Power station multi-robot collaborative inspection and task optimization scheduling system

Abstract

The invention relates to the technical field of power station operation and maintenance, and discloses a power station multi-robot collaborative inspection and task optimization scheduling system, which comprises a digital twinborn modeling module used for establishing a digital twinborn model of a power station environment, equipment and multi-type inspection robots, and realizing real-time mapping of a power station physical space and a virtual space through state synchronization; a task modeling and decomposition module; a task and capability matching scheduling module; a dynamic rescheduling module; a cloud edge cooperative control module; a data acquisition and feedback module; and a man-machine interaction and visualization module. Through integration of task node graph construction, robot capability vector modeling, local hot start rescheduling and a digital twin closed-loop self-learning mechanism, distribution continuity can be maintained when tasks change or robot states are updated, and parameters are continuously corrected based on operation data. And cooperative improvement of inspection task execution efficiency, scheduling response speed and system stability is realized.

Description

Technical Field

[0001] This invention relates to the field of power plant operation and maintenance technology, specifically to a power plant multi-robot collaborative inspection and task optimization scheduling system. Background Technology

[0002] Multi-robot collaborative inspection systems for power plants are intelligent systems designed to replace manual labor in inspecting, testing, and collecting data on power plant equipment. These systems typically consist of multiple mobile robots equipped with vision, infrared, temperature, or acoustic sensors. They can perform tasks such as equipment appearance inspection, temperature rise detection, and data transmission in complex power scenarios. With the development of digital and intelligent operation and maintenance, some systems have introduced task scheduling and path planning algorithms to enable multi-robot collaboration and inspection task allocation within the power plant environment, thereby improving inspection coverage and efficiency. Existing multi-robot inspection systems mainly rely on static task allocation or single-time global optimization algorithms for task scheduling and collaborative control.

[0003] However, current technologies, such as static task allocation or single-cycle global optimization algorithms, lack the ability to adapt to changes in tasks and dynamic updates in robot state. When new tasks are inserted, robot energy is insufficient, or equipment alarms occur, the global schedule needs to be recalculated, which can easily cause task interruption and response delay, resulting in the inability to dynamically correct parameters and limiting system stability and inspection efficiency. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a multi-robot collaborative inspection and task optimization scheduling system for power plants, which solves the problems of difficulty in adaptively updating scheduling results, easy interruption of task execution, and limited system stability and inspection efficiency.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a power plant multi-robot collaborative inspection and task optimization scheduling system, comprising: The digital twin modeling module is used to establish digital twin models of the power plant environment, equipment, and various types of inspection robots, and to realize real-time mapping between the physical space and virtual space of the power plant through state synchronization. The task modeling and decomposition module generates task node graph data based on the power plant virtual scene data, and structures the spatial location, type and priority of the tasks. The task and capability matching and scheduling module establishes the matching relationship between tasks and robots based on task node graph data and robot capability data, and performs multi-objective optimization calculations in the digital twin model environment to generate task allocation data and path planning data. The dynamic rescheduling module is used to perform local optimization of task allocation results based on real-time task change information during task execution. The cloud-edge collaborative control module performs global scheduling and policy updates in the cloud, as well as local control and path adjustment at the edge nodes, and performs bidirectional synchronization of task data and running status through the communication network. The data acquisition and feedback module collects robot operating status and multi-source perception data, and feeds the analysis results back to the task and capability matching and scheduling module for parameter updates. The human-computer interaction and visualization module is used to display the power plant environment, task nodes and robot operating status in the digital twin scenario, and to receive human input for adjusting task or scheduling parameters. The system forms a closed-loop data system that links the virtual and real worlds in a digital twin environment, and adaptively adjusts task allocation parameters based on the comparison results between actual operating data and simulation data.

[0006] Through the above technical solutions, a virtual power plant scene is constructed using a digital twin model. A multi-robot collaborative model is established by associating the task node graph with the robot capability vector. During task execution, a local hot-start rescheduling mechanism is adopted to achieve real-time response to task changes and state fluctuations. The system forms a collaborative decision-making structure between the cloud and edge layers. A virtual-real linkage closed loop is constructed through the data acquisition and feedback module. The scheduling parameters are adaptively adjusted based on the differences between simulation and operation data, thereby achieving continuous optimization of task planning, execution, and feedback.

