Method for realizing strip mine vehicle trajectory prediction through combination of random forest and Kalman

Abstract

The invention relates to the technical field of unmanned driving in strip mine areas, in particular to a method for realizing strip mine vehicle trajectory prediction by combining random forest and Kalman, which comprises the following steps: acquiring current motion state information of a vehicle in a mine area by sensing equipment; obtaining a candidate reference path based on the current motion state information and prior strip mine area high-precision map information, and obtaining the path curvature of the candidate reference path; the current motion state information and the path curvature are processed based on a random forest, and observed values of the longitudinal speed and the yaw angle speed of the reference path are obtained; establishing a state transition model and a measurement model of extended Kalman filtering, and taking observation values of the longitudinal velocity and the yaw velocity as measurement values of the extended Kalman filtering; carrying out state prediction and measurement updating of extended Kalman filtering, and finally obtaining a prediction track of the mining area vehicle in future time; the strip mine area vehicle trajectory prediction performance can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of driverless technology for open-pit mining areas, and particularly relates to a method for realizing the trajectory prediction of open-pit mining vehicles by combining random forest and Kalman filter. Background Art

[0002] The unmanned transportation in open-pit mines, as a key part of the construction of intelligent transportation systems, is not only one of the best scenarios for the wide application of driverless technology, but also an important driving force for promoting the development of the unmanned transportation industry. The driving safety of driverless mining trucks in open-pit mines is particularly crucial and is the core link to ensure the safe and efficient operation of the entire mine's unmanned transportation system. Trajectory prediction for vehicles in open-pit mining areas aims to predict the driving trajectories of other manned vehicles around driverless vehicles in the mine road scene, providing accurate predicted trajectory information for driverless vehicles to support downstream modules such as traffic decision-making, obstacle avoidance planning, and control execution.

[0003] The patent application with publication number CN118953406A discloses a vehicle trajectory prediction method based on an attention mechanism, and the patent with publication number CN114495036B discloses a method and system for predicting the trajectory of an autonomous driving vehicle based on a double-cross Transformer. The above technical solutions include, after obtaining the historical data of the target vehicle and surrounding vehicles, through deep learning methods, feature encoding of the interaction behavior between the ego vehicle and other vehicles in the scene, and also include encoding of vehicle-lane interaction features, vehicle trajectory feature encoding, and global environment feature encoding. These features need to be learned through a neural network model based on a large amount of scene data. Such methods are mainly applied to urban road environments for trajectory prediction of complex multimodal interactions of a large number of passenger vehicles. Directly applying these methods to trajectory prediction in open-pit mining areas will face significant adaptability challenges. One of the main reasons is that in urban road scenes, there are clear lane line information with constraints, while there is no clear lane constraint for vehicle operation on open-pit mining area roads. How to effectively utilize the mine map information to improve the trajectory prediction accuracy of the target vehicle is a difficult point to solve. Another main reason is that data-driven prediction methods usually rely on a large amount of labeled scene data for training and learning, resulting in high computational consumption and long required time for real-time calculation. At present, there is no publicly available motion trajectory dataset for use in mining areas. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method for realizing the trajectory prediction of open-pit mining vehicles by combining random forest and Kalman filter, which solves the technical problems in the prior art that vehicle trajectory prediction depends on labeled data and consumes a large amount of computing power.

[0005] The present invention provides a method for realizing the trajectory prediction of open-pit mine vehicles by combining random forest and Kalman, including the following steps:

[0006] Step S1: Obtain the current motion state information of the mining area vehicles by the sensing device, where the motion state information includes the position, heading angle, longitudinal speed, longitudinal acceleration and yaw angular velocity of the mining area vehicles;

[0007] Step S2: Based on the current motion state information and the prior high-precision map information of the open-pit mining area, obtain the candidate reference paths, and obtain the path curvature of the candidate reference paths, where the candidate reference paths are the paths that the mining area vehicles may travel in the future;

[0008] Step S3: Process the current motion state information and the path curvature based on the random forest to obtain the observed values of the longitudinal speed and yaw angular velocity of the mining area vehicles;

[0009] Step S4: Establish the state transition model and measurement model of the extended Kalman filter, and use the observed values of the longitudinal speed and yaw angular velocity as the measured values of the longitudinal speed and yaw angular velocity of the extended Kalman filter;

[0010] Perform state prediction and measurement update of the extended Kalman filter, and finally obtain the predicted trajectory of the mining area vehicles in the future.

