Human driver style online identification method

Abstract

The invention discloses a human driver style on-line identification method, which comprises the following steps of: 1, constructing a human driving behavior on-line identification model by adopting a modeling thought of inverse optimal control; step 2, simplifying the human driving behavior online identification model based on the Pontryagin minimum principle; 3, converting the model into a linear least square problem based on a model simplification result; and step 4, on the basis of model conversion, performing model solution by using a Lagrange multiplier method to obtain driving behavior parameters describing the style preference of the human driver, and further identifying the driving style of the human driver. According to the method, the driving style of the human driver of the environment vehicle can be accurately identified, so that the response performance and adaptability of an automatic driving system are effectively improved, and the driving safety and comfort are improved.

Description

Technical Field

[0001] The invention relates to a technical method for analyzing human driving behavior, and in particular to an online method for identifying the style of a human driver. Background Art

[0002] In the current research and update of autonomous driving systems, it is of great value to accurately identify the driving style of human drivers in environmental vehicles. By deeply identifying the driving styles of drivers, the response performance and adaptability of the autonomous driving system can be effectively improved, and driving safety and comfort can be enhanced.

[0003] At present, the analysis and modeling of human driving behavior are mainly carried out by mathematical modeling or machine learning methods. However, traditional mathematical modeling methods are usually based on many assumptions, so they have a narrow scope of application and weak generalization ability; while machine learning-based methods require massive data to drive, so the data demand is large and the interpretability is poor. In addition, the above methods all ignore the different driving styles of drivers and are unable to take individualized countermeasures to actively protect the right of way. There are also a few studies that consider the heterogeneous driving styles of drivers, but they mainly use offline data to refine driving styles, which is difficult to adapt to complex and changing driving scenarios.

[0004] Therefore, based on the above analysis, there is an urgent need for an online human driver style identification method that can adapt to highly complex and dynamic driving environments. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide an online human driver style identification method that can accurately identify the driving style of human drivers of environmental vehicles, effectively improve the response performance and adaptability of the automatic driving system, and improve driving safety and comfort. Based on the inverse optimal control theory, this method analyzes and characterizes the driving behavior characteristics of human drivers, and analyzes the multi-objective coupling mechanism of human driving behavior.

[0006] Technical solution: The method for online identification of human driver style of the present invention comprises the following steps:

[0007] Step 1: Build an online recognition model for human driving behavior;

[0008] Step 2: Based on the Pontryagin minimum principle, simplify the online recognition model of human driving behavior:

[0009] Step 3: Based on the model simplification results, transform the model into a linear least squares problem;

[0010] Step 4: Based on the model transformation, the Lagrange multiplier method is used to solve the model and obtain the driving behavior parameters that describe the style preferences of human drivers.

[0011] Preferably, in step 1, for building an online recognition model of human driving behavior:

[0012] Using the modeling idea of ​​inverse optimal control and based on the Pontryagin minimum principle, the Hamiltonian equation is listed:

[0013]

[0014] Among them, x n -System status;

[0015] a CV，n -Human-driven vehicle acceleration;

[0016] β-driving behavior parameters;

[0017] β T - Transpose of driving behavior parameter β;

[0018]

[0019] u EV，n -Control quantity of the main vehicle;

[0020] u CV，n -Amount of human-driven vehicle control;

[0021] A n , B n , C n -System state equation parameters;

[0022] - co-variables;

[0023] Based on the Hamiltonian equation, the necessary conditions that the optimal control input in the PMP method should satisfy are listed, which are the regular equation and the extreme value condition:

[0024]

[0025] in,

[0026] γ n - co-variables;

[0027] - Calculate the gradient of the Hamiltonian function using the system state as a variable;

[0028] H n -Hamiltonian function;

[0029] x n -System status;

[0030] a CV，n -Human-driven vehicle acceleration;

[0031] β-driving behavior parameters;

[0032] -Calculate the gradient of the Hamiltonian function with acceleration as the variable;

[0033] Preferably, the online identification model of human driving behavior in step 2 is simplified: First, the Hamiltonian equation is substituted into the canonical equation to obtain:

[0034]

[0035] Note n =(β,γ n ) T , we can get the following recursive equation:

