Mine electric locomotive unmanned comprehensive control method based on artificial intelligence

Abstract

The invention relates to the technical field of mine electric locomotives, and discloses a mine electric locomotive unmanned comprehensive control method based on artificial intelligence. The method comprises the following steps: S1, multi-modal environment perception: acquiring static and dynamic data of a mine environment by a multi-modal sensor, and processing the data through a sensor fusion technology; s2, real-time environment modeling: fusing multi-modal data to construct a mine three-dimensional real-time environment model; and S3, dynamic target trajectory prediction: predicting the motion trajectory of the dynamic target based on the deep learning model. And through a deep learning model, carrying out modeling and prediction on a historical motion track of a dynamic target. Through a multi-head self-attention mechanism, the model can accurately capture the time correlation and the multi-target interaction relationship in the target trajectory, the probability distribution of the future trajectory is generated, and important support is provided for obstacle avoidance decision and path optimization in a complex dynamic environment.

Description

Technical Field

[0001] The present invention relates to the technical field of mine electric locomotives, and in particular to an artificial intelligence-based unmanned driving integrated control method for mine electric locomotives. Background Art

[0002] Mine locomotives are important means of transportation in mines, responsible for transporting ore, personnel and materials from deep in the mine to the surface. With the continuous increase in the depth of mines and the complexity of the operating environment, the traditional locomotive driving method has been unable to meet the needs of safety, efficiency and automation. Therefore, it is particularly important to develop locomotive control technology that can perform unmanned operation in complex and dangerous mine environments.

[0003] Most mine electric locomotives rely on a single sensor for environmental perception, which cannot fully and accurately reflect the dynamic changes in the complex environment inside the mine. LiDAR can provide distance information in three-dimensional space, but cannot perceive temperature changes or close obstacles. Although visual cameras can provide image information, the image quality may be greatly affected due to the low light and complex viewing angles in the mine.

[0004] In complex mining environments, electric locomotives need to avoid dynamic targets. However, existing automation systems often rely on simple linear models or rule-based algorithms when predicting the trajectory of dynamic targets. They are unable to accurately handle the complex motion patterns of targets, especially in the case of multi-target interaction. The prediction error is large, increasing the risk of collision.

[0005] In a complex and dynamic environment like a mine, sudden or temporary obstacles often appear, which makes static path planning unable to adapt to environmental changes in real time. In addition, the dynamic adjustment strategy of the local path is relatively simple, which makes it difficult to deal with dynamic obstacles or unforeseen situations in high-speed movement, affecting the driving efficiency and safety of the electric locomotive. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides an integrated control method for unmanned driving of mine electric locomotives based on artificial intelligence, which solves the problems of insufficient perception ability, inaccurate path planning, and inaccurate dynamic target prediction during automatic driving in complex environments.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an artificial intelligence-based unmanned driving integrated control method for mine electric locomotives, comprising the following steps; S1. Multimodal environmental perception: Multimodal sensors acquire static and dynamic data of the mine environment and process the data through sensor fusion technology; S2. Real-time environment modeling: Fusing multimodal data to build a three-dimensional real-time environment model of the mine; S3, Dynamic target trajectory prediction: Predict the motion trajectory of dynamic targets based on deep learning models; S4, Path Planning: Combine static maps with dynamic perception data to calculate the global optimal path of the electric locomotive and dynamically adjust the local path during operation; S5, obstacle avoidance and decision control: Combined with the dynamic target prediction results, the speed and direction of the vehicle are adjusted in real time to achieve obstacle avoidance, and emergency obstacle avoidance is performed when an accident occurs; S6. Safety and fault tolerance: When communication is interrupted or the sensor fails, the system switches to safe mode and ensures the safety of system operation through redundant data or preset strategies.

[0008] Preferably, the multimodal sensor in step S1 includes: LiDAR, which is used to generate a 3D point cloud of the mine environment; RGB camera for collecting visual information Infrared camera, used to collect thermal imaging data; Ultrasonic sensors for detecting obstacles at close range; Inertial navigation unit, used to provide vehicle attitude, speed and acceleration information.

[0009] Preferably, the multimodal data fusion in step S2 is achieved by Kalman filtering, and the fusion formula is:

[0010] in: is the Kalman gain; is the estimated value of the current state, is the sensor measurement value, is the prediction covariance matrix, is the observation model, is the measurement noise covariance.

