阅读说明：本技术 燃料电池混合动力汽车的预测性能量管理方法 (Predictive energy management method for fuel cell hybrid electric vehicle ) 是由 马彦 李成 王思雨 于 2021-10-28 设计创作，主要内容包括：一种燃料电池混合动力汽车的预测性能量管理方法,属于新能源汽车电源技术领域。本发明的目的是在保持锂电池SOC处在参考值附近的同时,保证燃料电池以高效率运行,在满足上述条件的情况下使氢气消耗达到最小的燃料电池混合动力汽车的预测性能量管理方法。本发明步骤是：混合动力系统的拓扑结构；燃料电池混合动力汽车模型建立；车速预测与车辆驾驶模式识别；基于模型预测控制的燃料电池混合动力汽车在线能量管理方法设计。本发明达到了经济性、荷电状态和燃料电池效率三者之间的相对最优。设计了多目标的燃料电池混合动力汽车能量管理方法。在稳定保持锂电池SOC的前提下,大幅度提高了燃料电池效率,显著降低了氢气消耗量。(A predictive energy management method for a fuel cell hybrid electric vehicle belongs to the technical field of new energy vehicle power supplies. The invention aims to provide a predictive energy management method for a fuel cell hybrid electric vehicle, which can ensure that a fuel cell operates at high efficiency while keeping the SOC of a lithium battery close to a reference value, and can minimize the consumption of hydrogen under the condition of meeting the above conditions. The method comprises the following steps: topology of the hybrid power system; establishing a fuel cell hybrid electric vehicle model; vehicle speed prediction and vehicle driving mode recognition; the fuel cell hybrid electric vehicle on-line energy management method based on model predictive control is designed. The invention achieves the relative optimization among economy, state of charge and fuel cell efficiency. A multi-objective fuel cell hybrid electric vehicle energy management method is designed. On the premise of stably keeping the SOC of the lithium battery, the efficiency of the fuel battery is greatly improved, and the hydrogen consumption is obviously reduced.)

1. A predictive energy management method for a fuel cell hybrid vehicle,

s1, topological structure of the hybrid power system;

s2, establishing a fuel cell hybrid electric vehicle model;

the method is characterized in that:

s3, vehicle speed prediction and vehicle driving mode recognition

First, vehicle driving road condition recognition

Intercepting running condition section by random number method

T₀＝τ(T-ΔT) (21)

Wherein, T₀Is the start time of each sampling period, τ is a random number from 0 to 1, T is the length of each packet, Δ T is the length of each sample segment;

extracting feature vectors in each sampling segment

l(k)＝[v_ave,ε,a_max,a_min]^T (22)

Wherein v is_aveFor the average speed of each segment, ε is the idle time ratio, a_maxIs the maximum value of acceleration, a_minIs the minimum value of the acceleration;

normalizing the feature vector

Wherein l^*Is normalized characteristic value, i is original characteristic, l is_minAnd l_maxThe smallest and largest original features, respectively;

second, vehicle speed prediction

Discretizing acceleration and velocity of a data set into finite-length sequences

WhereinRepresents N_aThe number of discrete acceleration data is such that,represents N_rDiscrete velocity data;

the transition probability matrix is

Wherein the content of the first and second substances,indicates when the vehicle speed is equal to vⁱThe acceleration is measured in n sampling periods by aⁱIs changed into a^jThe probability of (d);

computing using maximum likelihood estimation

Wherein the content of the first and second substances,indicates when the vehicle speed is equal to vⁱAt an acceleration of aⁱThe number of the data of (2),indicates when the vehicle speed is equal to vⁱThe acceleration is measured in n sampling periods by aⁱIs changed into a^jThe number of data of (2);

satisfy the requirement of

S4 model prediction control-based online energy management method design for fuel cell hybrid electric vehicle

Defining performance indicators for battery SOC

Wherein the content of the first and second substances,is the two-norm of the matrix; s (SOC (k)) is a penalty term in the cost function, and the weight of the penalty term can be adjusted according to the balance relation between the SOC and other performance indexes;

defining a second performance index

Wherein, N-1 is the length of the prediction time domain;

defining a third performance index

Wherein eta is_fc(k) Is the fuel cell efficiency;

according to the performance indexes and the constraint conditions, the final objective function is

Wherein x is₁(k) Is SOC (k), x₂(k) Is P_fc(k) θ (k) is the vehicle speed at time k;

solving the objective function to obtain a control sequence

Wherein the content of the first and second substances,representing the fuel cell stack current at time k + 1. Solving an objective function at the moment k and obtaining a control sequence u_kThen, the first element in the control sequence is controlled at the time k +1Acting on the system and then solving for J_N(u_k+1,x₀) (ii) a By repeating the above processes, real-time energy distribution between the fuel cell and the lithium battery can be realized.

