Method and device for determining influence factors of vehicle endurance performance and electronic equipment
By constructing standards and feature vectors to be analyzed, and using a neural network model to analyze vehicle range performance, the systematic and consistent problems of factor identification in vehicle range prediction are solved, and accurate diagnosis and cause analysis of vehicle range performance are achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-03
- Publication Date
- 2026-04-14
AI Technical Summary
Existing technologies struggle to accurately identify key factors affecting driving range under specific vehicle operating conditions, resulting in a lack of systematicity and consistency in driving range prediction and cause analysis.
By obtaining the actual range achievement rate of sample vehicles, the vehicles are divided into vehicles to be analyzed and standard vehicles. Feature vectors of standard and vehicles to be analyzed are constructed, and the constructed feature vectors are analyzed using a pre-trained neural network model to determine the main influencing factors.
It enables accurate diagnosis and attribution analysis of vehicle range performance, and can automatically and quantitatively identify key variables that lead to low range, shortening the problem location time and improving operation and maintenance and R&D efficiency.
Smart Images
Figure CN121858882A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of automotive technology, specifically to a method, apparatus, and electronic device for determining factors affecting vehicle range performance. Background Technology
[0002] With the rapid development of the new energy vehicle industry, vehicle range and its stability have become crucial indicators for evaluating overall vehicle performance and user experience. In real-world use, a vehicle's range performance is influenced not only by the capacity and degradation of the battery, but also by various factors such as the vehicle's operating environment, driving habits, temperature variations, and operating conditions. To assess a vehicle's actual range performance, the "range achievement rate" is typically used as a metric, reflecting the degree to which the vehicle achieves its theoretical range in real-world scenarios. Analyzing the specific factors affecting the vehicle's range achievement rate is essential for improving vehicle performance.
[0003] In related technologies, the judgment is mostly based on the feature importance of the prediction model. This can only reflect the average influence of each influencing factor in the overall sample, but cannot reveal which influencing factor's sudden change or abnormal fluctuation caused the vehicle's range to decrease under specific operating conditions. Summary of the Invention
[0004] This invention provides a method, apparatus, and electronic device for determining the factors influencing vehicle range performance, aiming to solve the technical problem of difficulty in accurately extracting the important factors affecting vehicle range.
[0005] Firstly, a method for determining the factors influencing vehicle range performance is provided, including the following steps: Obtain the actual range achievement rate of the sample vehicles, and divide the sample vehicles into vehicles to be analyzed and standard vehicles based on the actual range achievement rate; Obtain initial data for the sample vehicles, wherein the initial data is data that affects the actual range achievement rate; Determine the standard feature vector of the standard vehicle; wherein the standard feature vector includes multiple standard data; the standard data are determined based on the initial data of the standard vehicle; Determine the feature vector to be analyzed for the vehicle to be analyzed; wherein the feature vector to be analyzed includes multiple data to be analyzed; the data to be analyzed is the initial data of the vehicle to be analyzed; The standard feature vector and the feature vector to be analyzed are combined to obtain multiple constructed feature vectors; Multiple constructed feature vectors are sequentially input into a pre-trained target neural network model to obtain the predicted range achievement rate corresponding to each constructed feature vector; Based on the predicted range achievement rate and the actual range achievement rate, the main influencing factors of range performance are determined.
[0006] In some embodiments, the sample vehicles are divided into vehicles to be analyzed and standard vehicles based on the actual range achievement rate, including: From the sample vehicles, vehicles whose actual range achievement rate is less than or equal to the first threshold are selected to obtain the vehicles to be analyzed. From the sample vehicles, vehicles whose actual range achievement rate is greater than the first threshold are selected to obtain standard vehicles.
[0007] In some embodiments, the first threshold is determined as follows: Calculate the average of the actual range achievement rates of the sample vehicles, and use it as the first average. The first threshold is obtained by multiplying the first average value by a preset proportional coefficient.
[0008] In some embodiments, determining the standard feature vector of the standard vehicle includes: Determine the initial data for the standard vehicle; Calculate the average value of the initial data for each category to obtain multiple second average values; Multiple second average values are combined to form a standard feature vector.
[0009] In some embodiments, the standard feature vector and the feature vector to be analyzed are combined to obtain multiple constructed feature vectors, including: The data to be analyzed in the feature vector to be analyzed is replaced sequentially with standard data of the same type in the standard feature vector to obtain multiple constructed feature vectors.
[0010] In some embodiments, the main influencing factors of range performance are determined based on the predicted range achievement rate and the actual range achievement rate, including: Calculate the difference between each predicted range achievement rate and the actual range achievement rate corresponding to the vehicle to be analyzed, and obtain multiple differences; The target difference is obtained by determining the maximum value among multiple differences. Determine the constructed feature vector corresponding to the target difference as the key feature vector; The standard data included in the key feature vector is determined, and the data type of the standard data is determined as the main influencing factor of battery life performance.
