CN110775065B - Hybrid electric vehicle battery life prediction method based on working condition recognition - Google Patents

Abstract

The invention discloses a hybrid electric vehicle battery life prediction method based on working condition identification. Training characteristic parameters of all working conditions by adopting a random forest model to realize the identification of the real-time working conditions; respectively making energy management control strategies under routes mainly under urban, suburban and high-speed working conditions for the hybrid electric vehicle, taking the sum of fuel consumption cost and battery life attenuation cost of each stage as an optimization objective function, solving based on a dynamic programming algorithm, dividing the whole vehicle working mode by using a Support Vector Machine (SVM), respectively training a neural network model by using the optimization results of the corresponding working modes under each route, and establishing the corresponding energy management control strategies based on the neural network to realize the prediction of the battery life; the method provided by the invention realizes the real-time prediction of the service life of the battery, improves the service life of the battery while ensuring the fuel economy of the whole vehicle, and reduces the use cost of the vehicle.

Description

Hybrid electric vehicle battery life prediction method based on working condition recognition

Technical Field

The invention belongs to the field of battery life of hybrid electric vehicles, and particularly relates to a hybrid electric vehicle battery life prediction method based on working condition identification.

Background

The hybrid electric vehicle has a plurality of power sources, the working state of each power source needs to be reasonably coordinated to meet the power performance requirement of the whole vehicle, and the energy-saving advantage of the hybrid electric vehicle is further fully exerted, wherein the performance of a power battery directly influences the performance of a driving motor, so that the fuel economy and the emission performance of the whole vehicle are influenced, and the key for realizing the performance of the whole vehicle is realized.

At present, research on the service life of the battery mainly focuses on establishing a battery service life model under a single working condition based on data under a single experimental condition, and the relation between the fuel economy and the service life of the battery in an energy management optimization management control strategy is rarely considered, so that the fuel economy is ensured and the service life of the battery cannot be considered under the actual driving working condition. Research shows that the fuel consumption and the attenuation degree of the battery life are in a contradictory relation, so that the attenuation degree of the battery life is considered in an energy optimization management control strategy, and the consideration is very necessary for improving the fuel economy of the whole vehicle, reducing the use cost of the vehicle and improving the use performance of the battery. The battery life model can be divided into a model method and a data driving method, wherein the model method is a model constructed based on a battery operation mechanism and aging, but the complexity is high, and the prediction error is large; the data-driven method is used for constructing a battery life model based on a large amount of experimental data, and can only accurately describe the battery life in a single experiment or under a single working condition. The actual running working conditions of the hybrid electric vehicle are complex and changeable, the real-time residual service life of the battery cannot be accurately predicted simply by using the two battery life models, and the adaptability to the working conditions is poor, so that the establishment of the battery life real-time prediction model becomes the key for improving the service performance of the battery and improving the fuel economy of the whole vehicle.

At present, the research on working condition identification mainly focuses on identifying the working condition type in real time and switching the control strategy under the corresponding type, so that the fuel economy of the whole vehicle is improved, for example, the Chinese patent publication No. CN106004865A, publication No. 2016-10-12, discloses a mileage self-adaptive hybrid power vehicle energy management method based on the working condition identification, aiming at a plug-in hybrid power vehicle, the fuel economy of the whole vehicle is improved by identifying the working condition and adapting to different driving mileage and working conditions, but the method does not consider the influence of the service life attenuation of a battery on the fuel economy; the research on energy management considering the battery life attenuation mainly focuses on predicting the battery life under a single working condition, for example, the Chinese patent publication number is CN107878445A, the publication date is 2018-04-06, and a hybrid electric vehicle energy management method considering the battery life is disclosed, the battery life attenuation is considered in offline global optimization control, the battery life is prolonged, and the influence of the change of the running working condition on the battery life is not considered; the existing patent does not fully consider the actual complex and changeable traveling working condition and the battery life prediction, so that under the actual traveling working condition, the capacity of a hybrid electric vehicle battery is difficult to accurately predict, the remaining service life of the battery cannot be accurately determined, the service performance of the battery is reduced, and the improvement of the fuel economy of the whole vehicle is not facilitated.

