Hydropower station inspection method, system and equipment based on multi-technology intensive multi-department cooperation and medium

Abstract

The invention discloses a hydropower station routing inspection method, system, equipment and medium based on multi-technology intensive multi-department cooperation, and relates to the technical field of hydropower station routing inspection, and the method comprises the steps: collecting multi-source data, constructing a standardized data set through fusion preprocessing, training a fusion analysis decision model, and inputting the model into real-time data to carry out cooperation state diagnosis. An analysis report is generated, a department capability portrait and a global scheduling strategy are combined, an intelligent inspection scheme is generated, multiple departments are guided to carry out cooperative execution, the system feeds back inspection data in real time, dynamic closed-loop optimization is carried out on the scheme, and intelligentization, collaboration and high efficiency of hydropower station inspection are achieved. According to the method, a closed-loop system integrating intelligent diagnosis, accurate decision and dynamic optimization is constructed, so that leap-type upgrading of hydropower station inspection from passive response to active prediction and from isolated execution to global coordination is realized, and the inspection efficiency, accuracy and risk prevention and control capability are improved.

Description

Technical Field

[0001] This invention relates to the field of hydropower station inspection technology, specifically to hydropower station inspection methods, systems, equipment, and media based on multi-technology integration and multi-departmental collaboration. Background Technology

[0002] With the advancement of intelligent and digital power plants, hydropower stations will gradually build various inspection systems and equipment that can be used for power production, such as inspection robots, drones, image recognition based on industrial television, and online monitoring of various equipment. However, these systems and equipment are often self-contained and scattered throughout the power plant without unified management. Individual maintenance personnel are only accustomed to using one or a few new devices and cannot easily use all the newly built inspection systems to maximize their comprehensive benefits, making it difficult to improve the inspection efficiency of power plant maintenance personnel.

[0003] On the one hand, power plant production systems are highly complex and interconnected, involving multiple departments and specialties. Departments such as electrical, mechanical, automation, protection, measurement and control, hydraulic engineering, and operation scheduling each bear important responsibilities, and there is a close interdependence among these departments. On the other hand, with the gradual aging of equipment, changes in the operating environment, and the application of new technologies, safety risks in power plant production are constantly increasing. Single-department inspections are difficult to fully cover all aspects and stages of power plant production, and often overlook important information or potential hazards. Through collaborative inspections, equipment failures and safety hazards can be identified and resolved in a timely manner, reducing the probability of accidents and losses, and improving the safety management level and inspection efficiency of hydropower stations. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention provides a method, system, equipment and medium for hydropower station inspection based on multi-technology integration and multi-departmental collaboration, in order to solve the problem that the existing technology is unable to fully cover all aspects and stages of power station production, and that important information or hidden dangers are missed.

[0005] To address the aforementioned technical challenges, a hydropower station inspection method based on multi-technology integration and multi-departmental collaboration is proposed, including: Multi-source data is collected from hydropower stations, and standardized multi-source inspection datasets are generated through multi-source data fusion preprocessing. A fusion analysis and decision-making model is constructed and pre-trained. New multi-source inspection data is acquired in real time and input into the pre-trained fusion analysis and decision-making model. Collaborative status diagnosis is performed on the new multi-source inspection data, and automated and manual data flows are analyzed to generate an inspection data analysis and diagnostic report. The inspection data analysis and diagnostic report is parsed and transformed into tasks. Combined with the multi-dimensional capability profile of the inspection department, a smart inspection plan is generated based on a global collaborative scheduling strategy. The inspection department executes collaborative inspection tasks according to the smart inspection plan and provides real-time feedback on the inspection data. Based on the feedback data, the smart inspection plan is dynamically optimized in a closed loop.

[0006] As a preferred embodiment of the hydropower station inspection method based on multi-technology integration and multi-departmental collaboration described in this invention, the step of model pre-training includes training the model using a dual-stream pre-training architecture and processing automated inspection data and manual inspection data. During the training process of the first data processing stream, the model is guided to distinguish between normal and abnormal states of equipment by constructing a comparison relationship between data samples, and the model is driven to learn autonomously from the automated data sequence and summarize the physical correlation law of the equipment system. During the training process of the second data processing stream, structured features are extracted from historical manual inspection records to obtain multi-dimensional feature vectors. The influence of different features on decision-making is quantified by calculating the correlation weights between feature vectors, and feature dependency patterns with the same decision-making logic are clustered and integrated into a set of inspection decision rules. The physical correlation patterns learned through the first data processing stream are integrated with the inspection decision rule set formed by the second data processing stream to build and improve the analytical and reasoning capabilities of data analysis.

[0007] As a preferred embodiment of the hydropower station inspection method based on multi-technology integration and multi-department collaboration described in this invention, the physical correlation law reflecting the connection between equipment systems includes learning the possibility that the current abnormal event will affect functionally related equipment on the equipment spatial layout network when an abnormal state of a piece of equipment occurs. Learn the data chain reactions and change patterns triggered by initial anomalies along the energy or signal transmission paths between devices, as well as the time delay characteristics and chain reaction sequence exhibited when anomalies propagate along the device-related network.

[0008] As a preferred embodiment of the hydropower station inspection method based on multi-technology integration and multi-department collaboration described in this invention, the collaborative status diagnosis includes splitting the real-time acquired multi-source inspection data into a real-time automated inspection data stream and a real-time manual inspection data stream. For the real-time automated inspection data stream, anomaly analysis is performed using a pre-trained model, and the fault radiation range is predicted; for the real-time manual inspection data stream, the pre-trained analysis model is used for processing, and fault points with weights higher than a threshold are identified. The fault points to be investigated, the probability of fault occurrence, the fault range, the preliminary results of fault investigation, and hidden fault points are summarized and packaged to generate a structured inspection data analysis and diagnostic report.

