Control system of composite intelligent maintenance robot for high-speed rail platform

Abstract

The invention relates to the technical field of intelligent maintenance of high-speed rail stations, and discloses a control system of a composite intelligent maintenance robot for a high-speed rail station, and the system comprises a visual detection module which is used for carrying out defect detection and fault recognition on high-speed rail station equipment through a deep learning algorithm; and the navigation and positioning module is used for carrying out environment mapping based on a synchronous positioning and map building technology of data fusion of the laser radar and the visual sensor, and realizing autonomous navigation and accurate positioning of the robot by combining a path planning algorithm. And by adopting the remote monitoring module, real-time monitoring, anomaly detection and remote control of the robot are realized, and the intelligence and safety of maintenance operation are improved. Through multi-sensor fusion, the state and environment of the robot are accurately sensed, and task scheduling and execution monitoring are optimized. And in combination with 5G and MQTT, stable and efficient data transmission is ensured, and the reliability of remote control is improved. Bayesian reasoning is adopted for anomaly detection and fault prediction, the fault rate is reduced, and the operation and maintenance efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent maintenance of high-speed railway platforms, and in particular to a control system of a composite intelligent maintenance robot for high-speed railway platforms. Background Art

[0002] As the scale of high-speed rail operations continues to expand, platform maintenance operations have become increasingly critical. Traditional maintenance methods mainly rely on manual inspections, where staff check the status of equipment one by one at the platform site. This method is not only labor-intensive, but also inefficient. Especially at night or in bad weather, the safety and accuracy of manual inspections are often difficult to guarantee. Manual monitoring methods are relatively simple, and usually only the operating status of equipment can be detected by the naked eye or simple instruments, which cannot achieve comprehensive perception and accurate analysis of the equipment.

[0003] Some automated maintenance equipment in the existing technology is already equipped with some sensors, but most of them rely on a single data source and cannot effectively integrate multiple sensor information. This makes it difficult for the equipment to make accurate judgments in complex environments. For example, the robot only relies on posture sensors for path planning, and it is easy to have problems such as path deviation or misoperation when the platform environment changes. Moreover, traditional communication systems generally use Wi-Fi or 4G networks. When faced with high data volume transmission, network instability and command delays often occur, affecting the response speed of remote control and the accuracy of operation.

[0004] Existing anomaly detection methods are usually based on set fixed thresholds, which trigger alarms when they exceed the range. This method lacks flexibility and is easily affected by environmental changes, resulting in false alarms. For example, when the external ambient temperature changes, the temperature sensor may falsely alarm. In addition, traditional anomaly detection methods rely on single-point judgments, cannot perform long-term data trend analysis, and are difficult to predict faults in advance, resulting in equipment problems usually being discovered after the failure occurs, increasing the passivity of operation and maintenance work.

[0005] In summary, the existing technology has certain deficiencies in data collection, environmental adaptability, communication stability and fault warning. With the development of intelligent maintenance systems, how to achieve all-round monitoring of robot status, improve the stability and real-time performance of remote control, and how to warn of faults in advance and improve operation and maintenance efficiency have become important technical issues that need to be solved urgently. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides a control system for a composite high-speed railway platform intelligent maintenance robot, which solves the problems of low manual inspection efficiency, insufficient perception capability of a single sensor, unstable remote communication and delayed fault warning in the existing high-speed railway platform maintenance method.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: A control system of a composite high-speed railway platform intelligent maintenance robot, including a visual inspection module for performing defect detection and fault identification on high-speed railway platform equipment through a deep learning algorithm;

[0008] The navigation and positioning module uses the simultaneous positioning and mapping technology based on the fusion of lidar and visual sensor data to build environmental maps, and combines the path planning algorithm to achieve autonomous navigation and precise positioning of the robot;

[0009] The task control module is used to dynamically allocate and schedule tasks based on the robot status and task priority based on the deep reinforcement learning algorithm, and optimize the task path by combining the task execution sequence, equipment status and environmental information;

[0010] The actuator module includes a six-degree-of-freedom robot arm, a fastening tool, a high-pressure water gun, and a spray device, which are used to perform cleaning, repair, and maintenance operations of the equipment respectively;

[0011] The remote monitoring module is used to collect robot status data in real time and exchange data with the control center.

[0012] Preferably, the visual inspection module combines an image segmentation algorithm to classify and annotate equipment defects such as cracks, looseness, and corrosion, identify the size, shape, and location of the defects, and train a deep learning model based on historical data to improve the generalization ability of defect recognition.

[0013] Preferably, the navigation and positioning module uses a combination of laser SLAM and visual SLAM to perform environmental modeling, and enhances positioning accuracy through multi-sensor data fusion technology, and uses A, Dijkstra or DWA algorithm to achieve robot path planning to ensure stable navigation of the robot in a dynamic environment.

[0014] Preferably, the task control module combines the robot energy consumption model, optimizes path planning through a reinforcement learning algorithm, adjusts task priority according to real-time task status, and uses a deep reinforcement learning method for dynamic scheduling to ensure the rationality and efficiency of task execution.

[0015] Preferably, the six-degree-of-freedom robotic arm of the actuator module is used for precise maintenance operations. The motion trajectory of the robotic arm is controlled based on inverse kinematics calculations and path planning algorithms, and combined with an adaptive impedance control strategy, the robotic arm can automatically adjust force control parameters when performing fastening, cleaning and spraying tasks, thereby improving the stability and safety of operations in complex environments.

[0016] Preferably, the tightening tool is used to tighten bolts and nuts of high-speed railway platform equipment, supports automatic identification and adaptation of bolts of multiple specifications, and detects the tightening status in real time in combination with a torque sensor to ensure that the tightening torque meets the equipment maintenance standards.

[0017] Preferably, the high-pressure water gun is used to clean stains and debris on the surface of the equipment, has an adjustable water spray pressure, automatically adjusts the cleaning mode according to different degrees of contamination, and combines with a visual detection module to identify the range of stains and optimize the cleaning path.

[0018] Preferably, the spraying device is used to perform protective coating spraying on the surface of the equipment, and the spraying parameters can be adjusted according to the material characteristics, and the spraying uniformity can be controlled in combination with visual feedback to ensure that the coating thickness meets the predetermined standards.

