Inspection task dynamic scheduling system based on endurance prediction and load balancing method

Abstract

The invention provides an inspection task dynamic scheduling system based on endurance prediction and a load balancing method, and belongs to the technical field of train inspection. The system comprises an endurance prediction module, a task decision-making module, a load balancing scheduling module, a credibility calculation module, a position correction module and the like. Residual endurance time and stop time are generated through an endurance prediction module, and a task decision module generates a task transfer data packet at the stop time; the load balancing scheduling module screens candidate robots based on the total demand energy consumption and the time adaptability, and determines a replacing robot by combining the comprehensive load value of task interval correction; the credibility calculation module generates a position credibility score through image matching and track calibration, and the position correction module dynamically adjusts the position to guarantee the precision. According to the method, the adaptation precision and the position evaluation accuracy of the candidate robots are improved, the global resource configuration is optimized, the task interruption rate is reduced, and the inspection task execution efficiency and stability are enhanced.

Description

Technical Field

[0001] This invention relates to the field of train inspection technology, specifically to a dynamic scheduling system for inspection tasks based on endurance prediction and a load balancing method. Background Technology

[0002] In the field of subway operation and maintenance, the use of mobile robots for automated train inspections has become an important means to ensure operational safety and improve maintenance efficiency. Trains need to complete rapid inspections during high-intensity, high-frequency operational breaks, thus placing extremely high demands on the continuity and reliability of inspection tasks.

[0003] However, existing technologies are insufficient to meet the dynamic scheduling requirements of subway scenarios. Most systems can only interrupt tasks and return when the robot's battery is low, leading to missed detection points and requiring subsequent manual re-inspection or a full-line re-inspection, resulting in low efficiency. Furthermore, the task handover process lacks a verification mechanism to ensure the accurate positioning of the replacement robot. In environments with variable lighting, similar structures like train undercarriages, or tunnels, positioning errors can easily occur, leading to incorrect task execution locations. In addition, existing scheduling strategies are mostly based on simple rules, failing to comprehensively consider the robot's real-time load, the spatiotemporal relationships between task sequences, and environmental interference factors, making it difficult to achieve efficient and reliable load balancing. Summary of the Invention

[0004] This invention provides a dynamic scheduling system and load balancing method for inspection tasks based on battery life prediction, which solves the problems in the prior art such as the forced interruption of inspection tasks due to power issues and the deviation of task execution location.

[0005] To achieve the above objectives, one embodiment of the present invention provides a dynamic scheduling system for inspection tasks based on endurance prediction, comprising: an endurance prediction module, used to generate the remaining endurance time of the inspection robot based on its remaining battery power and the distance between the robot's current location and the charging station, and to calculate the robot's stopping time; a task decision module, used to generate a stop operation command when the stopping time is reached, the stop operation command being used to control the robot to stop operation, record the location coordinates of the current inspection point and take an environmental marker image to generate a task handover data packet, and control the robot to return to the charging station; and a load balancing scheduling module, used to filter candidate robots based on the task handover data packet and evaluate the candidate robots. The system calculates the robot's overall load status to determine the replacement robot; it also sends the location coordinates and environmental marker images to the replacement robot to control the replacement robot to proceed to the inspection point marked by the location coordinates; the trust calculation module includes a trust generation unit and a trust correction unit. The trust generation unit calculates the matching relationship between the environmental marker image and the real-time environmental image using a navigation matching correction method after the replacement robot arrives at the inspection point, and generates a location trust score to evaluate the accuracy of the replacement robot's location; the trust correction unit corrects the location trust score based on the meeting position of the original inspection robot and the replacement robot, and determines whether to activate the inspection task based on the corrected location trust score.

[0006] Optionally, the endurance prediction module connects to the inspection robot's control system to obtain the robot's remaining battery power. The calculation process for the stop time includes: when the remaining battery power is lower than a preset battery power value, the endurance prediction module combines the distance between the inspection robot's current position and the charging station, the robot's moving speed, and a preset energy consumption model per unit distance to calculate the minimum endurance time required to return to the charging station from the current position. Based on the minimum endurance time and the theoretical endurance time corresponding to the robot's current remaining battery power, the module generates the actual remaining endurance time that can be used to continue inspection. The stop time is the end point of the duration corresponding to the generated remaining endurance time.

[0007] Optionally, the process of selecting candidate robots includes: parsing the task handover data packet, extracting the location coordinates of the current inspection point, and obtaining the mileage range and estimated time of the remaining inspection tasks; calculating the basic energy consumption required to complete the remaining inspection tasks based on the mileage range of the remaining inspection tasks and the energy consumption per unit mile of the inspection robots; adding the movement energy consumption from the current position of the candidate robot to the inspection point based on the calculated basic energy consumption to obtain the total energy consumption required by the candidate robots; calculating the time adaptability of each inspection robot based on the estimated time and the timeliness of the tasks assigned to each inspection robot; and selecting robots that simultaneously meet the total energy consumption requirement and time adaptability from all inspection robots as candidate robots.