Disclosure of Invention

In order to overcome the defects of the technology, the hybrid electric vehicle battery life prediction method based on the working condition identification provided by the invention has the advantages that the relation between the oil consumption and the battery life attenuation is balanced and coordinated in the energy optimization management control strategy, the online real-time prediction of the battery life of the hybrid electric vehicle under the actual driving working condition is realized based on the working condition identification module and the battery life prediction module, the battery life is accurately predicted on the basis of ensuring the fuel economy of the whole vehicle, the battery use performance is improved, and the adaptability of a control system to the actual complex driving working condition is improved.

The invention is realized by adopting the following technical scheme, which comprises the following steps:

(1) the method comprises the following steps of respectively carrying out global optimization control based on routes mainly comprising urban areas, suburbs and high-speed working conditions to obtain the change results of the working states of an engine and a battery of the hybrid power system along with the vehicle speed and the required power of the whole vehicle, and specifically comprises the following steps:

dividing routes mainly under urban, suburban and high-speed working conditions into 1 second step length to form an N-stage decision problem, and calculating the sum of the fuel consumption cost and the battery life attenuation cost of each stage according to the vehicle speed and the whole vehicle power requirement of each stage;

establishing a multi-objective optimization control model, which comprises the following steps: optimizing a target function and constraint conditions, and obtaining the optimal control quantity meeting the optimization target by adopting a dynamic programming algorithm;

the optimization objective function is:

in the formula, mu is a weight coefficient and the value range is 0-1; c_E(x_k,u_k) For fuel consumption costs, including engine fuel consumption and battery equivalent fuel consumption, C_H(x_k,u_k) Cost is attenuated for battery life; c_aIs the conversion coefficient; x is the number of_kTaking a state variable, namely the SOC of the battery, and gridding; u. of_kA decision variable, namely battery power, is taken, and gridding is carried out;

the state transition equation S_g[x_k,u_k]Comprises the following steps:

in the formula, SOC_kFor the battery SOC at time k, I_k+1Current flowing through the battery at time k +1, Q_batIs the battery capacity;

the k time fuel consumption cost C_E(x_k,u_k) Comprises the following steps:

C_E(x_k,u_k)＝W_fuel(k)+αP_bat(k)

in the formula, W_fuel(k) For the fuel consumption of the engine at time k, determined by the operating point of the engine, P_bat(k) Battery power at time k, α is the equivalent fuel factor;

the k-time battery life decay cost C_H(x_k,u_k) The following formula is used to obtain:

C_H(x_k,u_k)＝σ·|I_c(k)|

where σ is a battery life decay influence factor, I_c(k) The battery charge-discharge multiplying power at the moment k;

the battery life attenuation influence factor sigma is calculated by the following formula:

where τ is the rated Life of the battery, i.e., the total amount of electricity flowing through the battery at the End of Life (EOL) under rated operating conditions; gamma is the total amount of charge flowing through the battery at the end of its life under actual operating conditions; i is_c,nomThe rated charge-discharge rate of the battery; i is_cThe actual charge and discharge rate of the battery is obtained;

the constraint conditions are as follows:

P_{e_min}≤P_e(k)≤P_{e_max}

P_{bat_min}≤P_bat(k)≤P_{bat_max}

ω_{m_min}≤ω_m(k)≤ω_{m_max}

T_{m_min}(ω_m)≤T_m(k)≤T_{m_max}(ω_m)

in the formula, P_e(k) Engine power at time k; p_{e_min}At minimum power of the engine, P_{e_max}Is the maximum power of the engine, P_{bat_min}Is the minimum power of the battery, P_{bat_max}Is the maximum power of the battery, omega_m(k) Motor speed at time k, ω_{m_min}At minimum motor speed, ω_{m_max}At the maximum rotational speed of the motor, T_m(k) Motor torque at time k, T_{m_min}(ω_m) Is the minimum torque corresponding to the current rotating speed of the motor, T_{m_max}(ω_m) The maximum torque is the maximum torque corresponding to the current rotating speed of the motor;

thirdly, the minimum value of the optimized objective function of each stage of the state variable at each grid point is reversely calculated, and the result and the corresponding decision variable are stored;

forward calculation is carried out on the basis of the result stored by reverse calculation, namely the initial value of the SOC of the battery is known from the first moment to the N moment, the optimal control quantity of each moment under the routes mainly under urban, suburban and high-speed working conditions is obtained through interpolation respectively, and then the optimization results of the working states of the engine and the battery of the hybrid power system under each route along with the vehicle speed and the required power of the whole vehicle are obtained;

