prediction unit is constructed based on this target parameter configuration and trained using the training dataset to obtain the trained first primary prediction unit.

[0093] One possible implementation involves three types of individuals: a first type, a second type, and a third type. The first type of search individual corresponds to the discoverer individual and is primarily used to expand the parameter search range to find potential optimal parameter regions within the parameter value range. The second type of search individual corresponds to the follower individual and is primarily used to update based on the current best search individual, allowing for further searching near the currently optimal parameter region. The third type of search individual corresponds to the watchdog individual and is primarily used to perturb and update the parameter configuration when preset warning conditions are met, reducing the possibility of the parameter search process getting stuck in local optima.

[0094] Specifically, based on the individual type of each search individual, the parameter configurations corresponding to each search individual are iteratively updated, including: updating the parameter configurations for the first search individual of type 1 to expand the parameter search range; updating the parameter configurations for the second search individual of type 2 based on the current optimal search individual; and perturbing the parameter configurations for the third search individual of type 3 when preset warning conditions are met. The fitness of each search individual is then re-determined based on the updated parameter configurations.

[0095] For example, the first type of search individual can correspond to the discoverer individual in the parameter search group. The discoverer individual is used to perform a large-scale search in the parameter search space to identify potential preferred parameter regions.

[0096] Specifically, the parameter configuration of the first type of search individual can be updated based on the parameter configuration, current iteration number, maximum iteration number, and preset safety threshold of the first type of search individual in the current iteration round.

[0097] For example, the parameter configuration of a first-type search individual can be updated according to the following discoverer update formula.

[0098]

[0099] Where t represents the iteration round, and i represents the index of the individual being searched. This represents the current position of the i-th search individual in the j-th dimension at the (t+1)-th iteration in the population. This is a warning value. As a safety threshold, Let L be a random number and L be a unit vector. This represents the maximum number of iterations.

[0100] Through the aforementioned discoverer update formula, the first type of search individual can perform a global search within the parameter value range, enabling the parameter search process to cover a larger parameter space, thereby increasing the likelihood of discovering a better parameter configuration.

[0101] The second type of search individual corresponds to a follower individual in the parameter search group. Follower individuals are used to update their parameters based on the parameter configuration corresponding to the current best search individual, in order to perform a local search in the vicinity of the current better parameter region.

[0102] Specifically, the parameter configuration of the second type of search individual can be updated based on the current parameter configuration of the second type of search individual, the parameter configuration of the current best search individual, the parameter configuration of the current worst search individual, and the ranking information of the search individual in the parameter search group.

[0103] For example, the parameter configuration of a second-type search individual can be updated according to the following follower update formula:

[0104] in, This represents the position of the worst-performing search individual in generation t. This represents the position of the current best search individual, where n is the total number of search individuals.

[0105] Through the aforementioned follower update formula, the second type of search individual can perform follower search based on the current optimal parameter configuration, enabling the parameter search process to be further refined within the optimal parameter region, thereby improving the optimization accuracy of the target parameter configuration.

[0106] The third type of search individual corresponds to the vigilant individual in the parameter search group. The vigilant individual is used to perturb and update the corresponding parameter configuration when the preset warning conditions are met during the parameter search process, so as to improve the ability of the parameter search process to escape the local optimum.

[0107] Specifically, the preset warning conditions may include the fitness of the current search individual being worse than that of the current best search individual, or the fitness change of the current search population satisfying a preset stagnation condition. When the preset warning conditions are met, the parameter configurations of the third type of search individual can be updated based on the parameter configurations of the current best search individual, the current search individual, the current worst search individual, and a random perturbation factor.

[0108] For example, the parameter configuration of a third type of search individual can be updated according to the following vigilant update formula:

[0109] in, Let represent the fitness of the i-th search individual. This represents the current global optimal fitness, and randn is a random number that follows a standard normal distribution.

[0110] Through the aforementioned vigilant update formula, the third type of search individual can perturb and update the parameter configuration when the parameter search process encounters local stagnation or search risks, thereby enhancing the ability of the parameter search process to escape local optima and improving the reliability of the final determined target parameter configuration.

[0111] On one hand, embodiments of this application provide a method for predicting vehicle energy consumption. This method can be... Figure 1 The vehicle energy consumption prediction system 100 shown is executed. For example... Figure 3 As shown, the method may include the following steps.