[0009] As a preferred embodiment of the hydropower station inspection method based on multi-technology integration and multi-departmental collaboration described in this invention, the method for predicting the fault radiation range includes: using a spatiotemporal graph convolutional network to predict the fault radiation range, abstracting the hydropower station equipment system into a graph structure, with equipment as nodes, physical or functional connections between equipment as edges, and inputting real-time automated data as node attributes into the network. Based on pre-trained parameters, the network simulates the propagation of abnormal states along the edges and outputs a topology graph centered on the initial fault point, labeled with the probability of different associated equipment being affected and the expected impact delay, which is the fault radiation range. The prediction formula for spatiotemporal graph convolutional networks is expressed as: in, Let L be the feature matrix of the nodes in the l-th layer of the graph neural network. The node feature matrix of the (l+1)th layer of the graph neural network is the updated device state representation. This is the adjacency matrix of the hydropower station equipment system. The identity matrix has the same dimension as A. To add a self-connected adjacency matrix, for The degree matrix, The degree matrix Perform normalization operations. Let be the learnable weight matrix of the l-th layer. It is a non-linear activation function; The fault points whose identification weight is higher than the threshold include: using an attention mechanism to identify hidden fault points, constructing an attention network, using the feature vector extracted from real-time manual inspection data as the query, using the feature descriptions of various hidden faults stored in the knowledge base as keys and values, calculating the similarity score between the query and all keys, normalizing it into attention weights, and selecting the potential fault with the highest attention weight as the output hidden fault point. The formula for the attention mechanism is expressed as: in, Attention weight score This is a potential hidden fault in the knowledge base. This is the current symptom feature vector extracted from real-time manual inspection data. To conceal the fault The feature embedding vector, For the feature embedding vector of the current symptom s, For dot product, To take the exponent of the dot product result, This is the set of all potential hidden faults in the knowledge base. Index for potential hidden faults.

[0010] As a preferred embodiment of the hydropower station inspection method based on multi-technology integration and multi-departmental collaboration described in this invention, the generation of the intelligent inspection scheme includes parsing inspection data, analyzing and reporting diagnostic reports, converting each diagnostic conclusion into an inspection task, and forming an inspection task list. For each task in the inspection task list, calculate the three-dimensional capability profile index value of all candidate inspection departments, and assign each task to the inspection department with the best comprehensive index according to the preset optimization objective function. The formula for calculating the 3D capability profile index value is expressed as follows: in, Score the skill matching degree. The total number of qualification types required to perform the current inspection task. Let j be the weight coefficient of the j-th qualification in the current task. Let be the matching score of the inspection department for the j-th qualification, This represents the historical completion rate of similar tasks performed by the current inspection department. For load balancing, This represents the number of tasks currently being performed by the inspection department. To meet the maximum workload of the inspection department For qualification index; For each inspection department assigned a task, the ant colony optimization algorithm is used to minimize the total path length or total time required for the current department to complete all assigned tasks. Under the premise of satisfying the constraints of task execution order, geographical accessibility and safety procedures, the optimal inspection route is planned. The formula for planning the optimal inspection route is expressed as: in, To ultimately plan the optimal inspection path for the k-th inspection department, Let be the set of all possible inspection path sequences for the k-th inspection department. This is a candidate inspection path sequence to be assigned to the k-th inspection department. Path sequence The i-th position point in the array, From point To the next point The actual distance traveled For location point index; For shared inspection equipment that needs to be used sequentially by multiple departments, insert equipment handover nodes into the optimal inspection route, and define geographical coordinates and handover time windows for each node; Establish a unified time and space resource calendar to monitor the planned inspection time, routes and resource usage of all departments in real time. When it is detected that different departments plan to enter the same physical area or use the same exclusive resource at the same time, the task start time of a department is automatically adjusted, the detour route is replanned and the equipment handover time window is adjusted according to the preset priority rules.

[0011] As a preferred embodiment of the hydropower station inspection method based on multi-technology integration and multi-department collaboration described in this invention, the execution of dynamic closed-loop optimization includes calculating the progress deviation rate of each task and evaluating the task completion rate based on the inspection data fed back by each department in real time. The formula for calculating the schedule deviation rate is expressed as follows: in, This refers to the schedule deviation rate. This represents the percentage of planned progress. This represents the actual percentage of progress. When a department’s task progress deviation rate exceeds a preset threshold and is determined to be delayed, the system immediately scans the real-time status of the remaining departments, selects the department with the lowest current load rate and the shortest estimated time to reach the task point from the current location as the optimal replacement department, and dynamically reassigns the current task to the optimal replacement department. For departments where tasks are reassigned, an insertion heuristic algorithm is used to replan the original optimal inspection route. The newly inserted task is used as a node and inserted into the position of the existing task sequence. The path increase cost and time delay cost caused by each insertion scheme are calculated. The insertion point with the lowest overall cost is selected, and a new inspection route and schedule are generated and sent out in real time. The formula for calculating insertion cost is expressed as follows: in, These are the candidate insertion schemes. The overall cost of using insertion scheme O is as follows: This represents the increase in the total inspection path length after inserting a new task. This represents the increase in total task completion time after inserting a new task. Add a cost weighting factor to the path length. This is the weighting coefficient for time delay costs.

[0012] The beneficial effects of this preferred technical solution are as follows: by transforming diagnostic conclusions into intelligent task allocation, optimal path planning, and global resource coordination based on three-dimensional capability profiles, ant colony algorithms, and spatiotemporal resource calendars, it achieves scientific decision-making with optimal paths and no resource conflicts, maximizes human resource efficiency, reduces inspection costs, and ensures efficient and orderly multi-department collaborative operations.

[0013] As a preferred embodiment of the hydropower station inspection system based on multi-technology integration and multi-departmental collaboration described in this invention, it is characterized by including a data acquisition module, a model pre-training module, an inspection data analysis module, an inspection task generation module, and a feedback optimization module.

