CN-115271074-B - Man-machine sharing control regionalization decision method based on Gaussian hidden Markov model

CN115271074BCN 115271074 BCN115271074 BCN 115271074BCN-115271074-B

Abstract

The invention discloses a man-machine sharing control regional decision-making method based on a Gaussian hidden Markov model, which comprises the following steps of firstly collecting information, secondly processing information, thirdly judging grades, fourthly, primarily distributing, fifthly, finally distributing and updating in real time, wherein in the first step, the real-time driving information of a driver and the real-time driving information of an auxiliary driving system are collected from a control panel, sensing equipment and the auxiliary driving system module through the information collecting module, different response capacities and sensibility of the driver to environmental risks from different directions are comprehensively considered, the double-chain structure of the hidden Markov model is fully utilized, comprehensive consideration of the specificity of the driver, the relative driving capacity, the environmental risk degree of the vehicle, the region where the driver is located and the like is realized, and finally, the self-adaptive automatic adjustment of a control right switching strategy of the driver is realized, so that the driving is safer and more comfortable.

Inventors

LIU FANG
SU WEIXING
ZHU TIANHE

Assignees

天津工业大学

Dates

Publication Date: 20260505
Application Date: 20220704

Claims (8)

1. A man-machine sharing control regionalization decision method based on a Gaussian hidden Markov model comprises the following steps of firstly, information acquisition, secondly, information processing, thirdly, grade judgment, thirdly, preliminary distribution, fifth, final distribution and sixth, and is characterized in that: In the first step, the information acquisition module is used for collecting real-time driving information of a driver and real-time driving information of an auxiliary driving system from the control panel, the sensing equipment and the auxiliary driving system module; In the second step, the driver's real-time driving information is transmitted to the information processing module through the driver's pedestrian information unit of the information acquisition module, and the real-time driving ability and the real-time regional risk of driving are calculated respectively through the driver's real-time ability calculation unit and the regional real-time risk calculation unit based on the driving risk field, wherein the calculation process is as follows: first, the real-time driving ability of the driver is calculated: Firstly, seven characteristics representing the transverse and longitudinal running states of the vehicle are selected as indexes for evaluating the driving capacity of a driver, wherein the indexes comprise the standard deviation of the speed Standard deviation of acceleration Standard deviation of following distance Time distance of head of a vehicle Is a vehicle longitudinal state index and a vehicle lateral position deviation standard deviation Steering wheel corner entropy Steering wheel turn frequency Is a vehicle lateral state index of (a); Subjectively quantifying individual driving ability of the driver by using a hierarchical analysis algorithm based on group decision, and based on seven selected vehicle driving state characteristics, thereby the method is the first Expert in the bit according to' Evaluation matrix given by scale system Is that Dimension, and evaluation matrix consistency test is as follows: (1); In the formula, ; , For evaluating the maximum eigenvalue of the matrix, the corresponding eigenvector is , To evaluate the dimension of the matrix When the evaluation matrix meets the consistency requirement, namely passing the consistency test Normalized evaluation weight given by expert The calculation is as follows: (2); then objectively quantifying the individual driving ability of the driver by utilizing an entropy weight method to obtain each index weight value Information entropy for each index Corresponding difference coefficient The ratio of the occupied overall difference coefficients is as follows: (3); In the formula, Representing the coefficient of difference, i.e., ; And the first Information entropy of individual index The following are provided: (4); In the formula, Represents data normalized by the polar method, Representing the number of samples; and then calculating to obtain subjective and objective mixing weights: (5); In the formula, Representing the number of experts in the ANP algorithm, and obtaining the personal driving ability value of the driver so far: (6); In the formula, ; Secondly, calculating the risk of driving real-time areas: firstly analyzing traffic unit risk quantity, when a traffic unit collides, releasing kinetic energy through extrusion and collision, so that deformation occurs to both sides of the collision, and representing the traffic unit risk quantity, wherein the traffic unit risk quantity R is based on the analysis: (7); In the formula, The vehicle coefficient comprises a vehicle type coefficient and a vehicle loading coefficient, and the truck coefficient is larger than the car coefficient; For the quality of the traffic element, Is the speed of the vehicle in the traffic unit, For correction of the coefficients, here , The maximum value of the speed limit of the current road is set; and then carrying out distance-based risk correction, wherein for the same risk source, the closer the distance is, the larger the risk is, and the traffic unit can be used For traffic units Distance-based risk correction Expressed as: (8); In the formula, Is that Traffic unit to The distance vector of the traffic unit, For correction of the coefficients, here =1; Then, risk correction based on the motion state is carried out, and the traffic units are defined by considering the relative motion magnitude and direction among the traffic units For traffic units Is based on the risk correction of the movement state The method comprises the following steps: (9); In the formula, Is a traffic unit Traffic unit Is used for the relative velocity vector of (a), Is a relative velocity vector And distance vector Is included in the plane of the first part; Then carrying out risk correction based on traffic rules, namely considering that the environmental risk is constrained by the traffic rules when the traffic units run, and defining the risk correction based on the traffic rules The method comprises the following steps: (10); In the formula, As the lane line type coefficient, And when it is indicated by a broken line, The time is indicated by a solid line, Is the distance vector of the traffic unit to the lane line, If the traffic units are in a solid line, the behavior crossing the traffic line is generally not generated, and the risk is relatively low, but if the traffic units are in a broken line and the vehicle is close to the broken line, the probability of the lane change running is considered to be increased, and the risk is increased for the host vehicle; finally, comprehensively considering the risk quantification models defined by the formulas (7) to (10), the traffic unit can be obtained For traffic units Is the total risk of (2) The definition is as follows: (11); In the