A KNN-based heavy-load AGV lateral stability control method

Abstract

A KNN-based lateral stability control method for heavy-duty AGVs is disclosed, characterized by the following steps: 1. Classifying AGV operating conditions into five categories based on different loads, and collecting motion state data of the AGV under different operating conditions using simulation software; 2. Establishing a K-nearest neighbor (KNN) classifier, and training and validating it using the collected dataset; 3. Designing a set of nonlinear sub-controllers based on fuzzy PID, calculating the required yaw moment, and distributing the torque to the four drive wheels according to the torque distribution rule; 4. Introducing an error judgment strategy, activating the controller based on the centroid sideslip angle error to control its lateral stability. This invention establishes a load KNN classifier for heavy-duty AGVs, which can monitor the centroid sideslip angle error in real time. If the error exceeds a threshold, the controller is activated, the classifier's classification result is matched to the corresponding sub-controller, and an additional yaw moment is calculated and applied to the AGV.

Description

Technical Field

[0001] This invention relates to a lateral stability control method for heavy-duty AGVs, specifically a KNN-based lateral stability control method for heavy-duty AGVs. Background Technology

[0002] With the rapid development of the industrial AGV industry, heavy-duty AGVs are being used for product transfer and docking assembly in the final assembly and sub-assembly stages of large spacecraft and aircraft. However, during the transfer process, due to the uneven load distribution of the drive wheels of the heavy-duty AGV, the center of gravity shifts significantly between the unloaded and high-load states, causing a substantial decrease in the stability of the heavy-duty AGV or even instability. Therefore, it is crucial to invent a lateral stability control method.

[0003] Traditional PID control is primarily based on mathematical models. However, with technological advancements and societal development, controlled objects have become increasingly complex and uncertain, making it impossible to describe their nonlinear changes using precise mathematical models and methods. Furthermore, it fails to meet the demands of heavy-load flexible transport and the problem of reduced control response speed due to the large inertia of heavy loads. This paper proposes a KNN-based lateral stability control method for heavy-load AGVs. It utilizes a KNN classifier to categorize AGV load conditions and designs fuzzy PID sub-controllers for different categories. The corresponding controller is activated based on the load condition to control lateral stability, thus improving the adaptability of the PID controller. Summary of the Invention

[0004] The purpose of this invention is to address the problems of lateral stability and slow control response caused by the large inertia of heavy load AGVs when transporting heavy products, and to invent a lateral stability control method for heavy load AGVs based on KNN.

[0005] The technical solution of this invention is:

[0006] A KNN-based lateral stability control method for heavy-duty AGVs is characterized by the following steps: First, based on different loads, the AGV's operating conditions are divided into five categories, and motion state data of the AGV under different conditions are collected using simulation software. Second, a K-nearest neighbor (KNN) classifier is established and trained and validated using the collected dataset. Then, a set of nonlinear sub-controllers based on fuzzy PID is designed to calculate the required yaw torque and distribute the torque to the four drive wheels according to the torque distribution rule. Finally, an error judgment strategy is introduced, activating the controller based on the centroid sideslip angle error to control its lateral stability. Specifically, the method includes the following steps:

[0007] Step 1: Based on the different loads of the heavy-duty AGV, the AGV driving conditions are divided into 5 categories. The motion state data of the AGV under different conditions is collected using simulation software. The dataset is divided into training set and test set, and the dataset is preprocessed.

[0008] Step 2: Build a K-Nearest Neighbor (KNN) classifier and use the collected dataset for training and validation;

[0009] Step 3: Establish a two-degree-of-freedom dynamic model of the heavy-duty AGV, and calculate the centroid sideslip angle under ideal conditions based on the classification results of the classifier;

[0010] Step 4: Design a set of nonlinear subcontrollers based on fuzzy PID for the five typical load types in Step 1, and calculate the additional yaw moment;

[0011] Step 5: Introduce an error judgment strategy to calculate the error between the actual centroid sideslip angle and the ideal centroid sideslip angle. When the centroid sideslip angle error exceeds the threshold, activate the sub-controller corresponding to the classifier classification result.

[0012] Step 6: Establish torque distribution rules and distribute the calculated yaw moment to the four drive wheels according to the rules to control its lateral stability.

