A control method for intelligent passenger car to track a predetermined trajectory

Abstract

The invention discloses a control method for an intelligent bus to track a predetermined trajectory, comprising: establishing a local coordinate system of the intelligent bus and a global coordinate system of the earth; obtaining current posture information of the intelligent bus and posture information of a preview point; calculating the curvature of a road; establishing a Markov decision process (MDP) model of a path tracking control problem of the intelligent bus; deriving online learning rules of an actuator and an evaluator in an MKDHP algorithm; judging whether the preview point is the coordinate of an end point in the predetermined trajectory, and controlling the intelligent bus to decelerate to the end point; the MKDHP algorithm proposed in the invention has good generalization and self-optimization capabilities, and is therefore conducive to a path tracking controller based on the MKDHP algorithm to obtain higher control accuracy under different vehicle speeds and predetermined trajectory shape conditions, thereby improving the autonomous optimization capability and road adaptability of the intelligent bus.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent buses, and in particular relates to a control method for an intelligent bus to track a predetermined trajectory. Background Art

[0002] Smart buses can improve traffic safety and transportation efficiency. At the same time, the development of smart buses has promoted technological innovation in related fields such as artificial intelligence, sensors, and communication technology. Path tracking control is one of the key technologies for realizing smart buses. How to improve the adaptability of smart buses to complex road conditions is one of the current research hotspots.

[0003] Traditional control methods for intelligent buses to track predetermined trajectories, such as pure tracking, Stanley, and PID, can achieve the function of tracking predetermined trajectories, but the tracking accuracy is low under high speed and complex path conditions. Some scholars have proposed optimization control methods such as sliding mode control and model prediction based on the mathematical model of buses. Although they improve the tracking accuracy, they require high accuracy of the model and have high computational complexity, especially under complex road conditions, the control parameters need to be readjusted.

[0004] In order to improve the adaptability of smart buses to complex road conditions, some scholars have proposed intelligent path tracking control methods, such as neural network and fuzzy control, by combining traditional control theory with advanced artificial intelligence technology. The control method based on neural network takes a long time to train and requires a large amount of training data. The network model trained is greatly affected by the data and has poor interpretability. The control method based on fuzzy logic relies on expert experience to design fuzzy rules, which is difficult to adjust adaptively. Summary of the invention

[0005] In order to solve at least one of the problems raised in the above background technology, the present invention provides a control method for an intelligent bus to track a predetermined trajectory, so as to achieve higher control accuracy under different shape trajectories and speeds without repeatedly adjusting control parameters, thereby improving the autonomous optimization capability and adaptability of the intelligent bus under complex road conditions.

[0006] To achieve the above object, the present invention provides the following technical solution: a control method for an intelligent bus to track a predetermined trajectory, comprising the following steps:

[0007] S1: Establish a local coordinate system with the main antenna of the GPS sensor installed on the smart bus as the coordinate origin; establish a global coordinate system with the center of the earth as the origin;

[0008] S2: After driving the smart bus for a certain period of time, record n ​​lines of longitude, latitude, heading angle and speed information as a predetermined track;

[0009] S3: obtaining a kernel dictionary based on the approximate linear correlation analysis method for the collected smart bus state samples;

[0010] S4: Determine the current time The position information of the smart bus and the position information at the preview point are obtained, and the trajectory tracking error state equation in the local coordinate system is calculated, where i is a positive integer;

[0011] S5: Calculate the current time The curvature of the road ;

[0012] S6: Establish the MDP model of intelligent bus path tracking control problem;

[0013] S7: Design a controller for tracking the predetermined trajectory based on the MKDHP algorithm. The controller includes an actuator module and an evaluator module. The actuator module outputs the current moment The following control actions The evaluator module includes a first evaluator module and a second evaluator module; each of which outputs the current moment Next and next moment The derivative of the state value function with respect to the state and ;

[0014] S8: Establish online update rules for the actuator module and the evaluator module in the MKDHP algorithm;

[0015] S9: Determine the current moment Check whether the target point in the predetermined trajectory closest to the smart bus is the end point. If so, the smart bus stops driving; otherwise, assign i+1 to i and return to S4 for sequential execution.

