WO2021244632A1 - 一种电动汽车能耗预测方法及系统 - Google Patents

一种电动汽车能耗预测方法及系统 Download PDF

Info

Publication number
WO2021244632A1
WO2021244632A1 PCT/CN2021/098285 CN2021098285W WO2021244632A1 WO 2021244632 A1 WO2021244632 A1 WO 2021244632A1 CN 2021098285 W CN2021098285 W CN 2021098285W WO 2021244632 A1 WO2021244632 A1 WO 2021244632A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
driving
energy consumption
electric vehicle
driving state
Prior art date
Application number
PCT/CN2021/098285
Other languages
English (en)
French (fr)
Inventor
王震坡
刘鹏
张瑾
张照生
Original Assignee
北京理工大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京理工大学 filed Critical 北京理工大学
Publication of WO2021244632A1 publication Critical patent/WO2021244632A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00Computer-aided design [CAD]
    • G06F30/20Design optimisation, verification or simulation
    • G06F30/27Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00Administration; Management
    • G06Q10/04Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • YGENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02TCLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00Road transport of goods or passengers
    • Y02T10/60Other road transportation technologies with climate change mitigation effect
    • Y02T10/70Energy storage systems for electromobility, e.g. batteries

Definitions

  • the invention relates to the field of automobiles, in particular to a method and system for predicting energy consumption of electric vehicles.
  • the purpose of the present invention is to provide an electric vehicle energy consumption prediction method and system, which integrates the prediction of future driving conditions of the vehicle into the energy consumption prediction process, and at the same time considers the influence of driving environment factors to improve the accuracy of electric vehicle energy consumption prediction.
  • the present invention provides the following solutions:
  • An electric vehicle energy consumption prediction method including:
  • the travel segment data includes historical travel data of the electric vehicle during driving
  • the dynamic segment data includes the electric vehicle Historical driving data during constant speed driving or acceleration driving
  • the driving characteristic parameters of the prediction data of the working condition are input into the established energy consumption prediction model to obtain the energy consumption prediction value.
  • the use of the dynamic segment data and the Markov-Monte Carlo method to predict the operating conditions of the electric vehicle to obtain the operating condition prediction data of the electric vehicle specifically includes:
  • a Monte Carlo simulation method, the driving state transition probability matrix and the driving state flag are used to predict the working condition of the electric vehicle to obtain the working condition prediction data of the electric vehicle.
  • the calculation of the driving state transition probability matrix of the electric vehicle by using the time sequence of the dynamic segment data and the driving state mark specifically includes:
  • the transition probability between all the driving state flags is used to determine the driving state transition probability matrix of the electric vehicle.
  • the use of the Monte Carlo simulation method, the driving state transition probability matrix and the driving state flag to predict the working condition of the electric vehicle to obtain the working condition prediction data of the electric vehicle specifically includes:
  • an energy consumption prediction model specifically includes:
  • the grid search method is used to optimize the hyperparameters of the initial energy consumption prediction model to obtain the energy consumption prediction model.
  • An electric vehicle energy consumption prediction system including:
  • the acquisition module is used to acquire historical driving data of electric vehicles
  • the segmentation processing module is used to perform segmentation processing on the historical driving data to obtain travel segment data and dynamic segment data;
  • the travel segment data includes historical travel data of the electric vehicle during driving, and the dynamic segment
  • the data includes historical driving data of the electric vehicle in the process of driving at a constant speed or at an acceleration;
  • the working condition prediction module is used to predict the working condition of the electric vehicle by using the dynamic segment data and the Markov-Monte Carlo method to obtain the working condition prediction data of the electric vehicle;
  • the first acquisition module is used to acquire driving characteristic parameters and energy consumption data of the trip segment data
  • An energy consumption prediction model establishment module configured to take the driving characteristic parameters of the travel segment data as an input, and the energy consumption data as an output, and establish an energy consumption prediction model by using a machine learning method
  • the second acquisition module is used to acquire the driving characteristic parameters of the prediction data of the working condition
  • the energy consumption prediction module is used to input the driving characteristic parameters of the working condition prediction data into the established energy consumption prediction model to obtain the energy consumption prediction value.
  • the working condition prediction module specifically includes:
  • a driving state mark adding unit for adding driving state marks to different driving states in the dynamic fragment data by using the average speed in the dynamic fragment data
  • a driving state transition probability matrix calculation unit configured to use the time sequence of the dynamic segment data and the driving state mark to calculate the driving state transition probability matrix of the electric vehicle
  • the working condition prediction unit is used for predicting the working condition of the electric vehicle by using the Monte Carlo simulation method, the driving state transition probability matrix and the running state mark to obtain the working condition prediction data of the electric vehicle.
  • the driving state transition probability matrix calculation unit specifically includes:
  • the transition probability calculation subunit is used to use the time sequence of the kinetic segment data, according to the formula Calculate the transition probability of the driving state of the electric vehicle from the driving state mark i to the driving state mark j; in the formula, p ij represents the transition probability; N ij represents the number of events that transfer from the driving state mark i to the driving state mark j;
  • the driving state transition probability matrix calculation subunit is used for determining the driving state transition probability matrix of the electric vehicle by using the transition probabilities between all the driving state flags.
  • the working condition prediction unit specifically includes:
  • the next time driving state mark determining subunit is used to determine the next time driving state mark of the electric vehicle by using a Monte Carlo simulation method and the driving state transition probability matrix;
  • Predicted driving condition data determining subunit used to determine historical driving data in the dynamic segment data that is the same as the next-time driving state mark to obtain predicted driving condition data
  • the first obtaining subunit is used to obtain the current driving conditions and destination mileage of the electric vehicle
  • the second obtaining subunit is used to obtain the mileage length of the predicted data of the working condition
  • the first judgment subunit is used to judge whether the mileage length of the working condition prediction data is less than the destination mileage length, and obtain the first judgment result;
  • the return subunit is used to execute the next time driving state flag determination subunit when the first judgment result is yes, and update the working condition prediction data.
  • the energy consumption prediction model establishment module specifically includes:
  • the energy consumption prediction initial model training unit is used to train the driving characteristic parameters of the travel segment data and the energy consumption data by using a K-fold cross-validation method and an extreme gradient boosting algorithm to obtain an initial energy consumption prediction model;
  • the optimization unit is used to optimize the hyperparameters of the initial energy consumption prediction model by using a grid search method to obtain an energy consumption prediction model.
  • the present invention discloses the following technical effects:
  • the invention provides a method and system for predicting energy consumption of electric vehicles.
  • the method includes: obtaining historical driving data of the electric vehicle; segmenting the historical driving data to obtain travel segment data and dynamic segment data; the travel segment data includes historical travel data of the electric vehicle during driving, and the dynamic segment data includes Historical driving data of electric vehicles during constant speed driving or acceleration driving; use dynamic segment data and Markov-Monte Carlo method to predict the working conditions of electric vehicles to obtain the working conditions prediction data of electric vehicles; obtain travel distance
  • the driving characteristic parameters and energy consumption data of the segment data; the driving characteristic parameters of the itinerary segment data are taken as input, and the energy consumption data is used as the output, and the energy consumption prediction model is established by the machine learning method; the driving characteristic parameters of the working condition prediction data are obtained;
  • the driving characteristic parameters of the condition prediction data are input to the established energy consumption prediction model to obtain the energy consumption prediction value.
  • the invention extracts the driving characteristics of electric vehicles based on historical driving data.
  • energy consumption prediction is performed, the future driving conditions of the vehicle are first predicted based on the current state of the vehicle, and the prediction of the future driving conditions of the vehicle is integrated into the process of energy consumption prediction.
  • machine learning experience learning and iterative optimization can extract and fit the complex operating conditions and energy consumption based on a large number of vehicle historical driving data as training samples.
  • FIG. 1 is a flowchart of an electric vehicle energy consumption prediction method provided by an embodiment of the present invention
  • Fig. 2 is a schematic diagram of data division of driving segments provided by an embodiment of the present invention.
  • Fig. 3 is a schematic diagram of a working condition prediction process provided by an embodiment of the present invention.
  • Fig. 4 is a system diagram of an electric vehicle energy consumption prediction system provided by an embodiment of the present invention.
  • the purpose of the present invention is to provide an electric vehicle energy consumption prediction method and system, which integrates the prediction of future driving conditions of the vehicle into the process of energy consumption prediction, and at the same time considers the influence of factors such as the driving environment and the driving behavior of the driver to improve the electric vehicle Accuracy of energy consumption forecast.
  • FIG. 1 is a flowchart of an electric vehicle energy consumption prediction method provided by an embodiment of the present invention.
  • the electric vehicle energy consumption prediction method includes:
  • Step 101 Obtain historical driving data of an electric vehicle.
  • the data used in this embodiment are all data generated during the actual driving of an electric vehicle (hereinafter referred to as a vehicle), and the data items include: time, mileage, speed, latitude and longitude, voltage and current, etc.
  • the continuous historical actual driving data of the vehicle is first based on the year label, the month label and the day label. Divide the data into fragments, and then delete the data whose mileage is over 600 kilometers or less than 1 kilometer in a day.
