CN109733248B

CN109733248B - Pure electric vehicle remaining mileage model prediction method based on path information

Info

Publication number: CN109733248B
Application number: CN201910018307.7A
Authority: CN
Inventors: 郭建华; 王引航; 刘纬纶; 刘翠; 石大排; 刘昨非; 刘康杰; 初亮
Original assignee: Jilin University
Current assignee: Jilin University
Priority date: 2019-01-09
Filing date: 2019-01-09
Publication date: 2020-07-24
Anticipated expiration: 2039-01-09
Also published as: CN109733248A

Abstract

The invention discloses a pure electric vehicle remaining mileage model prediction method based on path information, which comprises the following steps of: analyzing historical driving data of a driver, extracting path information, and generating a state transition probability matrix according with the behavior characteristics of the driver; generating a predicted vehicle speed based on road information of a future path and a corresponding state transition probability matrix; establishing a parameter estimation model, and estimating driving parameters influencing the energy consumption and the remaining driving mileage of the automobile; establishing an RDR calculation model to predict the remaining driving mileage of the vehicle, and calculating the energy consumption rate of the vehicle by using the predicted vehicle speed obtained by the vehicle speed prediction model and the driving parameters estimated by the parameter estimation model as model inputs in the energy consumption prediction model; the residual energy prediction model is used for predicting the residual energy of the vehicle battery; and the remaining driving mileage of the vehicle can be predicted by integrating the energy consumption rate of the vehicle and the remaining energy of the battery, and is displayed through a remaining driving mileage display model.

Description

Pure electric vehicle remaining mileage model prediction method based on path information

Technical Field

The invention relates to a pure electric vehicle remaining mileage model prediction method based on path information, and belongs to the technical field of new energy vehicles.

Background

Compared with the conventional internal combustion engine automobile, the Battery Electric Vehicle (BEV) has obvious advantages in terms of energy consumption and emission, such as good dynamic property, low driving noise, energy conservation, zero emission and the like. However, the driving range of the electric vehicle is also short and the charging time is long due to the limitation of the development of the battery technology. Pure electric vehicle drivers are concerned about whether they can reach their destination under the current remaining energy, which is called "mileage anxiety" which is one of the main factors that currently limit the acceptance of electric vehicles. Obviously, installing large capacity batteries, rapidly charging and establishing more charging stations are effective means for effectively relieving and solving "mileage anxiety", but these methods still require a long time to be implemented due to limitations of the current state of the art and capital conditions. Another effective means is "accurate Remaining Driving Range prediction", and the driver can determine whether the vehicle can reach the destination or not by the predicted "Remaining Driving Range" (RDR), and plan the route and the charging location in advance. In addition, accurate mileage prediction is also the basis of electric vehicle energy management, and according to the remaining driving mileage, the BEV energy management system can reasonably optimize the use of electric energy, improve the driving mileage of the electric vehicle, and relieve the 'mileage anxiety' of a driver.

Currently, many researchers have proposed various BEV energy consumption and RDR prediction methods, which can be basically divided into two categories: historical data based RDR prediction and model based RDR prediction. The historical data based RDR prediction method is the RDR prediction method commonly used on BEVs currently in commercial operation. The method is used for counting historical energy consumption data of a period of time, supposing that future energy consumption is similar to current energy consumption, calculating current energy consumption rate, estimating residual energy according to the state of Charge (SOC), and finally obtaining the predicted residual driving mileage. The method has the advantages of simple calculation, good real-time performance and easy realization. Therefore, this method is used for the RDR prediction of most electric vehicles. However, the disadvantages of this approach are: when the future operating condition is greatly changed, the error of the prediction becomes large and the prediction result is completely untrustworthy. The BEV energy consumption is affected by many factors, such as operating conditions (vehicle speed), driver driving behavior (driving style), grade, temperature, battery SOC, battery State of Health (SOH), wind speed, road conditions, etc. The operating condition (vehicle speed) is one of the most main factors influencing energy consumption, and under different types of operating conditions, such as cities, suburbs, high speeds and the like, the energy consumption of electric vehicles has great difference, and obviously, when the operating condition changes, the RDR prediction based on historical data inevitably fails.

Disclosure of Invention

In order to solve the problems in the prior art, the invention provides a pure electric vehicle remaining mileage model prediction method based on path information. Analyzing a certain amount of historical driving data of drivers, extracting path information, and generating a state Transition Probability Matrix (TPM) which accords with the behavior characteristics of the drivers; then, based on the road information of the future path and the corresponding TPM, a predicted working condition (vehicle speed) controlled by the future road information is generated based on the Markov random theory. Next, based on the electric vehicle performance test, an electric vehicle accurate energy consumption model is established, which takes into account factors such as temperature, gradient, battery state of charge (SOC) that mainly affect energy consumption and RDR. And acquiring path information from a vehicle-mounted sensor, a weather forecast system and an electronic map, performing model estimation on related parameters in a vehicle model, and inputting the predicted vehicle speed generated in the first step into the energy consumption model to realize accurate prediction of BEV energy consumption and remaining driving mileage.

The purpose of the invention is realized by the following technical scheme:

a pure electric vehicle remaining mileage model prediction method based on path information comprises the following steps:

step one, establishing a vehicle speed prediction model to generate a future path predicted vehicle speed: analyzing a certain amount of historical driving data of drivers, extracting path information, and generating a state transition probability matrix according with the behavior characteristics of the drivers; generating a predicted vehicle speed controlled by the future road information based on the road information of the future path and a corresponding state transition probability matrix and on the basis of a Markov random theory;

establishing a parameter estimation model, and estimating driving parameters influencing the energy consumption and the remaining driving mileage of the automobile;

step three, establishing an RDR calculation model to predict the remaining driving mileage of the vehicle: the RDR calculation model comprises: the energy consumption prediction model, the residual energy prediction model and the residual travel mileage display model; the energy consumption prediction model takes the predicted vehicle speed obtained by the vehicle speed prediction model and the driving parameters estimated by the parameter estimation model as model inputs, and calculates the vehicle energy consumption rate; the residual energy prediction model is used for predicting the residual energy of the vehicle battery; and the remaining driving mileage of the vehicle can be predicted by integrating the energy consumption rate of the vehicle and the remaining energy of the battery, and is displayed through a remaining driving mileage display model.