[0014] The data acquisition module is used to acquire multi-source inspection data by integrating automated inspection systems and manual inspection terminals of various departments, perform spatiotemporal alignment and multimodal fusion preprocessing operations on the collected heterogeneous data, and monitor and clean the data sources based on the data quality assessment mechanism to generate a multi-source inspection dataset.

[0015] The model pre-training module is used to pre-train the intelligent inspection model using a dual-stream learning architecture and historical multi-source inspection datasets. The automated data stream learns the inherent physical correlation laws of the equipment system through contrastive learning and spatiotemporal graph convolutional networks, while the manual data stream extracts the expert's decision-making logic from historical work orders and records through natural language processing and attention mechanisms to form a structured cluster of inspection processing rules.

[0016] The inspection data analysis module is used to divide the data into automated and manual streams. For automated data, it identifies the fault points to be investigated, calculates the probability of fault occurrence, and predicts the range of impact. For manually reported data, it matches the pre-stored decision rules to generate preliminary investigation results and uses an attention mechanism to discover hidden fault points that the automated system has not captured. All analysis results are integrated and encapsulated to generate an inspection data analysis and diagnostic report.

[0017] The inspection task generation module is used to parse the report content, generate an inspection task list, combine the multi-dimensional capability profiles dynamically constructed by each inspection department, match the optimal execution department for each task, and generate a smart inspection plan based on a global collaborative scheduling strategy.

[0018] The feedback optimization module is used to receive feedback data uploaded in real time by each inspection department during the execution of the smart inspection plan, and to evaluate the task completion rate. When task delays, resource conflicts or new situations are found, the optimization program is immediately started to dynamically redistribute inspection tasks and replan inspection paths, and to dynamically adjust the plan based on real-time feedback.

[0019] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of a method for inspecting a hydropower station based on multi-technology integration and multi-departmental collaboration.

[0020] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for inspecting a hydropower station based on multi-technology integration and multi-departmental collaboration.

[0021] The beneficial effects of this invention are as follows: The dual-stream pre-training architecture of this invention deeply integrates the physical laws of equipment with expert decision-making logic, forming an interpretable and generalizable structured domain knowledge model, which enhances the ability to capture latent faults; the quantitative learning of physical correlation laws enables accurate spatiotemporal prediction of fault impacts, supporting risk-oriented proactive protection; collaborative state diagnosis provides precise input for automatic scheduling by processing multi-source data in parallel and outputting machine-readable structured reports, and transforms experience-dependent implicit knowledge into a computable process; intelligent task allocation and path planning, based on three-dimensional capability profiling and multi-objective optimization algorithms, achieves optimal coordination of global resources and refined management of scarce resources; the dynamic closed-loop optimization mechanism improves the system's resilience and self-healing ability in the face of emergencies through real-time progress monitoring and minimum disturbance replanning. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 The above is a flowchart of a hydropower station inspection method based on multi-technology integration and multi-departmental collaboration, provided as an embodiment of the present invention.

[0024] Figure 2 The flowchart of a hydropower station inspection system based on multi-technology integration and multi-departmental collaboration is provided as an embodiment of the present invention. Detailed Implementation

[0025] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0026] Example 1, referring to Figure 1 As one embodiment of the present invention, a method for inspecting hydropower stations based on multi-technology integration and multi-departmental collaboration is provided, comprising: S100: Collect multi-source data from hydropower stations, generate standardized multi-source inspection datasets by performing multi-source data fusion preprocessing, construct a fusion analysis decision model, and perform model pre-training.

[0027] S200: Acquires new multi-source inspection data in real time and inputs it into the pre-trained fusion analysis decision model to perform collaborative status diagnosis on the new multi-source inspection data, analyzes automated and manual data flows, and generates inspection data analysis and diagnosis reports.

[0028] S300: Analyzes and transforms inspection data analysis and diagnostic reports into tasks, combines the multi-dimensional capability profile of the inspection department, and generates a smart inspection plan based on a global collaborative scheduling strategy. The inspection department executes collaborative inspection tasks according to the smart inspection plan and provides real-time feedback on the inspection data. Based on the feedback data, the smart inspection plan is dynamically optimized in a closed loop.

[0029] It should be noted that the dual-stream pre-training architecture enables the model to combine the ability to understand physical laws with the ability to reason based on expert experience, accurately predict the range of fault radiation and identify hidden fault points, and transform the diagnostic conclusions into quantitative tasks. Based on the three-dimensional capability profile and optimization algorithm, scientific personnel path planning and global resource coordination are carried out, and the inspection plan is endowed with adaptive and self-repair capabilities through dynamic closed-loop optimization.

[0030] Example 2, refer to Figure 1 This is a second embodiment of the present invention, which provides a hydropower station inspection method based on multi-technology integration and multi-departmental collaboration, including: In step S100, the multi-source data acquisition includes collecting automated inspection data and manual inspection data, specifically including steps S101~S102: S101: Automated inspection data includes real-time sensor data consisting of the amplitude value output per second by the vibration acceleration sensor deployed on the generator bearing and the temperature reading returned every 30 seconds by the transformer top oil temperature sensor; real-time image data consisting of high-definition visible light photos of insulator strings taken by inspection drones along preset routes and infrared thermal video streams collected by robots moving along the pipe gallery; and real-time equipment operating condition data including the opening and closing coil current curves of the circuit breaker and the start-up logs of the protection devices.

[0031] S102: Manual inspection data includes equipment anomaly description data, inspection log data, and inspection processing data containing handling measures and replacement part model information.

[0032] Furthermore, in this embodiment of the application, in step S100, the multi-source data fusion preprocessing includes steps S111~S114: S111: Perform spatiotemporal alignment, that is, retain the original timestamp for automated inspection data, and for manual, delayed non-real-time data, backtrack according to the actual occurrence time and insert it into the corresponding time series database, and associate it with sensor data. Spatial alignment is achieved by mapping multi-source inspection data to a unified three-dimensional equipment model coordinate system.

