A distributed electric vehicle function agent behavior intention consistency determination method

Abstract

The application discloses a kind of distributed electric vehicle function Agent behavior intention consistency determination method, first to distributed electric vehicle according to electronic control unit function partition, the function area divided out is described using Agent language;Dynamics model representing function Agent system is constructed, and using torque, displacement sensor real-time collection vehicle function Agent behavior intention consistency determination required information, sensor acquisition data and vehicle dynamics equation are combined to construct function Agent system as power operator, real-time estimate different function Agent in past time to the pseudo work done to vehicle, finally based on Fisher linear discriminant algorithm establishes the linear classifier of function Agent consistency, completes the real-time determination and classification of function Agent target consistency degree in the process of vehicle driving.This method can effectively identify the behavior intention between different function Agents, is more simple, reliable, real-time and efficient to complete function Agent behavior intention consistency determination.

Description

Technical Field

[0001] This invention relates to the fields of automotive intelligent interaction and autonomous driving, and particularly to a distributed electric vehicle functional agent behavior intention. Figure 1 Methods for determining consistency. Background Technology

[0002] Currently, automotive intelligent chassis electronic control systems are becoming increasingly diversified. However, due to the lack of a reasonable overall model and control architecture, the functional contradictions among the basic control units are becoming more and more apparent. That is, current vehicle electronic control functions often adopt a relatively independent development model, making it difficult to achieve "intelligent" coordinated control of distributed electric vehicles. When the execution intentions of various chassis functional units conflict during autonomous driving, it often leads to redundancy or independent operation within the distributed electric vehicle. Therefore, accurately identifying the functional intentions of the control units is crucial for the comprehensive "intelligent" coordination of various control functions of the vehicle chassis under different environmental requirements and for more comfortable motion control in autonomous vehicles. In particular, rationally dividing the electronic control units of the entire chassis system of a distributed electric vehicle into functional parts and describing them using Agent language to construct the smallest unit of system integration control, thereby better identifying the execution intentions of functional units and ensuring the safe and efficient operation of the vehicle.

[0003] Chinese patent application number 201910730626.0 discloses "A driver intent understanding module for an unmanned driving control system," which discusses a method for understanding driver intent based on an unmanned driving system, significantly reducing the data computation load of the control system and improving processing efficiency. However, this method relies solely on a signal acquisition module to collect driver intent. Due to the limited scope of the acquired intent and the inconvenience of data acquisition, this method cannot better meet the various control functions of the vehicle chassis under different environmental conditions, and it also suffers from a low level of intelligence. Summary of the Invention

[0004] The technical problem this invention aims to solve is to provide a distributed electric vehicle functional agent that can accurately identify the behavioral intent of functional units, effectively eliminate coupling interference from execution units, and has high execution efficiency and practicality. Figure 1 Methods for determining consistency.

[0005] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0006] A distributed electric vehicle functional agent behavior intention Figure 1 The consistency determination method includes the following steps:

[0007] (1) For distributed drive electric vehicles, the functions are divided according to the electronic control unit, and the divided functional areas are described using Agent language;

[0008] (2) Construct dynamic equations representing the operations performed by different functional agent systems;

[0009] (3) Construct a pseudo-power operator for the functional agent system using the longitudinal and lateral dynamic equations of distributed electric vehicles and solve it in real time;

[0010] (4) Establish a linear classifier based on Fisher's linear discriminant algorithm to determine the consistency level of functional agent behavior intention.

[0011] Step (2) For different functional agent units, construct dynamic equations characterizing the operations performed by the functional agent units, including:

[0012] (21) Construct the driver agent dynamic equation. The driver's behavioral intention is determined by the maneuvering input torque τ of the distributed electric vehicle. d This indicates that it includes the steering torque input τ. dh , driving torque input τ dd Braking torque input τ db .

[0013] First, determine the driver's steering torque input. By combining the driver with the steer-by-wire system, the dynamic expression of the driver's hand torque input can be obtained:

[0014]

[0015] Among them, J dam For the equivalent inertia of the steering system, B inter For the steering system damping, τ f For the frictional torque of the steering system, τ e The steering resistance torque provided to the motor, δ is the steering wheel angle, and sign() is the sign function used to indicate the positive or negative sign of the parameter.

[0016] For distributed electric vehicles, electronic pedals are used to control the vehicle's drive and braking. The drive and braking torques can be determined based on different electronic pedal openings. Therefore, the formula for expressing the drive / braking torque is:

[0017]

[0018] Where, Δl d The actual accelerator pedal opening Δl d-real With respect to the target accelerator pedal opening Δl d-target The difference, Δl bFor the actual brake pedal opening Δl b-real With respect to the target brake pedal opening Δl b-target The difference, K b ,a b ,b b K is the coefficient for optimizing braking torque. d ,a d ,b d The driving torque optimization coefficient is determined based on the vehicle's control requirements. The relationship between the increase in the difference between the driver's input driving and braking torque and the electronic pedal opening is divided into three stages: 1. Slight increase in driving and braking torque: The opening curve is smoothed using a weighted moving average method to improve the rate of change of the electronic pedal opening characteristic curve and ensure a better driving experience; 2. Linear increase in driving and braking torque: The electronic pedal opening curve is smoothed using a simple translation method; 3. Saturation of driving torque increase.