Technical Field

The invention belongs to the technical field of new energy automobile power supplies.

Background

With the proposal of the national "double carbon" target, the emission of automobiles is facing to be more severe, and the electric driving and low emission of the automobiles are an irreversible development trend at present. Among various vehicle-mounted power supply types, the fuel cell has the characteristics of zero emission, high efficiency, high energy density and long endurance time, so that the fuel cell is a good vehicle-mounted power supply choice. However, since the fuel cell generates electric energy by means of electrochemical reaction, and air as reactant is delivered to the stack to participate in the reaction through the air compressor and the intake manifold, the single fuel cell power supply often has the characteristic of slow response, and is difficult to meet the rapidly changing vehicle energy demand. By establishing the hybrid power system of the fuel cell and the lithium battery, the energy response of the whole power system can be greatly improved, the lithium battery plays a role in recovering braking energy in the power supply process and serves as an auxiliary power supply when the power of the fuel cell is insufficient, and the power performance of the vehicle is greatly improved.

In a hybrid power system of a fuel cell and a lithium battery, the fuel cell is a main energy source and can output electric energy smoothly, and the lithium battery is an energy storage device which can serve as an auxiliary power supply. During the running process of the fuel cell hybrid electric vehicle, the electric energy required by the driving motor is determined by the vehicle speed and vehicle parameters and is provided by the fuel cell and the lithium battery together. If the output of the fuel cell is too large, the consumption of hydrogen is excessive, and the economy of a hybrid power system is poor; if the output of the lithium battery is too large, the SOC (State of Charge) of the lithium battery is too low, and the cruising ability of the vehicle is affected. Obviously, during the running process of the vehicle, the reasonable energy management method can improve the economy of the hybrid system and the power performance of the vehicle. Therefore, it is necessary to develop an effective energy management method to solve the problem of optimizing the real-time energy distribution between the fuel cell and the lithium battery.

Generally, energy management methods for hybrid vehicles are classified into two types, rule-based energy management methods and optimization-based energy management methods. The rule-based method distributes the power of the fuel cell and the power of the lithium battery according to the pre-designed energy management rule according to different types of running conditions of the automobile and different ranges of the SOC of the lithium battery. The rule-based method is simple and easy to implement and low in calculation amount, but the formulation of the energy management rule usually depends on expert experience, and the optimal solution of energy distribution cannot be obtained. The optimization-based method can be further divided into an offline optimization method and an online optimization method. Dynamic programming, a minimum principle and a series of bionic optimization algorithms such as a genetic algorithm all belong to offline optimization methods, and a global optimal solution of energy distribution can be obtained, but the offline optimization methods have great limitations in practical application. Compared with global optimization, the optimized solution of the online optimization method can be updated in real time, is more suitable for engineering practice and is a development trend in the present and future, but the online optimization energy management method has many problems to be solved:

1. in the design of optimization targets, the traditional method only considers the SOC of the lithium battery, is single-target optimization, and does not design an energy management method from the economical point of view, which can increase unnecessary hydrogen consumption.

2. Ohmic polarization, activation polarization and concentration polarization phenomena exist in the fuel cell, and the fuel cell has self energy-efficiency characteristic relation. When the fuel cell is operated in a low efficiency region, the polarization phenomenon may cause a large voltage loss, thereby reducing the energy utilization rate.

3. In the method following the principle of rolling time domain optimization, a local optimal solution in a future short time domain is obtained in each sampling period. In practical applications, the future vehicle speed information is unpredictable. The conventional processing method for the problem is to calculate the vehicle speed at the current moment as the future vehicle speed on the assumption that the vehicle runs at a constant speed, which obviously causes the optimization effect to be poor.

