Intelligent definite diagnosis and analysis method suitable for energy consumption abnormity of new energy vehicle

Abstract

The invention relates to the technical field of energy consumption analysis of new energy vehicles, in particular to an intelligent definite diagnosis and analysis method suitable for energy consumption abnormity of new energy vehicles, which comprises the following steps of: collecting historical data of cloud big data of the Internet of Vehicles, including CAN bus data, GPS position information, vehicle type and customer information, electric accessory operation parameters and air conditioner operation parameters of the vehicles; preprocessing the collected data, and designing an algorithm to extract multi-dimensional high-reliability characteristic factors influencing energy consumption; on the basis of a box plot method of a statistical principle, anomaly detection is carried out on the energy consumption value of one hundred kilometers per month of vehicles of the same vehicle type or the same order, abnormal points are identified, and data labels are generated. According to the invention, data acquisition is comprehensive, data in multiple aspects of vehicle operation, charging and the like are covered, comprehensiveness of analysis is ensured, and abundant input is provided for SHAP value calculation; moreover, the method can achieve the feature extraction of multiple dimensions, covers a plurality of influence factors, such as vehicle performance, operation conditions and driving behaviors, improves the precision of abnormality diagnosis, and guarantees the calculation reliability of the SHAP value.

Description

Technical Field

[0001] This invention relates to the field of energy consumption analysis technology for new energy vehicles, and in particular to an intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles. Background Technology

[0002] Against the backdrop of green transformation and the deepening of the "dual-carbon" strategy in the transportation sector, new energy buses have become the core carrier of urban and intercity green public transportation. Their energy efficiency is directly related to the cost control and energy conservation and emission reduction results of operators. With the rapid expansion of the new energy vehicle market and the continuous growth of vehicle ownership, the industry's need for monitoring and analyzing abnormal vehicle energy consumption is becoming increasingly urgent.

[0003] Currently, the industry's monitoring and analysis of abnormal energy consumption in new energy vehicles largely relies on maintenance personnel's experience-based judgment based on limited data and manual post-event troubleshooting. This traditional model has many shortcomings: 1. The analysis granularity is coarse, making it impossible to accurately capture energy consumption anomalies caused by multiple factors; 2. Poor timeliness; post-event investigation is insufficient to meet the needs of real-time energy consumption management. 3. It is highly dependent on manpower and cannot cope with the real-time, high-frequency data generated by a large number of vehicles, making it difficult to achieve large-scale, comprehensive, and refined energy consumption management.

[0004] Among existing related patented technologies, some solutions are designed for analyzing abnormal energy consumption in gasoline vehicles and cannot be adapted to the energy consumption characteristics of new energy vehicles; some solutions are limited to the analysis of specific fault causes or single influencing factors, resulting in a narrow scope of application; and some solutions use simple statistical methods for diagnosis, which have low accuracy and efficiency. Therefore, there is an urgent need for an intelligent analysis method that can accurately and quickly identify abnormal energy consumption in new energy vehicles and automatically attribute the causes, in order to overcome the shortcomings of existing technologies. Summary of the Invention

[0005] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description and other accompanying drawings.

[0006] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for intelligent diagnosis and analysis of abnormal energy consumption in new energy vehicles.

[0007] To achieve the above objectives, the technical solution of the present invention is: a method for intelligent diagnosis and analysis of abnormal energy consumption in new energy vehicles, comprising the following steps: S1. Collect historical data from the vehicle network cloud, including vehicle CAN bus data, GPS location information, vehicle model information, customer information, electrical accessory operating parameters, and air conditioning operating parameters; S2. Preprocess the data collected in step S1 and design an algorithm to extract multi-dimensional high-reliability feature factors that affect energy consumption. S3. Based on statistical principles, the box plot method is used to detect anomalies in the monthly energy consumption per 100 kilometers for vehicles of the same model or the same order (same customer and same batch), identify anomalies and generate data labels. S4. Construct a dataset, use a tree-based machine learning method to train and fit the dataset, and output the abnormal probability value of energy consumption of vehicles of the same model or the same order (same customer and same batch). S5. Initialize the TreeExplainer interpreter, calculate the SHAP value of each feature of each abnormal vehicle based on the Tree SHAP algorithm, and determine the feature with the greatest impact. S6. Convert and calculate the SHAP values ​​to obtain the influence weight index of each feature; S7. Output the abnormal vehicle number, abnormal probability, most influential feature, and influence weight.