[0112] S301, Based on the vehicle data of the target vehicle during its current driving process, determine the energy consumption impact data related to vehicle energy consumption during the driving process of the target vehicle under the current energy consumption prediction scenario.

[0113] The vehicle data can include vehicle operation data, driving environment data, battery status data, and energy-consuming component status data during the current driving process. Vehicle operation data can include vehicle speed, acceleration, road gradient, driving mode, vehicle load, vehicle mass, tire pressure, tire type, and drag coefficient. Driving environment data can include ambient temperature, wind speed, wind direction, and air humidity. Battery status data can include battery temperature, battery state of charge, battery voltage, and battery current. Energy-consuming component status data can include air conditioning status, regenerative braking status, cabin heating status, defrosting status, and the status of other electrical components.

[0114] For example, vehicle data of the target vehicle during its current driving process can be obtained through vehicle sensors, vehicle controllers, battery management systems, and energy consumption component controllers.

[0115] In one possible implementation, after acquiring vehicle data of the target vehicle during its current driving process, the current energy consumption prediction scenario for the target vehicle is determined based on data indicating vehicle driving-related aspects within that data. Based on this current energy consumption prediction scenario, energy consumption impact data related to vehicle energy consumption is filtered from the vehicle data.

[0116] One possible approach is to determine the driving condition characteristics of the target vehicle based on vehicle operation data from the vehicle data.

[0117] Specifically, driving condition characteristics can be used to characterize whether the target vehicle is currently in a state such as high-speed driving, low-speed driving, frequent acceleration and deceleration, driving on an incline, or smooth cruising. For example, the speed state of the target vehicle can be determined based on vehicle speed, the acceleration and deceleration state can be determined based on acceleration, the incline state can be determined based on road gradient, and the power output tendency of the target vehicle can be determined based on driving mode.

[0118] One possible approach is to determine the environmental operating condition characteristics of the target vehicle based on the driving environment data in the vehicle data.

[0119] Specifically, environmental operating condition characteristics can be used to characterize the impact of the external environment on vehicle energy consumption. For example, ambient temperature can be used to determine whether the vehicle is in a low-temperature or high-temperature environment, wind speed and direction can be used to determine the changes in air resistance, and air humidity can be used to determine the impact of the external environment on air conditioning or defogging requirements.

[0120] One possible implementation is to determine the battery operating characteristics of the target vehicle based on battery state data in the vehicle data.

[0121] Specifically, battery operating characteristics can be used to characterize the current operating efficiency of the power battery. For example, the battery temperature can be used to determine whether the battery is in a temperature-sensitive state, the battery's current remaining charge range can be determined based on the battery's state of charge, and the power battery's output state can be determined based on the battery voltage and battery current.

[0122] One possible implementation involves determining the additional load characteristics of the target vehicle based on the energy-consuming component status data in the vehicle data.

[0123] Specifically, additional load characteristics can be used to characterize energy consumption demands other than those for driving the vehicle. For example, the status of air conditioning, cabin heating, defrosting, and other electrical components can be used to determine whether the target vehicle currently has a high additional energy consumption load. The status of regenerative braking can be used to determine whether the target vehicle is currently undergoing an energy recovery process.

[0124] After obtaining the driving condition characteristics, environmental condition characteristics, battery operation characteristics, and additional load characteristics, the current energy consumption prediction scenario of the target vehicle can be determined based on one or more of the above characteristics.

[0125] For example, when the target vehicle's speed is high and wind speed or direction significantly affects the vehicle's drag, the current energy consumption prediction scenario can be identified as a wind resistance-sensitive scenario or a high-speed driving scenario. When the target vehicle frequently accelerates and decelerates and the regenerative braking state changes significantly, the current energy consumption prediction scenario can be identified as a start-stop driving scenario. When the ambient temperature is low, the battery temperature is low, or the air conditioning or cabin heating is in operation, the current energy consumption prediction scenario can be identified as a temperature-sensitive scenario or an additional load-sensitive scenario; when the road gradient is large, the current energy consumption prediction scenario can be identified as a gradient-sensitive scenario.