The invention has the beneficial effects that:

(1) the invention combines the historical working condition and the future path information, simultaneously considers the driving characteristics of the driver and the path information characteristics, and carries out controllable random working condition prediction. The method has the advantages of good instantaneity, high accuracy and good adaptability of the behavior characteristics of the driver.

(2) The driver historical working condition data is processed to form a Markov probability transfer matrix with the driving style and the road type as indexes, the memory space is small, the calculation is convenient, and the real-time updating can be realized. Along with the increase of the driving mileage, the probability transfer matrix representatively enhances the driving behavior characteristics of the driver, but the storage amount is kept unchanged; when the types of the stored roads are increased, the prediction accuracy is greatly improved, the storage amount is only slightly increased, and the vehicle-mounted system requirement is met.

(3) The proposed BEV energy consumption model considers the influence of road information such as temperature, gradient, battery state of charge (SOC) and the like on energy consumption, and accurately estimates model parameters. The model adopts a reverse modeling method, comprehensively considers the energy consumption of vehicle running, the transmission loss of a power system, the energy consumption of an auxiliary system, the energy recovery of regenerative braking and the like based on a vehicle performance test, and has high model precision, small calculated amount and good real-time property.

Drawings

Specific embodiments of the present invention will be described in detail below with reference to application examples.

FIG. 1 is a BEV and energy management system hardware architecture;

FIG. 2 is a remaining range prediction algorithm architecture;

FIG. 3 is a force balance diagram of an automobile;

FIG. 4 is a graph of a test of the relationship between the power loss of the transmission system and the input power of the motor at different rotation speeds;

FIG. 5 is a test plot of battery open circuit voltage versus SOC;

FIG. 6 is a flow chart of a vehicle speed generation algorithm;

FIG. 7 shows a road test route and route information for a city;

FIG. 8 shows the speed and path information for a city road test;

FIG. 9 is a segment of the "class II" road condition of an urban area;

FIG. 10 illustrates the acceleration and deceleration phases of a "class II" road in a city;

FIG. 11 is a schematic diagram of gridding and generating TPMs of vehicle speed data in an acceleration stage;

FIG. 12 is TPMs in the acceleration and deceleration phases of an urban "class II" road;

FIG. 13 shows the reference condition and predicted vehicle speed of an urban road;

FIG. 14 is a schematic diagram of a condition segment generation algorithm;

FIG. 15 is a curve of actual measurement and prediction of urban road energy consumption for different drivers;

fig. 16 is an actual and predicted curve of the RDR of the urban road.

Detailed Description

The invention is further described below with reference to the accompanying drawings. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way.

Fig. 1 is a hardware configuration of the BEV remaining mileage predicting system. The BEV power system in this example is composed of a Motor, a Motor Control Unit (MCU) battery, a Battery Management Unit (BMU), a reducer, and the like. In order to realize the remaining mileage prediction function, the BEV is equipped with a vehicle-mounted GPS navigation System (GVNS), an Intelligent Transportation System (ITS), a Geographic Information System (GIS), a Weather Report System (WRS), and the like. The information fusion processor is used for acquiring required path information from the system, performing data acquisition, storage, cleaning, format alignment and the like on the information, fusing different types of path information in different formats into a path information capable of being operated by the RDR prediction algorithm provided by the invention in an Energy Management Unit (EMU), and estimating the BEV energy consumption, calculating the residual driving range and managing the BEV energy. The EMU communicates with the BMU and the MCU through the CAN lines to coordinate and optimize BEV energy usage.

Fig. 2 is a remaining mileage prediction algorithm architecture, which includes three parts:

the first Part (Part1) RDR (remaining mileage) calculation model, which functions to calculate the BEV energy consumption rate and the battery remaining energy, and calculate the remaining mileage, and finally display the remaining mileage prediction result according to the driving demand. This section includes 3 sub-models: an energy consumption prediction model (vehicle model), a remaining energy prediction model (battery model), and a remaining mileage display model.

The second Part (Part2) is a parameter estimation model, which estimates the driving parameters required by the RDR calculation model, such as the whole vehicle mass, the road gradient and the air density. The partial model includes an estimation model of the above parameters.

The third Part (Part3) is a vehicle speed prediction model, also known as a condition prediction model. Vehicle speed is an input to the parameter estimation model and the RDR prediction model. The vehicle speed prediction model generates a future path prediction vehicle speed based on a Markov random theory according to the future path information and the historical driving data of the driver.

The predicted vehicle speed obtained by the vehicle speed prediction model and the driving parameters estimated by the parameter estimation model are input into the RDR calculation model, and the predicted value of the RDR is finally obtained and displayed on an instrument board, so that the accurate RDR prediction can effectively relieve the 'mileage anxiety' of a driver.

The three-part model is described in turn by way of example below.

Part 1RDR calculation model

(1) Energy consumption prediction model (vehicle model)

The target vehicle in the embodiment is a small pure electric car, the service mass of the car is 1060kg, the maximum speed of the car is 130km/h, the battery capacity is 91Ah, and the driving range can reach 228km under the constant speed driving condition of 69 km/h. The vehicle structure is shown in figure 1, and energy is consumed during the running process of the vehicleThe consumption can be divided into three parts: the first part is the loss of energy consumption of the motor and drive system, as indicated by points a-B in fig. 1; the second part is the energy consumed to overcome the running resistance, as indicated by point C in fig. 1; the third part is the auxiliary system energy consumption, as indicated by point D in fig. 1. In addition, there is a need to remove the energy recovered by the regenerative braking system. An energy consumption prediction model is established by adopting a reverse modeling method, namely the input of the model is the vehicle speed, and the output of the model is the battery output power P_bat(W) that is

Wherein, F_w(N) is the driving force of the automobile, which is equal to the running resistance of the automobile, namely F, as shown in the stress balance diagram of the automobile in fig. 6_w＝F_r+F_aero+F_g+F_m(ii) a v (m/s) is the running speed of the vehicle and is obtained by a vehicle speed prediction model of Part 3; p_{pt_loss}(W) BEV driveline power loss; p_aux(W) is the electrical accessory consumed power.