[0019] (22) Constructing the dynamic equations of the steering function agent

[0020] For distributed electric vehicle steer-by-wire systems, the dynamic characterization of the steering agent can use the torque output of the steering actuator motor as a key parameter characterizing the steering control behavior. Its dynamic expression is as follows:

[0021] τ a =I a K t η (3)

[0022] Where, τ a For steering system control output, I a K represents the actual current of the steering motor. t η is the torque coefficient of the steering motor, and η is the mechanical efficiency.

[0023] (23) Dynamic equations of the driving / braking agent

[0024] For distributed electric vehicles where all four wheels are independently controllable, their driving and braking torques can be controlled by full-vector torque. The torque response of the drive motor is used as a key parameter characterizing the intended driving / braking behavior, and its dynamic expression is as follows:

[0025]

[0026] Where, τ all τ represents the total output torque of the hub motor. t τ represents the target output torque of the motor. max τ represents the maximum output torque at the current motor speed; τ is a time constant, which can be obtained from the torque response data of the drive motor.

[0027] In step (3), the design based on the pseudo-power operator includes the following:

[0028] (31) Test data collection: Using on-board sensors, test data were collected for the distributed electric vehicle function Agent, including driver hand torque, steering control torque, drive pedal travel, brake pedal travel, steering wheel angle, steering wheel speed, wheel hub motor driving torque, wheel hub motor driving torque, and vehicle speed.

[0029] (32) Experimental data processing: Since data is collected through different sensors, some of the collected data is not easy to understand and observe intuitively. Therefore, it is necessary to convert the units of the data, convert the steering wheel angle and speed from radians to degrees, and the speed from m / s to km / h. The data is divided into four categories: driver operation data, steering function data, drive function data, and braking function data, so as to facilitate input into the four functional agent areas for identification. The data in each group is segmented, and each time period represents the operation behavior of the functional agent within a certain period of time. For each data segment, an adaptive Kalman filter algorithm is used to remove outliers in each data segment.

[0030] (33) To determine the behavioral intent of the execution unit, it is necessary to calculate the instantaneous pseudo-power of the driver agent, steering agent, and drive / brake agent systems.

[0031] The instantaneous pseudo-power expression for driver-agent steering control is as follows:

[0032]

[0033] Among them, w h (t) represents the instantaneous pseudo-power of the driver's steering maneuver, T w For the periodic time, v y Let t be the lateral speed of the vehicle and t be the travel time.

[0034] The method for calculating the pseudo-power of driver agent driving control is as follows:

[0035]

[0036] Among them, w dd (t) represents the instantaneous pseudo-power of the driver's driving operation, and vx represents the longitudinal speed of the vehicle.

[0037] The method for calculating the pseudo-power of driver-agent braking is as follows:

[0038]

[0039] Among them, w db (t) represents the instantaneous pseudo-power of the driver's braking operation;

[0040] The method for calculating the instantaneous pseudo-power of the steering system agent is as follows:

[0041]

[0042] Among them, w a (t) represents the instantaneous pseudo-power of the vehicle's auxiliary steering system.

[0043] The method for calculating the instantaneous pseudo-power of the control / drive system agent is as follows:

[0044]

[0045] Among them, w ab (t) represents the instantaneous pseudo-power of the vehicle drive assistance system, w ad (t) represents the instantaneous pseudo-power of the vehicle braking assist system.

[0046] Step (4) Establish a linear classifier based on Fisher's linear discriminant algorithm to determine the consistency level of functional agent behavior intent, including:

[0047] (41) First, the behavioral intentions of the functional agent are divided into steering intention, driving intention, and braking intention; the key parameter for determining the behavioral intentions of the functional agent is selected as: steering intention parameter w h ,w a Driver intent parameter w dd ,w ad Braking intention parameter w db ,w ab The evaluation level is divided into two levels (high and low).

[0048] (42) For the three pairs of key parameters for determining behavioral intentions obtained in the experiment, 40 sets of data were generated within two levels of standards. 20 sets of data were randomly extracted from each level. The first 15 sets of data were used for model training, and the last 5 sets of data were used for model testing.

[0049] (43) Based on the model data in (42), Fisher's decision analysis method is used to classify and determine the behavioral intentions of the functional agent:

[0050] The behavior of each pair of parameters in (42) Figure 1 Consistency levels are divided into two categories, defined as follows: and Where i = 1, 2, 3 represent steering intention, braking intention, and driving intention, respectively. Fisher's determination method is used to determine steering intention. Figure 1 Classify by consistency.