[0126] After determining the current energy consumption prediction scenario, energy consumption impact data related to vehicle energy consumption can be filtered from vehicle data based on this scenario. The key data involved in energy consumption prediction can differ depending on the current energy consumption prediction scenario. For example, in a high-speed driving scenario, energy consumption impact data may include one or more of the following: vehicle speed, air drag coefficient, wind speed, wind direction, vehicle mass, and road gradient; in a start-stop driving scenario, energy consumption impact data may include one or more of the following: acceleration, driving mode, regenerative braking status, vehicle load, and vehicle speed change information; in a temperature-sensitive scenario, energy consumption impact data may include one or more of the following: ambient temperature, battery temperature, battery state of charge, air conditioning status, and cabin heating status; in a gradient-sensitive scenario, energy consumption impact data may include one or more of the following: road gradient, vehicle mass, vehicle speed, and drive power related data.

[0127] Furthermore, after filtering out energy consumption impact data related to vehicle energy consumption, the energy consumption impact data can be preprocessed.

[0128] It should be noted that the preprocessing process for the energy consumption impact data is the same as the process described above, and will not be repeated here.

[0129] S302, input the energy consumption impact data into multiple pre-trained primary prediction units respectively, and obtain the primary energy consumption prediction results of each primary prediction unit.

[0130] The multiple primary prediction units can be those obtained during the aforementioned training process. For example, the multiple primary prediction units may include a first primary prediction unit, a second primary prediction unit, and a third primary prediction unit. The first primary prediction unit can be a random forest prediction unit, the second primary prediction unit can be a lightweight gradient boosting prediction unit, and the third primary prediction unit can be a support vector regression prediction unit. Each primary prediction unit can process the energy consumption impact data based on different prediction mechanisms, thereby obtaining primary energy consumption prediction results from different perspectives.

[0131] In one possible implementation, energy consumption impact data is input into multiple pre-trained primary prediction units to obtain primary energy consumption prediction results for each primary prediction unit. This includes: determining the scenario parameter configurations corresponding to the multiple primary prediction units in the current energy consumption prediction scenario based on the current energy consumption prediction scenario; and inputting the energy consumption impact data into the multiple primary prediction units configured with scenario parameter configurations to obtain the primary energy consumption prediction results output by each primary prediction unit.

[0132] The scenario parameter configuration can be used to characterize the configuration method adopted by multiple primary prediction units when making predictions under the current energy consumption prediction scenario. The scenario parameter configuration can include one or more of the following: the activation status of the primary prediction unit, the range of input features, model parameters, output data format, prediction window length, and prediction result normalization method. The scenario parameter configurations corresponding to each primary prediction unit can be different under different current energy consumption prediction scenarios.

[0133] For example, in high-speed driving scenarios, multiple primary prediction units can focus on receiving energy consumption data related to driving resistance, such as vehicle speed, wind speed, wind direction, and air drag coefficient; in temperature-sensitive scenarios, multiple primary prediction units can focus on receiving energy consumption data related to temperature and additional loads, such as ambient temperature, battery temperature, air conditioning status, and battery state of charge; in start-stop driving scenarios, multiple primary prediction units can focus on receiving energy consumption data related to vehicle dynamic changes, such as acceleration, vehicle speed changes, regenerative braking status, and driving mode.

[0134] After determining the scenario parameter configuration, multiple primary prediction units can be configured according to these parameters, and energy consumption impact data can be input into each of the configured primary prediction units. The first primary prediction unit can output a first primary energy consumption prediction result based on the energy consumption impact data; the second primary prediction unit can output a second primary energy consumption prediction result based on the energy consumption impact data; and the third primary prediction unit can output a third primary energy consumption prediction result based on the energy consumption impact data. Each primary energy consumption prediction result can be used to characterize the energy consumption prediction value of the target vehicle for the corresponding primary prediction unit under the current energy consumption prediction scenario.

[0135] For example, the first primary prediction unit can predict energy consumption impact data based on a random forest structure. The random forest prediction unit can perform regression prediction on the energy consumption impact data using multiple decision trees and determine the first primary energy consumption prediction result based on the outputs of the multiple decision trees. The second primary prediction unit can predict energy consumption impact data based on a gradient boosting structure and output the second primary energy consumption prediction result by progressively fitting the residuals. The third primary prediction unit can predict energy consumption impact data based on a support vector regression structure and output the third primary energy consumption prediction result through kernel function mapping and regression solving.

[0136] S303 inputs multiple primary energy consumption prediction results into a pre-trained fusion prediction unit to obtain the energy consumption prediction result of the target vehicle in the current energy consumption prediction scenario.