[0011] In some embodiments, the target neural network model is pre-trained in the following manner: Obtain the initial data of the sample vehicles; The initial data is segmented and outliers are removed to obtain the first feature vector; The initial neural network model is trained using the first feature vector to obtain the target neural network model.
[0012] Secondly, a device for determining factors affecting vehicle range performance is also provided, comprising: The first acquisition module is used to acquire the actual range achievement rate of the sample vehicles, and to divide the sample vehicles into vehicles to be analyzed and standard vehicles based on the actual range achievement rate. The second acquisition module is used to acquire the initial data of the sample vehicle, wherein the initial data is data that affects the actual range achievement rate; The first determining module is used to determine the standard feature vector of the standard vehicle; wherein the standard feature vector includes multiple standard data; the standard data is determined based on the initial data of the standard vehicle; The second determining module is used to determine the feature vector to be analyzed of the vehicle to be analyzed; wherein the feature vector to be analyzed includes multiple data to be analyzed; the data to be analyzed is the initial data of the vehicle to be analyzed; The combination module is used to combine the standard feature vector and the feature vector to be analyzed to obtain multiple constructed feature vectors; The prediction module is used to sequentially input multiple constructed feature vectors into a pre-trained target neural network model to obtain the predicted range achievement rate corresponding to each constructed feature vector; The analysis module is used to determine the main influencing factors of range performance based on the predicted range achievement rate and the actual range achievement rate.
[0013] Thirdly, an electronic device is provided, the electronic device including a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the steps of the method described in any of the above embodiments.
[0014] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, the computer program being loaded by a processor to perform the steps of the method described in any of the preceding embodiments.
[0015] The beneficial effects of this application include: a method for determining the influencing factors of vehicle range performance, which combines vehicle operating data (initial data) under different scenarios, and by comparing the difference between the predicted range achievement rate and the actual range achievement rate, can accurately select the main influencing factors of range performance, thereby achieving accurate diagnosis and attribution analysis of abnormal vehicle range. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a flowchart illustrating a method for determining factors influencing vehicle range performance according to an exemplary embodiment of this disclosure. Figure 2 This is a schematic diagram of a device for determining factors affecting vehicle range performance provided in an exemplary embodiment of this disclosure. Detailed Implementation
[0018] 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. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0019] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.
[0020] "A and / or B" includes the following three combinations: A only, B only, and a combination of A and B.
[0021] In this application, the term "exemplary" is used to mean "used as an example, illustration, or description." Any embodiment described as "exemplary" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use this application. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that this application can be made without using these specific details. In other instances, well-known structures and processes are not described in detail to avoid obscuring the description of this application with unnecessary detail. Therefore, this application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.
[0022] In related technologies, "range achievement rate" is commonly used as a metric to measure a vehicle's range performance (range capability), reflecting the degree to which a vehicle achieves its theoretical range in real-world usage scenarios. Range achievement rate = change in mileage / change in remaining mileage. For example, for vehicle A, before the start of this trip, vehicle A's predicted remaining mileage was 800km. After the trip, vehicle A actually traveled 30km, so the change in mileage is 30km, and the predicted remaining mileage is updated to 750km. Therefore, the change in remaining mileage = 800km - 750km = 50km, and vehicle A's range achievement rate = 30km / 50km = 60%. For example, regarding vehicle B, before the start of this trip, vehicle B's predicted remaining range was 800km. After the trip, vehicle B actually drove 30km, so the change in mileage is 30km, and the predicted remaining range is updated to 770km. Therefore, the change in remaining range = 800km - 770km = 30km, and vehicle B's range achievement rate = 30km / 30km = 100%. Vehicle B's range achievement rate is greater than vehicle A's. Therefore, vehicle B's range performance is better than vehicle A's because vehicle B can more accurately estimate its "predicted remaining range" using its own operating data. After driving 30km, its "predicted remaining range" decreased by 30km. Vehicle A, on the other hand, cannot accurately estimate its own "predicted remaining range," and after driving 30km, its "predicted remaining range" decreased by 50km.
[0023] In related technologies, range achievement calculations are mostly based on estimations of single-trip vehicle data, failing to comprehensively consider energy consumption differences across different road types, speed ranges, and ambient temperatures, thus lacking global representativeness. Furthermore, low range cause analysis often relies on the importance of model features, reflecting only the average impact of each influencing factor in the overall sample, but failing to reveal which factor's sudden change or abnormal fluctuation caused the range reduction under specific operating conditions. Finally, these technologies typically lack multi-condition fusion models, making it impossible to uniformly measure energy consumption performance across different driving scenarios or quantitatively compare the impact of each condition on overall range performance. This fragmented analytical approach results in a lack of systematicity and consistency in range prediction and cause identification.