[0011] Preferably, in the step S1, the motion state information of the target vehicle at the current moment t is expressed as: x t =[x t ,y t ,θ t ,v t ,a t ,ω t T , where x t ,y t ,θ t ,v t ,a t ,ω t respectively represent the eastward position coordinate, northward position coordinate, heading angle, longitudinal speed, longitudinal acceleration and yaw angular velocity at the current moment t.

[0012] Preferably, the step S2 specifically includes:

[0013] Step S2-1: According to the current motion state information and the prior high-precision map information of the open-pit mining area, obtain the possible candidate reference path set SET(Path i ), where Path i represents the i-th path;

[0014] ​Step S2-2: Divide the preset future time period into multiple time steps, denoted as t + kΔt, where t is the current time, Δt is the time step size, and k is the index of the time step; Based on the assumption of the target vehicle moving at a constant speed, obtain the position coordinates (px k , py k ) of the target vehicle at each time step, which are used as the preview points for the corresponding k time steps where are the position coordinates of the preview point for the k time step;

[0015] Step S2-3: Match the position coordinates of the preview point with the paths in the candidate reference path set SET(Path i ) to obtain the reference path closest to the position of the preview point, and determine the path curvature κ k of the preview point for the k time step from the reference path closest to the position of the preview point.

[0016] Preferably, the specific steps of Step S3 include:

[0017] Step S3-1: Based on the motion state of the target vehicle and the path curvature, establish the input of the random forest at the k time step

[0018] Step S3-2: Input the input of the random forest into the random forest for processing, and the N trained decision trees of the random forest generate corresponding prediction results where represents the prediction result generated by the Nth decision tree, v RF,N , ω RF,N are respectively the longitudinal speed and yaw angular velocity of the target vehicle generated by the Nth decision tree;

[0019] Step S3-3: Calculate the average value of the output results of the N decision trees to finally obtain the observed values of the longitudinal speed and yaw angular velocity at the k time step

[0020] Preferably, the specific steps of Step S4 include:

[0021] Step S4-1: Establish a state transition model for the extended Kalman filter. The state transition model calculates the state vector of the target vehicle at the next time step based on the nonlinear vehicle kinematic model for the state vector of the target vehicle at the current time step;

[0022] Step S4-2: Establish a measurement model for the extended Kalman filter. The measurement model uses the observed values of the longitudinal speed and yaw angular velocity as the measured values of the longitudinal speed and yaw angular velocity of the extended Kalman filter;

[0023] Step S4-3: Obtain the prior state estimation vector at the next time step using the state transition model, and calculate the prior error covariance matrix;

[0024] Step S4-4: Calculate the Kalman gain from the prior error covariance matrix, and obtain the state estimation vector at the next time step from the prior state estimation vector at the next time step; Obtain the error covariance matrix at the next time step from the prior error covariance matrix at the next time step;

[0025] Step S4-5: Return to Step S4-3 until the calculations for all time steps in the future time period are completed;

[0026] Step S4-6: Use the state estimation quantities from time steps k+1 to k+m in the future time period as the predicted trajectory Traj k+1,Kk+m 。

[0027] Preferably, in Step S4-1, the expression of the state transition model is:

[0028] x k+1 =f(x k )+q k =[f 1 ,f 2 ,f 3 ,f 4 ,f 5 ,f 6 +q k

[0029] f 1 :

[0030] f 2 :

[0031] f 3 :θ k+1 =θ k +ω k Δt

[0032] f 4 :v k+1 =v k +a k Δt

[0033] f 5 :a k+1 =a k

[0034] f 6 :ω k+1 =ω k

[0035] where, xk is the state vector of the target vehicle at the k-th time step, v k is the longitudinal speed of the target vehicle at the k-th time step, ω k is the yaw angular velocity of the target vehicle at the k-th time step, f(·) is the vehicle state transition model based on the non-linear vehicle kinematic model, q k is the process noise, which follows a normal distribution with a mean of 0 and a variance of Q k That is, q k ~N(0, Q k ), Q k = diag{0, 0, q dθ , q dv , q dω , q da}, q dθ , q dv , q dω , q da are the process noises of the heading angle, longitudinal speed, yaw angular velocity, and longitudinal acceleration, respectively.