Rolling resistance F_r(N) can be calculated from the following formula

F_r＝f_rm_vgcos(α_slop) (1.2)

Wherein f is_rThe coefficient is a rolling resistance coefficient, and is influenced by the ambient temperature and the road surface type, and the coefficient is obtained by a rolling resistance coefficient estimation model in Part 2; m_v(kg) is the BEV whole vehicle mass, which is obtained by a whole vehicle mass estimation model in Part 2; g (m/s)²) Taking the gravity acceleration as 9.81m/s²；α_slop(rad) is road slope, obtained from the road slope estimation model of Part 2.

Air resistance F_aero(N) is

Wherein rho (kg/m3) is the air density and is 1.29kg/m 3; a. the_f(m²) The frontal area of the vehicle is 1.97m in this example²；C_dIs the air resistance coefficient, which in this example is 0.3; v_win(m/s) is the wind speed in the direction of travel, obtained by the weather forecast system.

Slope resistance F_g(N) is

F_g＝m_vgsin(α_slop) (1.4)

Acceleration resistance F_m(N) is

Here J_w(kg·m²) Is the rotational inertia of the wheel and is 0.75kgm²；J_m(kg·m²) Is the rotational inertia of the motor and is 0.0384kgm²(ii) a r (m) is the tire radius, 0.278 m; i.e. i_gIs the transmission ratio of the gearbox, 8.654; dv/dt (m/s)²) The acceleration is the longitudinal automobile acceleration, and is obtained by differentiating the automobile speed, and the automobile speed is obtained by a automobile speed prediction model of Part 3.

System loss power P_{pt_loss}(W) can be measured by dynamometer experiment, and the value is motor output power P_elec(W) and Drum mechanical Power P_mecDifference of (W)

P_{pt_loss}＝P_elec-P_mec(1.6)

FIG. 4 is a test curve of the relationship between the transmission system loss power and the motor input power under different rotating speeds of the target vehicle. To simplify the calculations, the driveline power loss is described using a fitting equation. Fitting the test curve of fig. 4 by using a Matlab parameter estimation tool box to obtain a fitting formula and optimizing to obtain fitting parameters. Since the fitting formulas of the driving mode and the regenerative braking mode have different structures, the fitting formulas are respectively adopted for fitting, namely the fitting formulas are respectively matched with empirical formulas

Wherein the motor torque T_m(N.m) is

T_m＝F_wr/i_g(1.8)

Rotating speed of motor

Is composed of

P_c(W) is the idle loss power of the motor, and the empirical formula is

P_c＝c₁·v³+c₂·v²+c₃·v+c₄(1.10)

Wherein, a_i(i＝1～4)，b_i(i＝1～2)，c_i(i is 1-4) is a fitting coefficient.

The target vehicle driveline power loss fit equation in this example is

In an acceleration running mode

In regenerative braking mode

Wherein:

P_c＝0.06v³-4.85v²+116.93v+170 (1.13)

electric accessory energy consumption P_aux(W) has great randomness, and the average energy consumption of the electric accessory under various circulation conditions is used as the energy consumption of the electric accessory in the example, namely

P_aux＝P_{aux_avg}(1.14)

Wherein, P_{aux_avg}(W) is the average energy consumption of the electric accessory system under various cycle conditions, and is taken as 210W.

Finally obtaining the BEV energy consumption rate e_avg(kW/km), i.e.

Wherein S_rAnd (m) is the path length and is obtained from the navigation system.

(2) Residual energy prediction model (Battery model)

Since the battery is a complex electrochemical system, the remaining energy of the battery may vary greatly under different operating conditions. In the remaining energy prediction model, the present invention considers the influence of the state of health (SoH) of the battery, the battery temperature, and the like on the remaining energy of the battery. Battery residual energy E_rue(kWh) can be calculated from the following formula

E_rue＝Q₀·SoH·C_tempU_t,nom·(SOC-SOC_end,nom) (1.16)

Wherein Q is₀(Ah) is the rated capacity of the new battery, and is 91 Ah; c_tempThe battery temperature correction coefficient is determined by a battery characteristic test; u shape_t,nom(V) is the nominal terminal voltage of the battery; SOC_end，nomDischarging the lowest battery SOC; SoH is the state of health of the battery, defined as

Wherein Q is_bat(Ah) is the current battery rated capacity. The SoH is related to the number of charge and discharge cycles of the battery, the relationship between the SoH and the number of charge and discharge cycles can be determined by a battery test, and the SoH of the current battery is obtained from a battery management system.

The SOC of the battery with the formula (1.16) is estimated by adopting a simple battery internal resistance model, and the open-circuit voltage of the battery is

V_oc＝V_out+IR (1.18)

Wherein, V_out(V) is the battery output voltage; i (A) is the output current of the battery; r (omega) is the internal resistance of the battery. The open-circuit voltage versus SOC curve of the battery was determined by battery testing, as shown in fig. 5. The relation between the internal resistance R and the discharge current I of the battery can be determined by a battery charge-discharge test, and the fitting formula is

R＝d₁|I|³+d₂|I|²+d₃|I|+d₄(1.19)

Wherein d is_i(i is 1-4) is a fitting coefficient, and is obtained by optimizing a Matlab parameter estimation tool box according to a test curve, namely

R＝-3.84×10^-7|I|³+2.04×10^-5|I|²-3.7×10^-3|I|+0.41 (1.20)

Output power P of battery_bat(W) is

P_bat＝V_outI＝V_ocI-I²R (1.21)

Estimating SOC using an ampere-hour method, i.e.

(3) Display model for remaining driving mileage

Respectively calculating energy consumption rate e from the energy consumption prediction model and the residual energy prediction model_avgAnd battery residual energy E_rueThen, the current time t can be calculated by the following formula₂RDR of remaining driving range_cal(km), i.e.

Another method for calculating the remaining driving mileage is

Wherein RDR_cal(t₁) To be in the past t₁RDR prediction result calculated by formula (1.23) at time, delta L_cum(t₁,t₂) Is from t₁To t₂The actual distance traveled at the moment is obtained by integrating the actual vehicle speed.