[0024] On the one hand, this embodiment provides a method for determining the factors affecting vehicle range performance, such as... Figure 1 As shown, it includes the following steps: S101: Obtain the actual range achievement rate of the sample vehicles, and classify the sample vehicles into vehicles to be analyzed and standard vehicles based on the actual range achievement rate; S102: Obtain the initial data of the sample vehicle, wherein the initial data is data that affects the actual range achievement rate; S103: Determine the standard feature vector of the standard vehicle; wherein the standard feature vector includes multiple standard data; the standard data is determined based on the initial data of the standard vehicle; S104: Determine the feature vector to be analyzed for the vehicle to be analyzed; wherein, the feature vector to be analyzed includes multiple data to be analyzed; the data to be analyzed is the initial data of the vehicle to be analyzed; S105: Combine the standard feature vector and the feature vector to be analyzed to obtain multiple constructed feature vectors; S106: Input the multiple constructed feature vectors sequentially into the pre-trained target neural network model to obtain the predicted range achievement rate corresponding to each constructed feature vector; S107: Based on the predicted range achievement rate and the actual range achievement rate, determine the main influencing factors of range performance.
[0025] Based on the above embodiments, by combining the vehicle's operating data (initial data) under different scenarios, and by comparing the difference between the predicted range achievement rate and the actual range achievement rate, the main influencing factors of range performance can be effectively selected, thereby achieving accurate diagnosis and attribution analysis of abnormal vehicle range.
[0026] The following describes a specific implementation method for determining the factors affecting vehicle range performance.
[0027] S101: Obtain the actual range achievement rate of the sample vehicles, and divide the sample vehicles into vehicles to be analyzed and standard vehicles based on the actual range achievement rate.
[0028] In some embodiments, the sample vehicle refers to a vehicle already in use. Sample vehicles are typically pure electric vehicles. The actual range achievement rate of the sample vehicle = actual change in mileage / actual change in remaining mileage.
[0029] In some embodiments, the sample vehicles are divided into vehicles to be analyzed and standard vehicles based on the actual range achievement rate, specifically including: S1: From the sample vehicles, select the vehicles whose actual range achievement rate is less than or equal to the first threshold to obtain the vehicles to be analyzed; S2: From the sample vehicles, select the vehicles whose actual range achievement rate is greater than the first threshold to obtain standard vehicles.
[0030] In some embodiments, the vehicles to be analyzed are those with relatively low actual range achievement rates. These vehicles may have performance issues or be operating under abnormal conditions, thus requiring further analysis of the main factors influencing their range performance. Standard vehicles are those with relatively high actual range achievement rates; their actual range achievement rates meet the requirements and are not abnormal. Therefore, the data corresponding to the standard vehicles will be used as the "evaluation benchmark" to analyze the vehicles to be analyzed. The sample vehicles include multiple vehicles, the vehicles to be analyzed include multiple vehicles, and the standard vehicles include multiple vehicles.
[0031] In some embodiments, the first threshold is determined as follows: S1: Calculate the average of the actual range achievement rate of the sample vehicles, and use it as the first average; S2: Multiply the first average value by the preset proportional coefficient to obtain the first threshold.
[0032] In some embodiments, the preset proportional coefficient can be set to 50%, 60%, 70%, 80%, etc. The first threshold = first average value × preset proportional coefficient. In this way, vehicles whose actual range achievement rate is significantly lower than the first average value can be selected as vehicles to be analyzed.
[0033] The first threshold is determined by data from the vehicle group, avoiding misjudgments caused by the randomness of individual vehicles and enhancing the credibility of classification conclusions. It also reduces the uncertainty of manually setting thresholds and improves consistency across regions and time periods.
[0034] In some embodiments, a horizontal comparison is made between sample vehicles with the same configuration (such as vehicle model, battery type, energy density, drive system, etc.) in the market, and a first threshold is set for a group of sample vehicles with the same configuration.
[0035] S102: Obtain the initial data of the sample vehicle, wherein the initial data is data that affects the actual range achievement rate.
[0036] In some embodiments, the sample vehicles will upload the initial data generated during their operation to the power battery big data platform via technologies such as vehicle-to-everything (V2X) communication. The power battery big data platform is a data-sharing cloud platform used for uploading and storing initial data; users can also download and obtain initial data from the power battery big data platform.
[0037] In some embodiments, initial data refers to operating condition data that affects the range achievement rate. The types of initial data include: energy consumption, charging time, vehicle speed, battery pack cell voltage difference, SOC (state of charge), SOH (state of health), etc. Accordingly, initial data specifically includes data values corresponding to each data type, such as: energy consumption data, charging time data, vehicle speed data, battery pack cell voltage difference data, SOC data, and SOH data.
[0038] Each sample vehicle corresponds to a set of initial data, which is usually expressed as a vector. The elements in the vector are the data values of the initial data.