[0036] Preferably, the step S4-3 specifically includes:

[0037] Predict the state of the vehicle at the next time step using the state transition equation and calculate the prior error covariance matrix, and the expression is:

[0038]

[0039] Among them, represents the prior state estimation vector of the target vehicle at the k-th time step, represents the posterior state estimation vector of the known target vehicle at the (k - 1)-th time step, is the prior error covariance matrix of the target vehicle at the k-th time step, is the posterior error covariance matrix of the target vehicle at the (k - 1)-th time step, F k is the Jacobian matrix of the state transition equation at the k-th time step, is the transpose of F k , Q k-1 is the process noise covariance matrix at the (k - 1)-th time step.

[0040] Preferably, the step S4-4 specifically includes:

[0041] Calculate the Kalman gain and update the state estimation vector and the error covariance matrix, and the expression is:

[0042]

[0043] Among them, K k is the Kalman gain at the k-th time step, R kDenote the measurement noise covariance matrix, R k = diag{σ v,k 2 , σ ω,k 2}, where σ v,k 2 , σ ω,k 2 are the variances of v k , ω k respectively. Denote the state estimation vector of the target vehicle at time step k, is the state measurement value, and the state prediction result of the random forest is used as the measurement value, i.e., where represent the longitudinal speed and yaw angular velocity of the target vehicle at time step k respectively. is the posterior error covariance matrix at time step k, and I is the identity matrix.

[0044] Compared with the prior art, the present invention has at least the following beneficial effects:

[0045] (1) The present invention can generate multiple candidate reference paths and calculate the corresponding path curvatures by using high-precision map information and the current motion state. These information provide rich input features for the random forest, enabling the model to better adapt to the changes in complex environments. In this way, when the vehicle is driving in complex terrains such as mining areas, the prediction model can be adjusted in real time to make more accurate predictions for different road conditions.

[0046] (2) By combining the advantages of the random forest and the extended Kalman filter, the present invention significantly improves the accuracy and stability of vehicle trajectory prediction. The random forest model can effectively identify the driving intention of the vehicle, learn and analyze the complex relationship between the vehicle speed, direction and road curvature, so as to improve the prediction accuracy of the future driving speed and turning rate, and can better adapt to the trajectory prediction in complex scenarios such as mining areas.

[0047] (3) The combination of the random forest and the extended Kalman filter adopted by the present invention effectively reduces the accumulation of errors and ensures the stability of the prediction results. By establishing the state transition model and the measurement model, the extended Kalman filter can perform effective state estimation and update when dealing with nonlinear systems. The prediction results not only have high accuracy but also ensure the stability of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings are only for the purpose of illustrating specific embodiments and are not considered as a limitation to the present invention.

[0049] Figure 1Flowchart of the method for realizing the trajectory prediction of open-pit mine vehicles by combining random forest and Kalman provided by the present invention.

[0050] Figure 2 Schematic diagram of obtaining the candidate reference path of the target vehicle and the input information of the random forest model provided by the present invention.

[0051] Figure 3 Result graph of the long-term trajectory prediction of the target vehicle on the specific mining area road provided by the invention. Detailed implementation manners

[0052] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below in conjunction with the drawings and specific implementation manners. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. In addition, the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0053] The present invention provides a method for realizing the trajectory prediction of open-pit mine vehicles by combining random forest and Kalman. It mainly relies on the path information of the reference lane lines in the prior high-precision map of the open-pit mining area. Based on the random forest method, comprehensively considering the driving state of the vehicle and the curvature of the reference path, the future driving speed and yaw angular velocity of the vehicle are predicted to realize the state estimation of the vehicle. Then, the result of the state estimation is combined with the extended Kalman filter for trajectory prediction, thereby obtaining the long-term trajectory prediction result of the target vehicle.