(2) based on the optimization result of the dynamic programming, selecting a training sample, and dividing the working mode of the whole vehicle by using a support vector machine, wherein the method specifically comprises the following steps:

firstly, taking the required power of the whole vehicle, the vehicle speed and the battery SOC as input, judging whether an engine is started or not as output, classifying samples by adopting a fuzzy C-mean clustering algorithm, repeatedly adjusting a membership matrix and a clustering center, acquiring fuzzy classification of data, and uniformly selecting partial data from each class as training samples;

training a support vector machine model based on the selected training sample, and dividing the whole vehicle mode into a pure Electric (EV) mode and a Hybrid Electric (HEV) mode;

(3) respectively training a neural network model by utilizing the optimization results of the corresponding working modes under each route, and establishing a corresponding energy management control strategy based on the neural network, which specifically comprises the following steps:

respectively carrying out classification on samples in an HEV mode and a braking mode by using the required power of a whole vehicle, the vehicle speed and the battery SOC as input and the required power of the battery as output by using a fuzzy C-mean clustering algorithm, repeatedly adjusting a membership matrix and a clustering center, acquiring fuzzy classification of data, and uniformly selecting partial data from each class as training samples; selecting 3 neurons of an input layer, 1 neuron of an output layer and the number of neurons of a hidden layer based on the selected training sample, and training a neural network model;

applying the output of the trained neural network model in different modes to a developed energy management control strategy based on the neural network to determine the working point of the engine and the required power of the battery;

(4) training characteristic parameters of all working conditions by adopting a random forest model so as to realize the identification of real-time working conditions, and specifically comprising the following steps:

dividing all urban working conditions in three routes which mainly comprise urban working conditions, suburban working conditions and high-speed working conditions into one type, dividing all suburban working conditions into one type, and dividing all high-speed working conditions into one type;

determining characteristic parameters of the working condition, such as average vehicle speed, average acceleration, maximum acceleration, average deceleration, maximum deceleration and idle time;

thirdly, calculating characteristic parameters of various working conditions based on an optimization result of the dynamic programming, selecting a training sample, traversing the number of the decision tree, determining the number of the decision tree based on the classification accuracy obtained by traversal and the structural complexity of the random forest, and training a random forest model;

(5) under the operating mode of actually traveling, based on operating mode identification module and battery life prediction module, be applied to real vehicle control on line in real time, specifically include:

collecting data of a real vehicle under an actual driving condition, calculating characteristic parameters, identifying the working condition type of the real-time working condition by using a trained random forest model, counting the number of previously identified urban, suburban and high-speed working conditions, and determining the route type of the currently identified working condition according to the accumulated number;

determining the whole vehicle working mode under the current working condition by using a support vector machine model, determining the battery power at each moment by using a neural network model under a corresponding mode under a corresponding route, and then determining the working point of the engine at each moment by using the whole vehicle required power at each moment and the battery power at each moment, namely:

P_req(t)＝P_e(t)+P_bat(t)

in the formula, P_reqAnd (t) establishing an energy management control strategy based on a neural network for the required power of the whole vehicle at each moment, and realizing real vehicle control.

Compared with the prior art, the invention has the beneficial effects that:

(1) the battery life prediction method based on the working condition recognition coordinately controls the fuel consumption cost and the battery life, slows down the attenuation of the battery life and simultaneously considers the fuel economy of the whole vehicle;

(2) the working condition route selected by the battery life prediction method based on the working condition identification and the energy management control strategy based on the neural network corresponding to the working condition route have typical representativeness and are closer to the characteristics of actual running working conditions, and the adaptability of a control system to the actual complex running working conditions is effectively improved;

(3) the battery life prediction method based on the working condition recognition is based on real-time working condition recognition, can be applied to real-time vehicle control on line in real time, can obtain a control effect close to global optimization control, and is a suboptimal solution of the global optimization control;

(4) the battery life prediction method based on the working condition identification is beneficial to better managing and utilizing the battery, so that the performance of the battery is improved, and the cost of the battery life is reduced.