  • Use machine learning methods such as outlier detection to detect a large number of driving segments with continuous missing or abnormal data and delete them to obtain effective historical driving data.
  • Step 102 Perform segmentation processing on historical driving data to obtain travel segment data and dynamic segment data; travel segment data includes historical travel data of the electric vehicle during driving, and dynamic segment data includes electric vehicle driving at a constant speed or acceleration Historical driving data in the process.
  • Step 102 specifically includes: dividing the valid historical driving data obtained in step 101 into three levels of driving fragment data based on the speed characteristics and acceleration characteristics, including trip fragment data (trip_frag), micro trip fragment data (micro_frag), and dynamic fragment data ( kinematic_frag).
  • Table 1 lists the definitions of various driving segment data. The data division process of driving fragments is shown in Figure 2.
  • a typical trip_frag contains several micro_frags with different driving characteristics. According to the kinematic_frag division rules listed in Table 2, each micro_frag can be further divided into several kinematic_frags connected end to end.
  • the divided driving segment data is stored in the corresponding segment record tables trip_frag vin , micro_frag vin and kinematic_frag vin .
  • the driving segment characteristic parameters of the driving segment data include: vehicle number (vin), segment start time (start_time), segment end time (end_time), segment start mileage (start_range), segment end mileage (end_range), fragment type (frag_type) and vehicle state (vehicle_state) and other basic feature parameters of fragments.
  • the driving segment characteristic parameters of each driving segment data are extracted and recorded in the fragment statistical table frag_rec. Each record in the fragment statistical table frag_rec corresponds to a piece of driving segment data.
  • trip_frag vin (trip_frag 1 ,trip_frag 2 ,...,trip_frag nt ) T
  • micro_frag vin (micro_frag 1 ,micro_frag 2 ,...,micro_frag nm ) T
  • kinematic_frag vin (kinematic_frag 1 ,kinematic_frag 2 ,...,kinematic_frag nk ) T
  • frag_rec (vin,start_time,end_time,start_range,...,frag_type,vehicle_state) where nt, nm and nk are respectively the number of travel fragment data in the fragment record table trip_frag vin , micro_frag vin and kinematic_frag vin;
  • trip_frag 1 ,trip_frag 2 ,...,trip_frag nt represents the specific travel fragment data in the segment record table trip_frag vin obtained by dividing historical travel data, which is defined as trip segment data;
  • micro_frag 1 , micro_frag 2 ,..., micro_frag nm represents the specific driving fragment data in the fragment record table micro_frag vin obtained by dividing the historical driving data, which is defined as micro-travel fragment data;
  • kinematic_frag 1 ,kinematic_frag 2 ,...,kinematic_frag nk represents specific driving fragment data in the fragment record table kinematic_frag vin obtained by dividing historical driving data, which is defined as dynamic fragment data.
  • Step 103 Use the dynamic segment data and the Markov-Monte Carlo method to predict the operating conditions of the electric vehicle to obtain the operating condition prediction data of the electric vehicle.
  • the energy consumption of a vehicle is closely related to process parameters such as speed and acceleration during driving. Therefore, this embodiment first predicts the future operating conditions of the vehicle.
  • the speed change process of the vehicle is a process with no aftereffect of Markov property, which can be modeled and fitted through a Markov chain. Therefore, this embodiment uses a Markov Monte Carlo model to predict future driving conditions.
  • Step 103 specifically includes:
  • a driving state flag is added to frag_rec according to the average speed of each kinematic_frag.
  • numbers 1-9 are used to mark kinematic_frags with different average speeds.
  • the specific average speed of kinematic_frag and driving The corresponding relation of the status flags is shown in Table 3.
  • the driving state transition probability matrix of the electric vehicle is calculated. Specifically:
  • p ij represents the transition probability
  • N ij represents the number of events in which the vehicle transfers from the driving state mark i to the driving state mark j
  • i, j ⁇ [1, 9].
  • the probability of the vehicle transitioning from one driving state to another driving state is calculated, that is, electric
  • the transition probability of the traveling state of the car from the traveling state flag i to the traveling state flag j.
  • the transition probability between all driving state markers is used to determine the driving state transition probability matrix of the electric vehicle. Based on formula Calculate the transition probabilities between all driving state markers and fill them into the corresponding positions of the transition matrix to obtain a state transition probability matrix (TPM), which is used to characterize the historical driving characteristics of the vehicle. Use TPM, the current driving state of the vehicle, the current speed of the vehicle and the remaining mileage to reach the destination to predict the future operating conditions of the vehicle.
  • TPM state transition probability matrix
  • the Monte Carlo simulation method and the driving state transition probability matrix are used to determine the driving state flag of the electric vehicle at the next moment.
  • Vehicle condition prediction is a cyclic and iterative random process. In each cycle, a random number s is generated in the interval (0,1) each time based on the Monte Carlo simulation method. When s meets the following conditions At the time, select l as the vehicle's next running state mark.
  • P i1j represents the transition probability of the vehicle's current driving state mark i1 being transferred to the driving state mark j; i1 is the vehicle's current driving state mark, and l is the selected vehicle's next time driving state mark.
  • the predicted driving condition data is spliced with the current driving condition to obtain the operating condition prediction data of the electric vehicle.
  • the data of the speed column in the kinematic_frag corresponding to the kinematic_frag vin found in the kinematic_frag will be spliced to the end of the current driving condition of the vehicle.
  • 10 operating condition predictions are made for the current driving state of the vehicle to cover various driving possibilities that the vehicle may appear.
  • the working condition prediction data obtained in this step will be stored in the working condition prediction record table DC vin , and each DC n in the working condition prediction record table DC vin stores a predicted speed-time curve.
  • DC vin (DC 1 ,DC 2 ,...,DC n ,...,DC 10 ) T
  • DC n is the nth working condition prediction, which is composed of speed points in the order of several time dimensions; m is the total number of speed points in the working condition prediction; v 1 ,v 2 ,...,v m is the working condition prediction Speed data in.
  • the vehicle condition prediction process is the iterative splicing process of selecting the appropriate kinematic_frag, so every time a kinematic_frag is spliced to the end of the current driving condition, the current driving condition is updated to obtain the updated current driving condition and the updated current The final speed of the driving condition is updated to the final speed of the newly spliced kinematic_frag at the same time, and the mileage of kinematic_frag should be added to the length of the vehicle driving condition to obtain the predicted driving condition length.
  • the mileage length of the working condition prediction data is the length of the predicted driving condition.
  • the operating condition prediction process in step 103 will loop iteratively until the mileage of the operating condition prediction data is equal to the mileage of the vehicle to the destination.
  • An example of the working condition prediction process is shown in Figure 3, which includes the speed curve of the real working condition and the speed curve of the five working condition predictions.
  • Step 104 Acquire driving characteristic parameters and energy consumption data of the trip segment data.
  • the feature parameter extraction of the trip segment data is performed on the trip_frag vin obtained by segmenting the original historical traveling data in step 102, and the feature parameters of the trip segment data include traveling feature parameters and energy consumption data (EC).
  • the driving characteristic parameters selected in this embodiment include: driving time S, driving distance M, technical speed Acceleration 95% quantile a 0.95 , deceleration 5% quantile a 0.05 , average temperature And One-Hot code for weekdays/weekends (including holidays) and morning peak/evening peak/off-peak hours. Taking the driving characteristic parameters of the travel segment data as input and the energy consumption data of the travel segment data as output, the energy consumption prediction model is established.
  • One-Hot code for working days/weekends (including holidays) and morning peak/evening peak/non-peak hours includes morning peak 7:00-9:00 (MR workday ) on weekdays, and evening peak on weekdays 17 :00-19:00 (ER workday ), off-peak workday (NR workday ), weekend (including holidays) morning peak 9:00-11:00 (MR weekend ), weekend (including holidays) evening peak 15:00- 17:00 (ER weekend ), weekends (including holidays) off-peak (NR weekend ).
  • the calculation method of the above driving characteristic parameters and energy consumption data is as follows:
  • a 0.05 ⁇ a i'
  • a i' ⁇ 0 ⁇ 5% quantile (m/s 2 )(i' 1, 2,...,k-1)
  • S represents the driving time in seconds (s); k is the time length of the travel segment data or the working condition prediction data; t i'+1 represents the i'+1 time of the travel segment data or the working condition prediction data; T i 'denotes the i-th stroke fragment data or condition data predicted' time; t i '+ 1 -t i ' denotes the time difference in seconds; V i 'for the vehicle i' speed time, in km / h; U i 'and the I i' respectively, in the vehicle i 'battery voltage and current timing units are volt (V) and amperage (a); M represents a travel distance units in kilometers (km); Represents technology speed in kilometers per hour is (km / h); S d represents removed idling state travel time, in seconds (s); a i 'denotes the vehicle i' the time of acceleration values, in m / s 2 ; a 0.95 represents the 95% quantile of acceleration
  • the acceleration values a i'greater than 0 are arranged in ascending order, and then the 95% quantile is taken from the smallest to the largest.
  • a i' >0 ⁇ represents the collection of acceleration values a i'greater than 0;
  • a 0.05 represents the 5% quantile of deceleration, the unit is m/s 2 , specifically it will be less than 0 acceleration values a i 'larger in ascending order, and then small to large acceleration values taken corresponding to 5% quantile;
  • a i ' ⁇ 0 ⁇ indicates the value of the acceleration is less than 0 a i' of Set;
  • EC represents energy consumption data, the unit is kWh.
  • driving time and driving distance reflect the overall energy demand of the vehicle; speed and acceleration parameters characterize the state of the vehicle and the driving behavior of the driver, and reflect the depth of discharge of the power battery during driving; average temperature It will affect the performance of the power battery and the intensity of the use of auxiliary equipment; weekdays/weekends (including holidays) and the one-hot code for morning peak/evening peak/off-peak hours reflect the traffic conditions of the vehicle.
  • the above selected driving characteristic parameters comprehensively cover the factors that have an important influence on vehicle energy consumption.
  • trip_frag_feature vin (trip_frag_feature 1 ,...,trip_frag_feature q ,...,trip_frag_feature nt ) T
  • trip_frag_feature vin stores the trip fragment data extracted from all dynamic fragment data in trip_frag vin .
  • the feature parameters include driving time S i' , driving distance M i' , and technical speed Acceleration 95% quantile a 0.95i' , deceleration 5% quantile a 0.05i' , average temperature