[0054] In order to illustrate the effectiveness of the method proposed by the present invention, the above technical solutions of the present invention will be described in detail below through a specific embodiment, as Figure 1 shown, a method for realizing the trajectory prediction of open-pit mine vehicles by combining random forest and Kalman is disclosed, and the specific implementation steps are as follows:

[0055] Step S1: Obtain the current motion state information of the mining area vehicle by the sensing device, where the motion state information includes the position, heading angle, longitudinal speed, longitudinal acceleration and yaw angular velocity of the mining area vehicle.

[0056] The mining area vehicle for which trajectory prediction is performed is the target vehicle. The current motion state information of the target vehicle can be obtained by a variety of sensing devices. Specifically, it can be obtained through the sensing module equipped on the autonomous vehicle itself, or collected by the sensing unit deployed on the roadside, or relevant data can be obtained from other vehicles through vehicle-to-vehicle communication (V2V). In some embodiments, the information obtained from the above multiple information sources can be data-fused to obtain more accurate and reliable target vehicle motion state information.

[0057] The motion state information of the target vehicle at the current moment t is expressed as: x t =[x t , y t , θ t , v t , a t , ω t T , where x t , y t , θ t , v t , a t , ω t respectively represent the eastward position coordinate, northward position coordinate, heading angle, longitudinal speed, longitudinal acceleration, and yaw angular velocity at the current moment t.

[0058] By obtaining and updating the above state information in real time, the motion state of the target vehicle on the mining area road can be accurately grasped, providing reliable data support for subsequent trajectory prediction.

[0059] Step S2: Obtain a candidate reference path based on the current motion state information and the prior high-precision map information of the open-pit mining area, and obtain the path curvature of the candidate reference path. The candidate reference path is the path that the mining area vehicle may travel in the future.

[0060] As Figure 2 shown, according to the real-time state information of the target vehicle and the prior high-precision map information of the open-pit mining area, obtain a set SET(Path i ) of possible candidate reference paths in front of the target vehicle. Among them, Path i represents the i-th path, and each path is a feasible driving trajectory.

[0061] In some embodiments, the method for obtaining the candidate reference path set may be: perform positioning and matching in the high-precision map based on the current position of the target vehicle, obtain the current road section where the vehicle is located, and then search in the road network topology structure of the high-precision map with the current road section as the starting point. All drivable paths obtained are used as candidate paths. And the candidate paths can be preliminarily screened by combining the current state information such as the heading angle and speed of the vehicle, and the paths that have obvious conflicts with the current vehicle motion state are excluded, and finally a reasonable candidate reference path set is obtained.

[0062] Denote the current moment as t. Based on the assumption that the target vehicle moves at a constant speed, obtain the position of the target vehicle at each time step t + kΔt within the next 5 seconds, where Δt is the time step size and k is the index of the time step, Δt = 0.1s, k = 1, 2,..., 50. Obtain the position coordinates (px k , py k ​). Determine the preview point coordinates at the k-th time step (px k , py k ) as the preview point at the k-th time step , where are the position coordinates of the preview point at the k-th time step.

[0063] For each preview point, match the position of the preview point with the paths in the candidate reference path set SET(Path i ) to obtain the reference path closest to the position of the preview point, and determine the path curvature κ of the preview point at the k-th time step from the reference path closest to the position of the preview point k .

[0064] In some embodiments, the path curvature may be obtained from the high-precision mine map according to the reference path.

[0065] Step S3: Process the current motion state information and the path curvature based on a random forest to obtain observed values of the longitudinal speed and yaw angular velocity of the mine vehicle.

[0066] The random forest of the present invention is composed of multiple trained decision trees. N is the total number of decision trees. The trained decision trees can obtain predicted longitudinal speed and yaw angular velocity based on the input target vehicle information.

[0067] Based on the target vehicle motion state and the path curvature, establish the input of the random forest at the k-th time step Integrates the current vehicle state, time step, and curvature information of the reference path.

[0068] The N decision trees of the random forest generate corresponding prediction results where represents the prediction result generated by the N-th decision tree, v RF,N , ω RF,N are respectively the longitudinal speed and yaw angular velocity of the target vehicle generated by the N-th decision tree. Finally, the output result of the random forest is the average of the output results of the N decision trees, denoted as Variance

[0069]

[0070] By inputting the input of the random forest at the k-th time step into the random forest, obtain the longitudinal speed and yaw angular velocity at the k-th time step as the observed values at the k-th time step The variance of is used as the observation noise R at the k-th time step k = diag{σ v,k2 , σ ω,k 2}, where σ v,k 2 , σ ω,k 2 are respectively the variances, and diag{·} represents constructing a diagonal matrix.