[0033] S112: A multimodal data fusion strategy combining feature-level and decision-level fusion is adopted. In the feature-level fusion stage, time-domain and frequency-domain features are extracted from time-series data (such as vibration and temperature) through a sliding window. High-dimensional feature vectors are extracted from image data using a pre-trained convolutional neural network. Semantic vectors are generated from text data through the BERT semantic model. The feature vectors of each modality are normalized and concatenated by channel to form a joint feature vector describing the overall state of the equipment.

[0034] S113: Before fusion, the data quality is assessed. The assessment mechanism takes into account the completeness, timeliness and consistency of the data and generates a quality score on a 100-point scale. The formula for quality assessment is expressed as follows: in, Score the data quality. To score for completeness, For timeliness scoring, A consistency score is given.

[0035] S114: When When the response level reaches the yellow level, it is determined that the data source has a slight anomaly, and the remaining data sources with physical correlation or functional redundancy are automatically retrieved for cross-validation; when When the red-level response threshold is reached, it indicates that the data source is seriously abnormal or invalid. The system will directly dispatch the nearest inspection personnel to re-collect the data on-site and issue a work order to the maintenance team. At the same time, the current data source will be automatically frozen, and the data will be marked as invalid until the maintenance is completed and will no longer participate in subsequent model analysis.

[0036] In an optional implementation, in step S100, the multi-source data fusion preprocessing further includes setting a fixed time window, aggregating all automated data within the same window, and assigning manual records to the corresponding window according to the entry time. Spatial alignment is simplified to device code matching. During fusion, static weights are preset for different modal data (such as vibration, infrared, and text). The basic features of each modality after simple standardization are weighted and averaged to generate fusion features. The quality assessment is based solely on whether the data is missing and whether it is within a physically reasonable threshold using Boolean judgment.

[0037] In another optional implementation, in step S100, the multi-source data fusion preprocessing may further include performing preliminary feature extraction locally on each data source node (such as sensor edge computing unit, inspection terminal), encrypting and uploading the features to the central server, the server aggregating the feature models of the same type of equipment from multiple nodes through a federated averaging algorithm, updating the global feature extraction model and distributing it to each node, and the data quality is indirectly evaluated by the update gradient magnitude and consistency of the local model on the node.

[0038] It should be noted that, in the embodiments of this application, in step S100, the model pre-training includes adopting a dual-stream learning architecture, comprising two independent pre-training streams for processing automated and manual inspection data, specifically including steps S121~S122: S121: In the pre-training stream of automated inspection data, the contrastive learning paradigm is used to train the model to distinguish between the healthy and abnormal states of the equipment by constructing data samples of the same equipment under normal conditions as positive samples and combining fault and normal state data as negative samples.

[0039] Spatiotemporal graph convolutional networks are introduced to abstract the entire hydropower station equipment system into a graph structure, where nodes represent specific equipment and automated inspection data, and edges represent physical or functional relationships between equipment. Spatiotemporal graph convolutional networks capture the spatial and temporal dependencies of data through multiple spatiotemporal convolutional blocks and employ the Chebyshev multinomial approximation method, learning the physical correlation patterns between devices, including spatial topological correlation, energy conduction dependence, and fault propagation delay. The Chebyshev polynomial approximation formula is expressed as: in, Let G be the graph convolution operation performed on node features x by a convolution filter g with parameter θ defined on graph G. For learnable parameters, Let the order be the Chebyshev polynomial. Let be the Chebyshev polynomial of order y. The normalized graph Laplacian matrix after scaling. The original normalized graph Laplacian matrix, Let L be the largest eigenvalue of matrix L. Let h be the identity matrix. The number of nodes in the graph. This is the node feature vector input to the current graph convolutional layer; The spatial topological association formula is expressed as: in, For the equipment Equipment under abnormal conditions The conditional probability of being affected. and These are two device nodes in the graph structure. and For equipment and The feature representation vector learned by the model To make vector and Perform the splicing operation. To compute the learnable weight matrix of spatial associations, For bias terms, This is the Sigmoid activation function.

[0040] S122: In the pre-training stream of manual inspection data, extract three-dimensional features of environmental status, equipment status and abnormal handling decisions from historical work orders and inspection record text data, and use multi-head attention mechanism to quantify the dependencies between features and determine the importance of a certain handling decision in a specific context. The multi-head attention mechanism is represented as: in, The result is calculated based on the attention mechanism. For querying the matrix, The key matrix, For value matrices, This is the transpose of the key matrix. The feature dimension of the key vector. This refers to the Softmax function.

[0041] The attention weight distribution is transformed into an "IF-THEN" form rule by knowledge distillation technology, and similar rules are grouped into different inspection processing rule clusters by DBSCAN clustering algorithm.

[0042] Based on a comprehensive evaluation mechanism, the rules are value-assessed and optimized from multiple dimensions, including accuracy, coverage, and novelty. The formula is as follows: in, Score the overall value of rule R. For inspection and handling rules, Let R be the accuracy of rule R. For the coverage of rule R, For the novelty of rule R, These are the weighting coefficients for the accuracy dimension. These are the weighting coefficients for the coverage dimension. These are the weighting coefficients for the novelty dimension.

[0043] In an optional implementation, step S100, the model pre-training further includes, in the automated data stream, using a masked autoencoder to reconstruct and pre-train the device time-series data to learn robust feature representations, and obtaining the inter-device associations by calculating the correlation coefficient matrix of the feature vectors; in the manual data stream, using a bag-of-words model and a frequent pattern mining algorithm, directly mining the association rules of "IF condition THEN decision" from historical work order texts to form a rule base.