Drawings

The invention is further described with reference to the accompanying drawings in which:

FIG. 1 is a schematic flow diagram of the overall process of the present invention;

FIG. 2 is a vehicle speed diagram illustrating a selected suburban route of operation according to an embodiment of the present invention;

FIG. 3 is a vehicle speed map of a selected operating mode route based on a high speed operating mode in an embodiment of the present invention;

FIG. 4 is a vehicle speed diagram of a selected urban operating mode-dominated route in accordance with an embodiment of the present invention;

FIG. 5 is a flow chart of a hybrid vehicle condition identification module and a battery life prediction module of the present invention;

FIG. 6 is a diagram of classification correctness rates corresponding to different decision trees of a random forest model in an embodiment of the present invention;

fig. 7 is a flowchart illustrating actual vehicle control of the hybrid vehicle according to the present invention under actual driving conditions.

Detailed Description

The invention is described in detail below with reference to the following figures and detailed description:

referring to fig. 1 to 7, the invention provides a hybrid electric vehicle battery life prediction method based on working condition identification, which predicts the remaining service life of a battery of a hybrid electric vehicle. The invention discloses a hybrid electric vehicle battery life prediction method based on working condition identification, which comprises the following steps:

(1) referring to fig. 1, the global optimization control is respectively performed based on routes mainly including urban, suburban and high-speed working conditions to obtain the result of the change of the working states of the engine and the battery of the hybrid power system along with the vehicle speed and the required power of the whole vehicle, and the method specifically includes the following steps:

selecting three working condition routes which mainly comprise urban, suburban and high-speed working conditions, wherein each working condition route covers the urban, suburban and high-speed working conditions, the three working condition routes selected in the embodiment refer to fig. 2 to 4, the working condition 1 represents the suburban route 1, the suburban working conditions are taken as main conditions, the average speed is in a medium state, the working condition 2 represents the high-speed route 2, the high-speed working conditions are taken as main conditions, the average speed is high, the working condition 3 represents the urban route 3, the urban working conditions are taken as main conditions, the average speed is low, the routes which mainly comprise the urban, suburban and high-speed working conditions are divided into 1 second step length to form an N-stage decision problem, and the sum of the fuel consumption cost and the battery life attenuation cost of each stage is calculated according to the speed of each stage and the power requirement of the whole vehicle;

referring to fig. 1, a multi-objective optimization control model is established, which includes: optimizing a target function and constraint conditions, and obtaining the optimal control quantity meeting the optimization target by adopting a dynamic programming algorithm;

the optimization objective function is:

in the formula, mu is a weight coefficient and the value range is 0-1; c_E(x_k,u_k) For fuel consumption costs, including engine fuel consumption and battery equivalent fuel consumption, C_H(x_k,u_k) Cost is attenuated for battery life; c_aIs the conversion coefficient; x is the number of_kTaking a state variable, namely the SOC of the battery, and gridding; u. of_kA decision variable, namely battery power, is taken, and gridding is carried out;

the state transition equation S_g[x_k,u_k]Comprises the following steps:

in the formula, SOC_kFor the battery SOC at time k, I_k+1Current flowing through the battery at time k +1, Q_batIs the battery capacity;

the k time fuel consumption cost C_E(x_k,u_k) Comprises the following steps:

C_E(x_k,u_k)＝W_fuel(k)+αP_bat(k)

in the formula, W_fuel(k) For the fuel consumption of the engine at time k, determined by the operating point of the engine, P_bat(k) Battery power at time k, α is the equivalent fuel factor;

the k-time battery life decay cost C_H(x_k,u_k) The following formula is used to obtain:

C_H(x_k,u_k)＝σ·|I_c(k)|

where σ is a battery life decay influence factor, I_c(k) The battery charge-discharge multiplying power at the moment k;

the battery life attenuation influence factor sigma is calculated by the following formula:

where τ is the rated Life of the battery, i.e., the total amount of electricity flowing through the battery at the End of Life (EOL) under rated operating conditions; gamma is the total amount of charge flowing through the battery at the end of its life under actual operating conditions; i is_c,nomThe rated charge-discharge rate of the battery; i is_cThe actual charge and discharge rate of the battery is obtained;

the constraint conditions are as follows:

P_{e_min}≤P_e(k)≤P_{e_max}

P_{bat_min}≤P_bat(k)≤P_{bat_max}

ω_{m_min}≤ω_m(k)≤ω_{m_max}

T_{m_min}(ω_m)≤T_m(k)≤T_{m_max}(ω_m)

in the formula, P_e(k) Engine power at time k; p_{e_min}At minimum power of the engine, P_{e_max}Is the maximum power of the engine, P_{bat_min}Is the minimum power of the battery, P_{bat_max}Is the maximum power of the battery, omega_m(k) Motor speed at time k, ω_{m_min}At minimum motor speed, ω_{m_max}At the maximum rotational speed of the motor, T_m(k) Motor torque at time k, T_{m_min}(ω_m) Is the minimum torque corresponding to the current rotating speed of the motor, T_{m_max}(ω_m) The maximum torque is the maximum torque corresponding to the current rotating speed of the motor;

thirdly, the minimum value of the optimized objective function of each stage of the state variable at each grid point is reversely calculated, and the result and the corresponding decision variable are stored;

forward calculation is carried out on the basis of the result stored by reverse calculation, namely the initial value of the SOC of the battery is known from the first moment to the N moment, the optimal control quantity of each moment under the routes mainly under urban, suburban and high-speed working conditions is obtained through interpolation respectively, and then the optimization results of the working states of the engine and the battery of the hybrid power system under each route along with the vehicle speed and the required power of the whole vehicle are obtained;

(2) referring to fig. 5, based on the optimization result of the dynamic programming, a training sample is selected, and the whole vehicle working mode is divided by using the support vector machine, which specifically includes:

firstly, taking the required power of the whole vehicle, the vehicle speed and the battery SOC as input, judging whether an engine is started or not as output, classifying samples by adopting a fuzzy C-mean clustering algorithm, repeatedly adjusting a membership matrix and a clustering center, acquiring fuzzy classification of data, and uniformly selecting partial data from each class as training samples;

training a support vector machine model based on the selected training sample, and dividing the whole vehicle mode into a pure Electric (EV) mode and a Hybrid Electric (HEV) mode;

(3) respectively training a neural network model by utilizing the optimization results of the corresponding working modes under each route, and establishing a corresponding energy management control strategy based on the neural network, which specifically comprises the following steps:

respectively carrying out classification on samples in an HEV mode and a braking mode by using the required power of a whole vehicle, the vehicle speed and the battery SOC as input and the required power of the battery as output by using a fuzzy C-mean clustering algorithm, repeatedly adjusting a membership matrix and a clustering center, acquiring fuzzy classification of data, and uniformly selecting partial data from each class as training samples; selecting 3 neurons of an input layer, 1 neuron of an output layer and the number of neurons of a hidden layer based on the selected training sample, and training a neural network model;

applying the output of the trained neural network model in different modes to a developed energy management control strategy based on the neural network to determine the working point of the engine and the required power of the battery;

(4) referring to fig. 5, training the characteristic parameters of each working condition by using a random forest model to realize the identification of the real-time working condition specifically comprises:

dividing all urban working conditions in three routes which mainly comprise urban working conditions, suburban working conditions and high-speed working conditions into one type, dividing all suburban working conditions into one type, and dividing all high-speed working conditions into one type;

determining characteristic parameters of the working condition, such as average vehicle speed, average acceleration, maximum acceleration, average deceleration, maximum deceleration and idle time;

thirdly, calculating characteristic parameters of various working conditions based on an optimization result of dynamic programming, selecting training samples, traversing the number of the decision tree from 50 to 1000 in the embodiment, referring to fig. 6, when the number of the decision tree is 650, the classification accuracy reaches a highest value of 93.42%, and when the number of the decision tree is 150, the classification accuracy is still 93.35%, determining the number of the decision tree to be 150 based on the classification accuracy obtained by traversal and the structural complexity of a random forest, and training a random forest model;