[0013] In step 1: Based on the different loads of the heavy-duty AGV, the AGV's operating conditions are divided into 5 categories. Simulation software is used to collect the AGV's motion state data under different operating conditions. The dataset is divided into training and testing sets, and preprocessed. This invention applies to a heavy-duty AGV transporting a certain aerospace product. The AGV's transport load is divided into two categories: empty and high load. Under empty load conditions, the weight borne by the AGV's drive wheels is the vehicle's weight. Under high load conditions, there are four sub-conditions: loading product shelves, loading product shelves and product A, loading product shelves and product B, and loading product shelves, product A, and product B. For these different load conditions, this invention designs 5 typical loads based on the actual product weight: 1.5t, 2.5t, 3.5t, 4.5t, and 5.5t. To ensure safety, heavy-duty AGVs are required to travel at low speeds when transporting products. Therefore, three typical speeds of 10km / h, 25km / h, and 40km / h are designed for each load. Considering that the absence of magnetic strips during magnetic navigation can cause the AGV to struggle with left and right tracking, this invention designs different front wheel angles to simulate tracking. The above working conditions are designed using the simulation software Trucksim, with a simulation time of 30s and a sampling frequency of 1000Hz for each condition. The collected dataset is divided into training and testing sets in a 7:3 ratio, and the data is normalized by using a formula to normalize the original data to the [0,1] interval.

[0014] Step 2: Establish a K-Nearest Neighbor (KNN) classifier, and train and test the classifier using the collected dataset. The feature space of the KNN classifier is designed using the longitudinal velocity V measured by the inertial measurement unit of the AGV. x Lateral acceleration A y yaw rate AV x and yaw rate AV z These four feature vectors form the data, which are divided into five classes according to the feature space, corresponding to five different load conditions: L1, L2, L3, L4, and L5. The Euclidean distance is used as shown in equation (1), where d... 12 The distance between the new data and the feature space is represented by n, where n represents the number of feature vectors, and x represents the distance between the new data and the feature space. 1i Let x represent the i-th eigenvector in the feature space. 2i Let represent the i-th feature vector in the new data. In the feature space, find the k nearest data points to the new data, and select the class with the most categories among these k data points as the new data's category. This invention uses the preprocessed training set data from step 1 to establish the feature space, and uses the test set data to validate the established KNN classifier. Set k to 11; that is, for a test data point, calculate and find the 11 training data points with the closest Euclidean distance to that data. The category of the new data is determined by the class with the most category features among these 11 nearest data points. When the number of categories is the same among these 11 nearest data points, the point with the closest Euclidean distance is selected as the new data's category. Finally, the output category is stored in CLASS_ID.

[0015]

[0016] In step 3: a two-degree-of-freedom dynamic model of the heavy-duty AGV is established, and differential equations for the sideslip angle and yaw rate are established based on the model. The sideslip angle under ideal conditions is calculated. Ignoring the effects of the steering system and air resistance, only the lateral and yaw motions of the heavy-duty transport vehicle are considered, and a model incorporating the sideslip angle β and yaw rate ω is established. z The differential equation of motion is shown in equation (2):

[0017]

[0018] In equation (2), m is the total vehicle mass, selected according to the CLASS_ID value; a and b are the distances from the center of gravity to the front and rear axles, respectively; and k is the mass of the vehicle. f k r These are the front and rear axle lateral stiffness, I z Let v be the moment of inertia about the z-axis. x δ represents the longitudinal velocity, and δ represents the front wheel steering angle.

[0019] According to the two-degree-of-freedom dynamic differential equation of the AGV, under ideal conditions, the sideslip angular velocity and yaw angular velocity of the AGV are both 0. Therefore, the ideal yaw angular velocity and the ideal sideslip angle are determined as follows:

[0020]

[0021]

[0022] in, Here, L represents the steady-state response parameter, and L is the wheelbase between the front and rear axles of the AGV.

