A kind of inspection robot cooperative inspection task dynamic scheduling control system

Abstract

This invention provides a dynamic scheduling and control system for collaborative inspection tasks using inspection robots, comprising: a data fusion and optimization module configured to collect multi-source data from photovoltaic power plants; an environmental perception and modeling module configured to construct a digital environment model library and an equipment physical characteristic model library, coupled to form a LIM model; a fault diagnosis and task generation module configured to generate inspection task requirement lists and maintenance task requirement lists; a multi-objective collaborative scheduling module configured to generate inspection task scheduling schemes; and a robot execution and control module configured to receive the scheduling schemes, control at least one intelligent maintenance robot to autonomously navigate to the target location to perform inspection or maintenance operations, and provide real-time feedback on the execution status. This invention enables comprehensive perception, accurate diagnosis, optimized scheduling, and autonomous execution for the operation and maintenance of photovoltaic power plants, improving operation and maintenance efficiency, reducing safety risks, and saving operating costs, resulting in significant economic and social benefits.

Description

Technical Field

[0001] This application relates to the field of intelligent operation and maintenance technology for photovoltaic power plants, and more specifically, to a dynamic scheduling and control system for collaborative inspection tasks of inspection robots. Background Technology

[0002] Large-scale centralized photovoltaic (PV) power plants are typically characterized by a huge number of power generation units, a wide distribution range, and significant influence from terrain and topography during installation. PV arrays are built along mountain slopes, with varying module orientations and tilt angles, resulting in poor equipment consistency and significant differences in the inherent properties of different power generation units. Furthermore, PV power plants are often located in remote areas such as deserts and barren mountains, facing harsh natural environments, making it difficult for maintenance personnel to access them, and posing high safety risks.

[0003] The main problems currently existing in the operation and maintenance of photovoltaic power plants are as follows: First, the equipment status awareness capability is insufficient. The number of sensors deployed in the power plant is limited, the types of data collected are limited, and the data formats between different systems are inconsistent, making it difficult to form a comprehensive understanding of the equipment status. Maintenance personnel rely on regular inspections and post-fault repairs, which cannot detect early signs of equipment degradation in a timely manner.

[0004] Second, fault diagnosis relies on human experience. When equipment malfunctions, maintenance personnel need to combine historical data and on-site inspections to make judgments, resulting in low efficiency and poor accuracy. In particular, hidden defects such as hot spots and microcracks in photovoltaic modules are difficult to detect with the naked eye and are often only noticed after a significant drop in power generation has occurred.

[0005] Third, maintenance and repair tasks rely on manual operation. Maintenance and repair work at photovoltaic power plants, such as inverter switch operation and combiner box inspection, still requires maintenance personnel to be on-site. In complex terrain conditions, personnel travel back and forth, which is time-consuming and labor-intensive, and poses safety risks such as electric shock and falls.

[0006] Fourth, there is a lack of scientific methods for scheduling inspection resources. Although intelligent devices such as drones and inspection robots have been gradually applied, each device operates independently, and task allocation and route planning mainly rely on human experience. Faced with a large number of inspection points and diverse task requirements, it is difficult to achieve optimal resource allocation and coordinated task execution.

[0007] Therefore, how to fully grasp the environmental and equipment characteristics of photovoltaic power plants, achieve accurate perception of equipment status and early diagnosis of faults, and realize the collaborative operation of inspection resources through intelligent scheduling is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] This invention aims to overcome the shortcomings of the prior art and provide a dynamic scheduling and control system for collaborative inspection tasks of inspection robots, so as to solve the technical problems of photovoltaic power plant operation and maintenance relying on human experience, poor equipment coordination, and low fault diagnosis accuracy in the prior art.

[0009] This invention provides a dynamic scheduling and control system for collaborative inspection tasks of inspection robots, comprising: The data fusion and optimization module is configured to collect multi-source data from photovoltaic power plants, establish unified data standards, and fuse and optimize equipment data from different sources and time scales to form a full lifecycle information dataset. The environmental perception and modeling module is configured to extract environmental and equipment features of the photovoltaic power station based on the full life cycle information dataset, construct a digital environment model library and an equipment physical feature model library, and couple them to form a LIM model; The fault diagnosis and task generation module is configured to use the LIM model as data support, adopt a multimodal fusion fault diagnosis algorithm to identify the type and location of equipment faults, and combine equipment deterioration trend prediction to generate an inspection task requirement list and a maintenance task requirement list. The multi-objective collaborative scheduling module is configured to integrate the inspection task requirement list, the maintenance task requirement list, environmental information, robot status and resource constraints, and generate a scheduling scheme for the inspection tasks using a composite objective optimization algorithm. The scheduling scheme includes task allocation, path planning and resource scheduling. The robot execution and control module is configured to receive the scheduling scheme, control at least one intelligent maintenance and inspection robot to autonomously navigate to the target location to perform inspection or maintenance operations, and provide real-time feedback on the execution status.

[0010] Preferably, the data fusion and optimization module includes: The data acquisition unit is configured to acquire basic equipment data, equipment operation data, equipment anomaly data, maintenance and troubleshooting data, as well as drone inspection data and robot inspection data. The data preprocessing unit is configured to clean, standardize, and convert the format of the collected data. The fusion optimization unit is configured to use a unified data standard to associate and fuse device data from different sources and at different time scales to form the full lifecycle information dataset.