[0016] The construction method of the core dictionary in S3 is as follows,

[0017] Step 3.1, select formula (1) as the kernel function:

[0018] (1);

[0019] In formula (1), and is the state vector in the sample, is the kernel width of the Gaussian kernel function, and are the parameters of the polynomial kernel function;

[0020] Step 3.2: Sample set of smart bus status Based on the approximate linear correlation analysis method, it has been obtained feature vectors, is the obtained kernel dictionary, and determines whether to add the new feature vector Add to the kernel dictionary, first use equations (2) and (3) to calculate :

[0021] ;

[0022] In formula (2) , is an element in the kernel dictionary, then for:

[0023] ;

[0024] Then calculate the With predefined threshold Compare to update the kernel dictionary, if ,Will Added to the nuclear dictionary, i.e. ; otherwise there is no change to the nuclear dictionary.

[0025] The method for constructing the trajectory tracking error state equation in the local coordinate system in S4 is as follows:

[0026] Step 4.1: Define the current moment The position coordinates of the rear axle center of the next smart bus in the global coordinate system are: , the speed is , the heading angle is , denoted as ;

[0027] Step 4.2, Settings is the preview distance. In the predetermined trajectory, search for a target point that is greater than the preview distance L and closest to the smart bus. This target point is defined as the preview point. , denoted as ;

[0028] Step 4.3: Define the smart bus at the current moment Relative to the preview point The error state equation is:

[0029] ;

[0030] In formula (4), The current moment Lower longitudinal deviation, lateral deviation and heading angle deviation;

[0031] Step 4.4: According to formula (5), the distance between the smart bus and the preview in the local coordinate system is obtained: Preview point The rate of change of the lateral deviation and heading angle deviation is:

[0032] ;

[0033] In formula (5), and At the current moment The longitudinal and lateral speeds of the rear axle center point of the smart bus, The wheelbase of the smart bus. For the current moment The front wheel turning angle of the smart bus, Preview point The curvature of the road.

[0034] S5 Medium Road Curvature The calculation method is as follows,

[0035] Step 5.1: At the current moment Next, select the distance from the predefined trajectory to the smart bus The nearest target point , calculated using formula (6) To preview point Distance ,like , then a quadratic polynomial fit is used to Otherwise, search for a trajectory greater than the distance in the predetermined trajectory. And the target point closest to the smart bus , using a quadratic polynomial fit to The trajectory between

[0036] ;

[0037] In formula (6) is the number of target points of the fitted trajectory;

[0038] Step 5.2: Use a quadratic polynomial to fit the above trajectory. The fitted quadratic polynomial equation is shown in formula (7):

[0039] ;

[0040] In formula (7) is the constant term coefficient, is the coefficient of the first-order term, is the coefficient of the quadratic term;

[0041] The road curvature .

[0042] The method of establishing the MDP model in S6 is as follows:

[0043] Step 6.1: Define the smart bus at the current moment The state vector under , defines the smart bus at the current moment The control action under is the compensation amount of the expected driving curvature of the smart bus ;

[0044] Step 6.2: Calculate the current time using formula (8) Lower state vector The Return :

[0045] ;

[0046] In formula (8), is a semi-positive definite diagonal matrix associated with the state dimension, is a positive constant;

[0047] Step 6.3: Define the current moment The value of the next state function , the calculation formula is as follows:

[0048] ;

[0049] In formula (9), For the current moment The state vector of the smart bus under For the current moment Next control action, For the current moment The immediate return value under is the discount factor, For the next moment The objective function value under For the next moment The state vector of the smart bus under , is calculated by formula (10):

[0050] ;

[0051] In formula (10), To control the cycle, ;

[0052] Step 6.4: Calculate the current time using formula (11) The optimal control action :

[0053] (11) .