[0039] S103: Determine the standard feature vector of the standard vehicle; wherein the standard feature vector includes multiple standard data; the standard data are determined based on the initial data of the standard vehicle.
[0040] In some embodiments, determining the standard feature vector of the standard vehicle specifically includes: S1: Determine the initial data of the standard vehicle; S2: Calculate the average value of the initial data for each category to obtain multiple second average values; S3: Combine multiple second average values to form a standard feature vector.
[0041] In some embodiments, taking the initial data including three types of data—cell voltage difference, SOC, and SOH—as an example, the formation of the standard feature vector is explained. The average of the cell voltage difference data from multiple standard vehicles is calculated to obtain the first second average value, denoted as x1. The average of the SOC data from multiple standard vehicles is calculated to obtain the second second average value, denoted as x2. The average of the SOH data from multiple standard vehicles is calculated to obtain the third second average value, denoted as x3. The standard feature vector is represented as [x1, x2, x3]. It should be noted that, depending on the driving scenario of the sample vehicles, more types of initial data may be included. The method for obtaining the standard feature vector when including more types of initial data can be found in the aforementioned method and will not be elaborated here. Each element in the standard feature vector corresponds to a second average value.
[0042] S104: Determine the feature vector to be analyzed for the vehicle to be analyzed; wherein the feature vector to be analyzed includes multiple data to be analyzed; the data to be analyzed is the initial data of the vehicle to be analyzed.
[0043] In some embodiments, the initial data includes three types of data: battery pack cell voltage difference, SOC, and SOH. The battery pack cell voltage difference data is represented as z1, the SOC data as z2, and the SOH data as z3. The feature vector to be analyzed is then represented as [z1, z2, z3]. It should be noted that, depending on the driving scenario of the sample vehicle, more types of initial data may be included. Methods for obtaining the feature vector to be analyzed when including more types of initial data can be found in the aforementioned methods and will not be elaborated here. Each element in the feature vector to be analyzed corresponds to one piece of data to be analyzed.
[0044] The vehicles to be analyzed are usually multiple vehicles, and each vehicle corresponds to a feature vector [z1, z2, z3]. This application selects one of the multiple vehicles to be analyzed, and steps S105 to S107 will be described in detail later.
[0045] S105: Combine the standard feature vector and the feature vector to be analyzed to obtain multiple constructed feature vectors.
[0046] In some embodiments, the standard feature vector and the feature vector to be analyzed are combined to obtain multiple constructed feature vectors. Specifically, this includes: replacing the data to be analyzed in the feature vector to be analyzed with standard data of the same type in the standard feature vector to obtain multiple constructed feature vectors.
[0047] In some embodiments, taking the initial data including three types of data—cell voltage difference, SOC, and SOH—as an example, the first element in [z1, z2, z3] (representing the cell voltage difference data) is replaced with the first element in [x1, x2, x3] (representing the cell voltage difference data), resulting in the first constructed feature vector [x1, z2, z3]. The second element in [z1, z2, z3] (representing the SOC data) is replaced with the second element in [x1, x2, x3] (representing the SOC data), resulting in the second constructed feature vector [z1, x2, z3]. The third element in [z1, z2, z3] (representing the SOH data) is replaced with the third element in [x1, x2, x3] (representing the SOH data), resulting in the third constructed feature vector [z1, z2, x3]. In other words, multiple constructed feature vectors are obtained by sequentially replacing elements of the same data type in the feature vector to be analyzed with elements from the standard feature vector.
[0048] In each constructed feature vector, only one element is standard data, while the rest are potentially problematic data to be analyzed.
[0049] S106: Input the multiple constructed feature vectors sequentially into the pre-trained target neural network model to obtain the predicted range achievement rate corresponding to each constructed feature vector.
[0050] In some embodiments, the target neural network model is used to input feature vectors constructed from initial data, predict, analyze, and output the predicted range achievement rate corresponding to the vehicle. Therefore, the constructed feature vectors are sequentially input into the target neural network model, and the target neural network model outputs the predicted range achievement rate corresponding to each constructed feature vector.
[0051] For example, inputting [x1, z2, z3] into the target neural network model yields a predicted range achievement rate, denoted as y1. Inputting [z1, x2, z3] into the target neural network model yields a predicted range achievement rate, denoted as y2. Inputting [z1, z2, x3] into the target neural network model yields a predicted range achievement rate, denoted as y3. The actual range achievement rate of the vehicle being analyzed, corresponding to these multiple constructed feature vectors, is denoted as Y.
[0052] In some embodiments, the constructed feature vector can be normalized before being input into the target neural network model to obtain the predicted range achievement rate.
[0053] S107: Based on the predicted range achievement rate and the actual range achievement rate, determine the main influencing factors of range performance.