In the process of establishing the vehicle speed prediction model, the establishment of the transition probability matrix is an important step. Because the driving data of the vehicles under different road conditions are very different, the divergence degree of the transition probability matrix established by the traditional method is very high, and the prediction precision is insufficient.

Disclosure of Invention

The invention aims to provide a predictive energy management method for a fuel cell hybrid electric vehicle, which can ensure that a fuel cell operates at high efficiency while keeping the SOC of a lithium battery close to a reference value, and can minimize the consumption of hydrogen under the condition of meeting the above conditions.

The method comprises the following steps:

s1, topological structure of the hybrid power system;

s2, establishing a fuel cell hybrid electric vehicle model;

s3, vehicle speed prediction and vehicle driving mode recognition

First, vehicle driving road condition recognition

Intercepting running condition section by random number method

T₀＝τ(T-ΔT) (21)

Wherein, T₀Is the start time of each sampling period, τ is a random number from 0 to 1, T is the length of each packet, Δ T is the length of each sample segment;

extracting feature vectors in each sampling segment

l(k)＝[v_ave,ε,a_max,a_min]^T (22)

Wherein v is_aveFor the average speed of each segment, ε is the idle time ratio, a_maxIs the maximum value of acceleration, a_minIs the minimum value of the acceleration;

normalizing the feature vector

Wherein l^*Is normalized characteristic value, i is original characteristic, l is_minAnd l_maxThe smallest and largest original features, respectively; second, vehicle speed prediction

Discretizing acceleration and velocity of a data set into finite-length sequences

WhereinRepresents N_aThe number of discrete acceleration data is such that,represents N_rDiscrete velocity data;

the transition probability matrix is

Wherein the content of the first and second substances,indicates when the vehicle speed is equal to vⁱThe acceleration is measured in n sampling periods by aⁱIs changed into a^jThe probability of (d);

computing using maximum likelihood estimation

Wherein the content of the first and second substances,indicates when the vehicle speed is equal to vⁱAt an acceleration of aⁱThe number of the data of (2),indicates when the vehicle speed is equal to vⁱThe acceleration is measured in n sampling periods by aⁱIs changed into a^jThe number of data of (2);

satisfy the requirement of

S4 model prediction control-based online energy management method design for fuel cell hybrid electric vehicle

Defining performance indicators for battery SOC

Wherein the content of the first and second substances,is the two-norm of the matrix; s (SOC (k)) is a penalty term in the cost function, and the weight of the penalty term can be adjusted according to the balance relation between the SOC and other performance indexes;

defining a second performance index

Wherein, N-1 is the length of the prediction time domain;

defining a third performance index

Wherein eta is_fc(k) Is the fuel cell efficiency;

according to the performance indexes and the constraint conditions, the final objective function is

Wherein x is₁(k) Is SOC (k), x₂(k) Is P_fc(k) θ (k) is the vehicle speed at time k;

solving the objective function to obtain a control sequence

Wherein the content of the first and second substances,representing the fuel cell stack current at time k + 1. Solving an objective function at the moment k and obtaining a control sequence u_kThen, the first element in the control sequence is controlled at the time k +1Acting on the system and then solving for J_N(u_k+1,x₀) (ii) a By repeating the above processes, real-time energy distribution between the fuel cell and the lithium battery can be realized.

The invention has the beneficial effects that:

1. the method of the invention achieves the relative optimization among economy, state of charge and fuel cell efficiency. A multi-objective fuel cell hybrid electric vehicle energy management method is designed. On the premise of stably keeping the SOC of the lithium battery, the efficiency of the fuel battery is greatly improved, and the hydrogen consumption is obviously reduced;

2. by introducing the method for predicting the vehicle speed, the problem of the energy demand loss of the future vehicle in the optimal control problem solving is solved, and the economy of the energy management method is further improved;

3. a road condition identification method is introduced in the vehicle speed prediction, a driving information database is divided into three types, and three transition probability matrixes are respectively established, so that the accuracy of the vehicle speed prediction is improved.

