Planning method of pumped storage power station unmanned aerial vehicle inspection complex task

Abstract

The invention discloses a planning method for a pumped storage power station unmanned aerial vehicle inspection complex task. The method comprises the following steps: based on historical inspection data of the pumped storage power station, identifying key indexes and modes in inspection through big data analysis so as to construct a multi-dimensional historical inspection data set; based on historical maintenance records and real-time monitoring data of pumped storage power station equipment, using a time sequence analysis and neural network model to construct an equipment fault prediction model; obtaining potential faults and maintenance requirements of the equipment by using the equipment fault prediction model; and constructing an intelligent path inspection model by adopting a path planning algorithm so as to formulate an intelligent path inspection plan for the unmanned aerial vehicle inspection complex task. The unmanned aerial vehicle is used for collecting real-time data of inspection, and the inspection task path is dynamically adjusted. Through the method, the efficiency, the accuracy and the adaptability of the unmanned aerial vehicle to the completion of the complicated inspection task are ensured, and a powerful technical support is provided for the safe operation of the pumped storage power station.

Description

Technical Field

[0001] The present invention generally relates to the field of artificial intelligence technology. More specifically, the present invention relates to a planning method for complex tasks of unmanned aerial vehicle inspection of pumped storage power stations. Background Art

[0002] Pumped storage power stations are often located in areas with large terrain fluctuations and covering a wide area. The traditional manual inspection mode is inefficient and has blind spots. Especially in high-altitude areas or special meteorological conditions, such as low temperature, wind and sand, ice and snow, the harsh meteorological conditions in high-altitude areas make it very difficult to complete the inspection tasks.

[0003] Drones can quickly respond to special operations, take real-life photos of the risks of slope landslides and collapses at pumped-storage power stations, and promptly inform the on-site project department to prevent safety risks in advance. Therefore, it is particularly important to develop a reasonable inspection task route for drones to ensure that pumped-storage power station failures can be discovered in time and not missed. In addition, for inspections across highways, high-speed railways, and high-voltage lines that have a greater economic and social impact, as well as inspections in areas with a higher failure rate, drone inspections need to reduce interference with these areas and improve inspection efficiency.

[0004] In view of this, there is an urgent need to provide a planning method for complex inspection tasks of pumped-storage power stations by drones, so as to improve the accuracy and efficiency of drones in discovering problems when performing complex inspection tasks. Summary of the invention

[0005] In order to at least solve one or more of the technical problems mentioned above, the present invention provides a planning method for complex inspection tasks of unmanned aerial vehicles in pumped-storage power stations, including: based on the historical inspection data of the pumped-storage power station, identifying key indicators and patterns in the inspection through big data analysis to construct a multi-dimensional historical inspection data set. Based on the historical maintenance records and real-time monitoring data of the pumped-storage power station equipment, an equipment failure prediction model is constructed using time series analysis and neural network models. Based on the historical inspection data set, the equipment failure prediction model is used to obtain potential equipment failures and maintenance requirements. Based on the complex inspection tasks of the unmanned aerial vehicle and / or the potential failures and maintenance requirements of the equipment, a path planning algorithm is used to construct an intelligent path inspection model to formulate an intelligent path inspection plan for the complex inspection tasks of the unmanned aerial vehicle.

[0006] According to the intelligent path inspection planning, the real-time data of the inspection is collected by drones to perform data fusion analysis and / or equipment status diagnosis; wherein, the real-time data is fused and analyzed by artificial intelligence algorithms to detect pumped storage power station faults; according to the historical inspection data set and the real-time data, the deep learning model is used to accurately diagnose the equipment status and identify the fault type and location. According to the data fusion analysis and / or equipment status diagnosis results, the adaptive control algorithm is used to analyze to dynamically adjust the inspection task path. The results of the inspection tasks are collected and the results are analyzed through big data analysis to improve the inspection task planning scheme of the intelligent path inspection model.

[0007] In some embodiments, the planning method also includes: using expert-assisted decision-making to adjust the planning of complex drone inspection tasks; wherein, the expert-assisted decision-making includes using AR devices to present data analysis results and key information to inspection personnel to provide real-time auxiliary planning support.

