An energy recovery method and device based on PSO-SVM operating condition identification

Abstract

The application discloses an energy recovery method based on PSO-SVM working condition recognition, and comprises the following steps: collecting current driving data; loading a PSO-SVM working condition prediction model, inputting the driving data into the PSO-SVM working condition prediction model; obtaining a braking working condition type and a braking torque from the PSO-SVM working condition prediction model; selecting an energy recovery control strategy according to the braking working condition type, and executing energy recovery. According to the above technical scheme, the optimal recovery strategy of energy recovery can be formulated, the automobile in the braking process is safer, and the energy recovery effect is better.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent control of new energy vehicles, in particular to an energy recovery method and device based on PSO-SVM working condition recognition. BACKGROUND

[0002] In recent years, the driving range is one of the important factors restricting the promotion of electric vehicles. The breakthrough to solve this problem mainly focuses on: energy storage system, power system and energy recovery system. It takes a long time to accumulate technology to achieve technical innovation of energy storage system or significantly reduce energy consumption of power system, so improving the brake energy recovery efficiency in the energy recovery system is a practical approach at present.

[0003] In the prior art, the research on brake energy recovery technology mainly considers the following points: formulating a reasonable brake force distribution strategy under the premise of ensuring the safety and stable braking of the vehicle, building a simulation model and corresponding evaluation index, and improving the recovery rate. The most important of these key points is how to design an excellent brake force distribution control strategy, but these criteria cannot be applied to the scenarios encountered in actual driving. In actual application, there are also shortcomings such as limited energy recovery rate and poor brake stability. At the same time, although the existing electric vehicle brake energy recovery technology can recover part of the brake energy, the recovery efficiency is not very high, and some energy is wasted. In the braking process, the safety of the vehicle and the brake energy recovery cannot achieve ideal results at the same time, and the driver's braking experience is very poor, so a technical solution is needed to consider the braking performance of the vehicle and the safety of the driver while establishing an optimal recovery strategy for different working conditions to optimize the overall performance of the vehicle and maximize the vehicle's driving range. SUMMARY

[0004] To achieve the above-mentioned purpose, the present application provides an energy recovery method based on PSO-SVM working condition recognition, comprising the following steps:

[0005] Collecting current driving data, the driving data including speed v t , torque T t , brake intensity z t , current slope ω t , deceleration a t at time t;

[0006] Loading a PSO-SVM working condition prediction model, and inputting the driving data to the PSO-SVM working condition prediction model;

[0007] The brake working condition type and the brake torque are obtained from the PSO-SVM working condition prediction model, the brake working condition type includes: emergency brake working condition, ordinary brake working condition; the brake torque includes regenerative brake torque and mechanical brake torque;

[0008] According to the brake working condition type, an energy recovery control strategy is selected to perform energy recovery; wherein, the energy recovery is achieved by distributing the demand brake torque; the demand brake torque refers to adjusting the constituent ratio of the regenerative brake torque and the mechanical brake torque.

[0009] Further, before loading the PSO-SVM working condition prediction model, an optimal PSO-SVM working condition prediction model is obtained, including the following steps:

[0010] Setting algorithm parameters;

[0011] Defining and updating the penalty coefficient C and the kernel function parameter G;

[0012] Training and verifying the SVM identification model using sample data;

[0013] Outputting the optimal penalty coefficient C and the kernel function parameter G to obtain the optimal identification model.

[0014] Further, the sample data is a feature data set and a brake working condition type corresponding to the feature data;

[0015] The feature data set is Wherein z k , ω k , a k are the average brake intensity z k , the average slope ω k , and the average deceleration a k of the kth interception period, respectively.

[0016] The judgment basis of the brake working condition type is the brake intensity difference z, the slope difference ω, and the deceleration difference a.

[0017] The judgment method of the brake working condition type is:

[0018] If z>0 and ω>0, the brake working condition type is emergency brake working condition,

[0019] If z>0 and a>0, the brake working condition type is emergency brake working condition,

[0020] If ω>0 and a>0, the brake working condition type is emergency brake working condition,

[0021] Otherwise, the brake working condition type is general brake working condition.