[0051] Given a steering behavior dataset y j The instantaneous power occurs at time j = 1, 2, ..., N, where N represents the total number of samples. This refers to the instantaneous power during steering.

[0052] Let T h Indicates intention to change direction Figure 1 High affinity level The set of examples of each category, the sample set T h The mean vector is represented as Let T l Indicates intention to change direction Figure 1 Low affinity level The sample set of each category, sample set T l The mean vector is represented as

[0053] Assume the classification line is:

[0054]

[0055] in, Indicates intention to change direction Figure 1 The projection vector of the consistency determination sample; Indicates intention to change direction Figure 1 Consistency determination sample dataset;

[0056] Turning behavior intention Figure 1 The inter-class divergence of samples for consistency determination is:

[0057]

[0058] S b Representing vectors His own external product, that is, the turning towards behavioral intention Figure 1 Inter-class divergence for consistency determination;

[0059] Turning behavior intention Figure 1 The within-class divergence of samples for consistency determination is:

[0060]

[0061] Among them, S w The intra-class divergence of samples used to determine consistency in steering behavior;

[0062] The objective function of linear discriminant analysis is to maximize:

[0063]

[0064] Among them, J s Let S be the between-class divergence of the samples. bWith sample intraclass divergence S w The broad definition of Rayleigh.

[0065] Repeat step (54) to obtain the generalized Rayleigh quotient J of the braking determination sample. b and the generalized Rayleigh quotient J of the driving sample d .

[0066] Based on the above theory, the key parameters for determining the behavioral intent of the functional agent are: the turning intent parameter w. h ,w a Driver intent parameter w dd ,w ad Braking intention parameter w db ,w ab The optimal solution is obtained using Fisher's decision theory. The behavior of each functional agent unit can be determined if and only if the steering intention, driving intention, and braking intention are all consistent. Figure 1 To determine the intent of the behavior Figure 1 The consistency level is high; otherwise, the behavior is judged as intentional. Figure 1 The consistency level is low, at which point a functional agent unit needs to autonomously adjust until the behavioral intention is obtained. Figure 1 When the consistency is determined to be high, the mechanism is executed to complete the intelligent coordinated control of the distributed electric vehicle.

[0067] The above design takes into account the intelligent coordination and control requirements of distributed electric vehicles. By adjusting the individual functional agents as needed, vehicle safety can be effectively guaranteed.

[0068] Compared with the prior art, the beneficial effects of the present invention are:

[0069] This invention, when determining the consistency of the behavioral intentions of distributed electric vehicle (EV) control units, rationally divides the EV control units into functional parts and converts them into Agent language descriptions, thereby constructing the smallest unit for system integrated control. Based on the instantaneous pseudo-power operator, it identifies the behavioral intentions of functional Agent units, solving the problem of poor real-time performance in traditional algorithms. Utilizing Fisher's decision analysis method for consistency determination of functional Agent behavioral intentions has low computational cost, strong real-time performance, and the determined behavioral intention results are safer and more accurate than traditional methods. It can provide technical support for the efficient and intelligent coordinated control of distributed electric vehicles. Furthermore, based on accurate identification of functional unit behavioral intentions, it can effectively eliminate execution coupling conflicts within the EV control units, demonstrating strong practicality.

[0070] This invention relates to distributed electric vehicle functional agent behavior intent. Figure 1The consistency determination method divides functional areas, categorizing different functional units into driver functional areas, steering functional areas, and drive / braking functional areas. Corresponding functional agent units are established based on these areas. The driver agent is an intelligent agent capable of accurately representing the driver's characteristics, including cognitive and manipulative traits. The steering agent is an intelligent agent capable of autonomously controlling steering and responding precisely, exhibiting active steering and high execution accuracy. The drive / braking agent is an intelligent agent capable of distributed drive / braking control, possessing efficient and distributed collaborative control characteristics. A single functional agent represents the micro-level of a multi-agent system, possessing responsiveness, autonomy, and flexibility. The relationships between functional agents constitute the macro-level of the multi-agent system, enabling higher flexibility, environmental adaptability, and scalability through the organization and cooperation of each level. This coordinated control approach is highly suitable for the intelligent control of distributed electric vehicles. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the steer-by-wire system of the present invention;

[0072] Figure 2 This is a drive / brake pedal opening curve diagram processed by weighted processing and simple translation method according to the present invention;

[0073] Figure 3 This is a schematic diagram of the linear discriminant analysis (LDA) principle of the present invention;

[0074] Figure 4 This invention relates to the distributed electric vehicle functional agent behavior intent. Figure 1 Schematic diagram of the consistency determination method. Detailed Implementation

[0075] The specific embodiments of the present invention will now be described with reference to the accompanying drawings to enable those skilled in the art to better understand the invention. It should be noted that in the description of the present invention, detailed descriptions of known functions and designs that might obscure the main content of the invention will be omitted here.