[0054] The controller design method for tracking the predetermined trajectory in S7 is as follows,

[0055] Use formula (12) to calculate the actuator output in the MKDHP algorithm And as a smart bus at the moment The following control actions:

[0056] ;

[0057] In formula (12), is the maximum lateral acceleration, ; are the first weight and bias from the input layer to the hidden layer, ; are the second weight and bias from the hidden layer to the output layer, ;

[0058] Use formula (13) to calculate the output of the evaluator in the MKDHP algorithm And as a smart bus at the moment The following goals:

[0059] ;

[0060] In formula (13), For the current moment The weight matrix of the lower evaluator, For samples in the nuclear dictionary The basis functions formed, , is the number of samples in the kernel dictionary.

[0061] The online update rules of the actuator module and the evaluator module in S8 are as follows:

[0062] Step 8.1: Online update rules for actuator modules:

[0063] Minimize the error function established using equation (14) Learning objectives for the executor module:

[0064] ;

[0065] Update the current time using equations (15)-(18) Weight of the lower actuator module and bias , so as to obtain the next moment The weight under and bias :

[0066] ;

[0067] ;

[0068] ;

[0069] ;

[0070] In formula (15) - formula (18) is the learning step size of the evaluator module, ;

[0071] Step 8.2, online update rules of the evaluator module:

[0072] The error function is defined using formula (19): for:

[0073] ;

[0074] In formula (19), Represented as a vector Middle elements, Represented as a vector Middle elements, It can be obtained from formula (20):

[0075] ;

[0076] For status , assuming is its dimension, then , using formula (21) to define the intermediate variable :

[0077] ;

[0078] In formula (21) is a diagonal matrix Middle The diagonal element values, Indicates status Middle elements, Represents the weight matrix Middle The row weight vector is then used to obtain the error function: No. Row vector for:

[0079] (twenty two);

[0080] Defining intermediate variables , use equation (23)-(25) to update the current time The weight of the lower evaluator module , so as to obtain the next moment The weight under :

[0081] ;

[0082] (twenty four);

[0083] ;

[0084] In formula (23) to formula (25), is the learning step size of the evaluator module, For the forgetting factor, .

[0085] Compared with the prior art, the invention has the following beneficial effects: 1. The invention firstly establishes the MDP model of the intelligent bus tracking the predetermined trajectory control problem, selects the lateral deviation , heading angle deviation , road curvature and the current vehicle speed v as the state vector, and the reward function is designed by integrating the error performance index, safety and stability. Then, a control method for tracking the predetermined trajectory based on the MKDHP algorithm is proposed, in which the actuator is constructed by a three-layer feedforward neural network, and the evaluator is constructed by the basis function obtained based on the sparse kernel method. The weight update rules of the actuator and the evaluator are derived, which solves the problem that the traditional control method of intelligent buses needs to repeatedly adjust the control parameters under complex road conditions and lacks adaptability, and achieves higher control accuracy under different shape trajectories and speeds, and improves the autonomous optimization ability and adaptability of intelligent buses under complex road conditions.

[0086] 2. The present invention adopts a hybrid kernel function combining Gaussian kernel function and polynomial kernel function to construct basis function, which solves the problem of difficult feature selection and low learning efficiency of traditional approximate dynamic programming algorithm, and improves the generalization ability and real-time optimization ability of the path tracking controller based on MKDHP algorithm;

[0087] 3. The present invention only needs the coordinates of the target point relative to the smart bus in the process of tracking the predetermined trajectory. Therefore, it is not limited to GPS as a sensor, and other positioning means such as optical, radar and other sensors can be used to obtain the coordinates of the target point relative to the smart bus. Therefore, the method of the present invention has good versatility and universality, and is suitable for smart buses with different positioning means;

[0088] 4. The control method for the intelligent bus to track the predetermined trajectory of the present invention comprehensively considers the error performance index, safety and stability, and can realize the intelligent bus to track the predetermined trajectory with high accuracy while improving the driving safety and the comfort of the passengers;

[0089] 5. Since the present invention adopts a three-layer feedforward neural network to construct the actuator in the MKDHP algorithm, and constructs an evaluator based on the basis function obtained by the ALD analysis method, thereby constructing a path tracking controller framework based on the MKDHP algorithm, it not only has high reliability and robustness, but also is simple to implement. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] FIG1 is a structural diagram of a control system for an intelligent bus tracking a predetermined trajectory according to the present invention;