[0011] Preferably, the dynamic target trajectory prediction in step S3 is implemented by a deep learning model, and the prediction formula is:

[0012] in: The historical trajectory data of the target; For the future moment The probability distribution of the target location.

[0013] Preferably, the trajectory prediction model is built based on Transformer and uses a multi-head self-attention mechanism, and the formula is:

[0014] in: are query, key, and value matrices respectively; The dimension of the key.

[0015] Preferably, in step S4, the path planning is to calculate the optimal path from the starting point to the end point through an algorithm, and the path cost formula is:

[0016] in, From the starting point to the node The price, For Node Heuristic estimate of the target.

[0017] Preferably, in step S4, the local path is dynamically adjusted by optimizing the local path in combination with a dynamic window algorithm, and the objective function is:

[0018] in: is the safety distance of the path; is the driving speed; is the distance the path deviates from the global target.

[0019] Preferably, the obstacle avoidance and decision control in step S5 adjusts the vehicle behavior through a priority strategy, and the priority calculation formula is:

[0020] in: The risk level of the target; is the target speed; is the target distance.

[0021] Preferably, the emergency obstacle avoidance control in step S5 is implemented by calculating the braking acceleration, and the formula is:

[0022] in: is the current speed; It is a dynamic safety distance.

[0023] Preferably, the safety and fault tolerance in step S6 includes: When communications are interrupted, switch to redundant data mode and use the inertial navigation unit to maintain basic operations; When the sensor fails, the preset emergency stop strategy is adopted; The system regularly monitors the operating status of the modules and fixes potential problems through cloud-based diagnosis.

[0024] The present invention provides an artificial intelligence-based integrated control method for unmanned driving of mine electric locomotives. It has the following beneficial effects: 1. The present invention models and predicts the historical motion trajectory of dynamic targets through a deep learning model. Through the multi-head self-attention mechanism, the model can accurately capture the temporal correlation and multi-target interaction relationship in the target trajectory, generate the probability distribution of future trajectories, and provide important support for obstacle avoidance decisions and path optimization in complex dynamic environments.

[0025] 2. By combining dynamic target trajectory prediction with real-time perception data, the system can identify potential risks and generate safe obstacle avoidance paths in a timely manner. By adopting a priority evaluation mechanism and a dynamic speed direction adjustment strategy, the obstacle avoidance operation takes into account both efficiency and safety. In an emergency, the system triggers an emergency stop by calculating the braking acceleration to avoid serious collision accidents.

[0026] 3. The present invention achieves real-time adaptation to complex mine environments by combining global path planning with local path dynamic adjustment. Global path planning provides macro navigation directions, while local path adjustment quickly optimizes the driving path when dynamic obstacles appear, ensuring efficient driving of electric locomotives in dynamic environments and avoiding path deviations and collisions with obstacles. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a stereogram of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] Please see attached Figure 1 , an embodiment of the present invention provides an artificial intelligence-based integrated control method for unmanned driving of a mine electric locomotive, comprising the following steps; S1. Multimodal environmental perception: Multimodal sensors acquire static and dynamic data of the mine environment and process the data through sensor fusion technology; S2. Real-time environment modeling: Fusing multimodal data to build a three-dimensional real-time environment model of the mine; S3, Dynamic target trajectory prediction: Predict the motion trajectory of dynamic targets based on deep learning models; S4, Path Planning: Combine static maps with dynamic perception data to calculate the global optimal path of the electric locomotive and dynamically adjust the local path during operation; S5, obstacle avoidance and decision control: Combined with the dynamic target prediction results, the speed and direction of the vehicle are adjusted in real time to achieve obstacle avoidance, and emergency obstacle avoidance is performed when an accident occurs; S6. Safety and fault tolerance: When communication is interrupted or the sensor fails, the system switches to safe mode and ensures the safety of system operation through redundant data or preset strategies.