[0054] In some embodiments, the main influencing factors of range performance are determined based on the predicted range achievement rate and the actual range achievement rate, specifically including: S1: Calculate the difference between each predicted range achievement rate and the actual range achievement rate corresponding to the vehicle to be analyzed, and obtain multiple differences; S2: Determine the maximum value among multiple differences to obtain the target difference; S3: Determine the constructed feature vector corresponding to the target difference as the key feature vector; S4: Determine the standard data included in the key feature vector and determine the data type of the standard data as the main influencing factor of battery life performance.
[0055] Specifically, y1-Y is calculated to obtain the first difference Δ1, y2-Y is calculated to obtain the second difference Δ2, and y3-Y is calculated to obtain the third difference Δ3. Assuming the feature vector to be analyzed includes N different types of data (i.e., the number of data types equals N), then a total of N constructive feature vectors can be obtained, resulting in a total of N differences, Δi=yi-Y, where i takes values of 1, 2, 3, ..., N. Δi represents the i-th difference, and yi represents the i-th predicted range achievement rate. In this application, N=3.
[0056] Δi reflects the potential contribution of different data types to the improvement of battery life, and it is presented in a directly quantified form, thus achieving quantitative evaluation.
[0057] Assuming the maximum value of Δi (where i takes values of 1, 2, and 3) is Δ1, then the target difference is Δ1, and the corresponding constructed feature vector is [x1, z2, z3]. [x1, z2, z3] is the key feature vector. Since [x1, z2, z3] is obtained by replacing the "battery pack individual cell voltage difference data z1" in the data to be analyzed with the "battery pack individual cell voltage difference data x1" from the standard data (the standard data can be considered as the initial data when the range achievement rate is normal), after the replacement, it has the most significant effect on improving the predicted range achievement rate (Δ1 is the maximum value among multiple differences). This indicates that z1 has a greater influence than z2 and z3, causing the range achievement rate to be lower. Since [x1, z2, z3] includes the standard data x1, the "battery pack individual cell voltage difference" is taken as the main influencing factor on range performance.
[0058] Assuming the maximum value of Δi (where i takes values of 1, 2, and 3) is Δ2, then the target difference is Δ2, and the corresponding constructed feature vector is [z1, x2, z3]. [z1, x2, z3] is the key feature vector. Since [z1, x2, z3] is obtained by replacing the "SOC data z2" in the data to be analyzed with the "SOC data x2" in the standard data, it has the most significant effect on improving the predicted range achievement rate after the replacement (Δ2 is the maximum value among multiple differences). This indicates that z2 has a greater influence than z1 and z3, resulting in a lower range achievement rate. Since [z1, x2, z3] includes the standard data x2, "SOC" is taken as the main influencing factor on range performance.
[0059] Assuming the maximum value of Δi (where i takes values of 1, 2, and 3) is Δ3, then the target difference is Δ3, and the corresponding constructed feature vector is [z1, z2, x3], which is the key feature vector. Since [z1, z2, x3] is obtained by replacing the "SOH data z3" in the data to be analyzed with the "SOH data x3" from the standard data, it has the most significant effect on improving the predicted range achievement rate after the replacement (Δ3 is the maximum value among multiple differences). This indicates that z3 has a greater influence than z1 and z2, leading to a lower range achievement rate. Since [z1, z2, x3] includes the standard data x3, "SOH" is considered the main influencing factor on range performance.
[0060] For each of the multiple vehicles to be analyzed, steps S104 to S107 are executed respectively, thereby obtaining the main influencing factors of the range performance of each vehicle to be analyzed.
[0061] After identifying the main influencing factors, we can further analyze the initial data corresponding to these factors to diagnose potential problems with the vehicle under analysis.
[0062] It should be noted that the example given in this application is N equal to 3. Depending on the driving scenario of the sample vehicle, more types of initial data may be included (i.e., N may also be equal to other values). The method for obtaining the main influencing factors when including more types of initial data can be found in the foregoing embodiments, and will not be repeated here.
[0063] Based on the above embodiments, by comparing the magnitudes of multiple differences, the largest difference is determined as the target difference, and then the key feature vector is determined. The data types of the standard data in the key feature vector are taken as the main influencing factors (root cause features) of range performance. This process introduces the fluctuation contribution method to attribute the characteristics of vehicles with low range. By comparing the sensitivity and fluctuation contribution of different constructed feature vectors to range deviation, it is possible to quantitatively identify the dominant role of different features (different data types) in the range reduction process, realizing multi-level attribution from the feature layer to the system layer. This mechanism overcomes the defect of existing models that "can only predict but cannot explain", avoids the bias caused by simply relying on the importance of global features (global importance may not reflect the situation of individual vehicles), and makes the causes of abnormal range traceable and quantifiable. It can automatically and quantitatively identify the key variables causing low vehicle range without relying on human experience judgment, realizing automated diagnosis and explanation from prediction to finding the cause, and providing a reasonable basis for subsequent vehicle performance improvement and energy consumption optimization. Through an automated root cause identification mechanism, vehicle engineers can quickly pinpoint key issues affecting range (such as SOC estimation errors, high energy consumption, and deteriorating battery consistency), significantly reducing problem location time and improving the efficiency of operations and R&D. Furthermore, this method can be used for targeted diagnostics from single vehicle to single month, facilitating the development of precise rectification or optimization measures.