  • One-Hot code and energy consumption data EC reali' for weekdays/weekends (including holidays) and morning peaks/evening peaks/off peak hours, weekdays/weekends (including holidays) and morning peaks/evening peaks /One-Hot code during off-peak hours includes morning peak hours on weekdays 7:00-9:00 (MR workdayi' ), evening peak hours on weekdays 17:00-19:00 (ER workdayi' ), work Day off peak (NR workdayi' ), weekends (including holidays) morning peak 9:00-11:00 (MR weekendi' ), weekends (including holidays) evening peak 15:00-17:00 (ER weekendi' ), weekends (including holidays) off-peak (NR
  • Step 105 Taking the driving characteristic parameters of the trip segment data as input and the energy consumption data as output, and using the machine learning method to establish an energy consumption prediction model.
  • an extreme gradient boosting algorithm eXtreme Gradient Boosting, XGBoost
  • XGBoost extreme gradient boosting algorithm
  • Step 105 specifically includes:
  • the K-fold cross-validation method and the extreme gradient boosting algorithm are used to train the driving characteristic parameters and energy consumption data of the trip segment data to obtain the initial energy consumption prediction model.
  • a model optimization framework combining KFold cross-validation method and GridSearch method is used to optimize the parameters of the initial energy consumption prediction model.
  • the 10-fold method is used to train the prediction model.
  • trip_frag_feature vin_KFold ⁇ trip_frag_feature vin_1 ,...,trip_frag_feature vin_q1 ,...,trip_frag_feature vin_10 ⁇
  • trip_frag_feature vin_KFold represents the total set of training samples trip_frag_feature vin
  • trip_frag_feature vin_1 ,...,trip_frag_feature vin_q1 ,...,trip_frag_feature vin_10 represents 10 smaller sets
  • trip_frag_q1 vin_q1 represents the 1q1th set 10].
  • the grid search method is used to optimize the hyperparameters of the initial energy consumption prediction model, and the energy consumption prediction model is obtained.
  • the grid search method is used to optimize the XGBoost hyperparameters.
  • XGBoost hyperparameters refer to the parameters that need to be adjusted in the initial energy consumption prediction model (XGBoost model), including: number of trees (n_estimators), maximum tree Depth (max_depth), learning rate (learningrate), sampling rate (subsample), etc.
  • RMSE Root Mean Square Error
  • MAPE Mean Absolute Percentage Error
  • N is the number of samples used to test the accuracy of the energy consumption prediction model during the training process of the energy consumption prediction model
  • p represents the number of samples
  • Is the predicted value of the energy consumption prediction model
  • y p is the true value.
  • Step 106 Obtain driving characteristic parameters of the operating condition prediction data.
  • the feature parameter extraction is performed on the DC vin obtained by the prediction of the future driving condition in step 103, and the feature parameters of the DC vin include driving feature parameters and energy consumption data.
  • the driving characteristic parameters of the extracted working condition prediction data include: driving time S, driving distance M, technical speed Acceleration 95% quantile a 0.95 , deceleration 5% quantile a 0.05 , average temperature And One-Hotcode for weekdays/weekends (including holidays) and morning peak/evening peak/off-peak hours.
  • the calculation method of the driving characteristic parameter is the same as the calculation method of the driving characteristic parameter of the trip segment data in step 104.
  • the feature parameters extracted by DC vin will be stored in DC_feature vin , and DC_feature vin will preset the predicted energy consumption EC pred to 0.
  • DC_feature vin (DC_feature 1 ,...,DC_feature q' ,...,DC_feature 10 ) T
  • DC_feature vin characteristic parameters stored in the current position of the vehicle on the predicted future driving conditions 10 times the extracted DC vin, each operating mode DC vin prediction data corresponding to one line in DC_feature vin.
  • DC_feature q 'DC_feature vin to the first q' th prediction condition extracted feature parameters.
  • Step 107 Input the driving characteristic parameters of the working condition prediction data into the established energy consumption prediction model to obtain the energy consumption prediction value.
  • the driving characteristic parameters obtained in step 106 include S, M, a 0.95 , a 0.05 , MR workday , ER workday , NR workday , MR weekend , ER weekend and NR weekend are input into the energy consumption prediction model to predict future energy consumption. For each operating condition prediction, an energy consumption prediction value is obtained, and 10 operating conditions The mean value of the predicted energy consumption corresponding to the prediction As the final predicted value of vehicle energy consumption in the future.
  • N' represents the total number of working condition predictions
  • n' represents the number of working condition predictions
  • EC predn' represents the predicted value of energy consumption corresponding to the n'th working condition prediction.
  • FIG. 4 is a system diagram of the electric vehicle energy consumption prediction system provided by an embodiment of the present invention.
  • the electric vehicle energy consumption prediction system includes:
  • the obtaining module 201 is used to obtain historical driving data of the electric vehicle.
  • the segmentation processing module 202 is used to segment the historical driving data to obtain travel segment data and dynamic segment data; the travel segment data includes the historical travel data of the electric vehicle during the driving process, and the dynamic segment data includes the electric vehicle at a constant speed Historical driving data during driving or acceleration driving.
  • the working condition prediction module 203 is used to predict the working condition of the electric vehicle by using the dynamic segment data and the Markov-Monte Carlo method to obtain the working condition prediction data of the electric vehicle.
  • the working condition prediction module 203 specifically includes:
  • the driving state mark adding unit is used to add driving state marks to different driving states in the dynamic fragment data by using the average speed in the dynamic fragment data.
  • the driving state transition probability matrix calculation unit is used to calculate the driving state transition probability matrix of the electric vehicle by using the time sequence of the dynamic segment data and the driving state mark.
  • the driving state transition probability matrix calculation unit specifically includes:
  • Transition probability calculation subunit used to use the time sequence of dynamic segment data, according to the formula Calculate the transition probability of the driving state of the electric vehicle from the driving state mark i to the driving state mark j; in the formula, p ij represents the transition probability; N ij represents the number of events transitioning from the driving state mark i to the driving state mark j.
  • the driving state transition probability matrix calculation subunit is used to determine the driving state transition probability matrix of the electric vehicle by using the transition probabilities between all driving state flags.
  • the working condition prediction unit is used to predict the working condition of the electric vehicle by using the Monte Carlo simulation method, the driving state transition probability matrix and the driving state mark to obtain the working condition prediction data of the electric vehicle.
  • the working condition prediction unit specifically includes:
  • the next time driving state mark determination subunit is used to determine the next time driving state mark of the electric vehicle by using the Monte Carlo simulation method and the driving state transition probability matrix.
  • the predictive driving condition data determining subunit is used to determine the historical driving data with the same marking of the next moment in the dynamic segment data to obtain the predictive driving condition data.
  • the first acquisition subunit is used to acquire the current driving conditions and destination mileage of the electric vehicle.
  • the splicing subunit is used to splice the predicted driving condition data with the current driving condition in chronological order to obtain the operating condition prediction data of the electric vehicle.
  • the second acquisition subunit is used to acquire the mileage length of the working condition prediction data.
  • the first judgment subunit is used to judge whether the mileage length of the working condition prediction data is less than the destination mileage length, and obtain the first judgment result.
  • the return subunit is used to execute the next time driving state flag determination subunit when the first judgment result is yes, and update the working condition prediction data.
  • the first acquisition module 204 is configured to acquire driving characteristic parameters and energy consumption data of the travel segment data.
  • the energy consumption prediction model establishment module 205 is configured to take the driving characteristic parameters of the travel segment data as input and the energy consumption data as output, and establish an energy consumption prediction model by using a machine learning method.
  • the energy consumption prediction model establishment module 205 specifically includes:
  • the energy consumption prediction initial model training unit is used to train the driving characteristic parameters and energy consumption data of the travel segment data using the K-fold cross-validation method and the extreme gradient boosting algorithm to obtain the initial energy consumption prediction model.
  • the optimization unit is used to optimize the hyperparameters of the initial energy consumption prediction model by using the grid search method to obtain the energy consumption prediction model.
  • the second acquisition module 206 is used to acquire the driving characteristic parameters of the working condition prediction data.
  • the energy consumption prediction module 207 is used to input the driving characteristic parameters of the working condition prediction data into the established energy consumption prediction model to obtain the energy consumption prediction value.
  • the electric vehicle energy consumption prediction method and system of the present invention greatly improve the energy consumption prediction accuracy of the electric vehicle under actual driving conditions.
  • the present invention extracts driving characteristics based on the historical driving data of electric vehicles, and first predicts the future driving conditions of the vehicle based on the current state of the vehicle during energy consumption prediction , And fully consider the impact of driving environment factors, including environmental temperature, traffic conditions, and driver driving behavior, so as to ensure that the energy consumption prediction model has good accuracy in the actual application environment, and it also improves the actual driving conditions of electric vehicles. Accuracy of energy consumption prediction under.
  • the empirical learning and iterative optimization of machine learning can extract and fit the nonlinear coupling relationship between complex working conditions and energy consumption on the basis of a large number of historical vehicle driving data as training samples. Iteratively improve the accuracy, and finally realize the high-precision prediction of electric vehicles under actual working conditions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Medical Informatics (AREA)
  • Economics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Strategic Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Human Resources & Organizations (AREA)
  • Computer Hardware Design (AREA)
  • Quality & Reliability (AREA)
  • Development Economics (AREA)
  • Geometry (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Electric Propulsion And Braking For Vehicles (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种电动汽车能耗预测方法及系统,涉及汽车领域。该方法包括:获取电动汽车的历史行驶数据(101);对所述历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据(102);利用动力学片段数据和马尔可夫-蒙特卡洛方法对电动汽车进行工况预测,得到电动汽车的工况预测数据(103);获取所述行程片段数据的行驶特征参数和能耗数据(104);将行程片段数据的行驶特征参数作为输入,能耗数据作为输出,利用机器学习方法建立能耗预测模型(105);获取工况预测数据的行驶特征参数(106);将工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值(107)。该方法基于电动汽车的历史行驶数据,提取其行驶特征,在进行能耗预测时首先基于车辆当前的状态预测车辆的未来行驶工况,将车辆未来行驶工况预测融合到能耗预测的过程中,提高了电动汽车在实际行驶工况下的能耗预测精度。