[0011] Preferably, the environmental perception and modeling module includes: The digital environment model library construction unit is configured to combine GIS, satellite remote sensing imagery and UAV aerial survey data to extract the elevation changes, slope undulations and geographical obstacle distribution characteristics of photovoltaic power stations, construct a three-dimensional digital environment model library and photovoltaic area road network topology, and extract road material and slope characteristics. The equipment physical feature model library construction unit is configured to perform pixel-level classification of photovoltaic array images through a convolutional neural network, realize string area positioning and segmentation, and extract string azimuth, tilt angle, component elevation and installation gap feature parameters. The LIM model building unit is configured to couple the digital environment model library with the device physical feature model library to build a unified storage and management system for the full lifecycle information model of environment features, device features, and operational data.

[0012] Preferably, the fault diagnosis and task generation module includes: The multi-scale self-optimizing fault diagnosis unit is configured to establish a multi-scale self-optimizing fault diagnosis learning model based on the device's multi-channel time-series data stored in the LIM model, and to identify the root causes of device faults by training and optimizing the model through running data. The multimodal thermal anomaly detection unit is configured to integrate infrared thermal images and visible light images based on the device image data stored in the LIM model, use a small target detection algorithm to extract photovoltaic string regions and detect hot spot faults in the modules, and use a clustering algorithm to achieve automatic segmentation and location of faulty modules. The fault root cause localization unit is configured to integrate signal processing and image analysis techniques based on the historical operating data and image data stored in the LIM model, extract the edge contour features of the equipment in infrared and visible light images under fault conditions and compare them with the normal state to locate the faulty components and defect categories. The degradation trend prediction unit is configured to extract key equipment degradation electrical characteristics based on the equipment electrical operation data stored in the LIM model, construct a degradation trend prediction model, and predict early degradation status of equipment and generate early warning tasks through learning and training on the operation data.

[0013] Preferably, the multi-target cooperative scheduling module includes: The data integration unit is configured to integrate the environmental and equipment data output by the LIM model, the inspection task requirement list, the maintenance task requirement list, robot status, and resource scheduling constraint data, and perform cleaning and standardization. The decision rule base unit is configured to set operation and maintenance decision rules under different working conditions according to business scenarios and operation and maintenance strategies, and to transform composite objectives into decision variables and constraints. The optimization and solution unit is configured to construct a land-air integrated autonomous operation and maintenance decision optimization algorithm model with the goal of optimizing efficiency, safety, and cost, and solve for task allocation, path planning, and resource scheduling schemes; The dynamic rescheduling unit is configured to update the task queue and resource status when a new fault, robot fault, or environmental change is detected during task execution, trigger rescheduling optimization, and dynamically adjust the allocation scheme and execution path of unexecuted tasks.

[0014] Preferably, the robot execution and control module includes at least one intelligent maintenance and inspection robot, the intelligent maintenance and inspection robot comprising: The autonomous navigation chassis system is configured to build an environmental map based on multi-sensor fusion perception of environmental information, plan and track the optimal driving path, and achieve autonomous driving and dynamic obstacle avoidance; The dual-light gimbal vision system is configured to acquire two-dimensional or three-dimensional image information of the target device, identify the target and extract features through multi-view fusion, and achieve spatial pose calibration. A multi-degree-of-freedom robotic arm and an end effector are configured to plan the robotic arm's motion trajectory based on visual positioning results and drive the end effector to perform operations on the target device. The end effector includes a toggle switch mechanism and an insertion lock hole unlocking mechanism.

[0015] Preferably, the autonomous navigation chassis system is further configured to: perceive the environment by fusing LiDAR, camera and high-precision positioning system, search for drivable paths by graph search algorithm, generate optimized driving trajectory by combining chassis dynamic characteristics, and achieve trajectory tracking by control algorithm.

[0016] Preferably, the dual-light gimbal vision system is further configured to update the spatial pose information of the target device during the operation of the robotic arm, providing positioning for the real-time trajectory planning and autonomous operation of the robotic arm.

[0017] Preferably, the multi-objective collaborative scheduling module is further configured to: construct a power plant health evaluation model based on the equipment performance data stored in the LIM model and in combination with key operating indicators of the photovoltaic power plant, generate evaluation results of power plant performance indicators and macro-operating indicators, and provide a basis for scheduling decisions.

[0018] This invention provides a dynamic scheduling and control system for collaborative inspection tasks using inspection robots. Through a data fusion and optimization module, it achieves unified governance of multi-source heterogeneous data, laying the foundation for full lifecycle information management. An environmental perception and modeling module constructs a LIM model, forming a digital twin of the photovoltaic power station, coupling environmental features with equipment characteristics to provide standardized data support for precise analysis. Through a fault diagnosis and task generation module, it employs technologies such as multi-scale self-optimizing fault diagnosis, multi-modal thermal anomaly detection, fault root cause localization, and degradation trend prediction, significantly improving the accuracy of fault identification and early warning capabilities, transforming passive maintenance into proactive prevention. Through a multi-objective collaborative scheduling module, it comprehensively considers multiple objectives such as efficiency, safety, and cost, achieving optimal linkage and dynamic rescheduling of inspection personnel, drones, and intelligent maintenance robots, transforming the traditional operation and maintenance model relying on human experience into scientific operation and maintenance management guided by artificial intelligence. Through a robot execution and control module, it utilizes an autonomous navigation chassis, dual-light vision positioning, and a multi-degree-of-freedom robotic arm to replace manual labor in completing inspection and maintenance operations in complex environments, propelling photovoltaic power stations towards truly unmanned operation and maintenance. This invention enables comprehensive perception, accurate diagnosis, optimized scheduling, and autonomous execution of photovoltaic power plant operation and maintenance, effectively improving operation and maintenance efficiency, reducing safety risks, and saving operating costs, resulting in significant economic and social benefits.