[0064] The training process of the target neural network model will be explained in detail below.
[0065] In some embodiments, the target neural network model is pre-trained in the following manner: S1: Obtain the initial data of the sample vehicle; S2: Perform data segmentation and outlier removal on the initial data to obtain the first feature vector; S3: Use the first feature vector to train the initial neural network model to obtain the target neural network model.
[0066] First, feature processing is required on the initial data. The steps for feature processing are explained below.
[0067] In some embodiments, initial data of the sample vehicles is obtained from a power battery big data platform. This initial data may specifically include: energy consumption data, charging time data, vehicle speed data, battery pack cell voltage difference data, SOC data, and SOH data. This initial data may be presented in tabular form, and the tabular data is aggregated according to the sample vehicle's VIN code (a unique code representing vehicle identification information) and by month.
[0068] Furthermore, the initial data is segmented, with each sample vehicle's "complete driving trajectory" considered as a group of initial data. The "mileage traveled in this instance" corresponding to each group of initial data is then calculated.
[0069] Furthermore, outlier removal is performed on the initial data. Specific methods for outlier removal include: S1: Check whether the "mileage of this trip" of each initial data set is greater than the second threshold. If the "mileage of this trip" of a certain initial data set is greater than the second threshold, it is considered an outlier and needs to be removed. S2: Sequentially check whether the change in SOC data (change in SOC data = SOC data at the first moment - SOC data at the last moment) in each set of initial data exceeds the third threshold. If the change in SOC data in a set of initial data exceeds the third threshold, it is considered an outlier and needs to be removed. The greater the change in SOC data, the more accurate the estimation of battery life achievement rate will be. However, the greater the change in SOC data, the less usable data will be. Therefore, it is necessary to balance these two points when selecting the third threshold. S3: In each set of initial data, check whether the ratio between the change in remaining mileage and the change in SOC data meets the preset range. If it does not meet the preset range, it is considered an outlier and needs to be removed. S4: In each set of initial data, check whether there is charging time data marked as "driving status". If so, it is considered an outlier and needs to be removed.
[0070] Outlier removal increases the accuracy of the remaining initial data, ensuring that the calculated initial range achievement rate is not excessively biased. Data splitting and outlier removal ensure that the input values (first feature vector) and input values (first range achievement rate) come from continuous, realistic operating conditions, preventing isolated outliers from affecting subsequent model training. Higher-quality training samples result in a more robust CatBoost regression model and improved generalization ability.
[0071] After data segmentation and outlier removal, the remaining initial data can be represented as a vector to obtain the first feature vector.
[0072] For each first feature vector, calculate its corresponding change in mileage and change in remaining mileage. For each first feature vector, calculate its corresponding first range achievement rate, which is calculated as: First range achievement rate = Change in mileage / Change in remaining mileage. The first range achievement rate can be understood as the true range achievement rate after removing outlier data.
[0073] For each first feature vector, feature extraction is performed on data exhibiting collinearity (such as SOC data and battery pack cell voltage difference data). Simultaneously, all data are designed to have a monotonic impact on the first range achievement rate, ensuring consistent influence. Feature extraction refers to the collinearity between SOC data and battery pack cell voltage difference data; an increase in one will also increase the other. To separate their influence, a new indicator is constructed to avoid this interplay. All first feature vectors are aggregated monthly. For the same sample vehicle, there may be multiple first feature vectors within a given month, each corresponding to a complete driving trajectory. The first feature vectors are then normalized. By removing collinear terms through feature extraction, the replacement effect in step S105 becomes more comparable, the physical meaning of Δi becomes clearer, and the attribution results more reliable, facilitating engineering implementation (such as adopting targeted maintenance measures).
[0074] After the above feature processing is completed, the training process of the neural network model will begin.
[0075] In this application embodiment, the CatBoost (Categorical Boosting) regression model can be selected as the initial neural network model. The CatBoost regression model was chosen as the initial neural network model because of its high stability and generalization ability in handling nonlinear relationships. The CatBoost regression model performs excellently when handling categorical variables and nonlinear relationships. Benchmark training followed by artificial grid + Optuna optimization ensures search space coverage while saving computational resources. The built-in cross-validation method prevents overfitting, better ensuring a balance between performance and cost, and the model training process is reproducible and verifiable. This application can also select other machine learning regression models as the initial neural network model.
[0076] Before model training, the Scikit-learn library in Python is used to divide all the first feature vectors and their corresponding first endurance achievement rates into training set, test set and validation set to ensure that the training and evaluation process of the model is independent and representative.