[0022] Further, before calculating the braking intensity difference z, the slope difference ω and the deceleration difference a, preset characteristic data standards are respectively a standard braking intensity z1, a standard slope ω1 and a standard deceleration a1.

[0023] The method for calculating the braking intensity difference z, the slope difference ω and the deceleration difference a is:

[0024] z=z k -z1, ω=ω k -ω1, a=a k -a1.

[0025] The energy recovery control strategy is selected according to the braking condition type, that is, if the braking condition type is a normal braking condition, a global optimal energy recovery control strategy is selected, and if the braking condition type is an emergency braking condition, a front-rear axle braking force ideal distribution control strategy is selected.

[0026] In another aspect, the application provides an energy recovery device based on PSO-SVM working condition recognition, comprising:

[0027] The driving data acquisition module is used for acquiring current driving data, and the driving data includes the speed v t , the braking intensity z t , the current slope ω t , the torque T t and the deceleration a t at time t.

[0028] The prediction model loading module is used for loading a PSO-SVM working condition prediction model and inputting the driving data into the PSO-SVM working condition prediction model.

[0029] The prediction data processing module is used for obtaining output data from the prediction model loading module, and the output data is the braking condition type and the required braking torque, the braking condition type includes an emergency braking condition and a normal braking condition, and the braking torque includes a regenerative braking torque and a mechanical braking torque.

[0030] The energy recovery control module is used for selecting an energy recovery control strategy according to the braking condition type and performing energy recovery, wherein the energy recovery is achieved by distributing the required braking torque, and the method for distributing the required braking torque is to adjust the constituent ratio of the regenerative braking torque and the mechanical braking torque.

[0031] The prediction data processing module includes a working condition prediction model training unit and a braking condition type judgment unit, and the working condition prediction model training unit is used for generating an optimal PSO-SVM working condition prediction model, which includes:

[0032] The parameter preparation submodule is used for preparing algorithm parameters.

[0033] a coefficient control submodule for defining and updating a penalty coefficient C and a kernel function parameter G;

[0034] a training submodule for training and verifying an SVM identification model using sample data;

[0035] a model determination submodule for outputting an optimal penalty coefficient C and a kernel function parameter G and determining an optimal identification model.

[0036] Further, a sample preparation module is embedded in the driving data acquisition module for preparing a feature data set as a sample data set;

[0037] The brake working condition type judgment unit is configured to judge the brake working condition type according to the sample data set; the brake working condition type includes an emergency brake working condition and a general brake working condition.

[0038] Further, the brake working condition type judgment unit includes a judgment standard interface for obtaining a feature data standard and a calculation standard;

[0039] The feature data standard includes a standard brake intensity z1, a standard slope ω1 and a standard deceleration a1.

[0040] The calculation standard refers to a method for calculating a brake intensity difference z, a slope difference ω and a deceleration difference a, and includes:

[0041] z = z k - z1, ω = ω k - ω1, a = a k - a1.

[0042] Further, the energy recovery control module selecting an energy recovery control strategy refers to:

[0043] If the brake working condition type is a general brake working condition, a globally optimal energy recovery control strategy is selected; if the brake working condition type is an emergency brake working condition, a front-rear axle brake force ideal distribution control strategy is selected.

[0044] According to the present application, a working condition classification method can be defined by comprehensively considering multiple feature factors in the automobile driving process, the optimal energy recovery strategy can be formulated while the brake performance of the automobile and the safety of the driver are considered, and in the present application, the PSO-SVM algorithm is used to identify the working condition type and the brake torque, the identification accuracy is higher, the automobile in the braking process is safer, and the energy recovery effect is better. BRIEF DESCRIPTION OF DRAWINGS

[0045] Figure 1 is a step diagram of the working condition identification energy recovery method according to an embodiment of the present application;

[0046] Figure 2 is a working condition identification model training flowchart according to an embodiment of the present application;

[0047] Figure 3 It is the iteration process schematic view of PSO-SVM algorithm provided according to the embodiment of the application;

[0048] Figure 4 It is the recovery device structure diagram of working condition recognition provided according to the embodiment of the application. DETAILED DESCRIPTION

[0049] The application defines the type of braking working condition and the corresponding rules of different types, establishes a real-time working condition recognition model by using PSO-SVM algorithm, designs the target function of optimal energy recovery in different working conditions, realizes braking force distribution strategy simulation, optimizes the overall performance of the vehicle, and maximizes the vehicle's endurance as much as possible.