[0137] The fusion prediction unit can be the second-layer prediction structure obtained during the aforementioned training process. The fusion prediction unit can be used to fuse the primary energy consumption prediction results output by multiple primary prediction units to obtain the final energy consumption prediction result. For example, the fusion prediction unit can be a linear regression prediction unit.

[0138] Specifically, the first, second, and third primary energy consumption prediction results are combined into fused input data. This fused input data is then fed into a pre-trained fusion prediction unit. Based on the correspondence between the multiple primary prediction results learned during training and the actual energy consumption results, the fusion prediction unit performs regression processing on the fused input data to obtain the energy consumption prediction result of the target vehicle in the current energy consumption prediction scenario.

[0139] One possible implementation is that the fusion prediction unit can weightedly fuse multiple primary energy consumption prediction results. The primary energy consumption prediction results from different primary prediction units can have different levels of contribution to the fusion prediction unit. The fusion prediction unit can determine the influence relationship between multiple primary energy consumption prediction results and the final energy consumption prediction result based on the regression parameters formed during training, and output the final energy consumption prediction result. This energy consumption prediction result can be the energy consumption per unit mileage of the target vehicle in the current energy consumption prediction scenario, the predicted energy consumption within a preset driving distance, the predicted energy consumption within a preset time window, or the predicted energy consumption corresponding to the target driving segment.

[0140] In this process, the energy consumption prediction result of the target vehicle is not directly determined by a single prediction unit, but is obtained by multiple primary prediction units making predictions separately, and then merging them through a fusion prediction unit. This can combine the representation capabilities of multiple primary prediction units on the relationship of vehicle energy consumption changes, reduce the prediction bias caused by model structure limitations, sample fluctuations or scenario differences of a single prediction unit, and improve the accuracy and stability of vehicle energy consumption prediction results.

[0141] Furthermore, after obtaining the energy consumption prediction results for the target vehicle under the current energy consumption prediction scenario, corresponding energy consumption prompts or energy consumption analysis information can be generated based on the energy consumption prediction results. For example, the energy consumption prediction results can be combined with the target vehicle's current battery state of charge, remaining driving range, or target driving route to obtain the vehicle's range estimate; alternatively, based on the current energy consumption prediction scenario and energy consumption impact data, the main influencing factors of the current vehicle's high energy consumption can be identified, and corresponding energy consumption analysis results can be generated.

[0142] Furthermore, after obtaining multiple primary prediction units and a fusion prediction unit after training, the model is evaluated based on the test set to obtain the model evaluation results.

[0143] Specifically, historical energy consumption data from the test set are input into multiple trained primary prediction units to obtain multiple test prediction results. These multiple test prediction results are then input into a trained fusion prediction unit to obtain the test energy consumption prediction result. Based on the difference between the test energy consumption prediction result and the corresponding historical energy consumption result in the test set, the model evaluation result of the energy consumption prediction model is determined.

[0144] The model evaluation results may include one or more of the following: root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination. RMSE characterizes the overall deviation between the predicted energy consumption and historical energy consumption results. MAE characterizes the average deviation between the predicted energy consumption and historical energy consumption results. The coefficient of determination... () is used to characterize the degree of fit of the energy consumption prediction model to changes in vehicle energy consumption.

[0145] One possible implementation involves generating a comparison between predicted and actual values ​​based on the test energy consumption prediction results and the corresponding historical energy consumption results for the test set. Furthermore, it generates an error distribution based on the error between the test energy consumption prediction results and the historical energy consumption results. The comparison between predicted and actual values ​​provides a clear indication of the energy consumption prediction model's ability to predict trends in vehicle energy consumption. The error distribution reflects the model's predictive stability across different samples or energy consumption prediction scenarios.

[0146] Furthermore, the feature importance information corresponding to each energy consumption influencing factor in the historical energy consumption impact data can be determined based on the first primary prediction unit after training. Specifically, when the first primary prediction unit is a random forest prediction unit, the feature importance information corresponding to each energy consumption influencing factor can be determined based on the contribution of each energy consumption influencing factor in the node splitting process in the random forest prediction unit, and a feature importance ranking result can be generated based on the feature importance information.

[0147] Furthermore, based on battery state of charge (SOC) information from test sets or historical samples and corresponding test energy consumption prediction results, the correlation between battery SOC and vehicle energy consumption can be determined, and the relationship analysis results between battery SOC and vehicle energy consumption can be generated. These relationship analysis results can be used to characterize the changes in vehicle energy consumption per unit mileage under different battery SOC states.