[0008] Optionally, the navigation matching correction method includes: performing Scale Invariant Feature Transform (SIFT) keypoint extraction on the environmental marker image and the real-time environmental image respectively to obtain a first keypoint set and a second keypoint set; matching similar feature points in the first keypoint set and the second keypoint set using the K-nearest neighbor algorithm to generate a set of matching point pairs, and counting the number of valid matching point pairs in the set; calculating a basic trust score based on the number of valid matching point pairs as the initial value of the location trust score. Optionally, the inspection robot is equipped with a light sensor to collect ambient light intensity. The load balancing scheduling module is also used to analyze the integrity of feature contours in real-time environmental images to obtain the occlusion rate. The navigation matching correction method further includes: correcting the initial value of the location trust score based on the ambient light intensity or occlusion rate to generate the location trust score.

[0009] Optionally, the trust correction unit acquires the real-time position coordinates of the original inspection robot and the replacement robot. When the straight-line distance between the two enters a preset meeting range, a trust correction process is triggered. The trust correction process includes: the original inspection robot transmitting supplementary information to the replacement robot via short-range communication. The supplementary information includes a trajectory segment of the original robot moving to the meeting point after the inspection task is interrupted, a track feature image captured during the return journey, and the corresponding image coordinates; calculating a spatial deviation value by comparing the corresponding coordinates in the trajectory segment with the real-time coordinates of the replacement robot after it reaches the meeting point; calculating a feature overlap rate by comparing the image captured by the replacement robot after it reaches the image coordinates with the track feature image; and determining a correction coefficient based on the calculated spatial deviation value and feature overlap rate using a preset correction rule to correct the position trust score.

[0010] Optionally, a position correction module is also included, which is used to trigger a position correction process when the position trust score is lower than a preset threshold. The position correction process includes: controlling the replacement robot to collect real-time environmental images and generating corrected new coordinates through the navigation matching correction method; controlling the replacement robot to move to the new coordinate point and re-collect environmental images; and transmitting the re-collected environmental images to the trust calculation module to regenerate the position trust score.

[0011] On the other hand, a load balancing method for inspection tasks is also provided, applied to the load balancing scheduling module of the aforementioned dynamic scheduling system for inspection tasks, comprising: Step S1: obtaining real-time load parameters of candidate robots, and standardizing each parameter in the real-time load parameters to obtain the corresponding load coefficient; Step S2: calculating the load value of each candidate robot using a weighted algorithm based on the load coefficient of each parameter in the real-time load parameters and the corresponding preset weight; Step S3: obtaining the parameters of the next inspection task of each candidate robot and comparing them with the parameters of the task to be replaced, and calculating the task interval; Step S4: correcting the load value according to the task interval to obtain the comprehensive load value of each candidate robot, and selecting the candidate robot with the smallest comprehensive load value as the replacement robot.

[0012] Optionally, step S3 includes: obtaining the starting point coordinates of the next inspection task of each candidate robot, and at the same time obtaining the ending point coordinates of the task to be replaced, calculating the track path distance between the two as the task interval, and using the ratio of the task interval to the preset maximum path distance as the task interval coefficient. Optionally, the calculation of the comprehensive load value includes: selecting the corresponding correction weight according to the task interval coefficient, and correcting the load value through a preset correction form to obtain the comprehensive load value.

[0013] The present invention provides a dynamic scheduling system and load balancing method for inspection tasks based on endurance prediction. The technical advantage lies in the collaborative operation of an endurance prediction module, a task decision module, a load balancing scheduling module, and a trust calculation module, constructing a closed-loop intelligent task scheduling system. This system can accurately predict the remaining working time of a robot based on its real-time battery level and proactively trigger task interruption and handover processes at the optimal time, ensuring the continuity of inspection tasks. Through a load balancing algorithm, it comprehensively evaluates the energy consumption and time adaptability of candidate robots, intelligently selecting the most suitable replacement to achieve efficient utilization of cluster resources. Furthermore, it introduces a dual verification mechanism based on image matching and near-field communication to generate and dynamically correct location trust scores, ensuring that the replacement robot accurately arrives at the task interruption point, thereby greatly improving the reliability, automation, and operational efficiency of the entire inspection system. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a framework diagram of the dynamic scheduling system for inspection tasks provided in this embodiment of the invention; Figure 2 This is a flowchart of the candidate robot screening process provided in an embodiment of the present invention; Figure 3 This is a flowchart of the process for determining the replacement robot provided in an embodiment of the present invention; Figure 4 This is a flowchart of the location trust score generation process provided in an embodiment of the present invention; Figure 5 This is a flowchart of the inspection task load balancing method provided in the embodiments of the present invention. Detailed Implementation

[0015] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the scope of the present invention.

[0016] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0017] With the advancement of intelligent industrial inspection, existing inspection task scheduling systems suffer from several problems, including insufficient endurance prediction accuracy leading to frequent robot downtime, inadequate load balancing strategies resulting in resource allocation imbalances due to insufficient integration with task chain continuity, and a lack of dynamic correction mechanisms for position trust assessment causing task handover errors. Therefore, the development of a dynamic scheduling scheme and system for inspection tasks that deeply integrates endurance prediction, global load optimization, and accurate position verification is particularly urgent.

[0018] To address this issue, this invention provides a dynamic scheduling system and load balancing method for inspection tasks based on endurance prediction. The system uses an endurance prediction module to couple the calculation of remaining battery power and path distance to accurately generate stopping times. A load balancing algorithm based on a hierarchical screening mechanism and task interval coefficient optimization achieves efficient allocation of global resources. A dynamic trust correction system is constructed by combining encounter location trajectory matching and feature verification. Based on a position correction module and parameter iteration rules, a closed-loop control is formed from task handover to precise execution, achieving a synergistic improvement in inspection task continuity, robot load balancing, and position control accuracy.