[0023] Considering that the AGV's movement on actual roads is limited by the road surface adhesion, its lateral acceleration a y It should satisfy equation (5), where R is the wheel radius and μ is the road adhesion coefficient:

[0024] a y =v x 2 / R=ω z v x ≤μg (5)

[0025] Based on equations (3), (4), and (5), the side slip angle of the AGV's centroid under the limiting condition satisfies equation (6):

[0026]

[0027] In summary, the centroid sideslip angle should be the smaller of equations (4) and (6), as shown in equation (7):

[0028]

[0029] In step 4: a set of nonlinear sub-controllers based on fuzzy PID is designed, and the required yaw moment is calculated. The fuzzy PID nonlinear sub-controller set designed in this invention consists of 5 nonlinear sub-controllers. The inputs of each sub-controller are the error e between the ideal and actual centroid sideslip angles, and the error rate ec. The outputs are the changes in various parameter factors of the PID controller, including the proportional factor change ΔK. P , Change in integral factor ΔK i and the change in differential factor ΔK dFor each sub-controller, based on the deviation range of the centroid sideslip angle under the corresponding load, different basic universes of discourse for the error quantity e and the error rate ec are defined, and the fuzzy universe is unified as [-6, 6]. The fuzzy subset of the input variables is set as {NB, NM, NS, ZO, PS, PM, PB}, which is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}, and the membership function adopts the Gaussian function. Similarly, a unified basic universe of discourse, fuzzy universe of discourse and fuzzy subset are defined for the three output quantities of each sub-controller, and the membership function adopts the Gaussian function. The control factors of the PID controller are obtained by the incremental method, as shown in equations (8) to (10), where K p K i K d These are the controller proportional factor, controller integral factor, and controller derivative factor, respectively, K. p0 K i0 K d0 These are the initial values ​​for the corresponding factors:

[0030] K P =K P0 +ΔK P (8)

[0031] K i =K i0 +ΔK i (9)

[0032] K d =K d0 +ΔK d (10)

[0033] Finally, the required yaw moment M is calculated based on the PID controller expression. z (t), where t represents time t, as shown in equation (11), where e(t) is the centroid side deviation angle error at time t:

[0034]

[0035] In step 5: an error judgment strategy is introduced to calculate the error between the actual centroid sideslip angle and the ideal centroid sideslip angle. When the centroid sideslip angle error exceeds a threshold, the sub-controller corresponding to the classifier's classification result is activated. Using the ideal centroid sideslip angle calculated from the ideal AGV two-degree-of-freedom model in step 2 as a reference, the deviation between the actual and ideal centroid sideslip angles is used as the input to the error judgment strategy. This is compared with the AGV centroid sideslip angle judgment threshold. If the threshold is exceeded, the CLASS_ID value output by the classifier is assigned to CONTROL_ID as the output of the error judgment strategy, activating the sub-controller corresponding to the CONTROL_ID value. If the threshold is not exceeded, the CONTROL_ID value is set to 0, i.e., the sub-controller is shut down, reducing the burden on the control system.

[0036] In step 6: a torque distribution rule is established, and the calculated yaw torque is distributed to the four drive wheels according to the rule to control its lateral stability. This invention addresses the impact of different loads on the motion stability of heavy-duty AGVs; therefore, the torque distribution is based on the vertical load of the four drive wheels. The distribution rules are shown in equations (12) to (15):

[0037]

[0038]

[0039]

[0040]

[0041] In the formula, T ij This represents the torque of the four drive wheels, where i represents the front wheel (f) and j represents the rear wheel (r), and j represents the left wheel (l) and j represents the right wheel (r). M z R is the yaw moment calculated in step 5; F is the tire radius; z For the total vertical load, F zij The vertical load on the four drive wheels; δ is the front wheel steering angle; d f d is the front wheel track. r This refers to the rear wheel track.

[0042] The beneficial effects of this invention are:

[0043] (1) Establish a KNN classifier to classify the longitudinal velocity V measured by the inertial measurement unit of the AGV. x Lateral acceleration A y yaw rate AV x and yaw rate AV z As a feature space, based on the actual load conditions of heavy-duty AGVs during operation, the feature space is divided into 5 typical loads to meet the needs of heavy-duty flexible transfer.

[0044] (2) For the five typical load conditions in (1), design a set of nonlinear subcontrollers based on fuzzy PID and calculate the required yaw torque. Use fuzzy algorithm to tune the PID parameters, formulate fuzzy rules, membership functions and membership centers based on expert experience. Traditional PID control methods cannot meet the nonlinear conditions of heavy-load AGV motion;

[0045] (3) An error judgment strategy is introduced to calculate the error between the actual centroid sideslip angle and the ideal centroid sideslip angle. When the centroid sideslip angle error exceeds the threshold, the sub-controller is activated to control its lateral stability; if it does not exceed the threshold, the sub-controller is turned off to reduce the burden on the control system. Attached Figure Description

[0046] Figure 1 This is a flowchart of the present invention.

[0047] Figure 2 This is the design drawing of the present invention.