[0030] Specifically, by using multimodal sensors, deep learning models, path planning algorithms and safety control mechanisms, it is possible to achieve accurate path planning and obstacle avoidance for electric locomotives in dynamic environments, and ensure that the system can remain stable and safe in the face of sensor failure or other faults; In this embodiment, the environmental perception of the mine electric locomotive is accomplished through the collaborative work of multimodal sensors, using laser radar, camera, infrared sensor and inertial measurement unit to obtain static and dynamic information in the mine environment; Specifically, the LiDAR is used to provide three-dimensional information of the mine space, including distance, location of obstacles, and structure of the mine passages. The camera is used to capture surrounding image data to help identify obstacles, dynamic targets, and environmental features. The infrared sensor is used to detect temperature changes and further identify possible dynamic obstacles. According to preprocessing, the collected raw data is preprocessed; Specific: Filter and denoise the LiDAR data, remove outliers and normalize the spatial coordinates; Perform image enhancement, color normalization and background removal on images captured by RGB cameras and infrared cameras; Perform time synchronization processing on the ultrasonic sensor data and calculate the obstacle distance; Remove noise and compensate inertial navigation unit data; Sensor fusion technology is used to uniformly process the data collected by different sensors. The Kalman filter algorithm is used to fuse the different information output by the sensors, thereby improving the locomotive's perception accuracy of the mine environment. The Kalman filter estimates the locomotive's position and speed, and combines the measurement data of the laser radar, camera and infrared sensor for dynamic correction to obtain an accurate environmental model; The fusion formula of Kalman filter is:

[0031] in, is the estimated value of the state at the current moment; is the current sensor measurement value; is the estimated value of the state at the previous moment; is the Kalman gain, which indicates the credibility of the current state estimate; is the prediction covariance matrix of the previous moment; is the observation model; is the measurement noise covariance.

[0032] In this embodiment, the construction of the real-time environment model relies on the fusion of multimodal sensor data. Each sensor provides different information and has different measurement errors and noises, so it is crucial to effectively process them through data fusion technology; The system builds a three-dimensional environment model of the mine locomotive based on the fused sensor data. The model includes the static part of the mine and dynamic obstacles. The model is a grid-based three-dimensional map. The grid cell contains information about obstacles, free space, and the status of the target object at that location. Specifically, the environmental model needs to be updated in real time to reflect changes in the mine environment. Whenever new sensor data arrives, the system will promptly update the environmental model through a data fusion algorithm to ensure that the locomotive control system can make decisions based on the latest environmental information. This process ensures robustness and real-time performance in a dynamic environment. In order to further improve the accuracy of the environmental model, a joint data collaborative optimization method is adopted in this embodiment. This method comprehensively considers the characteristics of multimodal data and establishes a joint optimization model. The formula is as follows:

[0033] in: It is Prediction models for sensor data; It is Observations of various sensors; is the gradient of the sensor data, which indicates the rate at which the data changes; is the regularization coefficient for each sensor data; are model parameters.

[0034] In this embodiment, trajectory prediction is based on the analysis of multimodal perception data and deep learning models. By modeling the historical motion information of the target and combining it with environmental characteristics, the possible location and behavior trend of the target in multiple time periods in the future are calculated; Before trajectory prediction, input features are used to construct the input of dynamic target trajectory prediction, including the target's historical motion trajectory and environmental features; Historical trajectory data: Target in time The previous continuous position sequence ,in Indicates that the target is at time position, velocity and acceleration; Environmental features: information about the distribution and type of obstacles around the target obtained from the multimodal perception module; Data preprocessing: normalize historical trajectory data to eliminate unit and dimension differences and ensure the consistency of model input data; Trajectory prediction model; This example uses a Transformer-based deep learning model to predict the target trajectory. Model structure: The Transformer model consists of an encoder and a decoder. The encoder processes the input history trajectory and extracts the target motion. The feature,decoder combines the current environmental features to predict the target’s future,trajectory; Multi-head self-attention mechanism: The self-attention mechanism is used to capture the temporal correlation in the target trajectory, and its formula is:

[0035] in: query matrix; Key Matrix; Value matrix; : The dimension of the key, used for scaling to avoid the gradient vanishing problem caused by too large values; Time Series Modeling: The model uses the hidden state generated by the encoder and the decoder to predict the probability distribution of the target position at the future moment:

[0036] in The goal is to The probability distribution of the position at the time; Future trajectory prediction; Based on the output of the attention mechanism, the weight of each historical trajectory point is used to predict the target in the future The position of each time period, the prediction formula is:

[0037] in, The predicted The trajectory points of future time periods, is the weight of the corresponding historical trajectory point; Multi-objective interaction modeling; In a multi-target scenario, in order to consider the interaction between targets, this embodiment uses Social-LSTM to model the dynamic interaction of multiple targets:

[0038] in: ;Interaction characteristics between the target and its neighbor targets; Interaction weight matrix; The hidden state of the neighbor target; Trajectory uncertainty assessment; The predicted future trajectory is uncertain, so a probability distribution is generated for the prediction time period:

[0039] in, is the mean of the predicted trajectory, is the covariance matrix, which represents the uncertainty of the prediction.