[0008] In some embodiments, based on the historical inspection data of the pumped-storage power station, identifying key indicators and patterns in the inspection through big data analysis to construct a multi-dimensional historical inspection data set includes: preprocessing the historical inspection data, the preprocessing includes data cleaning, standardization and / or formatting; using big data statistical methods or machine learning algorithms to identify key indicators related to fault prediction in the preprocessed historical inspection data, and constructing fault association features; using clustering algorithms (such as K-means, hierarchical clustering) to classify equipment states to identify feature vectors under different state modes; using association rule learning algorithms to obtain the association relationship between the fault association features and the equipment state feature vectors to identify equipment component combinations with frequent faults; analyzing time series data to identify fault periodicity patterns and trend changes; and creating fact tables and dimension tables according to preset data models to construct a multi-dimensional historical inspection data set to store data.

[0009] In some embodiments, according to the complex inspection tasks of the UAV and / or the potential failures and maintenance needs of the equipment, a path planning algorithm is used to construct an intelligent path inspection model to formulate an intelligent path inspection plan for the complex inspection tasks of the UAV, including: using the path planning algorithm of the Dijkstra algorithm, taking different areas of the power station as nodes in the graph, the accessibility between areas as edges, and the priority as the weight of the edge, and setting priorities for different regional inspection tasks according to the fault prediction results of the equipment fault prediction model; according to the priority of the regional inspection task, using the A* algorithm and the TEB algorithm to obtain path planning in a complex environment; by simulating the inspection path of the UAV in the path planning in the complex environment in a virtual environment, the inspection path is optimized and adjusted to cover the preset key areas.

[0010] Through the above-provided planning method for complex inspection tasks of pumped storage power station drones, the embodiment of the present invention can construct a comprehensive scene information dataset by identifying key indicators and patterns in the inspection and constructing a multi-dimensional historical inspection dataset. This dataset will be used for subsequent task planning and risk assessment, and enhance the decision-making ability of drones in the inspection planning process.

[0011] Further using time series analysis and neural network models, we build equipment fault prediction models, and using path planning algorithms, we build intelligent path inspection models, so that we can formulate intelligent path inspection plans for the complex inspection tasks of the drones. By using drones to perform intelligent path inspection planning, we can reduce manual intervention and improve inspection efficiency. By using drones to automatically complete inspection tasks within a set time and upload data in real time, we can ensure the timeliness and accuracy of information, thereby ensuring the efficiency, accuracy and adaptability of drones in completing complex inspection tasks, and providing strong technical support for the safe operation of pumped storage power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0013] Figure 1 An exemplary flow chart of a method for planning a complex task of drone inspection of a pumped-storage power station according to some embodiments of the present invention is shown. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0015] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0016] Figure 1 An exemplary flow chart of a method 100 for planning a complex task of drone inspection of a pumped-storage power station according to some embodiments of the present invention is shown.

[0017] like Figure 1As shown, in step 101 of method 100, based on the historical inspection data of the pumped-storage power station, key indicators and patterns are obtained by identifying the historical inspection data through big data analysis, and a multi-dimensional historical inspection data set is constructed based on the key indicators and patterns.

[0018] Specifically in an embodiment scenario, the implementation process may include:

[0019] First, data is collected from historical inspection records, which may include but are not limited to: equipment status records, inspection time, inspection personnel, faults and defects found, maintenance and repair records, and environmental data (such as wind speed, temperature, humidity, etc.).

[0020] Then, the collected historical inspection data is preprocessed, which may include data cleaning, standardization and / or formatting. Data cleaning may include but is not limited to removing duplicate records, deleting missing values ​​or filling missing data, identifying and processing outliers. Standardization is the conversion of data into a unified metric and format to ensure that the data is comparable and consistent between different analysis tools and models, such as converting timestamps to a unified date and time format, standardizing or normalizing numerical data. Formatting can also include data classification and coding, which facilitates subsequent analysis and modeling processes by dividing data into different categories and coding methods.

[0021] Then, from the preprocessed data, statistical methods or machine learning algorithms are used to select the key indicators most relevant to fault prediction and construct fault association features. For example, correlation analysis, mutual information, chi-square test and other methods can be used to evaluate the importance of key indicators.