[0048] Figure 3 This is the two-degree-of-freedom dynamic model of the heavy-duty AGV of the present invention.

[0049] Figure 4 The PID controller parameter ΔK of this invention p ΔK i and ΔK d Fuzzy reasoning rules. Detailed Implementation

[0050] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0051] like Figure 1-4 As shown.

[0052] A KNN-based lateral stability control method for heavy-duty AGVs, the process of which is as follows: Figure 1 As shown, the specific steps include the following:

[0053] Step 1: Based on the different loads of the heavy-duty AGV, the AGV driving conditions are divided into 5 categories. The motion state data of the AGV under different conditions is collected using simulation software. The dataset is divided into training set and test set, and the dataset is preprocessed.

[0054] Taking a heavy-duty AGV for transporting a certain aerospace product as an example, the AGV's transport load is divided into two categories: empty and high load. In the empty load condition, the weight borne by the AGV's drive wheels is the vehicle's weight. In the high load condition, there are four scenarios: loading product shelves, loading product shelves and product A, loading product shelves and product B, and loading product shelves, product A, and product B. For these different load conditions, this invention designs five typical loads based on the actual product weight: 1.5t, 2.5t, 3.5t, 4.5t, and 5.5t. Since the heavy-duty AGV is designed to travel at low speeds to ensure safety during product transport, three typical speeds are designed for each load: 10km / h, 25km / h, and 40km / h. Considering that the absence of magnetic strips during magnetic navigation can cause the AGV to move in a left-right tracking state, this invention designs different front wheel angles to simulate the tracking state. The above working conditions were designed using the simulation software Trucksim. The simulation time for each working condition was set to 30 seconds, and the sampling frequency was set to 1000Hz. The collected dataset was divided into training and testing sets in a 7:3 ratio, and the data was normalized. That is, the original data was organized and normalized to the [0,1] interval using a formula.

[0055] Step 2: Build a K-Nearest Neighbor (KNN) classifier and train and test the classifier using the collected dataset.

[0056] like Figure 2 As shown, the feature space of the KNN classifier is designed using the longitudinal velocity V measured by the inertial measurement unit of the AGV. x Lateral acceleration A y yaw rate AV x and yaw rate AV z These four feature vectors form the data, which are divided into five classes according to the feature space, corresponding to five different load conditions: L1, L2, L3, L4, and L5. The Euclidean distance is used as shown in equation (1), where d... 12 The distance between the new data and the feature space is represented by n, where n represents the number of feature vectors, and x represents the distance between the new data and the feature space. 1i Let x represent the i-th eigenvector in the feature space. 2iLet represent the i-th feature vector in the new data. In the feature space, find the k nearest data points to the new data, and select the class with the most categories among these k data points as the new data's category. This invention uses the preprocessed training set data from step 1 to establish the feature space, and uses the test set data to validate the established KNN classifier. Set k to 11; that is, for a test data point, calculate and find the 11 training data points with the closest Euclidean distance to that data. The category of the new data is determined by the class with the most category features among these 11 nearest data points. When the number of categories is the same among these 11 nearest data points, the point with the closest Euclidean distance is selected as the new data's category. Finally, the output category is stored in CLASS_ID.

[0057]

[0058] Step 3: Establish a two-degree-of-freedom dynamic model of the heavy-duty AGV and calculate the centroid sideslip angle under ideal conditions based on the classification results of the classifier.

[0059] like Figure 3 As shown, ignoring the effects of the steering system and air resistance, and considering only the lateral and yaw motions of the heavy-duty transport vehicle, a model is established that includes the sideslip angle β and the yaw angle ω. z The differential equation of motion is shown in equation (2):

[0060]

[0061] In equation (2), m is the total vehicle mass, selected according to the CLASS_ID value; a and b are the distances from the center of gravity to the front and rear axles, respectively; and k is the mass of the vehicle. f k r These are the front and rear axle lateral stiffness, I z Let v be the moment of inertia about the z-axis. x δ represents the longitudinal velocity, and δ represents the front wheel steering angle.

[0062] According to the two-degree-of-freedom dynamic differential equation of the AGV, under ideal conditions, the sideslip angular velocity and yaw angular velocity of the AGV are both 0. Therefore, the ideal yaw angular velocity and the ideal sideslip angle are determined as follows:

[0063]

[0064]

[0065] in, Here, L represents the steady-state response parameter, and L is the wheelbase between the front and rear axles of the AGV.