[0019] A control method for an intelligent maintenance robot comprises the following steps:

[0020] S1. Use deep learning algorithms to detect defects in high-speed railway platform equipment through the visual inspection module, and classify and mark the defect information;

[0021] S2. The navigation and positioning module combines SLAM technology and path planning algorithms to complete the station environment mapping and dynamic path planning, ensuring the robot's autonomous navigation and precise positioning in complex environments;

[0022] S3. The task control module uses reinforcement learning algorithm to optimize task scheduling based on task priority, robot status and real-time environmental data, and dynamically adjusts the execution order and path of maintenance tasks;

[0023] S4. Perform cleaning, repair and maintenance operations through the actuator module, control the six-degree-of-freedom robot arm to complete precise operations based on inverse kinematics calculation and path planning algorithm, and adjust the force control parameters according to the adaptive impedance control strategy;

[0024] S5. Realize real-time data interaction between the robot and the control center through the remote monitoring module to perform task monitoring, fault warning and remote control.

[0025] Preferably, step S3 adopts a task allocation algorithm based on deep reinforcement learning, combines task priority with real-time environmental feedback, dynamically updates the task execution order, and adopts the DDPG method to achieve adaptive adjustment of tasks to optimize task scheduling efficiency.

[0026] The present invention provides a control system for a composite high-speed railway platform intelligent maintenance robot. It has the following beneficial effects:

[0027] 1. The present invention uses a remote monitoring module to perform real-time data collection, anomaly detection and remote control on the high-speed railway platform maintenance robot, achieving the technical effect of improving the intelligent management level of the system. Compared with the existing technical solutions that rely on on-site personnel inspections, it solves the problems of delayed manual monitoring and untimely fault detection, making maintenance operations more efficient and safer.

[0028] 2. The present invention uses multi-sensor fusion technology, combined with inertial measurement units, visual sensors, etc., to achieve accurate perception of the robot's motion state and intelligent analysis of the operating environment, achieving the technical effect of optimizing task scheduling and execution monitoring. Compared with the traditional single sensor monitoring method, it solves the shortcomings of one-sided data and poor environmental adaptability, allowing the robot to cope with complex operating scenarios more autonomously.

[0029] 3. The present invention adopts a remote communication method combining 5G network and MQTT protocol to ensure the real-time and stability of data transmission, and introduces a hybrid transmission strategy to achieve the technical effect of efficient remote data interaction. Compared with the existing solutions based on Wi-Fi or 4G, it solves the problem of data delay caused by limited bandwidth and network fluctuations, making remote control more reliable and command response faster.

[0030] 4. The present invention combines the Bayesian inference model for anomaly detection and fault prediction, which can provide early warning before a fault occurs, achieving the technical effect of reducing the robot failure rate and improving maintenance efficiency. Compared with the traditional anomaly detection method based on fixed thresholds, it solves the shortcomings of high false alarm rate and inability to perform long-term trend analysis, making robot operation and maintenance more intelligent and reducing unplanned downtime. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a schematic diagram of the overall framework of the control system of the high-speed railway platform intelligent maintenance robot in the present invention;

[0032] Figure 2 This is a system architecture diagram of the high-speed railway platform maintenance robot in the present invention;

[0033] Figure 3 A schematic diagram of robot motion state perception and multi-sensor fusion in the present invention;

[0034] Figure 4 It is a schematic flow chart of the steps of the control method of the high-speed railway platform intelligent maintenance robot in the present invention. DETAILED DESCRIPTION

[0035] 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.

[0036] Example:

[0037] Please see attached Figure 1 -Attached Figure 3 The embodiment of the present invention provides a control system for a composite high-speed railway platform intelligent maintenance robot, comprising:

[0038] The visual inspection module is used to detect defects in high-speed railway platform equipment through deep learning algorithms, identify faults such as cracks, looseness, and corrosion, use deep convolutional neural networks (CNNs) for target detection and classification, and combine Kalman filtering algorithms to fuse multi-sensor data to improve the accuracy of defect identification;

[0039] In the maintenance work of high-speed railway platforms, accurate and efficient equipment inspection is a key link to ensure the safety of facilities and stable operation. The traditional manual inspection method is limited by people's subjective judgment, prone to false detection or missed detection, and time-consuming, and it is difficult to meet the intelligent maintenance needs of modern high-speed railway stations. Therefore, the present invention provides a high-speed railway platform maintenance robot control system based on intelligent visual detection, which realizes automated and real-time defect detection of platform equipment through deep learning and multi-sensor fusion technology. The system can identify structural defects such as cracks, corrosion, looseness, and combine data processing and filtering algorithms to improve detection accuracy and stability, and provide reliable data support for subsequent maintenance work.

[0040] In this embodiment, the visual inspection module uses a high-resolution industrial camera and a depth sensor, combined with a deep convolutional neural network (CNN) for image processing and target recognition. The system mainly includes steps such as data acquisition, preprocessing, feature extraction, defect classification, defect location and measurement.

[0041] First, the robot's high-resolution camera acquires visible light images of the surface of the platform equipment, while the depth sensor acquires three-dimensional information of the target area. To adapt to different lighting environments, the camera has automatic exposure (AE) and automatic white balance (AWB) functions, combined with adaptive gain control to optimize image quality. In some embodiments, multispectral or infrared imaging devices can be used to improve detection capabilities in low-light environments.

[0042] In order to improve the accuracy of subsequent analysis, the system performs denoising, enhancement and geometric correction on the collected images. Specifically, image denoising uses a denoising method based on Gaussian filtering to remove high-frequency noise. The filtered image I ′ (x,y) is represented as:

[0043]

[0044] Among them, I(x,y) is the original image, G(i,j) is the two-dimensional Gaussian kernel function, which is defined as follows:

[0045]

[0046] Among them, σ is the standard deviation of the Gaussian distribution, and k represents the size of the filter window.

[0047] In addition, to enhance the contrast of defective areas, the system uses a CLAHE algorithm to adjust the brightness of local areas, making tiny cracks and corrosion marks more obvious.

[0048] After image preprocessing, the system uses deep convolutional neural network (CNN) to extract defect features. Generally, CNN consists of multiple convolutional layers, pooling layers, and fully connected layers. The convolutional layer is used to extract low-level features such as edges and textures, while the pooling layer reduces the data dimension and improves computing efficiency.

