Data-driven mountainous leaf blade transfer vehicle rollover prevention warning method and system

Abstract

The present application provides a kind of mountainous blade transfer car anti-rollover early warning method and system based on data driving, it is related to vehicle anti-rollover risk early warning technical field, it includes the following steps: S1, joint simulation generates neural network training data set;S2, constructs RBF neural network training as vehicle spring load parameter estimation model;S3, the parameter of RBF neural network is optimized using improved grey wolf optimization algorithm;S4, RBF neural network is optimized and based on the spring load prediction of optimized RBF neural network;S5, the state information of vehicle driving process is obtained by vehicle sensor, and the spring load distribution of vehicle is obtained and early warning signal is generated;S6, build the anti-rollover visualization platform of wind power blade mountainous transfer car based on ROS, and carry out anti-rollover early warning.The method of the present application carries out real-time vehicle anti-rollover early warning according to the parameters in the process of vehicle driving, and improves the safety in the process of vehicle transportation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle rollover risk early warning, in particular to a mountainous blade transport vehicle rollover early warning method and system based on data driving. BACKGROUND

[0002] With the increasing demand for energy, new green energy development shows a significant growth trend, among which wind power is an important source of new green energy, and the layout of wind turbines gradually extends from the coast to inland mountainous areas. The transportation of wind power devices is an important part of wind turbine installation, and the safety consideration during transportation is very critical. Due to the characteristics of large volume and mass of wind power devices, combined with the winding and tortuous roads in mountainous areas, there is a risk of rollover at any time during transportation, causing irreparable economic loss. Therefore, the rollover early warning of mountainous wind power blade transport vehicles is an important problem that needs to be solved to ensure the safety of goods and operating personnel during transportation.

[0003] The application number 201811296526.3 discloses a vehicle rollover control method, which obtains the air spring load value by obtaining the sensor value change of the electric control air suspension to realize rollover early warning. The application number 202311207832.6 discloses a vehicle rollover early warning method based on adaptive unscented Kalman filtering, which estimates the distance from the center of mass to the roll axis using unscented Kalman filtering, and then judges the rollover degree of the vehicle according to the change of lateral load to realize rollover early warning. The application number 202310668601.9 discloses a heavy truck rollover early warning method based on vehicle-road cooperation, which calculates the rollover degree and maximum safe speed of the heavy truck to achieve rollover early warning effect.

[0004] Currently, there are few studies on vehicle rollover early warning for oversized transportation. The vehicle rollover early warning mainly focuses on conventional vehicles, and some studies focus on heavy vehicle rollover early warning, but most of them are still in the theoretical stage and have few practical solutions. At the same time, the current vehicle rollover early warning method is mainly based on physical model method, which has high requirements for the accuracy of model establishment and the accurate identification of vehicle parameters. SUMMARY

[0005] In order to solve the above-mentioned deficiencies of the prior art, the purpose of the present application is to provide a data-driven mountain leaf blade transfer vehicle anti-rollover warning method and system, which uses sensors arranged on the vehicle to collect parameters during vehicle driving, such as vehicle longitudinal speed, front wheel steering angle, yaw angular velocity, lateral acceleration, roll angle, roll angular velocity, etc., based on improved and optimized RBF neural network training, obtains the spring load distribution of the vehicle through the state information of the vehicle during driving obtained by the vehicle-mounted sensor, obtains the rollover time TTR, and generates a rollover warning in combination with the rollover preparation. The anti-rollover warning method has the advantages of real-time, accuracy and easy implementation, and can more accurately evaluate the rollover risk of the vehicle.

[0006] Specifically, the present application provides a data-driven mountain leaf blade transfer vehicle anti-rollover warning method, characterized in that it comprises the following steps:

[0007] S1, joint simulation generates a data set;

[0008] S2, construct an RBF neural network training as a vehicle spring load parameter estimation model, determine the vehicle speed v x , the front wheel steering angle δ1, the trailer rear wheel active steering angle δ2 and the upper operation mechanism rotation angle θ four variables as the input of the prediction model; determine the spring load F i of each wheel of the vehicle as the output of the prediction model;

[0009] S3, optimize the parameters of the RBF neural network using the improved grey wolf optimization algorithm, define the levels as α, β, γ and η, and the levels decrease in turn, wherein α represents the optimal solution, β, γ and η represent the second solution, the third solution and the fourth solution, respectively;

[0010] In the iterative solution process, an adaptive convergence factor a is introduced to solve the problem of linear reduction of the convergence factor in the iterative process:

[0011]

[0012] Where rand represents a random number between 0 and 1; t represents the current iteration number, and T represents the maximum iteration number;

[0013] The optimal solution X(t+1) is optimized by introducing a Gauchy distribution function, and the optimized expression is:

[0014]

[0015] Where X1, X2 and X3 are the position vectors of the optimal solution α, the second solution β and the third solution γ, respectively;

[0016] S4, optimizing the RBF neural network and performing spring-on load prediction based on the optimized RBF neural network, specifically comprising the following sub-steps:

[0017] S41, determining the number of neurons of the hidden layer:

[0018]

[0019] Wherein, τ1, τ2 represents the number of neurons of the input layer, q∈[1, 20];

[0020] Represent the number of hidden neurons;

[0021] S42, data preprocessing, dividing the data into training samples and test samples and performing standardization processing;

[0022] S43, setting the parameters of grey wolf optimization algorithm, including the maximum number of iterations and the number of grey wolves;

[0023] S44, calculating the fitness value of each grey wolf, and sorting the grey wolves according to the fitness value;

[0024] Step S45, updating the parameters based on step S3 until the iteration is completed, and optimizing the parameters of the network according to the above parameters;

[0025] S46, performing spring-on load prediction based on the optimized RBF neural network;

[0026] S5, obtaining the state information of the vehicle during driving through the vehicle-mounted sensor to obtain the spring-on load distribution of the vehicle, obtaining the rollover time TTR, and generating a rollover warning signal based on the rollover time TTR and the rollover threshold value, wherein the rollover threshold value is calculated as:

[0027]

[0028] Wherein, k i Indicates the weight coefficient θ represents the turning angle of the upper working device, φ1, Indicates the yaw angle of the transport chassis tractor and trailer;

[0029] S6, building a ROS-based wind power blade mountain transport vehicle anti-rollover visualization platform to perform anti-rollover warning.

[0030] Preferably, in step S1, S1 utilizes Recurdyn multi-body dynamics simulation software and TruckSim vehicle dynamics simulation software to jointly simulate and generate a neural network training data set.

[0031] Preferably, step S1 specifically comprises the following sub-steps:

[0032] S11, build a 3D model of the wind turbine blade mountain transport vehicle upper loading mechanism in Recurdyn multi-body dynamics simulation software according to real vehicle parameters;

[0033] S12, build a wind turbine blade mountain transport vehicle dynamics model of the semi-trailer tractor vehicle in TruckSim software according to real vehicle parameters;

[0034] S13, build Recurdyn & TruckSim & MATLAB / Simulink three-party joint simulation environment in MATLAB / Simulink software;

[0035] MATLAB / Simulink three-party joint simulation environment;

[0036] S14, send the rotation angle information of the upper loading mechanism to Recurdyn through Simulink, and send the vehicle control parameters to TruckSim through Simulink;

[0037] S15, Recurdyn executes the action according to the received rotation angle information of the loading mechanism, generates the corresponding rotation angle, calls the output interface to send the load parameters of the calibration point to TruckSim, and TruckSim outputs the corresponding vehicle spring load value according to the received information;

[0038] S16, generate a data set for vehicle spring load parameter estimation according to the data of steps S11-S15;

[0039] S17, filter the data set generated in step S16 to finally generate a data set.

[0040] Preferably, step S2 specifically comprises the following steps:

[0041] S21, establish a nonlinear network architecture describing the change of the spring load as a vehicle spring load parameter estimation model, the nonlinear network architecture adopts a three-layer structure, the three-layer structure includes an input layer, a hidden layer and an output layer, the input vector is nonlinearly mapped to a hidden space through an activation function of the hidden layer, and the output of the prediction model is a linear weighted sum of the hidden layer neuron outputs;

[0042] S22, determine the input and output of the vehicle spring load parameter estimation model;

[0043] S23, select a Gaussian function as the activation function, obtain the estimated spring load value by using the approximate output of the RBF neural network, and further calculate the actual spring load value:

[0044]

[0045] wherein, denotes an estimated spring load value, c i and σ i denotes the center and width of the neuron, ω denotes the weight between the output and hidden layers, denotes an actual spring load value.