RDR from formula (1.23)_calThe energy state of the current vehicle and the energy consumption condition of the vehicle can be faithfully reflected. It can respond to the change of the driving cycle working condition quickly. However, suddenly under operating conditionsWhen the driving experience is influenced by anxiety of the driver, the RDR result predicted by the method has larger jump. RDR calculated from formula (1.24)_cumThe change is continuous and gentle, however, the influence of the change of the working condition on the RDR cannot be reflected, and a large error occurs in the predicted final result. In order to integrate the advantages and overcome the disadvantages of the two RDR prediction methods, the RDR prediction result RDR for final display_dis(km) can be calculated from the following formula

RDR_dis(t₂)＝w_disRDR_cal(t₂)+(1-w_dis)RDR_cum(t₂) (1.25)

Wherein, w_disIs a weight coefficient with a value range of [0, 1%]. The designer may select the value of this coefficient according to the particular RDR display requirements.

Part2 parameter estimation model

In the RDR calculation model of Part1, some vehicle driving parameters, such as rolling resistance coefficient, road gradient and vehicle mass, need to be calculated or estimated, and this section describes the calculation or estimation model of the above parameters.

(1) Initial value estimation model of rolling resistance coefficient

To calculate the rolling resistance F_r(N) (formula (1.2)) requires a coefficient f of rolling resistance_rPerforming an estimation of f_rWith respect to ambient temperature, tire temperature, vehicle speed, and road surface type, among others. F of the invention_rThe estimation model has two: f. of_rAn initial estimation model and a dynamic estimation model. Initial value estimation model for f_rAnd providing an initial value for the dynamic estimation model. The estimation method has small calculation amount and can be independently used as f under the condition of low precision requirement_rIs estimated. The study shows that the tire temperature and the vehicle speed are f_rIs much less than the influence of the road surface type and the ambient temperature. Therefore, only two factors of the ambient temperature and the road surface type are considered in this example. Under the conditions of different pavements and temperatures, target vehicles are adopted to carry out sliding experiments (Coast downsets) to obtain the relations among rolling resistance coefficients, environmental temperatures and pavement typesIs a test curve. Initial value f of rolling resistance coefficient_r0The fitting formula is

Wherein e is_i(i is 1 to 3) is a fitting coefficient, k_iThe road surface type correction coefficient.

In the embodiment, the sliding experiment is carried out on 5 different types of pavements at the temperature of 2-28 ℃, and f obtained by fitting is obtained_rIs expressed as

Wherein, the highway k ₁1 is ═ 1; smooth urban road k₂1.05; smooth country road k₃1.15; rough country road k₄1.35; belgium (road surface) road k₅＝1.40。

(2) Road surface gradient calculation model

On the premise that the path is known, the road surface gradient a can be calculated through a Geographic Information System (GIS) and GPS path longitude and latitude_slop(rad), i.e.

Where Δ h (m) is the height difference between two successive measurement points. The height data of the future path is obtained by a GIS system, such as the GIS system in this example is the SRTM (launch Radar Topographic mission) system of NASA, and the road elevation sampling point interval is 30 m.

(3) Dynamic estimation model for finished automobile mass and rolling resistance coefficient

BEV vehicle mass m_vThe energy consumption and the remaining driving mileage of the BEV are greatly influenced according to the number of passengers and different cargoes. However, the entire vehicle mass is difficult to directly measure while the vehicle is running, and therefore, the entire vehicle mass while running needs to be estimated. At the same time, the rolling resistance coefficient will also follow the tire temperatureThe road surface condition changes, and the rolling resistance coefficient initial value estimation model can only be used in occasions with low prediction precision requirements.

Therefore, in order to accurately estimate the overall vehicle mass and the rolling resistance coefficient when the vehicle is running, the present invention uses an m based Recursive least square (R L S) estimation algorithm_vAnd f_rAnd dynamically estimating the model.

The R L S estimation algorithm is based on a vehicle longitudinal dynamic model, and the output power P of the motor is obtained during the running process of the vehicle_m(W) is

Wherein, F_fric(N) is the driveline friction at the wheel, which can be obtained from a wheel spin test, in this example, F for the target vehicle_fricThe value was 15N.

The standard form of the equation (2.4) developed and written as a linear estimate is

Wherein the content of the first and second substances,

the classical R L S method was chosen to minimize the following

Its recursive solution is

Wherein

The initial value of P is set to 1, the signal sampling frequency is 25Hz, the initial value of the whole vehicle mass is set to 1250kg, and the initial value of the rolling resistance coefficient is determined by an equation (2.1). the R L S estimation method needs a certain time to stably converge, so that the RDR prediction model adopts the initial values of the whole vehicle mass and the rolling resistance coefficient before the predicted value is not converged.

Part3 vehicle speed prediction model

The vehicle speed-time history curve is also called as a driving cycle working condition, and is called as a working condition for short. It is the input of the RDR prediction model and is also one of the most important influencing factors of BEV energy consumption and RDR. Vehicle speed-time history is a typical markov process. Thus, the vehicle speed-time history is sampled as a discrete Markov Chain (Markov Chain), which is a random sequence of variables X that are characteristic of the Markov process₁,X₂,X₃,...X_n…, i.e. it is condition independent of past states given a current state, denoted as

Because the vehicle speed-time history is a Markov chain, the vehicle speed historical data can be counted based on a Markov random theory to generate a random vehicle speed-time history. And on the premise of operating for a long enough time or for a sufficient number of times, the generated vehicle speed curve and the historical vehicle speed curve have consistency on statistical characteristics. The discrete Markov chain generation predicted vehicle speed maintains randomness and can reflect the driving style of the driver. However, the vehicle speed generated by this method has a large randomness, and when the number of operations is small, the generated predicted vehicle speed and the actually measured vehicle speed have a large error. Therefore, the invention provides a random vehicle speed prediction method based on path information, which is used for acquiring future path information through a GPS and an ITS and generating a controllable random vehicle speed by combining a Markov method. The method has the advantages that random vehicle speed statistical characteristics generated at any time can be close to actual vehicle speed statistical characteristics, and the randomness of vehicle speed generation is kept.