[0019] The following is combined Figures 1-5 This invention is described in detail.

[0020] like Figure 1 As shown, this embodiment of the invention provides a dynamic scheduling system for inspection tasks based on endurance prediction, including: an endurance prediction module, used to generate the remaining endurance time of the inspection robot based on its remaining battery power and the distance between the robot's current location and the charging station, and to calculate the robot's stopping time; a task decision module, used to generate a stop operation command when the stopping time is reached, the stop operation command being used to control the robot to stop operation, record the location coordinates of the current inspection point and take an environmental marker image to generate a task handover data packet, and control the robot to return to the charging station; and a load balancing scheduling module, used to filter candidate robots based on the task handover data packet and evaluate the overall performance of the candidate robots. The system is configured to determine the replacement robot based on the load status; it also sends the location coordinates and environmental marker images to the replacement robot to control the replacement robot to proceed to the inspection point marked by the location coordinates; the trust calculation module includes a trust generation unit and a trust correction unit. The trust generation unit is used to calculate the matching relationship between the environmental marker image and the real-time environmental image through a navigation matching correction method after the replacement robot arrives at the inspection point, and generate a location trust score to evaluate the accuracy of the replacement robot's location; the trust correction unit is used to correct the location trust score based on the meeting position of the original inspection robot and the replacement robot, and determine whether to activate the inspection task based on the corrected location trust score.

[0021] The charging station is a fixed facility that provides power replenishment for the inspection robot. It has a built-in charging interface and communication module, which can automatically dock with the robot to complete charging and report the robot's charging status to the system. The remaining power refers to the remaining power of the inspection robot's current battery, which is collected in real time by the power sensor on the robot. The stopping time refers to the time point when the robot must stop its current task and start the return trip, calculated by the endurance prediction module based on the robot's remaining power, the minimum energy consumption requirement for returning to the charging station, and the remaining endurance that can continue to work. This time ensures that the robot has enough power to return to the charging station, while maximizing the use of the remaining power to complete part of the inspection task. The inspection robot is equipped with a locator and a camera, which can collect the location coordinates of the current inspection point and take pictures of environmental markers. The task handover data package refers to the specific data set generated by the original inspection robot when it stops working. It includes the location coordinates of the current inspection point, environmental marker images, and key information on unfinished inspection tasks, which is used to transmit the task start point and operation requirements to the replacement robot. The location trust score is used to evaluate the accuracy of the replacement robot's arrival at the location. Specifically, during operation of the dynamic scheduling system for inspection tasks based on endurance prediction provided in this embodiment of the invention: the endurance prediction module obtains the remaining battery power of the original inspection robot in real time, and calculates the remaining endurance time and the stop time when the operation must be stopped, based on its current location and the distance between its current location and the charging station; when the stop time is reached, the task decision module immediately generates a stop operation command, controls the original robot to record the location coordinates of the current inspection point, takes an environmental marker image containing fixed environmental features, packages it into a task handover data packet, and instructs the original robot to return to the charging station; after receiving the data packet, the load balancing scheduling module first filters out candidate robots, then determines the replacement robot by evaluating its comprehensive load status, and then sends the inspection point coordinates and environmental marker image to the replacement robot, controlling it to go to the target inspection point; after the replacement robot arrives, the trust level generation unit compares the environmental marker image with the real-time environmental image to generate an initial position trust level score; at the same time, the trust level correction unit monitors the positions of the original robot and the replacement robot in real time and corrects the score; if the corrected trust level score reaches a preset threshold, the system activates the inspection task, and the replacement robot performs the remaining inspection work.

[0022] The dynamic scheduling system for inspection tasks based on endurance prediction provided in this invention accurately calculates the stopping time through the endurance prediction module, avoiding robot shutdowns due to insufficient power and ensuring inspection continuity; it generates a task handover data package containing coordinates and environmental images through the task decision module, ensuring complete task handover information; it optimizes the selection and allocation of candidate robots through the load balancing scheduling module, solving the problem of uneven resource load; and it improves position accuracy and eliminates inspection blind spots through the dual verification mechanism of the trust degree calculation module. Ultimately, it achieves efficient connection of the inspection process, rational utilization of resources, and improved operational accuracy, reducing the need for manual intervention.

[0023] Preferably, the endurance prediction module is connected to the inspection robot's control system to obtain the remaining battery power of the inspection robot. The calculation process of the stop time includes: when the remaining battery power is lower than a preset battery power value, the endurance prediction module combines the distance between the inspection robot's current position and the charging station, the robot's moving speed, and a preset energy consumption model per unit distance to calculate the minimum endurance time required to return to the charging station from the current position, and generates the actual remaining endurance time that can be used to continue inspection based on the minimum endurance time and the theoretical endurance time corresponding to the robot's current remaining battery power; the stop time is the end point of the duration corresponding to the generated remaining endurance time.