[0046] Preferably, step S3 specifically comprises the following sub-steps:

[0047] S31, define the boundary prey, during the hunting process, the behavior of the hunting wolf is defined as follows:

[0048] D = |C · X P (t) - X(t) |;

[0049] The position update formula of the gray wolf is:

[0050] X(t+1) = X P (t) - A · D

[0051] Wherein, D represents the distance between the individual and the prey, t represents the current iteration number, A and C are coefficient vectors, X P and X are the position vectors of the prey and the gray wolf respectively, and the calculation formula of A and C is as follows:

[0052] A = 2a · r1-a

[0053] C = 2 · r2

[0054]

[0055] Wherein, a is a convergence factor, which decreases linearly from 2 to 0 with the increase of the iteration number, r1 and r2 are random numbers between 0 and 1;

[0056] S32, when the gray wolf determines the position of the prey, β and γ respectively surround the prey under the guidance of α, and the mathematical model describing the gray wolf individual tracking the prey position is described as follows:

[0057] D α = |C1 · X α -X|

[0058] D β = |C2 · X β -X|

[0059] D γ = |C3 · X γ -X|

[0060] X1 = X α -A1 · D α

[0061] X2 = X β -A2 · Dβ

[0062] X3 = X γ -A3 · D γ

[0063]

[0064] wherein D α , D β , D γ respectively represent the distance between α, β, γ and other individuals; X α , X β and X γ represent the current position vector of α, β, γ; C1, C2, C3 are random vectors; X is the current position of the grey wolf;

[0065] S33, in the iteration process, when the value of a is linearly reduced from 2 to 0, the corresponding a changes from -a to a, when |a| < 1, the wolf attacks the prey; when |A| > 1, the grey wolf is separated from the prey.

[0066] Preferably, the standardization expression in step S42 is:

[0067]

[0068] wherein X k represents the Kth feature vector;

[0069] In step S44, the expression of the fitness function is as follows:

[0070]

[0071] wherein y(i), respectively represent the actual value and the predicted value; i represents the ith sample, and n represents the total number of samples.

[0072] Preferably, step S5 specifically comprises the following sub-steps:

[0073] Step S51, vehicle rollover index selection is performed, and the lateral load transfer ratio LTR of the vehicle is defined as:

[0074]

[0075] Step S52, the vehicle spring load parameter estimation model after training is used to return the vehicle rollover time TTR according to the state information of the vehicle;

[0076] Step S53, generating a rollover early warning signal based on the rollover time TTR and the rollover threshold value, setting an alarm rollover time T=T1*epsilon, T1 is set according to requirements, and epsilon is the rollover threshold value, when the rollover time TTR is less than or equal to T, outputting a rollover early warning risk.

[0077] Preferably, step S6 specifically comprises the following sub-steps:

[0078] S61, establishing a three-dimensional model of the wind turbine blade mountain transport vehicle by using 3D modeling software;

[0079] S62, constraining and constructing the three-dimensional model in the ROS space to establish a multi-degree-of-freedom three-dimensional model of the wind turbine blade mountain transport vehicle;

[0080] S63, collecting signals by using a vehicle-mounted sensor and inputting the collected signals into a vehicle-mounted visual embedded system;

[0081] S64, reading corresponding signals by using a ROS node, inputting the signals into an interface for controlling the motion of the three-dimensional model, synchronizing the pose information of the vehicle in real time in the ROS space, and displaying on the screen;

[0082] S65, displaying the state information of the vehicle and the parameter change information of the on-spring load on the screen;

[0083] S66, according to the set threshold value, the indicator light of the embedded system will display different colors, green represents safety, yellow represents potential rollover risk, and red represents imminent rollover;

[0084] S67, when there is a rollover risk, the light of the vehicle-mounted embedded system flashes rapidly, and the buzzer emits an alarm sound, achieving the effect of rollover early warning.

[0085] Preferably, the vehicle control parameters in step S14 include vehicle speed and steering wheel angle.

[0086] In another aspect, the present application provides a warning system for the above-mentioned data-driven mountain blade transport vehicle rollover early warning method, which comprises a data set generation unit, a vehicle on-spring load parameter estimation model generation unit, an RBF neural network parameter optimization unit, an on-spring load prediction unit, a rollover early warning signal generation unit, and a rollover early warning unit.