[0022] Use big data statistical algorithms or machine learning algorithms to extract key features that may be relevant to fault diagnosis and prediction. For example: equipment operating parameters (such as temperature, pressure, etc.), frequency and time of fault occurrence, maintenance cycle and maintenance type, the impact of environmental changes on equipment performance, etc. And construct fault-related features, such as the year, month, week, etc. extracted from the timestamp, or construct new feature combinations from equipment parameters.

[0023] Then, use big data analysis techniques, such as clustering algorithms (such as K-means, hierarchical clustering), adaptive neural networks, or other statistical pattern recognition methods to identify key indicators and patterns in inspections. For example: use clustering algorithms to classify device states and identify device state feature vectors under different device state modes. Apply adaptive neural networks to perform pattern recognition processing on device states to complete the description of each sensor about the target.

[0024] Use association rule learning algorithms, such as the Apriori algorithm, to obtain the association between fault association features and device status feature vectors to identify combinations of device components with frequent failures. The basic steps may include:

[0025] Generate candidate itemsets: Starting from a single item, generate all possible itemsets.

[0026] Calculate support: Count the number of times each candidate item set appears in the data set.

[0027] Filter frequent item sets: Filter out frequent item sets based on the minimum support threshold.

[0028] Generate association rules: Generate all possible association rules from frequent itemsets.

[0029] Filter strong association rules: Filter out strong association rules based on the minimum confidence threshold.

[0030] In order to make it easier for those skilled in the art to understand the above association implementation, some of the terms are explained. Among them, support refers to the frequency of occurrence of an item set in all inspections. For example, if 10% of the inspection data in the data set contains a certain inspection task or fault combination, the support of the task or fault combination is 0.1.

[0031] Confidence is the ratio of the frequency of the condition item set and the result item set appearing together to the frequency of the condition item set appearing. For example, if 70% of the inspections that include task A also include task B, then the confidence of rule A->B is 0.7.

[0032] Lift is an indicator that measures the strength of association between item sets, and is calculated as the ratio of confidence to the support of the resulting item set. A lift greater than 1 indicates a positive correlation between the item sets.

[0033] Therefore, the planning method provided by the present invention can use the association rule learning algorithm to find the combination of components with frequent failures in power station equipment, thereby providing data support for inspection task planning. In addition, this method can also identify potential risk points, optimize inspection paths, and improve the reliability and safety of power station equipment.

[0034] Furthermore, the time series data in the historical inspection data are analyzed to identify the periodic patterns and trend changes of faults.

[0035] Periodic pattern recognition can be achieved by performing frequency domain analysis on time series data using Fast Fourier Transform (FFT) to obtain the basic frequency components of the data. Seasonal or cyclical patterns can be identified by describing the similarity between the time series and its own lagged values ​​using the autocorrelation function (ACF). Periodicity can be identified by observing the repetitive patterns of the data in different time periods through time domain analysis.

[0036] Trend changes can be achieved by using seasonal decomposition methods, such as Seasonal Decomposition of Time Series (STS), to decompose the time series into trend, seasonal, and random components. For linear trends, a linear regression model can be used to fit the data and identify trend changes. And for nonlinear trends, nonlinear models, such as the Holt-Winters nonlinear trend model, can be used to model and predict trend changes.

[0037] By effectively analyzing the time series data in the historical inspection data of power station equipment, the periodic patterns and trend changes of equipment can be identified, thus providing a scientific basis for inspection task planning.

[0038] Furthermore, based on the preset data model, fact tables and dimension tables are created to construct a multi-dimensional historical inspection data set to store the data.

[0039] In one embodiment, the metric data may be stored in the fact table, and the detailed information of the dimension may be stored in the dimension table; and in the data warehouse, a multi-dimensional historical inspection data set may be constructed through data aggregation and indexing based on predefined dimensions and metrics. In one embodiment of the present invention, the dimensions may include equipment type, location, inspection time, etc., and the metrics may include the number of failures, the number of maintenances, inspection results, etc. The preset data model may be designed as a data warehouse architecture of a star model or a snowflake model. The star model directly connects all dimension tables to a single fact table, while the snowflake model further normalizes the dimension table to form a hierarchical structure. Metrics are aggregated in the fact table to support fast queries. For example, the number of failures and maintenance per day can be pre-aggregated, and indexes can be established for key columns of the fact table and dimension table to improve query performance.