[0007] Preferably, the digital twin modeling includes: Spatial modeling of the power station's terrain, equipment layout, and channel structure is performed to create a three-dimensional environmental model of the power station. Collect the operating parameters and identification information of the main equipment of the power station, and create corresponding equipment objects in the virtual scene; Establish geometric models, motion constraints, and sensor models for various types of inspection robots, and perform real-time state mapping between physical robots and virtual models.

[0008] Preferably, the task modeling and decomposition includes: Extract inspection targets, cycles, and work areas from the inspection plan; Task nodes are generated based on device location and task type, and task attributes and priority weights are set for each task node; A task node graph is constructed based on the spatial distance between nodes and the task dependencies, providing input data for the scheduling module.

[0009] Preferably, the task and capability matching scheduling includes: A robot capability vector is established based on the motion performance, sensor configuration, energy status, and health status uploaded by each robot. The task node graph is associated and matched with the robot capability vector to determine the set of executable tasks; In the digital twin model environment, a multi-objective optimization algorithm is adopted to comprehensively consider path length, task priority, energy consumption and load balancing, and output task allocation data and path planning data.

[0010] Preferably, the dynamic rescheduling includes: Monitor task execution status, equipment alarms, and robot status changes; Rescheduling is triggered when a new task is inserted, a robot fails, or a path is blocked. While keeping the unaffected task allocation unchanged, perform local optimization calculations on the affected task regions and output updated task allocation data.

[0011] Preferably, the cloud-edge collaborative control includes: Perform periodic global scheduling and parameter learning in the cloud to generate a global task scheduling scheme; Perform real-time task control, path adjustment, and obstacle avoidance at edge nodes; The task data and the robot's operating status are synchronized bidirectionally through a communication network.

[0012] Preferably, the task and capability matching scheduling includes setting energy and health constraints, including: The task execution time is limited based on the robot's remaining battery power and a preset battery threshold. The continuous operating load is limited based on the health scores of key robot components; During the scheduling process, constraint parameters are dynamically updated based on real-time energy and health status data.

[0013] Preferably, the data acquisition and feedback includes: Collect visual, infrared, temperature, and equipment operation data from each inspection robot; The collected data is synchronized in time and registered spatially to form a unified dataset; Based on the fusion analysis results, the task node weights and robot capability parameters are updated, and feedback information is generated and provided to the scheduling module.

[0014] Preferably, the human-computer interaction and visualization include: Display the power plant environment, task nodes, and robot operation trajectory in a digital twin scenario; Real-time display of task progress, energy levels, and equipment status; Receive user input commands to adjust task priorities or issue pause / re-execution control commands.

[0015] This invention provides a multi-robot collaborative inspection and task optimization scheduling system for power plants. It has the following beneficial effects: 1. This invention integrates task node graph construction, robot capability vector modeling, local hot start rescheduling, and digital twin closed-loop self-learning mechanism. It can maintain the continuity of allocation when the task changes or the robot state is updated, and continuously correct parameters based on the running data, thereby achieving a synergistic improvement in the efficiency of inspection task execution, scheduling response speed, and system stability.

[0016] 2. This invention utilizes a collaborative mechanism of cloud-based global scheduling and edge node real-time control to enable the system to balance global optimization and real-time on-site response capabilities in complex power plant environments. The cloud is responsible for task balancing and parameter learning, while edge nodes perform local path adjustments and anomaly responses. This reduces communication latency and scheduling delays while ensuring inspection coverage, thereby improving the real-time performance and reliability of multi-robot collaborative operations.