[0091] FIG2 is a schematic diagram of the path tracking of the intelligent bus according to the present invention;

[0092] FIG3 is a schematic diagram of the structure of the intelligent bus path tracking controller of the present invention;

[0093] FIG4 is a diagram showing the effect of tracking a double-twisted road at different vehicle speeds based on the MKHDP algorithm and other algorithms of the present invention;

[0094] FIG5 is a diagram showing the effect of tracking urban roads at different vehicle speeds based on the MKHDP algorithm and other algorithms of the present invention; DETAILED DESCRIPTION

[0095] This embodiment takes a certain manufacturer's intelligent bus platform as an example to specifically illustrate how to use the method of the present invention to achieve tracking of a predetermined trajectory and its functional effects.

[0096] The intelligent bus platform used in this embodiment has a length of 5.99m, a width of 2.07m, and a wheelbase of 2.95m. It is equipped with a combined inertial navigation system named CGI-610 provided by Huace Company, which can obtain the longitude and latitude, speed and heading angle information of the intelligent bus, and the positioning accuracy can reach 0.02m. The computing platform uses an industrial computer named EPC-R4710 provided by Advantech Technology Company to implement a path tracking controller based on the MKDHP algorithm.

[0097] In this embodiment, the Figure 1 The control system structure diagram of the intelligent bus tracking the predetermined trajectory shown in the figure is a control method for the intelligent bus tracking the predetermined trajectory, which uses a computer control method with equal sampling cycles to achieve tracking of the predetermined trajectory. Each control cycle is carried out according to the following steps:

[0098] S1: Establish a local coordinate system with the main antenna of the GPS sensor installed on the smart bus as the coordinate origin, the forward direction of the smart bus as the x-axis, and the direction of the forward direction of the smart bus rotated 90 degrees counterclockwise as the y-axis; establish a global coordinate system with the center of the earth as the origin, the due east direction as the x-axis, and the due north direction as the y-axis;

[0099] S2: Drive the bus along a certain trajectory, and store the posture and speed information in the geodetic coordinate system as a predetermined trajectory, wherein the posture information includes the longitude and heading angle and the position coordinates; the predetermined trajectory is a sequence of n target points, and the target point sequence is described by an array of dimension , and the four elements of each row represent the longitude, latitude, heading angle and speed information in the trajectory respectively.

[0100] S3: obtaining a core dictionary based on the ALD analysis method for the collected smart bus state samples, where the samples are the smart bus states recorded when tracking a predetermined trajectory using other path tracking control methods;

[0101] The construction method of the core dictionary in S3 is as follows,

[0102] Step 3.1, select formula (1) as the kernel function:

[0103] (1);

[0104] In formula (1), and is the state vector in the sample, is the kernel width of the Gaussian kernel function, and are the parameters of the polynomial kernel function;

[0105] Step 3.2: Sample set of smart bus status Based on the approximate linear correlation analysis method, it has been obtained feature vectors, is the obtained kernel dictionary, and determines whether to add the new feature vector Add to the kernel dictionary, first use equations (2) and (3) to calculate :

[0106] ;

[0107] In formula (2) , is an element in the kernel dictionary, then for:

[0108] ;

[0109] Then calculate the With predefined threshold Compare to update the kernel dictionary, if ,Will Added to the nuclear dictionary, i.e. ; otherwise there is no change to the nuclear dictionary.

[0110] S4: Determine the current time The position and posture information of the intelligent bus and the position and posture information at the preview point are obtained, so as to establish the trajectory tracking error state equation in the local coordinate system;

[0111] The method for constructing the trajectory tracking error state equation in the local coordinate system in S4 is as follows:

[0112] Step 4.1: Figure 2 As shown in The position coordinates of the rear axle center of the next smart bus in the global coordinate system are: , the speed is , the heading angle is , denoted as ;

[0113] Step 4.2, Settings is the preview distance. In the predetermined trajectory, search for a target point that is greater than the preview distance L and closest to the smart bus. This target point is defined as the preview point. , denoted as ;

[0114] Step 4.3: Define the smart bus at the current moment Relative to the preview point The error state equation is:

[0115] ;

[0116] In formula (4), The current moment Lower longitudinal deviation, lateral deviation and heading angle deviation;

[0117] Step 4.4: According to formula (5), the distance between the smart bus and the preview in the local coordinate system is obtained: Preview point The rate of change of the lateral deviation and heading angle deviation is:

[0118] ;

[0119] In formula (5), and At the current moment The longitudinal and lateral speeds of the rear axle center point of the smart bus, The wheelbase of the smart bus. For the current moment The front wheel turning angle of the smart bus, Preview point The curvature of the road.