[0148] Furthermore, after obtaining the energy consumption prediction results of the target vehicle under the current energy consumption prediction scenario, the energy consumption analysis results of the target vehicle are determined based on the energy consumption impact data and the energy consumption prediction results.

[0149] The energy consumption analysis results can be used to characterize the energy consumption level of the target vehicle under the current energy consumption prediction scenario, as well as the correlation between various energy consumption influencing factors and vehicle energy consumption in the energy consumption impact data. The energy consumption analysis results may include one or more of the following: energy consumption prediction value, feature importance information, target influencing factor information, relationship information between battery state of charge and energy consumption, and model evaluation information.

[0150] In one possible implementation, the energy consumption analysis results of the target vehicle are determined based on energy consumption impact data and energy consumption prediction results, including: determining the energy consumption per unit mileage of the target vehicle under the current energy consumption prediction scenario based on the energy consumption prediction results; determining the feature importance information corresponding to each energy consumption impact factor based on the contribution degree of each energy consumption impact factor in the energy consumption prediction model based on the energy consumption impact data; and generating the energy consumption analysis results of the target vehicle based on the energy consumption per unit mileage and the feature importance information.

[0151] Specifically, the energy consumption prediction result can be the energy consumption per unit mileage of the target vehicle under the current energy consumption prediction scenario, such as Wh / km or kWh / 100km. The vehicle energy consumption prediction system can determine the energy consumption level of the target vehicle during the current driving process based on this energy consumption prediction result. Based on the model output information of multiple primary prediction units or fusion prediction units in the prediction process, the feature importance information corresponding to each energy consumption influencing factor in the energy consumption impact data is determined.

[0152] For example, when the first primary prediction unit is a random forest prediction unit, the feature importance information corresponding to energy consumption influencing factors such as vehicle speed, acceleration, road slope, ambient temperature, battery state of charge, and air conditioning status can be determined based on the contribution of each feature in the random forest prediction unit to the node splitting process.

[0153] After determining the importance of features, target influencing factors are identified from the energy consumption impact data based on this information. Target influencing factors can be one or more energy consumption factors that have a significant impact on the current energy consumption prediction results. For example, in a high-speed driving scenario, target influencing factors could include vehicle speed, wind speed, wind direction, and air resistance coefficient; in a temperature-sensitive scenario, target influencing factors could include ambient temperature, battery temperature, battery state of charge, and air conditioning status.

[0154] Another possible approach is to determine the correspondence between battery state of charge and vehicle energy consumption based on the battery state of charge information in the energy consumption impact data and the energy consumption prediction results, and then generate energy consumption analysis results based on this correspondence.

[0155] Specifically, the battery state of charge (SBC) information under the current energy consumption prediction scenario can be correlated with the corresponding energy consumption prediction results to obtain the relationship between battery SBC and energy consumption per unit mileage. This relationship information can be used to characterize the changes in vehicle energy consumption under different battery SBC ranges.

[0156] Furthermore, based on energy consumption impact data and energy consumption prediction results corresponding to multiple historical driving segments or multiple current sampling windows, it is possible to generate relationship curves between battery state of charge and energy consumption, feature importance ranking results, or energy consumption comparison results under different energy consumption prediction scenarios. Through the above analysis results, the correlation between vehicle operation data, driving environment data, battery state data, energy consumption component state data, and vehicle energy consumption can be reflected more intuitively.

[0157] For example, the energy consumption analysis results may include: the energy consumption per unit mile of the target vehicle under the current energy consumption prediction scenario; the top energy consumption influencing factors with a high degree of impact on energy consumption per unit mile; the ranking of the feature importance of each energy consumption influencing factor; and the relationship information between battery state of charge and energy consumption per unit mile. The vehicle energy consumption prediction system can send the energy consumption analysis results to a display device for display, or send the energy consumption analysis results to the vehicle controller, energy management system, or cloud server for use in vehicle range estimation, energy consumption statistics, or driver assistance analysis.

[0158] The above primarily describes the solutions provided in the embodiments of this application from the perspective of the working principle of the device. It is understood that, in order to achieve the above functions, the vehicle energy consumption prediction device includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the algorithm steps of the examples described in the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0159] This application embodiment can divide the vehicle energy consumption prediction device into functional modules according to the above method example. For example, each function can be divided into a separate functional module, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module.

