CN115186594A

CN115186594A - Energy-saving speed optimization method under influence of man-vehicle-road coupling

Info

Publication number: CN115186594A
Application number: CN202210892943.4A
Authority: CN
Inventors: 王金鹏; 赵海艳; 刘奇芳
Original assignee: Jilin University
Current assignee: Jilin University
Priority date: 2022-07-27
Filing date: 2022-07-27
Publication date: 2022-10-14

Abstract

The invention is applicable to the field of intelligent vehicle energy-saving optimization methods, and provides an energy-saving speed optimization method under the influence of human-vehicle-road coupling, which comprises the following steps: step (1): constructing a driving simulation scene, and summoning a plurality of drivers to collect driving data on a driving simulation rack; step (2): processing and analyzing the acquired data, and establishing a double model for identifying the driving condition and the driving style; and (3): adding influences of working conditions and driving styles into the problem of describing the speed optimization of the intelligent vehicle, obtaining a weight coefficient table about the working conditions and the driving styles by using an entropy weight method, and searching a speed track which enables the energy consumption in a time domain to be minimum under the condition of meeting constraint conditions; the energy-saving speed optimization method under the influence of human-vehicle-road coupling can realize the real-time optimization of the vehicle speed on line according to the driving style and the driving condition.

Description

Energy-saving speed optimization method under influence of man-vehicle-road coupling

Technical Field

The invention belongs to the field of intelligent vehicle energy-saving optimization methods, and particularly relates to an energy-saving speed optimization method under the influence of human-vehicle-road coupling.

Background

The road traffic system is a dynamic complex system which is influenced by multiple factors such as people, vehicles, roads, surrounding traffic environments and the like. The driving style and road condition information are two factors which are considered in the human-vehicle-road cooperative energy conservation, and are closely related to driving safety, fuel economy, automobile wear and the like.

The automobile driving condition is a curve describing the driving speed and time of the automobile, and is mainly used for determining the pollutant emission amount and the fuel consumption amount of the automobile and providing a reference basis for the technical development, evaluation and the like of new automobile types.

With the global energy crisis and the low-carbon travel demands, the energy consumption problem of passenger cars is always a common concern for users and manufacturers; with the great enrichment and the deepening of intellectualization of vehicle sensing information, more and more vehicle information and environment information can be introduced to further improve the energy-saving potential of the vehicle. In order to save energy and reduce consumption and find an optimal speed track to minimize energy consumption in a time domain, the invention researches an energy-saving speed optimization method under the influence of human-vehicle-road coupling.

Disclosure of Invention

The invention aims to provide an energy-saving speed optimization method under the influence of human-vehicle-road coupling, which aims to save energy, reduce consumption and search an optimal speed track of a vehicle so as to minimize energy consumption in a time domain.

The invention is realized in such a way that an energy-saving speed optimization method under the influence of man-vehicle-road coupling comprises the following steps:

step (1): a driving simulation scene is built, and a plurality of drivers are summoned to collect driving data on a driving simulation rack;

step (2): processing and analyzing the acquired data, and establishing a double model for identifying the driving condition and the driving style;

and (3): the influence of the working condition and the driving style is added in the problem of describing the speed optimization of the intelligent vehicle, a weight coefficient table about the working condition and the driving style is obtained by utilizing an entropy weight method, and a speed track which enables the energy consumption in a time domain to be minimum is searched under the condition of meeting the constraint condition.

According to the further technical scheme, in the step (1), the driving simulation scene is built through joint simulation of SCANeR and Matlab software.

According to a further technical scheme, in the step (1), the summoned drivers are grouped into different ages, driving ages and genders.

In the further technical scheme, in the step (2), the processing and analyzing of the collected data specifically comprises: firstly, filtering and standardizing by a filter, and then extracting characteristic parameters of the driving condition and the driving style; and under the condition of identifying different working conditions, identifying the driving style.

According to the further technical scheme, the different working conditions comprise urban, suburban and high-speed working conditions.

In a further technical scheme, in the step (3), the weight coefficient table mainly has the following functions: the magnitude of the driving braking force is calculated in the objective function by the weight coefficient.