[0036] z n+1 =K n z n

[0037] in,

[0038]

[0039] The derived recursive equation is reversed and recursively converted to obtain the following equation:

[0040] z n+1 =K n K n-1 K n-2 ...K0z0=K n z0

[0041] z0=(β,γ0) T

[0042] in,

[0043] I - unit vector;

[0044] 0 - zero matrix;

[0045] γ0-initial costate vector;

[0046] β-driving behavior parameters;

[0047] A n -System state equation parameters;

[0048] x n -System status;

[0049]

[0050] - Calculate the gradient of the Hamiltonian function using the system state as a variable;

[0051] Similarly, substituting the Hamiltonian equation into the extreme value condition, we can obtain:

[0052]

[0053] remember Then we can get the following equation:

[0054] L n z n+1 =0

[0055] in,

[0056] -Find the gradient of the Hamiltonian function with acceleration as the variable.

[0057]

[0058] Substituting the equation derived by reverse recursion into the above equation, we can get the following equation:

[0059] L n K n z0=0

[0060] Preferably, the simplified model for online identification of human driving behavior in step 3 is transformed:

[0061] Considering the noise of the collected data and the error of the model, it is actually difficult to solve a set of β according to the last equation obtained in step 2. Therefore, the equation obtained in step 2 is enlarged and converted into a linear least squares problem.

[0062]

[0063] stez0=β1=a

[0064] in,

[0065] remember

[0066] P n -semi-positive definite matrix;

[0067]

[0068] a (a>0) - magnification factor, so that the value of the first variable of z0 is equal to a.

[0069] β1 - the first term of driving behavior parameters;

[0070]

[0071] K=K n K n-1 Kn-2 ...K0;

[0072]

[0073] Preferably, after the simplified model of online identification of human driving behavior in step 4 is transformed, the following solution is obtained:

[0074] For the above least squares problem, the Lagrange multiplier method is used to solve it. The Lagrange equation is:

[0075] 2P n z0+pe T =0

[0076] in,

[0077] p is the Lagrange multiplier, P n - As defined above, z0-same as above, z0=(β,γ0) T , e T - As defined above, The transpose of

[0078] To express z0 more clearly, let the matrix P k Block, get the following formula:

[0079]

[0080] in,

[0081] -Matrix P k The element in the first row and first column;

[0082] q k - Remove elements The matrix P k The first column of

[0083] -q k The transpose of

[0084] -Matrix P k submatrix of ;

[0085] -The vector z0 minus the element a.

[0086] Compared with the prior art, the present invention has the following beneficial effects:

[0087] 1. Traditional methods for analyzing and modeling human driving behavior mainly use mathematical modeling or machine learning methods, which often ignore the different driving styles of drivers and fail to adopt driving strategies that vary from person to person. However, the online human driver style identification algorithm proposed in the present invention takes into account the heterogeneous driving styles of drivers and uses online data to refine driving styles, which can adapt to complex and changeable driving scenarios.

[0088] 2. The online human driver style identification method proposed in this invention deconstructs the multi-objective coupling mechanism of human driving behavior, allowing the autonomous driving vehicle to have the ability to identify the driving style and behavior of human drivers and rival vehicles in real time. Therefore, the autonomous driving vehicle can adopt driving strategies that vary from person to person, thereby improving driving efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 Flowchart of the algorithm for online identification of human driver style;

[0090] Figure 2 Diagram of the modeling ideas for inverse optimal control. DETAILED DESCRIPTION

[0091] The following will be described in more detail with reference to the schematic diagram of the human driver style online identification method of the present invention, wherein the preferred embodiment of the present invention is shown, and it should be understood that those skilled in the art can modify the present invention described herein, and still achieve the advantageous effects of the present invention. Therefore, the following description should be understood as being widely known to those skilled in the art, and not as a limitation of the present invention.

[0092] like Figures 1-2 The method for online identification of human driver style in this embodiment specifically includes the following steps:

[0093] Step 1: Build an online recognition model for human driving behavior.