[0049] Assume that the input feature map is F l , the convolution operation can be expressed as:

[0050] F l+1 =σ(W*F l + b);

[0051] Among them, W is the convolution kernel, * represents the convolution operation, b is the bias term, and σ is the activation function (such as ReLU).

[0052] In some embodiments, in order to improve the robustness of detection, the system uses multi-scale feature extraction technology, that is, constructing a pyramid network (FPN) to fuse features of different scales, so that the system can simultaneously detect large-scale structural defects (such as deformation) and small-scale detail defects (such as microcracks). The calculation method of pyramid feature fusion is as follows:

[0053]

[0054] Among them, F s represents the feature map of the sth layer, α s is the weight coefficient, α s is the number of layers of the pyramid.

[0055] After defect identification, the system uses the target segmentation algorithm to accurately mark the defect area. Generally, a semantic segmentation network with a U-Net structure is used to segment the defect area at the pixel level and calculate parameters such as crack length and width.

[0056] The crack length L and width W can be calculated by pixel scale conversion:

[0057]

[0058] Among them, d i is the pixel spacing along the center line of the crack, w i is the width of the crack at each location, N is the total number of pixels of the crack, and M is the number of sampling points of the crack.

[0059] In some cases, the system also combines deep learning with traditional edge detection algorithms (such as the Canny algorithm) to improve the accuracy of crack detection. The gradient calculation method of the Canny operator is as follows:

[0060]

[0061] Among them, G x and G y are the gradients of the image in the x and y directions, respectively.

[0062] In order to further improve the detection accuracy, the system combines the Kalman filter algorithm to optimize the detection results. The state prediction equation of the Kalman filter is:

[0063] X k|k-1 =AX k-1 +BU k +W k ;

[0064] P k|k-1 =AP k-1 A T +Q;

[0065] Among them, X k is the current state estimate, A is the state transfer matrix, P k is the error covariance matrix, Q is the process noise covariance matrix, U k is the control input, W k is the process noise.

[0066] In addition, in order to predict the development trend of defects, the system uses a long short-term memory network (LSTM) to analyze historical data. Assume that the defect feature vector is X t , the state update equation of LSTM is as follows:

[0067] h t =σ(W hX t +U h h t-1 +b h );

[0068] Among them, h t is the current hidden state, W h and U h is the weight matrix, b h is the bias term.

[0069] In one possible implementation, the system analyzes the crack growth rate through historical data and predicts its possible future growth trend to plan maintenance plans in advance. For example, assuming that the crack growth model is:

[0070] L(t) = L0 + vt;

[0071] Where L0 is the initial crack length, v is the crack growth rate, and t is time.

[0072] This embodiment realizes intelligent defect detection of high-speed railway platform equipment through deep learning, multi-sensor fusion, SLAM and Kalman filtering technologies. The system can operate efficiently in complex environments, improve the accuracy and reliability of detection, and provide strong technical support for the intelligent maintenance of high-speed railway platforms.

[0073] The navigation and positioning module uses the simultaneous localization and mapping (SLAM) technology based on the fusion of LiDAR and visual sensor data to build environmental maps, and combines the path planning algorithm to achieve autonomous navigation and precise positioning of the robot. It uses particle filtering and visual inertial odometry (VIO) to fuse multi-source data, and corrects positioning errors through graph optimization algorithms.

[0074] In the maintenance work of high-speed railway platforms, the accuracy of the navigation and positioning system directly affects the efficiency and accuracy of the maintenance robot. In order to ensure that the robot can accurately locate and navigate in complex environments, the present invention combines multi-sensor information fusion technology, and realizes high-precision spatial positioning and path planning through the collaborative work of multiple sensors such as vision, laser radar (LiDAR), and inertial measurement unit (IMU). Compared with traditional positioning technology, the present invention proposes a navigation and positioning method based on visual SLAM (simultaneous localization and mapping) and deep sensor fusion, which can maintain stable and real-time positioning accuracy in the dynamic and complex high-speed railway platform environment.

[0075] In this embodiment, the positioning system carried by the robot integrates multiple devices such as laser radar, visual sensors and inertial sensors. The laser radar generates high-precision environmental point cloud data by scanning the platform environment, providing long-distance environmental information; the visual sensor extracts landmark features through image information to provide short-distance positioning accuracy; the inertial measurement unit (IMU) provides dynamic state information by detecting the acceleration and angular velocity of the robot. Through the data fusion of these sensors, the system can achieve efficient positioning and path planning in a changing environment.

[0076] Generally, the system performs positioning through the following steps: First, the point cloud data obtained by the lidar is synchronized with the image taken by the visual sensor, and the key points are extracted through the visual SLAM algorithm and matched with the previously established map. During the matching process, the Kalman filter is used to correct the errors of each sensor to ensure the accuracy of positioning.

[0077] Specifically, the application of the visual SLAM algorithm in the present invention includes processes such as image feature extraction, feature matching, pose estimation, and map updating. In the image feature extraction stage, the system uses a deep learning network to extract key feature points in the image and matches them with previously constructed map feature points. After feature matching, based on the matching results, the system performs pose estimation and calculates the current pose increment of the robot. In some embodiments, assuming that the current pose of the robot is T k , then the update of the pose can be expressed as:

[0078] T k =T k-1 ·exp(ξ k );

[0079] Among them, ξ k represents the position increment of the robot at time k, and exp(·) is the exponential mapping from Lie algebra to Lie group. This update process ensures that the position of the robot on the map can be accurately updated as the movement changes.

[0080] As an option, the present embodiment also combines the three-dimensional information of the depth sensor, and further improves the positioning accuracy by combining the point cloud data with the visual image. In a three-dimensional environment, the depth sensor can provide depth information for each pixel, and by combining it with the visual image, the system can achieve more accurate positioning. In the sensor fusion stage, the point cloud data of the lidar and the depth information of the visual image are processed by weighted fusion, and the fused result is used to accurately update the robot's position and posture.

[0081] In order to further improve the stability of positioning, this embodiment combines the Kalman filter algorithm to optimize the data of multiple sensors. The Kalman filter algorithm achieves higher precision positioning by estimating the state of sensor data and updating the error covariance matrix. Specifically, the state update equation of the Kalman filter is:

[0082] X k|k-1 =AX k-1 +BU k +W k ;

[0083] P k|k-1 =AP k-1 A T +Q;

[0084] Among them, X k is the current state estimate, A is the state transfer matrix, P k is the error covariance matrix, Q is the process noise covariance matrix, U k is the control input, W k is process noise. By updating the state at each moment, Kalman filtering can effectively reduce the impact of sensor noise on positioning accuracy.