[0066] Considering that the AGV's movement on actual roads is limited by the road surface adhesion, its lateral acceleration a y It should satisfy equation (5), where R is the wheel radius and μ is the road adhesion coefficient:

[0067] a y =v x 2 / R=ω z v x ≤μg (5)

[0068] Based on equations (3), (4), and (5), the side slip angle of the AGV's centroid under the limiting condition satisfies equation (6):

[0069]

[0070] In summary, the centroid sideslip angle should be the smaller of equations (4) and (6), as shown in equation (7):

[0071]

[0072] Step 4: Design a set of nonlinear subcontrollers based on fuzzy PID for the five typical load types in Step 1, and calculate the additional yaw moment.

[0073] The fuzzy PID nonlinear subcontroller set in this embodiment consists of 5 nonlinear subcontrollers, such as... Figure 2 As shown, the inputs to each sub-controller are the error e between the ideal and actual centroid sideslip angles, and the error rate ec. The outputs are the parameter factor changes of the PID controller, including the proportional factor change ΔK. P , Change in integral factor ΔK i and the change in differential factor ΔK d For each sub-controller, based on the deviation range of the centroid sideslip angle under the corresponding load, different basic universes of discourse for the error quantity *e* and the error rate *ec* are defined, with the fuzzy universes uniformly set to [-6, 6]. The fuzzy subsets of the input variables are set to {NB, NM, NS, ZO, PS, PM, PB}, which are {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}, and the membership function is a Gaussian function. Similarly, for the three output quantities of each sub-controller, a unified basic universe of discourse, fuzzy universe of discourse, and fuzzy subset are defined, with the membership function using a Gaussian function, and the fuzzy inference rules are as follows: Figure 4 As shown. The control factors of the PID controller are obtained by the incremental method, as shown in equations (8) to (10), where K p K i K d These are the controller proportional factor, controller integral factor, and controller derivative factor, respectively, K. p0 K i0 K d0 These are the initial values ​​for the corresponding factors:

[0074] K P =K P0 +ΔKP (8)

[0075] K i =K i0 +ΔK i (9)

[0076] K d =K d0 +ΔK d (10)

[0077] Finally, the required yaw moment M is calculated based on the PID controller expression. z (t), where t represents time t, as shown in equation (11), where e(t) is the centroid side deviation angle error at time t:

[0078]

[0079] Step 5: Introduce an error judgment strategy to calculate the error between the actual centroid sideslip angle and the ideal centroid sideslip angle. When the centroid sideslip angle error exceeds a certain threshold, activate the sub-controller corresponding to the classifier's classification result.

[0080] Using the ideal centroid sideslip angle calculated from the ideal AGV two-degree-of-freedom model in step 2 as a reference, the deviation between the actual and ideal centroid sideslip angles of the AGV is used as the input to the error judgment strategy. This is compared with the AGV centroid sideslip angle judgment threshold. If the threshold is exceeded, the CLASS_ID value output by the classifier is assigned to CONTROL_ID as the output of the error judgment strategy, activating the sub-controller corresponding to the CONTROL_ID value. If the threshold is not exceeded, the CONTROL_ID value is assigned to 0, i.e., the sub-controller is turned off to reduce the burden on the control system.

[0081] Step 6: Establish torque distribution rules and distribute the calculated yaw moment to the four drive wheels according to the rules to control its lateral stability.

[0082] To address the impact of different loads on the motion stability of heavy-duty AGVs, torque distribution is based on the vertical loads on the four drive wheels. The distribution rules are shown in equations (12) to (15):

[0083]

[0084]

[0085]

[0086]

[0087] In the formula, T ijThis represents the torque of the four drive wheels, where i represents the front wheel (f) and j represents the rear wheel (r), and j represents the left wheel (l) and j represents the right wheel (r). M z R is the yaw moment calculated in step 5; F is the tire radius; z For the total vertical load, F zij The vertical load on the four drive wheels; δ is the front wheel steering angle; d f d is the front wheel track. r This refers to the rear wheel track.

[0088] The parts not covered in this invention are the same as or can be implemented using existing technologies.