[0017] 3. This invention uses a multi-source sensing data fusion and feedback update mechanism to map inspection results to task nodes and robot capability parameters in real time, enabling adaptive updates of task weights and scheduling models. It can automatically adjust task priorities and execution strategies based on changes in equipment status, reducing the frequency of manual intervention, enabling the inspection process to have dynamic learning and continuous evolution capabilities, and enhancing the stability and self-maintenance of the system in long-term operation. Attached Figure Description

[0018] Figure 1 This is an architecture diagram of a power plant multi-robot collaborative inspection and task optimization scheduling system according to the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see the appendix Figure 1 This invention provides a power plant multi-robot collaborative inspection and task optimization scheduling system, comprising: The digital twin modeling module is used to establish digital twin models of the power plant environment, equipment, and various types of inspection robots, and to realize real-time mapping between the physical space and virtual space of the power plant through state synchronization. Furthermore, digital twin modeling includes: Spatial modeling of the power station's terrain, equipment layout, and channel structure is performed to create a three-dimensional environmental model of the power station. Collect the operating parameters and identification information of the main equipment of the power station, and create corresponding equipment objects in the virtual scene; Establish geometric models, motion constraints, and sensor models for various types of inspection robots, and perform real-time state mapping between physical robots and virtual models.

[0021] Specifically, by spatially modeling the power station's terrain, equipment layout, and channel structure, a three-dimensional environmental model of the power station is generated. This model is used to reflect the actual structural relationships of the power station in virtual space. The operating parameters and identification information of the main equipment of the power station are collected by the monitoring system and corresponding equipment objects are established in the virtual scene, so that the virtual equipment can reflect its operating status.

[0022] Meanwhile, geometric models, motion constraints, and sensor models are established for various types of inspection robots. Based on the pose information uploaded by the robots, the position and posture of the corresponding virtual robots are updated in the virtual scene, thereby performing real-time state mapping between physical robots and virtual models. The system obtains a unified virtual-real correspondence data environment, providing basic support for subsequent task node generation, scheduling calculation, and operation visualization.

[0023] The task modeling and decomposition module generates task node graph data based on the power plant virtual scene data, and structures the spatial location, type and priority of the tasks. Furthermore, task modeling and decomposition include: Extract inspection targets, cycles, and work areas from the inspection plan; Task nodes are generated based on device location and task type, and task attributes and priority weights are set for each task node; A task node graph is constructed based on the spatial distance between nodes and the task dependencies, providing input data for the scheduling module.

[0024] Specifically, the inspection first extracts the inspection targets, cycle, and work area information from the inspection plan. The inspection plan can be generated by the power plant operation and maintenance system, including the names of the equipment to be inspected, the inspection cycle, and the corresponding area. Based on this information, the system determines the set of equipment to be inspected and the time window, allowing inspection requirements to be input into the system in a data-driven manner. For example, in a substation, the main transformer, switchgear, and surge arresters listed in the inspection plan are identified as inspection target nodes. Subsequently, the module generates task nodes based on device location and task type, and sets task attributes and priority weights for each task node. Task attributes include task type (such as infrared temperature measurement, image recognition, or status detection) and the category of the device to which it belongs. Priority is determined based on the importance of the device or alarm status. The generated task nodes thus contain both spatial location information and task execution weights, providing structured input for subsequent scheduling. For example, device nodes with abnormal temperature alarms will be given higher priority so that they are preferentially allocated to robots capable of performing that task in the scheduling calculation. After generating task nodes, a task node graph is constructed based on the spatial distance and task dependencies between nodes. The edges between nodes represent feasible robot paths or task sequence constraints. The task node graph thus reflects the spatial distribution and operational logic. For example, in the main transformer area, there are sequence constraints between the main transformer temperature detection node and the adjacent switch cabinet detection node. The system establishes corresponding directed edges in the task node graph for subsequent scheduling calculations.

[0025] The task and capability matching and scheduling module establishes the matching relationship between tasks and robots based on task node graph data and robot capability data, and performs multi-objective optimization calculations in the digital twin model environment to generate task allocation data and path planning data. Furthermore, task and capability matching scheduling includes: A robot capability vector is established based on the motion performance, sensor configuration, energy status, and health status uploaded by each robot. The task node graph is associated and matched with the robot capability vector to determine the set of executable tasks; In the digital twin model environment, a multi-objective optimization algorithm is adopted to comprehensively consider path length, task priority, energy consumption and load balancing, and output task allocation data and path planning data.