FIG. 6 is a flow chart of a vehicle speed generation algorithm. The vehicle speed generation algorithm includes two parts of "generating Transition Probability Matrices (TPMs)" and "generating predicted vehicle speeds". The first part is mainly to generate transition probability matrixes of the driver on different road surfaces according to historical vehicle speed data and path information of the driver. The second part is to generate a predicted vehicle speed based on future path information in combination with the TPMs.

(1) Generating transition probability matrices

Step 1, working condition data acquisition and path information extraction

The acquisition of historical vehicle speed data is divided into two stages: acquiring real-time working condition test data of a real vehicle and acquiring real-time working condition data of a user.

In the first stage, the whole factory or system manufacturer should use the matching sample vehicle to perform the road test of the real vehicle, and the test vehicle should be equipped with a GPS, electronic map, GIS and CAN data acquisition system. The data to be acquired include vehicle speed time (v-t) history data, vehicle travel track (GPS coordinates), road information (path length, road type, signal light position and traffic light change time, corner position and corner radius, zone position, traffic flow data, etc.), driver driving data (steering wheel corner, pedal opening, gear, etc.), and vehicle travel data (battery SOC, battery temperature, motor torque, battery terminal voltage, battery terminal current, electric accessory status, power consumption, etc.). The road test is completed by drivers of different driving styles driving sample vehicles on different types of roads, and the driving mileage of the test is enough. The experimental data collected at this stage are mainly used for initial calibration of the RDR calculation model and the parameter estimation model and for generating the basic TPMs database.

In order to generate TPMs for drivers of different driving styles, the driving styles need to be evaluated and classified. Driving style index J of driver on certain type of road i_d(i) Is composed of

J_d(i)＝w₁·e_avg(i)+w₂v_m(i)+w₃v_max(i)+w₄a_am(i)+w₅a_bm(i) (3.2)

Wherein e is_avgAverage energy consumption rate for the driver on this type of road, (kW/km); v. of_mAverage speed, (km/h); v. of_maxMaximum speed, (km/h); a is_amIs the average acceleration, (m/s)²)；；a_bm(m/s) is the average deceleration (absolute value)²)；w_1～5Are weight coefficients.

After the test data of the sample vehicle roads of a plurality of drivers are collected, the J of each driver on each type of road is calculated by the formula (3.2)_d(i) And performing cluster analysis. The road test data of the same type are combined together to be used as the real vehicle road test data of the type of drivers. J at the center of the cluster_d(i) I.e. the typical driving style index J_dnor(i) In that respect In this example, the driving style is divided into three categories: general type J_dnor1(i) Radical J_dnor2(i) And mild form J_dnor3(i) In that respect Compared with the common type, the aggressive driver has higher energy consumption, average vehicle speed, maximum vehicle speed and average acceleration and deceleration under the same driving environment, and the mild type is opposite.

In the second phase, the BEV energy consumption prediction and management system still needs to collect the driving data of the user, and the specific data is basically the same as the first phase. The user driving data collected at this stage is mainly used for online calibration of the RDR calculation model and the parameter estimation model and for generating the user-specific TPMs database.

TABLE 1 road type and working condition segmenting method

In this example, a first stage road test mode is used to obtain condition test data. On a typical public road of a certain country, a target vehicle is adopted to carry out actual measurement road test, the total mileage of working condition data is 600km, and the working condition data is subjected to data processing, such as filtering, outlier rejection, data alignment and the like. Classifying the data according to road types, analyzing the vehicle speed data of the same road type, fitting the vehicle speed data after obtaining the vehicle speed frequency distribution, and obtaining the maximum vehicle speed v of the road of the type_max(including the vehicle speed value at 80% of the vehicle speed data), "minimum" vehicle speed v_min(vehicle speed value containing 20% vehicle speed data) and "average" vehicle speed v_nom(vehicle speed value including 50% vehicle speed data), (note that "maximum", "minimum", and "average" here are different from the usual meaning, and they are derived from the vehicle speed frequency distribution) is shown in table 1. The road types of the country are divided into three types of city, suburb and high speed, and the city road is further divided into a residential area, a third type, a second type and a main road.

Fig. 7 is a schematic diagram of a route and route information for a road test in a city using a target vehicle. The path of the test is shown in the figure, and the path comprises three urban road types of 'residential area', 'three types' and 'main road', and the type and the length of the road can be obtained from an electronic map. The map also shows the path turn angle (+) and the position of the traffic light (□), as well as the current wind direction, etc. The above-mentioned route information is integrated with the test vehicle speed data on the road at a certain time, as shown in fig. 8. The maximum vehicle speed v for the three road types is also shown_maxMinimum vehicle speed v_minAnd an average vehicle speed v_nom。

Step 2, dividing working condition sections according to road types

As can be seen from fig. 8, the driving condition (vehicle speed) is composed of the condition segments. In the present invention, the operating regime is defined as 2 "minimum" vehicle speeds v_minThe vehicle speed history in between. Dividing the whole working condition into working condition segments according to the definitionThe method comprises the steps of solving a second derivative of the vehicle speed process, wherein the point where the derivative is zero is an inflection point of a vehicle speed process curve, and then the vehicle speed in the inflection point is less than the minimum vehicle speed v_minIs the segment break point v_bI.e. by

The fragmentary discontinuities of the operating conditions of fig. 8 are indicated by the ". about.. And dividing the whole test working condition into a plurality of working condition sections according to the segment discontinuity points. And classifying the working condition sections according to the type of the road surface where the working condition sections are located. FIG. 9 is a segment of the behavior of the "class II" road of FIG. 8.

Step 3 of dividing acceleration and deceleration phases

Each operating mode section consists of an acceleration phase and a deceleration phase. The classified working condition segments need to be continuously divided into an acceleration stage and a deceleration stage, and the distance ratio of the acceleration stage is calculated. The stage break point is the maximum point of each operating condition segment, as indicated by the symbol in fig. 9. FIG. 10 is a graph of the "second class" road condition segment of FIG. 9 divided into an acceleration phase and a deceleration phase and clustered in one graph. Distance ratio r of acceleration section_daIs defined as

Wherein n is the number of the working condition sections; s_da(i) (km) is the length of the acceleration segment of the ith working condition segment; s_d(i) (km) is the length of the i-th operating regime section.

