Vehicle motion trail prediction method and system

Abstract

The invention discloses a vehicle motion trail prediction method and system, and belongs to the technical field of surface mine unmanned driving. The method comprises the following steps: preprocessing vehicle motion trail data collected in real time, inputting the preprocessed vehicle motion trail data into a pre-trained machine learning model, and predicting to obtain a motion trail of a vehicle at a future moment; determining the grade of the collision risk according to the predicted motion track of the vehicle at the future moment, and providing an early warning signal according to the grade of the collision risk; the method is suitable for mining vehicles, and can solve the problems of low prediction precision and slow response speed.

Description

Technical Field

[0001] The invention relates to a vehicle motion trajectory prediction method and system, belonging to the technical field of unmanned driving in open-pit mines. Background Art

[0002] With the continuous improvement of the degree of automation in mines, the automated driving technology of mining trucks has been widely used in actual production, and the safety of unmanned mining trucks has become the focus of research and practice. In a complex mining environment, how to accurately predict the motion trajectory of mining vehicles is crucial to ensuring the safe operation of vehicles and improving transportation efficiency. At present, traditional trajectory prediction methods mainly rely on simple mathematical models or empirical rules, which are difficult to accurately reflect the actual driving conditions of mining vehicles in complex mining environments, and have problems such as low prediction accuracy and poor adaptability. Therefore, there is an urgent need for a method that can accurately predict the future motion trajectory of mining vehicles and effectively apply it to mining operations to improve the overall safety and efficiency of mining truck automatic driving systems.

[0003] Patent No. 202210578437.8 provides a vehicle trajectory prediction method, device, chip, terminal and computer equipment. The invention includes: determining the relative coordinate system relative to the vehicle body coordinate system at the starting time according to the geodetic coordinates of the vehicle at different driving times; obtaining the lateral displacement, longitudinal velocity and longitudinal acceleration of the vehicle in the relative coordinate system at different driving times, and generating an observation state sequence according to the lateral displacement, longitudinal velocity and longitudinal acceleration; identifying the vehicle's driving behavior based on the observation state sequence and the behavior recognition model corresponding to different driving behaviors; generating a set of trajectory curves of different driving behaviors based on the observation state sequence and a preset polynomial, and determining a target trajectory curve set corresponding to the driving behavior of the identified vehicle in the set of trajectory curves of different driving behaviors; determining the vehicle's trajectory curve in the target trajectory curve set according to the observation state sequence.

[0004] Patent No. 202210578437.8 provides a method for generating driving trajectories of unmanned vehicles in open-pit mines. The driving trajectory generation method of the invention is to collect the road boundaries and intersection boundaries of the mining area, and generate driving trajectories in road scenes and intersection scenes based on the sequential quadratic programming method and the geometric curve method. It can meet the high-frequency changeability of roads in open-pit mining scenes and can generate mining area vehicle driving trajectories in real time, thereby improving the generation rate of mining area vehicle driving trajectories. The generated vehicle driving trajectory meets the requirements of vehicle dynamics and is easy for unmanned vehicles in mining areas to preview and track.

[0005] A Learning-Based Approach for Lane Departure Warning Systems with a Personalized Driver Model A learning-based approach is proposed to predict unexpected lane departure behavior and a warning strategy based on a model prediction algorithm is developed.

[0006] The patent No. 202210578437.8 provides a vehicle trajectory prediction method, device, chip, terminal and computer equipment. The patent has complex data processing and relies on the conversion of geodetic coordinates and relative coordinate systems, which increases the risk of error; the behavior recognition model is not accurate enough; the method of generating trajectory curves by pre-setting polynomials is difficult to cope with complex trajectory changes and cannot effectively capture nonlinear characteristics and long-term dependencies.

[0007] Patent No. 202210578437.8 provides a method for generating driving trajectories for unmanned vehicles in open-pit mines. Although it can generate driving trajectories that are suitable for open-pit mine roads, it has problems such as complex calculations, limited satisfaction of vehicle dynamics, poor algorithm versatility, and accuracy and stability during long-term operation, which cannot meet actual needs.