[0076] A distributed electric vehicle functional agent behavior intention Figure 1 Methods for determining consistency, such as Figure 4 It includes the following steps;

[0077] Step (1): Divide the distributed electric vehicle functional agent area. Divide the different execution units of the distributed electric vehicle into driver function area, steering control function area, drive control function area and braking control function area according to their functions.

[0078] Step (2): Use Agent language to describe the functional areas of the execution unit:

[0079] Based on the functional areas of the execution unit in step (1), corresponding functional agent units are established. The driver functional area corresponds to the driver agent, which is an intelligent agent that can accurately represent the characteristics of the driver, including thinking and manipulation characteristics. The steering control functional area corresponds to the steering agent, which can autonomously control the steering and has the characteristics of active steering and high execution accuracy. The drive control functional area corresponds to the drive agent, and the braking control area corresponds to the braking agent. They can perform distributed collaborative control of the vehicle and have the characteristics of efficient, distributed intelligent collaborative control. A single functional agent is the micro level of the multi-agent system, with responsiveness, autonomy, and flexibility. The relationship between functional agents constitutes the macro level of the multi-agent system. Through the organization and cooperation of each level, a comprehensive function with higher flexibility, environmental adaptability, and scalability can be achieved. This coordinated control idea is very suitable for the intelligent control of distributed electric vehicles.

[0080] Step (3): Perform dynamic description on the constructed functional agent.

[0081] See Figure 1 As can be seen, based on the principle of steer-by-wire control, the steering system can be divided into two parts: the driver control area and the intelligent control area. The driver control area can collect manipulation parameters to characterize the driver's intention. This invention utilizes the input torque τ of the distributed electric vehicle... d The analysis is performed to characterize the driver's behavioral intentions, including the steering torque input τ. dh .

[0082] (31) Constructing the driver agent's dynamic equations. First, determine the driver's steering torque input. By combining the driver with the steer-by-wire system, the dynamic expression of the driver's hand torque input can be obtained:

[0083]

[0084] Among them, J dam For the equivalent inertia of the steering system, B inter For the steering system damping, τ f For the frictional torque of the steering system, τ e The steering resistance torque provided to the motor, δ is the steering wheel angle, and sign() is the sign function used to indicate the positive or negative sign of the parameter.

[0085] See Figure 2The design of the relationship between electronic pedal opening and driving / braking torque takes into account the actual torque of the wheel hub motor. The designed curve has been smoothed to ensure a good driving experience. The curve design includes the following:

[0086] For distributed electric vehicles, electronic pedals are used to control the vehicle's driving and braking, and the driving torque input τ is selected. dd Braking torque input τ db This represents the driver's driving and braking intentions. The driving / braking torque is primarily determined by different electronic pedal openings, and the formula for expressing the driving / braking torque is:

[0087]

[0088] Where, Δl d The actual accelerator pedal opening Δl d-real With respect to the target accelerator pedal opening Δl d-target The difference, Δl b For the actual brake pedal opening Δl b-real With respect to the target brake pedal opening Δl b-target The difference, K b ,a b ,b b K is the coefficient for optimizing braking torque. d ,a d ,b d The driving torque optimization coefficient is determined based on the vehicle's control requirements. The relationship between the increase in the difference between the driver's input driving and braking torque and the electronic pedal opening is divided into three stages: 1. Slight increase in driving and braking torque: The opening curve is smoothed using a weighted moving average method to improve the rate of change of the electronic pedal opening characteristic curve and ensure a better driving experience; 2. Linear increase in driving and braking torque: The electronic pedal opening curve is smoothed using a simple translation method; 3. Saturation of driving torque increase.

[0089] For the slight increase in drive / braking torque, the opening curve is smoothed using a weighted moving average method to improve the rate of change of the electronic pedal opening characteristic curve and ensure a better driving experience.

[0090] The slight intervention phase of the drive / braking torque distribution curve is smoothed using a weighted average method. The calculation process of the weighted moving average method is as follows:

[0091]

[0092] Where, r d1 ,r b1 These are the weights of the actual values ​​of the first point of the driving and braking torques, r. d2 ,r b2These are the weights of the actual values ​​of the second point for the driving and braking torques, respectively, r. dn ,r bn These are the weights of the actual values ​​of the driving and braking torques at the nth point, respectively, where n is the predicted value and r is the weight of the actual values. b1 +r b2 +…r bn =1, r d1 +r d2 +…r dn =1. Select n τ' dd_n-1 ,τ' dd_n-2 ,τ' dd_n-3 …,τ' db_n-1 ,τ' db_n-2 ,τ' db_n-3 …, thus obtaining a series of points (x dd1 ,y dd1 ),(x dd2 ,y dd2 ),(x dd3 ,y dd3 )...,(x db1 ,y db1 ),(x db2 ,y db2 ),(x db3 ,y db3 ..., select initial values ​​T respectively dd1 ,T db1 Substituting the selected points into formula (3) one by one, we can obtain:

[0093]

[0094] For the linear increase in driving and braking torque stage, the electronic pedal opening curve is smoothed using a simple translation method. The calculation for the simple translation method is as follows:

[0095]

[0096] Where τ dd ,τ db These are the predicted values ​​for the next point, where n is the number of points in the moving average, and f is the predicted value. dd1 (Δl d ),f db1 (Δl b f is the actual value of the previous point. dd2 (Δl d ), f dd3 (Δl d ),...f ddn (Δl d );f db2 (Δl b ),f db3 (Δl b ),...fdbn (Δl b Let τ represent the actual values ​​of the first two points, the first three points, and so on up to the nth point. Select n points τ respectively. dd1 ,τ dd2 ,τ dd3 …τ ddn ;τ db1 ,τ db2 ,τ db3 …τ dbn The resulting series of points is: (x dd1 ,y dd1 ),(x dd2 ,y dd2 ),(x dd3 ,y dd3 ...; (x db1 ,y db1 ),(x db2 ,y db2 ),(x db3 ,y db3 ...

[0097] Choose initial values ​​as T dd1 ,T db1 Substituting the selected points into equation (5) one by one, we get:

[0098]

[0099] (32) Steering Agent Dynamics Equation. For a distributed electric vehicle steer-by-wire system, the dynamics of the steering agent can be characterized by the torque output of the steering actuator motor as a key parameter representing the intention of steering control behavior. Its dynamic expression is as follows:

[0100] τ a =I a K t η (7)

[0101] Where, τ a For steering system control output, I a K represents the actual current of the steering motor. t η is the torque coefficient of the steering motor, and η is the mechanical efficiency.

[0102] (33) Dynamic equations for the driving / braking agent. For a distributed electric vehicle where all four wheels are independently controllable, its driving and braking torques can be controlled by full vector torque. The torque response of the drive motor is used as a key parameter characterizing the driving / braking behavior. Its dynamic expression is:

[0103]

[0104] Where, τ allτ represents the total output torque of the hub motor. t τ represents the target output torque of the motor. max τ represents the maximum output torque at the current motor speed; τ is a time constant, which can be obtained from the torque response data of the drive motor.

[0105] Step (4) Test Data Acquisition. Using onboard sensors, test data is collected for the distributed electric vehicle function agent, including driver hand torque, steering control torque, drive pedal travel, brake pedal travel, steering wheel angle, steering wheel speed, hub motor driving torque, hub motor driving torque, and vehicle speed;

[0106] Step (5) Experimental Data Processing. Since data is collected through different sensors, some of the collected data is not easily understood or observed intuitively. Therefore, it is necessary to convert the units of the data, converting the steering wheel angle and speed from radians to degrees, and the speed from m / s to km / h. The data is divided into four categories: driver control data, steering function data, drive function data, and braking function data, to facilitate input into the four functional agent areas for identification. The data in each group is segmented, with each time period representing the functional agent's control behavior over a period of time. For each data segment, an adaptive Kalman filter algorithm is used to remove outliers.

[0107] The filter state equation is:

[0108] p k =Ap k-1 +w k (9)

[0109] Where: p k Let p be the system state at time k. k-1 Let A be the system state at time k-1, and let A be the state transition matrix; w k This system noise can be attributed to sensor error; it is assumed to be zero-mean Gaussian white noise, and w k ∈(0,M). The observation equation of the system is:

[0110] q k =Hp k-1 +v k (10)

[0111] Where q k Let v be the observation value at time k, and H be the parameters of the observation system; k The observation noise is mainly time-quantization noise, and v k ∈(0,T0). The state update equation for the Kalman filter is:

[0112]

[0113] The measurement update equation for the Kalman filter is:

[0114]

[0115] in, For prior state estimation, For posterior state estimation, B' k For the prior estimation of error covariance, B k To estimate the error covariance in the posterior timescale,

[0116] When the prediction result indicates an increasing error, the Taylor series expansion algorithm at the current iteration number is determined to be in a divergent state. The measured value of the current state is then considered a bad point and is removed.

[0117] Step (6), the design based on the pseudo-power operator, includes the following:

[0118]

[0119] Where w is the pseudo-power operator.

[0120] To more clearly express the execution intent of each functional unit, the instantaneous pseudo-power of the driver agent, steering agent, and drive / braking agent systems needs to be calculated for the operation input of the execution unit.

[0121] The instantaneous pseudo-power expression for driver-agent steering control is as follows:

[0122]

[0123] Among them, w h (t) represents the instantaneous pseudo-power of the driver's steering maneuver, T w For the periodic time, v y Let t be the lateral speed of the vehicle and t be the travel time.