Description

一种电动汽车能耗预测方法及系统
本申请要求于2020年06月05日提交中国专利局、申请号为202010505376.3、发明名称为“一种电动汽车能耗预测方法及系统”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本发明涉及汽车领域,特别是涉及一种电动汽车能耗预测方法及系统。
背景技术
近年来,交通电气化逐渐成为实现节能减排和提高能量效率的有效措施。截至2019年底,中国新能源汽车数量已达381万辆。然而,受限于当前的动力电池技术的发展,电动汽车续驶里程较短、充电时间较长,以及基础设施不足等缺陷大大限制了电动汽车的推广应用。考虑到电动汽车在现实应用中的局限性,电动汽车在实际行驶工况下的能耗已经成为电动汽车用户、汽车制造厂商和政府十分关注的关键性能指标,其对电动汽车交通运输系统的能源效率、环境效益和经济效益有着重要的影响。准确预测电动汽车的能耗对缓解驾驶员的行驶里程焦虑至关重要,并能够为电池容量优化设计、绿色路线规划以及充电基础设施的运营管理提供有力支持。因此,实际行驶工况下的电动汽车能耗准确估算与预测的需求日益增长。
现有的电动汽车能耗预测技术多采用基于车辆动力学模型的方法。在该方法中,车辆纵向动力学模型(longitudinal dynamics model,LDM)和车辆比功率模型(vehicle specific power,VSP)通常被用于车辆能耗估计,在应用该方法进行能耗估计之前需要获取或假设大量车辆参数包括车辆迎风面积、质量和滚动阻力系数等,而在实际应用中很难提前精确获取这些参数,尤其是在应用于大量车辆如物流车群的情况时,获取每辆车的详细参数几乎不具有可行性,同时车辆动力学模型方法往往通过固定的工况如NEDC(New European Driving Cycle,新欧洲驾驶周期)模拟车辆工况,但是车辆的实际行驶工况十分复杂,基于车辆动力学模型的方法无法考虑动态的车辆工况的影响,因此预测精度较差。
发明内容
本发明的目的是提供一种电动汽车能耗预测方法及系统,将车辆未来行驶工况预测融合到能耗预测的过程中,同时考虑行驶环境因素的影响,提高电动汽车能耗预测的精度。
为实现上述目的,本发明提供了如下方案:
一种电动汽车能耗预测方法,包括:
获取电动汽车的历史行驶数据;
对所述历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据;所述行程片段数据包括所述电动汽车在行驶过程中的历史行驶数据,所述动力学片段数据包括所述电动汽车在恒定速度行驶或加速度行驶的过程中的历史行驶数据;
利用所述动力学片段数据和马尔可夫-蒙特卡洛方法对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据;
获取所述行程片段数据的行驶特征参数和能耗数据;
将所述行程片段数据的行驶特征参数作为输入,所述能耗数据作为输出,利用机器学习方法建立能耗预测模型;
获取所述工况预测数据的行驶特征参数;
将所述工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值。
可选的,所述利用所述动力学片段数据和马尔可夫-蒙特卡洛方法对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据,具体包括:
利用所述动力学片段数据中的平均速度,对所述动力学片段数据中不同的行驶状态添加行驶状态标记;
利用所述动力学片段数据的时间顺序和所述行驶状态标记,计算得到所述电动汽车的行驶状态转移概率矩阵;
利用蒙特卡洛模拟方法、所述行驶状态转移概率矩阵和所述行驶状态标记对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据。
可选的,所述利用所述动力学片段数据的时间顺序和所述行驶状态标记,计算得到所述电动汽车的行驶状态转移概率矩阵,具体包括:
利用所述动力学片段数据的时间顺序,根据公式
Figure PCTCN2021098285-appb-000001
计算所述电动汽车的行驶状态从行驶状态标记i转移至行驶状态标记j的转移概率;式中,p ij表示转移概率;N ij表示从行驶状态标记i转移至行驶状态标记j的事件数;
利用所有所述行驶状态标记之间的转移概率确定所述电动汽车的行驶状态转移概率矩阵。
可选的,所述利用蒙特卡洛模拟方法、所述行驶状态转移概率矩阵和所述行驶状态标记对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据,具体包括:
利用蒙特卡洛模拟方法和所述行驶状态转移概率矩阵确定所述电动汽车的下一时刻行驶状态标记;
确定所述动力学片段数据中与所述下一时刻行驶状态标记相同的历史行驶数据,得到预测行驶工况数据;
获取所述电动汽车的当前行驶工况和目的地里程长度;
按照时间顺序将所述预测行驶工况数据与所述当前行驶工况进行拼接,得到所述电动汽车的工况预测数据;
获取所述工况预测数据的里程长度;
判断所述工况预测数据的里程长度是否小于所述目的地里程长度,得到第一判断结果;
若所述第一判断结果为是,则返回“利用蒙特卡洛模拟方法和所述行驶状态转移概率矩阵确定所述电动汽车的下一时刻行驶状态标记”,更新所述工况预测数据。
可选的,所述将所述行程片段数据的行驶特征参数作为输入,所述能耗数据作为输出,利用机器学习方法建立能耗预测模型,具体包括:
采用K折交叉验证方法和极端梯度提升算法对所述行程片段数据的行驶特征参数和所述能耗数据进行训练,得到能耗预测初始模型;
采用网格搜索方法对所述能耗预测初始模型的超参数进行优化,得到 能耗预测模型。
一种电动汽车能耗预测系统,包括:
获取模块,用于获取电动汽车的历史行驶数据;
分割处理模块,用于对所述历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据;所述行程片段数据包括所述电动汽车在行驶过程中的历史行驶数据,所述动力学片段数据包括所述电动汽车在恒定速度行驶或加速度行驶的过程中的历史行驶数据;
工况预测模块,用于利用所述动力学片段数据和马尔可夫-蒙特卡洛方法对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据;
第一获取模块,用于获取所述行程片段数据的行驶特征参数和能耗数据;
能耗预测模型建立模块,用于将所述行程片段数据的行驶特征参数作为输入,所述能耗数据作为输出,利用机器学习方法建立能耗预测模型;
第二获取模块,用于获取所述工况预测数据的行驶特征参数;
能耗预测模块,用于将所述工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值。
可选的,所述工况预测模块,具体包括:
行驶状态标记添加单元,用于利用所述动力学片段数据中的平均速度,对所述动力学片段数据中不同的行驶状态添加行驶状态标记;
行驶状态转移概率矩阵计算单元,用于利用所述动力学片段数据的时间顺序和所述行驶状态标记,计算得到所述电动汽车的行驶状态转移概率矩阵;
工况预测单元,用于利用蒙特卡洛模拟方法、所述行驶状态转移概率矩阵和所述行驶状态标记对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据。
可选的,所述行驶状态转移概率矩阵计算单元,具体包括:
转移概率计算子单元,用于利用所述动力学片段数据的时间顺序,根据公式
Figure PCTCN2021098285-appb-000002
计算所述电动汽车的行驶状态从行驶状态标记i转移 至行驶状态标记j的转移概率;式中,p ij表示转移概率;N ij表示从行驶状态标记i转移至行驶状态标记j的事件数;
行驶状态转移概率矩阵计算子单元,用于利用所有所述行驶状态标记之间的转移概率确定所述电动汽车的行驶状态转移概率矩阵。
可选的,所述工况预测单元,具体包括:
下一时刻行驶状态标记确定子单元,用于利用蒙特卡洛模拟方法和所述行驶状态转移概率矩阵确定所述电动汽车的下一时刻行驶状态标记;
预测行驶工况数据确定子单元,用于确定所述动力学片段数据中与所述下一时刻行驶状态标记相同的历史行驶数据,得到预测行驶工况数据;
第一获取子单元,用于获取所述电动汽车的当前行驶工况和目的地里程长度;
拼接子单元,用于按照时间顺序将所述预测行驶工况数据与所述当前行驶工况进行拼接,得到所述电动汽车的工况预测数据;
第二获取子单元,用于获取所述工况预测数据的里程长度;
第一判断子单元,用于判断所述工况预测数据的里程长度是否小于所述目的地里程长度,得到第一判断结果;
返回子单元,用于当所述第一判断结果为是时,执行下一时刻行驶状态标记确定子单元,更新所述工况预测数据。
可选的,所述能耗预测模型建立模块,具体包括:
能耗预测初始模型训练单元,用于采用K折交叉验证方法和极端梯度提升算法对所述行程片段数据的行驶特征参数和所述能耗数据进行训练,得到能耗预测初始模型;
优化单元,用于采用网格搜索方法对所述能耗预测初始模型的超参数进行优化,得到能耗预测模型。
根据本发明提供的具体实施例,本发明公开了以下技术效果:
本发明提供了一种电动汽车能耗预测方法及系统。该方法包括:获取电动汽车的历史行驶数据;对历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据;行程片段数据包括电动汽车在行驶过程中的历史行驶数据,动力学片段数据包括电动汽车在恒定速度行驶或加速度行驶的 过程中的历史行驶数据;利用动力学片段数据和马尔可夫-蒙特卡洛方法对电动汽车进行工况预测,得到电动汽车的工况预测数据;获取行程片段数据的行驶特征参数和能耗数据;将行程片段数据的行驶特征参数作为输入,能耗数据作为输出,利用机器学习方法建立能耗预测模型;获取工况预测数据的行驶特征参数;将工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值。本发明基于电动汽车的历史行驶数据,提取其行驶特征,在进行能耗预测时首先基于车辆当前的状态预测车辆的未来行驶工况,将车辆未来行驶工况预测融合到能耗预测的过程中,大大提高了电动汽车在实际行驶工况下的能耗预测精度;机器学习的经验学习和迭代优化能够在大量车辆历史行驶数据作为训练样本的基础上提取、拟合复杂工况与能耗之间的非线性耦合关系,并随着车辆不断产生的行程片段进行迭代提高精度,最终实现实际工况下的电动汽车高精度预测。
说明书附图
为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1为本发明实施例所提供的电动汽车能耗预测方法的流程图;
图2为本发明实施例所提供的行驶片段数据划分示意图;
图3为本发明实施例所提供的工况预测过程示意图;
图4为本发明实施例所提供的电动汽车能耗预测系统的系统图。
具体实施方式
下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。
本发明的目的是提供一种电动汽车能耗预测方法及系统,将车辆未来行驶工况预测融合到能耗预测的过程中,同时考虑行驶环境和驾驶员驾驶行为等因素的影响,提高电动汽车能耗预测的精度。
为使本发明的上述目的、特征和优点能够更加明显易懂,下面结合附图和具体实施方式对本发明作进一步详细的说明。
本实施例提供一种电动汽车能耗预测方法,图1为本发明实施例所提供的电动汽车能耗预测方法的流程图,参见图1,电动汽车能耗预测方法包括:
步骤101,获取电动汽车的历史行驶数据。本实施例所使用的数据均为电动汽车(以下简称车辆)的实际行驶过程中产生的数据,数据项包括:时间、里程、速度、经纬度、电压和电流等。
考虑到在车辆复杂工况下数据采集传感器和数据无线传输装置在数据采集和传输过程中会产生数据丢失或异常等情况,首先根据年标签、月标签和日标签将车辆连续的历史实际行驶数据分割为片段数据,然后将一天内行驶里程在600公里以上或1公里以下的数据删除。通过离群点检测等机器学习方法检测大量、连续缺失或异常数据的行驶片段并删除,得到有效的历史行驶数据。
步骤102,对历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据;行程片段数据包括电动汽车在行驶过程中的历史行驶数据,动力学片段数据包括电动汽车在恒定速度行驶或加速度行驶的过程中的历史行驶数据。