[0008] In some embodiments, in step S1, the CAN bus data includes mileage data, vehicle speed data, battery voltage data, battery current data, motor operating parameters, and transmission system operating parameters.

[0009] In some embodiments, in step S2, the multi-dimensional high reliability characteristic factors include motor efficiency, transmission efficiency, road slope, air conditioning energy consumption ratio, electrical accessory energy consumption ratio, charging current, recycling efficiency, water-cooled unit energy consumption ratio during charging, acceleration characteristics, high speed ratio, idling speed ratio, and parking air conditioning start-up recognition ratio.

[0010] In some embodiments, the high-speed ratio is calculated as follows: extract the number of values ​​N1 with a vehicle speed greater than 50km / h, divide it by the number of values ​​N2 with a non-zero vehicle speed, and obtain the high-speed ratio = N1 / N2; the idling ratio is calculated as follows: count the total idling time of the vehicle, divide it by the total power-on time of the vehicle, and obtain the idling ratio = total idling time / total power-on time.

[0011] In some embodiments, the recycling efficiency is calculated as follows: identify the non-charging section of the vehicle, count the power value product of the total voltage and current, calculate the absolute value E1 of the sum of negative power and the sum of positive power E2, and the recycling efficiency = E1 / E2; the charging current is determined as follows: identify the start and end points of the charging time, collect the current value during the charging process, and calculate the magnitude of the charging current through probability density distribution.

[0012] In some embodiments, the energy consumption in step S2 includes battery thermal management energy consumption. The extraction method for battery thermal management energy consumption related features is as follows: when the vehicle adopts independent water cooling mode, the "TMS demand power" value is counted and integral calculation is performed; the calculation method for the parking air conditioning on recognition ratio is as follows: the percentage of time the air conditioning is not turned off when the vehicle is parked is counted, and the parking air conditioning on recognition ratio is determined in combination with the vehicle's usage habits.

[0013] In some embodiments, the anomaly detection process using the box plot method in step S3 includes: S31. Calculate the first quartile Q1 and the third quartile Q3 of the energy consumption data of the same batch of vehicles; S32. Calculate the interquartile range IQR = Q3 - Q1; S33. Determine the critical values: lower critical value = Q1 - 1.5 × IQR, upper critical value = Q3 + 1.5 × IQR; S34. Determine whether the monthly energy consumption per 100 kilometers is higher than the upper limit threshold. If so, mark it as an upper limit abnormal value and generate an abnormal label; otherwise, mark it as a normal label.

[0014] In some embodiments, in step S4, the tree-based machine learning method is a random forest classifier, and the training process of the random forest classifier includes dataset partitioning, hyperparameter tuning, and model validation.

[0015] In some embodiments, in step S5, the SHAP value is calculated using the following formula: The feature with the greatest impact is identified; where ϕi is the SHAP value of feature i, F is the set of all features, S is a subset of features, and f(S) is the marginal contribution of subset S; the TreeExplainer interpreter directly calculates the feature contribution through tree structure features, which include node splitting conditions and leaf node weights, and can obtain accurate Shapley values ​​without sampling.

[0016] In some embodiments, in step S6, the influence weight index is calculated by dividing the single feature SHAP value of an abnormal sample by the sum of the SHAP values ​​of all features of the abnormal sample to obtain the influence weight index corresponding to that feature.

[0017] By adopting the above technical solution, the beneficial effects of the present invention are: 1. Comprehensive data collection, covering various aspects such as vehicle operation, charging, and auxiliary systems, ensures comprehensive analysis and provides rich input for SHAP value calculation; 2. Feature extraction is multi-dimensional, covering multiple influencing factors such as vehicle performance, operating conditions, and driving behavior, improving the accuracy of anomaly diagnosis and ensuring the reliability of SHAP value calculation; 3. Anomaly detection employs statistical methods and automatic labeling, reducing manual costs, improving detection efficiency, and providing high-quality samples for model training and SHAP value calculation; 4. The model training uses a random forest classifier, which has high prediction accuracy and can quantify the probability of anomalies, providing high-quality model support for SHAP value calculation. 5. Combining the Tree SHAP algorithm and explicit calculation formulas to achieve model interpretability analysis, accurately locate the root cause of anomalies, provide clear directions for investigation, and ensure the rigor and repeatability of the method; 6. The output results are comprehensive, including the abnormal vehicle number, the probability of abnormality, the most influential feature and the influence weight, providing full support for operation and maintenance decisions.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0019] Undoubtedly, such and other objects of the present invention will become more apparent after the following detailed description of the preferred embodiments, which are illustrated in various accompanying drawings and figures.