[0085] After obtaining the precise position of the robot, the system can perform path planning and navigation based on the environmental map. During the path planning process, the system first generates the optimal path based on the A algorithm or Dijkstra algorithm through the map. During the path planning process, the system takes into account the obstacles on the platform, the layout of the station, and possible dynamic environmental changes to ensure that the robot can travel smoothly in a complex environment.

[0086] Specifically, assuming the current position of the robot is T k , the target position is T target , then the objective function in the path planning process can be expressed as:

[0087]

[0088] Among them, T i is the pose of each point in the path, α i is the distance weight to obstacles or inaccessible areas, and n is the number of points in the path. By minimizing this objective function, the system can generate the optimal path.

[0089] During navigation, the robot may encounter dynamic obstacles (such as pedestrians, vehicles, etc.). In order to improve the dynamic obstacle avoidance capability of the system, this embodiment combines real-time obstacle detection and avoidance algorithms. Through the laser radar and visual sensors, the system can detect obstacles in front of the robot in real time and adjust the path through the dynamic obstacle avoidance algorithm. Assume that the distance of the obstacle at the current moment is dobs , then the dynamic adjustment rule in the obstacle avoidance process can be expressed as:

[0090] new path = current path + ΔT;

[0091] Among them, ΔT is the position increment after obstacle avoidance, currentpath is the current path, and newpath is the new path after obstacle avoidance.

[0092] In some embodiments, the system can achieve sub-centimeter positioning accuracy by fusing visual SLAM with data from multiple sensors such as lidar and IMU. Through Kalman filtering and other optimization algorithms, the system can further reduce sensor errors and ensure positioning stability. Specifically, through multi-sensor data fusion, the system can effectively handle errors caused by factors such as sensor noise and dynamic environmental changes, ensuring the stable operation of the robot in complex environments such as high-speed rail platforms.

[0093] In summary, this embodiment realizes a high-precision and high-stability navigation and positioning system for high-speed railway platform maintenance robots through multi-sensor fusion, visual SLAM, Kalman filtering and other technologies. The system can achieve accurate positioning in complex and dynamic environments, providing reliable technical support for robots to complete maintenance tasks.

[0094] The task control module is used to dynamically allocate and schedule tasks based on the deep reinforcement learning algorithm according to the robot status and task priority, optimize the task path by combining the task execution order, device status and environmental information, and use the deep deterministic policy gradient (DDPG) method to dynamically learn and update task allocation, and automatically adapt to task adjustments in complex environments;

[0095] In the maintenance work of high-speed railway platforms, the task control module is the core part to ensure that the robot can perform tasks efficiently. This module not only needs to handle complex task scheduling, but also needs to work closely with the navigation and positioning system to ensure that the robot can complete various maintenance tasks according to the predetermined plan. In the present invention, the task control module is based on a multi-level control architecture, and works together at the three levels of perception, decision-making and execution to achieve intelligent scheduling and execution of tasks. Compared with traditional rule-based task control, the present invention combines methods such as reinforcement learning and adaptive planning to make task control more intelligent and able to adapt to dynamic changes in the platform environment.

[0096] In this embodiment, the task control module first decomposes the overall maintenance task to generate executable subtasks. In general, the task decomposition process is optimized based on factors such as task priority, operation area, and execution time. Specifically, assuming that the maintenance task set is T = {T1, T2, ..., T n}, where Ti represents the i-th task, then the goal of task scheduling is to minimize the total execution time:

[0097]

[0098] Among them, C i Represents task T i By analyzing the task dependencies, the system establishes the constraints for task scheduling:

[0099]

[0100] Among them, S j For task T j The start time of It is a set of task dependencies. By optimizing task scheduling, we can ensure that each task can be completed in the shortest time.

[0101] As an option, in some embodiments, task scheduling is dynamically optimized using a reinforcement learning method. The system uses a deep reinforcement learning (DRL) model to dynamically adjust the order of task execution based on the robot's current state and task priority. Specifically, define the state s t is the task currently performed by the robot, action a t For the next task selected, the reward function R t Assuming it is the profit after the task is completed, the intelligent task scheduling strategy can be expressed as:

[0102]

[0103] Among them, γ is a discount factor, which is used to weigh short-term benefits and long-term benefits. Through continuous training of reinforcement learning, the system can learn the optimal task scheduling strategy and improve task execution efficiency.

[0104] After the task decomposition and scheduling are completed, the task control module adjusts the execution strategy in real time according to the current position and environmental status of the robot. In general, the task execution process includes task instruction parsing, subtask dispatching, status monitoring and other links. The robot executes the corresponding operation instructions based on the posture information provided by the navigation system. In one possible implementation, the task control module implements task execution process control based on a finite state machine (FSM). Assume that the task state set is S = {s1, s2, ..., s m}, the state transition is determined by the input signal I = {i1,i2,...,i k} is triggered, the state transition function of task execution can be expressed as:

[0105] S t+1 =f(S t ,I t );

[0106] Among them, S t is the current state, I t is the input signal, and f(·) is the state transition function. Through the state management of FSM, the task control module can ensure the stability and reliability of the task execution process.

[0107] Specifically, during the robot's task execution, the task control module continuously monitors the task execution status and makes adjustments through the feedback mechanism. As an option, this embodiment adopts an adaptive control method to adjust the task plan based on the real-time feedback of task execution. Assuming that the error of task execution is e(t), the error correction control law can be expressed as:

[0108]

[0109] Among them, K p , K i , K d To control the gain parameters, by adjusting these parameters, the system can optimize the task execution process and improve the task completion rate.

[0110] In the actual operation process, the execution of the task may be disturbed by the external environment, such as obstacles, equipment failures, etc. In order to ensure the smooth progress of the task, this embodiment introduces an abnormality detection and recovery mechanism. In general, anomaly detection is based on the change pattern of sensor data. For example, in some embodiments, the robot detects its own motion state through IMU. When an abnormal acceleration change is detected, the system triggers the abnormality recovery mechanism.