[0071] Step S4: Establish the state transition model and measurement model of the extended Kalman filter, and use the observed values of the longitudinal speed and yaw angular velocity as the measured values of the longitudinal speed and yaw angular velocity of the extended Kalman filter;

[0072] Perform state prediction and measurement update of the extended Kalman filter, and finally obtain the predicted trajectory of the mining area vehicle at future time.

[0073] In this step, the present invention uses the driving speed and yaw angular velocity of the target vehicle at future time steps obtained by state estimation based on the random forest method as the measured values, and updates other state quantities through the extended Kalman filter to obtain the estimated values of other state quantities of the target vehicle within the prediction time domain, and further obtains the long-term prediction trajectory of the target vehicle. The specific steps include:

[0074] (1) Construct the state transition model

[0075] The expression of the state transition model of the extended Kalman filter is:

[0076] x k+1 = f(x k ) + q k = [f 1 , f 2 , f 3 , f 4 , f 5 , f 6 + q k

[0077] where x k is the state vector of the target vehicle at the k-th time step, v k is the longitudinal speed of the target vehicle at the k-th time step, ω k is the yaw angular velocity of the target vehicle at the k-th time step, f(·) is the vehicle state transition model based on the non-linear vehicle kinematic model, q k is the process noise, which follows a normal distribution with a mean of 0 and a variance of Q k , that is, q k ~ N(0, Q k ), and Q k = diag{0, 0, q dθ , q dv , qdω , q da},q dθ , q dv , q dγ , q da are the process noises of the heading angle, longitudinal speed, yaw angular velocity, and longitudinal acceleration, respectively.

[0078] The vehicle state transition equation f(x k ) is:

[0079] f 1 :

[0080] f 2 :

[0081] f 3 : θ k+1 = θ k + ω k Δt

[0082] f 4 : v k+1 = v k + a k Δt

[0083] f 5 : a k+1 = a k

[0084] f 6 : ω k+1 = ω k

[0085] (2) Construct the measurement model

[0086] The expression of the measurement model of the extended Kalman filter is:

[0087] z k = Hx k + γ k

[0088] where is the measurement matrix, γk is the measurement noise, and γ k ~ N(0, R k ) indicates that the measurement noise satisfies a normal distribution with a variance of σ v,k 2 , σ ω,k 2 , and R k = diag{σ v,k 2 , σ ω,k 2}.

[0089] (3) State prediction process

[0090] In the state prediction process, the state transition equation is used to predict the state of the vehicle at the next time step, and the prior error covariance matrix is calculated. The expressions are as follows:

[0091]

[0092] Where, represents the prior state estimation vector of the target vehicle at time step k, represents the posterior state estimation vector of the known target vehicle at time step k - 1, is the prior error covariance matrix of the target vehicle at time step k, is the posterior error covariance matrix of the target vehicle at time step k - 1, F k is the Jacobian matrix of the state transition equation at time step k, which is obtained by solving the first-order partial derivative of the state transition equation with respect to the state vector of the target vehicle, is F k transpose of, Q k-1 is the process noise covariance matrix at time step k - 1.

[0093] (4) Measurement update process

[0094] In the measurement update process, the Kalman gain is calculated, and the state estimation vector and the error covariance matrix are updated. The expressions are as follows:

[0095]

[0096] Where, K k is the Kalman gain at time step k, R k represents the measurement noise covariance matrix, R k = diag{σ v,k 2 , σ ω,k 2}}, σ v,k 2 , σ ω,k 2 are the variances of v k , ω k respectively, represents the state estimation vector of the target vehicle at time step k, is the state measurement value, and the state prediction result of the random forest is used as the measurement value, that is Where, represent the longitudinal speed and yaw angular velocity of the target vehicle at time step k respectively, is the posterior error covariance matrix at time step k, and I is the identity matrix.