Drawings

FIG. 1 is a topological block diagram of a hybrid powertrain system;

FIG. 2 is a diagram of a transition probability matrix;

FIG. 3 is a road condition recognition accuracy test cycle chart;

fig. 4 is a road condition recognition result diagram;

FIG. 5 is a diagram of vehicle speed prediction results;

FIG. 6 is a flow diagram of a predictive energy management method implementation;

fig. 7 is a hydrogen consumption map during vehicle travel;

FIG. 8 is a graph showing SOC variation during driving of a vehicle;

fig. 9 is a graph showing the operating efficiency of a fuel cell in a conventional method;

fig. 10 is a graph of the operating efficiency of a fuel cell in accordance with the present invention.

Detailed Description

The invention aims to solve the problems that the existing energy management method is not good enough in economy, low in fuel cell efficiency and lack of future vehicle speed information.

The invention aims to solve the problems in the traditional energy management method that: 1. the economics of the energy management process, which would increase the unnecessary consumption of hydrogen, are not taken into account; 2. ohmic polarization, activation polarization and concentration polarization phenomena exist in the fuel cell, and the fuel cell has self energy-efficiency characteristic relation. When the fuel cell operates in a low-efficiency region, the polarization phenomenon can cause larger voltage loss, thereby reducing the energy utilization rate; 3. in practical applications, future vehicle speed information is unpredictable. The traditional processing method for the problem is to assume that the vehicle runs at a constant speed and then perform calculation, which obviously causes the optimization effect to be poor; 4. the establishment of the transition probability matrix is an important step in the process of establishing the vehicle speed prediction model. Because the driving information of the vehicles under different road conditions is very different, the divergence degree of the transition probability matrix established by the traditional method is very high, and the prediction precision is insufficient.

The invention is described in four parts: topology of the hybrid power system; establishing a fuel cell hybrid electric vehicle model; vehicle speed prediction and vehicle driving mode recognition; the fuel cell hybrid electric vehicle on-line energy management method based on model predictive control is designed.

Topology of the hybrid system: and the connection mode of the fuel cell and the lithium battery and the bus.

Establishing a fuel cell hybrid electric vehicle model: and establishing an integral model of the fuel cell hybrid electric vehicle, wherein the integral model comprises a vehicle longitudinal dynamics model, a driving motor model, a fuel cell model and a lithium battery internal resistance model.

Vehicle speed prediction and vehicle driving pattern recognition: in order to solve the problem that the conventional energy management method is lack of the future vehicle speed, a transition probability matrix of the vehicle speed and the acceleration is established according to the historical driving information of the vehicle to predict the future vehicle speed. Considering that the divergence degree of the transition probability matrix is too large due to the large difference of the vehicle driving information under different road conditions, the road conditions of the vehicle driving are divided into three categories: the method is characterized in that a vehicle driving mode identification method is also designed, three transition probability matrixes corresponding to different road conditions are respectively established according to the three road conditions, and the accuracy of vehicle speed prediction is improved by adopting a mode of firstly identifying and then predicting.

The fuel cell hybrid electric vehicle on-line energy management method based on model predictive control is designed as follows: in order to solve the problem that the traditional energy management method does not consider the economical efficiency and the efficiency of the fuel cell, a multi-objective optimization method is designed. The fuel cell is guaranteed to operate at high efficiency while keeping the SOC of the lithium battery in the vicinity of the reference value, and consumption of hydrogen is minimized while satisfying the above conditions.

The technical solution proposed by the present invention is further illustrated and explained with reference to the accompanying drawings:

1. topology of hybrid power system

In the fuel cell hybrid electric vehicle, a fuel cell, a lithium battery and a driving motor are connected through a bus. According to the connection mode of the fuel cell and the bus, the topological structure of the fuel cell-lithium battery hybrid power system can be divided into two types: the first is direct mixing and the second is indirect mixing. The invention adopts an indirect hybrid mode, the topological structure of the hybrid system is shown in figure 1, the fuel cell is connected with the bus through the unidirectional DC-DC converter, and the lithium cell is directly connected with the bus. The control of the fuel cell stack current can be realized by controlling the duty ratio of the DC-DC converter.