[0008] A Learning-Based Approach for Lane Departure Warning Systems with a Personalized Driver Model proposes a learning-based approach to predict unexpected lane departure behavior and the driver's attempt to pull the vehicle back to the correct lane. However, it is mainly designed for the safety of passenger car drivers to warn drivers if lane departure occurs, and the prediction method cannot be generally applied to mining vehicles. Summary of the invention

[0009] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a vehicle motion trajectory prediction method and system, which collects multi-feature motion trajectory data of mining vehicles, uses a sliding window method to construct a large number of time series samples, and selects a machine learning algorithm to improve the accuracy of trajectory prediction, so as to solve the problems of low prediction accuracy and slow response speed; In order to achieve the above purpose / solve the above technical problems, the present invention is implemented by adopting the following technical solutions: In a first aspect, a vehicle motion trajectory prediction method is disclosed, the method comprising: Collect vehicle motion trajectory data in real time; Preprocess the real-time collected vehicle motion trajectory data, and input the preprocessed vehicle motion trajectory data into a pre-trained machine learning model to predict the vehicle's motion trajectory at future moments; Expanding the predicted motion trajectory according to the vehicle width to generate an outer envelope of the vehicle trajectory; Calculate the overlap degree and time of the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; Based on the degree and time of overlap, the level of collision risk is determined, and based on the level of collision risk, a warning signal is provided.

[0010] Optionally, the real-time collection of vehicle motion data is performed by installing a GPS system and sensors on the vehicle to collect the vehicle's motion trajectory data in real time, and the data includes at least position coordinates, speed, acceleration, direction and load.

[0011] Optionally, the vehicle motion trajectory data collected in real time is preprocessed, including data normalization and feature extraction; The data normalization normalizes the motion trajectory data to The formula is as follows:

[0012] in, is the minimum value of the data, is the maximum value of the data, for The data at the moment, is the normalized data; The feature extraction is used to extract the coordinates, speed, acceleration, direction, and load physical quantities from the normalized motion trajectory data as input features to form a feature vector. The formula is as follows:

[0013] in, is the feature vector, Representative vehicle The longitude of the moment, Representative vehicle The latitude of the moment, Representative vehicle The speed of time, Representative vehicle The acceleration of time, Representative vehicle The direction of time, Representative vehicle Load capacity at the moment.

[0014] Optionally, the machine learning model adopts a long short-term memory network model, and the long short-term memory network model includes an input layer, multiple hidden layers and an output layer; Among them, the input layer is used to receive the feature vector as input; The hidden layer consists of multiple LSTM units, which are used to extract time series features. Each LSTM unit has memory capabilities and can capture long-term dependencies in time series. The output layer is used to generate predictions for future trajectories.

[0015] Optionally, the machine learning model is trained using a training set, using mean square error as a loss function, optimizing model parameters through gradient descent and back propagation algorithms, using a validation set to evaluate model performance, adjusting hyperparameters, and using a test set to evaluate the final performance and generalization ability of the model.

[0016] Optionally, the mean square error formula is:

[0017] Where N is the number of samples, is the true value, is the predicted value.

[0018] Optionally, use a sliding window method to construct a time series sample, where each window contains The input features at each moment are The output features at each moment, the sliding step length is , generate training set and validation set for training model.

[0019] Optionally, the steps of the sliding window method are as follows: choose and :Set the number of input points And output points , the path contains data points; Set the sliding window: the window size is , the sliding step length is ; Generate samples: Each window generates an input-output sample pair.

[0020] Optionally, the vehicle collision warning includes: Trajectory outer envelope generation: The trajectory of each vehicle is expanded according to the vehicle width to form the outer envelope of the vehicle trajectory; Outer envelope overlap calculation: Calculate the overlap between the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; Collision risk level assessment: Determine the level of collision risk based on the overlap degree and time of the outer envelope; Warning signal: According to the collision risk level, a warning signal is provided, including sound alarm and flash alarm.