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

the energy-saving speed optimization method under the influence of human-vehicle-road coupling can realize the real-time optimization of the vehicle speed according to the driving style and the driving condition on line;

the energy-saving speed optimization method under the influence of human-vehicle-road coupling reduces the energy-saving difference of vehicle running among different drivers;

the energy-saving speed optimization method under the influence of man-vehicle-road coupling provided by the invention has extremely important value for research and development of automatic driving of L3 and above.

Drawings

FIG. 1 is a general scheme diagram;

FIG. 2 is a schematic diagram of a simulation scene construction environment;

FIG. 3 is a closed loop diagram of a driver's stand system;

FIG. 4 is the PCA principal component cumulative contribution;

FIG. 5 is a model diagram of a driving condition prediction;

FIG. 6 is a flow chart of a driving style recognition model;

FIG. 7A, FIG. 7B, and FIG. 7C are the original data under different working conditions, respectively;

FIGS. 8A and 8B are raw driving data for different drivers, respectively;

FIG. 9 is a graph of a comparison of data after preprocessing;

FIG. 10A is a real-time monitoring diagram of the driving speed; FIG. 10B is a diagram of the predicted result of the driving mode;

FIG. 11 is a driver style identification result;

FIG. 12 is a graph showing the relationship between the speed of 0-90km/h and the fuel consumption.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

Specific implementations of the present invention are described in detail below with reference to specific embodiments.

S1, establishing an analog simulation scene and acquiring and processing data

A driver in-loop simulation platform is set up, and the test equipment comprises a scene simulation system, a double-station driving simulator, a sensing simulation system and a vehicle dynamics simulation system. The scene simulation system simulates the traffic scene in the driving process of the vehicle in real time in an image form, for example, fig. 2 shows a simulation scene constructed by combining SCANER and matlab. Firstly, adding a simulink simulation module in the CONFIGURATION of SCANeR, and simultaneously, generating the simulation module of SCANeR in a simulink module library. After the software configuration is successful, the SCANeR module in the simulink exports the required output data type, and a controller of the SCANeR module can be added to realize the simulation scene required by the joint simulation construction; a real-person driver observes a real-time driving visual scene and operates a steering wheel, an accelerator pedal and a brake pedal in a driving simulator according to own driving intention; the sensing simulation system simulates sensors such as a camera and a millimeter wave radar and sends information such as traffic signals, pedestrians, roads and traffic signals detected by the virtual sensors to the automatic driving controller; the vehicle controller receives driving signals from the driving simulator and sensing signals from the sensing simulation system, plans, decides and controls according to an automatic driving strategy, and sends control signals to the vehicle dynamics model; the vehicle dynamics simulation system simulates the real vehicle motion process, after receiving the control signal, the vehicle dynamics model calculates that the position and attitude of the vehicle change, and the vehicle position and attitude information is transmitted to the scene simulation system in real time; the scene simulation system updates the scene in real time according to the change of the pose of the vehicle, so that a closed loop diagram of the driving stand system is formed, and is shown in figure 3.

Screening drivers according to basic information of the recruited drivers, such as age, gender, occupation, driving age, driving type, driving mileage, whether traffic accidents occur or not, and then carrying out self-evaluation on driving styles to finally form a test sample of 30 drivers with different ages, driving ages and genders. In the experimental process, the driving style of the driver is preliminarily judged according to experience through partial driving behaviors of the driver, and a subjective evaluation result of the driving style is formed by combining a self-evaluation questionnaire and is shown in the following table 1. The effective and sufficient characteristics are basic conditions for establishing the recognizer, the data of the Internet of vehicles is different from the traditional data, most of the data of the Internet of vehicles is continuous/long-section time sequence data, the dimensionality is extremely large, the signal quantity is excessive, most of the data is irrelevant to the identification establishing process, therefore, before the characteristics are selected, relevant documents are read in signals of a database, a part of parameters for representing the posture and the motion state of the vehicle and a driver operation signal are picked to serve as the basis of the identification experiment characteristic set, and the parameters comprise transverse and longitudinal acceleration, transverse and longitudinal vehicle speed, accelerator pedal opening, braking force, transverse and longitudinal lateral displacement and yaw angular speed.