formula, ; And Respectively represent traffic units calculated by the formula (7) Traffic unit Is a traffic unit risk amount; In the third step, the real-time driving ability of the driver is input into the real-time driving ability relative evaluation module, the driving ability of all drivers for safe driving in the statistical data set, i.e. the gaussian distribution statistical diagram formed by the public driving ability is utilized, and the real-time relative driving ability of the driver, i.e. the probability of the driving ability value in the overall distribution, is obtained according to the position of the driving ability quantized value in the real-time driving ability calculation unit in the gaussian distribution statistical diagram, and the driving real-time regional risk is input into the regional quantization risk field module to obtain the intelligent automobile Is the center of a circle and the radius is All non-shielding risk sources in a circle of a meter, namely other traffic elements, dividing the circle into six areas, calculating total amount risk fields generated by all non-shielding risk sources in each area to obtain regional surrounding environment quantitative risk fields of a main traffic unit, obtaining real-time regional driving risks, and then respectively judging the relative driving capability level of a driver and the real-time important corresponding area risk level, wherein the calculation formula of the total amount risk fields is as follows: (12); In the formula, Represent the first The total amount of risk field for each region, ; Represent the first Number of risk sources for each region, then coefficient for 6 regions Respectively take out ; In the fourth step, the relative driving capability level of the driver and the risk level of the real-time important coping region are input into the control right calculation module, and the maximum value in the total amount risk field of six regions is taken through the Gaussian distribution relative capability and risk calculation unit The emphasis faced for the primary traffic unit is on risk, In order to deal with the region where the risk is important, the distribution level of the important risk is based The relative driving ability level of the driver is checked, and a preliminary control right allocation strategy of man-machine co-driving is determined by Gaussian distribution; in the fifth step, the preliminary control right allocation strategy is input into the hidden markov calculation unit, the preliminary control right allocation strategy is refined by using the hidden markov model, the driving right conversion between the intelligent driving model and the driver is described by using the hidden markov chain, and the driving right conversion is performed according to the state set And state transition matrix Wherein, the 1 state indicates that the driver obtains the main driving right, the 2 state indicates that the intelligent driving model obtains the main driving right, and the observation probability matrix is integrated: (13); I.e. the probability of driving in a smart driving model is proportional to the current emphasis on risk of the main traffic unit, Is the area of Related coefficients; and gives the flexible control right distribution coefficient of the man-machine co-driving control at the moment t based on the Viterbi algorithm: (14); The servo stage shares the actual output: (15); In the formula, In order for the control amount to be in effect, The control amount is input for the driver, Inputting control quantity for the intelligent driving model, and further calculating a final driving control right allocation strategy; In the sixth step, the control right calculation module inputs the relative driving capability level of the driver into the absolute driving capability updating module, and the state transition matrix in the hidden markov model is utilized to evaluate the long-term absolute driving capability of the driver, wherein the state transition matrix a adopts the following updating mode with forgetting factors: (16); In the formula, Subscript of (2) Respectively represent And A master state of time; Representing the maximum value in the vector and taking out the position of the maximum value As the root cause The main state of the moment of time, (17); It can be seen from the updated formulae (16) and (17) of the state transition matrix that, for a driver with better driving ability, he can grasp more driving initiative during long-term driving, then he is in the state transition matrix And Compared with And The final driving control right distribution strategy is updated in real time, so that the long-term objective evaluation of the absolute driving ability of the driver is realized.
2. The method for localized decision making of human-machine sharing control based on Gaussian hidden Markov model as set forth in claim 1, wherein in said step one, the input terminals of the information acquisition module are connected with the output terminals of the control panel, the sensing device and the driving assistance system module, respectively.
3. The method for localized decision-making of human-machine sharing control based on Gaussian hidden Markov model of claim 1, wherein in the second step, the input end of the driver information unit is connected with the output end of the control panel, and the output end of the driver information unit is connected with the input end of the information processing module.
4. The method for localized decision-making based on Gaussian hidden Markov model of claim 1, wherein in the third step, the input end of the real-time driving capability relative evaluation module is connected with the output end of the driver real-time capability calculation unit, and the input end of the localized quantized risk field module is connected with the output end of the localized real-time risk calculation unit based on driving risk field.
5. The method for man-machine sharing control regionalization decision-making based on Gaussian hidden Markov model according to claim 1, wherein in the fourth step, the input end of the control right calculation module is respectively connected with the output ends of the information processing module, the real-time driving capability relative evaluation module and the regionalization quantization risk field module.
6. The method for man-machine sharing control regionalization decision-making based on Gaussian hidden Markov model according to claim 1, wherein in the fourth step, the output ends of the control right calculation module, the driver information unit and the auxiliary driving system module are all connected with the input end of the control panel.
7. The method for localized decision making of human-machine sharing control based on Gaussian hidden Markov model of claim 1, wherein in the fifth step, the input end of the hidden Markov calculation unit is connected with the output end of the Gaussian distribution relative capacity and risk calculation unit.
8. The method for localized decision-making of human-machine sharing control based on Gaussian hidden Markov model of claim 1, wherein in the sixth step, the input end and the output end of the absolute driving ability update module are connected with the output end and the input end of the control right calculation module, respectively.