[0020] To make the above-mentioned beneficial effects and other objects, features and advantages of the present invention more apparent and understandable, one or more preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0022] In the accompanying drawings, the same parts use the same reference numerals, and the drawings are schematic and not necessarily drawn to actual scale.

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only one or more embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on such drawings without creative effort.

[0024] Explanation of key figure labels: Figure 1 This is a flowchart of an intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to the present invention. Detailed Implementation

[0025] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings and examples, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly. It should be noted that, as long as there is no conflict, the various embodiments and features in the various embodiments of the present invention can be combined with each other, and the resulting technical solutions are all within the protection scope of the present invention.

[0026] Furthermore, numerous specific details are set forth in the following description for illustrative purposes to provide a thorough understanding of the embodiments of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without the specific details or particular methods described herein.

[0027] Please see Figure 1 This invention provides a method for intelligent diagnosis and analysis of abnormal energy consumption in new energy vehicles, comprising the following steps: S1. Collect historical data from the vehicle network cloud, including vehicle CAN bus data, GPS location information, vehicle model information, customer information, electrical accessory operating parameters, and air conditioning operating parameters; S2. Preprocess the data collected in step S1 and design an algorithm to extract multi-dimensional high-reliability feature factors that affect energy consumption. S3. Based on statistical principles, the box plot method is used to detect anomalies in the monthly energy consumption per 100 kilometers for vehicles of the same model or the same order (same customer and same batch), identify anomalies and generate data labels. S4. Construct a dataset, use a tree-based machine learning method to train and fit the dataset, and output the abnormal probability value of energy consumption of vehicles of the same model or the same order (same customer and same batch). S5. Initialize the TreeExplainer interpreter, calculate the SHAP value of each feature of each abnormal vehicle based on the Tree SHAP algorithm, and determine the feature with the greatest impact. S6. Convert and calculate the SHAP values ​​to obtain the influence weight index of each feature; S7. Output the abnormal vehicle number, abnormal probability, most influential feature, and influence weight.

[0028] When in use, the system first acquires multi-source data generated during the operation of new energy vehicles through the vehicle-to-everything (V2X) cloud, providing a data foundation for subsequent analysis. Then, the raw data undergoes preprocessing and feature extraction to identify key factors affecting energy consumption. Box plots are used to quickly identify and mark vehicles with abnormal energy consumption. A random forest classifier is used to train the data, enabling a quantitative assessment of the probability of energy consumption anomalies. The TreeExplainer interpreter, combined with the SHAP algorithm, is used to uncover the root causes of anomalies and determine the influence weights of each feature. Finally, comprehensive anomaly diagnosis results are output, providing a basis for operation and maintenance decisions. This process automates the entire process from data acquisition to result output, avoiding the subjectivity and limitations of traditional manual analysis, thereby improving the efficiency and accuracy of energy consumption anomaly diagnosis. Simultaneously, interpretability analysis enables anomaly detection and root cause localization.

[0029] According to some embodiments of this application, optionally, in step S1, the CAN bus data includes mileage data, vehicle speed data, battery voltage data, battery current data, motor operating parameters, and transmission system operating parameters.

[0030] Specifically, the CAN bus is the core data transmission channel for new energy vehicles, capable of collecting key parameters during vehicle operation in real time. Mileage data reflects the vehicle's cumulative mileage, speed data reflects driving status, battery voltage and current data are directly related to battery energy consumption, and motor and transmission system operating parameters reflect the energy consumption characteristics of the powertrain. These data collectively constitute the foundational data support for energy consumption analysis. Its role is to clarify the specific types of CAN bus data, ensuring the comprehensiveness and relevance of data collection, providing high-quality foundational data for subsequent feature extraction and anomaly analysis, and avoiding analytical errors caused by missing data.

[0031] According to some embodiments of this application, optionally, in step S2, the multi-dimensional high reliability characteristic factors include motor efficiency, transmission efficiency, road slope, air conditioning energy consumption ratio, electrical accessory energy consumption ratio, charging current, recycling efficiency, water-cooled unit energy consumption ratio during charging, acceleration characteristics, high speed ratio, idling speed ratio, and parking air conditioning start-up recognition ratio.