TABLE 1 subjective evaluation results of typical conditions and driving style

The yaw rate signal comes from a gravity sensor in an electronic stability program system, and is a sensor raw signal which contains part of high-frequency noise and has a zero-point drift phenomenon. Therefore, before the signal is used, the maximum value of the signal needs to be filtered by a Butterworth low-pass filter. Generally, due to the problems of signal abnormality, working condition division or working condition identification errors and the like, abnormal values may be contained in the feature set, and if feature selection is directly carried out without processing, valuable features may be removed. Therefore, a series of processes still need to be performed on the indicators in the feature set to meet the requirement of feature selection. The preprocessing process of the data is selected and standardized, and the formula is as follows:

wherein E (X) and S (X) are the mean and variance of the vector X, and sample data before and after processing are X and X'.

S2, feature extraction and recognizer construction

S2.1 feature extraction and dimension reduction

The driving data of each driver are screened, and the driving data corresponding to the driver under the two conditions that the driver drives on a specified road and the driver does not drive according to the specified driving speed are taken as unreasonable data and are eliminated. Processing the driving data of the drivers meeting the requirements and extracting characteristic parameters capable of reflecting the driving style of each driver; wherein the maximum vehicle speed v is selected _max Average vehicle speed

Average acceleration

Average deceleration

Idle time ratio p _i And a uniform time ratio P _c As the characteristic parameters for identifying the working conditions, 12 indexes of a transverse and longitudinal vehicle speed v, a longitudinal and transverse acceleration a, an impact degree J, a yaw angular velocity r, an accelerator opening degree and a braking force are selected as the characteristic parameters for identifying the driving style. Wherein the vehicle speed and addThe speed can fully reflect the driving habit of a driver, the impact degree reflects the pressure intensity applied by the driver to an acceleration pedal or a brake pedal, the yaw rate deflection represents the stability degree of the automobile, and the latter two can express the driving personality of the driver. These signals are combined into the original driving signal vector, reflecting the characteristics of the driver style from different aspects. The characteristic parameters are extracted as follows:

note: maximum Max, minimum Min, mean, standard deviation SD, root Mean square RMS, coefficient of variation (ratio of standard deviation to Mean) C.V, median Me, range R, difference of maximum and Mean Max-Mean, difference of Mean and minimum Mean-Min, difference of maximum and median Max-Me, lower quartile Q1, middle quartile Q2, upper quartile Q3 and Mean greater than upper quartile (> Q3).

For a certain sample of the original data, respectively calculating 6-dimensional characteristic indexes of the working conditions and 112 characteristic indexes of 8 characteristics representing the driving style to obtain a characteristic index sample, further calculating the characteristic index samples of all the working conditions, and constructing a driving condition and driving style identification characteristic set. Considering that multiple collinearity exists between high-dimensional vector data, instability of a feature space can be caused, and therefore incoherence of an output result is caused; the structure of the recognition model may be complicated, the input with high dimensionality is liable to cause slow calculation speed, waste of storage space, and even bad recognition results. The invention adopts a principal component analysis method to reduce the dimensionality of data, and adopts the basic principle that the original variable matrix is processed by adopting a matrix transformation and linear combination method in consideration of the possible internal implicit connection among original variables, thereby forming a plurality of new comprehensive indexes. The specific process is mainly that each sample firstly removes the average value, calculates the eigenvalue and the eigenvector of the covariance matrix among the samples, and reserves the eigenvector corresponding to the former N largest eigenvalues. And then, converting the original features into a new space constructed by the N feature vectors obtained above, thereby realizing feature compression.

Feature extraction processing is performed on a high-dimensional vector formed by driving style multi-feature parameters by using PCA, and the high-dimensional vector is introduced into a PCA model to obtain a relation between an accumulated contribution rate and a principal component, as shown in FIG. 4. The cumulative contribution rate calculation formula:

the accumulated contribution rate of the principal components is continuously increased along with the accumulation of the principal components in the early stage, but the accumulated contribution rate of the principal components is slowly increased along with the accumulation of the principal components in the later stage, and the accumulated contribution is almost unchanged until the final stage, so that the principal components in the later stage can be ignored in the later calculation analysis to achieve the effect of reducing the data dimension. In general, the principal component corresponding to the cumulative contribution rate of 90% is selected as a subject to be studied later.