[0024] Among them, the unit distance energy consumption model is a mathematical model based on the hardware parameters and environmental characteristics of the inspection robot. It is used to quantify the power consumption per meter of movement. For example, the model output value is 50mAh / m in a flat environment, and is corrected to 60mAh / m in a 3° slope. The minimum endurance time refers to the shortest time required for the robot to return to the charging station from its current position. The theoretical endurance time refers to the maximum working time corresponding to the current remaining power, which is calculated by dividing the remaining power by the energy consumption per unit time.

[0025] For example, suppose an inspection robot has a total battery capacity of 10000mAh, a preset battery level of 20% (2000mAh), and a current remaining battery level of 2500mAh, which is higher than the preset value. The current distance from the charging station is 500 meters, the robot's speed is 0.5m / s, and the energy consumption per unit distance in the current flat terrain is 50mAh / m, resulting in an energy consumption per unit time of 1000mAh / h. First, calculate the minimum runtime: 500 ÷ 0.5 = 1000 seconds ≈ 0.28 hours; theoretical runtime: 2500 ÷ 1000m = 2.5 hours; actual remaining runtime: 2.5 - 0.28 ≈ 2.22 hours. If the current time is 10:00, the stopping time is 10:00 + 2.22 ≈ 12:13, meaning the robot can continue inspections until 12:13, after which it must stop and return.

[0026] The preferred embodiment of this invention uses a predetermined power level trigger mechanism to accurately calculate the minimum runtime by combining distance traveled, speed of movement, and energy consumption per unit distance model. Then, the actual remaining runtime that can be operated is determined by the difference between the theoretical runtime and the minimum runtime. This makes the calculation of the stopping time more consistent with the robot's actual energy consumption characteristics and environmental differences. It avoids premature return due to relying solely on the absolute value of the remaining power level and prevents mid-journey stoppage caused by insufficient power level assessment. This significantly improves the accuracy of runtime prediction and power utilization efficiency, ensuring the continuity of inspection tasks and the safety of the robot's return journey.

[0027] like Figure 2As shown, preferably, the process of selecting candidate robots includes: parsing the task handover data packet, extracting the location coordinates of the current inspection point, and obtaining the mileage range and estimated time of the remaining inspection tasks; calculating the basic energy consumption required to complete the remaining inspection tasks based on the mileage range of the remaining inspection tasks and the unit mileage energy consumption parameter of the inspection robots; adding the movement energy consumption from the current position of the candidate robot to the inspection point based on the calculated basic energy consumption to obtain the total energy consumption required by the candidate robots; calculating the time adaptability of each inspection robot based on the estimated time and the timeliness of the tasks assigned to each inspection robot; and selecting robots that simultaneously meet the total energy consumption requirement and time adaptability from all inspection robots as candidate robots. Among them, the remaining inspection task mileage range refers to the total actual path length from the current inspection point to the end point of the remaining unfinished inspection task, which is obtained by parsing the task handover data packet; the estimated time refers to the estimated time required to complete the remaining inspection task; the basic energy consumption refers to the power consumption required to complete the remaining inspection task itself, which is calculated by multiplying the remaining inspection task mileage range by the energy consumption parameter per unit mileage; the movement energy consumption refers to the power consumption required for the candidate robot to move from its current position to the current inspection point, and the calculation formula is: the path distance from the candidate robot's current position to the inspection point × the energy consumption parameter per unit mileage, reflecting the energy cost of the robot reaching the task starting point; the time adaptability is a quantitative indicator used to evaluate whether the candidate robot can complete the remaining tasks within the specified time, and the calculation logic is: the estimated end time of the candidate robot's current task + the estimated time of the remaining inspection task ≤ the deadline of the remaining task, and at the same time, the candidate robot must not have any higher priority task conflicts.

[0028] Specifically, the process of selecting candidate robots is as follows: The load balancing scheduling module first parses the task handover data packet, extracts the coordinates of the current inspection point, the mileage range of the remaining inspection task (e.g., 1000 meters), and the estimated time (e.g., 60 minutes); based on the mileage range and the energy consumption per unit mileage (e.g., 50mAh / m), the basic energy consumption to complete the remaining task is calculated as 1000×50=50000mAh; then, the energy consumption of the candidate robot moving from its current position to the inspection point is added. For example, if a robot is 500 meters from the inspection point, its energy consumption is 500×50=25000mAh, resulting in a total energy requirement of 50000+25000=75000mAh; simultaneously, the estimated time and the timeliness of the tasks already assigned to the candidate robot are considered. For example, if the robot's current task ends at 14:00 and the deadline for the remaining tasks is 15:00, the time adaptability is calculated: 14:00+60 minutes=15:00≤Deadline; finally, robots with remaining battery power ≥ total energy requirement and time adaptability are selected from all robots as candidate robots.

[0029] The preferred embodiment of this invention, by clearly defining the dual screening criteria of total energy consumption and time adaptability, ensures that candidate robots can not only meet the power requirements of the remaining tasks and avoid task interruption due to insufficient power, but also complete the work within the deadline. Compared with the existing technology that only selects based on distance or the absolute value of remaining power, this significantly improves the adaptability accuracy of candidate robots, reduces the task allocation error rate, provides a reliable candidate pool for subsequent load balancing scheduling, optimizes global resource allocation, and improves the execution efficiency and stability of inspection tasks.