[0040] Optimization and output of trajectory prediction; In the prediction phase, the model generates multiple trajectory candidates and selects the optimal trajectory as output:

[0041] in, For the The probability of candidate trajectories is calculated, and the trajectory with the largest probability is selected as the final result.

[0042] In this embodiment, in a complex mine environment, the global optimal path of the electric locomotive is calculated by combining static maps with dynamic perception data, and the local path is adjusted in real time according to dynamic obstacles or environmental changes. Through path planning, it is ensured that the electric locomotive can complete the transportation task efficiently and safely, while avoiding collisions and other potential risks; Based on the mine's infrastructure information, a static map of the mine is constructed using pre-collected LiDAR point cloud data. The static map includes the mine's tunnel layout, location information of fixed obstacles, and access restrictions in various areas; The static map is represented as a node-based map structure, where each node corresponds to a location point and the weight of the edge represents the cost of moving from one node to an adjacent node; Global path planning; Use algorithms to perform global path planning on static maps and calculate the optimal path from the starting point to the target point; The cost function of path planning is:

[0043] in: : From the starting point to the node the cumulative cost of : From the node Heuristic estimate of the cost to the target point, usually calculated using Euclidean distance or Manhattan distance; In a mine environment, heuristic estimation needs to be adjusted based on actual conditions such as tunnel width and turning radius to improve the reliability of path planning; Fusion of dynamic perception data; During the operation of the electric locomotive, the dynamic perception module obtains the information of dynamic targets and sudden obstacles in the environment in real time; The dynamic perception data is integrated with the static map to generate an updated map with dynamic obstacle markings for local path adjustment; Dynamic adjustment of local paths In this embodiment, the local path adjustment is based on a dynamic window algorithm to optimize the path in real time to avoid dynamic obstacles; The objective function of the dynamic window algorithm is:

[0044] in: : obstacle safety distance of the path; : driving speed on the path; : The degree of deviation between the path and the global goal; ; Weight coefficient, dynamically adjusted according to the mine environment; By optimizing this objective function, the electric vehicle can quickly avoid obstacles in a dynamic environment while minimizing deviations from the global path; Path tracking and error correction; The planned path is tracked and controlled to ensure that the electric locomotive travels along the predetermined path. The path tracking error is corrected by the PID controller. The control formula is:

[0045] in: The deviation between the current position and the target position; ; Proportional, differential, and integral gain coefficients are dynamically adjusted to suit different environments; The PID controller adjusts the speed and direction of the electric locomotive in real time to ensure that the vehicle maintains smooth driving in complex and dynamic environments.

[0046] Global and local path coordination; Global path planning provides the locomotive with a macroscopic navigation direction, while local path adjustment optimizes the global path based on dynamic perception data. During the path planning process, the global path and local path collaborate through an interactive mechanism: When the deviation of the local path exceeds the set threshold, the global path planning is re-triggered to ensure that the electric locomotive can return to the optimal driving path; Real-time path update mechanism; In a mine, due to the frequent dynamic changes in the environment, the path planning module needs to be regularly updated in real time: When a new obstacle is detected, a local path adjustment is immediately triggered; If the dynamic perception data indicates that the global path is no longer feasible, the global path is recalculated and synchronized with the local path adjustment module; In this embodiment, by combining the dynamic target trajectory prediction results and real-time perception data, the system can identify potential dangers during the locomotive's driving process and achieve obstacle avoidance operations by adjusting the speed and direction. The goal of obstacle avoidance and decision control is to adjust the locomotive's driving state in real time according to dynamic environmental information, avoid collisions, and ensure the vehicle's stable operation in a complex mine environment; Dynamic target priority assessment; During the operation of the electric locomotive, the system evaluates the priority of the dynamic targets based on their trajectory prediction results to determine the urgency of obstacle avoidance; The priority calculation formula is:

[0047] in: Target risk level; Target speed; Target distance; , , ;The weight coefficient is dynamically adjusted according to the characteristics of the mine environment; Through the above formula, the system can classify and sort different targets, providing a basis for the subsequent obstacle avoidance strategy selection; Obstacle avoidance path generation; Based on the priority evaluation results, the obstacle avoidance path is generated in combination with the dynamic window algorithm. The dynamic window algorithm calculates the feasible control set of the vehicle in terms of speed and direction in real time, and selects the optimal path to optimize the target; The function is:

[0048] in: : obstacle safety distance of the path; : driving speed on the path; : The degree of deviation between the path and the target point; ; Weight coefficient, adjusted according to different scenarios to optimize safety and efficiency; Through this optimization process, the system is able to generate safe and efficient obstacle avoidance paths in real time; Emergency obstacle avoidance control; When the system detects a high-risk target, it triggers the emergency obstacle avoidance mechanism The calculation formula for the braking acceleration of emergency obstacle avoidance is:

[0049] in: ; Current speed of the electric locomotive; : Dynamically calculated safety distance, adjusted in real time based on the priority of the target and the dynamic characteristics of the vehicle; If the target distance is less than the safety threshold and cannot be avoided, the system will first trigger an emergency stop to avoid collision; Dynamic speed and direction adjustment; According to the optimization results of the obstacle avoidance path, the speed and direction of the vehicle are adjusted in real time; The speed adjustment formula is:

[0050] in: : The maximum permissible speed of the vehicle; ; Direction adjustment is achieved by controlling the steering angle of the vehicle. The steering angle calculation formula is:

[0051] in, is the target point coordinate, is the current vehicle position; Coordinated control of dynamic obstacle avoidance and path planning; The system avoids obstacles while maintaining coordination with the global path: If the obstacle avoidance path deviates from the global path by more than the set threshold, the local path planning is retriggered; Ensure that after the vehicle avoids obstacles, it can smoothly return to the main path of the global plan; Real-time obstacle avoidance status monitoring and feedback; During the obstacle avoidance process, the system continuously monitors the distance between the vehicle and the obstacle and dynamically updates the obstacle avoidance status.

[0052] The obstacle avoidance strategy is adjusted through the feedback mechanism. When the target leaves the safe range of the vehicle, the system automatically releases the obstacle avoidance state and resumes normal driving.

[0053] In this embodiment, a safety and fault-tolerant mechanism is designed to ensure the continuous and stable operation of the mine electric locomotive unmanned driving system in the event of communication interruption, sensor failure or other abnormal conditions. This mechanism switches to a safe mode in abnormal conditions through redundant data, preset safety strategies and system self-diagnosis functions to maintain basic system operation or quickly respond to emergency conditions. The goal of safety and fault tolerance is to ensure the high reliability of the electric locomotive in a complex mine environment and the safety of personnel and equipment. Redundant sensors and data redundancy mechanisms; Multi-sensor redundant configuration: In system design, sensors have redundant configuration. For example, lidar, ultrasonic sensors and cameras work together, and when one sensor fails, other sensors can provide supplementary data; Data redundancy processing: Compensate for missing data from failed sensors through time series data recording and prediction models; For the position data, the extended Kalman filter is used to predict the current state of the vehicle:

[0054] in, is the state transition matrix, is the control matrix, It is noise; The environmental data is supplemented by the environmental model to ensure that the local path planning module can continue to operate Security strategies in case of communication interruption; When a communication interruption is detected, the system automatically switches to local control mode, using the locally cached global path and real-time perception data to maintain normal operation of the vehicle; If the communication interruption exceeds the set time threshold and the dynamic sensing data is insufficient, the emergency stop mechanism is triggered to avoid dangerous situations; Safe handling of sensor failures Single sensor failure handling: Ensure the integrity of key data through multimodal data fusion and redundancy mechanisms. For example, when the lidar fails, the visual information of the camera and the data of the ultrasonic sensor are used to replace the obstacle detection function.

[0055] Multi-sensor failure handling: When multiple key sensors fail and the redundancy mechanism cannot supplement the data, the system initiates an emergency parking strategy based on the vehicle's current location and environmental information; Emergency stop and safety mode switching; The triggering conditions for emergency stop include: The distance to the obstacle is less than the safety threshold; The communication interruption exceeds the maximum tolerance time; The critical sensor fails completely and cannot be supplemented by redundancy; The brake control formula for emergency stop is:

[0056] in: ; Current speed of the electric locomotive; Dynamically calculated safety distance.