[0120] S5 Medium Road Curvature The calculation method is as follows,

[0121] Step 5.1: At the current moment Next, select the distance from the predefined trajectory to the smart bus The nearest target point , calculated using formula (6) To preview point Distance ,like , then a quadratic polynomial fit is used to Otherwise, search for a trajectory greater than the distance in the predetermined trajectory. And the target point closest to the smart bus , using a quadratic polynomial fit to The trajectory between

[0122] ;

[0123] In formula (6) is the number of target points of the fitted trajectory;

[0124] Step 5.2: Use a quadratic polynomial to fit the above trajectory. The fitted quadratic polynomial equation is shown in formula (7):

[0125] ;

[0126] In formula (7) is the constant term coefficient, is the coefficient of the first-order term, is the coefficient of the quadratic term;

[0127] The road curvature .

[0128] S6: Establish the MDP model of intelligent bus path tracking control problem;

[0129] The method of establishing the MDP model is as follows: Step 6.1: Define the smart bus at the current moment The state vector under , defines the smart bus at the current moment The control action under is the compensation amount of the expected driving curvature of the smart bus ;

[0130] Step 6.2: Calculate the current time using formula (8) Lower state vector The Return :

[0131] ;

[0132] In formula (8), is a semi-positive definite diagonal matrix associated with the state dimension, is a positive constant;

[0133] Step 6.3: Define the current moment The value of the next state function , the calculation formula is as follows:

[0134] ;

[0135] In formula (9), For the current moment The state vector of the smart bus under For the current moment Next control action, For the current moment The immediate return value under is the discount factor, For the next moment The objective function value under For the next moment The state vector of the smart bus under , is calculated by formula (10):

[0136] ;

[0137] In formula (10), To control the cycle, ;

[0138] Step 6.4: Calculate the current time using formula (11) The optimal control action :

[0139] (11) .

[0140] S7: Build Figure 3 The control framework based on the MKDHP algorithm is shown in the figure. The MKDHP algorithm is used to obtain the optimal control action of the above formula (11), and the actuator output in the MKDHP algorithm is calculated using formula (12). And as a smart bus at the moment The following control actions:

[0141] (12);

[0142] In formula (12), is the maximum lateral acceleration, ; are the first weight and bias from the input layer to the hidden layer, ; are the second weight and bias from the hidden layer to the output layer, .

[0143] Use formula (13) to calculate the output of the evaluator in the MKDHP algorithm And as a smart bus at the moment The following goals:

[0144] ;

[0145] In formula (13), For the current moment The weight matrix of the lower evaluator, For samples in the nuclear dictionary The basis functions formed, , is the number of samples in the kernel dictionary. Figure 3 The structures and parameters of evaluator 1 and evaluator 2 are exactly the same, and they estimate the derivatives of the value functions of the current state and the predicted state with respect to the state respectively.

[0146] S8: Establish online update rules for the actuator module and the evaluator module in the MKDHP algorithm;

[0147] The online update rules are as follows: Step 8.1, the online update rules of the actuator module, are essentially the online update process of the network weights and biases, which is gradually implemented through iterative operations. The initial weights are obtained by using the data of other path following control methods through the weight update rules derived online or offline in this step:

[0148] Minimize the error function established using equation (14) Learning objectives for the executor module:

[0149] ;

[0150] Update the current time using equations (15)-(18) Weight of the lower actuator module and bias , so as to obtain the next moment The weight under and bias :

[0151] ;

[0152] ;

[0153] ;

[0154] ;

[0155] In formula (15) - formula (18) is the learning step size of the evaluator module, ;