[0030] like Figure 4 As shown, preferably, the navigation matching correction method includes: performing Scale Invariant Feature Transform (SIFT) keypoint extraction on the environmental marker image and the real-time environmental image respectively to obtain a first keypoint set and a second keypoint set; matching similar feature points in the first keypoint set and the second keypoint set using the K-nearest neighbor algorithm to generate a set of matching point pairs, and counting the number of valid matching point pairs in the set of matching point pairs; calculating a basic trust score based on the number of valid matching point pairs as the initial value of the location trust score.

[0031] Further preferably, the inspection robot is equipped with a light sensor to collect ambient light intensity, and the load balancing scheduling module is also used to analyze the integrity of feature contours in real-time environmental images to obtain the occlusion rate. The navigation matching correction method further includes: correcting the initial value of the location trust score based on the ambient light intensity or occlusion rate to generate the location trust score.

[0032] The first keypoint set originates from the geometric feature set of the environmental marker image; the second keypoint set originates from the dynamic feature set of the real-time environmental image; the number of effective matching point pairs refers to the number of cross-image consistent feature point pairs selected by the k-NN algorithm, which is used to eliminate positioning drift caused by mismatches.

[0033] Specifically, the location trust score generation process is as follows: The trust score generation unit first uses a navigation matching correction method to extract key points from the environmental marker images in the task handover data package and the environmental images collected in real time by the replacement robot using scale-invariant feature transformation. This process generates a first key point set and a second key point set by detecting stable features such as corners and edges in the images. Then, the K-nearest neighbor algorithm is used to perform similarity matching on the feature point sets. This algorithm selects point pairs with highly consistent spatial location and texture features by calculating the Euclidean distance of feature descriptors, forming a set of matching point pairs. Next, the system dynamically corrects the basic score by integrating environmental parameters: real-time illumination intensity is obtained through the light sensor on the inspection robot, and the occlusion rate is calculated by analyzing the contour integrity of features in the real-time environmental image. Based on the coupled influence model of illumination intensity and occlusion rate, the final location trust score is generated. This score will be used as a core parameter input to the load balancing scheduling module to drive the task activation decision of the replacement robot.

[0034] A preferred embodiment of this invention constructs a location trust score generation system, which significantly improves the accuracy of positioning reliability assessment in complex scenarios through multimodal feature fusion and environmental adaptive correction mechanisms. The system first performs scale-invariant feature transformation on environmental marker images and real-time environmental images to extract key points, generating two sets of feature points. An efficient matching algorithm is then used to construct a set of matching point pairs, converting the matching results into a basic trust score to achieve preliminary quantification of positioning reliability. The key point matching process effectively overcomes interference from metallic reflections and weak textures, maintaining a high matching success rate even in scenarios where traditional methods fail, such as tunnels and low light. A dynamic environmental compensation mechanism is also introduced. Real-time ambient light intensity data is collected by sensors. Under conditions of strong backlight or extreme low light, optical correction coefficients are triggered, and the integrity of feature contours is analyzed to calculate occlusion parameters for secondary calibration of the basic score. This dual correction mechanism effectively suppresses positioning drift caused by weather changes and object occlusion, significantly reducing positioning errors under adverse conditions such as heavy rain and dense fog.

[0035] Preferably, the trust correction unit acquires the real-time position coordinates of the original inspection robot and the replacement robot. When the straight-line distance between the two enters a preset meeting range, a trust correction process is triggered. The trust correction process includes: the original inspection robot transmitting supplementary information to the replacement robot via short-range communication. The supplementary information includes a trajectory segment of the original robot moving to the meeting point after the inspection task is interrupted, a track feature image captured during the return journey, and the corresponding image coordinates; calculating a spatial deviation value by comparing the corresponding coordinates in the trajectory segment with the real-time coordinates of the replacement robot after it reaches the meeting point; calculating a feature overlap rate by comparing the image captured by the replacement robot after it reaches the image coordinates with the track feature image; and determining a correction coefficient based on the calculated spatial deviation value and feature overlap rate using a preset correction rule to correct the position trust score.

[0036] Among them, the preset encounter range refers to the distance threshold, such as 10 meters, that is pre-set by the system to trigger the trust correction process; the trajectory segment refers to the coordinate sequence of the original robot's movement path from the current inspection point to the encounter point; the track feature image refers to the image taken by the original robot during its return trip, which includes track fixed markers such as mileage markers and support numbers; the image coordinates refer to the coordinates of the shooting position corresponding to the track feature image; the spatial deviation value refers to the difference between the real-time coordinates of the replacement robot arriving at the encounter point and the corresponding position coordinates in the trajectory segment, with a smaller value indicating a higher degree of position overlap; the feature overlap rate refers to the matching ratio between the real-time image taken by the replacement robot at the image coordinates and the track feature image of the original robot, calculated by the SIFT method as the percentage of effective feature point overlaps out of the total number of feature points.

[0037] The correction rules are as follows: if the spatial deviation value is ≤5cm and the feature overlap rate is ≥80%, the correction coefficient is 1.2; if the spatial deviation value is 5-10cm and the feature overlap rate is 60%-80%, the correction coefficient is 1.0; if the spatial deviation value is >10cm or the feature overlap rate is <60%, the correction coefficient is 0.8.