[0050] The specific implementation mode of the application will be described in detail below with reference to the drawings in the specification.

[0051] Figure 1 It is the energy recovery method step diagram of working condition recognition, as shown in the figure, comprising the following steps:

[0052] Step S100: real-time acquisition of current driving data, the driving data including speed v t , torque T t , braking intensity z t , current slope ω t , deceleration a t at the time t;

[0053] For a hybrid vehicle, the torque T t in this step includes motor torque and engine torque, and for a pure energy vehicle, the torque T t is motor torque.

[0054] Step S110: loading a working condition prediction model, inputting driving data to the working condition prediction model, and obtaining braking working condition type and braking torque;

[0055] In the common model training, through K-CV cross validation method, a better penalty coefficient C and kernel function parameter g are obtained, but the search time is long, the search precision is low, and the recognition accuracy needs to be improved; the particle swarm optimization algorithm (PSO) is a bionic algorithm for optimization, which is established by simulating the behavior of some creatures in nature, has the advantages of simple search mechanism, fast convergence speed and small amount of calculation, and can reduce and avoid the optimization process from falling into local optimal solution, therefore, the penalty coefficient C and the kernel function parameter g in the particle swarm optimization support vector machine (SVM) algorithm are used in this step to avoid the problems of large amount of calculation and long search time of K-CV cross validation method, and further improve the accuracy of the offline driving working condition recognition model.

[0056] The working condition prediction model loaded in this step adopts a PSO-optimized SVM identification algorithm (referred to as PSO-SVM algorithm), as shown in S111 in the following figure: Figure 1

[0057] The basic flow of the PSO-SVM algorithm is as shown in the following figure: Figure 2

[0058] First, set the algorithm parameters: set the number of population particles to s, initialize the speed and position of each particle in the population, and obtain the first generation population

[0059] The search space of the present application is 2-dimensional, so each particle contains 2 variables, and the historical optimal position P best of each particle is set as the initial position, and the global optimal position of the particle group is taken as the optimal value in P best ;

[0060] Define the penalty coefficient C and the kernel function parameter G, and in the subsequent iteration process, update the penalty coefficient C and the kernel function parameter G according to the obtained population;

[0061] Train and verify the SVM identification model using sample data: the verification process is to test the SVM algorithm accuracy using test samples, obtain the fitness function value, and the fitness function value calculation formula is as follows:

[0062]

[0063] In the formula: Z rec is the working condition type of the identification model; Z act is the actual working condition type; and n is the number of test samples;

[0064] In the iteration process, update the particle speed and position according to formula (2) and formula (3) to obtain the t+1 generation population:

[0065]

[0066]

[0067] Wherein, ω is the inertia weight; d=1, 2, …D; i=1, 2, …, n; m is the current iteration number; V id is the speed of the particle; c1 and c2 are non-negative constants, referred to as acceleration factors; r1 and r2 are random numbers distributed in the interval [0, 1]. The position update range is usually set to [-X max , X max ], and the speed update range is set to -V max ​​, V max ;

[0068] Calculate the fitness of each particle after updating, and compare it with the best position P best corresponding to the fitness of the previous experience, if better, then its current position as the particle P best ;

[0069] Compare the fitness of each particle with the best position g best experienced by the whole particle, if better, update the value of g best ;

[0070] Check the final value condition, if the accuracy meets the preset condition, stop iteration; if the accuracy does not meet the preset condition, update the penalty coefficient C and the kernel function parameter G; if the maximum iteration number is exceeded, stop iteration.