[0156] Step 8.2, online update rules of the evaluator module:

[0157] The error function is defined using formula (19): for:

[0158] ;

[0159] In formula (19), Represented as a vector Middle elements, Represented as a vector Middle elements, It can be obtained from formula (20):

[0160] ;

[0161] For status , assuming is its dimension, then , using formula (21) to define the intermediate variable :

[0162] ;

[0163] In formula (21) is a diagonal matrix Middle The diagonal element values, Indicates status Middle elements, Represents the weight matrix Middle The row weight vector is then used to obtain the error function: No. Row vector for:

[0164] (twenty two);

[0165] Defining intermediate variables , use equation (23)-(25) to update the current time The weight of the lower evaluator module , so as to obtain the next moment The weight under :

[0166] ;

[0167] (twenty four);

[0168] ;

[0169] In formula (23) to formula (25), is the learning step size of the evaluator module, For the forgetting factor, .

[0170] S9: Control action obtained by the actuator module , then use formula (26) to get the steering wheel angle of the smart bus :

[0171] (26);

[0172] Determine the current moment Check whether the target point in the predetermined trajectory closest to the smart bus is the end point. If so, the smart bus stops driving; otherwise, assign i+1 to i and return to S4 to execute subsequent steps in sequence.

[0173] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.

Claims

1. A control method for an intelligent bus to track a predetermined trajectory, characterized in that: The steps include: S1: Establish a local coordinate system with the main antenna of the GPS sensor installed on the smart bus as the coordinate origin; establish a global coordinate system with the center of the earth as the origin; S2: After driving the smart bus for a certain period of time, record n ​​lines of longitude, latitude, heading angle and speed information as a predetermined track; S3: obtaining a kernel dictionary based on the approximate linear correlation analysis method for the collected smart bus state samples; S4: Determine the current time The position information of the smart bus and the position information at the preview point are obtained, and the trajectory tracking error state equation in the local coordinate system is calculated, where i is a positive integer; S5: Calculate the current time The curvature of the road ; S6: Establish the MDP model of intelligent bus path tracking control problem; S7: Design a controller for tracking the predetermined trajectory based on the MKDHP algorithm. The controller includes an actuator module and an evaluator module. The actuator module outputs the current moment The following control actions The evaluator module includes a first evaluator module and a second evaluator module; each of which outputs the current moment Next and next moment The derivative of the state value function with respect to the state and ; S8: Establish online update rules for the actuator module and the evaluator module in the MKDHP algorithm; S9: Determine the current moment Check whether the target point in the predetermined trajectory closest to the smart bus is the end point. If so, the smart bus stops driving; otherwise, assign i+1 to i and return to S4 for sequential execution.

2. A control method for an intelligent bus to track a predetermined trajectory according to claim 1, characterized in that: The method for constructing the core dictionary in S3 is as follows: Step 3.1, select formula (1) as the kernel function: （1）； In formula (1), and is the state vector in the sample, is the kernel width of the Gaussian kernel function, and are the parameters of the polynomial kernel function; Step 3.2: Smart bus status sample set Based on the approximate linear correlation analysis method, it has been obtained feature vectors, ; is the obtained kernel dictionary, , to determine whether to add the new feature vector Add to the kernel dictionary, first use equations (2) and (3) to calculate : ； In formula (2) , is an element in the kernel dictionary, ,but for: ； Then calculate the With predefined threshold Compare to update the kernel dictionary, if ,Will Added to the nuclear dictionary, i.e. ; otherwise there is no change to the nuclear dictionary.

3. The control method for an intelligent bus to track a predetermined trajectory according to claim 1, characterized in that: The method for constructing the trajectory tracking error state equation in the local coordinate system in S4 is as follows: Step 4.1: Define the current moment The position coordinates of the rear axle center of the next smart bus in the global coordinate system are: , the speed is , the heading angle is , denoted as ; Step 4.2, Settings is the preview distance. In the predetermined trajectory, search for a target point that is greater than the preview distance L and closest to the smart bus. This target point is defined as the preview point. , denoted as ; Step 4.3: Define the smart bus at the current moment Relative to the preview point The error state equation is: ； In formula (4), The current moment Lower longitudinal deviation, lateral deviation and heading angle deviation; Step 4.4: According to formula (5), the distance between the smart bus and the preview in the local coordinate system is obtained: Preview point The rate of change of the lateral deviation and heading angle deviation is: ； In formula (5), and At the current moment The longitudinal and lateral speeds of the rear axle center point of the smart bus, The wheelbase of the smart bus. For the current moment The front wheel turning angle of the smart bus, Preview point The curvature of the road.