[0087] The data set generation unit is used for joint simulation to generate a data set; the vehicle spring-on load parameter estimation model generation unit is used for constructing RBF neural network training as a vehicle spring-on load parameter estimation model; the RBF neural network parameter optimization unit is used for optimizing the parameters of the RBF neural network by using an improved grey wolf optimization algorithm; the spring-on load prediction prediction unit is used for optimizing the RBF neural network and performing spring-on load prediction based on the optimized RBF neural network; the rollover early warning signal generation unit is used for obtaining the spring-on load distribution of the vehicle by acquiring state information of the vehicle in the driving process through the vehicle-mounted sensor and generating an early warning signal; and the rollover early warning unit is used for building a ROS-based wind power blade mountain transport vehicle anti-rollover visualization platform to perform anti-rollover early warning.

[0088] Compared with the prior art, the method has the following beneficial effects:

[0089] (1) The method of the present application uses sensors arranged on the vehicle to collect parameters in the driving process of the vehicle, such as the longitudinal speed of the vehicle, the front wheel angle, the yaw angular velocity, the lateral acceleration, the roll angle, the roll angular velocity, etc., based on the improved and optimized RBF neural network training, the state information of the vehicle in the driving process is acquired through the vehicle-mounted sensor to obtain the spring-on load distribution of the vehicle, the rollover time TTR is obtained, and the rollover early warning is generated in combination with the rollover preparation. The anti-rollover early warning method has the advantages of real-time, accuracy and easy implementation, and can more accurately evaluate the rollover risk of the vehicle.

[0090] (2) The method of the present application uses Recurdyn multi-body dynamics simulation software and TruckSim vehicle dynamics simulation software to jointly simulate to generate a neural network training data set, which can ensure the authenticity and diversity of the training data and further improve the prediction accuracy.

[0091] (3) The method of the present application uses the improved grey wolf optimization algorithm (IGWO) to optimize the parameters of the RBF neural network to improve the estimation accuracy of the model, and uses the optimized RBF neural network to predict the vehicle load, which can greatly improve the prediction accuracy and ensure the accuracy of the rollover early warning. BRIEF DESCRIPTION OF DRAWINGS

[0092] Figure 1 The method flowchart of the present application is shown in the figure;

[0093] Figure 2 The overall flowchart of the method of the present application is shown in the figure;

[0094] Figure 3 The data generation method of the present application is shown in the figure;

[0095] Figure 4 The neural network structure of the present application is shown in the figure;

[0096] Figure 5 The schematic diagram of the vehicle-mounted embedded system structure of the present application. DETAILED DESCRIPTION

[0097] Hereinafter, the embodiments of the present application will be described with reference to the accompanying drawings.

[0098] As Figure 2 shown, the present application provides a data-driven mountain blade transfer vehicle anti-rollover warning method. It uses Recurdyn multi-body dynamics simulation software and TruckSim vehicle dynamics simulation software to jointly simulate the generation of neural network training data set; through the improved RBF neural network training, the vehicle spring load parameter estimation model is generated; the parameter estimation model is deployed to the vehicle, the state information of the vehicle in the driving process is obtained through the sensor information of the vehicle, the spring load distribution of the vehicle is obtained, and the rollover time TTR is calculated, and according to the sensor parameters, the pose of the vehicle and the load distribution information are generated in the visual embedded system in the cockpit, thereby playing the effect of anti-rollover warning.

[0099] Specifically, as Figure 1 shown, the steps of the data-driven mountain blade transfer vehicle anti-rollover warning method of the present application are as follows:

[0100] S1, data set generation, as Figure 3 shown, specifically includes the following steps:

[0101] S11, according to the real vehicle parameters, a 3D model of the wind power blade mountain transfer vehicle upper loading operation mechanism is built in the Recurdyn multi-body dynamics simulation software.

[0102] S12, according to the real vehicle parameters, a wind power blade mountain transfer vehicle transportation chassis semi-trailer vehicle dynamics model is built in the TruckSim software.

[0103] S13, Recurdyn&TruckSim

[0104] MATLAB / Simulink three-party joint simulation environment.

[0105] S14, through Simulink, the slewing angle information of the upper loading operation mechanism is sent to Recurdyn; the speed, steering wheel turning angle and other control parameters of the vehicle are sent to TruckSim.

[0106] S15, Recurdyn executes actions according to the received working mechanism rotation angle information, generates the corresponding rotation angle, and calls the output interface to send the load parameters of the calibration point position to TruckSim; TruckSim outputs the corresponding whole vehicle spring load value according to the received information.

[0107] S16, generate a data set for vehicle spring load parameter estimation according to the data of steps S11-S15.

[0108] S17, filter the data set generated in step S16 to finally generate a training set that can be used for neural network training.