Claims

1. A KNN-based lateral stability control method for heavy-duty AGVs, characterized by: First, based on the different loads of the heavy-duty AGV, the AGV's operating conditions are divided into 5 categories, and motion state data of the AGV under different operating conditions are collected using simulation software. Second, a K-nearest neighbor (KNN) classifier is established and trained and validated using the collected dataset. Then, a set of nonlinear sub-controllers based on fuzzy PID is designed to calculate the required yaw torque and distribute the torque to the four drive wheels according to the torque distribution rules. Finally, an error judgment strategy is introduced to activate the controller based on the centroid side slip angle error to control its lateral stability. Includes the following steps: Step 1: Based on the different loads of the heavy-duty AGV, the AGV driving conditions are divided into 5 categories. The motion state data of the AGV under different conditions is collected using simulation software. The dataset is divided into training set and test set, and the dataset is preprocessed. Step 2: Build a K-Nearest Neighbor (KNN) classifier and train and validate it using the collected dataset. Building the KNN classifier means training and testing it using the collected dataset. The feature space of the KNN classifier is designed using the longitudinal velocity measured by the AGV's inertial measurement unit. Lateral acceleration yaw rate and yaw rate These four feature vectors are used to divide the data into five categories according to the feature space, corresponding to five different load conditions, namely L1, L2, L3, L4 and L5; the Euclidean distance is calculated as shown in equation (1): （1） where d 12 represents the distance between the new data and the feature space, n represents the number of feature vectors, x 1i represents the i-th feature vector in the feature space, x 2i represents the i-th feature vector in the new data; Find the k nearest data points to the new data in the feature space, and select the class with the most categories among the k nearest data points as the class of the new data. Construct the feature space using the preprocessed training set data from step 1, and validate the constructed KNN classifier using the test set data. Set the value of k to 11, that is, for a test data point, calculate and find the 11 nearest training data points in Euclidean distance to that data, and determine the class of the new data by selecting the class with the most category features among the 11 nearest data points. When the same number of categories appears among these 11 nearest data points, select the point with the closest Euclidean distance as the class of the new data. Finally, save the output class in CLASS_ID. Step 3: Establish a two-degree-of-freedom dynamic model of the heavy-duty AGV, and calculate the centroid sideslip angle under ideal conditions based on the classification results of the classifier; Step 4: Design a set of nonlinear subcontrollers based on fuzzy PID for the five typical load types in Step 1, and calculate the additional yaw moment; Step 5: Introduce an error judgment strategy to calculate the error between the actual centroid sideslip angle and the ideal centroid sideslip angle. When the centroid sideslip angle error exceeds a set threshold, activate the sub-controller corresponding to the classifier classification result. Step 6: Establish torque distribution rules and distribute the calculated yaw moment to the four drive wheels according to the rules to control its lateral stability.

2. The method according to claim 1, characterized in that, In step 1: Based on the different loads of heavy-duty AGVs, the AGV operating conditions are divided into 5 categories. This involves using simulation software to collect the motion state data of the AGV under different operating conditions, dividing the dataset into training and testing sets, and preprocessing the dataset. AGV transfer load conditions are divided into two categories: empty and high load. In the empty load condition, the weight borne by the AGV drive wheels is the vehicle weight. In the high load condition, there are four sub-conditions: loading product shelves, loading product shelves and product A, loading product shelves and product B, and loading product shelves, product A, and product B. For these different load conditions, based on the actual product weight, 5 typical loads are designed: 1.5t, 2.5t, 3.5t, ... 4.5t and 5.5t; due to the requirement of low speed for safety when the AGV is transporting products under heavy load, three typical speeds of 10km / h, 25km / h and 40km / h are designed for each load; considering that the absence of magnetic strips will cause the AGV to move left and right when moving through magnetic navigation, different front wheel angles are designed to simulate the tracking state; the above working conditions are designed using the simulation software Trucksim, and the simulation time for each working condition is set to 30s and the sampling frequency is set to 1000Hz; the collected dataset is divided into training set and test set in a 7:3 ratio and the data is normalized, that is, the original data is organized by formula and all are normalized to the [0,1] interval.