[0111] As an option, this embodiment combines Bayesian reasoning to perform abnormality judgment. Assuming that the sensor measurement data is X and the abnormal event is H, based on the Bayesian theorem, the abnormal probability is calculated as follows:

[0112]

[0113] Among them, P(X|H) is the distribution of measurement data when an anomaly occurs, P(H) is the prior probability of an abnormal event, and P(X) is the total probability of measurement data. By calculating the probability of anomaly, when P(H|X) exceeds the set threshold, the mission control module triggers the anomaly handling strategy.

[0114] During the abnormal recovery process, the system first tries to automatically adjust the task execution parameters to restore normal operation. For example, if the path is blocked, the task control module calls the path planning algorithm to regenerate a feasible path. If the automatic recovery fails, the system sends an alarm message to the operation and maintenance personnel and waits for manual intervention.

[0115] The task control module can support the collaborative work of multiple robots. In a multi-robot system, task control needs to consider issues such as task allocation and communication coordination. This embodiment adopts a distributed task scheduling method so that multiple robots can collaboratively complete tasks.

[0116] Specifically, the task control module performs task allocation based on game theory. Assume that there are N robots, each robot R i The utility function is U i , the task allocation goal is to maximize the overall utility:

[0117]

[0118] Among them, U i Affected by factors such as task completion time and energy consumption. In one possible implementation, task allocation is based on a market bidding mechanism, where each robot bids for tasks and tasks are ultimately allocated based on the bidding results. This approach can effectively improve task execution efficiency and ensure the rational allocation of resources.

[0119] In summary, this embodiment realizes the functions of task decomposition, scheduling, execution, exception handling and collaborative control through a multi-level task control architecture. By combining reinforcement learning, adaptive control, Bayesian reasoning and other methods, the system can efficiently complete maintenance tasks in complex dynamic environments, improve the robot's task execution capability, and provide a stable and reliable task control solution for high-speed railway platform maintenance operations.

[0120] The actuator module includes a six-degree-of-freedom robot arm, a fastening tool, a high-pressure water gun and a spray device, which are used to perform cleaning, repair and maintenance operations of the equipment. The six-degree-of-freedom robot arm realizes path planning based on the inverse kinematics algorithm, and adopts an adaptive impedance control strategy to adjust the force control parameters in real time according to environmental feedback to ensure the accuracy and stability of the maintenance operation;

[0121] In high-speed railway platform maintenance operations, the actuator module is a key component to ensure that the robot can perform various tasks accurately and efficiently. This module needs to ensure that the robot can complete a variety of operations in a complex platform environment, such as cleaning, inspection, and maintenance. The actuator module directly affects the accuracy, speed, and stability of task execution. In conjunction with the scheduling strategy of the task control module, the actuator module can perform flexible and efficient operations according to task requirements.

[0122] In this embodiment, the actuator module includes multiple mechanical arms, drive systems and sensor modules, which work together through an intelligent control system to achieve high-precision execution of maintenance tasks. The module can automatically select the most appropriate execution method according to task requirements, environmental conditions and robot status to ensure the smooth completion of the task.

[0123] Generally, the actuator module includes multiple working units, including the manipulator unit, the drive unit and the sensor feedback unit. Specifically, the manipulator unit operates flexibly according to the task requirements, the drive unit is responsible for providing the power required for movement, and the sensor feedback unit monitors the working status of the manipulator and equipment in real time for dynamic adjustment. Each unit works together through an efficient communication mechanism to ensure the accuracy and efficiency of task execution.

[0124] As an option, in this embodiment, the robot arm unit adopts a multi-degree-of-freedom design to achieve complex motion control. The movement of each robot arm is controlled by the drive system through motors, gears and other devices, and the movement accuracy of the robot arm is adjusted in combination with sensor feedback. In order to improve execution efficiency and flexibility, the end effector of the robot arm is designed with a modular structure, which can be quickly replaced according to task requirements.

[0125] During the task execution, the actuator module needs to ensure that the robot arm can accurately complete the specified action. Generally, motion control uses an algorithm based on inverse kinematics (IK) to ensure that the end effector can reach the target position by calculating the angles of each joint of the robot arm in real time.

[0126] Specifically, let the target position of the end effector of the robot arm be P target , the joint angles of the robot are θ1,θ2,...,θ n , then the inverse kinematics equation is:

[0127] P target =f(θ1,θ2,...,θ n );

[0128] Among them, f(·) is the kinematic model of the robot arm. By solving this equation, the specific angle θ of the robot arm joint can be obtained. i , thereby controlling the robot to reach the target position. In order to improve the accuracy, iterative optimization methods such as gradient descent are usually used to adjust the motion trajectory to reduce the error.

[0129] In some embodiments, in order to further improve the execution accuracy, an adaptive control algorithm is used in this embodiment to dynamically adjust the control parameters according to the error between the current position and the target position of the robot. Assuming that the current position error is e(t) and the control input is u(t), the control law can be expressed as:

[0130] u(t)=-Ke(t);

[0131] Among them, K is the adaptive gain. By adjusting the K value, the control system can effectively reduce the execution error and improve the accuracy of task completion.

[0132] During the execution of the mission, the actuator module not only needs to complete the predetermined actions, but also needs to make real-time adjustments based on the instructions of the mission control module and environmental changes. Generally, the execution of the mission will be affected by external factors, such as environmental changes, equipment failures, etc. At this time, the actuator needs to respond quickly and adjust the execution strategy.

[0133] Specifically, this embodiment adopts a dynamic adjustment mechanism based on sensor feedback. The sensor feedback unit monitors the real-time status of the robot arm and transmits the data to the execution control system. When the system detects execution errors or external interference, it will automatically adjust the execution strategy. For example, during the execution of a task, if it is detected that the motion trajectory of the robot arm deviates from the target path, the system will correct the deviation by correcting the motion control parameters to ensure the smooth progress of the task.

[0134] As an option, the sensor feedback unit includes force / torque sensor, visual sensor, acceleration sensor, etc. The force / torque sensor can detect the force applied to the robot arm during the execution of the task and adjust the execution force in time; the visual sensor can help the robot identify and avoid obstacles to ensure the safety of operation; the acceleration sensor can monitor the movement state of the robot arm in real time and provide dynamic adjustment data.