[0032] Specifically, the energy consumption of new energy vehicles is influenced by a combination of factors. Vehicle performance, such as motor efficiency and transmission efficiency, directly determines energy loss during power transmission. Operating conditions, including road gradient, highway speed, and idling speed, affect energy consumption during driving. Driving behavior, such as acceleration characteristics and the use of air conditioning while parked, influences additional energy consumption. Charging current and recycling efficiency affect energy replenishment and recovery. The energy consumption of electrical accessories and air conditioning directly reflects the energy consumption of auxiliary systems. By extracting these multi-dimensional characteristic factors, we can avoid the limitations of single-feature analysis, ensuring a comprehensive capture of key factors affecting energy consumption. This provides a basis for accurate subsequent anomaly diagnosis, making the identification of anomaly causes more comprehensive and accurate.

[0033] According to some embodiments of this application, optionally, the high-speed ratio is calculated as follows: extract the number of values ​​N1 with a vehicle speed greater than 50km / h, divide it by the number of values ​​N2 with all non-zero vehicle speeds, and obtain the high-speed ratio = N1 / N2; the idling ratio is calculated as follows: count the total idling time of the vehicle, divide it by the total power-on time of the vehicle, and obtain the idling ratio = total idling time / total power-on time.

[0034] Specifically, at high speeds, vehicle wind resistance increases, resulting in higher energy consumption compared to low and medium speeds. At idle, the vehicle's powertrain operates but without effective mileage, representing ineffective energy consumption. By statistically analyzing the percentage of speeds exceeding 50 km / h, the impact of high-speed driving can be quantified. Similarly, by statistically analyzing the proportion of total idling time to total power-on time, the energy consumption impact of idling can be quantified. Standardized calculation methods ensure the comparability of high-speed and idling percentages across different vehicles and time periods, thereby avoiding feature distortion caused by differences in calculation methods and improving the reliability and effectiveness of the features.

[0035] According to some embodiments of this application, optionally, the recycling efficiency is calculated as follows: identify the non-charging section of the vehicle, count the power value product of the total voltage and current, calculate the absolute value E1 of the sum of negative power and the sum of positive power E2, and the recycling efficiency = E1 / E2; the charging current is determined as follows: identify the start and end points of the charging time, collect the current value during the charging process, and calculate the magnitude of the charging current through probability density distribution.

[0036] Specifically, the energy recovery system of new energy vehicles can recover some energy during braking or deceleration, and the recovery efficiency directly affects the vehicle's overall energy consumption. The magnitude of the charging current affects charging efficiency and battery life, thus indirectly affecting the vehicle's energy consumption performance. By calculating the energy recovery ratio during non-charging periods, the performance of the energy recovery system can be accurately evaluated; by analyzing charging current data through probability density distribution, the vehicle's charging habits and charging status can be accurately reflected. This ensures the accuracy and effectiveness of charging-related characteristics, provides a reliable basis for analyzing the impact of the charging process on energy consumption, and enables the diagnosis of abnormal causes to cover charging-related factors.

[0037] According to some embodiments of this application, optionally, the energy consumption in step S2 includes battery thermal management energy consumption. The extraction method for battery thermal management energy consumption related features is as follows: when the vehicle adopts independent water cooling mode, the "TMS demand power" value is counted and integral calculation is performed; the calculation method for the parking air conditioning on recognition ratio is as follows: the proportion of time when the air conditioning is not turned off when the vehicle is parked is counted, and the parking air conditioning on recognition ratio is determined in combination with the vehicle's usage habits.

[0038] Specifically, the battery thermal management system aims to maintain the battery's optimal operating temperature, and its energy consumption is a significant component of the energy consumption of new energy vehicles. In independent water-cooling mode, this energy consumption can be accurately quantified by integrating the "TMS demand power" value. Leaving the air conditioning on while parked results in additional battery discharge, which is considered ineffective energy consumption. By statistically analyzing the percentage of this time and combining it with vehicle usage habits, the energy consumption impact of such undesirable driving behaviors can be accurately identified. This feature extraction method fills the gap in the analysis of the energy consumption impact of the battery thermal management system and the air conditioning while parked, making the feature factors more comprehensive and further improving the accuracy of anomaly diagnosis. It can identify energy consumption anomalies that are easily overlooked in traditional analysis.

[0039] According to some embodiments of this application, optionally, the outlier detection principle in step S3 is to construct a box plot based on the quartiles (Q1, Q3) of the same batch of vehicle energy consumption data, where Q1 is the first quartile, Q3 is the third quartile, the box represents the range from Q1 to Q3, and the length is the interquartile range (IQR = Q3 - Q1); this principle is summarized in the following steps: S31. Calculate the first quartile Q1 and the third quartile Q3 of the energy consumption data of the same batch of vehicles; S32. Calculate the interquartile range IQR = Q3 - Q1; S33. Determine the critical values: lower critical value = Q1 - 1.5 × IQR, upper critical value = Q3 + 1.5 × IQR; S34. Determine whether the monthly energy consumption per 100 kilometers is higher than the upper limit threshold. If so, mark it as an upper limit abnormal value and generate an abnormal label; otherwise, mark it as a normal label.