Step 4, gridding vehicle speed data

And (3) repeating the steps 1-3 until enough test vehicle speed data are obtained, then centralizing the vehicle speed data of the acceleration section and the deceleration section of the same driver on the same road, and regridding and interpolating the vehicle speed data. In this example, the original vehicle speed data has a distance unit of km and a vehicle speed unit of km/s, and the units are converted into m and m/s in gridding. FIG. 11 is a schematic diagram of gridding vehicle speed data in an acceleration phase. In this example, the distance sampling interval is 1m, and the vehicle speed sampling interval is 0.1 m/s. The method for gridding the vehicle speed data is to sample stage vehicle speed data according to a distance interval of 1m, and if the vehicle speed data is not on a grid point, the vehicle speed data needs to be rounded.

Step 5 statistics of the number of state transitions

The gridded phase vehicle speed data is scanned, and the state transition number is counted, as shown in fig. 11. First, a TPMs table is generated, and the ordinate of the table is the current state vehicle speed and the abscissa is the next state vehicle speed. The scanned vehicle speed interval is 0.1m/s, and the state vehicle speeds are scanned from the lowest vehicle speed to the highest vehicle speed in sequence. As shown in FIG. 11, if the current scanning vehicle speed status is 5.0m/s, there are 3 transition states: 5.1m/s, 5.2m/s and 5.3 m/s. Wherein 5.0m/s- >5.1m/s and 5.3m/s have a number of transfers of 1, 5.0m/s- >5.2m/s and 5.2 have a number of transfers of 2. In the state without transition, the number of transitions is 0. The above transfer numbers are sequentially filled into the TPMs table. And repeating the steps to complete the scanning of the vehicle speed data in all the stages and obtain all the state transition numbers (frequency).

Step 6 generating a transition probability matrix

Transition probability P_ijCalculated from the following equation

Wherein n is_ijIs from v_iTo v_jS is the total number of transition states.

Calculating all transition probabilities P in TPMs according to equation (3.4)_ijI.e. to obtain TPMs in v-v form, which can be converted into TPMs in v-a form from an acceleration a ═ Δ v/Δ t and Δ t ═ 1s, to give a ═ Δ v. Fig. 12 is a diagram of the urban "class two" road acceleration and deceleration phases TPMs. The intensity of the gray scale in the figure represents the magnitude of the transition probability.

The steps are repeated to generate the acceleration and deceleration stages TPMs with different driving styles (normal type, aggressive type and mild type) on various roads. And stores and indexes these TPMs classifications into a database.

(2) Generating a predicted vehicle speed

The process of generating the predicted vehicle speed is shown on the right side of fig. 6, and is essentially the reverse of the process of generating the TPMs. The method comprises the following specific steps:

step 1, obtaining future path coordinates and path information

When a driver inputs a destination in a vehicle-mounted navigation system, the system acquires future path coordinates (GPS coordinates) and path information (path length, road type, signal lamp position and traffic light conversion time, corner position and corner radius, deceleration strip position, traffic flow data and the like) from a GPS and an electronic map. The system processes the data, eliminates the abnormal points and resamples according to the length of the driving path.

Step 2 generating reference conditions

And generating a modal working condition called a reference working condition according to the path coordinates and the path information. The method for generating modal conditions will be described below by taking a city path shown in fig. 7 as an example. The route in fig. 7 includes 3 road surface types, and the route is divided into a plurality of segments by route nodes (traffic lights, corners, speed bumps, and the like). Generally, a driver may decelerate or stop while passing through a path node, and then accelerate the vehicle. Therefore, an acceleration and deceleration phase, i.e., a period of operation, may be included between the 2 nodes. The road type, the latitude lat and the longitude lon of the node can be obtained from an electronic map and a navigation system.

A. Distance L between two nodes B_ABCan be calculated from the coordinates of the nodes, i.e.

Wherein R (m) is the radius of the earth, lat_AAnd lat_BLatitude values of points A and B, lon_AAnd lon_BThe longitude values of points A and B are respectively.

When the radius of a corner on a path is large, or when there is a traffic light at an intersection, there are multiple nodes_AB<L_minAnd selecting the middle position of the 2 nodes as the position of the fusion node.

When the node is only a traffic light (vehicle going straight), the driver will go straight through the node since there is a certain probability that it will be displayed as a green light. Through the ITS system, the traffic light and traffic light duration can be obtained, and the occurrence probability of the green light is calculated, namely

Generating a random number n of 0-1 by a Matalb random function_lgIf n is_lg＜p_lgThen the traffic light node will be cancelled (considered a green light), otherwise the traffic light node will be retained.

The nodes are arranged according to positions, and the nodes of the same road type are connected by the average vehicle speed of the type to form the reference working condition of the future path, as shown in fig. 13. The reference condition shows the road type, length, average vehicle speed, node position and the like. It is the 'skeleton' of the predicted working condition to be generated, and a working condition segment is formed between 2 nodes.

Step 3TPMs type matching

After generating the "modal conditions," the system identifies the driving style of the current vehicle driver and matches the corresponding types of TPMs.

When the driving range of the vehicle on a certain road type is less than a certain threshold value S_r1(e.g., 50km) the system may not be able to obtain sufficient driver travel data, and at this point the system employs TPMs of a "normal type" driving style.

When the driving mileage exceeds S_r1The system may then identify the driving style of the driver. The driving style index J of the driving on the road type i is calculated according to the formula (3.2)_d(i) And calculate and J_dnor1(i)、J_dnor2(i) And J_dnor3(i) The absolute value of the difference. The driving style type with the smallest absolute value of the difference is the driving style of the driver. The TPMs of that driving style are used as the TPMs for that driver on that type of road. Repeating the above stepsThe driving style of the driver on each type of road is identified and the TPMs are matched.

When the driving mileage of the vehicle on the road of the type exceeds S_r2(e.g., 100km), the system may generate the TPM for the driver on the class of roads, as described above. The system will expand the TPMs and use the driver specific TPMs.

Therefore, the driver historical working condition data are counted to form TPMs taking the driving style and the road type as indexes, the storage capacity is small, the calculation is convenient, and the TPMs can be updated in real time. Along with the increase of the driving mileage, the representativeness of the probability transfer matrix to the driving behavior characteristics of the driver is enhanced, but the storage capacity is not greatly increased; when the stored road types and driving styles TPMs are increased, the prediction accuracy is greatly improved, and the storage amount is only slightly increased, so that the vehicle-mounted system requirement is met.