[0077] Training and testing sets were selected, with the first feature vector as the input and the first battery life achievement rate as the output (i.e., the target variable), to train the CatBoost regression model. First, the basic CatBoost model (with only necessary parameters set) was used for initial training to obtain the model's baseline performance. Then, a two-stage optimization strategy was adopted for the model's key hyperparameters (such as learning rate, tree depth, L2 regularization coefficient, number of iterations, etc.). In the first stage, the grid search method is used to explore the optimal combination of hyperparameters (such as learning rate, tree depth, L2 regularization coefficient, number of iterations, etc.) within a finite parameter space. Grid search is a hyperparameter optimization method that selects the optimal combination of parameters on the test set by exhaustively searching in a predefined parameter grid. In the second stage, based on the optimal hyperparameter combination obtained in the first stage, the Optuna framework is introduced, and Bayesian optimization algorithm is used for automated tuning to determine more refined and accurate hyperparameters, thereby improving the prediction accuracy and stability of the model.
[0078] After the hyperparameters are optimized, the model is validated on the validation set using the built-in cross-validation mechanism of the CatBoost regression model. This effectively prevents overfitting and ensures that the model has good generalization ability for unknown data.
[0079] Once the final hyperparameters of the CatBoost regression model are determined, the training and test sets are redefined, and the final model training and prediction experiments are conducted. By calculating and analyzing the root mean square error (RMSE) of the first result output by the CatBoost regression model on the test set (i.e., the predicted value of the first range achievement rate) and the first range achievement rate on the test set, the model's fitting ability and prediction performance are comprehensively evaluated, thereby determining the model's performance in the range achievement rate prediction task.
[0080] If the RMSE between the first result and the first range achievement rate is less than the fourth threshold (the fourth threshold is set to 0.005), the model can be considered to be reliable and effective in predicting the range achievement rate, and this trained CatBoost regression model can be used as the target neural network model.
[0081] Based on the above steps, a regression model, namely the target neural network model, can be obtained that can predict the range achievement rate based on vehicle operating condition driving characteristic indicators. The range achievement rate prediction framework built on the target neural network model can achieve unified prediction for different vehicle models, identify potential low range risks in advance, and utilize multi-source data from different operating conditions, time dimensions, and battery states during the training process of the target neural network model. This effectively avoids the bias and instability caused by single-condition data, providing support for subsequent problem early warning and optimization.
[0082] Through the above embodiments, this application constructs a big data modeling system integrating multi-condition features. By standardizing and filtering multi-dimensional vehicle condition features such as battery pack cell voltage difference data, SOC data, and SOH data, and combining machine learning / neural network algorithms to establish a range achievement rate prediction model, it achieves accurate quantitative evaluation of the vehicle's actual range performance. This system effectively overcomes the bias problem caused by traditional methods that calculate range achievement rate based on only a single condition. The comprehensive modeling using multi-condition features significantly improves the stability and reliability of range evaluation, while making the model training process more representative and generalizable. Furthermore, this application designs an automated processing and analysis workflow for vehicle lifecycle data, enabling unified modeling, diagnosis, and result visualization output for multiple vehicle models and batches. By introducing a big data computing and parallel model training framework, it can achieve rapid iterative analysis of hundreds of thousands of vehicle data points, providing real-time decision-making support. The range achievement prediction and root cause analysis method based on multi-condition fusion features achieves high-precision quantitative prediction of the actual range performance of vehicles, overcomes the problems of large calculation errors and poor generalization ability of traditional single-condition calculation, forms an attribution path for single vehicles and multiple conditions, and realizes closed-loop analysis from "abnormal results" to "factor diagnosis", providing an operable basis for vehicle design optimization, fault location and energy consumption management.
[0083] This application provides a device for determining the factors affecting vehicle range performance, see below. Figure 2 As shown, it includes: The first acquisition module 201 is used to acquire the actual range achievement rate of the sample vehicles, and divide the sample vehicles into vehicles to be analyzed and standard vehicles based on the actual range achievement rate. The second acquisition module 202 is used to acquire the initial data of the sample vehicle, wherein the initial data is data that affects the actual range achievement rate; The first determining module 203 is used to determine the standard feature vector of the standard vehicle; wherein the standard feature vector includes multiple standard data; the standard data is determined based on the initial data of the standard vehicle; The second determining module 204 is used to determine the feature vector to be analyzed of the vehicle to be analyzed; wherein the feature vector to be analyzed includes multiple data to be analyzed; the data to be analyzed is the initial data of the vehicle to be analyzed; The combination module 205 is used to combine the standard feature vector and the feature vector to be analyzed to obtain multiple constructed feature vectors; Prediction module 206 is used to sequentially input multiple constructed feature vectors into a pre-trained target neural network model to obtain the predicted range achievement rate corresponding to each constructed feature vector; Analysis module 207 is used to determine the main influencing factors of range performance based on the predicted range achievement rate and the actual range achievement rate.