S2.2, establishing a model for identifying driving conditions and driving styles

Next, a training model of the BP neural network is performed as shown in fig. 5.

1) And (4) initializing the network. 4 characteristic parameters and 3 driving condition types are determined, so that the number N of BP neural network input nodes constructed by the method _in Number of output layer nodes N of 4 _out 3, and selecting the number N of the neurons in the hidden layer according to an empirical formula _h Is 8. Initializing connection weight omega between layers _ij And ω _jk Hidden layer threshold a, output layer threshold b.

1) The hidden layer output is computed. From an input vector x ₁ ,x ₂ …x _i ]Connecting the weight ω _ij And a threshold value a _i The hidden layer output z can be obtained _j 。

Wherein f is a stimulus function, and the sigmoid function is selected as follows:

2) And calculating an output layer result. Outputting z from the hidden layer _j And a connection weight ω _jk And a threshold value b _k The output layer y can be calculated _k 。

3) And (4) error calculation. Calculating the prediction error e _k ＝Y _k -y _k 。

4) And updating the weight and the threshold. According to the prediction error e _k Updating each weight value omega _ij ，ω _jk And a threshold value a _i ，b _k 。

And (4) judging whether the algorithm reaches the target precision, and if not, returning to the step (2).

The invention adopts the SOM network model in the artificial neural network when clustering the driving style, and can automatically classify the input modes according to the learning rules, namely, the input modes are self-organized under the unsupervised condition, the coefficients of the connection weights are repeatedly adjusted to finally enable the coefficients to reflect the mutual relation among the input samples, and the classification result is represented in a competition layer, and the identification process is shown as figure 6.

1) And (6) vector normalization. For the current input mode vector X in the self-organizing network, the weight vector W corresponding to each neuron in the competition layer _j = (j =1,2, \8230;, n), all normalized to obtain input vector

And weight vector

2) Winning neurons were sought. Weight vector W to be corresponding to all neurons of the competition layer _j = (j =1,2, \8230;, n) similarity comparisons were performed. The most similar neuron wins with a weight vector of

3) Network output and rights adjustment

According to WTA learning rule, the winning neuron output is 1, and the others are 0, i.e.

Only the winning neuron has the right to adjust its weight vector, which is learned and adjusted as follows:

alpha is learning efficiency, generally decreases with the progress of learning multidimensional, namely, the degree of adjustment is smaller and smaller, and tends to the center of a cluster.

4) And (5) renormalizing. After the normalized weight vector is adjusted, the obtained new vector is no longer a unit vector, so the learning adjusted vector needs to be normalized again, and the operation is circulated until the learning rate is attenuated to 0.

S2.3 analysis of results

As can be seen from the raw data graphs of fig. 7A, 7B, and 7C and fig. 8A and 8B, the driving data of different drivers under different operating conditions all show different differences. The result of preprocessing the raw data is shown in fig. 9, filtering can eliminate noise and interference of zero drift, dimensions on each dimension in the data set are different, magnitude difference of values is large, and mutual interference between data can be eliminated by utilizing standardization. And finally, importing the created characteristic data set into BP and SOM neural networks for training, wherein a running speed real-time monitoring graph is shown as a graph in FIG. 10A, and working condition prediction is shown as a graph in FIG. 10B, so that the trained model under the test working condition can predict the running working condition. The recognition result is shown in fig. 11, the neural network divides different drivers into three categories, and it can be known from the clustering centers of the first category to the third category that the X, Y, and Z axis parameters all have increasing trends. In the process of changing the driving style from the first type to the third type, the requirement on the dynamic performance of the driver is higher and higher, and the first type can be considered as a mild driver, the second type is a normal driver, and the third type is an aggressive driver. Then, test sample data is imported into the trained SOM neural network model, and normalization is firstly carried out on the mode input to be recognized; secondly, comparing the normalized mode input with each weight of the SOM neural network to select the neuron of the competition layer with the minimum Euclidean distance; and thirdly, searching the cluster class number to which the neuron belongs to obtain a driving style verification result.