[0021] In another aspect, a vehicle motion trajectory prediction system is provided, the system comprising: A data acquisition module is configured to collect vehicle motion trajectory data in real time; A prediction module is configured to pre-process the vehicle motion trajectory data collected in real time, and input the pre-processed vehicle motion trajectory data into a pre-trained machine learning model to predict the vehicle's motion trajectory at a future moment; An expansion module is configured to expand the predicted motion trajectory according to the width of the vehicle to generate an outer envelope of the vehicle trajectory; A calculation module, configured to calculate the overlap degree and time between the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; The warning module is configured to determine the level of collision risk according to the overlap degree and time, and provide a warning signal according to the collision risk level.

[0022] Compared with the prior art, the present invention has the following beneficial effects: The present invention collects multi-feature driving data of mining vehicles, constructs time series samples using the sliding window method, and constructs models using machine learning algorithms. It can accurately predict the driving trajectory of vehicles at multiple moments in the future, improve prediction accuracy and the ability to adapt to complex environments. The present invention meets the real-time scheduling needs of mining operations through real-time data collection and fast prediction algorithms. The system can accurately predict vehicle operation trajectories, improve transportation efficiency, effectively prevent potential collision risks, and reduce collision accidents, and has strong practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The figure is a flow chart of an embodiment of the vehicle motion trajectory prediction system of the present invention; Figure 2 Shown is a schematic diagram of input and output parameters of the machine learning model of the present invention; Figure 3 Shown is a schematic diagram of the long short-term memory network model of the present invention; Figure 4 Shown is a schematic diagram of the vehicle motion trajectory prediction effect of the present invention. DETAILED DESCRIPTION

[0024] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods. Example

[0025] like Figure 1-Figure 4 As shown, a vehicle motion trajectory prediction method is disclosed, the method comprising: Step 1: Collect vehicle motion trajectory data in real time; Step 2: preprocess the real-time collected vehicle motion trajectory data, and input the preprocessed vehicle motion trajectory data into a pre-trained machine learning model to predict the vehicle's motion trajectory at a future moment; Step 3, expanding the predicted motion trajectory according to the vehicle width to generate an outer envelope of the vehicle trajectory; Step 4, calculating the overlap degree and time of the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; Step 5, determining the level of collision risk based on the overlap degree and time, and providing a warning signal based on the collision risk level.

[0026] Regarding step 1, in the specific implementation process of this embodiment, the following processing is performed: (1) Data collection: By installing a GPS system and a variety of sensors on mining vehicles, multi-feature motion trajectory data of mining vehicles in different scenarios are collected, including position coordinates, speed, acceleration, direction, load, etc. The default data collection frequency is 10Hz.

[0027] (2) Data preprocessing: preprocess the collected data to ensure the accuracy and standardization of the data: 1) Data normalization: Convert data of different dimensions to a unified scale, for example, normalize feature values ​​such as position, velocity, acceleration, and direction to The formula is as follows: in, and are the minimum and maximum values ​​of the data, respectively. for The data at the moment, is the normalized data.

[0028] 2) Feature extraction: Extract physical quantities such as position coordinates, speed, acceleration, direction, load and so on from the processed data as input features. Form feature vector , as input data for subsequent model training.

[0029]

[0030] in, Represents vehicles Longitude, latitude, speed, acceleration, direction and load at the moment.

[0031] In this embodiment, the specific steps for training the machine learning model are as follows: 2. Construction of data set for training model; For the collected paths, each path has data points. Assume that each input sequence contains The data of consecutive moments is output as follows In the process of data set construction, in order to improve the effect of model training and prediction, the sliding window method is used to construct time series samples.