[0026] Specifically, the task and capability matching and scheduling module establishes a robot capability vector based on the motion performance, sensor configuration, energy status, and health status uploaded by each robot. Motion performance includes parameters such as maximum speed and turning radius; sensor configuration reflects the types of targets the robot can perceive and the measurement range; energy status represents the remaining battery power and endurance; and health status is represented by the lifespan or failure probability of key components. These data are used to form the capability vector. ,in Indicates athletic performance. Indicates sensor adaptability, Indicates the energy coefficient. Indicates the health coefficient; Subsequently, the module matches the task node graph with the robot capability vector to determine the set of executable tasks. That is, the system judges the reachability of each robot within the allowable range of energy and health status based on the notification type, spatial location and execution constraints required by the task node. After matching, the set of executable tasks is formed, which provides candidate data for subsequent path calculation and scheduling. For example, if the task node is the main variable infrared detection, the task will only be included in the robot's task set if the robot has an infrared sensor, its energy status is higher than the threshold and it is located in the reachable area. Based on this, multi-objective optimization is performed in the digital twin environment, comprehensively considering four indicators: path length, task priority, energy consumption, and load balancing. The objective function can be expressed as: ; in, For path length, For energy consumption, Assign task priority to the architecture-level weighted value. For load balance deviation, to To correspond to the weighting coefficients, the solution process is constrained by the unique task allocation constraint and the upper limit of robot capability constraint. The optimization algorithm can adopt an improved ant colony algorithm or genetic algorithm to simulate and calculate multiple scheduling schemes in the digital twin virtual scene, and select the scheme with the minimum comprehensive cost as the task allocation and path planning result. For example, in a power plant scenario with three robots, the system automatically assigns the infrared detection task to the robot with infrared temperature measurement function and the image acquisition task to the robot equipped with a camera through this optimization process, thereby achieving a balance between energy, load and priority.

[0027] Furthermore, energy and health constraints are set in the task and capability matching and scheduling, including: The task execution time is limited based on the robot's remaining battery power and a preset battery threshold. The continuous operating load is limited based on the health scores of key robot components; During the scheduling process, constraint parameters are dynamically updated based on real-time energy and health status data.

[0028] Specifically, based on the robot's remaining battery power and a preset battery threshold, the system limits the task execution time. It calculates the robot's sustainable working time in real time based on the energy monitoring data reported by each robot. The remaining power is The average power consumption per unit time is Then its sustainable mission duration is Based on the execution time required for the task Determine if it satisfies If the conditions are not met, the task will not be included in the robot's executable set. For example, if robot A's remaining battery power can only sustain 20 minutes of operation, and a temperature measurement task is expected to take 25 minutes, the match will be automatically excluded to avoid downtime.

[0029] Secondly, the continuous workload is limited based on the health scores of the robot's key components. The health score is calculated from the operating status indicators of the transmission mechanism, sensors, and control system, and can be expressed as follows: The value 1 represents the optimal health status. The system sets a health threshold. ,when When the robot's continuous working time or number of tasks is limited, the health score can be calculated based on vibration sensors, temperature sensors or maintenance records. For example, when the health score of a tracked robot drops to 0.6 due to the increase in the temperature of the drive motor, the system limits its task allocation to a single inspection area to avoid continuously performing high-load tasks. During scheduling, constraint parameters are dynamically updated based on real-time energy and health status data. Energy status is sampled periodically by the power sensor module, and health status is provided by the device self-check and status assessment module. The scheduling algorithm reads the latest constraint values ​​in each iteration, verifies the feasibility of the current task allocation scheme, and simulates the task execution process in a digital twin environment to verify the constraint satisfaction. When the robot's energy consumption or health status drops beyond the threshold during execution, its task priority or work range is automatically adjusted in the next scheduling cycle. For example, in a long-term inspection task, when the robot's battery drops below 30%, the system marks it as "acceptable light-load task" to extend the overall work continuity.

[0030] The dynamic rescheduling module is used to perform local optimization of task allocation results based on real-time task change information during task execution. Furthermore, dynamic rescheduling includes: Monitor task execution status, equipment alarms, and robot status changes; Rescheduling is triggered when a new task is inserted, a robot fails, or a path is blocked. While keeping the unaffected task allocation unchanged, perform local optimization calculations on the affected task regions and output updated task allocation data.