[0044] In another optional implementation, in step S100, the model pre-training may further include: the automated data stream uses a static graph neural network to learn feature propagation on the device topology, with the initial features of the nodes being the device static parameters and operating averages; the manual data stream constructs a case library, storing historical inspection cases (environment, phenomena, processing, results) in vector form; during the pre-training phase, the matching network is trained to learn how to retrieve the most similar historical cases and solutions from the case library based on the current state features.

[0045] It should be noted that, in the embodiments of this application, the physical correlation between the learning devices in step S121 includes implementation through a spatiotemporal graph convolutional network, specifically including steps A1 to A3: A1: Spatial topological association: The conditional probability of how a learning device failure affects the state of adjacent nodes (devices) in the graph.

[0046] A2: Energy conduction dependence: Learning the functional relationship in which changes in upstream equipment parameters trigger a chain reaction in downstream equipment parameters along energy flow paths such as electric current, hydraulic power, and torque.

[0047] A3: Fault propagation delay: Learn the distribution of time delay required for a fault signal to propagate along a specific edge (connection); the network learns the patterns by performing convolution operations on device state graphs in multiple time periods, capturing both spatial proximity and temporal dynamics.

[0048] In an optional implementation, in step S121, the learning of the physical association between devices further includes initializing the network structure according to the device physical connection diagram, where nodes represent device state variables and edges represent possible causal or influence relationships, using historical automated data, and calculating the conditional probability distribution between node states through a parameter learning algorithm (maximum likelihood estimation), quantifying the probability (spatial association) of the remaining devices becoming abnormal when a device is abnormal, and how this probability evolves over time.

[0049] In another optional implementation, in step S121, the learning of the physical correlation between devices may further include modeling the influence relationship between devices as an approximate solution of a set of partial differential equations or ordinary differential equations, using a neural network (PINNs) to learn the parameters and functional forms in the equations, so that the network, under known initial conditions and boundary conditions, outputs a trajectory that can predict the change of the state of the remaining devices in the future, expressing the laws of energy conduction and fault propagation.

[0050] In step S200, the generation of the inspection data analysis and diagnostic report includes S201~S204: S201: Real-time automated inspection data includes real-time automated inspection data from sensors, drones, and robotic automation systems; real-time manual inspection data includes data entered by inspection personnel through terminals.

[0051] S202: For real-time automated inspection data, at the level of a single device, the degree of abnormal deviation of the real-time data is identified by comparing and learning the normal pattern. When the abnormal score exceeds the preset threshold, the current device is marked as a point to be investigated, and the probability of failure is calculated by combining historical data. The formula for calculating the degree of abnormal deviation is expressed as follows: in, The degree of abnormal deviation of the equipment at time t. This refers to the current automated inspection data of the equipment collected at time t. These are the feature representation vectors obtained after passing through the encoder network. This is the mean vector of the feature representation vectors of the device under normal (healthy) operating conditions. The square of the L2 norm; The formula for calculating the probability of failure is expressed as: in, Let D be the probability that the current equipment is indeed faulty given that an anomaly score D is observed. These are learnable weight coefficients. For learnable bias terms, It is a natural exponential function.

[0052] Based on the spatial topological associations, energy conduction dependencies, and fault propagation delay patterns learned during the pre-training phase, a spatiotemporal graph convolutional network is used for inference simulation to generate a fault radiation range map centered on the initial fault point and labeled with different impact probabilities and delays. The prediction formula for spatiotemporal graph convolutional networks is expressed as: in, Let L be the feature matrix of the nodes in the l-th layer of the graph neural network. The node feature matrix of the (l+1)th layer of the graph neural network is the updated device state representation. This is the adjacency matrix of the hydropower station equipment system. The identity matrix has the same dimension as A. To add a self-connected adjacency matrix, for The degree matrix, The degree matrix Perform normalization operations. Let be the learnable weight matrix of the l-th layer. It is a non-linear activation function.

[0053] S203: For real-time manual inspection data, natural language processing technology is used to perform feature parsing, generate feature vectors, and then match them with pre-trained rule clusters. The rule with the highest matching degree is selected as the fault diagnosis result.

[0054] The correlation between currently manually reported symptoms and various hidden faults in the knowledge base is quantified by using an attention mechanism, and hidden fault points with attention weights higher than the threshold are output as potential risks. The formula for the attention mechanism is expressed as: in, Attention weight score This is a potential hidden fault in the knowledge base. This is the current symptom feature vector extracted from real-time manual inspection data. To conceal the fault The feature embedding vector, For the feature embedding vector of the current symptom s, For dot product, To take the exponent of the dot product result, This is the set of all potential hidden faults in the knowledge base. Index for potential hidden faults.

[0055] S204: Integrate and encapsulate the information obtained from the analysis, including the fault points to be investigated, the probability of fault occurrence, the fault radiation range, the fault investigation results, and the information on hidden fault points, to generate an inspection data analysis and diagnosis report.

[0056] In step S300, generating the intelligent inspection plan includes steps S301 to S305: S301: Analyze each conclusion in the diagnostic report and translate it into a specific, executable inspection task that includes location, type, priority, and estimated time.

[0057] S302: To achieve intelligent allocation, a multi-dimensional capability profile is constructed for each inspection department, including three core dimensions: skill matching degree, location timeliness, and load balancing degree.

[0058] S303: Adopts a smart inspection collaboration strategy, uses ant colony optimization algorithm to plan the optimal inspection route for each department, taking into account task order, geographical accessibility and safety procedures, and sets up handover nodes containing time windows, locations and equipment identifiers for shared equipment that needs to be used across departments. The formula for planning the optimal inspection route is expressed as: in, To ultimately plan the optimal inspection path for the k-th inspection department, Let be the set of all possible inspection path sequences for the k-th inspection department. This is a candidate inspection path sequence to be assigned to the k-th inspection department. Path sequence The i-th position point in the array, From point To the next point The actual distance traveled For location point index.