[0097] By continuously iterating the state prediction process and the measurement update process, the predicted trajectory Traj formed by the motion state of the target vehicle in the k+1 to k+m time steps can be obtained. k+1,Kk+m .

[0098] like Figure 3 As shown, it is a result diagram of the long-term trajectory prediction of the target vehicle on a specific mining area road according to the present invention.

[0099] Through the method described in this embodiment, the vehicle trajectory prediction in the open-pit mine area is realized. By combining the advantages of random forest and extended Kalman filter, compared with the vehicle trajectory prediction method based on physical kinematic model, the driving state of the vehicle and the prior map information are comprehensively considered. The relationship between the vehicle speed, direction and road curvature is learned through random forest, which effectively improves the prediction accuracy of the vehicle's future driving speed and turning rate, and significantly improves the accuracy, stability and adaptability of vehicle trajectory prediction. At the same time, the combination of random forest and extended Kalman filter reduces error accumulation, ensures the stability of the prediction results, and provides strong support for unmanned driving trajectory prediction technology in complex environments such as mining areas.

[0100] Although the specific embodiments of the present invention depict various actions or steps in a specific order, this should be understood as requiring such actions or steps to be performed in the specific order shown or in a sequential order, or requiring all illustrated actions or steps to be performed to obtain the desired results. Under certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the present disclosure. Certain features described in the context of a separate embodiment can also be implemented in a single implementation in combination. On the contrary, the various features described in the context of a single implementation can also be implemented in multiple implementations individually or in any suitable sub-combination. The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or replacements that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

[0101] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting vehicle trajectories in open-pit mines by combining random forest and Kalman, characterized in that: The following steps are involved: Step S1, obtaining the current motion state information of the mining vehicle by the sensing device, wherein the motion state information includes the position, heading angle, longitudinal speed, longitudinal acceleration and yaw angular velocity of the mining vehicle; Step S2, obtaining a candidate reference path based on the current motion state information and a priori high-precision map information of the open-pit mine area, and obtaining the path curvature of the candidate reference path, wherein the candidate reference path is a path that the mining area vehicle may travel in the future; Step S3, processing the current motion state information and the path curvature based on random forest to obtain observation values ​​of the longitudinal velocity and yaw angular velocity of the mining vehicle; Step S4, establishing a state transition model and a measurement model of an extended Kalman filter, and using the observed values ​​of the longitudinal velocity and the yaw angular velocity as the measured values ​​of the longitudinal velocity and the yaw angular velocity of the extended Kalman filter; Perform state prediction and measurement update of the extended Kalman filter to finally obtain the predicted trajectory of the mining vehicle in the future.

2. The method for predicting vehicle trajectories in an open-pit mine by combining random forest and Kalman according to claim 1, characterized in that: In step S1, the motion state information of the target vehicle at the current time t is expressed as: t =[x t ,y t ,θ t ,v t ,a t ,ω t ] T , where x t ,y t ,θ t ,v t ,a t ,ω t They represent the east position coordinate, north position coordinate, heading angle, longitudinal velocity, longitudinal acceleration and yaw angular velocity at the current time t respectively.

3. The method for predicting open-pit mine vehicle trajectories by combining random forest and Kalman according to claim 2, characterized in that: The step S2 specifically includes: Step S2-1, based on the current motion state information and the prior high-precision map information of the open-pit mine area, obtain a possible candidate reference path set SET (Path i ), where Path i represents the i-th path; Step S2-2, divide the preset future time period into multiple time steps, expressed as t+kΔt, where t is the current time, Δt is the time step, and k is the index of the time step; based on the assumption that the target vehicle moves at a uniform speed, obtain the target vehicle position coordinates (px) at each time step k ,py k ), as the preview point of the corresponding k time step in is the position coordinate of the preview point in k time step; Step S2-3: Based on the position coordinates of the preview point and the candidate reference path set SET (Path i ) to obtain a reference path closest to the preview point, and determine the path curvature κ of the preview point at time step k by the reference path closest to the preview point. k .