步骤102具体包括:基于速度特征和加速度特征将步骤101所得的有效的历史行驶数据划分为3级行驶片段数据,包括行程片段数据(trip_frag)、微行程片段数据(micro_frag)和动力学片段数据(kinematic_frag)。表1列出了各种行驶片段数据的定义。行驶片段数据划分过程如图2所示,一段典型的trip_frag包含若干行驶特征各异的micro_frag,根据表2所列出的kinematic_frag划分规则,每一个micro_frag可以进一步划分为若干首尾相连的kinematic_frag。
表1 3级行驶片段数据的定义
Figure PCTCN2021098285-appb-000003
表2 kinematic_frag划分规则
Figure PCTCN2021098285-appb-000004
划分好的行驶片段数据储存在相应的片段记录表trip_frag vin,micro_frag vin和kinematic_frag vin中。对于每一条行驶片段数据,行驶片段数据的行驶片段特征参数均包括:车辆编号(vin)、片段起始时间(start_time)、片段结束时间(end_time)、片段起始里程(start_range)、片段结束里程(end_range)、片段类型(frag_type)和车辆状态(vehicle_state)等片段基础特征参数。提取每条行驶片段数据的行驶片段特征参数,并记录在片段统计表frag_rec中,片段统计表frag_rec中每一条记录对应一条行驶片段数据。这些行驶片段数据将分别用于后续的工况预测、行程片段数据的行驶特征参数的提取和能耗数据的提取以及能耗预测模型的建立等过程。
trip_frag vin=(trip_frag 1,trip_frag 2,...,trip_frag nt) T
micro_frag vin=(micro_frag 1,micro_frag 2,...,micro_frag nm) T
kinematic_frag vin=(kinematic_frag 1,kinematic_frag 2,...,kinematic_frag nk) T
frag_rec=(vin,start_time,end_time,start_range,...,frag_type,vehicle_state)其中,nt,nm和nk分别为片段记录表trip_frag vin,micro_frag vin和kinematic_frag vin中行驶片段数据的数量;
trip_frag 1,trip_frag 2,...,trip_frag nt表示历史行驶数据划分得到的片段记录表trip_frag vin中具体的行驶片段数据,定义为行程片段数据;
micro_frag 1,micro_frag 2,...,micro_frag nm表示历史行驶数据划分得到的片段记录表micro_frag vin中具体的行驶片段数据,定义为微行程片段数据;
kinematic_frag 1,kinematic_frag 2,...,kinematic_frag nk表示历史行驶数据划分得 到的片段记录表kinematic_frag vin中具体的行驶片段数据,定义为动力学片段数据。
步骤103,利用动力学片段数据和马尔可夫-蒙特卡洛方法对电动汽车进行工况预测,得到电动汽车的工况预测数据。车辆的能耗与行驶过程中的速度和加速度等过程参数密切相关,因此本实施例首先对车辆未来的工况进行预测。车辆的速度变化过程是具有马尔可夫(Markov)性质的无后效性过程,可以通过马尔可夫链(Markov chain)进行建模拟合。所以本实施例使用马尔可夫-蒙特卡洛模型(Markov Monte Carlo model)进行未来行驶工况的预测。
步骤103具体包括:
利用动力学片段数据中的平均速度,对动力学片段数据中不同的行驶状态添加行驶状态标记。针对步骤102中历史行驶数据分割处理得到的kinematic_frag,根据每个kinematic_frag的平均速度在frag_rec中增加行驶状态标记,本实施例使用数字1-9标记不同平均速度的kinematic_frag,具体kinematic_frag的平均速度与行驶状态标记对应关系如表3所示。
表3 kinematic_frag的行驶状态标记
Figure PCTCN2021098285-appb-000005
利用动力学片段数据的时间顺序和行驶状态标记,计算得到电动汽车的行驶状态转移概率矩阵。具体包括:
利用动力学片段数据的时间顺序,根据下式计算电动汽车的行驶状态从行驶状态标记i转移至行驶状态标记j的转移概率:
Figure PCTCN2021098285-appb-000006
式中,p ij表示转移概率;N ij表示车辆从行驶状态标记i转移至行驶状态标记j的事件数;i,j∈[1,9]。基于kinematic_frag在时间维度上的排列顺序依次统计出车辆在各行驶状态标记之间的转移次数,进而计算车辆从一个行驶状态转移到另一个行驶状态(包括停留在相同行驶状态)的概率,即电动汽车的行驶状态从行驶状态标记i转移至行驶状态标记j的转移概率。
利用所有行驶状态标记之间的转移概率确定电动汽车的行驶状态转移概率矩阵。基于公式
Figure PCTCN2021098285-appb-000007
计算所有行驶状态标记之间的转移概率并填入转移矩阵对应的位置中,可以得到状态转移概率矩阵(transition probability matrix,TPM),状态转移概率矩阵用于表征车辆的历史行驶特征。利用TPM、车辆的当前行驶状态、车辆的当前速度和到达目的地的剩余里程进行车辆未来工况预测。
Figure PCTCN2021098285-appb-000008
利用蒙特卡洛模拟方法、行驶状态转移概率矩阵和行驶状态标记对电动汽车进行工况预测,得到电动汽车的工况预测数据。具体包括:
利用蒙特卡洛模拟方法和行驶状态转移概率矩阵确定电动汽车的下一时刻行驶状态标记。车辆工况预测是一个循环迭代的随机过程,在每次循环中,首先基于蒙特卡洛(Monte Carlo)模拟方法每次在(0,1]区间内生成一个随机数s。当s满足以下条件时,选定l作为车辆的下一时刻行驶状态标记。
Figure PCTCN2021098285-appb-000009
式中,P i1j表示车辆当前的行驶状态标记i1转移至行驶状态标记j的转移概率;i1为车辆当前的行驶状态标记,l为选定的车辆下一时刻行驶状态标记。
确定动力学片段数据中与下一时刻行驶状态标记相同的历史行驶数据,得到预测行驶工况数据。确定车辆的下一状态后,从frag_rec中行驶状态标记为l的kinematic_frag中选取合适的kinematic_frag,要求所选的kinematic_frag的初始速度与车辆当前行驶工况末速度的差值小于1km/h,并从kinematic_frag vin中找到对应的kinematic_frag作为预测行驶工况数据。
获取电动汽车的当前行驶工况和目的地里程长度。
按照时间顺序将预测行驶工况数据与当前行驶工况进行拼接,得到电动汽车的工况预测数据。将从kinematic_frag vin中找到对应的kinematic_frag中的速度一列的数据拼接到车辆当前行驶工况的末尾。为支持后续的能耗预测,在步骤103中针对车辆当前行驶状态进行10次工况预测以覆盖车辆可能出现的各种行驶可能性。该步骤得到的工况预测数据将存储在工况预测记录表DC vin中,工况预测记录表DC vin中的每个DC n存储预测的一条速度时间曲线。
DC vin=(DC 1,DC 2,...,DC n,...,DC 10) T
DC n=(v 1,v 2,...,v m)
其中,DC n为第n次工况预测,由若干时间维度顺序上的速度点组成;m为工况预测中速度点的总数;v 1,v 2,...,v m是工况预测中的速度数据。
更新车辆行驶状态、车辆行驶工况末速度以及车辆行驶工况长度。
车辆工况预测过程是选取合适的kinematic_frag迭代拼接的过程,因此每次向当前行驶工况末尾拼接一个kinematic_frag后都对当前行驶工况进行更新,得到更新后的当前行驶工况,更新后的当前行驶工况的末速度同时更新为新拼接的kinematic_frag的末速度,且要把kinematic_frag的行驶里程累加到车辆行驶工况长度中得到预测行驶工况长度。
获取工况预测数据的里程长度。工况预测数据的里程长度即预测行驶工况长度。
判断工况预测数据的里程长度是否小于目的地里程长度,得到第一判断结果。
若第一判断结果为是,则返回“利用蒙特卡洛模拟方法和行驶状态转移概率矩阵确定电动汽车的下一时刻行驶状态标记”,更新工况预测数据。
步骤103的工况预测过程将循环迭代直至工况预测数据的里程长度等于车辆至目的地的里程长度。工况预测过程示例如图3所示,图3包括真实工况的速度曲线和5次工况预测的速度曲线。
步骤104,获取行程片段数据的行驶特征参数和能耗数据。针对在步骤102中由原始历史行驶数据切分得到的trip_frag vin进行行程片段数据的特征参数提取,行程片段数据的特征参数包括行驶特征参数和能耗数据(EC)。
本实施例选取的行驶特征参数包括:行驶时长S、行驶距离M、技术速度
Figure PCTCN2021098285-appb-000010
加速度95%分位数a 0.95、减速度5%分位数a 0.05、平均温度
Figure PCTCN2021098285-appb-000011
以及工作日/周末(包括节假日)和早高峰/晚高峰/非高峰时段的独热编码(One-Hot code)。将行程片段数据的行驶特征参数作为输入,行程片段数据的能耗数据作为输出,建立能耗预测模型。工作日/周末(包括节假日)和早高峰/晚高峰/非高峰时段的独热编码(One-Hot code)包括工作日早高峰7:00-9:00(MR workday),工作日晚高峰17:00-19:00(ER workday),工作日非高峰(NR workday),周末(包括节假日)早高峰9:00-11:00(MR weekend),周末(包括节假日)晚高峰15:00-17:00(ER weekend),周末(包括节假日)非高峰(NR weekend)。上述行驶特征参数和能耗数据的计算方法如下:
Figure PCTCN2021098285-appb-000012
Figure PCTCN2021098285-appb-000013
Figure PCTCN2021098285-appb-000014
Figure PCTCN2021098285-appb-000015
a 0.95={a i'|a i'>0}95%分位数(m/s 2)(i'=1,2,...,k-1)
a 0.05={a i'|a i'<0}5%分位数(m/s 2)(i'=1,2,...,k-1)
Figure PCTCN2021098285-appb-000016
其中,S表示行驶时长,单位为秒(s);k为行程片段数据或工况预测数据的时长;t i'+1表示行程片段数据或工况预测数据的第i’+1个时刻;t i'表示行程片段数据或工况预测数据的第i’个时刻;t i'+1-t i'表示时间差,单位为秒;v i'为车辆在i’时刻的速度,单位为km/h;U i'和I i'分别为车辆在i’时刻的电池电压和电流,单位分别为伏特(V)和安培(A);M表示行驶距离,单位为千米(km);
Figure PCTCN2021098285-appb-000017
表示技术速度,单位为千米每小时(km/h);S d表示除去怠速状态的行驶时间,单位为秒(s);a i'表示车辆在i’时刻的加速度值,单位为m/s 2;a 0.95表示加速度95%分位数,单位为m/s 2,具体为将大于0的加速度值a i'按照从小到大的顺序排列,然后从小到大取95%分位数对应的加速度值;{a i'|a i'>0}表示大于0的加速度值a i'的集合;a 0.05表示减速度5%分位数,单位为m/s 2,具体为将小于0的加速度值a i'按照从小到大的顺序排列,然后从小到大取5%分位数对应的加速度值;{a i'|a i'<0}表示小于0的加速度值a i'的集合;EC表示能耗数据,单位为kWh。
在所选取的行驶特征参数中,行驶时长和行驶距离反映了车辆的总体能量需求;速度和加速度参数表征了车辆状态和驾驶员的驾驶行为,反映了行驶过程中的动力电池放电深度;平均温度会影响动力电池性能和辅助设备的使用强度;工作日/周末(包括节假日)和早高峰/晚高峰/非高峰时段的独热编码反映了车辆行驶的交通状况。上述选定的行驶特征参数全面涵盖了对车辆能耗具有重要影响的因素。由trip_frag vin提取的行程片段数据的特征参数将存储在trip_frag_feature vin中,trip_frag_feature vin包含 片段的真实能耗(Energyconsumption)数据EC real,EC real=EC。