[0135] During the execution of a task, the actuator module may fail, such as damage to the robotic arm, failure of the drive system, etc. Therefore, in order to ensure the stability of task execution, this embodiment introduces a fault detection and recovery mechanism. In general, fault detection relies on real-time monitoring of the device status by sensors. For example, when the torque of the robotic arm exceeds the preset range, the system will trigger a fault alarm and start a recovery mechanism based on the fault type.

[0136] As an option, this embodiment combines a fault diagnosis algorithm to predict potential faults by analyzing sensor data. Let the sensor measurement data be X and the fault event be H, then the fault detection probability P(H|X) can be calculated by Bayesian reasoning:

[0137]

[0138] Among them, P(X|H) is the distribution of sensor data when the fault occurs, P(H) is the prior probability of the fault, and P(X) is the total probability of the measured data. When P(H|X) exceeds the set threshold, the system will initiate corresponding recovery measures, such as automatic shutdown, alarm notification, etc.

[0139] During the recovery process, the system will first try to automatically repair or adjust the task execution parameters, such as adjusting the operation mode of the robot arm or replacing the damaged execution unit. If the automatic recovery fails, the system will complete the troubleshooting through remote monitoring and operation and maintenance personnel.

[0140] In some embodiments, the task control module works in conjunction with multiple execution agencies to achieve multi-task parallel execution. Specifically, when multiple execution agencies work in conjunction, the task control module allocates tasks according to different task requirements and coordinates the work of each execution agency through communication. Each execution agency independently executes the corresponding subtask according to the task requirements and execution status, and feeds back the execution results to the task control module.

[0141] In a multi-actuator system, task scheduling needs to take into account factors such as the dependencies between tasks, the status of each actuator, and task priority. Through efficient scheduling and collaborative work, multiple actuators can jointly complete complex maintenance tasks and improve overall execution efficiency.

[0142] In summary, this embodiment ensures that the actuator module can complete maintenance tasks efficiently and accurately through precise motion control, sensor feedback and intelligent scheduling mechanism. Through dynamic adjustment and fault recovery mechanism, the actuator can adapt to complex and changing task environments and ensure the smooth completion of tasks. The close cooperation between this module and the task control module further improves the efficiency and stability of high-speed railway platform maintenance operations, and provides important technical support for future intelligent high-speed railway maintenance systems.

[0143] Remote monitoring module, used to collect robot status data in real time and interact with the control center, supports 5G communication protocol, and performs remote control, task monitoring and fault diagnosis based on MQTT and HTTP / 2 protocol

[0144] In the maintenance of high-speed railway platforms, the remote monitoring module is responsible for real-time monitoring and management of the robot's operating status, task execution, and fault information. This module can not only improve the safety of maintenance operations, but also provide data support for task scheduling, actuator control, and fault recovery. The remote monitoring module works in conjunction with the task control module and actuator module to achieve remote data collection, status analysis, abnormal alarm, and operation and maintenance management, ensuring the stability and intelligent operation of the system.

[0145] In this embodiment, the remote monitoring module includes a data acquisition unit, a communication unit, a monitoring and analysis unit, and a remote control terminal. The data acquisition unit is used to obtain data from various sensors of the robot, including motion status, task execution progress, and equipment operation status. The communication unit is responsible for uploading the collected data to a remote server or cloud system and receiving control instructions. The monitoring and analysis unit evaluates the robot's operating status based on historical data and real-time data, and performs anomaly detection and prediction. The remote control terminal provides an interactive interface to support manual intervention and remote operation.

[0146] In general, the remote monitoring module obtains the robot's operating status data through a variety of sensors. Specifically, the data acquisition unit includes environmental sensors, posture sensors, force / torque sensors, temperature sensors, etc. Each sensor is used to monitor the robot's surrounding environment, the movement status of the robotic arm, the execution force, and the temperature changes inside the device.

[0147] In one possible implementation, the robot's posture is tracked in real time by an inertial measurement unit (IMU). Let the robot's posture in three-dimensional space be denoted as q(t), which includes the position p(t) and the attitude angle φ(t):

[0148] q(t)=[p(t),φ(t)];

[0149] Among them, p(t) is the displacement vector, φ(t) is the posture expressed by Euler angles. The inertial measurement unit obtains the acceleration a of the robot in different axes through the accelerometer and gyroscope. x , a y , a z and angular velocity ω x ,ω y ,ω z The data fusion algorithm is used to estimate the changing trend of q(t) to ensure that the remote monitoring system can accurately grasp the robot status.

[0150] As an option, the remote monitoring module combines visual sensors to obtain images of the work site and analyzes the robot's surrounding environment through image processing algorithms. For example, the image data I (x, y, t) obtained by the camera can be used to identify key information such as platform obstacles and the scope of the work area through feature extraction and pattern recognition, providing support for task scheduling.

[0151] During remote monitoring, the communication unit is responsible for data transmission and command issuance. Generally, the unit uses wireless communication methods such as 5G, Wi-Fi or industrial Ethernet to ensure the real-time and stability of data transmission.

[0152] Specifically, in this embodiment, the communication unit of the robot interacts with the remote server through the MQTT (Message Queuing Telemetry Transport) protocol. Suppose the data collected by the robot is D(t), and the data that the remote server expects to receive is D server (t), the data transmission model can be expressed as:

[0153] D server (t) = T(D(t));

[0154] Among them, T(·) represents the data transmission process, including data encoding, compression and encryption operations. In order to improve the reliability of communication, the remote monitoring module adopts a hybrid transmission method combining TCP / UDP to ensure that the control instructions can be issued with low latency and that the large amount of monitoring information can be uploaded stably.

[0155] As an option, in high bandwidth demand scenarios, this embodiment uses 5G network for data transmission, taking advantage of 5G's ultra-low latency characteristics (generally less than 1ms) to ensure rapid response to remote monitoring instructions. In addition, the historical data stored in the remote server can be statistically processed through big data analysis methods to optimize robot task planning.

[0156] During the execution of the task, the remote monitoring module needs to detect abnormal situations and trigger the alarm mechanism in time. Generally, anomaly detection is based on real-time analysis of sensor data, such as detecting abnormal movement of the robot arm and drive system failure.