[0040] Specifically, the box plot method, based on the statistical distribution characteristics of data, uses quartiles to define the normal distribution range of the data. Energy consumption data from the same batch of vehicles typically exhibits a certain distribution pattern. The interval between the first quartile (Q1) and the third quartile (Q3) contains 75% of the normal data. A critical value obtained by expanding the range by 1.5 times the interquartile range effectively distinguishes between normal and abnormal data. Data points exceeding the upper critical value usually deviate from the normal distribution and can be identified as having abnormal energy consumption. This box plot method, based on statistical principles, effectively avoids the high cost and subjectivity of manual labeling, automatically and quickly identifying vehicles with abnormal energy consumption while ensuring the accuracy and objectivity of anomaly detection, providing reliable labeled data for subsequent model training.

[0041] According to some embodiments of this application, optionally, in step S4, the machine learning method based on the tree model is a random forest classifier, and the training process of the random forest classifier includes dataset partitioning, hyperparameter tuning, and model validation.

[0042] Specifically, the random forest classifier consists of multiple decision trees and improves the model's generalization ability and prediction accuracy through ensemble learning. During training, the dataset is first divided into a training set and a validation set. The training set is used for model training, and the validation set is used for model performance evaluation. The model structure is optimized through hyperparameter tuning (such as the number of decision trees and maximum depth). Finally, the model's prediction accuracy is verified using the validation set to ensure that the model can accurately output the probability of abnormal vehicle energy consumption. The random forest classifier has strong anti-overfitting ability and high prediction accuracy, effectively handling multi-dimensional feature data. By quantifying the output of abnormal probabilities, it provides a basis for assessing the severity of energy consumption anomalies, making anomaly diagnosis more precise and quantifiable.

[0043] According to some embodiments of this application, optionally, in step S5, the SHAP value is calculated using the following formula: The TreeExplainer directly calculates feature contributions through tree structure features, including node splitting conditions and leaf node weights, to obtain accurate Shapley values ​​without sampling.

[0044] Specifically, the TreeExplainer interpreter is based on the Tree SHAP algorithm and is specifically designed for interpretability analysis of tree models. The structure of a tree model consists of node splitting conditions and leaf node weights. By parsing the tree structure, TreeExplainer directly calculates the contribution value (i.e., SHAP value) of each feature to the prediction result without sampling, thus obtaining accurate Shapley values ​​and avoiding errors caused by approximate calculations.

[0045] The formula for calculating the SHAP value is: The meanings of each parameter are as follows: ϕi: SHAP value of feature i, used to quantify the contribution of feature i to the energy consumption anomaly result. The positive or negative value indicates whether the feature promotes or inhibits the anomaly, and the larger the absolute value, the stronger the influence. F: The set of all features involved in the analysis, covering multi-dimensional features such as motor efficiency, transmission efficiency, and road slope extracted in this invention; S: Any subset of the feature set F that does not contain feature i. By traversing all possible subsets S, we ensure that the contribution of each feature is fairly distributed. f(S): The marginal contribution of the feature subset S to the model's prediction results, i.e., the model's prediction output when only the subset S is considered; f(S∪{i}): The marginal contribution of feature subset S to the model prediction result after adding feature i. The marginal contribution of feature i is obtained by comparing the difference between f(S∪{i}) and f(S), and then combined with the combination coefficients. (Based on the principle of permutation and combination, ensure that the weights of each subset S are fair), and sum them to obtain the SHAP value of feature i.

[0046] The above SHAP value calculation formula can accurately uncover the core impact characteristics of abnormal energy consumption in each abnormal vehicle, solving the "black box" problem of traditional machine learning models, making the location of abnormal causes more scientific and reliable, providing maintenance personnel with a clear direction for investigation, and the publication of the formula can also ensure the repeatability and rigor of the method.

[0047] According to some embodiments of this application, optionally, in step S6, the influence weight index is calculated by dividing the single feature SHAP value of the abnormal sample by the sum of the SHAP values ​​of all features of the abnormal sample to obtain the influence weight index corresponding to the feature.