[0059] S304: Monitors all departments' plans in real time through a unified spatiotemporal resource calendar. When conflicts in location or resource usage are detected, the conflict is automatically resolved by delaying the start, replanning the route, or adjusting the handover time, based on preset priority rules.

[0060] S305: For each task in the task list, calculate the capability profile indicators of all candidate departments, and through a multi-objective optimization function, with the goal of minimizing total skill cost, response time and load pressure, achieve the globally optimal task allocation under the constraints that each task is assigned only once and each department's tasks are not overloaded, thus forming a smart inspection solution. The multi-objective optimization function is expressed as follows: in, This represents the total number of inspection tasks. The total number of candidate inspection departments. This serves as an index for inspection tasks. For the inspection department's index, The decision variable is a binary variable. For department k regarding the execution of tasks Skill matching score, For department k, how do you reach the task from the current position? Estimated response time required at the location The current load balancing degree of department k, The weighting coefficient of skill cost in the overall objective function. The weighting coefficient of response time cost in the overall objective function. This represents the weighting coefficient of load cost in the overall objective function; The constraints are expressed as follows: in, This represents the maximum task capacity of department k.

[0061] Furthermore, in this embodiment of the application, in step S302, constructing a multi-dimensional capability profile includes steps B1 to B3: B1: Skills matching is quantified by comprehensively evaluating personnel qualification level and historical task completion rate. Higher qualifications score higher than lower qualifications, and excellent historical performance also improves the score. The formula for calculating skills matching is as follows: in, Score the skill matching degree. The total number of qualification types required to perform the current inspection task. Let j be the weight coefficient of the j-th qualification in the current task. Let be the matching score of the inspection department for the j-th qualification, This represents the historical completion rate of similar tasks performed by the current inspection department.

[0062] B2: Location timeliness is calculated by using the shortest path algorithm that takes into account terrain factors to determine the estimated response time of the department to the task point.

[0063] B3: Load balancing is assessed based on the relationship between the current number of tasks in a department and its maximum capacity to determine the level of busyness; The formula for calculating load balancing is as follows: in, For load balancing, This represents the number of tasks currently being performed by the inspection department. To meet the maximum workload of the inspection department For qualification index.

[0064] In an optional implementation, in step S302, constructing a multi-dimensional capability profile further includes: skill matching degree is simplified to whether the personnel possess the key qualifications required for the task, if yes, it is 1, otherwise it is 0; location timeliness is obtained by querying the standard travel time matrix between various areas within a preset fixed site; load balancing degree is simplified to whether the department currently has any available personnel (if yes, it is 1, otherwise it is 0), and the three are added together with fixed weights to obtain a total score.

[0065] In another optional implementation, in step S302, constructing a multi-dimensional capability profile may further include using the three dimensions of the capability profile as state features, the task allocation result as an action, and the task completion efficiency and quality as a reward, training a reinforcement learning agent, dynamically adjusting the weight allocation strategy of the three dimensions, and calculating the location timeliness considering the mutual influence of paths during multi-department collaborative inspections, and performing joint path planning to estimate a more accurate arrival time.

[0066] Furthermore, in step S300, the execution of dynamic closed-loop optimization includes steps S311 to S314: S311: During the inspection execution phase, a dynamic closed-loop optimization mechanism ensures the efficiency and agility of the plan. Each inspection department carries out collaborative work based on the smart inspection plan and uploads feedback data on task progress, newly discovered anomalies, actual completion time, and resource usage in real time through mobile terminals.

[0067] S312: Calculate the progress deviation rate of each task based on feedback data. When the progress deviation of a department's tasks exceeds a preset threshold, the task reassignment mechanism will be triggered. The formula for calculating the schedule deviation rate is expressed as follows: in, This refers to the schedule deviation rate. This represents the percentage of planned progress. This represents the actual percentage of progress.

[0068] S313: The system will scan all candidate departments with sufficient load and timely response, and select the optimal replacement department based on the comprehensive evaluation function from the dimensions of skill matching, response time and load balancing. The comprehensive evaluation function is expressed as: in, Score the skill matching degree. The best successor department selected after comprehensive evaluation. The estimated response time required to reach the mission location from the current location. For load balancing, This represents the weighting coefficient of the skills matching dimension in the dynamic redistribution assessment. In response to the weighting coefficient of the timeliness dimension in the dynamic redistribution assessment, This represents the weighting coefficient of the load balancing dimension in the dynamic redistribution evaluation.

[0069] S314: When tasks are reassigned, an insertion heuristic algorithm is used to replan the inspection routes for the affected departments. By calculating the path and time costs of inserting new tasks into different positions in the existing sequence, the solution with the lowest total cost is selected for path adjustment. The formula for calculating insertion cost is expressed as follows: in, These are the candidate insertion schemes. The overall cost of using insertion scheme O is as follows: This represents the increase in the total inspection path length after inserting a new task. This represents the increase in total task completion time after inserting a new task. Add a cost weighting factor to the path length. This is the weighting coefficient for time delay costs.

[0070] It should be noted that, for example, if Maintenance Department 1 experiences a serious delay due to a complex fault, the system will automatically reassign the subsequent task A, which has a high risk of delay, to Maintenance Department 2, which has just completed a task nearby and has a lighter workload. The system will also adjust the inspection routes and schedules of the two departments in real time and push the new optimization plan to the mobile terminal, thereby realizing dynamic and adaptive closed-loop optimization of the inspection process and ensuring efficient collaboration and timely response of the overall inspection work.

[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0072] Example 3, referring to Figure 2 The third embodiment of the present invention provides a hydropower station inspection system based on multi-technology integration and multi-departmental collaboration, including a data acquisition module, a model pre-training module, an inspection data analysis module, an inspection task generation module, and a feedback optimization module.