[0094] Using the modeling idea of ​​inverse optimal control and based on the Pontryagin minimum principle, the Hamiltonian equation is listed:

[0095]

[0096] Among them, x n -System status;

[0097] a CV，n -Human-driven vehicle acceleration;

[0098] β-driving behavior parameters;

[0099] β T - Transpose of driving behavior parameter β;

[0100]

[0101] u EV，n -Control quantity of the main vehicle;

[0102] u CV，n -Amount of human-driven vehicle control;

[0103] A n , B n , C n -System dynamic parameters.

[0104] Based on the Hamiltonian equation, the necessary conditions that the optimal control input in the PMP method should satisfy are listed, which are the regular equation and the extreme value condition:

[0105]

[0106] Among them, λ n - a co-state vector, the number of parameters in the co-state vector is equal to the number of variables in the system model;

[0107] - Calculate the gradient of the Hamiltonian function using the system state as a variable;

[0108] H n -Hamiltonian function;

[0109] x n -System status;

[0110] a CV，n -Human-driven vehicle acceleration;

[0111] β-driving behavior parameters;

[0112] -Find the gradient of the Hamiltonian function with acceleration as the variable.

[0113] Step 2: Based on the Pontryagin minimum principle, simplify the online recognition model of human driving behavior.

[0114] First, substitute the Hamiltonian equation into the canonical equation to obtain:

[0115]

[0116] Note n =(β,γ n ) T , we can get the following recursive equation:

[0117] z n+1 =K n z n

[0118] in,

[0119]

[0120] The derived recursive equation is reversed and recursively converted to obtain the following equation:

[0121] z n+1 =K n K n-1 K n-2 ...K0z0=K n z0

[0122] z0=(β,γ0) T

[0123] Where, I-unit vector;

[0124] 0 - zero matrix;

[0125] γ0-initial costate vector.

[0126] β-driving behavior parameters;

[0127] A n -System state equation parameters;

[0128] x n -System status;

[0129]

[0130] - Calculate the gradient of the Hamiltonian function using the system state as a variable;

[0131] Similarly, substituting the Hamiltonian equation into the extreme value condition, we can obtain:

[0132]

[0133] remember Then we can get the following equation:

[0134] L n z n+1 =0

[0135] in,

[0136] -Calculate the gradient of the Hamiltonian function with acceleration as the variable;

[0137]

[0138] Substituting the equation derived by reverse recursion into the above equation, we can get the following equation:

[0139] L n K n z0=0

[0140] Step 3: Based on the model simplification results, transform the model into a linear least squares problem.

[0141] Considering the noise of the collected data and the error of the model, it is actually difficult to solve a set of β according to the last equation obtained in step 2. Therefore, the equation obtained in step 2 is enlarged and converted into a linear least squares problem.

[0142]

[0143] stez0=β1=a

[0144] Among them, P n -semi-positive definite matrix;

[0145]

[0146] a (a>0) - magnification, so that the value of the first variable of z0 is equal to a;

[0147] β1 - the first term of driving behavior parameters;

[0148]

[0149] K=K n K n-1 K n-2 ...K0;

[0150]

[0151] Step 4: Based on the model transformation, the Lagrange multiplier method is used to solve the model and obtain the driving behavior parameters that describe the style preferences of human drivers.

[0152] For the above least squares problem, the Lagrange multiplier method is used to solve it. The Lagrange equation is:

[0153] 2P n z0+pe T =0

[0154] Where p is the Lagrange multiplier.

[0155] To express z0 more clearly, let the matrix P k Block, get the following formula:

[0156]

[0157] in, -Matrix P k The element in the first row and first column;

[0158] q k- Remove elements The matrix P k The first column of

[0159] -Matrix P k submatrix of ;

[0160] -The vector z0 minus the element a.

[0161] The goal of the present invention for parameter identification is to find the parameter β that can make the optimal solution meet the necessary conditions by obtaining the optimal trajectory data online. The modeling idea is as follows: Figure 2 shown.

[0162] Figure 2 The framework for solving human driver style parameters based on trajectory data is presented, including forward and reverse optimization control analysis. Both are centered on the optimal control condition and form a closed-loop feedback. The forward analysis calculates driving behavior parameters through trajectory data and the optimal control algorithm, while the reverse analysis adopted by the present invention derives the parameter β that enables the optimal solution to meet the necessary conditions through trajectory data and the reverse optimization algorithm.