Step 4 generating working condition section

The following description takes the first working condition section on the "type II" road of FIG. 13 as an example to illustrate the algorithm for generating the working condition section, FIG. 14 is a schematic diagram of the algorithm for generating the working condition section, a working condition section exists between a first node A (corner) and a second node B (corner) on the "type II" road of FIG. 13, and the distance L between the two nodes_ABCalculated from equation (3.5) and rounding the result to 1m as L in FIG. 14_AB10 m; the distance ratio of the acceleration section is obtained by the statistics of the formula (3.3), and r is set_da0.6; the length of the acceleration section is 6m and the length of the deceleration section is 4 m.

Let the interval of generating vehicle speed points be 1m and the speed value interval be 0.1 m/s. The speed v of the vehicle at the starting point of the working condition section₀The vehicle speed at the end of the previous operating range, v in this example₀5.0 m/s. The vehicle speed of the next meter will be determined by the transition probabilities of the corresponding state points of the acceleration TPMs for that road type, in this example, 5.0m/s of vehicle speed has three states 5.1m/s, 5.2m/s and 5.3m/s that can be transitioned with transition probabilities of 0.25, 0.5 and 0.25. Generating a random number between 0 and 1 by using a random number generation function Round () of Matlab to determine a transition state, namely

As shown in fig. 14, if the generated random number is 0.54, the next state v is based on equation (3.7)₁5.2 m/s. And repeating the steps to sequentially generate the speed points of the acceleration section until the length of the acceleration section is met. Then the last speed point v of the acceleration section_endAs the starting point of the deceleration section, the deceleration TPMs of the type of road are selected, and the vehicle speed point of the deceleration section is generated according to the same algorithm as that of the acceleration section (FIG. 14). Until the deceleration section length is satisfied.

Step 5 working condition section integration and filtering

And (4) repeating the step (4) and sequentially generating the speeds of other working condition sections until the length of the section of the type is met. The generated road segment speed needs to be smoothed and resampled, in this example, the generated speed is smoothed by a Butterworth filter.

Step 6, generating a predicted vehicle speed curve

When the road type changes, corresponding TPMs are required to be selected, and the vehicle speed curves of all the road sections are sequentially generated until the predicted vehicle speed distance is equal to the future path length. The predicted vehicle speed curve in fig. 13 is the predicted vehicle speed generated on the city path.

And establishing a path information-based electric vehicle remaining driving mileage prediction simulation model by adopting Matlab/Simulink according to the remaining driving mileage prediction algorithm framework and the three mathematical models. The urban road working condition shown in FIG. 7 is used for simulation and verification of the effectiveness and precision of the model. In the simulation, the actual measurement driving data of 2 drivers with different driving styles are respectively adopted, and the TPMs of the two drivers under the urban working condition are generated. Fig. 15 is a measured and predicted curve of energy consumption of 2 drivers on an urban road. It can be seen that driver 2 has a higher energy consumption than driver 1, with a difference of about 8%, which proves that different driving styles have a greater impact on energy consumption. The model provided by the text is adopted to predict the energy consumption, and as can be seen from fig. 15, the predicted energy consumption of 2 drivers can be well matched with the respective actually measured energy consumption, which shows that the proposed algorithm can adapt to the influence of different driving styles on the energy consumption. Fig. 16 shows the RDR actual and predicted curves of the driver 1 on the urban road, which are well matched. To verify the accuracy of the RDR prediction model presented herein, an end point relative error (TRE) and a Root Mean Square Error (RMSE) were used for evaluation.

TRE is defined as

Wherein, P_tTo predict curve endpoint values (energy consumption values to take trip end points for energy consumption prediction, RDR values to take start points for RDR prediction); m_tAre actual curve endpoint values. This value characterizes the accumulated prediction error for the entire run.

RMSE is defined as

Wherein i is a curve sampling point; n is the total number of sampling points; p_iThe predicted value at a curve sampling point i is obtained; m_iIs the measured value at the sampling point i of the curve. This value characterizes the goodness of fit of the predicted curve to the measured value.

Because the prediction curves obtained by the method are randomly generated, the prediction curves obtained by each operation are not completely the same. Therefore, the prediction program was run a plurality of times, and the prediction error of the driver 1 on the urban road was calculated, with the results shown in table 2. It can be seen that the energy consumption prediction TRE is 0.86% and the RDR is 1.1%, indicating that the cumulative error is small. In addition, the error from the RMSE is small, and the matching degree between the predicted curve and the measured value is good.

TABLE 2 prediction error of energy consumption and RDR under urban road conditions

Claims

1. A pure electric vehicle remaining mileage model prediction method based on path information is characterized by comprising the following steps:

step one, establishing a vehicle speed prediction model to generate a future path predicted vehicle speed: analyzing a certain amount of historical driving data of drivers, extracting path information, and generating a state transition probability matrix according with the behavior characteristics of the drivers; generating a predicted vehicle speed controlled by the future road information based on the road information of the future path and the corresponding state transition probability matrix;

the method comprises the following steps:

1) generating a transition probability matrix of the driver on different road surfaces according to the historical vehicle speed data and the path information of the driver:

1.1) collecting working condition data and extracting path information, respectively collecting real vehicle working condition data and driver working condition data, evaluating and classifying driver driving style, and indicating J driving style of driver on a certain type of road i_d(i) Comprises the following steps:

J_d(i)＝w₁·e_avg(i)+w₂v_m(i)+w₃v_max(i)+w₄a_am(i)+w₅a_bm(i)

wherein e is_avgAverage energy consumption rate for the driver on this type of road, (kW/km); v. of_mAverage speed, (km/h); v. of_maxMaximum speed, (km/h); a is_amIs the average acceleration, (m/s)²)；a_bm(m/s) is the average deceleration (absolute value)²)；w_1～5Is a weight coefficient;

1.2) dividing the running condition sections according to the road types;

1.3) continuously dividing the classified working condition segments into an acceleration stage and a deceleration stage and calculating the distance ratio of the acceleration stage;