2. Fuel cell hybrid electric vehicle model establishment

Vehicle longitudinal dynamics model

The vehicle longitudinal dynamic model describes the relation between the stress condition of the vehicle and the vehicle speed. When the vehicle is running, the stress condition is as follows

F_trac＝F_roll+F_aero+F_grade+F_inertia (1)

Wherein, F_tracIs the driving force applied to the vehicle, F_rollTo rolling resistance, F_aeroAs air resistance, F_gradeAs ramp resistance, F_inertiaFor acceleration resistance, respectively calculated as

F_roll＝fmg cosθ (2)

F_grade＝mg sinθ (4)

Wherein m is the half-load mass of the vehicle, f is the wheel rolling resistance coefficient, theta is the gradient of the running road surface, A_fArea facing the wind, C_dIs the air resistance coefficient, u is the running speed, δ is the rotating mass conversion coefficient, g is the gravitational acceleration, and t is the time.

It follows that the driving force is a variable related to the vehicle speed, i.e.

F_trac＝h(u) (6)

The moment T borne by the vehicle wheel_wheelAnd angular velocity of rotation omega_wheelThe relationship with the running driving force of the automobile and the vehicle speed u can be expressed as

T_wheel＝F_trac·r (7)

Wherein r is the wheel radius.

Second, drive the motor model

The modeling method of the driving motor can be generally divided into a theoretical model built according to a driving principle and an efficiency model according to a table look-up method. The theoretical model of the motor is generally used for researching the working characteristics and the control performance of the motor, more detailed parameter information of the motor needs to be known, and the modeling is more complex; the efficiency model according to the table look-up method is mainly based on the Map of the motor, the simulation speed is high, and the method is suitable for researching the energy management method of the whole vehicle. Because the research of the invention focuses on the working principle of driving the motor but the energy distribution of the fuel cell hybrid electric vehicle, the efficiency model based on the table lookup is selected to build the motor model. The motor Map table information is provided by a manufacturer or an actual vehicle experiment and can be regarded as a table look-up function of the motor output characteristics, namely

η_MG＝f_MG(T_MG,ω_MG) (9)

Wherein, T_MGFor output of torque, omega, of the motor_MGIs angular velocity, η_MGThe working efficiency of the motor is improved. And searching an efficiency MAP table according to the output torque and the output rotating speed of the motor at a certain moment to obtain the working efficiency of the motor at the moment. f. of_MG(. cndot.) is an efficiency look-up table function.

According to the dynamics of the whole vehicle, the torque and the rotating speed of the motor can be obtained by the following formula

ω_MG＝ω_wheeli_fd (11)

Wherein i_fdIs a main reduction ratio, η_DLIs the transmission efficiency of the main speed reduction differential.

The mechanical power P of the motor_MGAnd electric power P_eAre respectively

Wherein, P_MGWhen the power is less than 0, regenerative braking is performed, and energy flows to the storage battery from the motor.

Power battery model

In the invention, an internal resistance model is adopted to model the power battery system. The storage battery system can be equivalent to a series circuit of a power supply and an internal resistance, so that the dynamic processes of terminal voltage and SOC of the storage battery system in the charging and discharging processes are simulated.

According to circuit principles, a model can be represented as

V_BP＝V_OC-I_BP·R (13)

Wherein, V_OCIs the open circuit voltage of the battery, I_BPIs the current of the battery, R is the internal resistance of the battery, V_BPIs the voltage across the cell, P_BPIs the power of the battery.

The SOC of the battery is used for representing the residual capacity condition of the battery and is expressed by formula

Wherein Q represents the remaining battery power, Q_cRepresenting the battery capacity.

In the internal resistance model, the estimation of the SOC selects an ampere-hour integral method, and the equation of the method is expressed as

Wherein, I is the charging and discharging current of the battery, I is more than or equal to 0 to represent discharging, I is less than 0 to represent charging, and Q is_intIs the initial charge of the battery, η_BPThe value at discharge is 1 for the coulombic efficiency of the cell.

Since there is a limit in the capacity of the battery, the maximum charge and discharge power thereof may be expressed as

Wherein, P_BP,maxThe maximum power V that can be reached during the charging and discharging of the battery_minIs the lowest value of the battery terminal voltage, V_maxThe highest value of the battery voltage. The open-circuit voltage of the battery is a function related to the SOC of the batteryThe internal resistances at the battery charging and discharging times are different, but can be expressed as a function of the SOC.