[0160] It should be noted that the module division in this embodiment is illustrative and represents only one logical functional division; in actual implementation, other division methods may be used. When dividing functional modules according to their respective functions, Figure 4 A schematic diagram of a possible composition of the vehicle energy consumption prediction device involved in the above and embodiment is shown. Figure 4 As shown, the vehicle energy consumption prediction device 400 may include a determination module 401 and an input module 402.

[0161] The determining module 401 is used to support the execution of the vehicle energy consumption prediction device 400. Figure 3 S301 in the illustrated vehicle energy consumption prediction method.

[0162] Input module 402 is used to support the execution of vehicle energy consumption prediction device 400. Figure 2 S302 and S303 in the illustrated vehicle energy consumption prediction method.

[0163] One possible implementation involves jointly training multiple primary prediction units and a fusion prediction unit. This joint training is achieved through the following steps: acquiring historical energy consumption impact data related to vehicle energy consumption during vehicle operation under historical energy consumption prediction scenarios, and the corresponding historical energy consumption results; constructing a training dataset based on the historical energy consumption impact data and historical energy consumption results; and training multiple primary prediction units and a fusion prediction unit based on the training dataset, resulting in trained primary prediction units and a trained fusion prediction unit.

[0164] In one possible implementation, before training multiple primary prediction units and the fusion prediction unit based on the training dataset, the device is further configured to: determine the target parameter configuration of the first primary prediction unit based on the training dataset and a preset parameter search strategy; and construct the first primary prediction unit based on the target parameter configuration.

[0165] One possible implementation involves a determination module, used to determine the target parameter configuration of the first primary prediction unit based on the training dataset and a preset parameter search strategy. Specifically, this module: determines multiple model parameters to be optimized for the first primary prediction unit, and the corresponding parameter value ranges for each parameter. Based on the values ​​of the multiple model parameters within their respective value ranges, it generates multiple parameter configurations for the first primary prediction unit. Based on each parameter configuration, it constructs candidate first primary prediction units. It trains and validates each candidate first primary prediction unit using the training dataset, obtaining the prediction error of each candidate first primary prediction unit. Based on the prediction errors of each candidate first primary prediction unit, it determines the target parameter configuration from the parameter configurations and identifies the candidate first primary prediction unit corresponding to the target parameter configuration as the first primary prediction unit.

[0166] One possible implementation is that a set of parameter configurations corresponds to one search individual. The determination module, used to determine the target parameter configuration from the parameter configurations based on the prediction errors of each candidate first primary prediction unit, specifically performs the following steps: determining the fitness of each search individual based on the prediction errors of the candidate first primary prediction units corresponding to each search individual; determining the individual type of each search individual based on its fitness; iteratively updating the parameter configurations corresponding to each search individual based on its individual type; and determining the target parameter configuration based on the fitness of each search individual, provided that the iterative update meets a preset iteration stopping condition.

[0167] One possible implementation involves a module that, when iteratively updating the parameter configurations for each search individual based on its individual type, specifically: updating the parameter configurations for the first search individual (individual type 1) to expand the search range; updating the parameter configurations for the second search individual (individual type 2) based on the current optimal search individual; and, under preset warning conditions, perturbing the parameter configurations for the third search individual (individual type 3) with a perturbation update. Finally, the fitness of each search individual is re-determined based on the updated parameter configurations.

[0168] One possible implementation involves a determination module that, based on vehicle data during the current driving process of the target vehicle, determines energy consumption impact data related to vehicle energy consumption during the current energy consumption prediction scenario. Specifically, this module is used to: determine the current energy consumption prediction scenario for the target vehicle based on vehicle data indicating vehicle driving-related data during the current driving process; and, based on the current energy consumption prediction scenario, filter out energy consumption impact data related to vehicle energy consumption from the vehicle data.

[0169] One possible implementation involves a determination module that, based on vehicle data indicating vehicle driving-related data during the current driving process, determines the current energy consumption prediction scenario for the target vehicle. Specifically, this module: determines the target vehicle's driving condition characteristics based on vehicle operation data within the vehicle data; determines the target vehicle's environmental condition characteristics based on driving environment data within the vehicle data; determines the target vehicle's battery operating characteristics based on battery status data within the vehicle data; determines the target vehicle's additional load characteristics based on energy-consuming component status data within the vehicle data; and finally, based on the driving condition characteristics, environmental condition characteristics, battery operating characteristics, and additional load characteristics, determines the target vehicle's current energy consumption prediction scenario.