[0157] Specifically, suppose a key parameter x(t) (such as motor temperature, actuator torque, etc.) should satisfy the setting range [x min ,x max ], then the anomaly detection condition can be expressed as:

[0158]

[0159] If the condition is met, the alarm signal A(t) is triggered and the abnormal event is recorded. In order to improve the accuracy of abnormality detection, the present embodiment adopts a fault prediction model based on Bayesian reasoning to calculate the probability of fault occurrence P(H|X). When the probability exceeds the set threshold, the system issues an early warning to reduce the impact of the fault.

[0160] In one possible implementation, the alarm mechanism includes two modes: local alarm and remote notification. The local alarm alerts the on-site personnel through buzzers and warning lights, while the remote notification sends alarm information to maintenance personnel through SMS, email or mobile applications to ensure that abnormal situations can be handled in a timely manner.

[0161] In some embodiments, the remote monitoring module is not only responsible for data monitoring, but also supports remote control and manual intervention. The remote control terminal provides a visual interface, allowing operation and maintenance personnel to remotely control the robot and adjust the task execution strategy.

[0162] As an option, in this embodiment, the remote control terminal adopts a dual interface design of Web terminal + mobile terminal to support operation requirements in different scenarios. Through the remote interface, the operation and maintenance personnel can view the status of the robot and send control instructions when necessary, such as pausing the task, replanning the path, etc.

[0163] In order to ensure the safety of operation during remote control, this embodiment introduces an access rights management mechanism, using user identity authentication (such as OAuth2.0) to ensure that only authorized personnel can control the robot. At the same time, the system will record all remote operation logs for subsequent analysis and optimization.

[0164] In a multi-robot system, the remote monitoring module needs to support the status management of multiple robots running simultaneously. Generally, the task control module assigns independent tasks to different robots, while the remote monitoring module monitors the status of each robot and provides global data analysis.

[0165] Specifically, in a multi-robot scenario, the status data S of each robot i (t) will be uploaded to the remote server and form the global state matrix S total (t):

[0166] S total (t)=[S1(t),S2(t),...,S N (t)];

[0167] Where N is the total number of robots. Based on this data, the remote monitoring system can optimize tasks, such as adjusting task scheduling strategies and avoiding path conflicts between robots.

[0168] In summary, the remote monitoring module of this embodiment realizes the intelligent management of the high-speed railway platform maintenance robot through functions such as data collection, wireless communication, anomaly detection, and remote control. Combined with real-time status monitoring and remote control capabilities, the remote monitoring module can ensure the safety, stability, and efficiency of the robot's task execution, providing technical support for the further optimization of the intelligent maintenance system.

[0169] Please refer to the attached Figure 4 , a control method of an intelligent maintenance robot comprises the following steps:

[0170] S1. Use deep learning algorithms to detect defects in high-speed railway platform equipment through the visual inspection module, and classify and mark the defect information;

[0171] In an embodiment of the present invention, the visual inspection module is mainly composed of a deep neural network (DNN) and a target detection algorithm. Generally, a convolutional neural network (CNN) is used to extract image features, and a region proposal network (RPN) is combined to detect and classify defective targets.

[0172] As an option, in the inspection task, target detection models such as YOLO (You Only Look Once) or Faster R-CNN can be used to improve the accuracy and real-time performance of defect identification. Specifically, the YOLO network can complete target positioning and category recognition in a single forward propagation process, which is suitable for maintenance scenarios with high real-time requirements.

[0173] Defect information classification uses Softmax classifier for multi-category discrimination, and combines with the fully connected layer to output defect category labels. Position labeling uses the bounding box regression algorithm, which uses the least squares optimization method to accurately regress the position coordinates of the target defect, allowing the robot to perform targeted processing on different types of defects.

[0174] S2. The navigation and positioning module combines SLAM technology and path planning algorithms to complete the station environment mapping and dynamic path planning, ensuring the robot's autonomous navigation and precise positioning in complex environments;

[0175] The present invention adopts simultaneous localization and mapping (SLAM) technology, combined with laser radar (LiDAR) and visual sensors, to achieve station environment mapping. Specifically, the laser radar is used to scan the surrounding environment and generate a point cloud map, and the visual sensor is used to obtain environmental texture information. The combination of the two can improve the accuracy of environmental mapping.

[0176] In some embodiments, the SLAM algorithm uses an extended Kalman filter (EKF-SLAM) or a graph optimization method (Graph-SLAM) for pose estimation to reduce cumulative errors and improve navigation accuracy. In terms of path planning, the present invention introduces the D*Lite algorithm for global path planning and combines the artificial potential field method (APF) for local obstacle avoidance to improve the robot's autonomous operation capability in a dynamic environment.

[0177] S3. The task control module uses reinforcement learning algorithm to optimize task scheduling based on task priority, robot status and real-time environmental data, and dynamically adjusts the execution order and path of maintenance tasks;

[0178] In some embodiments of the present invention, task scheduling uses reinforcement learning (RL) methods to improve the autonomous optimization capability of scheduling strategies. Generally, a deep Q network (DQN) or proximal policy optimization (PPO) algorithm is used to train strategies in a simulated environment, so that the robot can dynamically adjust the execution order under different task conditions.

[0179] Specifically, task scheduling can be defined as a Markov decision process (MDP), where the task state is S, the action space is A, and the state transition probability is P(S ′ |S,A), and the reward function is R(S,A). By iteratively updating the Q-value function, the task scheduling strategy is optimized so that the robot can complete the optimal task allocation in the shortest time.

[0180] As an option, the attention mechanism can be combined to optimize task scheduling, so that the scheduling model can dynamically pay attention to key status information during task execution and improve the accuracy of task priority judgment.

[0181] S4. Perform cleaning, repair and maintenance operations through the actuator module, control the six-degree-of-freedom robot arm to complete precise operations based on inverse kinematics calculation and path planning algorithm, and adjust the force control parameters according to the adaptive impedance control strategy;

[0182] In an embodiment of the present invention, the motion control of the six-degree-of-freedom robotic arm adopts inverse kinematics (IK) calculation, and the joint angles are solved by the Jacobian Matrix, so that the robotic arm can accurately reach the target position.

[0183] As an option, in high-precision manipulation tasks, the Newton-Raphson iteration method or the Pseudo-Inverse Jacobian method can be used to improve the solution accuracy. Specifically, let the end position of the robot be x and the joint angle be q, then the inverse kinematics equation is expressed as:

[0184] q k+1 =q k +J + (q k )(x d -x k );

[0185] Among them, J + (q k ) is the pseudo-inverse of the Jacobian matrix, x d is the target pose.