[0031] Specifically, the module performs local optimization of task allocation results based on real-time task change information. When new tasks are inserted, robots fail, or paths are blocked during system operation, the module quickly redistributes the affected task areas without changing the overall scheduling structure, so as to maintain the continuity of task execution. First, the system monitors the task execution status, equipment alarms, and robot status changes. The monitoring data comes from the operation feedback of each robot, the task execution progress, and alarm signals from the power plant monitoring system. By continuously comparing the planned task with the actual task progress, when a task delay, a new equipment alarm, or an abnormal robot status is detected, a task change event is generated and uploaded to the scheduling layer. For example, during a power plant inspection, if the main transformer temperature sensor triggers a high temperature alarm, the system immediately generates a new "infrared retest" task node. When the dynamic rescheduling module detects a new task insertion, robot failure, or path obstruction, it triggers the rescheduling process. The rescheduling logic first determines the set of affected tasks and related robots, and then recalculates the task allocation and path planning within a local area. For new tasks, the system determines whether idle or low-load robots in the nearby area can undertake the task. For failed robots, their unfinished tasks will be assigned to standby robots with the same capabilities. For example, if a robot performing a task in the main transformer area enters standby mode due to insufficient power, the system selects a robot with the same sensor configuration from the surrounding robots to take over the task and recalculates its local path.

[0032] While maintaining the unaffected task allocation unchanged, the dynamic rescheduling module performs local optimization calculations on the affected task areas. The optimization objective is to restore scheduling feasibility in the shortest response time while maintaining system load balance. This can be achieved through a multi-objective optimization method based on a warm-start mechanism, using the previous task allocation result as the initial solution. Only the paths and allocation parameters are recalculated for the changed areas, thereby reducing global recalculation time. A new local task allocation scheme is calculated, and relevant robot path planning data is updated to ensure that inspection tasks continue to execute under changing conditions. For example, when a new infrared retesting task is added, the system quickly simulates and compares multiple allocation schemes in a digital twin environment and selects the scheme with the shortest response time for actual execution.

[0033] The cloud-edge collaborative control module performs global scheduling and policy updates in the cloud, as well as local control and path adjustment at the edge nodes, and performs bidirectional synchronization of task data and running status through the communication network. Furthermore, cloud-edge collaborative control includes: Perform periodic global scheduling and parameter learning in the cloud to generate a global task scheduling scheme; Perform real-time task control, path adjustment, and obstacle avoidance at edge nodes; The task data and the robot's operating status are synchronized bidirectionally through a communication network.

[0034] Specifically, periodic global scheduling and parameter learning are performed in the cloud. The cloud server periodically analyzes the task execution efficiency, energy consumption, and path distribution of each robot based on virtual operating data in the digital twin environment, updates the scheduling algorithm parameters, and generates a new global scheduling scheme. This scheme includes information such as task allocation priority, robot working periods, and area division, which guides the execution control of edge nodes. On the edge node side, the module is responsible for real-time task control and path adjustment. After receiving the cloud scheduling plan, the edge computing device dynamically fine-tunes the local task based on the actual data on site. Its main functions include obstacle avoidance path adjustment, task status tracking, and abnormal response. When the robot encounters path obstruction or local communication delay during execution, the edge node can immediately adjust its movement path or execution order to maintain the real-time response capability of the system. For example, when the inspection robot detects construction obstacles in the channel, the edge node recalculates the local path based on the surrounding environment information to avoid global rescheduling. Furthermore, the communication network enables bidirectional synchronization of task data and operational status between the cloud and edge nodes. The cloud periodically receives and summarizes the status information reported by each robot, while the edge nodes update execution instructions according to the global strategy issued by the cloud. Thus, the bidirectional communication structure ensures a balance between global optimization and local real-time performance, making the scheduling process both globally coordinated and adaptable to changes on-site.

[0035] The data acquisition and feedback module collects robot operating status and multi-source perception data, and feeds the analysis results back to the task and capability matching and scheduling module for parameter updates. Furthermore, data collection and feedback include: Collect visual, infrared, temperature, and equipment operation data from each inspection robot; The collected data is synchronized in time and registered spatially to form a unified dataset; Based on the fusion analysis results, the task node weights and robot capability parameters are updated, and feedback information is generated and provided to the scheduling module.