[0124] The method for calculating the pseudo-power of driver agent driving control is as follows:

[0125]

[0126] Among them, w dd (t) represents the instantaneous pseudo-power of the driver's driving operation, v x The longitudinal speed of the vehicle;

[0127] The method for calculating the pseudo-power of driver-agent braking is as follows:

[0128]

[0129] Among them, wdb (t) represents the instantaneous pseudo-power of the driver's braking operation;

[0130] The method for calculating the instantaneous pseudo-power of the steering system agent is as follows:

[0131]

[0132] Among them, w a (t) represents the instantaneous pseudo-power of the vehicle's auxiliary steering system.

[0133] The method for calculating the instantaneous pseudo-power of the control / drive system agent is as follows:

[0134]

[0135] Among them, w ab (t) represents the instantaneous pseudo-power of the vehicle drive assistance system, w ad (t) represents the instantaneous pseudo-power of the vehicle braking assist system.

[0136] The instantaneous pseudo-powers of the driver system, steering system, and drive / brake system were determined through the above calculations, providing a basis for the execution intentions of the functional units. Figure 1 The consistency determination provides technical support.

[0137] Step (7): Functional Agent Behavior Intent Based on Fisher Algorithm Figure 1 Consistency determination.

[0138] See Figure 3 — Figure 4 It can be seen that, regarding the functional agent's behavioral intent Figure 1 Consistency determination includes the following steps:

[0139] (51) First, the functional agent's behavioral intentions are divided into steering intention, driving intention, and braking intention.

[0140] (52) The key parameter for determining the behavioral intent of the functional agent is: the turning intent parameter w. h ,w a Driver intent parameter w dd ,w ad Braking intention parameter w db ,w ab The evaluation is graded on a two-level scale (high and low).

[0141] (53) The experimental data collected in step (4) is preprocessed in step (5) to obtain 3 pairs of key parameters for determining behavioral intention. 40 sets of data are generated within the range of 2 level standards. 20 sets of data are randomly extracted from each level. The first 15 sets of data are used for model training and the last 5 sets of data are used for model testing.

[0142] (54) Based on the model data in (53), Fisher's decision theory is used to classify and determine the behavioral intentions of each functional agent:

[0143] The behavior of each pair of parameters in (53) Figure 1 Consistency levels are divided into two categories, defined as follows: and Where i = 1, 2, 3 represent steering intention, braking intention, and driving intention, respectively. Fisher's determination method is used to determine steering intention. Figure 1 Classify by consistency.

[0144] Given a steering behavior dataset y j The instantaneous power occurs at time j = 1, 2, ..., N, where N represents the total number of samples. This refers to the instantaneous power during steering.

[0145] Let T h Indicates intention to change direction Figure 1 High affinity level A set of samples of each category, whose mean vector is represented as... Let T l Indicates intention to change direction Figure 1 Low affinity level The sample set of each category is represented by the mean vector as follows:

[0146] Assume the classification line is:

[0147]

[0148] in, Indicates intention to change direction Figure 1 The projection vector of the consistency determination sample; Indicates intention to change direction Figure 1 Consistency determination sample dataset; Linear discriminant analysis (LDA) of the two classes of samples, such as... Figure 3 As shown.

[0149] Projecting the data onto a straight line, we can obtain the projections of the centers of the two types of samples onto the line as follows: and The variances of the two types of samples are as follows: and

[0150] To more accurately classify and determine agent behavioral intentions, it's necessary to make similar samples as close as possible and dissimilar samples as far apart as possible. This can be achieved by minimizing the variance of the projection points of similar samples. Minimize the size; this allows the projection point of the center of the outlier sample to be as far away as possible. To be as large as possible. Therefore, the objective of Linear Discriminant Analysis (LDA) is to maximize:

[0151]

[0152] Turning behavior intention Figure 1 The inter-class divergence of samples for consistency determination is:

[0153]

[0154] Its representation vector His own external product, that is, the turning behavior intention Figure 1 The inter-class divergence of samples used for consistency determination.

[0155] Turning behavior intention Figure 1 The within-class divergence of samples for consistency determination is:

[0156]

[0157] Where, Σ h To change behavior intention Figure 1 The covariance matrix (n×n) among features of high-consistency class examples, Σ l To change behavior intention Figure 1 The covariance matrix (n×n) among features of low-consistency class examples.

[0158] The objective of Linear Discriminant Analysis (LDA) is to maximize:

[0159]

[0160] Among them, J s For S b With S w The broad definition of Rayleigh.

[0161] The optimization problem to be solved is:

[0162]

[0163] To indicate the intention to change direction Figure 1 The optimal projection vector of the consistency determination sample;

[0164] Since both the numerator and denominator are about The quadratic term, therefore the solution is the same as... The length is irrelevant. Let The optimization problem can then be written as:

[0165]

[0166] Applying the Lagrange multiplier method:

[0167]

[0168] Where λ is a real number.