[0057] After parking, the system enters safety mode, where the vehicle remains stationary and sounds the alarm while attempting to reestablish communications and restore sensor functionality; System self-diagnosis and fault recovery; The system performs self-diagnosis regularly to detect the status of sensors, communication modules and control units: Detect sensor deviations or failures through health monitoring algorithms; Monitor the quality of communications to determine if there is a risk of delay or interruption; When the fault is diagnosed and located, the system attempts to recover by restarting the software or hardware or switching redundant devices; If the fault cannot be recovered, the system will automatically report the abnormal information and prompt manual intervention; Safe Space Management and Risk Assessment Calculate the safe space between the vehicle and obstacles based on the dynamic environment model; Dynamic adjustment formula of safety space:

[0058] in, is the current distance, is the minimum safe distance; If there is insufficient safe space, slow down immediately or trigger the obstacle avoidance mechanism; The system simultaneously evaluates the risk level of the current path and environment, generates a risk score, and reports it to the central control system in a visual form.

[0059] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An integrated control method for unmanned driving of mine electric locomotives based on artificial intelligence, characterized in that: The steps include: S1. Multimodal environmental perception: Multimodal sensors acquire static and dynamic data of the mine environment and process the data through sensor fusion technology; S2, Real-time environment modeling: Fusing multimodal data to build a three-dimensional real-time environment model of the mine; S3, Dynamic target trajectory prediction: Predict the motion trajectory of dynamic targets based on deep learning models; S4, Path Planning: Combine static maps with dynamic perception data to calculate the global optimal path of the electric locomotive and dynamically adjust the local path during operation; S5, obstacle avoidance and decision control: Combined with the dynamic target prediction results, the speed and direction of the vehicle are adjusted in real time to achieve obstacle avoidance, and emergency obstacle avoidance is performed when an accident occurs; S6. Safety and fault tolerance: When communication is interrupted or the sensor fails, the system switches to safe mode and ensures the safety of system operation through redundant data or preset strategies.

2. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: The multimodal sensor in step S1 includes: LiDAR, which is used to generate a 3D point cloud of the mine environment; RGB camera for collecting visual information Infrared camera, used to collect thermal imaging data; Ultrasonic sensors for detecting obstacles at close range; Inertial navigation unit, used to provide vehicle attitude, speed and acceleration information.

3. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: In step S2, multimodal data fusion is achieved through Kalman filtering, and the fusion formula is: in: is the Kalman gain; is the estimated value of the current state, is the sensor measurement value, is the prediction covariance matrix, is the observation model, is the measurement noise covariance.

4. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: The dynamic target trajectory prediction in step S3 is achieved through a deep learning model, and the prediction formula is: in: The historical trajectory data of the target; For the future moment The probability distribution of the target location.

5. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 4 is characterized in that: The trajectory prediction model is built based on Transformer and uses a multi-head self-attention mechanism. The formula is: in: query matrix; ;Key Matrix; Value matrix; The key dimension is used for scaling to avoid the vanishing gradient problem caused by too large values; In this embodiment, the input historical trajectory data and environmental features are encoded as The importance of the target motion trajectory time segment is calculated through the attention mechanism, capturing the dynamic changes of the target motion pattern.

6. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: In step S4, the path planning is to calculate the optimal path from the starting point to the end point through an algorithm, and the path cost formula is: in, From the starting point to the node The price, For Node Heuristic estimate of the target.

7. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: In step S4, the local path is dynamically adjusted by combining a dynamic window algorithm to optimize the local path, and the objective function is: in: is the safety distance of the path; is the driving speed; is the distance the path deviates from the global target.

8. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: The obstacle avoidance and decision control in step S5 adjusts the vehicle behavior through a priority strategy, and the priority calculation formula is: in: The risk level of the target; is the target speed; is the target distance.

9. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: The emergency obstacle avoidance control in step S5 is achieved by calculating the braking acceleration, and the formula is: in: is the current speed; It is a dynamic safety distance.

10. The method for unmanned driving integrated control of mine electric locomotive based on artificial intelligence according to claim 1 is characterized in that: The safety and fault tolerance in step S6 include: When communications are interrupted, switch to redundant data mode and use the inertial navigation unit to maintain basic operations; When the sensor fails, the preset emergency stop strategy is adopted; The system regularly monitors the operating status of the modules and fixes potential problems through cloud-based diagnosis.

Images