[0036] Specifically, the system first collects multi-source sensory data from each inspection robot, including visual images, infrared temperature, ambient temperature, and equipment operating status information. Visual data can be obtained by camera sensors and used to identify the appearance of the equipment, the status of indicator lights, and abnormal phenomena. Infrared temperature data is used to detect the surface temperature distribution of the equipment. Equipment operating status data can be read from the monitoring system through a wireless communication interface, including the current, voltage, and alarm signals of the equipment. For example, when a switch cabinet temperature sensor reports overheating information, the system simultaneously calls up the robot's infrared image to confirm the actual temperature rise area. Subsequently, the module performs time synchronization and spatial registration on the collected data to form a unified dataset. Time synchronization is based on a unified timestamp mechanism to align data from different robots and monitoring systems. Spatial registration is based on the robot's position and posture information to map image, infrared, and temperature data into the three-dimensional coordinates of the digital twin scene. Through time and space consistency processing, the system can accurately display the real-time operating status of the equipment in the virtual environment. For example, when two robots collect images and temperature data of the same transformer from different directions, they can be merged in the virtual scene to generate a complete monitoring view of the same equipment. After forming a unified dataset, the fusion results are analyzed to update task node weights and robot capability parameters. Based on the data fusion results, the system evaluates the completion status of task nodes and changes in equipment status. When the detection results of a task node show equipment abnormalities, the weight of that node is increased, allowing it to obtain higher priority in subsequent scheduling. When a robot's energy consumption or sensor performance declines, the corresponding capability parameters are automatically lowered, affecting its task allocation in the next round. For example, after multiple rounds of inspections, if the system detects a decrease in the infrared temperature measurement accuracy of a robot, it automatically lowers its "infrared task adaptability" parameter, so that subsequent scheduling algorithms will prioritize allocating such tasks to other robots.

[0037] The human-computer interaction and visualization module is used to display the power plant environment, task nodes and robot operating status in the digital twin scenario, and to receive human input for adjusting task or scheduling parameters. Furthermore, human-computer interaction and visualization include: Display the power plant environment, task nodes, and robot operation trajectory in a digital twin scenario; Real-time display of task progress, energy levels, and equipment status; Receive user input commands to adjust task priorities or issue pause / re-execution control commands.

[0038] Specifically, the digital twin scenario displays the main equipment of the power plant, the inspection area, and the robot's operating trajectory. Task nodes are marked with identifiers to facilitate operators' observation of the inspection range and execution progress. The system can display task progress, energy status, and equipment operating information in real time, and provide graphical prompts for anomalies. It also receives user input commands to adjust task priority, execution status, or scheduling parameters. Operators can issue commands such as pause, re-execute, or prioritize scheduling through the interface, and the system updates the execution strategy in the next scheduling cycle based on the input commands.

[0039] The system forms a closed data loop that links the virtual and real worlds in a digital twin environment, and adaptively adjusts task allocation parameters based on the comparison results between actual operating data and simulation data.

[0040] Specifically, during task execution, the robot's operating status, task progress, and environmental perception information are synchronously transmitted back to the digital twin environment. The virtual model updates the robot's position, task node status, and equipment operating parameters based on the received real-time data, ensuring that the virtual scene remains consistent with the physical power station. At the same time, the system performs simulation calculations corresponding to the actual task in the virtual environment, generating predicted task progress and energy consumption data. By comparing the differences between simulation data and actual operating data, the deviation of task execution is evaluated. Deviations may include path execution errors, task completion time differences, or energy consumption differences. When the deviation exceeds a set threshold, the task allocation parameters and robot capability parameters are adjusted according to the difference results to achieve adaptive updates of the task scheduling model. For example, when the simulation predicts that the task time is shorter than the actual execution time, the system automatically increases the task weight of the relevant area and can allocate more robot resources in the next scheduling cycle. Thus, the simulation model parameters can be optimized according to the actual operating conditions in each scheduling cycle, so that the digital twin environment continues to closely approximate the physical operating state, providing more accurate data support for task planning and scheduling.