[0071] In this step, the population size of the particle swarm algorithm is set to 20, the maximum iteration number is 200, and the algorithm convergence accuracy is set the same as the BP neural network, and its iteration process is shown in Figure 3 .

[0072] After 200 iterations, the PSO-SVM algorithm finds the optimal particle with an identification accuracy of 94%, corresponding to the optimal particle C=4.5, g=7.2, and the optimal identification model is obtained at this time.

[0073] In this step, the sample data for training and verifying the SVM identification model is the feature data set and the braking condition type corresponding to the feature data:

[0074] During the sample collection process, the feature parameters of the vehicle driving in a period of time are intercepted, the time slice is θ, the interception period is k, the total interception number is k a , initialize k=1, and take [θ(k-1), θk] as the kinematics segment interval to intercept the current working condition data, extract data and standardize, and obtain the feature data set as where z k , ω k , a k are the average values of the kth interception period, i.e. the average braking intensity z k , the average slope ω k , and the average deceleration a k ;

[0075] In order to ensure the accuracy of the average feature parameters of the working condition, it is judged whether the interception period k is equal to the total interception number k a , if yes, data integration and working condition optimization identification are performed by PSO-SVM, otherwise, k+1 is assigned to k, and then the working condition data set in the kinematics segment interval [θ(k-1), θk] is re-intercepted.

[0076] After the sample data is defined, the judgment basis of the braking condition type, i.e., the braking intensity difference z, the slope difference ω, and the deceleration difference a, is defined.

[0077] The braking intensity difference z, the slope difference ω, and the deceleration difference a at each time are calculated, and the method for judging the braking condition type according to the judgment basis is as follows:

[0078] If z>0 and ω>0, the braking condition type is the emergency braking condition,

[0079] If z>0 and a>0, the braking condition type is the emergency braking condition,

[0080] If ω>0 and a>0, the braking condition type is the emergency braking condition,

[0081] Otherwise, the braking condition type is the general braking condition.

[0082] That is, if two of the above three parameters are greater than 0 at the same time, it can be judged that the braking condition type at this time is the emergency braking condition, otherwise it is the general braking condition.

[0083] The calculation method of the braking intensity difference z, the slope difference ω, and the deceleration difference a can be defined according to the laboratory conditions and environmental conditions, for example, a preset standard method can be used, and the standard braking intensity z1, the standard slope ω1, and the standard deceleration a1 are defined as specified values through a management interface to preset characteristic data standards, and then the calculation method of the braking intensity difference z, the slope difference ω, and the deceleration difference a is as follows:

[0084] z=z k -z1, ω=ω k -ω1, a=a k -a1.

[0085] In the above steps, the best PSO-SVM model is obtained, and in the model, the driving data is input, and the optimal condition classification and braking torque F can be output in step S112.

[0086] Specifically, the braking condition type includes the emergency braking condition D and the ordinary braking condition N, and the braking torque F includes the regenerative braking torque and the mechanical braking torque.

[0087] Step S120: According to the braking condition type, an energy recovery control strategy is selected, and energy recovery is performed.

[0088] The energy recovery control strategy includes a global optimal energy recovery control strategy and a front-rear axle braking force ideal distribution control strategy, if the braking condition type is the ordinary braking condition, the global optimal energy recovery control strategy is selected, and if the braking condition type is the emergency braking condition, the front-rear axle braking force ideal distribution control strategy is selected.

[0089] Step S130: energy recovery is realized by distributing required braking torque;

[0090] In step S100, the torque T t In step S112, the braking torque F is obtained; the braking torque F of the energy vehicle includes the regenerative braking torque L and the mechanical braking torque T; after the energy recovery control strategy is determined, the regenerative braking torque and the mechanical braking torque are adjusted according to F=L+T, and the composition ratio is finally realized.