[0032] The samples are generated by sliding window method, each window contains The input features at each moment are The output features at each moment, the sliding step length is , thereby generating a large number of samples. The specific steps of the sliding window method are as follows: (1) Selection and :Set the number of input points And output points , the path contains data points.

[0033] (2) Sliding window: The window size is , the sliding step length is .

[0034] (3) Generate samples: Each window generates an input-output sample pair. The input and output parameters of the sample are as follows: Figure 2 shown.

[0035] For example, for a path containing 100 data points ( ), choose , , , that is, each input contains data from 10 consecutive moments, and the output is the predicted trajectory for the next 10 moments, with a sliding step size of 1.

[0036] The data points of the path are as follows:

[0037] The samples generated by the sliding window method are as follows: Sample 1: enter:

[0038] Output:

[0039] Sample 2: enter:

[0040] Output: ... Sample 81: enter:

[0042] Output:

[0043] In this way, we can generate samples for model training and validation.

[0044] The above processing is performed on each collected path to generate a data set, and the data set is divided into a training set, a validation set, and a test set. The commonly used division ratio is 60% (training set), 20% (validation set), and 20% (test set).

[0045] 3. Model selection and training; (1) Model selection: According to the spatiotemporal characteristics of the mining vehicle's motion trajectory, a model suitable for processing sequence data is selected, such as a long short-term memory network (LSTM) or a convolutional neural network (CNN) model. The advantages and disadvantages of different models are compared. For example, LSTM has advantages in processing long sequence data, while CNN performs better in extracting spatial features. The present invention selects a long short-term memory network (LSTM) model, which is suitable for processing time series data and can capture long-term dependencies and nonlinear features in the vehicle's motion trajectory. Long short-term memory network (LSTM) model is Figure 3 shown.

[0046] (2) Data input: The preprocessed historical driving data is input into the model in time series. The data format is:

[0047] in, is the input step size, is the prediction step length.

[0048] (3) LSTM model construction: Build a deep learning network consisting of multiple LSTM units, including an input layer, multiple hidden layers, and an output layer: Input layer: receives feature vector As input; Hidden layer: composed of multiple LSTM units to extract time series features. Each LSTM unit has memory ability and can capture long-term dependencies in time series; Output layer: Generates the prediction results of future trajectories, that is, the predicted location information .

[0049] (4) Model training: The model is trained using the training set and validated on the validation set to adjust hyperparameters and prevent overfitting. The mean square error (MSE) is used to measure the prediction performance of the model. The mean square error (MSE) between the prediction result and the true value is calculated as follows:

[0050] Where N is the number of samples, is the true value, is the predicted value.

[0051] According to the performance of the validation set, the gradient descent and back propagation algorithms are used to gradually improve the prediction accuracy of the model by continuously adjusting the network weights and minimizing the loss function.

[0052] 4. Model testing; (1) Loading test data: The test dataset undergoes the same preprocessing operations as the training data, such as normalization and feature extraction.

[0053] (2) Predicting vehicle trajectory: Use the trained model to predict the validation set and obtain the predicted trajectory of the mining vehicle.

[0054] (3) Calculate evaluation indicators: 1) Mean Squared Error (MSE): Calculate the mean squared error between the predicted trajectory and the actual trajectory.

[0055] 2) Mean Absolute Error (MAE): Calculate the mean absolute error between the predicted trajectory and the actual trajectory. The formula is:

[0056] in, is the sample size, is the true value, is the predicted value.

[0057] Analyze the performance of the model on the test set to determine whether the model meets the actual application requirements. If the model performance is poor, return to model training to tune the model until satisfactory results are achieved.

[0058] In the specific implementation process, the specific predictions in step 2 are as follows: In practical applications, the real-time trajectory prediction part mainly includes acquiring real-time data, real-time data processing, trajectory prediction, and collision warning. Figure 4 shown.

[0059] The following are the specific steps for vehicle motion trajectory prediction: 1. Get real-time data; Sensors installed on mining vehicles collect real-time information about the current time and the preceding The vehicle motion data at each moment includes position coordinates, speed, acceleration, direction, load, etc.