[0019] The further effects of the aforementioned non-conventional preferred method will be explained below in conjunction with specific embodiments. Attached Figure Description

[0020] To more clearly illustrate the embodiments of the present invention or the existing technical solutions, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram illustrating the composition of a dynamic scheduling and control system for collaborative inspection tasks of inspection robots, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the composition of a data fusion and optimization module in a dynamic scheduling and control system for collaborative inspection tasks of inspection robots, provided in an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the composition of the environmental perception and modeling module in a dynamic scheduling and control system for collaborative inspection tasks of inspection robots, provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of the composition of a fault diagnosis and task generation module in a collaborative inspection task dynamic scheduling and control system for inspection robots, provided in an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the composition of a multi-target collaborative scheduling module in a dynamic scheduling and control system for collaborative inspection tasks of inspection robots, provided in an embodiment of the present invention. Figure 6 This is a schematic diagram of the composition of the robot execution and control module in a dynamic scheduling and control system for collaborative inspection tasks of inspection robots, provided in an embodiment of the present invention. Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0022] The attached diagram is labeled as follows: 100 - Data Fusion and Optimization Module, 110 - Data Acquisition Unit, 120 - Data Preprocessing Unit, 130 - Fusion and Optimization Unit, 200 - Environmental Perception and Modeling Module, 210 - Digital Environment Model Library Construction Unit, 220 - Equipment Physical Feature Model Library Construction Unit, 230 - LIM Model Construction Unit, 300 - Fault Diagnosis and Task Generation Module, 310 - Multi-scale Self-optimizing Fault Diagnosis Unit, 320 - Multi-modal Thermal Anomaly Detection Unit, 330 - Fault... Root cause localization unit, 340-deterioration trend prediction unit, 400-multi-target collaborative scheduling module, 410-data integration unit, 420-decision rule base unit, 430-optimization solution unit, 440-dynamic rescheduling unit, 500-robot execution and control module, 510-autonomous navigation chassis system, 520-dual-light gimbal vision system, 530-multi-degree-of-freedom robotic arm and end effector, 700-electronic equipment, 701-processor, 702-memory, 703-computer program. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0024] This embodiment provides a dynamic scheduling and control system for collaborative inspection tasks using inspection robots, applicable to intelligent operation and maintenance scenarios in large-scale centralized photovoltaic power plants. This system, through a data-driven approach, integrates multi-source heterogeneous data fusion, environmental and equipment modeling, fault diagnosis, task generation, collaborative scheduling, and robot execution into a closed-loop operation and maintenance management system. It aims to address issues in the background technology such as insufficient perception capabilities, reliance on manual diagnosis, reliance on manual maintenance, and a lack of scientific scheduling methods. The following detailed description, in conjunction with the accompanying drawings, describes a dynamic scheduling and control system for collaborative inspection tasks using inspection robots according to an embodiment of this application.

[0025] like Figure 1 As shown, the system includes: a data fusion and optimization module 100, an environmental perception and modeling module 200, a fault diagnosis and task generation module 300, a multi-objective collaborative scheduling module 400, and a robot execution and control module 500. These modules are connected sequentially to form a complete closed loop of data and business flows, from data fusion, environmental modeling, fault diagnosis, task generation, collaborative scheduling to autonomous execution. During system operation, the modules interact through standardized data interfaces and iteratively optimize their internal models using continuously accumulated operational data, achieving adaptive intelligent operation and maintenance.

[0026] The data fusion and optimization module 100 is configured to collect multi-source data from photovoltaic power plants, establish unified data standards, and fuse and optimize equipment data from different sources and time scales to form a full lifecycle information dataset. This module addresses the problem of insufficient equipment status awareness mentioned in the background technology, providing a high-quality data foundation for subsequent modeling.

[0027] Specifically, such as Figure 2 As shown, the data fusion and optimization module 100 includes: Data acquisition unit 110 is configured to continuously acquire basic equipment data, equipment operation data, equipment anomaly data, maintenance and troubleshooting data, and drone and robot inspection data. Basic equipment data includes static information such as equipment model, factory parameters, and installation location; equipment operation data includes high-frequency real-time monitoring data (on a second or minute scale) such as current, voltage, power, and temperature; equipment anomaly data records changes such as equipment replacement and relocation; maintenance and troubleshooting data records historical repair records and defect handling status; and drone and robot inspection data includes visible light images, infrared thermal images, and laser point clouds. These data come from diverse sources, have different formats, and different sampling frequencies, constituting a typical multi-source heterogeneous data set for power plants.

[0028] The data preprocessing unit 120 is configured to clean, standardize, and convert the format of the collected data. The cleaning operation uses statistical analysis and rule-based judgment to remove outliers (such as readings exceeding the measurement range) and noisy data (such as garbled characters caused by communication interruptions). The standardization operation converts electrical data of different dimensions (current, voltage, power, etc.) into uniform per-unit values ​​or normalized ranges. The format conversion converts proprietary protocol data output from different manufacturers' devices into a standard data structure defined internally by the system. The preprocessed data is stored in a unified format, providing a clean and consistent data source for subsequent fusion.