[0170] One possible implementation involves an input module that, when inputting energy consumption impact data into multiple pre-trained primary prediction units to obtain the primary energy consumption prediction results of each unit, specifically: determining the scenario parameter configurations corresponding to the multiple primary prediction units under the current energy consumption prediction scenario; inputting the energy consumption impact data into the multiple primary prediction units configured with scenario parameters to obtain the primary energy consumption prediction results output by each primary prediction unit.

[0171] It should be noted that all relevant content of each step involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0172] The vehicle energy consumption prediction device 400 provided in this application embodiment is used to perform the above-mentioned... Figure 2 or Figure 3 The vehicle energy consumption prediction method shown can therefore achieve the same effect as the vehicle energy consumption prediction method described above.

[0173] This application also provides a vehicle energy consumption prediction device, which can execute the vehicle energy consumption prediction method and related steps in the above method embodiments.

[0174] This application also provides a computer-readable storage medium storing instructions thereon, which, when executed, perform the vehicle energy consumption prediction method and related steps in the above method embodiments.

[0175] This application also provides a computer program product that, when run on a computer, causes the computer to execute the vehicle energy consumption prediction method and related steps described in the above method embodiments.

[0176] In some embodiments, the methods shown in this application can be implemented as computer program instructions encoded in a machine-readable format on a computer-readable storage medium or on other non-transitory media or articles of art.

[0177] This application also provides a vehicle energy consumption prediction system 100, such as... Figure 5 As shown, the vehicle energy consumption prediction system 100 includes at least one processor 501 and at least one interface circuit 502.

[0178] As an example, when the vehicle energy consumption prediction system 100 includes a processor and an interface circuit, the processor can be... Figure 5 The processor 501 shown in the solid box (or the processor 501 shown in the dashed box) can be an interface circuit. Figure 5 The interface circuit 502 is shown in the solid box (or the dashed box). When the vehicle energy consumption prediction system 100 includes two processors and two interface circuits, the two processors include... Figure 5 The processor 501 shown in the solid box and the processor 501 shown in the dashed box, these two interface circuits include Figure 5 Interface circuit 502 is shown in both solid and dashed boxes. No limitations are imposed on this.

[0179] The processor 501 and the interface circuit 502 can be interconnected via a line. For example, the interface circuit 502 can be used to receive signals. Alternatively, the interface circuit 502 can be used to send signals to other devices (such as the processor 501). For instance, the interface circuit 502 can read computer instructions stored in memory and send those instructions to the processor 501. The processor 501 executes the instructions and, in conjunction with input / output devices, implements the various steps in the above embodiments, such as implementing... Figure 2-3 The methods illustrated are the steps performed in the embodiments shown. Of course, this vehicle energy consumption prediction system may also include other discrete components, and this application embodiment does not specifically limit this.

[0180] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0181] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0182] The units described as separate components may or may not be physically separate. A component shown as a unit can be one or more physical units; that is, it can be located in one place or distributed in multiple different locations. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0183] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0184] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, in essence, or the part that contributes, or all or part of the technical solution, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0185] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A vehicle energy consumption prediction method characterized by, The method includes: Based on the vehicle data of the target vehicle during its current driving process, determine the energy consumption impact data related to vehicle energy consumption during the driving process of the target vehicle under the current energy consumption prediction scenario; The energy consumption impact data is input into multiple pre-trained primary prediction units to obtain the primary energy consumption prediction results of each primary prediction unit. The multiple primary energy consumption prediction results are input into a pre-trained fusion prediction unit to obtain the energy consumption prediction result of the target vehicle in the current energy consumption prediction scenario.

2. The method of claim 1, wherein, The multiple primary prediction units and the fusion prediction unit are jointly trained, and are obtained through the following steps: Acquire historical energy consumption impact data related to vehicle energy consumption during vehicle operation under historical energy consumption prediction scenarios, as well as historical energy consumption results corresponding to the historical energy consumption impact data; A training dataset is constructed based on the historical energy consumption impact data and the historical energy consumption results. Multiple primary prediction units and the fusion prediction unit are trained based on the training dataset to obtain multiple trained primary prediction units and the trained fusion prediction unit.