(5) referring to fig. 7, under an actual driving condition, the real-time online application to the real vehicle control based on the condition recognition module and the battery life prediction module specifically includes:

firstly, at the beginning of vehicle running, the initial recognition delay time is delta P, no working condition recognition is carried out in the [0, delta P ] time period, in the embodiment, the vehicle works under the urban area route in the [0, delta P ] time period, the data of the actual vehicle under the actual running working condition is collected, the characteristic parameters in the [ t-delta P, t ] time period are calculated, the working condition type of the real-time working condition is recognized by using a trained random forest model, the working condition type is used as the working condition type in the [ t, t + delta f ] time after the t moment, the number of the previously recognized urban area, suburb and high-speed working conditions is counted, the route type to which the currently recognized working condition belongs is determined according to the accumulated number, and the sampling step length delta f is usually taken as 1 second;

determining the whole vehicle working mode under the current working condition by using a support vector machine model, determining the battery power at each moment by using a neural network model under a corresponding mode under a corresponding route, and then determining the working point of the engine at each moment by using the whole vehicle required power at each moment and the battery power at each moment, namely:

P_req(t)＝P_e(t)+P_bat(t)

in the formula, P_reqAnd (t) establishing an energy management control strategy based on a neural network for the required power of the whole vehicle at each moment, and realizing real vehicle control.

Claims (1)

1. A hybrid electric vehicle battery life prediction method based on working condition identification is characterized by comprising the following steps:

(1) the method comprises the following steps of respectively carrying out global optimization control based on routes mainly comprising urban areas, suburbs and high-speed working conditions to obtain the change results of the working states of an engine and a battery of the hybrid power system along with the vehicle speed and the required power of the whole vehicle, and specifically comprises the following steps:

dividing routes mainly under urban, suburban and high-speed working conditions into 1 second step length to form an N-stage decision problem, and calculating the sum of the fuel consumption cost and the battery life attenuation cost of each stage according to the vehicle speed and the whole vehicle power requirement of each stage;

establishing a multi-objective optimization control model, which comprises the following steps: optimizing a target function and constraint conditions, and obtaining the optimal control quantity meeting the optimization target by adopting a dynamic programming algorithm;

the optimization objective function is:

in the formula, mu is a weight coefficient and the value range is 0-1; c_E(x_k,u_k) For fuel consumption costs, including engine fuel consumption and battery equivalent fuel consumption, C_H(x_k,u_k) Cost is attenuated for battery life; c_aIs the conversion coefficient; x is the number of_kIs a state variable, namely a battery state of charge (SOC), and is subjected to gridding; u. of_kA decision variable, namely battery power, is taken, and gridding is carried out;

the state transition equation S_g[x_k,u_k]Comprises the following steps:

in the formula, SOC_kFor the battery SOC at time k, I_k+1Current flowing through the battery at time k +1, Q_batIs the battery capacity;

the k time fuel consumption cost C_E(x_k,u_k) Comprises the following steps:

C_E(x_k,u_k)＝W_fuel(k)+αP_bat(k)

in the formula, W_fuel(k) For the fuel consumption of the engine at time k, determined by the operating point of the engine, P_bat(k) Battery power at time k, α is the equivalent fuel factor;

the k-time battery life decay cost C_H(x_k,u_k) The following formula is used to obtain:

C_H(x_k,u_k)＝σ·|I_c(k)|

where σ is a battery life decay influence factor, I_c(k) The battery charge-discharge multiplying power at the moment k;

the battery life attenuation influence factor sigma is calculated by the following formula:

where τ is the rated Life of the battery, i.e., the total amount of electricity flowing through the battery at the End of Life (EOL) under rated operating conditions; gamma is the total amount of charge flowing through the battery at the end of its life under actual operating conditions; i is_c,nomThe rated charge-discharge rate of the battery; i is_cThe actual charge and discharge rate of the battery is obtained;

the constraint conditions are as follows:

P_{e_min}≤P_e(k)≤P_{e_max}

P_{bat_min}≤P_bat(k)≤P_{bat_max}

ω_{m_min}≤ω_m(k)≤ω_{m_max}

T_{m_min}(ω_m)≤T_m(k)≤T_{m_max}(ω_m)

in the formula, P_e(k) Engine power at time k; p_{e_min}At minimum power of the engine, P_{e_max}Is the maximum power of the engine, P_{bat_min}Is the minimum power of the battery, P_{bat_max}Is the maximum power of the battery, omega_m(k) Motor speed at time k, ω_{m_min}At minimum motor speed, ω_{m_max}At the maximum rotational speed of the motor, T_m(k) Motor torque at time k, T_{m_min}(ω_m) Is the minimum torque corresponding to the current rotating speed of the motor, T_{m_max}(ω_m) The maximum torque is the maximum torque corresponding to the current rotating speed of the motor;

thirdly, the minimum value of the optimized objective function of each stage of the state variable at each grid point is reversely calculated, and the result and the corresponding decision variable are stored;

forward calculation is carried out on the basis of the result stored by reverse calculation, namely the initial value of the SOC of the battery is known from the first moment to the N moment, the optimal control quantity of each moment under the routes mainly under urban, suburban and high-speed working conditions is obtained through interpolation respectively, and then the optimization results of the working states of the engine and the battery of the hybrid power system under each route along with the vehicle speed and the required power of the whole vehicle are obtained;

(2) based on the optimization result of the dynamic programming, selecting a training sample, and dividing the working mode of the whole vehicle by using a support vector machine, wherein the method specifically comprises the following steps:

firstly, taking the required power of the whole vehicle, the vehicle speed and the battery SOC as input, judging whether an engine is started or not as output, classifying samples by adopting a fuzzy C-mean clustering algorithm, repeatedly adjusting a membership matrix and a clustering center, acquiring fuzzy classification of data, and uniformly selecting partial data from each class as training samples;

training a support vector machine model based on the selected training sample, and dividing the whole vehicle mode into a pure Electric (EV) mode and a Hybrid Electric (HEV) mode;

(3) respectively training a neural network model by utilizing the optimization results of the corresponding working modes under each route, and establishing a corresponding energy management control strategy based on the neural network, which specifically comprises the following steps:

respectively carrying out classification on samples in an HEV mode and a braking mode by using the required power of a whole vehicle, the vehicle speed and the battery SOC as input and the required power of the battery as output by using a fuzzy C-mean clustering algorithm, repeatedly adjusting a membership matrix and a clustering center, acquiring fuzzy classification of data, and uniformly selecting partial data from each class as training samples; selecting 3 neurons of an input layer, 1 neuron of an output layer and the number of neurons of a hidden layer based on the selected training sample, and training a neural network model;

applying the output of the trained neural network model in different modes to a developed energy management control strategy based on the neural network to determine the working point of the engine and the required power of the battery;

(4) training characteristic parameters of all working conditions by adopting a random forest model so as to realize the identification of real-time working conditions, and specifically comprising the following steps:

dividing all urban working conditions in three routes which mainly comprise urban working conditions, suburban working conditions and high-speed working conditions into one type, dividing all suburban working conditions into one type, and dividing all high-speed working conditions into one type;

determining characteristic parameters of the working condition, such as average vehicle speed, average acceleration, maximum acceleration, average deceleration, maximum deceleration and idle time;

thirdly, calculating characteristic parameters of various working conditions based on an optimization result of the dynamic programming, selecting a training sample, traversing the number of the decision tree, determining the number of the decision tree based on the classification accuracy obtained by traversal and the structural complexity of the random forest, and training a random forest model;

(5) under the operating mode of actually traveling, based on operating mode identification module and battery life prediction module, be applied to real vehicle control on line in real time, specifically include:

collecting data of a real vehicle under an actual driving condition, calculating characteristic parameters, identifying the working condition type of the real-time working condition by using a trained random forest model, counting the number of previously identified urban, suburban and high-speed working conditions, and determining the route type of the currently identified working condition according to the accumulated number;

determining the whole vehicle working mode under the current working condition by using a support vector machine model, determining the battery power at each moment by using a neural network model under a corresponding mode under a corresponding route, and then determining the working point of the engine at each moment by using the whole vehicle required power at each moment and the battery power at each moment, namely:

P_req(t)＝P_e(t)+P_bat(t)

in the formula, P_reqAnd (t) establishing an energy management control strategy based on a neural network for the required power of the whole vehicle at each moment, and realizing real vehicle control.

Images