S3, optimizing energy-saving speed under influence of man-vehicle road coupling

The vehicle speed optimization problem is generally described as: and a starting point and an end point are given, and under the condition that certain constraint conditions are met, the vehicle running time is finally minimized or the energy consumption is finally minimized. Under urban working conditions, the phenomenon of quick start and stop occurs in the running process of a vehicle due to the existence of traffic lights; under high-speed working conditions, the conditions of acceleration and deceleration and lane change and overtaking can be continuously caused by different styles of drivers; under the rural working condition, due to the fact that the road surface is uneven and the motor vehicle and non-motor vehicle roads are not distinguished, a plurality of driving obstacles exist or the phenomenon of rapid start and stop is caused. The above circumstances may cause unnecessary energy consumption and emission. As can be seen from the average fuel consumption at various vehicle speeds of 0 to 90km/h in fig. 11 below, fuel consumption increases both at low speed and high speed, and fuel consumption increases significantly when the vehicle speed fluctuation is large. Therefore, it is desirable to solve an optimal speed in consideration of the work condition and the driving style, and in order to reduce fuel consumption as much as possible, the vehicle should be driven at the optimal speed as much as possible, and the optimal speed should be maintained. This patent proposes speed optimization problem purpose under people's vehicle and road coupling influences improves fuel economy when guaranteeing the travelling comfort, through reducing the vehicle and go the in-process frequently with the speed reduction proportion, reduces unnecessary energy resource consumption and emission.

The speed optimization objective is to optimize braking and driving forces in the prediction horizon to minimize vehicle energy consumption over the entire horizon, with the control variable u = [ F = [ ] _t ,F _b ] ^T ,t′∈[t,t+T]Then the optimization problem can be described as

L(·)＝ωP _loss (x,u,t)+ω _dr ·max{F _t ² ,F _b ² }

φ(x(t _f ))＝(v(t _f )-v _f ) ² +(s(t _f )-s _f ) ²

F _t ∈[0,F _t,max ]F _b ∈[0,F _b,max ]

Wherein x = [ v, s] ^T Is state variable, v (t), s (t) are vehicle speed and running distance, M is vehicle mass, F _t (τ) is the driving force of the entire vehicle, F _b (τ) is braking force, c _d ρA _f [ 2 ] is air resistance, and gfcos (θ) + gsin (θ) is rolling resistance and gradient resistance due to gravity and gradient (θ represents road gradient).

Wherein the weight coefficient related to the driving condition and the driving style is omega _dr The present invention is embodied in control of driving force and braking force. During the experiment, the obtained omega _dr Tabulated as table 3, the driving speed is adjusted by predicting the driving condition and driving style at the next moment to make the vehicle runThe process is safe and economical. For example, if aggressive drivers demand more power for the vehicle and there is a high risk of excessive speed during high speed, then ω is used _dr7 The driving force is adjusted to achieve the purpose of speed optimization.

TABLE 3 Driving Condition and Driving Style weight coefficient Table

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents and improvements made within the spirit and principle of the present invention are intended to be included within the scope of the present invention.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims

1. An energy-saving speed optimization method under the influence of man-vehicle-road coupling is characterized by comprising the following steps of:

2. The energy-saving speed optimization method under the influence of human-vehicle-road coupling according to claim 1, wherein in the step (1), the driving simulation scene is constructed through joint simulation of SCANeR and Matlab software.

3. The energy saving speed optimization method under the influence of man-vehicle-road coupling according to claim 1, wherein in the step (1), the summoned drivers are grouped into different ages, driving ages and sexes.

4. The energy-saving speed optimization method under the influence of human-vehicle-road coupling according to claim 1, wherein in the step (2), the processing and analyzing of the collected data specifically comprises: firstly, filtering and standardizing by a filter, and then extracting characteristic parameters of the driving condition and the driving style; and under the condition of identifying different working conditions, identifying the driving style.

5. The energy-saving speed optimization method under the influence of human-vehicle-road coupling according to claim 4, wherein the different operating conditions comprise urban, suburban and high-speed operating conditions.

6. The energy-saving speed optimization method under the influence of human-vehicle-road coupling according to claim 1, wherein in the step (3), the weight coefficient table is used for calculating the magnitude of the driving braking force through the weight coefficient in the objective function.