Description

Man-machine sharing control regionalization decision method based on Gaussian hidden Markov model Technical Field The invention relates to the technical field of automatic driving and man-machine co-driving, in particular to a man-machine sharing control regional decision method based on a Gaussian hidden Markov model. Background With the continuous evolution of automatic driving technology, an auxiliary driving system has become a standard of intelligent automobiles, but the shared control driving system represented by man-machine co-driving under an open road scene still has the problems of safety, reliability and the like. Therefore, under the 'double drivers' environment of man-machine co-driving, the high reliable and reasonable allocation of driving rights is important for the development and driving safety of intelligent automobiles. The current main stream research thought is mostly focused on comprehensively considering the state of a driver, and the deviation degree of the decision or intention of the driver and the decision of an intelligent model is utilized as a main basis for distributing the servo sharing control right, wherein the process only considers the state or intention of the driver and the state of a vehicle to distribute the driving control right, and does not consider the specificity of the driver (the specificity refers to the correction capability of the driver with different driving capabilities for coping with the same driving risk to different degrees). The driving control right allocation strategy for evaluating the state of the driver only easily generates the defect of excessive intervention or insufficient intervention, so that the self driving capability of the driver is weakened, the human-computer interaction comfortableness is reduced, the original purpose of the man-machine co-driving problem research is overcome, secondly, the existing servo level sharing control problem research is too general in consideration of the surrounding environment risk, the detail is not generated, and the vigilance and the coping capability of the driver for the front risk and the rear risk are different in practice, so that if the general environment total risk is considered, the distinction is not reasonable. Disclosure of Invention The invention aims to provide a man-machine sharing control regionalization decision method based on a Gaussian hidden Markov model, so as to solve the problems in the background art. The man-machine sharing control regionalization decision method based on the Gaussian hidden Markov model comprises the following steps of first step information acquisition, second step information processing, third step grade judgment, fourth step preliminary distribution, fifth step final distribution, sixth step real-time updating; In the first step, the information acquisition module is used for collecting real-time driving information of a driver and real-time driving information of an auxiliary driving system from the control panel, the sensing equipment and the auxiliary driving system module; In the second step, the driver's real-time driving information is transmitted to the information processing module through the driver's pedestrian information unit of the information acquisition module, and the real-time driving ability and the real-time regional risk of driving are calculated respectively through the driver's real-time ability calculation unit and the regional real-time risk calculation unit based on the driving risk field, wherein the calculation process is as follows: first, the real-time driving ability of the driver is calculated: Firstly, seven characteristics representing the transverse and longitudinal running states of a vehicle are selected as indexes for evaluating the driving capability of a driver, wherein the indexes comprise a speed standard deviation Std v, an acceleration standard deviation Std a, a following distance standard deviation Std DHW, a vehicle longitudinal state index of a vehicle headway THW, a vehicle transverse position offset standard deviation Std d, a steering wheel corner entropy SE and a vehicle transverse state index of a steering wheel transfer frequency TF; Subjectively quantifying individual driving capacity of the driver by using a hierarchical analysis algorithm based on group decision, and based on the seven selected vehicle driving state characteristics, so that the ith expert can evaluate the matrix according to a 5/5-9/1 scale system For 7 x 7 dimensions, the evaluation matrix consistency test is as follows: wherein ri= 0.4007; For evaluating the maximum eigenvalue of the matrix, the corresponding eigenvector is M is the dimension of the evaluation matrix, when CR i < = 0.1, the evaluation matrix meets the consistency requirement, namely passes the consistency test, and the i-th expert gives the normalized evaluation weightThe calculation is as follows: Then, the individual driving capability of the driver is objecti