3. The method according to claim 1, characterized in that, In step 3: A two-degree-of-freedom dynamic model of a heavy-duty AGV is established, and differential equations for the sideslip angle and yaw rate are derived based on the model. When calculating the sideslip angle under ideal conditions, the effects of the steering system and air resistance are ignored, and only the lateral and yaw motions of the heavy-duty transport vehicle are considered. A model incorporating the sideslip angle β and yaw rate ω is then established. z The differential equation of motion is shown in equation (2): （2） In equation (2), m is the total vehicle mass, selected according to the CLASS_ID value; a and b are the distances from the center of gravity to the front and rear axles, respectively; and k is the mass of the vehicle. f k r These are the front and rear axle lateral stiffness, I z Let v be the moment of inertia about the z-axis. x Where δ is the longitudinal velocity and δ is the front wheel steering angle; According to the two-degree-of-freedom dynamic differential equation of the AGV, under ideal conditions, the sideslip angular velocity and yaw angular velocity of the AGV are both 0. Therefore, the ideal yaw angular velocity and the ideal sideslip angle are determined as follows: （3） （4） in, Here, L represents the steady-state response parameter, and L is the wheelbase between the front and rear axles of the AGV. Considering that the AGV's movement on actual roads is limited by the road surface adhesion, its lateral acceleration a y It should satisfy equation (5), where R is the wheel radius and μ is the road surface adhesion coefficient: （5） Based on equations (3), (4), and (5), it is derived that the AGV's centroid side slip angle satisfies equation (6) under the extreme condition: （6） In summary, the centroid sideslip angle should be the smaller of equations (4) and (6), as shown in equation (7): （7）。 4. The method according to claim 1, characterized in that, In step 4: Design a set of nonlinear subcontrollers based on fuzzy PID to calculate the required yaw moment. The fuzzy PID nonlinear subcontroller set consists of 5 nonlinear subcontrollers. The inputs of each subcontroller are the error e between the ideal and actual centroid sideslip angles and the error rate ec. The outputs are the changes in various parameter factors of the PID controller, including the change in the proportional factor. Changes in integral factors and the change in differential factor ; For each sub-controller, based on the deviation range of the centroid side slip angle under the corresponding load, different basic universes of discourse for the error quantity e and the error rate ec are defined, and the fuzzy universe is unified as [-6,6]; the fuzzy subset of the input variables is set as {NB,NM,NS,ZO,PS,PM,PB}, which is {negative large, negative medium, negative small, zero, positive small, positive medium, positive large}, and the membership function adopts the Gaussian function; similarly, a unified basic universe of discourse, fuzzy universe of discourse and fuzzy subset are defined for the three output quantities of each sub-controller, and the membership function adopts the Gaussian function; the control factors of the PID controller are obtained by the incremental method, as shown in equations (8) to (10), where K p K i K d These are the controller proportional factor, controller integral factor, and controller derivative factor, respectively, K. p0 K i0 K d0 These are the initial values ​​for the corresponding factors: （8） （9） （10） Finally, the required yaw moment M is calculated based on the PID controller expression. z (t), where t represents time t, as shown in equation (11), where e(t) is the centroid side deviation angle error at time t: （11）。 5. The method according to claim 1, characterized in that, In step 5: An error judgment strategy is introduced to calculate the error between the actual centroid sideslip angle and the ideal centroid sideslip angle. When the centroid sideslip angle error exceeds a threshold, the sub-controller corresponding to the classifier's classification result is activated. The ideal centroid sideslip angle calculated based on the ideal AGV two-degree-of-freedom model in step 2 is used as a reference. The deviation between the actual and ideal centroid sideslip angles of the AGV is used as the input of the error judgment strategy and compared with the AGV centroid sideslip angle judgment threshold. If the threshold is exceeded, the CLASS_ID value output by the classifier is assigned to CONTROL_ID as the output of the error judgment strategy, and the sub-controller corresponding to the CONTROL_ID value is activated. If the threshold is not exceeded, the CONTROL_ID value is set to 0, i.e., the sub-controller is turned off to reduce the burden on the control system.

6. The method according to claim 1, characterized in that, In step 6: Establishing torque distribution rules and distributing the calculated yaw torque to the four drive wheels according to the rules to control its lateral stability refers to the impact of different loads on the motion stability of the heavy-duty AGV. Therefore, the torque distribution is based on the vertical load of the four drive wheels; the distribution rules are shown in equations (12) to (15): （12） （13） （14） （15） In the formula, T ij This represents the torque of the four drive wheels, where i represents the front wheel (f) and j represents the rear wheel (r), and j represents the left wheel (l) and j represents the right wheel (r). M z R is the yaw moment calculated in step 5; F is the tire radius; z For the total vertical load, F zij The vertical load on the four drive wheels; δ is the front wheel steering angle; d f d is the front wheel track. r This refers to the rear wheel track.

Images