Fourth, fuel cell model

Since the optimization objectives of the energy management method of the present invention include improving the fuel cell efficiency and reducing the hydrogen consumption, the fuel cell system model establishes a correspondence relationship that mainly describes the hydrogen consumption rate and the power and efficiency of the fuel cell, wherein the hydrogen consumption rate is

Wherein eta is_fcFor fuel cell efficiency, η, is obtained from the fuel cell system efficiency-fuel cell system power output curve_fc＝f_fc(P_fcs)，P_fcsLHV is the lower heating value of hydrogen for the power of the fuel cell system.

Output power P of action of direct current converter (DC-DC)_DCDC

P_DCDC＝η_DCDCP_fcs (19)

Wherein eta is_DCDCFor the DC-DC model conversion efficiency, the table can be looked up for calculation.

The power balance relationship among the components of the fuel cell-lithium battery hybrid power system is

P_e+P_aux＝P_BP+P_DCDC (20)

Wherein, P_auxFor accessory power, the vehicle is set to a fixed value while traveling.

3. Vehicle speed prediction and vehicle driving pattern recognition

First, vehicle driving road condition recognition

Generally, driving scenes of automobiles include expressways and urban roads. Automobiles on urban roads may encounter traffic jams or be clear. Different driving scenes can generate different driving data, and the driving modes of the automobile can be classified according to the driving data. The classified driving data contains the characteristics of different driving scenes, so that the vehicle speed is more easily predicted. The invention divides the driving mode of the automobile into three types: highway road conditions, urban congestion road conditions and urban unblocked road conditions.

In the face of the classification problem of small samples and low dimension, the support vector machine has the characteristics of high training speed, high recognition speed and high accuracy. Therefore, the invention adopts a Support Vector Machine (SVM) to solve the classification problem. It is worth mentioning that SVM can only solve the binary classification problem, whereas the classification problem herein is the ternary classification problem. Therefore, the present invention adopts a one-to-one method to solve this problem. The one-to-one method is to train three support vector machines, classify any two of the three modes, and finally adopt a hand-holding voting mode, wherein the mode with more votes is the classification result.

In the present invention, the entire data set includes 600 standard cycle segments 150 seconds long, which are divided into 10 packets, one packet being a test set and the remaining packets being a training set. 30 fragments are truncated in each training packet. Intercepting running condition section by random number method

T₀＝τ(T-ΔT) (21)

Wherein, T₀Is the start time of each sampling period, τ is a random number from 0 to 1, T is the length of each packet, and Δ T is the length of each sample segment, 150s in the present invention.

Extracting feature vectors in each sampling segment

l(k)＝[v_ave,ε,a_max,a_min]^T (22)

Wherein v is_aveFor the average speed of each segment, ε is the idle time ratio, a_maxIs the maximum value of acceleration, a_minIs the minimum value of the acceleration.

Since the magnitude of each feature is different, the feature vector is normalized

Wherein l^*Is normalized characteristic valueL is the original characteristic, l_minAnd l_maxThe smallest and largest original features, respectively. The preprocessed training set may be used to train the support vector machine. To maximize the use of the data set, the present invention uses a k-fold cross validation algorithm in support vector machine training. The SVM needs two training parameters, namely a penalty coefficient and a kernel function width. The k-fold cross validation algorithm takes 10 data packets in the invention as a test set in sequence, trains 10 different classifiers, and takes the highest precision as the final training result.

Second, vehicle speed prediction

On the premise of knowing the driving road condition of the vehicle, the vehicle speed can be respectively predicted according to different road conditions. Various speed prediction methods such as neural network prediction methods, markov prediction methods, deep learning methods, and methods based on multi-source information fusion exist. Among these methods, the markov predictor has advantages of high accuracy and high efficiency. In order to improve the accuracy of speed prediction, the invention has designed a road condition identification method. In the part, a Markov predictor is designed for three different driving road conditions (freeways, urban congestion and urban traffic), and a transition probability matrix is constructed by taking six standard driving conditions as a database. Discretizing acceleration and velocity of a data set into finite-length sequences

WhereinRepresents N_aThe number of discrete acceleration data is such that,represents N_rDiscrete speed data.