[0060] 2. Real-time data processing; Preprocess the data collected in real time to ensure the accuracy and consistency of the data. The steps include: (1) Data normalization: Normalize the data to scope; (2) Feature extraction: Extract physical quantities such as position coordinates, speed, acceleration, direction, and load from the processed data as input features. Form a feature vector , as input data for subsequent model training.

[0061]

[0062] in, Represents vehicles Longitude, latitude, speed, acceleration, direction and load at the moment.

[0063] 3. Trajectory prediction; Using the trained LSTM model, input the current time and the previous The processed feature vector at each moment is used to predict the future The trajectory of movement at a moment.

[0064] (1) Set the prediction step size: determine the future that needs to be predicted a moment.

[0065] (2) Execute prediction: Input real-time data into the model to obtain future The prediction results at that moment.

[0066] Assume that the current time is , the input sequence is:

[0067] The prediction results are:

[0068] 4. Collision warning; After obtaining the future trajectories of all mining vehicles, we expand the trajectory of each vehicle according to the vehicle width to form the outer envelope of the vehicle trajectory. Calculate the overlap between the outer envelope of the current vehicle trajectory and the outer envelope of other vehicles to determine the collision risk level of the vehicle. Based on the collision risk level, provide corresponding early warning signals to prevent potential safety risks. The specific steps are as follows: (1) Trajectory outer envelope generation; The predicted trajectory of each mining vehicle is expanded to take into account the width of the vehicle , generating the outer envelope of the trajectory. The outer envelope of the trajectory can be generated by offsetting the trajectory points on both sides. to form.

[0069] (2) Calculation of outer envelope coincidence; Calculate the overlap between the outer envelope of the current vehicle's trajectory and the outer envelope of other vehicles' trajectories. If the two outer envelopes have overlapping areas, it is determined that there is a collision risk.

[0070] (3) Collision risk level assessment; The level of collision risk is determined based on the overlap degree and time of the outer envelopes. The risk level can be divided into three levels: low, medium and high.

[0071] (4) Early warning signals; According to the collision risk level, corresponding warning signals are provided. Warning signals include sound alarms and flashing alarms to remind drivers and dispatchers in time. Example

[0072] A vehicle motion trajectory prediction system is disclosed, the system comprising: A data acquisition module is configured to collect vehicle motion trajectory data in real time; A prediction module is configured to pre-process the vehicle motion trajectory data collected in real time, and input the pre-processed vehicle motion trajectory data into a pre-trained machine learning model to predict the vehicle's motion trajectory at a future moment; An expansion module is configured to expand the predicted motion trajectory according to the width of the vehicle to generate an outer envelope of the vehicle trajectory; A calculation module, configured to calculate the overlap degree and time between the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; The warning module is configured to determine the level of collision risk according to the overlap degree and time, and provide a warning signal according to the collision risk level.

[0073] The present invention aims to solve the problems of low prediction accuracy and slow response speed. By collecting multi-feature motion trajectory data of mining vehicles, a large number of time series samples are constructed using the sliding window method, and a machine learning algorithm is selected to improve the accuracy of trajectory prediction. The vehicle's current driving data at multiple moments is input into a trained machine learning model to predict the vehicle's motion trajectory at multiple moments in the future in real time. The collision risk is evaluated based on the predicted trajectory data to improve the safety and efficiency of mining operations. In general, the present invention provides a high-precision, real-time mining vehicle motion trajectory prediction system, which provides a strong guarantee for improving the safety and efficiency of mining truck automatic driving systems.

[0074] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A vehicle motion trajectory prediction method, characterized in that: The method comprises: Collect vehicle motion trajectory data in real time; Preprocess the real-time collected vehicle motion trajectory data, and input the preprocessed vehicle motion trajectory data into a pre-trained machine learning model to predict the vehicle's motion trajectory at future moments; Expanding the predicted motion trajectory according to the vehicle width to generate an outer envelope of the vehicle trajectory; Calculate the overlap degree and time of the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; Based on the degree and time of overlap, the level of collision risk is determined, and based on the level of collision risk, a warning signal is provided.