[0029] The fusion optimization unit 130 is configured to use a unified data standard to correlate and fuse equipment data from different sources and time scales, forming a full lifecycle information dataset. This unit uses timestamps and device IDs as indexes and employs a spatiotemporal alignment algorithm to precisely match operational data, environmental data, and image data. For example, it correlates component current data at the same time with the infrared thermal imaging temperature data of the corresponding string; it correlates images of a certain area captured by a drone with historical fault records for that area. The fused dataset not only contains the original data but also derives various composite features (such as the temperature-current ratio and power-irradiance deviation rate), providing rich sample features for subsequent model training. This dataset is continuously updated as the power plant operates, becoming the data foundation for all upper-layer applications of the system.

[0030] The Environmental Perception and Modeling Module 200 is configured to extract environmental and equipment features of photovoltaic power plants based on a full lifecycle information dataset, construct a digital environment model library and an equipment physical feature model library, and couple them to form a LIM model. This module transforms raw data into structured model knowledge, providing a digital twin foundation for accurate diagnosis.

[0031] Specifically, such as Figure 3 As shown, the environmental perception and modeling module 200 includes: The digital environment model library construction unit 210 is configured to combine GIS, satellite remote sensing imagery, and UAV aerial survey data to extract the elevation changes, slope undulations, and geographical obstacle distribution characteristics of the photovoltaic power station, construct a 3D digital environment model library and the road network topology of the photovoltaic area, and extract road material and slope characteristics. GIS data provides basic geographic information (such as latitude, longitude, and altitude), satellite remote sensing imagery provides macroscopic terrain features (such as mountain orientation and vegetation cover), and UAV aerial survey data (oblique photography, LiDAR) provides high-precision local terrain information (such as micro-topography between component arrays). Through multi-source data fusion and 3D reconstruction algorithms, a digital 3D terrain model of the power station is generated. Simultaneously, based on this model, the road network topology is automatically extracted, passable areas are identified, and key parameters such as road slope and material are calculated, providing environmental constraints for subsequent robot path planning.

[0032] The equipment physical feature model library construction unit 220 is configured to perform pixel-level classification of photovoltaic array images using a convolutional neural network, achieving string region localization and segmentation, and extracting string azimuth, tilt angle, component elevation, and installation gap feature parameters. This unit first trains a semantic segmentation CNN model using historically accumulated photovoltaic array image samples (including images under different terrain and illumination conditions) to accurately identify photovoltaic string regions in the image. The training samples are manually labeled, including categories such as string boundaries, obstructions, and background. The trained model performs pixel-by-pixel classification on the input image, outputting a precise mask for the strings, and then extracts the azimuth (relative to due south), tilt angle (relative to the horizontal plane), component center point elevation, and installation gap between strings for each string through geometric calculations. These parameters reflect the natural endowment differences between different strings; for example, the tilt angle of components on a hillside may vary with the terrain, resulting in different actual received irradiance. These features are stored in the equipment physical feature model library, providing a basis for subsequent output consistency analysis.

[0033] LIM model building unit 230 is configured to couple the digital environment model library and the equipment physical characteristic model library, constructing a unified storage and management model for the entire lifecycle information of environmental characteristics, equipment characteristics, and operational data. The LIM model adopts a graph database or spatial database structure, associating and storing environmental elements (terrain, road network), equipment elements (string location, attitude parameters), dynamic operational data (electrical quantities, temperature), and historical events (faults, maintenance). For example, each string node not only contains its physical characteristic parameters but also links to its location's terrain data, historical fault records, and real-time monitoring data streams. The LIM model supports multi-dimensional queries (such as "querying the current power of all strings in an area with a slope greater than 15 degrees") and spatiotemporal analysis (such as "comparing the power generation curves of strings facing different directions under the same inverter"). This model essentially constitutes a digital twin of the photovoltaic power station, providing standardized data service interfaces for upper-level applications. The model itself is continuously enriched with the addition of new data; for example, newly added fault samples will update fault association knowledge.

[0034] The fault diagnosis and task generation module 300 is configured to use the LIM model as data support, employ a multimodal fusion fault diagnosis algorithm to identify equipment fault types and locations, and combine this with equipment degradation trend prediction to generate inspection task requirement lists and maintenance task requirement lists. This module directly addresses the problem of fault diagnosis relying on human experience in the background technology, improving the accuracy and automation level of diagnosis through intelligent algorithms.

[0035] Specifically, such as Figure 4 As shown, the fault diagnosis and task generation module 300 includes: The multi-scale self-optimizing fault diagnosis unit 310 is configured to establish a multi-scale self-optimizing fault diagnosis learning model based on multi-channel time-series data of devices stored in the LIM model. This model is trained and optimized using running data to identify the root causes of device faults. The core of this unit is a deep learning model, whose input is time-series data fragments extracted from multiple related devices from the LIM model (e.g., current, voltage, and temperature data of all strings under an inverter, as well as ambient irradiance and wind speed). The multi-scale aspect is reflected in the following: on the time scale, the model simultaneously processes second-level transient changes (for detecting instantaneous arcs), minute-level fluctuations (for identifying MPPT tracking anomalies), and day-level trends (for detecting performance degradation); on the spatial scale, the model analyzes the correlation between data from a single device and data from adjacent devices (e.g., when the current of a certain string decreases while the surrounding strings are normal, it may indicate a fault in that string; if the current of all strings in the entire area decreases, it may be due to shading or reduced irradiance). Self-optimization is reflected in the model's online learning mechanism. When a new fault occurs and is manually confirmed, the sample is added to the training set, triggering incremental updates to the model and allowing it to continuously adapt to newly emerging fault modes. In the initial stages of model training, a historical fault sample library is used. As the system operates, the sample library is continuously enriched, and the diagnostic accuracy continues to improve. The output of this unit is the fault type (e.g., string short circuit, inverter overvoltage) and fault probability, as well as possible root causes (e.g., "hot spot in the second module of the third string").