The transition probability matrix is

Wherein the content of the first and second substances,indicates when the vehicle speed is equal to vⁱThe acceleration is measured in n sampling periods by aⁱIs changed into a^jThe probability of (c).

Computing using maximum likelihood estimation

Wherein the content of the first and second substances,indicates when the vehicle speed is equal to vⁱAt an acceleration of aⁱThe number of the data of (2),indicates when the vehicle speed is equal to vⁱThe acceleration is measured in n sampling periods by aⁱIs changed into a^jThe number of data of (2).

Satisfy the requirement of

The established transition probability matrix is shown in fig. 2.

In order to verify the validity of road condition identification, a custom driving cycle as shown in fig. 3 is designed, and the cycle is composed of three representative standard cycle working conditions. The recognition result is shown in fig. 4, and in the vertical axis category, 1 represents highway road conditions, 2 represents urban congestion road conditions, and 3 represents urban smooth road conditions. The road condition identification method provided by the invention has the advantages that the accuracy can reach more than 95%, the identification error segment basically only appears during mode switching, and the conformity between the identification result and the real road condition is good when the mode switching does not occur. The accurate road condition identification result provides guarantee for vehicle speed prediction. Fig. 5 shows the result of vehicle speed prediction under the expressway condition. It can be seen that although the predicted vehicle speed has some error from the actual vehicle speed, the predicted vehicle speed and the actual vehicle speed have approximately the same tendency.

4. The specific design process of the fuel cell hybrid electric vehicle on-line energy management method based on model predictive control is designed in this section. The three optimization objectives in the present invention are first described in formula form. In order to ensure that the battery does not exhaust the electric quantity in the driving process of the automobile, the performance index of the SOC of the battery is defined

Wherein the content of the first and second substances,is the two-norm of the matrix. The performance index represents the error magnitude of the actual SOC and the reference SOC. Sometimes this performance index is also rewritten as an inequality constraint, but this has the disadvantage that the weighting relationship between the SOC error index and other performance indexes cannot be adjusted. S (SOC (k)) is a penalty term in the cost function. The weight of the SOC can be adjusted according to the balance relation between the SOC and other performance indexes.

Considering the economy of a hybrid system, hydrogen consumption should be as low as possible, and therefore a second performance index is defined

Wherein N-1 is the length of the prediction time domain. Such performance indicators can ensure that hydrogen consumption is minimized within each prediction horizon.

In order to improve the efficiency of the fuel cell and prolong the service life of the electric pile, a third performance index is defined

Wherein eta is_fc(k) Is the fuel cell efficiency. E (P) since the solution to the optimal control problem is a minimum finding process_fc(k) The sign of) is negative to ensure maximum efficiency.

There are certain physical constraints on the equipment in the fuel cell hybrid vehicle electrical system. It is necessary to incorporate these constraints in the design process of the optimal controller in view of the safety and lifetime of the components. The pem fuel cell system consists of an air compressor and other auxiliary devices, and the power change rate of the fuel cell should be limited in order to protect the motor of the air compressor. In addition, to prevent frequent activation and deactivation of the air compressor, the fuel cell power limit should also be limited. According to the performance indexes and the constraint conditions, the final objective function is

Wherein x is₁(k) Is SOC (k), x₂(k) Is P_fc(k) And θ (k) is the vehicle speed at time k.

Solving the objective function to obtain a control sequence

Wherein the content of the first and second substances,representing the fuel cell stack current at time k + 1. Solving an objective function at the moment k and obtaining a control sequence u_kThen, the first element in the control sequence is controlled at the time k +1Acting on the system and then solving for J_N(u_k+1,x₀). By repeating the above processes, real-time energy distribution between the fuel cell and the lithium battery can be realized. Hair brushA flow chart illustrating an implementation of the energy management method is shown in fig. 6.

Simulation process and results

1. Vehicle driving pattern recognition:

the first step is to establish a database required by SVM training. The data sources are standard cycle conditions, where highway condition data comes from the cycle: US06_ HWY, HWFET; the city unblocked road condition data comes from circulation: UDDS, INDIA _ URBAN _ SAMPLE; the urban congestion road condition data source is as follows: MANHATTAN, NYCC are provided. For the above six cycles, each cycle intercepts 100 driving data segments with the length of 150s by a random number method. Then, feature extraction is carried out on the 600 segments, and feature vectors have four dimensions including average speed, maximum acceleration, minimum acceleration and idle time ratio. Since the dimensions between the features are not uniform, normalization processing is also performed. This step is performed off-line.