[0040] The following is an example of creating these tables:

[0041] First, clarify the goals and business requirements of data analysis. For example, it is necessary to analyze the failure rate, maintenance cycle, and inspection frequency of pumped storage power station equipment.

[0042] Then design dimension tables to contain descriptive information to provide context for data analysis. Here are some possible dimension tables:

[0043] The device dimension table (Devices) may contain the following fields: DeviceID (device ID), DeviceType (device type), Location (location), InstallationDate (installation date), and MaintenanceTeam (maintenance team).

[0044] The time dimension table (Time) can contain fields: DateID (date ID), Date (actual date), DayOfWeek (day of the week), Month (month), Year (year), Hour (hour).

[0045] The maintenance dimension table (Maintenance) can contain fields: MaintenanceID (maintenance ID), MaintenanceDate (maintenance date), and MaintenanceType (maintenance type).

[0046] Design a fact table to contain quantitative business indicators for analysis. A fact table is usually associated with one or more dimension tables. The following are possible fact tables:

[0047] The inspection fact table (Inspections) may contain the following fields: InspectionID (inspection ID), DeviceID (foreign key, associated with the device dimension table), DateID (foreign key, associated with the time dimension table), MaintenanceID (foreign key, associated with the maintenance dimension table), InspectionDate (inspection date), FaultFound (whether a fault was found), NumberOfIssues (number of issues), and InspectionType (inspection type).

[0048] The above can use SQL language to create the structure of these tables in the database. Then fill the historical inspection data into these tables. In some implementation scenarios, data migration, transformation and loading (ETL) process may be involved.

[0049] The multi-dimensional historical inspection data set constructed by the present invention can be used to analyze failure modes, predict maintenance needs, and optimize inspection plans.

[0050] Further, the process proceeds to step 102: Based on the historical maintenance records and real-time monitoring data of the pumped storage power station, an equipment failure prediction model is constructed through time series analysis and a neural network model.

[0051] In one embodiment, the above-mentioned construction of the equipment failure prediction model may include: extracting features related to equipment failure prediction, such as equipment operating parameters, historical failure frequencies, maintenance cycles, etc. Statistical methods or machine learning algorithms may be used for feature selection and construction.

[0052] Then, a time series analysis algorithm is used to analyze the historical operation data of the equipment to identify potential failure modes. In some embodiment scenarios, the analysis method formula may include:

[0053]

[0054] Among them, B is the back-off operator, φ is the autoregressive term, θ is the moving average term, d is the difference order, which is used to improve the stability of the time series; p and q are the orders of the autoregressive term and the moving average term, respectively; D is the seasonal difference order; s is the seasonal period; Φ i and θ j are seasonal autoregressive and moving average parameters; k and m are the orders of seasonal autoregressive and moving average terms; Z t is a vector of external variables, which may include equipment parameters and historical failure rates; β is a coefficient vector of external variables; ε t It's white noise.

[0055] The parameters that have a direct impact on the health status of the equipment set by the above analysis formula may include the operating temperature, pressure, vibration frequency, etc. of the equipment. In the analysis model, Z t As external variables, they can help the model capture device-specific operating patterns and abnormal behaviors. For example, if the operating temperature of a device frequently exceeds a threshold, this may be a precursor to failure.

[0056] In addition, historical failure data can provide the frequency and pattern of equipment failures. Using historical failure data as input to the above formula can help predict future failure trends. In the model of the above analysis method, the historical failure rate can be used as Z t A part of it is input into the analysis model together with the time series data to improve the accuracy of fault prediction.

[0057] For time series data with seasonal characteristics, the seasonal difference order D and the seasonal period s are key parameters. They are used to capture and handle seasonal changes in the data. For example, if equipment is more likely to fail during a certain period of the year (such as winter), the seasonal component can help the model identify this pattern.

[0058] The parameters of the autoregressive term (p) and the moving average term (q) determine the order of the autoregressive and moving average terms in the model, which will affect the model's dependence on historical data and the model's response speed to past errors.