2. The vehicle motion trajectory prediction method according to claim 1, characterized in that: The real-time collection of vehicle motion data collects the vehicle's motion trajectory data in real time by installing a GPS system and sensors on the vehicle, and the data at least includes position coordinates, speed, acceleration, direction and load.

3. The vehicle motion trajectory prediction method according to claim 1, characterized in that: The preprocessing of the real-time collected vehicle motion trajectory data includes data normalization and feature extraction; The data normalization normalizes the motion trajectory data to The formula is as follows: ； in, is the minimum value of the data, is the maximum value of the data, for The data at the moment, is the normalized data; The feature extraction is used to extract the coordinates, speed, acceleration, direction, and load physical quantities from the normalized motion trajectory data as input features to form a feature vector. The formula is as follows: ； in, is the feature vector, Representative vehicle The longitude of the moment, Representative vehicle The latitude of the moment, Representative vehicle The speed of time, Representative vehicle The acceleration of time, Representative vehicle The direction of time, Representative vehicle Load capacity at the moment.

4. The vehicle motion trajectory prediction method according to claim 1, characterized in that: The machine learning model adopts a long short-term memory network model, and the long short-term memory network model includes an input layer, multiple hidden layers and an output layer; Among them, the input layer is used to receive the feature vector as input; The hidden layer consists of multiple LSTM units, which are used to extract time series features. Each LSTM unit has memory capabilities and can capture long-term dependencies in time series. The output layer is used to generate predictions for future trajectories.

5. The vehicle motion trajectory prediction method according to claim 1, characterized in that: The machine learning model is trained using a training set, using mean square error as the loss function, optimizing model parameters through gradient descent and back propagation algorithms, evaluating model performance using a validation set, adjusting hyperparameters, and using a test set to evaluate the final performance and generalization ability of the model.

6. The vehicle motion trajectory prediction method according to claim 5, characterized in that: The mean square error formula is: ； Where N is the number of samples, is the true value, is the predicted value.

7. The vehicle motion trajectory prediction method according to claim 5, characterized in that: Use the sliding window method to construct time series samples, each window contains The input features at each moment are The output features at each moment, the sliding step length is , generate training set and validation set for training model.

8. The vehicle motion trajectory prediction method according to claim 7, characterized in that: The steps of the sliding window method are as follows: choose and :Set the number of input points And output points , the path contains data points; Set the sliding window: the window size is , the sliding step length is ; Generate samples: Each window generates an input-output sample pair.

9. The vehicle motion trajectory prediction method according to claim 1, characterized in that: The vehicle collision warning includes: Trajectory outer envelope generation: The trajectory of each vehicle is expanded according to the vehicle width to form the outer envelope of the vehicle trajectory; Outer envelope overlap calculation: Calculate the overlap between the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; Collision risk level assessment: Determine the level of collision risk based on the overlap degree and time of the outer envelope; Warning signal: According to the collision risk level, a warning signal is provided, including sound alarm and flash alarm.

10. A vehicle motion trajectory prediction system, characterized in that: The system comprises: A data acquisition module is configured to collect vehicle motion trajectory data in real time; A prediction module is configured to pre-process the vehicle motion trajectory data collected in real time, and input the pre-processed vehicle motion trajectory data into a pre-trained machine learning model to predict the vehicle's motion trajectory at a future moment; An expansion module is configured to expand the predicted motion trajectory according to the width of the vehicle to generate an outer envelope of the vehicle trajectory; A calculation module, configured to calculate the overlap degree and time between the outer envelope of the current vehicle trajectory and the outer envelope of other vehicle trajectories; The warning module is configured to determine the level of collision risk according to the overlap degree and time, and provide a warning signal according to the collision risk level.

Images