[0048] Specifically, the SHAP value reflects the contribution of a feature to anomalies, but the magnitude of SHAP values ​​may differ across features. By dividing the SHAP value of a single feature by the sum of the SHAP values ​​of all features in the anomaly sample, the feature contribution can be standardized into a weighted index between 0 and 1, intuitively reflecting the proportion of each feature's impact on energy consumption anomalies. This calculation method further quantifies and normalizes the degree of influence. Quantifying the influence weighted index allows maintenance personnel to clearly understand the degree of influence of each anomaly cause, prioritize factors with high influence weights, improve the efficiency and targeting of energy consumption anomaly troubleshooting, and provide a clear direction for vehicle performance optimization. Example

[0049] The application targets vehicles of the same model or from the same order (same customer, same batch), and the operation route covers urban main roads, suburban roads, and intercity highways. The operation time is 6 months, and the specific implementation process is as follows: 1. Data Collection: Collect 6 months of historical data for vehicles of the same model or from the same order through the vehicle network cloud, including: CAN bus data: Mileage data (unit: km), vehicle speed data (unit: km / h), battery voltage data (unit: V), battery current data (unit: A), motor speed (unit: r / min), motor torque (unit: N·m), transmission system speed (unit: r / min). GPS location information: latitude, longitude, and altitude; Vehicle information: vehicle model, battery capacity (kWh), motor power (kW); Customer information: operating company, route to which the vehicle belongs; Electrical accessory operating parameters: Lights on / off status, audio operating power (unit: W); Air conditioner operating parameters: air conditioner on status, air conditioner set temperature (unit: ℃), air conditioner operating power (unit: kW).

[0050] 2. Feature Extraction: The collected data is preprocessed, including data cleaning (removing missing and outliers) and data standardization, and then multi-dimensional, highly reliable feature factors are extracted. Motor efficiency: Calculated based on motor input power and output power, motor efficiency = motor output power / motor input power; Transmission efficiency: Calculated based on the input and output power of the transmission system, transmission efficiency = output power of transmission system / input power of transmission system; Road gradient: Calculated based on GPS altitude data and driving distance, road gradient = (end point altitude - start point altitude) / driving distance; Air conditioning energy consumption percentage: Total air conditioning energy consumption / Total vehicle energy consumption; Electric accessory energy consumption percentage: Total energy consumption of electric accessories / Total energy consumption of the vehicle; Charging current: By identifying the start and end times of charging, current data during the charging process is collected, and the average charging current is calculated using probability density distribution. Recovery efficiency: Identify the non-charging section, calculate the absolute value of the sum of negative power E1 and the sum of positive power E2, recovery efficiency = E1 / E2; Energy consumption percentage of water-cooled unit during charging: Total energy consumption of water-cooled unit during charging / Total energy consumption during charging; Acceleration characteristics: Calculated based on the rate of change of vehicle speed, acceleration = (current vehicle speed - previous vehicle speed / time interval); High-speed percentage: The number of values ​​with a speed greater than 50km / h N1 is extracted and divided by the number of all non-zero speed values ​​N2. High-speed percentage = N1 / N2. Idle speed percentage: The total idle time (the time when the vehicle speed is 0 and the motor is running) is calculated and divided by the total power-on time of the vehicle. Idle speed percentage = Total idle time / Total power-on time; Air conditioning on / off recognition ratio: The time the air conditioning is on while the car is parked is counted and divided by the total parking time. Parking air conditioning on / off recognition ratio = Time the air conditioning is on / off / Total parking time.

[0051] 3. Anomaly Detection: The box plot method is used to detect anomalies in the monthly energy consumption per 100 kilometers for each vehicle. The first quartile Q1 of the monthly energy consumption data of 100 buses is 25 kWh / 100km, and the third quartile Q3 is 30 kWh / 100km. Calculate the interquartile range (IQR): IQR = Q3 - Q1 = 5 kWh / 100 km; Determine the critical values: Lower critical value = 25 - 1.5 × 5 = 17.5 kWh / 100km, Upper critical value = 30 + 1.5 × 5 = 37.5 kWh / 100km; The monthly energy consumption per 100 kilometers for each vehicle is assessed. If a vehicle's monthly energy consumption per 100 kilometers is 40 kWh / 100 km, which is higher than the upper limit threshold of 37.5 kWh / 100 km, it is marked as abnormal. If a vehicle's monthly energy consumption per 100 kilometers is 28 kWh / 100 km, which is within the normal range, it is marked as normal.