3. The method of claim 2, wherein, Before training the multiple primary prediction units and the fusion prediction unit based on the training dataset, the method further includes: Based on the training dataset and the preset parameter search strategy, the target parameter configuration of the first primary prediction unit is determined; Based on the target parameter configuration, the first primary prediction unit is constructed.

4. The method of claim 3, wherein, The step of determining the target parameter configuration of the first primary prediction unit based on the training dataset and a preset parameter search strategy includes: Determine multiple model parameters to be optimized for the first primary prediction unit, and the range of parameter values ​​corresponding to each of the model parameters to be optimized; Based on the values ​​of multiple model parameters to be optimized within their respective parameter ranges, multiple parameter configurations for the first primary prediction unit are generated. Based on the parameter configurations described above, candidate first primary prediction units are constructed respectively; The candidate first primary prediction units are trained and validated based on the training dataset to obtain the prediction error of each candidate first primary prediction unit. Based on the prediction errors of each candidate first primary prediction unit, a target parameter configuration is determined from the parameter configuration, and the candidate first primary prediction unit corresponding to the target parameter configuration is determined as the first primary prediction unit.

5. The method of claim 4, wherein, A set of parameter configurations corresponds to one search entity; determining the target parameter configuration from the parameter configurations based on the prediction errors of each candidate first primary prediction unit includes: The fitness of each search individual is determined based on the prediction error of the candidate first primary prediction unit corresponding to each search individual. Based on the fitness of each search individual, determine the individual type of each search individual; Based on the individual type of each search individual, the parameter configuration corresponding to each search individual is iteratively updated; If the iterative update satisfies the preset iteration stop condition, the target parameter configuration is determined based on the fitness of each search individual.

6. The method of claim 5, wherein, The iterative update of parameter configurations for each search individual based on its individual type includes: Update the parameter configuration for the first search individual of type 1 to expand the parameter search range; Based on the current best search individual, the parameter configuration corresponding to the second search individual of the second type is updated accordingly; Under the condition that the preset warning conditions are met, the parameter configuration corresponding to the third search individual of the third type is perturbed and updated; The fitness of each search individual was reassessed based on the updated parameter configuration.

7. The method of claim 1, wherein, The step of determining energy consumption impact data related to vehicle energy consumption during the current energy consumption prediction scenario, based on vehicle data of the target vehicle during its current driving process, includes: Based on the vehicle data indicating vehicle driving-related data in the current driving process of the target vehicle, the current energy consumption prediction scenario of the target vehicle is determined; Based on the current energy consumption prediction scenario, energy consumption impact data related to vehicle energy consumption is filtered out from the vehicle data.

8. The method of claim 7, wherein, The step of determining the current energy consumption prediction scenario for the target vehicle based on vehicle data indicating vehicle driving-related data during the current driving process includes: Based on the vehicle operation data in the vehicle data, determine the driving condition characteristics of the target vehicle; Based on the driving environment data in the vehicle data, the environmental operating condition characteristics of the target vehicle are determined; Based on the battery status data in the vehicle data, determine the battery operating characteristics of the target vehicle; Based on the energy-consuming component status data in the vehicle data, the additional load characteristics of the target vehicle are determined; Based on the driving condition characteristics, environmental condition characteristics, battery operating characteristics, and additional load characteristics, the current energy consumption prediction scenario for the target vehicle is determined.

9. The method of claim 1, wherein, The step of inputting the energy consumption impact data into multiple pre-trained primary prediction units to obtain the primary energy consumption prediction results of each primary prediction unit includes: Based on the current energy consumption prediction scenario, determine the scenario parameter configurations corresponding to the multiple primary prediction units under the current energy consumption prediction scenario; The energy consumption impact data is input into multiple primary prediction units configured with the scenario parameters, respectively, to obtain the primary energy consumption prediction results output by each primary prediction unit.

10. A vehicle energy consumption prediction device characterized by comprising: The device includes: The determination module is used to determine the energy consumption impact data related to vehicle energy consumption during the current energy consumption prediction scenario based on the vehicle data of the target vehicle during its current driving process. The input module is used to input the energy consumption impact data into a plurality of pre-trained primary prediction units to obtain the primary energy consumption prediction results of each primary prediction unit. The input module is also used to input multiple primary energy consumption prediction results into a pre-trained fusion prediction unit to obtain the energy consumption prediction result of the target vehicle in the current energy consumption prediction scenario.

Images