[0041] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-robot collaborative inspection and task optimization scheduling system for power plants, characterized in that, include: The digital twin modeling module is used to establish digital twin models of the power plant environment, equipment, and various types of inspection robots, and to realize real-time mapping between the physical space and virtual space of the power plant through state synchronization. The task modeling and decomposition module generates task node graph data based on the power plant virtual scene data, and structures the spatial location, type and priority of the tasks. The task and capability matching and scheduling module establishes the matching relationship between tasks and robots based on task node graph data and robot capability data, and performs multi-objective optimization calculations in the digital twin model environment to generate task allocation data and path planning data. The dynamic rescheduling module is used to perform local optimization of task allocation results based on real-time task change information during task execution. The cloud-edge collaborative control module performs global scheduling and policy updates in the cloud, as well as local control and path adjustment at the edge nodes, and performs bidirectional synchronization of task data and running status through the communication network. The data acquisition and feedback module collects robot operating status and multi-source perception data, and feeds the analysis results back to the task and capability matching and scheduling module for parameter updates. The human-computer interaction and visualization module is used to display the power plant environment, task nodes and robot operating status in the digital twin scenario, and to receive human input for adjusting task or scheduling parameters. The system forms a closed-loop data system that links the virtual and real worlds in a digital twin environment, and adaptively adjusts task allocation parameters based on the comparison results between actual operating data and simulation data.

2. The power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 1, characterized in that, The digital twin modeling includes: Spatial modeling of the power station's terrain, equipment layout, and channel structure is performed to create a three-dimensional environmental model of the power station. Collect the operating parameters and identification information of the main equipment of the power station, and create corresponding equipment objects in the virtual scene; Establish geometric models, motion constraints, and sensor models for various types of inspection robots, and perform real-time state mapping between physical robots and virtual models.

3. The power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 1, characterized in that, The task modeling and decomposition includes: Extract inspection targets, cycles, and work areas from the inspection plan; Task nodes are generated based on device location and task type, and task attributes and priority weights are set for each task node; A task node graph is constructed based on the spatial distance between nodes and the task dependencies, providing input data for the scheduling module.

4. The power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 1, characterized in that, The task and capability matching and scheduling includes: A robot capability vector is established based on the motion performance, sensor configuration, energy status, and health status uploaded by each robot. The task node graph is associated and matched with the robot capability vector to determine the set of executable tasks; In the digital twin model environment, a multi-objective optimization algorithm is adopted to comprehensively consider path length, task priority, energy consumption and load balancing, and output task allocation data and path planning data.

5. A power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 1, characterized in that, The dynamic rescheduling includes: Monitor task execution status, equipment alarms, and robot status changes; Rescheduling is triggered when a new task is inserted, a robot fails, or a path is blocked. While keeping the unaffected task allocation unchanged, perform local optimization calculations on the affected task regions and output updated task allocation data.

6. A power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 1, characterized in that, The cloud-edge collaborative control includes: Perform periodic global scheduling and parameter learning in the cloud to generate a global task scheduling scheme; Perform real-time task control, path adjustment, and obstacle avoidance at edge nodes; The task data and the robot's operating status are synchronized bidirectionally through a communication network.

7. A power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 4, characterized in that, The task and capability matching and scheduling includes energy and health constraints, including: The task execution time is limited based on the robot's remaining battery power and a preset battery threshold. The continuous operating load is limited based on the health scores of key robot components; During the scheduling process, constraint parameters are dynamically updated based on real-time energy and health status data.

8. A power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 1, characterized in that, The data collection and feedback include: Collect visual, infrared, temperature, and equipment operation data from each inspection robot; The collected data is synchronized in time and registered spatially to form a unified dataset; Based on the fusion analysis results, the task node weights and robot capability parameters are updated, and feedback information is generated and provided to the scheduling module.

9. A power plant multi-robot collaborative inspection and task optimization scheduling system according to claim 1, characterized in that, The human-computer interaction and visualization include: Display the power plant environment, task nodes, and robot operation trajectory in a digital twin scenario; Real-time display of task progress, energy levels, and equipment status; Receive user input commands to adjust task priorities or issue pause / re-execution control commands.

Images