4. The method for predicting open-pit mine vehicle trajectories by combining random forest and Kalman according to claim 3 is characterized in that: The step S3 specifically includes: Step S3-1: Based on the target vehicle motion state and the path curvature, establish the input of the random forest under k time steps Step S3-2: Input the input of the random forest into the random forest for processing, and the N trained decision trees of the random forest generate corresponding prediction results. in, Represents the prediction result generated by the Nth decision tree, v RF,N ,ω RF,N are the longitudinal velocity and yaw angular velocity of the target vehicle generated by the Nth decision tree respectively; Step S3-3: Calculate the average value of the output results of N decision trees, and finally obtain the longitudinal velocity and yaw angular velocity observation values ​​of k time steps.

5. The method for predicting open-pit mine vehicle trajectories by combining random forest and Kalman according to claim 4, characterized in that: The step S4 specifically includes: Step S4-1, establishing a state transfer model of an extended Kalman filter, wherein the state transfer model calculates the state vector of the target vehicle at the current time step based on a nonlinear vehicle kinematic model to obtain the state vector of the target vehicle at the next time step; Step S4-2, establishing a measurement model of an extended Kalman filter, wherein the measurement model uses the observed values ​​of the longitudinal velocity and the yaw angular velocity as the measured values ​​of the longitudinal velocity and the yaw angular velocity of the extended Kalman filter; Step S4-3, using the state transfer model to obtain the prior state estimation vector of the next time step, and calculating the prior error covariance matrix; Step S4-4, calculating the Kalman gain by the prior error covariance matrix, obtaining the state estimation vector of the next time step by the prior state estimation vector of the next time step; obtaining the error covariance matrix of the next time step by the prior error covariance matrix of the next time step; Step S4-5, return to step S4-3 until the calculation of all time steps in the future time period is completed; Step S4-6: The state estimation value of the k+1 to k+m time steps in the future time period is used as the predicted trajectory Traj k +1,Kk+m .

6. The method for predicting vehicle trajectories in an open-pit mine by combining random forest and Kalman according to claim 5, characterized in that: In step S4-1, the state transition model expression is: x k+1 =f(x k )+q k =[f1,f2,f3,f4,f5,f6]+q k f3:θ k+1 =θ k +oh k Δt f4:v k+1 =v k +a k Δt f5: a k+1 =a k f6:oh k+1 =ω k Among them, x k is the state vector of the target vehicle at time step k, v k is the longitudinal velocity of the target vehicle at time step k, ω k is the yaw rate of the target vehicle at time step k, f(·) is the vehicle state transition model based on the nonlinear vehicle kinematics model, and q k is process noise, with mean 0 and variance Q k The normal distribution of q k ~N(0,Q k ), Q k =diag{0,0,q dθ ,q dv ,q dω ,q da }, q dθ ,q dv ,q dω ,q da They are the process noise of heading angle, longitudinal velocity, yaw angular velocity and longitudinal acceleration respectively.

7. The method for predicting open-pit mine vehicle trajectories by combining random forest and Kalman according to claim 6, characterized in that: The step S4-3 specifically includes: Use the state transfer equation to predict the state of the vehicle at the next time step and calculate the prior error covariance matrix, which is expressed as: in, represents the prior state estimate vector of the target vehicle at time step k, represents the posterior state estimate vector of the target vehicle known at k-1 time steps, is the prior error covariance matrix of the target vehicle at k time steps, is the posterior error covariance matrix of the target vehicle at k-1 time steps, F k is the Jacobian matrix of the state transfer equation for k time steps, F k The transpose of Q k-1 is the process noise covariance matrix for k-1 time steps.

8. The method for predicting vehicle trajectories in an open-pit mine by combining random forest and Kalman according to claim 7, characterized in that: The step S4-4 specifically includes: Calculate the Kalman gain and update the state estimation vector and error covariance matrix. The expression is: Among them, K k is the Kalman gain at time step k, R k represents the measurement noise covariance matrix, R k =diag{σ v,k 2 ,σ ω,k 2 },σ v,k 2 ,σ ω,k 2 They are v k ,ω k The variance of represents the state estimation vector of the target vehicle at time step k, is the state measurement value, and the state prediction result of random forest is used as the measurement value, that is, in, They represent the longitudinal velocity and yaw rate of the target vehicle at k time steps, is the posterior error covariance matrix for k time steps, and I is the identity matrix.

Images