[0169] make in It is a real number.

[0170]

[0171] Combining equation (26) and equation (27), we can obtain:

[0172]

[0173] Due to the solution and The length is irrelevant, therefore let We can obtain:

[0174]

[0175] Repeat step (43) to obtain the generalized Rayleigh quotient J of the braking determination sample. b Generalized Rayleigh quotient J of driving samples d And two optimal solutions.

[0176] Based on the above theory, the key parameters for determining the behavioral intent of the functional agent are: the turning intent parameter w. h ,w a Driver intent parameter w dd ,w ad Braking intention parameter w db ,w ab By solving for the optimal solution using Fisher's decision theory, we can determine whether the steering intention, braking intention, and driving intention are consistent. The behavior of each functional agent unit can be determined if and only if all three intentions are consistent. Figure 1 To determine the intent of the behavior Figure 1 The consistency level is high; otherwise, the behavior is judged as intentional. Figure 1 The consistency level is low, at which point a functional agent unit needs to autonomously adjust until the behavioral intention is obtained. Figure 1 When the consistency is determined to be high, the mechanism is executed to complete the intelligent coordinated control of the distributed electric vehicle.

[0177] See Figure 4 As can be seen, this example illustrates the behavioral intent of a distributed electric vehicle functional agent. Figure 1 Design a consistency determination method, behavioral intention Figure 1The consistency determination process first involves functionally dividing the various electronic control units (ECUs) of the distributed electric vehicle. The vehicle dynamics model is then integrated with Agent language to describe the divided functional areas. Real-time data collection ensures the accuracy and timeliness of vehicle behavior intent recognition. A pseudo-power algorithm is used to intuitively describe the execution intent of each control unit, and the Fisher algorithm is used to analyze the execution intent of the functional units. Figure 1 Consistency is assessed. When the consistency level is high, the execution unit executes directly; when the consistency level is low, the functional agent autonomously adjusts until the consistency level is high, at which point the execution unit is activated and begins execution. This design effectively avoids misinterpretation of the functional unit's execution intent due to driver error or the inability to recognize mechanical operations.

[0178] The advantage of this example is:

[0179] The functional agent behavior employed in this method Figure 1 Consistency determination methods can effectively identify the execution intentions of different functional units, which is of great significance for the development of safety assistance systems that require adaptation to different driving styles. The consistency determination method based on pseudo-power algorithm and Fisher's determination criterion has low computational load, is easy to control in real time, and the determined behavioral intention structure is safer and more reliable than traditional methods. It is applicable to various driving conditions and can meet the needs of intelligent coordinated control of distributed electric vehicles.

[0180] The above description is merely a preferred embodiment of the present invention and does not limit the scope of the present invention. All equivalent structural changes made based on the description and drawings of the present invention are included within the scope of the present invention.

Claims

1. A method for determining the consistency of behavioral intent of a distributed electric vehicle functional agent, characterized in that, Includes the following steps: (1) For distributed drive electric vehicles, the functions of the electronic control unit are divided into driver function area, steering function area and drive / braking function area, and corresponding functional agent units are established based on the divided functional areas. (2) For different functional agent units, construct dynamic equations to characterize the operations performed by the functional agent units; (3) Using the longitudinal and lateral dynamic equations of distributed electric vehicles, a pseudo-power operator for the functional agent system is constructed and solved in real time; (4) Establish a linear classifier for functional agent consistency based on Fisher's linear discriminant algorithm to determine the consistency level of functional agent behavior intention judgment; Step (3) Utilize the longitudinal and lateral dynamic equations of the distributed electric vehicle to construct a pseudo-power operator for the functional agent system, and solve it in real time, including: (31) Collect test data, including driver hand torque, steering control torque, drive pedal travel, brake pedal travel, steering wheel angle, steering wheel speed, hub motor driving torque, hub motor driving torque, and vehicle speed; (32) The collected test data are divided into four categories: driver steering control data, driver braking control data, driver driving control data, steering system execution data, driving system execution data, and braking system execution data. These data are then input into four functional agent units for identification. The data in each category is segmented, with each time period representing the control behavior of the functional agent over a period of time. (33) Based on the pseudo-power judgment method, the instantaneous pseudo-power of the driver agent, steering agent, and drive / braking agent systems is calculated as follows: The instantaneous pseudo-power expression for driver-agent steering control is as follows: (5) in, For the instantaneous pseudo-power of the driver's steering maneuver, For periodic time, Let be the lateral speed of the vehicle. Runtime; The method for calculating the pseudo-power of driver agent driving control is as follows: (6) in, The instantaneous pseudo-power of the driver's driving operation. The longitudinal speed of the vehicle; The method for calculating the pseudo-power of driver-agent braking is as follows: (7) in, The instantaneous pseudo-power for driver braking control; The method for calculating the instantaneous pseudo-power of the steering system agent is as follows: (8) in, This refers to the instantaneous pseudo-power of the vehicle's auxiliary steering system. The method for calculating the instantaneous pseudo-power of the control / drive system agent is as follows: (9) in, For the instantaneous pseudo power of the vehicle drive assist system, This refers to the instantaneous pseudo-power of the vehicle's braking assist system. Step (4) Establish a linear classifier for functional agent consistency based on Fisher's linear discriminant algorithm, and determine the consistency level of functional agent behavior intent judgment, including: (41) The behavioral intentions of the functional agents are divided into steering intention, driving intention, and braking intention; the key parameters for determining the behavioral intentions of the functional agents are: steering intention parameters. Driver intent parameters Braking intention parameters The three pairs of behavioral intent judgment levels are divided into two levels: high behavioral intent consistency level and low behavioral consistency level. (42) Generate model data for model training and testing within two evaluation level standards (high and low) for the key parameters of the selected 3 pairs of functional agents to determine their behavioral intentions. (43) Based on the model data in step (42), Fisher's decision theory is used to classify and determine the behavioral intentions of each functional agent. Only when the steering intention, driving intention, and braking intention are consistent are the behavioral intentions of each functional agent unit consistent, and the behavioral intention consistency level is determined to be high; otherwise, the behavioral intention consistency level is determined to be low. When the behavioral intention consistency is determined to be high, the mechanism is executed to complete the intelligent coordination control of the distributed electric vehicle.