[0091] Meanwhile, the application provides an energy recovery device corresponding to the energy recovery method based on working condition recognition, as shown in the figure, comprising the following parts: Figure 4

[0092] P410 driving data acquisition module: used for acquiring current driving data, the driving data including speed v t , braking intensity z t , current slope ω t , torque T t , deceleration a t ;

[0093] The sample preparation unit P411 is embedded in the driving data acquisition module, which is used for preparing the feature data set as the sample data set;

[0094] P420 prediction model loading module: used for loading the PSO-SVM working condition prediction model, and inputting the driving data acquired by the data acquisition module into the PSO-SVM working condition prediction model;

[0095] P430 prediction data processing module: used for acquiring output data from the prediction model loading module as the input of the prediction data processing module, and outputting data including braking working condition type and braking torque; the braking working condition type includes: emergency braking working condition, ordinary braking working condition; the braking torque includes regenerative braking torque and mechanical braking torque;

[0096] The working condition prediction model training unit P431 and the braking working condition type judgment unit P436 are embedded in the prediction data processing module; the braking working condition type judgment unit P436 is used for judging the braking working condition type according to the sample data set; the output data is the emergency braking working condition and the ordinary braking working condition.

[0097] The braking working condition type judgment subunit includes a judgment standard interface, which is used for acquiring feature data standards and calculating standards;

[0098] The feature data standards include standard braking intensity z1, standard slope ω1 and standard deceleration a1;

[0099] ​The method for calculating the standard includes:

[0100] z=z k -z1, ω=ω k -ω1, a=a k -a1.

[0101] The P431 working condition prediction model training unit is used for generating an optimal PSO-SVM working condition prediction model, and includes:

[0102] The P432 parameter preparation sub-module is used for preparing algorithm parameters.

[0103] The P433 coefficient control sub-module is used for defining and updating a penalty coefficient C and a kernel function parameter G.

[0104] The P434 training sub-module is used for training and verifying an SVM identification model by using sample data.

[0105] The P435 model determination sub-module is used for outputting an optimal penalty coefficient C and a kernel function parameter G and determining an optimal identification model.

[0106] The P440 energy recovery control module is used for selecting an energy recovery control strategy according to a brake working condition type output by the prediction data processing module and executing energy recovery.

[0107] The energy recovery control module selecting an energy recovery control strategy refers to:

[0108] If the brake working condition type is a normal brake working condition, a globally optimal energy recovery control strategy is selected, and if the brake working condition type is an emergency brake working condition, a front-rear axle brake force ideal distribution control strategy is selected.

[0109] In the application, a plurality of characteristic factors in the automobile driving process, such as deceleration, vehicle speed, road slope and brake intensity, are comprehensively considered as the feedforward conditions of control, a working condition classification method is proposed, the brake performance of the automobile and the safety of the driver are considered, the current driving condition is predicted, the brake torque is predicted, and the optimal energy recovery strategy is formulated.

[0110] The above disclosed are only several specific embodiments of the present application, but the present application is not limited thereto, and any changes that can be thought of by those skilled in the art shall fall within the protection scope of the present application.

Claims

1. An energy recovery method based on PSO-SVM operating condition identification, characterized in that, Includes the following steps: Collect current driving data, including the speed at time t. Torque Braking strength Current slope deceleration ; Load the PSO-SVM working condition prediction model and input the driving data into the PSO-SVM working condition prediction model; The braking condition type and braking torque are obtained from the PSO-SVM condition prediction model. The braking condition type includes: emergency braking condition and normal braking condition; the braking torque includes regenerative braking torque and mechanical braking torque. Based on the braking condition type, an energy recovery control strategy is selected and energy recovery is executed; wherein, the energy recovery is achieved by allocating the required braking torque; the allocation of the required braking torque refers to adjusting the composition ratio of regenerative braking torque and mechanical braking torque; The process of obtaining the optimal PSO-SVM load condition prediction model before loading the PSO-SVM load condition prediction model includes the following steps: Set algorithm parameters; Define and update the penalty coefficient C and the kernel function parameter G; Use sample data to train and validate the SVM recognition model; Output the optimal penalty coefficient C and kernel function parameters G to obtain the optimal recognition model; The sample data consists of a feature dataset and the braking condition type corresponding to the feature data. The feature dataset is ,in The average braking intensity of the k-th intercept period is respectively Average slope Average deceleration ; The criteria for determining the braking condition type are based on the difference in braking intensity. Slope difference Deceleration difference ; The method for determining the braking condition type is as follows: if and The braking condition type is emergency braking condition. if and The braking condition type is emergency braking condition. if and The braking condition type is emergency braking condition. Otherwise, the braking condition type is normal braking condition.