trip_frag_feature vin=(trip_frag_feature 1,...,trip_frag_feature q,...,trip_frag_feature nt) T
Figure PCTCN2021098285-appb-000018
trip_frag_feature vin存放trip_frag vin中所有动力学片段数据提取的行程片段数据的特征参数包括行驶时长S i'、行驶距离M i'、技术速度
Figure PCTCN2021098285-appb-000019
加速度95%分位数a 0.95i'、减速度5%分位数a 0.05i'、平均温度
Figure PCTCN2021098285-appb-000020
工作日/周末(包括节假日)和早高峰/晚高峰/非高峰时段的独热编码(One-Hot code)以及能耗数据EC reali',工作日/周末(包括节假日)和早高峰/晚高峰/非高峰时段的独热编码(One-Hot code)包括工作日早高峰7:00-9:00(MR workdayi'),工作日晚高峰17:00-19:00(ER workdayi'),工作日非高峰(NR workdayi'),周末(包括节假日)早高峰9:00-11:00(MR weekendi'),周末(包括节假日)晚高峰15:00-17:00(ER weekendi'),周末(包括节假日)非高峰(NR weekendi'),trip_frag vin中的每一条动力学片段数据对应trip_frag_feature vin中的一行数据。其中trip_frag_feature q为trip_frag vin中第q个动力学片段数据提取的行程片段数据的特征参数。
步骤105,将行程片段数据的行驶特征参数作为输入,能耗数据作为输出,利用机器学习方法建立能耗预测模型。本实施例采用极端梯度提升算法(eXtreme GradientBoosting,XGBoost)挖掘步骤104中获取的行程片段数据的行驶特征参数和能耗数据之间的关联关系。步骤104中得到的trip_frag_feature vin将用于训练能耗预测模型和参数调优。
步骤105具体包括:
采用K折交叉验证方法和极端梯度提升算法对行程片段数据的行驶特征参数和能耗数据进行训练,得到能耗预测初始模型。在基于XGBoost的预测模型的训练过程中采用K折(KFold)交叉验证方法和网格搜索(GridSearch)方法相结合的模型优化框架对能耗预测初始模型进行参数优化。本实施例采用10折方法进行预测模型训练,首先将训练样本 trip_frag_feature vin分解成10个较小的集合:
trip_frag_feature vin_KFold={trip_frag_feature vin_1,...,trip_frag_feature vin_q1,...,trip_frag_feature vin_10}
其中,trip_frag_feature vin_KFold表示训练样本trip_frag_feature vin的总集,trip_frag_feature vin_1,...,trip_frag_feature vin_q1,...,trip_frag_feature vin_10表示10个较小的集合,trip_frag_feature vin_q1表示第q1个集合,q1∈[1,10]。
在每次训练时使用其中的9个集合作为训练数据对基于XGBoost的预测模型进行训练,并用剩余的1个集合对预测模型进行验证。依次将trip_frag_feature vin_1到trip_frag_feature vin_10作为测试样本进行预测模型训练,以10次循环计算的平均水平作为预测模型的最终评价。
采用网格搜索方法对能耗预测初始模型的超参数进行优化,得到能耗预测模型。在KFold迭代训练的基础上,采用网格搜索方法对XGBoost超参数进行优化,XGBoost超参数指能耗预测初始模型(XGBoost模型)中需要调整的参数,具体包括:树数量(n_estimators)、最大树深度(max_depth)、学习率(learningrate)和采样比例(subsample)等。以学习率(learningrate)为例说明参数优化步骤,依据经验预设XGBoost模型的学习率取值范围,采用10折方法对不同学习率下的XGBoost模型预测性能使用MAPE和RMSE进行评价,寻找预测性能达到最优对应的学习率,并以同样的方式测试XGBoost模型超参数的不同组合以确定最优超参数组合,并最终获取最优的能耗预测模型。
本实施例使用评价指标均方根误差(Root Mean Squared Error,RMSE)和相对百分误差(MeanAbsolutePercentageError,MAPE)作为能耗预测模型预测性能的评价指标。RMSE和MAPE的计算方法为:
Figure PCTCN2021098285-appb-000021
Figure PCTCN2021098285-appb-000022
式中,N为能耗预测模型训练过程中用于测试能耗预测模型精度的样本数,p表示样本数序号,
Figure PCTCN2021098285-appb-000023
为能耗预测模型的预测值,y p为真实值。RMSE和MAPE越小说明能耗预测模型的预测精度越高。
步骤106,获取工况预测数据的行驶特征参数。针对在步骤103中未来行驶工况预测得到的DC vin进行特征参数提取,DC vin的特征参数包括行驶特征参数和能耗数据。提取的工况预测数据的行驶特征参数包括:行驶时长S、行驶距离M、技术速度
Figure PCTCN2021098285-appb-000024
加速度95%分位数a 0.95、减速度5%分位数a 0.05、平均温度
Figure PCTCN2021098285-appb-000025
以及工作日/周末(包括节假日)和早高峰/晚高峰/非高峰时段的独热编码(One-Hotcode)。行驶特征参数的计算方法同步骤104中行程片段数据的行驶特征参数的计算方法。由DC vin提取的特征参数将存储在DC_feature vin中,DC_feature vin将预测能耗EC pred预设为0。
DC_feature vin=(DC_feature 1,...,DC_feature q',...,DC_feature 10) T
Figure PCTCN2021098285-appb-000026
DC_feature vin中存放车辆在当前位置对未来10次行驶工况预测的DC vin提取的特征参数,DC vin中的每一条工况预测对应DC_feature vin中的一行数据。其中DC_feature q'为DC_feature vin中第q’个工况预测提取的特征参数。
步骤107,将工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值。
将步骤106获取的行驶特征参数包括S、M、
Figure PCTCN2021098285-appb-000027
a 0.95、a 0.05
Figure PCTCN2021098285-appb-000028
MR workday、ER workday、NR workday、MR weekend、ER weekend和NR weekend输入到能耗预测模型中对未来能耗进行预测,对每个工况预测均获得一个能耗预测值,将10个工况预测对应的能耗预测值的均值
Figure PCTCN2021098285-appb-000029
作为最终的车辆未来能耗预测值。
Figure PCTCN2021098285-appb-000030
其中,N’表示工况预测的总数,n’表示工况预测的次数,EC predn'表示第n’次工况预测对应的能耗预测值。
本实施例提供一种电动汽车能耗预测系统,图4为本发明实施例所提供的电动汽车能耗预测系统的系统图。参见图4,电动汽车能耗预测系统包括:
获取模块201,用于获取电动汽车的历史行驶数据。
分割处理模块202,用于对历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据;行程片段数据包括电动汽车在行驶过程中的历史行驶数据,动力学片段数据包括电动汽车在恒定速度行驶或加速度行驶的过程中的历史行驶数据。
工况预测模块203,用于利用动力学片段数据和马尔可夫-蒙特卡洛方法对电动汽车进行工况预测,得到电动汽车的工况预测数据。
工况预测模块203具体包括:
行驶状态标记添加单元,用于利用动力学片段数据中的平均速度,对动力学片段数据中不同的行驶状态添加行驶状态标记。
行驶状态转移概率矩阵计算单元,用于利用动力学片段数据的时间顺序和行驶状态标记,计算得到电动汽车的行驶状态转移概率矩阵。行驶状态转移概率矩阵计算单元具体包括:
转移概率计算子单元,用于利用动力学片段数据的时间顺序,根据公式
Figure PCTCN2021098285-appb-000031
计算电动汽车的行驶状态从行驶状态标记i转移至行驶状态标记j的转移概率;式中,p ij表示转移概率;N ij表示从行驶状态标记i转移至行驶状态标记j的事件数。
行驶状态转移概率矩阵计算子单元,用于利用所有行驶状态标记之间的转移概率确定电动汽车的行驶状态转移概率矩阵。
工况预测单元,用于利用蒙特卡洛模拟方法、行驶状态转移概率矩阵和行驶状态标记对电动汽车进行工况预测,得到电动汽车的工况预测数据。
工况预测单元具体包括:
下一时刻行驶状态标记确定子单元,用于利用蒙特卡洛模拟方法和行驶状态转移概率矩阵确定电动汽车的下一时刻行驶状态标记。
预测行驶工况数据确定子单元,用于确定动力学片段数据中与下一时刻行驶状态标记相同的历史行驶数据,得到预测行驶工况数据。
第一获取子单元,用于获取电动汽车的当前行驶工况和目的地里程长度。
拼接子单元,用于按照时间顺序将预测行驶工况数据与当前行驶工况进行拼接,得到电动汽车的工况预测数据。
第二获取子单元,用于获取工况预测数据的里程长度。
第一判断子单元,用于判断工况预测数据的里程长度是否小于目的地里程长度,得到第一判断结果。
返回子单元,用于当第一判断结果为是时,执行下一时刻行驶状态标记确定子单元,更新工况预测数据。
第一获取模块204,用于获取行程片段数据的行驶特征参数和能耗数据。
能耗预测模型建立模块205,用于将行程片段数据的行驶特征参数作为输入,能耗数据作为输出,利用机器学习方法建立能耗预测模型。
能耗预测模型建立模块205具体包括:
能耗预测初始模型训练单元,用于采用K折交叉验证方法和极端梯度提升算法对行程片段数据的行驶特征参数和能耗数据进行训练,得到能耗预测初始模型。
优化单元,用于采用网格搜索方法对能耗预测初始模型的超参数进行优化,得到能耗预测模型。
第二获取模块206,用于获取工况预测数据的行驶特征参数。
能耗预测模块207,用于将工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值。
本发明的电动汽车能耗预测方法及系统大大提高了电动汽车实际行驶工况下的能耗预测精度。相比于传统方法使用固定工况模拟电动汽车的能耗,本发明基于电动汽车的历史行驶数据,提取其行驶特征,在进行能 耗预测时首先基于车辆当前的状态预测车辆的未来行驶工况,并充分考虑行驶环境因素包括环境温度、交通状况以及驾驶员驾驶行为等因素的影响,从而保证能耗预测模型在实际应用环境下具有良好的精度,同时也提高了电动汽车在实际行驶工况下的能耗预测精度。此外机器学习的经验学习和迭代优化能够在大量车辆历史行驶数据作为训练样本的基础上提取、拟合复杂工况与能耗之间的非线性耦合关系,并随着车辆不断产生的行程片段进行迭代提高精度,最终实现实际工况下的电动汽车高精度预测。
本说明书中各个实施例采用递进的方式描述,每个实施例重点说明的都是与其他实施例的不同之处,各个实施例之间相同相似部分互相参见即可。对于实施例公开的系统而言,由于其与实施例公开的方法相对应,所以描述的比较简单,相关之处参见方法部分说明即可。
本文中应用了具体个例对本发明的原理及实施方式进行了阐述,以上实施例的说明只是用于帮助理解本发明的方法及其核心思想;同时,对于本领域的一般技术人员,依据本发明的思想,在具体实施方式及应用范围上均会有改变之处。综上所述,本说明书内容不应理解为对本发明的限制。