[0036] The multimodal thermal anomaly detection unit 320 is configured to store device image data based on the LIM model, fusing infrared thermal images and visible light images. It employs a small target detection algorithm to extract photovoltaic string regions and detect hot spot faults in modules, and uses a clustering algorithm to automatically segment and locate faulty modules. This unit is specifically designed for hidden defects such as hot spots in photovoltaic modules. First, it acquires pairs of infrared and visible light images captured by drones or robots during inspections from the LIM model. Since photovoltaic modules occupy a small portion of the distant image, the YOLOv8 algorithm, optimized for small targets, is used for string region detection. This algorithm uses a large number of labeled infrared-visible light paired image samples during training to learn how to accurately locate strings in complex backgrounds. After detecting a string, the infrared temperature matrix of each string is extracted. Thresholding segmentation and DBSCAN clustering are used to identify pixel clusters with significantly higher temperatures than the surrounding pixels; these clusters correspond to candidate hot spot regions. Then, the visible light image is used to verify the region, eliminating false detections caused by reflection and shadows. Finally, the precise location of the hot spot module (e.g., "second row, third module") and the hot spot level are output. The detection results are stored in the LIM model as part of the fault samples.

[0037] The fault root cause localization unit 330 is configured to integrate historical operational data and image data stored in the LIM model. It combines signal processing and image analysis techniques to extract edge contour features of the equipment in infrared and visible light images under fault conditions and compare them with those under normal conditions to locate the faulty component and defect category. This unit is used to further analyze the detected faults and determine the specific defect type (e.g., cracks, obstructions, aging, diode short circuits). The method involves comparing images at the time of the fault with images from a normal period (historical normal images of the same equipment retrieved from the LIM model). After image registration, edge contour features are extracted (e.g., crack lines on the component surface, changes in solder ribbon areas), and signal processing techniques are used to analyze the electrical waveforms before and after the fault (e.g., IV curve distortion features). By inputting the image features and electrical features into a multimodal classifier (e.g., support vector machine or random forest), which is trained on a large number of labeled samples (samples including image-electrical feature pairs corresponding to various defects), the classifier can output the defect category and confidence level. This process advances fault detection from "discovery" to "understanding," providing precise guidance for subsequent maintenance and inspection.

[0038] The degradation trend prediction unit 340 is configured to extract key equipment degradation electrical characteristics from equipment electrical operation data stored in the LIM model, construct a degradation trend prediction model, and predict early degradation states of equipment and generate early warning tasks through learning and training on operational data. Degradation electrical characteristics include photovoltaic module power degradation rate (calculated by comparing actual power with theoretical power under the same irradiance), inverter efficiency degradation rate (comparing DC input with AC output), and branch current dispersion rate (the coefficient of variation of branch currents under the same combiner box), etc. These indicators are extracted from long-term series data in the LIM model. Then, a time-series prediction model (such as LSTM or Prophet) is constructed and trained using historical indicator data to learn the degradation patterns of each piece of equipment. The training data comes from historical maintenance records marked with each stage: "normal - slight degradation - severe degradation - failure". The model can predict the changing trends of each indicator over a future period. When the predicted value is about to exceed a preset threshold, an early warning task is generated to prompt maintenance personnel to intervene in advance. This unit transforms passive maintenance into proactive early warning, effectively reducing the rate of sudden failures.

[0039] The four units work collaboratively: multi-scale self-optimizing fault diagnosis and multi-modal thermal anomaly detection to identify anomalies in real time, root cause localization for in-depth analysis of causes, and degradation trend prediction and early warning of potential risks. All detected anomalies, faults, and warnings are summarized to generate a list of inspection task requirements (e.g., "perform close-range infrared re-inspection of suspected hot spot areas") and a list of maintenance task requirements (e.g., "replace the third component in the second row" and "clean obstructions from the surface of the fifth string"). The task list includes information such as task type, target equipment, location coordinates, and priority.

[0040] The multi-objective collaborative scheduling module 400 is configured to integrate the inspection task requirement list, maintenance task requirement list, environmental information, robot status, and resource constraints. It employs a composite objective optimization algorithm to generate an inspection task scheduling scheme, which includes task allocation, path planning, and resource scheduling. This module directly addresses the problem of "lack of scientific methods for inspection resource scheduling" in the background technology, achieving optimal resource allocation and dynamic task collaboration.

[0041] Specifically, such as Figure 5 As shown, the multi-target cooperative scheduling module 400 includes: The data integration unit 410 is configured to integrate environmental and equipment data, inspection task requirement lists, maintenance task requirement lists, robot status, and resource scheduling constraint data output by the LIM model, and then clean and standardize them. Robot status includes the current location, battery level, current task execution status (idle / busy / faulty), and types of sensors and tools carried by each intelligent maintenance robot. Resource scheduling constraints include maintenance personnel working hours, drone flight restrictions (such as no-fly zones and maximum flight radius), and safety regulations (such as prohibiting high-altitude operations in rainy weather). All data is uniformly formatted into the input parameters required by the optimization algorithm.