[0073] The data acquisition module is used to acquire multi-source inspection data by integrating automated inspection systems and manual inspection terminals of various departments, perform spatiotemporal alignment and multimodal fusion preprocessing operations on the collected heterogeneous data, and monitor and clean the data sources based on the data quality assessment mechanism to generate a multi-source inspection dataset.

[0074] The model pre-training module is used to pre-train the intelligent inspection model using a dual-stream learning architecture and historical multi-source inspection datasets. The automated data stream learns the inherent physical correlation laws of the equipment system through contrastive learning and spatiotemporal graph convolutional networks, while the manual data stream extracts the expert's decision-making logic from historical work orders and records through natural language processing and attention mechanisms to form a structured cluster of inspection processing rules.

[0075] The inspection data analysis module is used to divide the data into automated and manual streams. For automated data, it identifies the fault points to be investigated, calculates the probability of fault occurrence, and predicts the range of impact. For manually reported data, it matches the pre-stored decision rules to generate preliminary investigation results and uses an attention mechanism to discover hidden fault points that the automated system has not captured. All analysis results are integrated and encapsulated to generate an inspection data analysis and diagnostic report.

[0076] The inspection task generation module is used to parse the report content, generate an inspection task list, combine the multi-dimensional capability profiles dynamically constructed by each inspection department, match the optimal execution department for each task, and generate a smart inspection plan based on a global collaborative scheduling strategy.

[0077] The feedback optimization module is used to receive feedback data uploaded in real time by each inspection department during the execution of the smart inspection plan, and to evaluate the task completion rate. When task delays, resource conflicts or new situations are found, the optimization program is immediately started to dynamically redistribute inspection tasks and replan inspection paths, and to dynamically adjust the plan based on real-time feedback.

[0078] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

[0079] Example 4, the fourth embodiment of the present invention, differs from the previous three embodiments in that: If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0080] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0081] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0082] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

Claims

1. A hydropower station inspection method based on multi-technology integration and multi-departmental collaboration, characterized by: include, Multi-source data is collected from hydropower stations, and standardized multi-source inspection datasets are generated by performing multi-source data fusion preprocessing. A fusion analysis and decision-making model is constructed and the model is pre-trained. New multi-source inspection data is acquired in real time and input into a pre-trained fusion analysis and decision model to perform collaborative status diagnosis on the new multi-source inspection data, analyze automated and manual data flows, and generate inspection data analysis and diagnosis reports. The inspection data analysis and diagnostic report is parsed and transformed into tasks. Combined with the multi-dimensional capability profile of the inspection department, a smart inspection plan is generated based on the global collaborative scheduling strategy. The inspection department executes collaborative inspection tasks according to the smart inspection plan and feeds back the inspection data in real time. Based on the feedback data, the smart inspection plan is dynamically optimized in a closed loop.

2. The hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in claim 1, characterized in that: The process of model pre-training includes training the model using a dual-stream pre-training architecture and processing automated inspection data and manual inspection data. During the training process of the first data processing stream, the model is guided to distinguish between normal and abnormal states of equipment by constructing a comparison relationship between data samples, and the model is driven to learn autonomously from the automated data sequence and summarize the physical correlation law of the equipment system. During the training process of the second data processing stream, structured features are extracted from historical manual inspection records to obtain multi-dimensional feature vectors. The influence of different features on decision-making is quantified by calculating the correlation weights between feature vectors, and feature dependency patterns with the same decision-making logic are clustered and integrated into a set of inspection decision rules. The physical correlation patterns learned through the first data processing stream are integrated with the inspection decision rule set formed by the second data processing stream to build and improve the analytical and reasoning capabilities of data analysis.

3. The hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in claim 2, characterized in that: The inductive physical correlation laws reflecting the connection between equipment systems include learning the probability that when a device experiences a state abnormality, the current abnormal event will affect functionally related devices on the device spatial layout network. Learn the data chain reactions and change patterns triggered by initial anomalies along the energy or signal transmission paths between devices, as well as the time delay characteristics and chain reaction sequence exhibited when anomalies propagate along the device-related network.

4. The hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in claim 3, characterized in that: The collaborative status diagnosis includes splitting the real-time multi-source inspection data into a real-time automated inspection data stream and a real-time manual inspection data stream. For real-time automated inspection data streams, anomaly analysis is performed using pre-trained models, and the fault radiation range is predicted. The real-time manual inspection data stream is processed using a pre-trained analysis model, and fault points with weights higher than a threshold are identified. The fault points to be investigated, the probability of fault occurrence, the fault range, the preliminary results of fault investigation, and hidden fault points are summarized and packaged to generate a structured inspection data analysis and diagnostic report.

5. The hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in claim 4, characterized in that: The predicted fault radiation range includes using a spatiotemporal graph convolutional network to predict the fault radiation range. The hydropower station equipment system is abstracted into a graph structure, with equipment as nodes and physical or functional connections between equipment as edges. Real-time automated data is input into the network as node attributes. Based on the parameters learned through pre-training, the network simulates the propagation of abnormal states along the edges and outputs a topology graph centered on the initial fault point, labeled with the probability of different associated equipment being affected and the expected impact delay, which is the fault radiation range. The prediction formula for spatiotemporal graph convolutional networks is expressed as: in, Let L be the feature matrix of the nodes in the l-th layer of the graph neural network. The node feature matrix of the (l+1)th layer of the graph neural network is the updated device state representation. This is the adjacency matrix of the hydropower station equipment system. The identity matrix has the same dimension as A. To add a self-connected adjacency matrix, for The degree matrix, The degree matrix Perform normalization operations. Let be the learnable weight matrix of the l-th layer. It is a non-linear activation function; The fault points whose identification weight is higher than the threshold include: using an attention mechanism to identify hidden fault points, constructing an attention network, using the feature vector extracted from real-time manual inspection data as the query, using the feature descriptions of various hidden faults stored in the knowledge base as keys and values, calculating the similarity score between the query and all keys, normalizing it into attention weights, and selecting the potential fault with the highest attention weight as the output hidden fault point. The formula for the attention mechanism is expressed as: in, Attention weight score This is a potential hidden fault in the knowledge base. This is the current symptom feature vector extracted from real-time manual inspection data. To conceal the fault The feature embedding vector, For the feature embedding vector of the current symptom s, For dot product, To take the exponent of the dot product, This is the set of all potential hidden faults in the knowledge base. Index for potential hidden faults.