[0109] S2, construct an RBF neural network structure, as shown in Figure 4 , the specific steps include:

[0110] S21, establish a nonlinear network architecture to describe the change of spring load. A three-layer structure is adopted, including an input layer, a hidden layer, and an output layer. The input vector is nonlinearly mapped to the hidden space through the activation function of the hidden layer. The output of the prediction model is the linear weighted sum of the hidden layer neuron output.

[0111] S22, select four variables as the input of the network: vehicle speed v x , front wheel steering angle δ1, trailer rear wheel active steering angle δ2, and upper working mechanism rotation angle θ; select the output of the network as the spring load F i of each wheel of the vehicle.

[0112] S23, select the Gaussian function as the activation function, and the approximate output of the RBF neural network is:

[0113]

[0114] wherein, represents the output of the hidden neuron, c i and σ i represent the center and width of the neuron, and ω represents the weight between the output and the hidden layer.

[0115] S3, use the improved grey wolf optimization algorithm (IGWO) to optimize the parameters of the RBF neural network to improve the estimation accuracy of the model. There is a strict hierarchical distribution relationship between wolves, and the social hierarchy is divided into α, β, γ and η, with the hierarchy decreasing in turn. Among them, α is the head wolf, representing the solution with the highest fitness value, β, γ and η represent the second, third and other solutions respectively. The specific steps include the following sub-steps:

[0116] S31, define the boundary prey:

[0117] In the hunting process, the behavior of the hunting wolf is defined as follows:

[0118] D = |C · X P (t)-X(t)|.

[0119] The position update formula of the gray wolf is:

[0120] X(t+1) = X P (t)-A·D

[0121] Where D represents the distance between the individual and the prey, t represents the current iteration number, A and C are coefficient vectors, X P and X are the position vectors of the prey and the gray wolf respectively. The calculation formula of A and C is as follows:

[0122] A = 2a·r1-a

[0123] C = 2·r2

[0124]

[0125] Where a is the convergence factor, which decreases linearly from 2 to 0 as the iteration number increases, r1 and r2 are random numbers between 0 and 1.

[0126] S32, hunting:

[0127] The wolf can identify the prey and surround them, when the gray wolf determines the position of the prey, β and γ surround the prey under the guidance of α respectively, the mathematical model of the gray wolf individual tracking the prey position is described as follows:

[0128] D α = |C1·X α -X|

[0129] D β = |C2·X β -X|

[0130] D γ = |C3·X γ -X|

[0131] X1 = X α -A1·D α

[0132] X2 = X β -A2·D β

[0133] X3 = X γ -A3·D γ

[0134]

[0135] Where D α, D β , D γ respectively represent the distance between a, b, g and other individuals; X α , X β and X γ represent the current position vector of a, b, g; C1, C2, C3 are random vectors; X is the current position of the grey wolf

[0136] The adaptive convergence factor a is introduced to solve the problem of linear reduction of convergence factor in the iteration process

[0137]

[0138] where rand represents a random number between 0 and 1.

[0139] The Gauchy distribution function is introduced to optimize the optimal solution X(t+1), and the optimized expression is:

[0140]

[0141] S33, attacking prey:

[0142] When the prey stops moving, the grey wolf completes the hunting process by attacking. In order to simulate the process of approaching the prey, the decrease of a is gradually controlled. At this time, the fluctuation range of a is correspondingly reduced. That is, during the iteration process, when the value of a decreases linearly from 2 to 0, the corresponding a also changes from -a to a. When |a|<1, the wolf attacks the prey. When |A|>1, the grey wolf and the prey are separated, which enhances the global search ability of the algorithm and finds the most suitable prey.

[0143] S4, spring load estimation based on improved RBF neural network

[0144] S41, determine the number of neurons in the hidden layer

[0145]

[0146] where τ1τ2 represents the number of neurons in the input layer, q∈[1,20].

[0147] S42, data preprocessing, divide the data into training samples and test samples, and the standardization expression is:

[0148]

[0149] where X k represents the Kth feature vector.

[0150] S43, set the parameters of grey wolf optimization algorithm, including the maximum number of iterations and the number of grey wolves.

[0151] S44, calculate the fitness value of each gray wolf, and sort the gray wolves according to the fitness value. The expression of the fitness function is as follows:

[0152]

[0153] wherein, respectively represent the actual value and the predicted value.

[0154] S45, update the parameters according to step S3 until the iteration is completed, and optimize the parameters of the network according to the above parameters.