2. The method for determining the consistency of behavioral intent of a distributed electric vehicle functional agent according to claim 1, characterized in that: Step (2) For different functional agent units, construct dynamic equations characterizing the operations performed by the functional agent units, including: (21) Construct the dynamic equations representing the driver agent unit: The driver's behavioral intention is determined by the control input torque they exert on the distributed electric vehicle. This indicates that it includes steering torque input. Drive torque input Braking torque input ; The dynamic expression for the driver's hand torque input: (1) in, For steering torque input, For the equivalent inertia of the steering system, For steering system damping, For the friction torque of the steering system, The steering assist torque provided to the steering system motor, The value is the steering wheel angle; sign() is the sign function used to indicate the positive or negative sign of the parameter. The formula for expressing the driving / braking torque is: (2) in, To optimize the braking torque coefficient, Optimize the driving torque coefficient; For actual accelerator pedal opening With the target accelerator pedal opening The difference, For actual brake pedal opening relative to the target brake pedal opening The difference, where and The formula for expressing it is: (22) Construct the dynamic equations of the steering function agent unit: The dynamic equation of the steering agent unit is expressed as follows: (3) in, For steering system control output, This is the actual current of the steering motor. This refers to the torque coefficient of the steering motor. For mechanical efficiency; (23) Construct the dynamic equations of the driving / braking agent unit: The dynamic equation expression for the drive / braking agent unit is as follows: (4) in, This indicates the total output torque of the hub motor; The target output torque for the motor; This represents the maximum output torque at the current motor speed. The time constant is obtained from the torque response data of the drive motor.

3. The method for determining the consistency of behavioral intent of a distributed electric vehicle functional agent according to claim 2, characterized in that, The behavioral intent consistency level of each pair of parameters in step (41) is divided into two categories, defined as follows: and ,in, These represent the intentions of steering, braking, and driving, respectively.

4. The method for determining the consistency of behavioral intent of a distributed electric vehicle functional agent according to claim 3, characterized in that, Step (43) uses the Fisher determination method to classify the consistency of turning behavior intentions, including: Given a steering behavior dataset , For instantaneous power during steering, At the moment when instantaneous power is present, , Indicates the total number of samples; set up This indicates a high level of consistency between the intention to change course and the actual behavior. The set of examples of a category, the sample set The mean vector is represented as ;set up This indicates a low level of consistency between the intention and the behavioral shift. The sample set of the category, the sample set The mean vector is represented as ; Assume the classification line is: (10) in, This represents the projection vector of the sample used to determine the consistency of the turning behavior intention; This represents a sample dataset for determining consistency between the intention and the turning behavior. The inter-class divergence of the samples used to determine the consistency of the turning behavior intention is: (11) Representing vectors The product of itself and its own, which is the inter-class divergence of the sample class for determining the consistency of the turning behavior intention; The within-class divergence of the consistency determination of turning behavior intention is: （12） in, The intra-class divergence of samples used to determine consistency in steering behavior; The objective function of linear discriminant analysis is to maximize: （13） in, For the divergence between sample classes Within-class divergence of samples The broad definition of Rayleigh.

5. The method for determining the consistency of behavioral intent of a distributed electric vehicle functional agent according to claim 4, characterized in that, Given a braking determination sample dataset and a driving sample dataset, the generalized Rayleigh quotient of the braking determination samples is obtained according to the method steps in step (43). and the generalized Rayleigh quotient of driving samples .

Images