Claims (10)

  1. 一种电动汽车能耗预测方法,其特征在于,包括:
    获取电动汽车的历史行驶数据;
    对所述历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据;所述行程片段数据包括所述电动汽车在行驶过程中的历史行驶数据,所述动力学片段数据包括所述电动汽车在恒定速度行驶或加速度行驶的过程中的历史行驶数据;
    利用所述动力学片段数据和马尔可夫-蒙特卡洛方法对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据;
    获取所述行程片段数据的行驶特征参数和能耗数据;
    将所述行程片段数据的行驶特征参数作为输入,所述能耗数据作为输出,利用机器学习方法建立能耗预测模型;
    获取所述工况预测数据的行驶特征参数;
    将所述工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值。
  2. 根据权利要求1所述的电动汽车能耗预测方法,其特征在于,所述利用所述动力学片段数据和马尔可夫-蒙特卡洛方法对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据,具体包括:
    利用所述动力学片段数据中的平均速度,对所述动力学片段数据中不同的行驶状态添加行驶状态标记;
    利用所述动力学片段数据的时间顺序和所述行驶状态标记,计算得到所述电动汽车的行驶状态转移概率矩阵;
    利用蒙特卡洛模拟方法、所述行驶状态转移概率矩阵和所述行驶状态标记对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据。
  3. 根据权利要求2所述的电动汽车能耗预测方法,其特征在于,所述利用所述动力学片段数据的时间顺序和所述行驶状态标记,计算得到所述电动汽车的行驶状态转移概率矩阵,具体包括:
    利用所述动力学片段数据的时间顺序,根据公式
    Figure PCTCN2021098285-appb-100001
    计算所述电动汽车的行驶状态从行驶状态标记i转移至行驶状态标记j的转移概 率;式中,p ij表示转移概率;N ij表示从行驶状态标记i转移至行驶状态标记j的事件数;
    利用所有所述行驶状态标记之间的转移概率确定所述电动汽车的行驶状态转移概率矩阵。
  4. 根据权利要求2所述的电动汽车能耗预测方法,其特征在于,所述利用蒙特卡洛模拟方法、所述行驶状态转移概率矩阵和所述行驶状态标记对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据,具体包括:
    利用蒙特卡洛模拟方法和所述行驶状态转移概率矩阵确定所述电动汽车的下一时刻行驶状态标记;
    确定所述动力学片段数据中与所述下一时刻行驶状态标记相同的历史行驶数据,得到预测行驶工况数据;
    获取所述电动汽车的当前行驶工况和目的地里程长度;
    按照时间顺序将所述预测行驶工况数据与所述当前行驶工况进行拼接,得到所述电动汽车的工况预测数据;
    获取所述工况预测数据的里程长度;
    判断所述工况预测数据的里程长度是否小于所述目的地里程长度,得到第一判断结果;
    若所述第一判断结果为是,则返回“利用蒙特卡洛模拟方法和所述行驶状态转移概率矩阵确定所述电动汽车的下一时刻行驶状态标记”,更新所述工况预测数据。
  5. 根据权利要求1所述的电动汽车能耗预测方法,其特征在于,所述将所述行程片段数据的行驶特征参数作为输入,所述能耗数据作为输出,利用机器学习方法建立能耗预测模型,具体包括:
    采用K折交叉验证方法和极端梯度提升算法对所述行程片段数据的行驶特征参数和所述能耗数据进行训练,得到能耗预测初始模型;
    采用网格搜索方法对所述能耗预测初始模型的超参数进行优化,得到能耗预测模型。
  6. 一种电动汽车能耗预测系统,其特征在于,包括:
    获取模块,用于获取电动汽车的历史行驶数据;
    分割处理模块,用于对所述历史行驶数据进行分割处理,得到行程片段数据和动力学片段数据;所述行程片段数据包括所述电动汽车在行驶过程中的历史行驶数据,所述动力学片段数据包括所述电动汽车在恒定速度行驶或加速度行驶的过程中的历史行驶数据;
    工况预测模块,用于利用所述动力学片段数据和马尔可夫-蒙特卡洛方法对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据;
    第一获取模块,用于获取所述行程片段数据的行驶特征参数和能耗数据;
    能耗预测模型建立模块,用于将所述行程片段数据的行驶特征参数作为输入,所述能耗数据作为输出,利用机器学习方法建立能耗预测模型;
    第二获取模块,用于获取所述工况预测数据的行驶特征参数;
    能耗预测模块,用于将所述工况预测数据的行驶特征参数输入建立的能耗预测模型,得到能耗预测值。
  7. 根据权利要求6所述的电动汽车能耗预测系统,其特征在于,所述工况预测模块,具体包括:
    行驶状态标记添加单元,用于利用所述动力学片段数据中的平均速度,对所述动力学片段数据中不同的行驶状态添加行驶状态标记;
    行驶状态转移概率矩阵计算单元,用于利用所述动力学片段数据的时间顺序和所述行驶状态标记,计算得到所述电动汽车的行驶状态转移概率矩阵;
    工况预测单元,用于利用蒙特卡洛模拟方法、所述行驶状态转移概率矩阵和所述行驶状态标记对所述电动汽车进行工况预测,得到所述电动汽车的工况预测数据。
  8. 根据权利要求7所述的电动汽车能耗预测系统,其特征在于,所述行驶状态转移概率矩阵计算单元,具体包括:
    转移概率计算子单元,用于利用所述动力学片段数据的时间顺序,根 据公式
    Figure PCTCN2021098285-appb-100002
    计算所述电动汽车的行驶状态从行驶状态标记i转移至行驶状态标记j的转移概率;式中,p ij表示转移概率;N ij表示从行驶状态标记i转移至行驶状态标记j的事件数;
    行驶状态转移概率矩阵计算子单元,用于利用所有所述行驶状态标记之间的转移概率确定所述电动汽车的行驶状态转移概率矩阵。
  9. 根据权利要求7所述的电动汽车能耗预测系统,其特征在于,所述工况预测单元,具体包括:
    下一时刻行驶状态标记确定子单元,用于利用蒙特卡洛模拟方法和所述行驶状态转移概率矩阵确定所述电动汽车的下一时刻行驶状态标记;
    预测行驶工况数据确定子单元,用于确定所述动力学片段数据中与所述下一时刻行驶状态标记相同的历史行驶数据,得到预测行驶工况数据;
    第一获取子单元,用于获取所述电动汽车的当前行驶工况和目的地里程长度;
    拼接子单元,用于按照时间顺序将所述预测行驶工况数据与所述当前行驶工况进行拼接,得到所述电动汽车的工况预测数据;
    第二获取子单元,用于获取所述工况预测数据的里程长度;
    第一判断子单元,用于判断所述工况预测数据的里程长度是否小于所述目的地里程长度,得到第一判断结果;
    返回子单元,用于当所述第一判断结果为是时,执行下一时刻行驶状态标记确定子单元,更新所述工况预测数据。
  10. 根据权利要求6所述的电动汽车能耗预测系统,其特征在于,所述能耗预测模型建立模块,具体包括:
    能耗预测初始模型训练单元,用于采用K折交叉验证方法和极端梯度提升算法对所述行程片段数据的行驶特征参数和所述能耗数据进行训练,得到能耗预测初始模型;
    优化单元,用于采用网格搜索方法对所述能耗预测初始模型的超参数进行优化,得到能耗预测模型。
PCT/CN2021/098285 2020-06-05 2021-06-04 一种电动汽车能耗预测方法及系统 WO2021244632A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010505376.3A CN111666715B (zh) 2020-06-05 2020-06-05 一种电动汽车能耗预测方法及系统
CN202010505376.3 2020-06-05

Publications (1)

Publication Number Publication Date
WO2021244632A1 true WO2021244632A1 (zh) 2021-12-09

Family

ID=72386394

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/098285 WO2021244632A1 (zh) 2020-06-05 2021-06-04 一种电动汽车能耗预测方法及系统

Country Status (2)