[0155] S5, calculate the rollover warning time of the wind turbine blade transfer vehicle based on the change of the spring load:

[0156] S51, select the vehicle rollover index, and define the lateral load transfer ratio LTR of the vehicle as:

[0157]

[0158] S52, use the trained vehicle spring load parameter estimation model to return the vehicle rollover time TTR according to the state information of the vehicle.

[0159] S53, generate a rollover warning signal based on the rollover time TTR and the rollover threshold, set the alarm rollover time T=T1*ε, T1 is set according to the demand, and ε is the rollover threshold, when the rollover time TTR is less than or equal to T, output the rollover warning risk.

[0160] S6, establish a ROS-based wind turbine blade mountain transfer vehicle anti-rollover visualization platform, the visualization platform is as shown in Figure 5 , and specifically includes the following steps:

[0161] S61, use 3D modeling software to establish a three-dimensional model of the wind turbine blade mountain transfer vehicle.

[0162] S62, constrain the three-dimensional model in the ROS space to realize the movable multi-degree-of-freedom wind turbine blade mountain transfer vehicle three-dimensional model.

[0163] S63, collect signals through vehicle-mounted sensors and input them into the vehicle-mounted visualization embedded system.

[0164] S64, read the corresponding signals through the ROS node, input them into the interface for controlling the motion of the three-dimensional model, synchronize the pose information of the vehicle in real time in the ROS space, and display on the screen.

[0165] S65, display the state information of the vehicle and the parameter change information of the spring load on the screen.

[0166] S66, according to the set threshold, the indicator light of the embedded system will display different colors, green represents safety, yellow represents potential rollover risk, and red represents imminent rollover.

[0167] S67, when there is a rollover risk, the light of the vehicle embedded system flashes rapidly, and the buzzer emits an alarm sound, achieving the effect of rollover warning.

[0168] In another aspect, the present application provides a warning system for the above-mentioned data-driven mountain blade transport vehicle rollover warning method, which comprises a data set generation unit, a vehicle spring load parameter estimation model generation unit, an RBF neural network parameter optimization unit, a spring load prediction unit, a rollover warning signal generation unit and a rollover warning unit.

[0169] The data set generation unit is used to generate a data set jointly; the vehicle spring load parameter estimation model generation unit is used to construct an RBF neural network training as a vehicle spring load parameter estimation model; the RBF neural network parameter optimization unit is used to optimize the parameters of the RBF neural network by using the improved grey wolf optimization algorithm; the spring load prediction unit is used to optimize the RBF neural network and predict the spring load based on the optimized RBF neural network; the rollover warning signal generation unit is used to obtain the spring load distribution of the vehicle by acquiring the state information of the vehicle during driving through the vehicle-mounted sensor and generate a warning signal; the rollover warning unit is used to build a ROS-based wind power blade mountain transport vehicle rollover prevention visualization platform for rollover prevention warning.

[0170] Specific example application:

[0171] First, according to the steps shown in the invention, install the required sensors on the transport vehicle, and read the corresponding data through the display screen of the embedded system to determine whether the sensor installation is normal. When the vehicle is running, the sensor normally reads the vehicle state information. When the vehicle encounters a possible rollover scene, such as driving on a rugged mountain road, in order to avoid possible collision obstacles, the vehicle-mounted working mechanism rotates to avoid, generating a turning angle. In the driver's cabin, the driver monitors the vehicle body posture in real time in the screen of the embedded system, and finds that the model in the picture also produces corresponding animation effects with the actual action of the vehicle. The vehicle continues to travel, and when the vehicle body tilts at a certain section, the driver displays the data returned by the screen, the vehicle body load distribution is abnormal, the left wheel load is significantly greater than the right wheel load, and according to the signal detected by the sensor, the vehicle-mounted embedded system indicator light displays yellow, indicating that there is a rollover risk, the actual TTR of the rollover is 30 seconds, the driver receives the corresponding rollover signal, and immediately takes action to adjust the vehicle pose, and after the execution is completed, the signal light returns to green, the load value returns to normal, and the rollover risk is eliminated.

[0172] The above described embodiments are only to illustrate the preferred embodiments of the present application, and are not intended to limit the scope of the present application. Any modification and improvement of the technical solutions of the present application made by those skilled in the art without departing from the design spirit of the present application shall fall within the protection scope of the present application.