[0038] Specifically, when the real-time straight-line distance between the original inspection robot and the replacement robot enters a preset meeting range, such as 10 meters, the trust correction process is initiated: the original robot transmits supplementary information to the replacement robot via short-range communication, including trajectory segments from the inspection point to the meeting point: such as the coordinate sequence X=100.00, Y=20.00, Z=3.00 to X=105.00, Y=20.00, Z=3.00, and the trajectory feature images and corresponding shooting coordinates captured during the return journey; the replacement robot arrives... After the meeting point, the system compares its real-time coordinates (X=105.02, Y=20.01, Z=3.00) with the corresponding coordinates in the trajectory segment, calculating a spatial deviation of 2.2cm. Simultaneously, a real-time image is captured at the shooting coordinates and compared with the trajectory feature image, yielding a feature overlap rate of 85%. According to the preset correction rule: when the spatial deviation is ≤5cm and the overlap rate is ≥80%, the correction coefficient is 1.2. If the initial position confidence score is 70 points, then the corrected score is 70×1.2=84 points.

[0039] The preferred embodiment of this invention proposes a location trust score correction process that compares the trajectory segments of the original inspection robot and the replacement robot within their encounter range, matches the trajectory feature images, and dynamically adjusts the location trust score by combining spatial deviation values ​​and feature overlap rates. This not only compensates for the shortcomings of single image matching, which is susceptible to the effects of lighting and occlusion, but also corrects the cumulative errors in navigation and positioning, thus significantly improving the accuracy of location assessment. Even in complex industrial environments, the reliability of the score can still be ensured through dual verification, effectively avoiding the omission of inspection points or repeated operations due to location deviations, providing a key guarantee for seamless task handover, and significantly improving the system's adaptability and stability in dynamic environments.

[0040] Preferably, the inspection task dynamic scheduling system further includes a position correction module, which is used to trigger a position correction process when the position trust score is lower than a preset threshold. The position correction process includes: controlling the replacement robot to collect real-time environmental images and generating corrected new coordinates through the navigation matching correction method; controlling the replacement robot to move to the new coordinate point and re-collect environmental images; and transmitting the re-collected environmental images to the trust calculation module to regenerate the position trust score.

[0041] The process of generating new coordinates through navigation matching correction is as follows: SIFT key point extraction is performed on the real-time environmental image currently collected by the replacement robot and the environmental marker image in the task handover data package. The positional deviation between the two is calculated by matching effective feature points, such as X-axis offset of 3cm and Y-axis offset of 2cm, to generate coordinate correction amount, and then further generate the corrected new coordinates.

[0042] Specifically, when the location trust score, after correction, is still below a preset threshold such as 80 points, the location correction module initiates the location correction process: First, it controls the replacement robot to re-acquire a real-time image containing fixed environmental features. Then, it compares this image with the environmental marker image using SIFT feature matching, calculates the position deviation, and generates a corrected new coordinate, such as X=100.10m in the original coordinate and X=100.05m in the corrected coordinate. Next, it drives the robot to move stepwise to the new coordinate, making fine adjustments of ≤5cm each time to avoid overshoot, and takes another environmental image at the new location. The new image is then transmitted to the trust calculation module, and the trust generation and correction process is repeated to re-score.

[0043] A preferred embodiment of the present invention provides a complete position correction process. This process involves a position correction module initiating a closed-loop adjustment when the score fails to meet the standard, using step-by-step fine-tuning to ensure that the robot moves accurately to the corrected coordinates, avoiding task delays caused by direct rescheduling and reducing resource waste. At the same time, through repeated feature matching and score correction, the system balances correction efficiency and task continuity while ensuring position accuracy, effectively solving the problem of single matching failure in complex environments, reducing the probability of rescheduling, and ultimately improving the system's adaptability to dynamic scenarios and the execution stability of inspection tasks.

[0044] like Figure 3 and Figure 5 As shown, this embodiment of the invention also provides a load balancing method for inspection tasks, applied to the load balancing scheduling module of the above-mentioned dynamic scheduling system for inspection tasks, including: Step S1: Obtain the real-time load parameters of the candidate robot, and standardize each parameter in the real-time load parameters to obtain the corresponding load coefficient; Step S2: Calculate the load value of each candidate robot using a weighted algorithm based on the load coefficient of each parameter in the real-time load parameters and the corresponding preset weight; Step S3: Obtain the parameters of the next inspection task for each candidate robot, compare them with the parameters of the task to be replaced, and calculate the task interval; Step S4: Adjust the load value according to the task interval to obtain the comprehensive load value of each candidate robot, and select the candidate robot with the smallest comprehensive load value as the replacement robot.

[0045] Specifically, real-time load parameters refer to the quantitative indicators collected by the load balancing scheduling module that reflect the current task pressure and status of candidate robots. These include: the number of unfinished tasks, the remaining time of the current task, the straight-line distance to the inspection point of the task to be replaced, and the average delay time of the last 3 tasks. Standardization is the process of converting real-time load parameters of different dimensions into values ​​between 0 and 1. The formula is load coefficient = (actual parameter value - minimum parameter value) / (maximum parameter value - minimum parameter value). For example, if a robot has 3 unfinished tasks, and the minimum value of this parameter among candidate robots is 1 and the maximum value is 5, then its load coefficient = (3-1) / (5-1) = 0.5. The preset weights are: unfinished task quantity weight 0.3, current task remaining time weight 0.2, distance weight 0.3, and average delay time weight 0.2. The weighted algorithm is the process of multiplying the above load parameters by their corresponding weights and then summing them. The task interval refers to the track path distance between the endpoint of the next task of the candidate robot and the endpoint of the task to be replaced.