[0042] The decision rule base unit 420 is configured to set operation and maintenance decision rules under different working conditions based on business scenarios and operation and maintenance strategies, transforming complex objectives into decision variables and constraints. Decision rules are stored in the form of a knowledge base, such as: priority rules (critical fault tasks have higher priority than general inspection tasks, and early warning tasks have the lowest priority), safety rules (drones are prohibited from taking off when wind speeds exceed level 6, and robots are prohibited from operating outdoors at night), efficiency rules (prioritizing the allocation of the nearest available robot, and path planning should avoid repeatedly passing through the same area), and economic rules (considering both robot and labor costs, prioritizing high-value areas), etc. These rules can be extracted from historical operation and maintenance experience and continuously adjusted through simulation optimization. The rule base supports dynamic configuration to adapt to strategy changes in different seasons and different operation and maintenance stages.

[0043] The optimization unit 430 is configured to construct a land-air integrated autonomous operation and maintenance decision optimization algorithm model with the objectives of optimizing efficiency, safety, and cost, and to solve for task allocation, path planning, and resource scheduling schemes. This problem is modeled as a multi-objective combinatorial optimization problem, with objective functions including: minimizing the total task completion time (efficiency), maximizing the safety margin (safety), and minimizing energy consumption and manpower costs (cost). Decision variables include: which robot (or personnel) each task is assigned to, the order in which the robots execute tasks, and the paths to each task point. Constraints include robot power limits, task time windows, equipment compatibility (e.g., only robots equipped with infrared cameras can perform hot spot re-inspection), obstacle avoidance constraints, etc. The solution algorithm uses an improved genetic algorithm or ant colony algorithm to search for the Pareto optimal solution set in the solution space. During algorithm execution, environmental data from the LIM model (e.g., the impact of terrain slope on robot movement speed) is called in real time to calculate path costs. The solution outputs a series of task allocation schemes and execution paths, such as: "Robot A: Current position - Task point 2 (replace component) - Task point 5 (inspection), estimated time 45 minutes, power consumption 30%; Drone B: Task point 1 (infrared inspection) - Task point 3 (visible light photography), estimated time 20 minutes, be careful to avoid high-voltage lines." This scheduling scheme is sent to the robot execution and control module.

[0044] The multi-objective collaborative scheduling module 400 is also configured to store equipment performance data based on the LIM model, and combine it with key operational indicators of the photovoltaic power plant to construct a power plant health evaluation model. This generates evaluation results for power plant performance indicators and macro-operational indicators, providing a basis for scheduling decisions. This health evaluation model uses fuzzy set theory to transform performance indicators such as equipment conversion efficiency, failure rate, dispersion rate, and maintenance costs extracted from the LIM model into fuzzy sets. The weights of each indicator are determined through the analytic hierarchy process combined with expert experience, a fuzzy evaluation matrix is ​​constructed, and fuzzy comprehensive evaluation is performed. After defuzzification, the health level of the equipment and the overall power plant is obtained. The evaluation results include a health ranking for each equipment type, a list of inefficient equipment, predicted power generation losses, and suggested optimization measures. These results are directly input into the decision rule base unit 420 to influence scheduling priority settings: faulty equipment has the highest task priority, deteriorated equipment generates preventative maintenance tasks, sub-healthy equipment has increased inspection frequency, and a global operation and maintenance strategy adjustment is triggered when the overall health of the power plant declines. This function combines technical and operational indicators, enabling scheduling decisions to consider both current task execution and long-term operational benefits of the power plant, achieving a leap from equipment-level fault diagnosis to power plant-level health assessment. For example, equipment in a "faulty" state has its maintenance tasks prioritized to the highest level; equipment in a "deteriorated" state generates preventative maintenance tasks with a higher priority than routine inspections; equipment in a "sub-healthy" state has its inspection frequency increased and is included in the key monitoring list; when the overall health of the power plant declines, it triggers adjustments to the global operation and maintenance strategy, such as increasing the density of drone inspections and arranging special investigations. Furthermore, health evaluation results are also used for long-term operation and maintenance planning. For example, based on equipment deterioration trends, health changes over the next six months can be predicted, allowing for advance equipment replacement plans; comparing the health performance of equipment from different manufacturers provides a basis for procurement decisions; and analyzing differences in equipment health across different regions optimizes power plant layout and operation and maintenance resource allocation.

[0045] The dynamic rescheduling unit 440 is configured to update the task queue and resource status when a new fault, robot malfunction, or environmental change is detected during task execution. This triggers rescheduling optimization and dynamically adjusts the allocation scheme and execution path of unexecuted tasks. For example, if a robot malfunctions and cannot continue performing a task, or a new emergency fault is detected requiring immediate attention, or a sudden downpour causes outdoor operations to halt, the dynamic rescheduling unit will capture these events in real time, pause the original plan, re-input the remaining unfinished tasks and new tasks into the optimization solution unit, generate a new scheduling scheme, and notify the relevant execution units. This mechanism ensures the system can quickly respond to dynamic changes and always maintain the optimality of the scheduling scheme.

[0046] Robot Execution and Control Module 500 The robot execution and control module 500 is configured to receive a scheduling scheme, control at least one intelligent maintenance robot to autonomously navigate to the target location to perform inspection or maintenance operations, and provide real-time feedback on the execution status. This module achieves automated replacement of the "maintenance task execution relying on manual operation" in the background technology.