Claims

1. A data-driven based anti-rollover warning method for a mountainous leaf blade transfer vehicle, characterized in that: It comprises the following steps: S1, joint simulation generates a data set; S2, construct RBF neural network training as a vehicle on-spring load parameter estimation model, determine the vehicle speed , front wheel steering angle , trailer rear wheel active steering angle , and the upper operation mechanism rotation angle Four variables as the input of the prediction model; determine the on-spring load of each wheel of the vehicle The output of the prediction model; Step S2 specifically comprises the following steps: S21, a nonlinear network architecture describing the change of the load on the spring is established as a vehicle load-on-spring parameter estimation model, the nonlinear network architecture adopts a three-layer structure, the three-layer structure includes an input layer, a hidden layer and an output layer, an input vector is nonlinearly mapped to a hidden space through an activation function of the hidden layer, and an output of the prediction model is a linear weighting sum of neuron outputs of the hidden layer; S22, the input and output of the vehicle load-on-spring parameter estimation model are determined; S23, a Gaussian function is selected as the activation function, an estimated load-on-spring value is obtained by using an approximate output of the RBF neural network, and an actual load-on-spring value is further calculated: ； wherein, represents an estimated on-spring load value, and represents the center and width of a neuron, represents the weights between the output and hidden layers, represents an actual on-spring load value; S3, the parameters of the RBF neural network are optimized by using the improved grey wolf optimization algorithm, and the definition levels are 、 、 and , the levels decrease in turn, wherein represents the optimal solution, 、 、 respectively represent the second solution, the third solution and the fourth solution; Introducing adaptive convergence factor in iterative solution process Solving the problem of linear reduction of convergence factor in iterative process: ； wherein represents a random number between 0 and 1 ; t represents the current iteration number, and T represents the maximum iteration number. Introduction Distribution function to optimal solution Optimization, the optimized expression is: ； wherein , , are position vectors of the optimal solution , the second solution and the third solution , respectively; S4, the RBF neural network is optimized, and the load-on-spring is predicted based on the optimized RBF neural network, specifically comprising the following sub-steps: S41, the number of neurons of the hidden layer is determined: ； wherein, , represents the number of neurons of the input layer, ; represents the number of hidden neurons; S42, data preprocessing, the data is divided into training samples and test samples and is subjected to standardization processing; S43, parameters of the grey wolf optimization algorithm are set, including the maximum number of iterations and the number of grey wolves; S44, the fitness value of each grey wolf is calculated, and the grey wolves are sorted according to the fitness value; S45, the parameters are updated based on step S3 until the iteration is completed, and the parameters of the network are optimized according to the above parameters; S46, the load-on-spring is predicted based on the optimized RBF neural network; S5, the state information of the vehicle in the driving process is obtained by the vehicle-mounted sensor to obtain the load-on-spring distribution of the vehicle, the rollover time TTR is obtained, the rollover early warning signal is generated based on the rollover time TTR and the rollover threshold, and the rollover threshold calculation method is: ； wherein denotes a weight coefficient, denotes the upper working device corner, , denotes the yaw angle of the transport chassis tractor and trailer; S6, a rollover prevention visualization platform of the wind power blade mountain transport vehicle based on ROS is constructed, and the rollover prevention early warning is performed.

2. The data-driven based anti-rollover early warning method for mountainous terrain blade transport vehicle according to claim 1, characterized in that: In step S1, a data set is generated by joint simulation of Recurdyn multi-body dynamics simulation software and TruckSim vehicle dynamics simulation software.

3. The data-driven based anti-rollover early warning method for mountainous terrain blade transport vehicle according to claim 2, characterized in that: Step S1 specifically comprises the following sub-steps: S11, a three-dimensional model of the wind power blade mountain transport vehicle loading operation mechanism is built in the Recurdyn multi-body dynamics simulation software according to the real vehicle parameters; S12, a vehicle dynamics model of the wind power blade mountain transport vehicle transport chassis semi-trailer tractor is built in the TruckSim software according to the real vehicle parameters; S13, a Recurdyn & TruckSim & MATLAB / Simulink joint simulation environment is built in the MATLAB / Simulink software; S14, the rotation angle information of the loading operation mechanism is sent to Recurdyn through Simulink, and the vehicle control parameters are sent to TruckSim through Simulink; S15, Recurdyn executes the action according to the received operation mechanism rotation angle information, generates the corresponding rotation angle, calls the output interface to send the load parameters of the calibration point position to TrcukSim, and TruckSim outputs the corresponding vehicle load-on-spring value according to the received information; S16, generate a data set for vehicle spring load parameter estimation according to the data of steps S11-S15; S17, filter the data set generated in step S16 to finally generate a training data set for neural network training.