[0052] 4. Model Training: Construct a dataset, using data from the first 5 months as the training set and data from the 6th month as the validation set. A random forest classifier is used for training and fitting. Dataset split: The training set contains 500 samples (100 cars × 5 months), and the validation set contains 100 samples (100 cars × 1 month). Hyperparameter tuning: Determine the optimal hyperparameters of the random forest classifier through grid search, with the number of decision trees = 100, maximum depth = 10, and minimum number of sample splits = 5; Model validation: The model accuracy on the validation set is 92%, and it can accurately output the abnormal probability value of energy consumption for each vehicle. For example, the abnormal probability of an abnormal vehicle is 0.95, and the abnormal probability of a normal vehicle is 0.03.

[0053] 5. Feature Influence Analysis: Initialize the TreeExplainer interpreter based on the Tree SHAP algorithm and formulas. Calculate the SHAP value for each characteristic of each abnormal vehicle: The feature set F of the abnormal vehicle contains 12 features such as motor efficiency, transmission efficiency, and road gradient. Iterate through all subsets S that do not contain a single feature i, calculate the marginal contribution f(S) of each subset S and the marginal contribution f(S∪{i}) after adding feature i, and sum the combination coefficients to obtain the SHAP value of each feature. The calculated SAP values ​​for the abnormal vehicle are as follows: road gradient SAP value = 0.3, air conditioning energy consumption ratio SAP value = 0.25, motor efficiency SAP value = 0.15, transmission efficiency SAP value = 0.1, high speed ratio SAP value = 0.08, idling ratio SAP value = 0.07, and total SAP value for other characteristics = 0.05. The most significant feature affecting this vehicle is the road gradient (the absolute value of SHAP is the largest).

[0054] 6. Impact Weight Calculation: Calculate the impact weight index for each feature: The weighting index for the impact of road slope = 0.3 / (0.3 + 0.25 + 0.15 + 0.1 + 0.08 + 0.07 + 0.05) = 0.3 / 1 = 0.3; The weighting index for the proportion of air conditioning energy consumption is 0.25 / 1 = 0.25; The weighting index for the impact of motor efficiency is 0.15 / 1 = 0.15; The weighting index for the impact of transmission efficiency is 0.1 / 1 = 0.1; The weighting index for the proportion of high-speed traffic is 0.08 / 1 = 0.08; The weighting index for the impact of idling rate is 0.07 / 1 = 0.07; The weighting index for other features is 0.05 / 1 = 0.05.

[0055] 7. Output Results: Output the diagnostic results for the abnormal vehicle: Abnormal vehicle number = XXX, abnormal probability = 0.95, maximum influencing feature = road slope, influence weight = 0.3.

[0056] Example Effect Analysis: This invention's method was used to diagnose energy consumption anomalies in vehicles of the same model or from the same order, identifying 12 vehicles with abnormal energy consumption. The anomalies in 8 vehicles were primarily caused by road gradient and air conditioning energy consumption ratios, in 3 vehicles by low motor efficiency, and in 1 vehicle by abnormal charging current. Maintenance personnel conducted targeted investigations based on the diagnostic results: for vehicles significantly affected by road gradient, optimized driving routes; for vehicles with high air conditioning energy consumption ratios, standardized air conditioning usage procedures; for vehicles with low motor efficiency, performed motor maintenance; and for vehicles with abnormal charging current, inspected and repaired charging system faults. After rectification, the average energy consumption per 100 kilometers of the 12 abnormal vehicles decreased by 8%, and the overall vehicle energy consumption decreased by 3%, significantly improving operational economy and energy conservation and emission reduction effects, validating the effectiveness and practicality of this invention's method (including the application of the SHAP value calculation formula).

[0057] It should be understood that the embodiments disclosed herein are not limited to the specific processing steps or materials disclosed herein, but should be extended to equivalent substitutions of such features as understood by those skilled in the art. It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting.

[0058] The term "embodiment" in this specification refers to a specific feature or characteristic described in connection with an embodiment that is included in at least one embodiment of the invention. Therefore, phrases or "embodiments" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0059] Furthermore, the described features or characteristics can be incorporated into one or more embodiments in any other suitable manner. In the above description, specific details, such as thickness, quantity, etc., are provided to provide a comprehensive understanding of embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented without the aforementioned specific details or may be implemented using other methods, components, materials, etc.