[0163] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any technician in the relevant technical field, without departing from the scope of the technical solution of the present invention, makes any form of equivalent replacement or modification to the technical solution and technical content disclosed in the present invention, which does not depart from the content of the technical solution of the present invention and still falls within the protection scope of the present invention.

Claims

1. A method for online identification of human driver style, characterized in that: The following steps are involved: Step 1: Use the inverse optimal control modeling approach to build an online recognition model for human driving behavior; Step 2: Based on the Pontryagin minimum principle, the online identification model of human driving behavior is simplified by combining the canonical equation and the extreme value condition; Step 3: Convert the online identification model of human driving behavior into a linear least squares problem; Step 4: Based on the model transformation, the Lagrange multiplier method is used to solve the model and obtain the driving behavior parameters that describe the style preferences of human drivers.

2. The method for online identification of human driver style according to claim 1, characterized in that: The method of step 1: adopt the modeling idea of ​​inverse optimal control, based on the Pontryagin minimum principle, list the Hamiltonian equation: in, x n -System status; a CV，n -Human-driven vehicle acceleration; β-driving behavior parameters; β T - Transpose of driving behavior parameter β L n - u EV，n -Main vehicle control quantity; u CV，n -Amount of human-driven vehicle control; A n , B n , C n -System state equation parameters; - co-variables; Based on the Hamiltonian equation, the necessary conditions that the optimal control input in the PMP method should satisfy are listed, which are the regular equation and the extreme value condition: in, γ n - co-variables; - Calculate the gradient of the Hamiltonian function using the system state as a variable; H n -Hamiltonian function; x n -System status; a CV，n -Human-driven vehicle acceleration; β-driving behavior parameters; -Find the gradient of the Hamiltonian function with acceleration as the variable.

3. The method for online identification of human driver style according to claim 2, characterized in that: The method for simplifying the online recognition model of human driving behavior in step 2 is as follows: First, substitute the Hamiltonian equation into the canonical equation to obtain: Remember n =(β,γ n ) T , we can get the following recursive equation: With n+1 =K n With n in, The derived recursive equation is reversed and recursively converted to obtain the following equation: from n+1 =K n TO n-1 TO n-2 ...K0z0=K n z0 z0=(β,γ0) T in, I - unit vector; 0 - zero matrix; γ0-initial costate vector; β-driving behavior parameters; A n -System state equation parameters; x n -System status; L n - - Calculate the gradient of the Hamiltonian function using the system state as a variable; Similarly, substituting the Hamiltonian equation into the extreme value condition, we can obtain: remember Then we can get the following equation: L n z n+1 =0 in, -Calculate the gradient of the Hamiltonian function with acceleration as the variable; With n+1 -With n+1 =(β,γ n+1 ) T ; Substituting the equation derived by reverse recursion into the above equation, we can get the following equation: L n K n z0=0。 4. The method for online identification of human driver style according to claim 3, characterized in that: The conversion method in step 3: Considering the noise of the collected data and the error of the model, it is actually difficult to solve a set of β according to the equation finally obtained in step 2. Therefore, the equation obtained in step 2 is amplified and converted into a linear least squares problem. in, remember P n -Positive semidefinite matrix a (a>0) - magnification, so that the value of the first variable of z0 is equal to a, β1 - the first term of driving behavior parameters L n - K=K n K n-1 K n-2 …K0 z0-z0=(β,γ0) T 。 5. The method for online identification of human driver style according to claim 4, characterized in that: After the simplified model of online identification of human driving behavior in step 4 is transformed, the solution is obtained: the Lagrange multiplier method is used to solve it, and the Lagrange equation is: 2P n z0+on T =0 in, p is the Lagrange multiplier, P n - As defined above, z0-same as above, z0=(β,γ0) T , e T - As defined above, The transpose of To express z0 more clearly, let the matrix P k Block, get the following formula: in, -Matrix P k The element in the first row and first column; q k - Remove elements The matrix P k The first column of -q k The transpose of -Matrix P k submatrix of ; -The vector z0 minus the element a.

Images