4. The control method for an intelligent bus to track a predetermined trajectory according to claim 1, characterized in that: The road curvature in S5 The calculation method is as follows, Step 5.1: At the current moment Next, select the distance from the predefined trajectory to the smart bus The nearest target point , calculated using formula (6) To preview point Distance ,like , then a quadratic polynomial fit is used to The trajectory between Otherwise, search for a distance greater than And the target point closest to the smart bus , using a quadratic polynomial fit to The trajectory between ； In formula (6) is the number of target points of the fitted trajectory; Step 5.2: Use a quadratic polynomial to fit the above trajectory. The fitted quadratic polynomial equation is shown in formula (7): ； In formula (7) is the constant term coefficient, is the coefficient of the first-order term, is the coefficient of the quadratic term; The road curvature .

5. The control method for an intelligent bus to track a predetermined trajectory according to claim 1, characterized in that: The method for establishing the MDP model in S6 is as follows Step 6.1: Define the smart bus at the current moment The state vector under , defines the smart bus at the current moment The control action under is the compensation amount of the expected driving curvature of the smart bus ; Step 6.2: Calculate the current time using formula (8) Lower state vector The Return : ； In formula (8), is a semi-positive definite diagonal matrix associated with the state dimension, is a positive constant; Step 6.3: Define the current moment The value of the next state function , the calculation formula is as follows: ； In formula (9), For the current moment The state vector of the smart bus under For the current moment Next control action, For the current moment The immediate return value under is the discount factor, For the next moment The objective function value under For the next moment The state vector of the smart bus under , is calculated by formula (10): ； In formula (10), To control the cycle, ; Step 6.4: Calculate the current time using formula (11) The optimal control action : (11) 。 6. The control method for an intelligent bus to track a predetermined trajectory according to claim 1, characterized in that: The controller design method for tracking the predetermined trajectory in S7 is as follows Use formula (12) to calculate the actuator output in the MKDHP algorithm And as a smart bus at the moment The following control actions: ； In formula (12), is the maximum lateral acceleration, ; are the first weight and bias from the input layer to the hidden layer, ; are the second weight and bias from the hidden layer to the output layer, ; Use formula (13) to calculate the output of the evaluator in the MKDHP algorithm And as a smart bus at the moment The following goals: ； In formula (13), For the current moment The weight matrix of the lower evaluator, For samples in the nuclear dictionary The basis functions formed, , is the number of samples in the kernel dictionary.

7. The control method for an intelligent bus to track a predetermined trajectory according to claim 1, characterized in that: The online update rules of the actuator module and the evaluator module in S8 are as follows: Step 8.1: Online update rules for actuator modules: Minimize the error function established using equation (14) Learning objectives for the executor module: ； Update the current time using equations (15)-(18) Weight of the lower actuator module and bias , so as to obtain the next moment The weight under and bias : ； ； ； ； In formula (15) - formula (18) is the learning step size of the evaluator module, ; Step 8.2, online update rules of the evaluator module: The error function is defined using formula (19): for: ； In formula (19), Represented as a vector Middle elements, Represented as a vector Middle elements, It can be obtained from formula (20): ； For status , assuming is its dimension, then , using formula (21) to define the intermediate variable : ； In formula (21) is a diagonal matrix Middle The diagonal element values, Indicates status Middle elements, Represents the weight matrix Middle The row weight vector is then used to obtain the error function: No. Row vector for: (22)； Defining intermediate variables , use equation (23)-(25) to update the current time The weight of the lower evaluator module , so as to obtain the next moment The weight under : ； （24）； ； In formula (23) to formula (25), is the learning step size of the evaluator module, For the forgetting factor, .

Images