1.4) stage working condition data gridding: the method comprises the steps that speed data of an acceleration connection section and a deceleration connection section of the same driver on the same road are collected together, and the speed data are subjected to gridding and interpolation again;

1.5) scanning the stage working condition data after gridding, and counting the state transfer number;

1.6) generating a transition probability matrix;

2) generating a predicted vehicle speed based on the future path information in combination with the TPMs:

2.1) acquiring future path coordinates and path information of the vehicle from the vehicle-mounted system;

2.2) generating a reference working condition according to the path coordinate and the path information, wherein the reference working condition displays the road type, the length, the average speed and the node position, and a working condition segment is formed among 2 nodes;

2.3) after the reference working condition is generated, the system identifies the driving style of the current vehicle driver and matches a transition probability matrix of a corresponding type;

2.4) sequentially generating working condition sections on various types of roads according to the reference working condition;

2.5) carrying out working condition section integration and filtering;

2.6) generating a predicted vehicle speed curve: when the road type changes, selecting a corresponding transition probability matrix, and sequentially generating a vehicle speed curve of each road section until the predicted vehicle speed distance is equal to the future path length;

2. The pure electric vehicle remaining mileage model prediction method based on path information as set forth in claim 1, wherein the estimation of the driving parameters by the second-step parameter estimation model specifically comprises:

1) initial value estimation of the rolling resistance coefficient:

initial value f of rolling resistance coefficient_r0The fitting formula is

Wherein e is_i(i is 1 to 3) is a fitting coefficient, k_iCorrecting the coefficient for the road surface type;

2) calculating the road surface gradient:

the road surface gradient a can be calculated through a geographic information system and GPS path longitude and latitude_slop(rad), i.e.

Wherein Δ h (m) is the height difference between two consecutive measurement points;

3) and (3) dynamically estimating the finished automobile mass and rolling resistance coefficient:

dynamic estimation of whole vehicle mass m based on recursive least square estimation algorithm_vAnd coefficient of rolling resistance f_r，

During the running of the vehicle, the motor outputs power P_m(W) is

Wherein, T_m(N.m) is motor torque;

the motor rotating speed; f_r(N) is rolling resistance; f_aero(N) is air resistance; f_g(N) is the slope drag; f_m(N) is acceleration resistance; f_fric(N) is driveline friction at the wheels;

the standard form of writing the above equation as a linear estimate is

Wherein the content of the first and second substances,

wherein, J_w(kg·m²) Is the rotational inertia of the wheel; r (m) is the tire radius; i.e. i_gIs the transmission ratio of the gearbox; a is_x(m/s²) Is the longitudinal vehicle acceleration; v (m/s) is the vehicle running speed;

g(m/s²) Acceleration of gravity, α (rad) road gradient, f_rIs the rolling resistance coefficient; m (kg) is the BEV whole vehicle mass;

the following equation is minimized:

its recursive solution is

Wherein the content of the first and second substances,

3. the pure electric vehicle remaining mileage model prediction method based on path information as claimed in claim 1, wherein the energy consumption prediction model in the third step is established by:

establishing an energy consumption prediction model by adopting a reverse modeling method, wherein the input of the model is the vehicle speed, and the output of the model is the battery output power P_bat(W), namely:

wherein, F_w(N) is the driving force of the automobile; v (m/s) is the running speed of the vehicle and is obtained by the vehicle speed prediction model;

rolling resistance F_r(N) is calculated from the following formula:

F_r＝f_rm_vgcos(α_slop)

in the formula (f)_rIs a coefficient of rolling resistance, m_v(kg) vehicle mass α_slop(rad) is the road surface gradient, all calculated by the parameter estimation model; g (m/s)²) Is the acceleration of gravity;

air resistance F_aero(N) is calculated from the following formula:

in the formula, rho (kg/m)³) Is the air density; a. the_f(m²) The frontal area of the automobile; c_dIs the air resistance coefficient; v_win(m/s) is the wind speed in the direction of travel;

slope resistance F_g(N) is calculated from the following formula:

F_g＝m_vgsin(α_slop)

acceleration resistance F_m(N) is calculated from the following formula:

in the formula, J_w(kg·m²) Is the rotational inertia of the wheel; j. the design is a square_m(kg·m²) Is the rotational inertia of the motor; r (m) is the tire radius; i.e. i_gIs the transmission ratio of the gearbox; dv/dt (m/s)²) Is the longitudinal vehicle acceleration;

P_{pt_loss}(W) power loss for the vehicle driveline;

P_aux(W) consumes power for the electrical accessory.

4. The pure electric vehicle remaining mileage model prediction method based on path information as claimed in claim 1, wherein the establishment process of the remaining energy prediction model in the third step is as follows:

battery residual energy E_rue(kWh) can be calculated from the following formula:

E_rue＝Q₀·SoH·C_tempU_t,nom·(SOC-SOC_end,nom)

wherein Q is₀(Ah) is the new battery rated capacity; c_tempA battery temperature correction coefficient; u shape_t,nom(V) is the nominal terminal voltage of the battery; SOC is the state of charge of the battery; SOC_end，nomDischarging the lowest battery SOC; SoH is the state of health of the battery, defined as:

in the formula, Q_bat(Ah) is the current battery rated capacity;

estimating the SOC by adopting an ampere-hour method:

in the formula, I (A) is the output current of the battery.

5. The pure electric vehicle remaining mileage model prediction method based on path information as claimed in claim 1, wherein the calculation method of the remaining mileage in the third step is:

separating from the energy consumption prediction model and the residual energy prediction modelRespectively calculating to obtain the energy consumption rate e_avgAnd battery residual energy E_rueThen, the current time t can be calculated by the following formula₂RDR of remaining driving range_cal(km), i.e.

Further calculated by:

wherein RDR_cal(t₁) To be in the past t₁RDR prediction result of time delta L_cum(t₁,t₂) Is from t₁To t₂The actual distance traveled at any moment is obtained by the integration of the actual vehicle speed;

RDR prediction result RDR finally displayed_dis(km) can be calculated from the following formula:

RDR_dis(t₂)＝w_disRDR_cal(t₂)+(1-w_dis)RDR_cum(t₂)

wherein, w_disIs a weight coefficient with a value range of [0, 1%]。