[0084] This application provides an electronic device, which includes a processor and a memory for storing processor-executable instructions. When the processor executes the instructions, it implements the steps of the method described in any of the above embodiments.
[0085] This embodiment also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps of any of the methods in the above embodiments.
[0086] In the embodiments of this application, the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM), etc.
[0087] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0088] The foregoing has provided a detailed description of a method, apparatus, and electronic device for determining factors affecting vehicle range performance, as provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for determining an influencing factor of a vehicle's range performance, characterized in that include: Obtain the actual range achievement rate of the sample vehicles, and divide the sample vehicles into vehicles to be analyzed and standard vehicles based on the actual range achievement rate; Obtain initial data for the sample vehicles, wherein the initial data is data that affects the actual range achievement rate; Determine the standard feature vector of the standard vehicle; wherein the standard feature vector includes multiple standard data; the standard data are determined based on the initial data of the standard vehicle; Determine the feature vector to be analyzed for the vehicle to be analyzed; wherein the feature vector to be analyzed includes multiple data to be analyzed; the data to be analyzed is the initial data of the vehicle to be analyzed; The standard feature vector and the feature vector to be analyzed are combined to obtain multiple constructed feature vectors; Multiple constructed feature vectors are sequentially input into a pre-trained target neural network model to obtain the predicted range achievement rate corresponding to each constructed feature vector; Based on the predicted range achievement rate and the actual range achievement rate, the main factors affecting range performance are determined.
2. The method of claim 1, wherein, Based on the actual range achievement rate, the sample vehicles are divided into vehicles to be analyzed and standard vehicles, including: From the sample vehicles, vehicles whose actual range achievement rate is less than or equal to the first threshold are selected to obtain the vehicles to be analyzed. From the sample vehicles, vehicles whose actual range achievement rate is greater than the first threshold are selected to obtain standard vehicles.
3. The method of claim 2, wherein, The first threshold is determined in the following way: Calculate the average of the actual range achievement rates of the sample vehicles, and use it as the first average. The first threshold is obtained by multiplying the first average value by a preset proportional coefficient.
4. The method of claim 1, wherein, Determining the standard feature vector of the standard vehicle includes: Determine the initial data for the standard vehicle; Calculate the average value of the initial data for each category to obtain multiple second average values; Multiple second average values are combined to form a standard feature vector.
5. The method of claim 1, wherein, The standard feature vector and the feature vector to be analyzed are combined to obtain multiple constructed feature vectors, including: The data to be analyzed in the feature vector to be analyzed is replaced sequentially with standard data of the same type in the standard feature vector to obtain multiple constructed feature vectors.
6. The method according to claim 1, characterized in that, Based on the predicted range achievement rate and the actual range achievement rate, the main influencing factors of range performance are determined, including: Calculate the difference between each predicted range achievement rate and the actual range achievement rate corresponding to the vehicle to be analyzed, and obtain multiple differences; The target difference is obtained by determining the maximum value among multiple differences. Determine the constructed feature vector corresponding to the target difference as the key feature vector; The standard data included in the key feature vector is determined, and the data type of the standard data is determined as the main influencing factor of battery life performance.
7. The method according to claim 1, characterized in that, The target neural network model is pre-trained in the following manner: Obtain the initial data of the sample vehicles; The initial data is segmented and outliers are removed to obtain a first feature vector; The initial neural network model is trained using the first feature vector to obtain the target neural network model.
8. A device for determining factors affecting vehicle range performance, characterized in that, include: The first acquisition module is used to acquire the actual range achievement rate of the sample vehicles, and to divide the sample vehicles into vehicles to be analyzed and standard vehicles based on the actual range achievement rate. The second acquisition module is used to acquire the initial data of the sample vehicle, wherein the initial data is data that affects the actual range achievement rate; The first determining module is used to determine the standard feature vector of the standard vehicle; wherein the standard feature vector includes multiple standard data; the standard data is determined based on the initial data of the standard vehicle; The second determining module is used to determine the feature vector to be analyzed of the vehicle to be analyzed; wherein the feature vector to be analyzed includes multiple data to be analyzed; the data to be analyzed is the initial data of the vehicle to be analyzed; The combination module is used to combine the standard feature vector and the feature vector to be analyzed to obtain multiple constructed feature vectors; The prediction module is used to sequentially input multiple constructed feature vectors into a pre-trained target neural network model to obtain the predicted range achievement rate corresponding to each constructed feature vector; The analysis module is used to determine the main influencing factors of range performance based on the predicted range achievement rate and the actual range achievement rate.
9. An electronic device, characterized in that, The electronic device includes a processor and a memory for storing processor-executable instructions, wherein the processor executes the instructions to implement the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It stores a computer program, which is loaded by a processor to perform the steps of the method according to any one of claims 1 to 7.