[0046] The inspection task load balancing method provided in this invention accurately quantifies robot load through standardized processing and weighted algorithms, and dynamically corrects the load value by combining task intervals. This avoids the one-sidedness of scheduling based on a single parameter and significantly reduces the error of comprehensive load assessment. At the same time, it prioritizes the selection of the robot with the smallest comprehensive load value to achieve global resource optimization, reduce task waiting time, and improve the overall execution efficiency of inspection tasks and the system load balancing capability.

[0047] Preferably, step S3 includes: obtaining the starting point coordinates of the next inspection task for each candidate robot, and simultaneously obtaining the ending point coordinates of the task to be replaced, calculating the track path distance between the two as the task interval, and using the ratio of the task interval to the preset maximum path distance as the task interval coefficient.

[0048] More preferably, the calculation of the comprehensive load value includes: selecting the corresponding correction weight according to the task interval coefficient, and correcting the load value through a preset correction form to obtain the comprehensive load value.

[0049] Among them, the starting point coordinates of the next inspection task refer to the starting position coordinates of the first task currently assigned to the candidate robot but not yet executed; the ending point coordinates of the task to be replaced refer to the final completion point coordinates of the remaining inspection tasks to be handed over; the track path distance is different from the straight line distance, and refers to the actual path length along the preset inspection track from the starting point of the candidate robot's next task to the ending point of the task to be replaced, including the turning distance of track curves and branch nodes; the task interval coefficient refers to the index that quantifies the impact of task interval on load, and the calculation formula is: task interval coefficient = track path distance ÷ preset maximum path distance; the correction formula is the comprehensive load value = correction weight × load value.

[0050] The specific relationship between the task interval coefficient and the correction weight is as follows: When the task interval coefficient ∈ [0, 0.3], the correction weight = 0.8. In this range, the tasks are closely connected, the robot's movement loss is small, and the correction weight needs to be reduced to highlight the load advantage; when the task interval coefficient ∈ (0.3, 0.6], the correction weight = 1.0. In this range, the connection distance is medium, the load impact is moderate, and the baseline weight is maintained; when the task interval coefficient ∈ (0.6, 1.0], the correction weight = 1.2. In this range, the tasks are loosely connected, the movement energy consumption and time cost increase, and the correction weight needs to be increased to reflect the additional load.

[0051] Specifically, the calculation process for the comprehensive load value is as follows: First, the starting point coordinates of the next inspection task of the candidate robot are extracted from its task list, and the ending point coordinates of the task to be replaced are parsed from the task handover data packet. Next, based on the system's built-in electronic track map, the actual path distance along the track from the starting point of the candidate robot's next inspection task to the ending point of the task to be replaced is calculated using a path planning algorithm; this distance is used as the task interval. Then, this task interval is compared with a preset maximum path distance, such as 1000 meters, and the resulting ratio is the task interval coefficient. Subsequently, a corresponding correction weight is selected based on the task interval coefficient; for example, a coefficient of 0.2 corresponds to a correction weight of 0.8, and a coefficient of 0.8 corresponds to a correction weight of 1.1. The load values ​​of the candidate robots are then corrected using a preset correction formula, ultimately obtaining the comprehensive load value for each candidate robot. Finally, the candidate robot with the smallest comprehensive load value is selected as the replacement robot.

[0052] In a preferred embodiment of the present invention, task intervals are calculated by track path distance, which is more in line with actual inspection scenarios than straight-line distance. The standardized processing of task interval coefficients makes the impact of different distances quantifiable and comparable. Combined with coefficient selection to correct weights and adjust load values, the comprehensive load value can accurately reflect the smoothness of task connection and resource consumption, reduce load assessment deviations caused by unreasonable path planning, improve the scientific nature of replacement robot selection, and optimize the balance and efficiency of global task scheduling.

[0053] The above description is merely a preferred embodiment of the technical solution of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic scheduling system for inspection tasks based on endurance prediction, characterized in that, include: The endurance prediction module is used to generate the remaining endurance of the inspection robot based on its remaining battery power and the distance between the robot's current location and the charging station, and to calculate the robot's stopping time. The task decision module is used to generate a stop operation command when the stop time is reached. The stop operation command is used to control the robot to stop operation, record the position coordinates of the current inspection point and take an environmental marker image to generate a task handover data packet, and control the robot to return to the charging station. The load balancing scheduling module is used to filter candidate robots based on the task handover data packet and evaluate the comprehensive load status of the candidate robots to determine the replacement robot; it is also used to send the location coordinates and environmental marker image to the replacement robot to control the replacement robot to go to the inspection point marked by the location coordinates. The trust calculation module includes a trust generation unit and a trust correction unit. The trust generation unit is used to calculate the matching relationship between the environmental marker image and the real-time environmental image through a navigation matching correction method after the replacement robot arrives at the inspection point, and generate a location trust score to evaluate the accuracy of the replacement robot's location. The trust correction unit is used to correct the location trust score based on the meeting position of the original inspection robot and the replacement robot, and determine whether to activate the inspection task based on the corrected location trust score.