Country Link
CN (1) CN111666715B (zh)
WO (1) WO2021244632A1 (zh)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114035098A (zh) * 2021-12-14 2022-02-11 北京航空航天大学 一种融合未来工况信息和历史状态信息的锂电池健康状态预测方法
CN114407661A (zh) * 2022-02-10 2022-04-29 北京理工大学重庆创新中心 基于数据驱动的电动车能耗预测方法、系统、设备及介质
CN114492169A (zh) * 2021-12-28 2022-05-13 国网河北省电力有限公司营销服务中心 考虑用户市场响应潜力的电动汽车用户画像方法及系统
CN114781245A (zh) * 2022-03-28 2022-07-22 北京理工大学 基于油耗强相关参数修正的瞬态汽车油耗估计方法及系统
CN114999134A (zh) * 2022-05-26 2022-09-02 北京新能源汽车股份有限公司 一种驾驶行为预警方法、装置及系统
CN115027489A (zh) * 2022-07-08 2022-09-09 中国科学技术大学 一种基于循环预测时域的车辆减排预测控制方法
CN115071505A (zh) * 2022-06-20 2022-09-20 武汉理工大学 燃料电池汽车分层规划方法、系统、装置及存储介质
CN115188205A (zh) * 2022-07-04 2022-10-14 武汉理工大学 一种基于道路信息的汽车行驶工况修正方法
CN115424443A (zh) * 2022-09-20 2022-12-02 安徽江淮汽车集团股份有限公司 一种基于行驶数据的车辆异常监控方法
DE102022004321A1 (de) 2022-11-21 2023-01-05 Mercedes-Benz Group AG Verfahren zur Verbrauchsschätzung von Elektrofahrzeugen und dessen Verwendung zum Ermitteln einer Navigationsroute
CN115587327A (zh) * 2022-08-27 2023-01-10 长安大学 一种考虑Jerk的汽车油耗预测方法
CN115808922A (zh) * 2022-01-07 2023-03-17 宁德时代新能源科技股份有限公司 商用电动车辆能耗预测方法、装置和计算机设备
CN116109025A (zh) * 2023-04-04 2023-05-12 深圳天溯计量检测股份有限公司 基于大数据的电池续航测试方法
WO2023222372A1 (de) * 2022-05-17 2023-11-23 Mercedes-Benz Group AG Verfahren zur planung von ladevorgängen für ein fahrzeug und fahrzeug
WO2023240919A1 (zh) * 2022-06-17 2023-12-21 中汽研汽车检验中心(天津)有限公司 车辆行驶工况确定方法、设备和存储介质

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111666715B (zh) * 2020-06-05 2023-04-25 北京理工大学 一种电动汽车能耗预测方法及系统
CN112560186B (zh) * 2020-10-15 2022-08-05 吉林大学 电动汽车行程能耗预测方法、装置、设备及可存储介质
CN113076697B (zh) * 2021-04-20 2022-11-29 潍柴动力股份有限公司 典型驾驶工况的构建方法、相关装置及计算机存储介质
CN113313309B (zh) * 2021-05-31 2022-08-23 河南职业技术学院 一种用于新能源汽车的充电方法及系统
CN113642768A (zh) * 2021-07-12 2021-11-12 南京航空航天大学 一种基于工况重构的车辆行驶能耗预测方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180118033A1 (en) * 2016-10-27 2018-05-03 Hefei University Of Technology Method and device for on-line prediction of remaining driving mileage of electric vehicle
CN110610260A (zh) * 2019-08-21 2019-12-24 南京航空航天大学 行车能耗预测系统、方法、存储介质和设备
CN110728772A (zh) * 2019-10-15 2020-01-24 西南交通大学 一种有轨电车典型行驶工况构建方法
CN111666715A (zh) * 2020-06-05 2020-09-15 北京理工大学 一种电动汽车能耗预测方法及系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108198425A (zh) * 2018-02-10 2018-06-22 长安大学 一种电动汽车行驶工况的构建方法
CN110675619B (zh) * 2018-07-02 2021-06-15 上海汽车集团股份有限公司 一种车辆行程能耗预测方法和装置
CN109558988B (zh) * 2018-12-13 2021-08-06 北京理工新源信息科技有限公司 一种基于大数据融合的电动汽车能耗预测方法及系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180118033A1 (en) * 2016-10-27 2018-05-03 Hefei University Of Technology Method and device for on-line prediction of remaining driving mileage of electric vehicle
CN110610260A (zh) * 2019-08-21 2019-12-24 南京航空航天大学 行车能耗预测系统、方法、存储介质和设备
CN110728772A (zh) * 2019-10-15 2020-01-24 西南交通大学 一种有轨电车典型行驶工况构建方法
CN111666715A (zh) * 2020-06-05 2020-09-15 北京理工大学 一种电动汽车能耗预测方法及系统

Cited By (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114035098A (zh) * 2021-12-14 2022-02-11 北京航空航天大学 一种融合未来工况信息和历史状态信息的锂电池健康状态预测方法
CN114492169B (zh) * 2021-12-28 2024-03-29 国网河北省电力有限公司营销服务中心 考虑用户市场响应潜力的电动汽车用户画像方法及系统
CN114492169A (zh) * 2021-12-28 2022-05-13 国网河北省电力有限公司营销服务中心 考虑用户市场响应潜力的电动汽车用户画像方法及系统
CN115808922B (zh) * 2022-01-07 2023-10-27 宁德时代新能源科技股份有限公司 商用电动车辆能耗预测方法、装置和计算机设备
CN115808922A (zh) * 2022-01-07 2023-03-17 宁德时代新能源科技股份有限公司 商用电动车辆能耗预测方法、装置和计算机设备
CN114407661A (zh) * 2022-02-10 2022-04-29 北京理工大学重庆创新中心 基于数据驱动的电动车能耗预测方法、系统、设备及介质
CN114407661B (zh) * 2022-02-10 2023-11-07 北京理工大学重庆创新中心 基于数据驱动的电动车能耗预测方法、系统、设备及介质
CN114781245B (zh) * 2022-03-28 2024-05-28 北京理工大学 基于油耗强相关参数修正的瞬态汽车油耗估计方法及系统
CN114781245A (zh) * 2022-03-28 2022-07-22 北京理工大学 基于油耗强相关参数修正的瞬态汽车油耗估计方法及系统
WO2023222372A1 (de) * 2022-05-17 2023-11-23 Mercedes-Benz Group AG Verfahren zur planung von ladevorgängen für ein fahrzeug und fahrzeug
CN114999134B (zh) * 2022-05-26 2024-05-28 北京新能源汽车股份有限公司 一种驾驶行为预警方法、装置及系统
CN114999134A (zh) * 2022-05-26 2022-09-02 北京新能源汽车股份有限公司 一种驾驶行为预警方法、装置及系统
WO2023240919A1 (zh) * 2022-06-17 2023-12-21 中汽研汽车检验中心(天津)有限公司 车辆行驶工况确定方法、设备和存储介质
CN115071505A (zh) * 2022-06-20 2022-09-20 武汉理工大学 燃料电池汽车分层规划方法、系统、装置及存储介质
CN115188205B (zh) * 2022-07-04 2024-03-29 武汉理工大学 一种基于道路信息的汽车行驶工况修正方法
CN115188205A (zh) * 2022-07-04 2022-10-14 武汉理工大学 一种基于道路信息的汽车行驶工况修正方法
CN115027489A (zh) * 2022-07-08 2022-09-09 中国科学技术大学 一种基于循环预测时域的车辆减排预测控制方法
CN115027489B (zh) * 2022-07-08 2023-03-07 中国科学技术大学 一种基于循环预测时域的车辆减排预测控制方法
CN115587327A (zh) * 2022-08-27 2023-01-10 长安大学 一种考虑Jerk的汽车油耗预测方法
CN115424443B (zh) * 2022-09-20 2024-04-26 安徽江淮汽车集团股份有限公司 一种基于行驶数据的车辆异常监控方法
CN115424443A (zh) * 2022-09-20 2022-12-02 安徽江淮汽车集团股份有限公司 一种基于行驶数据的车辆异常监控方法
DE102022004321A1 (de) 2022-11-21 2023-01-05 Mercedes-Benz Group AG Verfahren zur Verbrauchsschätzung von Elektrofahrzeugen und dessen Verwendung zum Ermitteln einer Navigationsroute
CN116109025A (zh) * 2023-04-04 2023-05-12 深圳天溯计量检测股份有限公司 基于大数据的电池续航测试方法

Also Published As

Publication number Publication date
CN111666715A (zh) 2020-09-15
CN111666715B (zh) 2023-04-25

Similar Documents

Publication Publication Date Title
WO2021244632A1 (zh) 一种电动汽车能耗预测方法及系统
CN111497679B (zh) 一种纯电动汽车能耗监测优化方法及系统
CN109558988B (zh) 一种基于大数据融合的电动汽车能耗预测方法及系统
Kivekäs et al. Stochastic driving cycle synthesis for analyzing the energy consumption of a battery electric bus
Yan et al. EV charging load simulation and forecasting considering traffic jam and weather to support the integration of renewables and EVs
US9759573B2 (en) Route based energy consumption estimation using physical models
JP7163099B2 (ja) エネルギー管理装置、モデル管理方法及びコンピュータプログラム
CN110570014A (zh) 一种基于蒙特卡洛和深度学习的电动汽车充电负荷预测方法
CN110667434A (zh) 一种自适应工况的纯电动汽车行驶里程估计方法及系统
CN104123833A (zh) 一种道路状况的规划方法和装置
CN111670340A (zh) 一种车辆剩余行驶里程的获取方法、电子设备及车辆
CN104442825A (zh) 一种电动汽车剩余行驶里程的预测方法和系统
CN113263954B (zh) 电动汽车续驶里程预测方法、装置、设备和可读存储介质
CN103838868B (zh) 一种基于多公交线路运行数据融合的城市公交到站时间预测方法
CN108345987B (zh) 一种道路基础设施建设项目影响评价决策支持系统及方法
CN109484187A (zh) 信息处理方法和具有电驱动功能的车辆
CN109409571A (zh) 一种电动汽车充电需求预测方法及装置
Xing et al. Operation Energy Consumption Estimation Method of Electric Bus Based on CNN Time Series Prediction
CN110674990B (zh) 设置滑动窗口更新机制的即时配送路径选择方法及系统
Zhang et al. Neural network based vehicle speed prediction for specific urban driving
CN115759347A (zh) 一种基于特征数据的电动公交车行程能耗快速预测方法
Chen et al. The framework of increasing drivers’ income on the online taxi platforms
CN116205370A (zh) 一种基于地区大数据的电动汽车续航里程预测方法及装置
CN114722487A (zh) 一种基于大数据的新能源汽车车型参数预测估计方法
Wang et al. Modeling and algorithms of VMT and AADT estimation for community area traffic networks

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21817733

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 03.04.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 21817733

Country of ref document: EP

Kind code of ref document: A1