6. The hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in claim 5, characterized in that: The intelligent inspection solution includes parsing inspection data, analyzing diagnostic reports, converting each diagnostic conclusion into an inspection task, and forming an inspection task list. For each task in the inspection task list, calculate the three-dimensional capability profile index value of all candidate inspection departments, and assign each task to the inspection department with the best comprehensive index according to the preset optimization objective function. The formula for calculating the 3D capability profile index value is expressed as follows: in, Score the skill matching degree. The total number of qualification types required to perform the current inspection task. Let j be the weight coefficient of the j-th qualification in the current task. Let be the matching score of the inspection department for the j-th qualification, This represents the historical completion rate of similar tasks performed by the current inspection department. For load balancing, This represents the number of tasks currently being performed by the inspection department. To meet the maximum workload of the inspection department For qualification index; For each inspection department assigned a task, the ant colony optimization algorithm is used to minimize the total path length or total time required for the current department to complete all assigned tasks. Under the premise of satisfying the constraints of task execution order, geographical accessibility and safety procedures, the optimal inspection route is planned. The formula for planning the optimal inspection route is expressed as: in, To ultimately plan the optimal inspection path for the k-th inspection department, Let be the set of all possible inspection path sequences for the k-th inspection department. This is a candidate inspection path sequence to be assigned to the k-th inspection department. Path sequence The i-th position point in the array, From point To the next point The actual distance traveled For location point index; For shared inspection equipment that needs to be used sequentially by multiple departments, insert equipment handover nodes into the optimal inspection route, and define geographical coordinates and handover time windows for each node; Establish a unified time and space resource calendar to monitor the planned inspection time, routes and resource usage of all departments in real time. When it is detected that different departments plan to enter the same physical area or use the same exclusive resource at the same time, the task start time of a department is automatically adjusted, the detour route is replanned and the equipment handover time window is adjusted according to the preset priority rules.

7. The hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in claim 6, characterized in that: The execution of dynamic closed-loop optimization includes calculating the progress deviation rate of each task and evaluating the task completion rate based on the inspection data fed back by each department in real time. The formula for calculating the schedule deviation rate is expressed as follows: in, Schedule deviation rate This represents the percentage of planned progress. This represents the actual percentage of progress. When a department’s task progress deviation rate exceeds a preset threshold and is determined to be delayed, the system immediately scans the real-time status of the remaining departments, selects the department with the lowest current load rate and the shortest estimated time to reach the task point from the current location as the optimal replacement department, and dynamically reassigns the current task to the optimal replacement department. For departments where tasks are reassigned, an insertion heuristic algorithm is used to replan the original optimal inspection route. The newly inserted task is used as a node and inserted into the position of the existing task sequence. The path increase cost and time delay cost caused by each insertion scheme are calculated. The insertion point with the lowest overall cost is selected, and a new inspection route and schedule are generated and sent out in real time. The formula for calculating insertion cost is expressed as follows: in, These are the candidate insertion schemes. The overall cost of using insertion scheme O, This represents the increase in the total inspection path length after inserting a new task. This represents the increase in total task completion time after inserting a new task. Add a cost weighting factor to the path length. This is the weighting coefficient for time delay costs.

8. A hydropower station inspection system based on multi-technology integration and multi-departmental collaboration, employing the hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in any one of claims 1 to 7, characterized in that, It includes a data acquisition module, a model pre-training module, an inspection data analysis module, an inspection task generation module, and a feedback optimization module; The data acquisition module is used to acquire multi-source inspection data by integrating the automated inspection system and the manual inspection terminals of various departments, perform spatiotemporal alignment and multimodal fusion preprocessing operations on the collected heterogeneous data, and monitor and clean the data source based on the data quality assessment mechanism to generate a multi-source inspection dataset. The model pre-training module is used to pre-train the intelligent inspection model using a dual-stream learning architecture and historical multi-source inspection datasets. The automated data stream learns the inherent physical correlation laws of the equipment system through contrastive learning and spatiotemporal graph convolutional networks. The manual data stream extracts the expert's decision-making logic from historical work orders and records through natural language processing and attention mechanisms to form a structured cluster of inspection processing rules. The inspection data analysis module is used to divide the data into automated and manual streams. For automated data, it identifies the fault points to be investigated, calculates the probability of fault occurrence and predicts the range of impact. For manually reported data, it matches the pre-stored decision rules to generate preliminary investigation results and uses the attention mechanism to discover hidden fault points that the automated system has not captured. All analysis results are integrated and packaged to generate an inspection data analysis and diagnostic report. The inspection task generation module is used to parse the report content, generate an inspection task list, combine the multi-dimensional capability profiles dynamically constructed by each inspection department, match the optimal execution department for each task, and generate a smart inspection plan based on a global collaborative scheduling strategy. The feedback optimization module is used to receive feedback data uploaded in real time by each inspection department during the execution of the smart inspection plan, and to evaluate the task completion rate. When task delays, resource conflicts or new situations are found, the optimization program is immediately started to dynamically redistribute inspection tasks and replan inspection paths, and to dynamically adjust the plan based on real-time feedback.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the hydropower station inspection method based on multi-technology integration and multi-departmental collaboration as described in any one of claims 1 to 7.

Images