[0047] Specifically, such as Figure 6 As shown, the robot execution and control module 500 includes at least one intelligent maintenance and inspection robot, which includes: The Autonomous Navigation Chassis System 510 is configured to build an environmental map based on multi-sensor fusion perception of environmental information, plan and track the optimal driving path, and achieve autonomous driving and dynamic obstacle avoidance. The system uses LiDAR, cameras, and high-precision positioning systems (such as RTK-GPS) to fuse environmental perception, and employs the SLAM algorithm to construct a local environmental map in real time and match it with the global map (from a LIM model) for local positioning. Upon receiving a path from the scheduling plan, the chassis system uses a graph search algorithm (such as A*) to search for a drivable path on the real-time map, combines chassis dynamics characteristics (such as minimum turning radius and maximum climbing angle) to generate a smooth driving trajectory, and uses a PID control algorithm to achieve trajectory tracking. During driving, the system uses sensors to detect obstacles (such as rocks, potholes, and people) in real time, and uses the Dynamic Window Method (DWA) to perform local path replanning to achieve obstacle avoidance. The system also has an autonomous charging function, automatically returning to a charging station when the battery level falls below a threshold.

[0048] The Dual-Light Gimbal Vision System 520 is configured to acquire 2D or 3D image information of target devices, accurately identify targets and extract features through multi-view fusion, and achieve spatial pose calibration. The Dual-Light Gimbal Vision System includes a high-resolution infrared thermal imager and a visible light camera, mounted on a rotatable gimbal. When the robot approaches the task point, the vision system performs coarse alignment based on the device's prior position information (such as GPS coordinates and altitude) in the LIM model, and then accurately identifies the target device (such as an inverter switch or combiner box door lock) through image recognition algorithms (such as template matching or deep learning target detection). The system employs a multi-view geometric algorithm, combined with gimbal angle and laser ranging, to calculate the target's 3D position and orientation (i.e., spatial pose) relative to the robot base, providing precise coordinates for the robotic arm's operation. During robotic arm operation, the vision system continuously tracks the target, updates pose information in real time, and compensates for deviations caused by minor movements of the robot chassis or target vibrations, achieving dynamic closed-loop control.

[0049] A multi-DOF robotic arm and end effector 530 are configured to plan the robotic arm's motion trajectory based on visual positioning results, driving the end effector to perform operations on the target device. The robotic arm employs a 6-DOF design, providing sufficient flexibility and workspace. The motion planning module plans the trajectory in joint space based on the target pose, uses inverse kinematics to solve for the joint angles, and generates a smooth motion curve through interpolation algorithms to ensure the robotic arm reaches the target position smoothly and without collisions. The end effector includes two quick-change modules: a toggle switch mechanism and a keyhole unlocking mechanism. The toggle switch mechanism, consisting of a geared motor and an adapter, is used to rotate the DC-side switch (similar to a knob) of the string inverter; the keyhole unlocking mechanism, consisting of a miniature linear actuator and a key head, is used to insert into the door lock of the DC combiner box and rotate to unlock it. The control system pre-activates the corresponding actuator according to the task type ("operating the switch" or "opening the door"). When the robotic arm end effector reaches the predetermined position, it drives the actuator to complete the action and confirms the success of the operation through force feedback or visual feedback. After the operation is completed, the robot reports the execution result to the control center and updates the task status.

[0050] During system operation, the various modules collaborate closely: the data fusion and optimization module continuously gathers data, the environmental perception and modeling module constantly updates the LIM model, the fault diagnosis and task generation module analyzes data in real time and generates a task list, the multi-objective collaborative scheduling module dynamically optimizes resource allocation, and the robot execution and control module precisely executes tasks and feeds the results back to the data fusion module, forming a closed loop. All models in the system (CNN segmentation model, fault diagnosis model, degradation prediction model, and optimized scheduling model) are periodically or online retrained using continuously accumulated sample data (including manually confirmed fault samples and successfully executed scheduling cases), achieving self-learning and self-optimization, and continuously improving the system's intelligence level.

[0051] Figure 7 This is a schematic diagram of the electronic device 700 provided in an embodiment of this application. For example... Figure 7 As shown, the electronic device 700 of this embodiment includes a processor 701, a memory 702, and a computer program 703 stored in the memory 702 and executable on the processor 701. When the processor 701 executes the computer program 703, it implements the steps of the system executing the corresponding method described above.

[0052] Electronic device 700 can be a desktop computer, laptop, handheld computer, cloud server, or other electronic device. Electronic device 700 may include, but is not limited to, a processor 701 and a memory 702. Those skilled in the art will understand that... Figure 7 This is merely an example of electronic device 700 and does not constitute a limitation on electronic device 700. It may include more or fewer parts than shown, or different parts.

[0053] The processor 701 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0054] The memory 702 can be an internal storage unit of the electronic device 700, such as a hard disk or RAM of the electronic device 700. The memory 702 can also be an external storage device of the electronic device 700, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the electronic device 700. The memory 702 can also include both internal and external storage units of the electronic device 700. The memory 702 is used to store computer programs and other programs and data required by the electronic device.

[0055] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0056] If integrated modules / units are implemented as software functional units and sold or used as independent products, they can be stored in a readable storage medium (e.g., a computer-readable storage medium). Based on this understanding, all or part of the processes executed by the methods of each module / unit in this application can also be implemented by a computer program instructing related hardware. The computer program can be stored in a readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable storage medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0057] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A dynamic scheduling and control system for collaborative inspection tasks of inspection robots, characterized in that, include: The data fusion and optimization module is configured to collect multi-source data from photovoltaic power plants, establish unified data standards, and fuse and optimize equipment data from different sources and time scales to form a full lifecycle information dataset. The environmental perception and modeling module is configured to extract environmental and equipment features of the photovoltaic power station based on the full life cycle information dataset, construct a digital environment model library and an equipment physical feature model library, and couple them to form a LIM model; The fault diagnosis and task generation module is configured to use the LIM model as data support, adopt a multimodal fusion fault diagnosis algorithm to identify the type and location of equipment faults, and combine equipment deterioration trend prediction to generate an inspection task requirement list and a maintenance task requirement list. The multi-objective collaborative scheduling module is configured to integrate the inspection task requirement list, the maintenance task requirement list, environmental information, robot status and resource constraints, and generate a scheduling scheme for the inspection tasks using a composite objective optimization algorithm. The scheduling scheme includes task allocation, path planning and resource scheduling. The robot execution and control module is configured to receive the scheduling scheme, control at least one intelligent maintenance and inspection robot to autonomously navigate to the target location to perform inspection or maintenance operations, and provide real-time feedback on the execution status.