4. The data-driven based anti-rollover early warning method for mountainous terrain blade transport vehicle according to claim 1, characterized in that: Step S3 specifically includes the following sub-steps: S31, define the boundary prey, and the behavior of the hunting wolf is defined as follows during the hunting process: ； The position update formula of the gray wolf is: ； wherein, denotes the distance between the individual and the prey, denotes the current iteration number, and is a coefficient vector, and are the position vectors of the prey and the grey wolf, respectively, and are calculated as follows: ； ； ； wherein, is a convergence factor that linearly decreases from 2 to 0 as the number of iterations increases, , is a random number between [0, 1]; S32, when the gray wolf determines the position of the prey, and respectively under the guidance of surround the prey, the mathematical model of the gray wolf individual tracking the prey position is described as follows: ； ； ； ； ； ； ； wherein , , respectively represent , , distance from other individuals; , and represent , , current position vector; , , is a random vector; is the current position of the grey wolf; S33, in the iteration process, when The value decreases linearly from 2 to 0, and the corresponding From To When The wolf attacks the prey; when The gray wolf is separated from the prey.

5. The data-driven based anti-rollover early warning method for mountainous terrain blade transport vehicle according to claim 1, characterized in that: The standardization expression in step S42 is: ； wherein, represents the Kth eigenvector; In step S44, the expression of the fitness function is as follows: ； wherein, respectively represent actual and predicted values; i represents the i-th sample, and n represents the total number of samples.

6. The data-driven based anti-rollover early warning method for mountainous terrain blade transport vehicle according to claim 1, wherein: Step S5 specifically includes the following sub-steps: Step S51, vehicle rollover index selection, the lateral load transfer ratio LTR of the vehicle is defined as: ； Step S52, using the trained vehicle spring load parameter estimation model, the vehicle rollover time TTR is returned according to the state information of the vehicle; Step S53, generating a rollover warning signal based on the rollover time TTR and the rollover threshold, setting the warning rollover time , According to the demand setting, The rollover threshold, when the rollover time TTR is less than or equal to T, output the rollover warning risk.

7. The data-driven based anti-rollover early warning method for mountainous terrain blade transport vehicle according to claim 1, wherein: Step S6 specifically includes the following sub-steps: S61, a 3D modeling software is used to establish a 3D model of the wind power blade mountain transport vehicle; S62, constraint construction is performed on the 3D model in the ROS space to establish a multi-degree-of-freedom 3D model of the wind power blade mountain transport vehicle; S63, signals are collected through vehicle-mounted sensors and input into the vehicle-mounted visual embedded system; S64, the corresponding signals are read through the ROS node, the signals are input into the interface for controlling the motion of the 3D model, the pose information of the vehicle is synchronized in real time in the ROS space, and the signals are displayed on the screen; S65, the state information of the vehicle and the parameter change information of the spring load are displayed on the screen; S66, according to the set threshold value, the indicator light of the embedded system will display different colors, green represents safety, yellow represents potential rollover risk, and red represents imminent rollover; S67, when there is a rollover risk, the light of the vehicle-mounted embedded system flashes rapidly and the buzzer emits an alarm sound, achieving the effect of rollover warning.

8. The data-driven based anti-rollover early warning method for mountainous terrain blade transport vehicle of claim 3, wherein: The vehicle control parameters in step S14 include vehicle speed and steering wheel angle.

9. A warning system for the data-driven based anti-rollover warning method of mountainous terrain blade transport vehicle of claim 1, characterized in that: It includes a data set generation unit, a vehicle spring load parameter estimation model generation unit, an RBF neural network parameter optimization unit, a spring load prediction unit, a rollover warning signal generation unit, and a rollover warning unit; The data set generation unit is used to generate a data set through joint simulation; the vehicle spring load parameter estimation model generation unit is used to construct an RBF neural network training as a vehicle spring load parameter estimation model; the RBF neural network parameter optimization unit is used to optimize the parameters of the RBF neural network using the improved gray wolf optimization algorithm; the spring load prediction unit is used to optimize the RBF neural network and predict the spring load based on the optimized RBF neural network; The rollover warning signal generation unit is used to obtain the state information of the vehicle during driving through vehicle-mounted sensors to obtain the spring load distribution of the vehicle and generate a warning signal; the rollover warning unit is used to build a ROS-based wind power blade mountain transport vehicle anti-rollover visualization platform for anti-rollover warning.

Images