Claims

1. A method for intelligent diagnosis and analysis of abnormal energy consumption in new energy vehicles, characterized in that, Includes the following steps: S1. Collect historical data from the vehicle network cloud, including vehicle CAN bus data, GPS location information, vehicle model information, customer information, electrical accessory operating parameters, and air conditioning operating parameters; S2. Preprocess the data collected in step S1 and design an algorithm to extract multi-dimensional high-reliability feature factors that affect energy consumption. S3. Based on statistical principles, the box plot method is used to detect anomalies in the monthly energy consumption per 100 kilometers for vehicles of the same model or the same order (same customer and same batch), identify anomalies and generate data labels. S4. Construct a dataset, use a tree-based machine learning method to train and fit the dataset, and output the abnormal probability value of energy consumption of vehicles of the same model or the same order (same customer and same batch). S5. Initialize the TreeExplainer interpreter, calculate the SHAP value of each feature of each abnormal vehicle based on the Tree SHAP algorithm, and determine the feature with the greatest impact. S6. Convert and calculate the SHAP values ​​to obtain the influence weight index of each feature; S7. Output the abnormal vehicle number, abnormal probability, most influential feature, and influence weight.

2. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 1, characterized in that, In step S1, the CAN bus data includes mileage data, vehicle speed data, battery voltage data, battery current data, motor operating parameters, and transmission system operating parameters.

3. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 1, characterized in that, In step S2, the multi-dimensional high reliability characteristic factors include motor efficiency, transmission efficiency, road slope, air conditioning energy consumption ratio, electrical accessory energy consumption ratio, charging current, recycling efficiency, water-cooled unit energy consumption ratio during charging, acceleration characteristics, high speed ratio, idling speed ratio, and parking air conditioning start-up recognition ratio.

4. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 3, characterized in that, The high-speed percentage is calculated as follows: extract the number of values ​​N1 with a vehicle speed greater than 50km / h, divide it by the number of values ​​N2 with a non-zero vehicle speed, and get the high-speed percentage = N1 / N2; the idling percentage is calculated as follows: count the total idling time of the vehicle, divide it by the total power-on time of the vehicle, and get the idling percentage = total idling time / total power-on time.

5. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 3, characterized in that, The recycling efficiency is calculated as follows: identify the non-charging section of the vehicle, count the power value product of the total voltage and current, calculate the absolute value E1 of the sum of negative power and the sum of positive power E2, and the recycling efficiency = E1 / E2; the charging current is determined as follows: identify the start and end points of the charging time, collect the current value during the charging process, and calculate the magnitude of the charging current through probability density distribution.

6. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 3, characterized in that, The energy consumption in step S2 includes battery thermal management energy consumption. The extraction method for battery thermal management energy consumption related features is as follows: when the vehicle is in independent water cooling mode, the "TMS demand power" value is counted and integral calculation is performed. The calculation method for the parking air conditioning on recognition ratio is as follows: the percentage of time the air conditioning is not turned off when the vehicle is parked is counted, and the parking air conditioning on recognition ratio is determined in combination with the vehicle's usage habits.

7. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 1, characterized in that, The anomaly detection process using the box plot method in step S3 includes: S31. Calculate the first quartile Q1 and the third quartile Q3 of the energy consumption data of the same batch of vehicles; S32. Calculate the interquartile range IQR = Q3 - Q1; S33. Determine the critical values: lower critical value = Q1 - 1.5 × IQR, upper critical value = Q3 + 1.5 × IQR; S34. Determine whether the monthly energy consumption per 100 kilometers is higher than the upper limit threshold. If so, mark it as an upper limit abnormal value and generate an abnormal label; otherwise, mark it as a normal label.

8. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 1, characterized in that, In step S4, the tree-based machine learning method is the random forest classifier. The training process of the random forest classifier includes dataset partitioning, hyperparameter tuning, and model validation.

9. A method for intelligent diagnosis and analysis of abnormal energy consumption in new energy vehicles according to claim 1, characterized in that, In step S5, the SHAP value is calculated using the following formula: The feature with the greatest impact is identified; where ϕi is the SHAP value of feature i, F is the set of all features, S is a subset of features, and f(S) is the marginal contribution of subset S; the TreeExplainer interpreter directly calculates the feature contribution through tree structure features, which include node splitting conditions and leaf node weights, and can obtain accurate Shapley values ​​without sampling.

10. The intelligent diagnostic analysis method for abnormal energy consumption in new energy vehicles according to claim 1, characterized in that, In step S6, the influence weight index is calculated by dividing the single feature SHAP value of the abnormal sample by the sum of the SHAP values ​​of all features of the abnormal sample to obtain the influence weight index corresponding to that feature.

Images