2. The system according to claim 1, characterized in that, The data fusion and optimization module includes: The data acquisition unit is configured to acquire basic equipment data, equipment operation data, equipment anomaly data, maintenance and troubleshooting data, as well as drone inspection data and robot inspection data. The data preprocessing unit is configured to clean, standardize, and convert the format of the collected data. The fusion optimization unit is configured to use a unified data standard to associate and fuse device data from different sources and at different time scales to form the full lifecycle information dataset.

3. The system according to claim 1, characterized in that, The environmental perception and modeling module includes: The digital environment model library construction unit is configured to combine GIS, satellite remote sensing imagery and UAV aerial survey data to extract the elevation changes, slope undulations and geographical obstacle distribution characteristics of photovoltaic power stations, construct a three-dimensional digital environment model library and photovoltaic area road network topology, and extract road material and slope characteristics. The equipment physical feature model library construction unit is configured to perform pixel-level classification of photovoltaic array images through a convolutional neural network, realize string area positioning and segmentation, and extract string azimuth, tilt angle, component elevation and installation gap feature parameters. The LIM model building unit is configured to couple the digital environment model library with the device physical feature model library to build a unified storage and management system for the full lifecycle information model of environment features, device features, and operational data.

4. The system according to claim 1, characterized in that, The fault diagnosis and task generation module includes: The multi-scale self-optimizing fault diagnosis unit is configured to establish a multi-scale self-optimizing fault diagnosis learning model based on the device's multi-channel time-series data stored in the LIM model, and to identify the root causes of device faults by training and optimizing the model through running data. The multimodal thermal anomaly detection unit is configured to integrate infrared thermal images and visible light images based on the device image data stored in the LIM model, use a small target detection algorithm to extract photovoltaic string regions and detect hot spot faults in the modules, and use a clustering algorithm to achieve automatic segmentation and location of faulty modules. The fault root cause localization unit is configured to integrate signal processing and image analysis techniques based on the historical operating data and image data stored in the LIM model, extract the edge contour features of the equipment in infrared and visible light images under fault conditions and compare them with the normal state to locate the faulty components and defect categories. The degradation trend prediction unit is configured to extract key equipment degradation electrical characteristics based on the equipment electrical operation data stored in the LIM model, construct a degradation trend prediction model, and predict early degradation status of equipment and generate early warning tasks through learning and training on the operation data.

5. The system according to claim 1, characterized in that, The multi-objective cooperative scheduling module includes: The data integration unit is configured to integrate the environmental and equipment data output by the LIM model, the inspection task requirement list, the maintenance task requirement list, robot status, and resource scheduling constraint data, and perform cleaning and standardization. The decision rule base unit is configured to set operation and maintenance decision rules under different working conditions according to business scenarios and operation and maintenance strategies, and to transform composite objectives into decision variables and constraints. The optimization and solution unit is configured to construct a land-air integrated autonomous operation and maintenance decision optimization algorithm model with the goal of optimizing efficiency, safety, and cost, and solve for task allocation, path planning, and resource scheduling schemes; The dynamic rescheduling unit is configured to update the task queue and resource status when a new fault, robot fault, or environmental change is detected during task execution, trigger rescheduling optimization, and dynamically adjust the allocation scheme and execution path of unexecuted tasks.

6. The system according to claim 1, characterized in that, The robot execution and control module includes at least one intelligent maintenance and inspection robot, which includes: The autonomous navigation chassis system is configured to build an environmental map based on multi-sensor fusion perception of environmental information, plan and track the optimal driving path, and achieve autonomous driving and dynamic obstacle avoidance; The dual-light gimbal vision system is configured to acquire two-dimensional or three-dimensional image information of the target device, identify the target and extract features through multi-view fusion, and achieve spatial pose calibration. A multi-degree-of-freedom robotic arm and an end effector are configured to plan the robotic arm's motion trajectory based on visual positioning results and drive the end effector to perform operations on the target device. The end effector includes a toggle switch mechanism and an insertion lock hole unlocking mechanism.

7. The system according to claim 6, characterized in that, The autonomous navigation chassis system is also configured to: perceive the environment by fusing LiDAR, cameras and high-precision positioning system, search for drivable paths using graph search algorithm, generate optimized driving trajectories by combining chassis dynamics characteristics, and achieve trajectory tracking through control algorithm.

8. The system according to claim 6, characterized in that, The dual-light gimbal vision system is also configured to update the spatial pose information of the target device during the operation of the robotic arm, providing positioning for the real-time trajectory planning and autonomous operation of the robotic arm.

9. The system according to claim 1, characterized in that, The multi-objective collaborative scheduling module is also configured to: construct a power plant health evaluation model based on the equipment performance data stored in the LIM model and in combination with key operating indicators of the photovoltaic power plant, generate evaluation results of power plant performance indicators and macro-operating indicators, and provide a basis for scheduling decisions.

Images