The second step is SVM training. Because a single SVM can only solve the two-classification problem, but the invention relates to the three-classification problem, three SVM's are trained, any two of the three modes are identified pairwise, and the training process is optimized by using a k-fold cross validation method. This step is also performed off-line.

The third step is the identification process, which is performed online. Assuming that the current time is kth, firstly, intercepting the driving sections from the kth to the 1 st, carrying out feature extraction and normalization processing on the intercepted driving sections, and finally, identifying.

The recognition results are shown in fig. 3 and 4 in the specification. Where fig. 3 is a speed profile of a custom driving cycle and fig. 4 is a recognition result of the driving pattern of fig. 3. In the vertical axis category, 1 represents highway road conditions, 2 represents urban congestion road conditions, and 3 represents urban smooth road conditions. The road condition identification method provided by the invention has the advantages that the accuracy can reach more than 95%, the identification error segment basically only appears during mode switching, and the conformity between the identification result and the real road condition is good when the mode switching does not occur. The accurate road condition identification result provides guarantee for vehicle speed prediction.

2. Vehicle speed prediction

The first step is to establish a database required for vehicle speed prediction. The data sources are standard cycle conditions, where highway condition data comes from the cycle: US06_ HWY, HWFET; the city unblocked road condition data comes from circulation: UDDS, INDIA _ URBAN _ SAMPLE; the urban congestion road condition data source is as follows: MANHATTAN, NYCC are provided. For the above six cycles, the acceleration and the speed are discretized into integers. This step is performed off-line.

The second step is to establish a transition probability matrix. The invention predicts the vehicle speed according to the road condition, and establishes three transition probability matrixes, wherein each road condition corresponds to one transition probability matrix. A transition probability matrix of 1s is established, and the establishment process is described by equations (24) - (27). This step is also performed off-line.

The third step is vehicle speed prediction, which is performed online. And (3) assuming that the current moment is the kth second, calculating the acceleration and the speed of the vehicle at the current moment, and then combining the driving mode identification results to select a corresponding transition probability matrix. And (5) calculating to obtain a predicted vehicle speed sequence of t seconds in the future by searching for matching.

The vehicle speed prediction result is shown in figure 5 in the specification. It can be seen that although the predicted vehicle speed has some error from the actual vehicle speed, the predicted vehicle speed and the actual vehicle speed have approximately the same tendency.

3. The simulation flow of the online energy management method of the fuel cell hybrid electric vehicle based on model predictive control is shown in figure 6 in the specification. Assuming that the current time is kth second, firstly intercepting the driving sections from the kth to the 1 st second, carrying out feature extraction and normalization processing on the intercepted driving sections, and finally carrying out driving mode identification. And then calculating the acceleration and the speed of the vehicle at the current moment, and selecting a corresponding transition probability matrix by combining the driving mode recognition result. And (5) calculating to obtain a predicted vehicle speed sequence of t seconds in the future by searching for matching. And substituting the predicted vehicle speed sequence of the future t seconds into the model to obtain the vehicle energy demand sequence of the future t seconds. And then substituting the vehicle energy demand sequence of t seconds into the objective function to solve the optimal control problem, and solving a numerical solution sequence, wherein the first element in the sequence is the energy distribution scheme at the next moment.

The simulation results are shown in figures 7-10 in the specification.As shown in FIGS. 7 and 8, in the formula (31), λ is taken₁＝0.5,λ₂＝100，λ₃The introduction of the vehicle speed prediction allows the hydrogen consumption to be reduced by 2.5% or more, while the error between the SOC and the reference value 0.6 is smaller compared to the energy management method without vehicle speed prediction. Considering the efficiency of the fuel cell, the conventional method as shown in fig. 9 cannot keep the fuel cell operating at a high efficiency, which is distributed between 0 and 44%. As can be seen in fig. 10, the energy management method of the present invention can achieve an average operating efficiency of 38.4% for the fuel cell.