2. The dynamic scheduling system for inspection tasks according to claim 1, characterized in that, The endurance prediction module is connected to the inspection robot's control system to obtain the robot's remaining battery power. The calculation process for the stop time includes: when the remaining battery power is lower than a preset battery power value, the endurance prediction module combines the distance between the inspection robot's current position and the charging station, the robot's moving speed, and a preset energy consumption model per unit distance to calculate the minimum endurance time required to return to the charging station from the current position. Based on the minimum endurance time and the theoretical endurance time corresponding to the robot's current remaining battery power, the module generates the actual remaining endurance time that can be used to continue inspection. The stop time is the end point of the duration corresponding to the generated remaining endurance time.

3. The dynamic scheduling system for inspection tasks according to claim 1, characterized in that, The process of screening candidate robots includes: Parse the task handover data packet, extract the location coordinates of the current inspection point, and obtain the mileage range and estimated time of the remaining inspection tasks; Based on the remaining inspection task's mileage range and the inspection robot's energy consumption per unit mileage, the basic energy consumption required to complete the remaining inspection task is calculated. Based on the calculated basic energy consumption, the total energy consumption required by the candidate robot is obtained by adding the movement energy consumption from the current position of the candidate robot to the inspection point. Based on the estimated time consumption and the timeliness of the tasks assigned to each inspection robot, the time adaptability of each inspection robot is calculated. From all the inspection robots, the robots that simultaneously meet the requirements of total energy consumption and time adaptability are selected as candidate robots.

4. The dynamic scheduling system for inspection tasks according to claim 2, characterized in that, The navigation matching correction method includes: The environmental marker image and the real-time environmental image are respectively subjected to scale-invariant feature transform (SIFT) keypoint extraction to obtain a first keypoint set and a second keypoint set. The K-nearest neighbor algorithm is used to match similar feature points between the first key point set and the second key point set to generate a set of matching point pairs, and the number of valid matching point pairs in the set of matching point pairs is counted. The base trust score is calculated based on the number of valid matching point pairs, and is used as the initial value for the location trust score.

5. The dynamic scheduling system for inspection tasks according to claim 4, characterized in that, The inspection robot is equipped with a light sensor to collect ambient light intensity. The load balancing scheduling module is also used to analyze the integrity of feature contours in real-time environmental images to obtain the occlusion rate. The navigation matching correction method further includes: correcting the initial value of the location trust score based on the ambient light intensity or occlusion rate to generate the location trust score.

6. The dynamic scheduling system for inspection tasks according to claim 5, characterized in that, The trust correction unit acquires the real-time position coordinates of the original inspection robot and the replacement robot. When the straight-line distance between the two enters a preset encounter range, the trust correction process is triggered. The trust correction process includes: The original inspection robot transmits supplementary information to the replacement robot via short-range communication. The supplementary information includes the trajectory segment of the original robot moving to the meeting point after the inspection task was interrupted, the trajectory feature images taken during the return trip, and the corresponding image coordinates. Based on the real-time coordinates of the replacement robot after it arrives at the meeting point, the spatial deviation value is calculated by comparing it with the corresponding coordinates in the trajectory segment. Based on the image acquired after the replacement robot reaches the image coordinates, the feature overlap rate is calculated by comparing it with the track feature image. Based on the calculated spatial deviation value and feature overlap rate, a correction coefficient is determined through a preset correction rule to correct the location trust score.

7. The dynamic scheduling system for inspection tasks according to claim 1, characterized in that, It also includes a location correction module, used to trigger a location correction process when the location trust score is lower than a preset threshold, the location correction process including: The robot is controlled to acquire real-time environmental images, and the corrected new coordinates are generated using the navigation matching correction method. Control the replacement robot to move to the new coordinate point and re-acquire environmental images; The re-acquired environmental images are transmitted to the trust calculation module to regenerate the location trust score.

8. A method for load balancing of inspection tasks, applied to the load balancing scheduling module of the dynamic scheduling system for inspection tasks as described in any one of claims 1-7, characterized in that, include: Step S1: Obtain the real-time load parameters of the candidate robot, and standardize each parameter in the real-time load parameters to obtain the corresponding load coefficient; Step S2: Calculate the load value of each candidate robot using a weighted algorithm based on the load coefficient of each parameter in the real-time load parameters and the corresponding preset weight; Step S3: Obtain the parameters of the next inspection task for each candidate robot, compare them with the parameters of the task to be replaced, and calculate the task interval; Step S4: Adjust the load value according to the task interval to obtain the comprehensive load value of each candidate robot, and select the candidate robot with the smallest comprehensive load value as the replacement robot.

9. The inspection task load balancing method according to claim 8, characterized in that, Step S3 includes: obtaining the starting point coordinates of the next inspection task for each candidate robot, and simultaneously obtaining the ending point coordinates of the task to be replaced, calculating the track path distance between the two as the task interval, and using the ratio of the task interval to the preset maximum path distance as the task interval coefficient.

10. The inspection task load balancing method according to claim 9, characterized in that, The calculation of the comprehensive load value includes: selecting the corresponding correction weight according to the task interval coefficient, and correcting the load value through a preset correction form to obtain the comprehensive load value.

Images