2. The energy recovery method according to claim 1, characterized in that, Calculate the braking intensity difference Slope difference Deceleration difference Prior to this, preset characteristic data standards are established, namely, standard braking intensity. Standard slope Standard deceleration ; Calculate the difference in braking strength Slope difference Deceleration difference The method is as follows: - ， - ， - 。 3. The energy recovery method according to claim 1, characterized in that, The selection of the energy recovery control strategy based on the braking condition type refers to: If the braking condition type is normal braking condition, select the globally optimal energy recovery control strategy; If the braking condition type is emergency braking, select the ideal distribution control strategy for braking force between the front and rear axles.

4. An energy recovery device based on PSO-SVM operating condition identification, characterized in that, include: Driving data acquisition module: used to collect current driving data, including the speed at time t. Braking strength Current slope Torque deceleration ; Prediction model loading module: used to load the PSO-SVM working condition prediction model and input the driving data into the PSO-SVM working condition prediction model; Predictive data processing module: used to obtain output data from the predictive model loading module. The output data includes braking condition type and required braking torque. The braking condition type includes emergency braking condition and normal braking condition. The braking torque includes regenerative braking torque and mechanical braking torque. Energy recovery control module: used to select an energy recovery control strategy and execute energy recovery according to the braking condition type; wherein, the energy recovery is achieved by allocating the required braking torque; the allocation of the required braking torque refers to adjusting the composition ratio of regenerative braking torque and mechanical braking torque; The prediction data processing module includes a working condition prediction model training unit and a braking working condition type judgment unit. The working condition prediction model training unit is used to generate the optimal PSO-SVM working condition prediction model, including: The parameter preparation submodule is used to prepare algorithm parameters; The coefficient control submodule is used to define and update the penalty coefficient C and the kernel function parameter G; The training submodule is used to train and validate the SVM recognition model using sample data; The model determination submodule is used to output the optimal penalty coefficient C and kernel function parameters G to determine the optimal recognition model; The sample data consists of a feature dataset and the braking condition type corresponding to the feature data. The feature dataset is ,in The average braking intensity of the k-th intercept period is respectively Average slope Average deceleration ; The criteria for determining the braking condition type are based on the difference in braking intensity. Slope difference Deceleration difference ; The method for determining the braking condition type is as follows: if and The braking condition type is emergency braking condition. if and The braking condition type is emergency braking condition. if and The braking condition type is emergency braking condition. Otherwise, the braking condition type is normal braking condition.

5. The energy recovery device according to claim 4, characterized in that, The driving data acquisition module includes an embedded sample preparation module for preparing a feature dataset as a sample dataset. The braking condition type determination unit is used to determine the braking condition type based on the sample dataset; The braking conditions include emergency braking and normal braking.

6. The energy recovery device according to claim 5, characterized in that, The braking condition type determination unit includes a determination standard interface for obtaining feature data standards and calculation standards; The characteristic data standard includes standard braking strength. Standard slope Standard deceleration ; The calculation standard refers to the calculation of braking intensity difference. Slope difference Deceleration difference The methods include: - 、 - 、 - 。 7. The energy recovery device according to claim 4, characterized in that, The energy recovery control module selects an energy recovery control strategy, which means: If the braking condition is a normal braking condition, select the globally optimal energy recovery control strategy; if the braking condition is an emergency braking condition, select the ideal distribution control strategy for braking force between the front and rear axles.

Images