[0186] In terms of force control strategy, the present invention adopts adaptive impedance control (Adaptive Impedance Control), which adjusts the stiffness and damping parameters of the robot arm so that it can adaptively adjust the applied force when contacting surfaces of different materials, thereby improving the flexibility of the operation.

[0187] S5. Realize real-time data interaction between the robot and the control center through the remote monitoring module to perform task monitoring, fault warning and remote control.

[0188] The remote monitoring module uses 5G+MQTT protocol for data transmission to ensure low latency and high reliability of remote control. Generally, the robot interacts with the cloud server through the IoT Gateway to achieve remote monitoring of the task execution status.

[0189] As an option, during the data transmission process, the present invention adopts an adaptive bandwidth allocation algorithm to dynamically adjust the data transmission rate according to the task importance and network load to improve network utilization. In some embodiments, edge computing can be combined to move part of the data processing tasks to the robot side to reduce the network communication burden and improve the system response speed.

[0190] In terms of fault warning, the present invention adopts Bayesian Inference Model to predict the equipment status. Specifically, assuming that the robot operation status is X and the fault event is F, the fault probability is calculated as follows:

[0191]

[0192] Where P(F|X) represents the probability of a failure in state X, P(X|F) is the probability of observing state X in the event of a failure, and P(F) and P(X) are the prior probability of failure and the marginal probability of the state, respectively.

[0193] As an option, fault prediction can also be combined with long short-term memory networks (LSTM) for time series analysis to predict possible future equipment failures based on historical sensor data and improve warning accuracy.

[0194] 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. A control system for a composite high-speed railway platform intelligent maintenance robot, characterized in that: include, Visual inspection module, used to detect defects and identify faults in high-speed railway platform equipment through deep learning algorithms; The navigation and positioning module uses the simultaneous positioning and mapping technology based on the fusion of lidar and visual sensor data to build environmental maps, and combines the path planning algorithm to achieve autonomous navigation and precise positioning of the robot; The task control module is used to dynamically allocate and schedule tasks based on the robot status and task priority based on the deep reinforcement learning algorithm, and optimize the task path by combining the task execution sequence, equipment status and environmental information; The actuator module includes a six-degree-of-freedom robot arm, a fastening tool, a high-pressure water gun, and a spray device, which are used to perform cleaning, repair, and maintenance operations of the equipment respectively; The remote monitoring module is used to collect robot status data in real time and exchange data with the control center.

2. The control system of a composite high-speed railway platform intelligent maintenance robot according to claim 1 is characterized in that: The visual inspection module combines image segmentation algorithms to classify and annotate equipment defects such as cracks, looseness, and corrosion, identify the size, shape, and location of defects, and train a deep learning model based on historical data to improve the generalization ability of defect identification.

3. The control system of a composite high-speed railway platform intelligent maintenance robot according to claim 1 is characterized in that: The navigation and positioning module adopts a combination of laser SLAM and visual SLAM to perform environmental modeling, and enhances positioning accuracy through multi-sensor data fusion technology. It adopts A, Dijkstra or DWA algorithm to realize robot path planning, ensuring the stable navigation of the robot in a dynamic environment.

4. The control system of a composite high-speed railway platform intelligent maintenance robot according to claim 1 is characterized in that: The task control module combines the robot energy consumption model, optimizes path planning through reinforcement learning algorithm, adjusts task priority according to real-time task status, and uses deep reinforcement learning method for dynamic scheduling to ensure the rationality and efficiency of task execution.

5. The control system of a composite high-speed railway platform intelligent maintenance robot according to claim 1, characterized in that: The six-degree-of-freedom robotic arm of the actuator module is used for precise maintenance operations. The motion trajectory of the robotic arm is controlled based on inverse kinematics calculation and path planning algorithm, and combined with adaptive impedance control strategy, the robotic arm can automatically adjust force control parameters when performing fastening, cleaning and spraying tasks, thereby improving the stability and safety of operations in complex environments.

6. The control system of a composite high-speed railway platform intelligent maintenance robot according to claim 1, characterized in that: The tightening tool is used to tighten bolts and nuts of high-speed railway platform equipment, supports automatic identification and adaptation of bolts of multiple specifications, and combines with a torque sensor to detect the tightening status in real time to ensure that the tightening torque meets the equipment maintenance standards.

7. The control system of a composite high-speed railway platform intelligent maintenance robot according to claim 1, characterized in that: The high-pressure water gun is used to clean stains and debris on the surface of the equipment. It has an adjustable water spray pressure, automatically adjusts the cleaning mode according to different degrees of contamination, and combines with a visual detection module to identify the scope of stains and optimize the cleaning path.

8. The control system of a composite high-speed railway platform intelligent maintenance robot according to claim 1, characterized in that: The spraying device is used to spray a protective coating on the surface of the equipment. The spraying parameters can be adjusted according to the material characteristics, and the spraying uniformity can be controlled in combination with visual feedback to ensure that the coating thickness meets the predetermined standards.

9. A control method for an intelligent maintenance robot, characterized in that: A control method for an intelligent maintenance robot as described in any one of claims 1 to 8 comprises the following steps: S1. Use deep learning algorithms to detect defects in high-speed railway platform equipment through the visual inspection module, and classify and mark the defect information; S2. The navigation and positioning module combines SLAM technology and path planning algorithms to complete the station environment mapping and dynamic path planning, ensuring the robot's autonomous navigation and precise positioning in complex environments; S3. The task control module uses reinforcement learning algorithm to optimize task scheduling based on task priority, robot status and real-time environmental data, and dynamically adjusts the execution order and path of maintenance tasks; S4. Perform cleaning, repair and maintenance operations through the actuator module, control the six-degree-of-freedom robot arm to complete precise operations based on inverse kinematics calculation and path planning algorithm, and adjust the force control parameters according to the adaptive impedance control strategy; S5. Realize real-time data interaction between the robot and the control center through the remote monitoring module to perform task monitoring, fault warning and remote control.

10. The control method of the intelligent maintenance robot according to claim 9, characterized in that: The step S3 adopts a task allocation algorithm based on deep reinforcement learning, combines task priority with real-time environmental feedback, dynamically updates the task execution order, and adopts the DDPG method to achieve adaptive adjustment of tasks to optimize task scheduling efficiency.

Images