[0059] Furthermore, in some implementation scenarios, a recurrent neural network LSTM can be used, and on this basis, by introducing factors such as equipment parameters and historical failure rates, and considering the seasonal characteristics of time series data, the seasonal difference order D and the seasonal cycle s are introduced to build an equipment failure prediction model. The construction steps include:

[0060] Implement seasonal difference operation to handle seasonal changes: Δ D x t =(1-B s ) D x t ,

[0061] Where Δ represents the differential operation, B s represents the seasonal backward operator, which moves the data backward by s time units, x t is the time series data at the current time step, such as temperature readings; D is the seasonal difference order. Seasonal difference operations help eliminate seasonal effects, allowing the model to focus on non-seasonal trends and patterns. For example, if equipment is more likely to fail during certain periods of the year (such as winter), the seasonal component can help the model identify this pattern.

[0062] In the forget gate of the LSTM unit, a sigmoid activation function is used to output a value between 0 and 1. The formula includes:

[0063] f t =σ(w f ·[h t-1 ,Δ D x t ,x t-s ,Z t ]+b f ),

[0064] Among them, f t is the activation vector of the forget gate, w f is the weight matrix of the forget gate, b f is the bias vector, σ is the sigmoid activation function; x t is the time series data of the current time step, x t-s represents the time series data before s period, h t-1 is the hidden state of the previous time step, Z t are external variables, which can be equipment parameters and / or historical failure rates.

[0065] At the input gate of the LSTM unit, the formula for calculating how much new information should be updated to the unit state at the current time step includes:

[0066] i t =σ(W i ·[h t-1 ,Δ D x t ,x t-s ,Z t ]+b i ),

[0067]

[0068] Among them, i t is time series data, is the candidate cell state, W i and W c are the weight matrices of the input gate and candidate cell states, respectively, and b i and b c are the bias vectors of the input gate and candidate cell state respectively; tanh is the hyperbolic tangent activation function, which is used to map the input to the (-1,1) interval, representing the value of the cell state.

[0069] The cell state is updated by combining the results of the forget gate and the input gate. The formula includes:

[0070]

[0071] Among them, C t is the cell state at the current time step, C t-1 is the cell state at the previous time step; * indicates element-wise multiplication.

[0072] At the output gate of the LSTM unit, it is determined whether the information in the cell state is passed to the hidden state of the next moment or is used as the model output at the current moment. The formula includes:

[0073] o t =σ(W o ·[h t-1 ,Δ D x t ,x t-s ,Z t ]+b o ),

[0074] Among them, the output activation value o t From the current input x t and the previous hidden state h t-1 It is calculated through a fully connected layer with a sigmoid activation function; W o is the weight matrix of the output gate, b o is the bias vector for the output gate.

[0075] Then, the final hidden state h is obtained by element-wise multiplication of the value of the cell state after the tanh function. t , the formula is:

[0076] h t =o t *tanh(C t ),

[0077] Among them, t is the activation vector of the outputs; C t is the cell state at the current time step, h t is the hidden state at the current time step.

[0078] In the analysis method of the embodiment of constructing the equipment failure prediction model, the equipment parameter Z t The device parameter Z introduced in the same embodiment scenario as above t , which may include the operating temperature, pressure, vibration frequency, etc. of the equipment, are parameters that have a direct impact on the health of the equipment. t As additional feature inputs, they can help the model capture device-specific operating patterns and abnormal behaviors. For example, if a device’s operating temperature frequently exceeds a threshold, this could be a precursor to failure.

[0079] Historical failure data is used as model input to provide the frequency and pattern of equipment failures, thereby helping to predict future failure trends. In the above formula, the historical failure rate can be used as Z t A part of it is input into the model together with the time series data to improve the accuracy of fault prediction.

[0080] In addition, the parameter weight matrix (W) and bias vector (b) need to be learned and optimized through training data. Among them, the weight matrix determines the influence of the input data on the cell state and hidden state, while the bias vector provides adjustment space for the model, so that the model can better adapt to different data characteristics.

[0081] In the above-mentioned equipment failure prediction model, the seasonal adjustment of seasonal changes is handled by introducing the seasonal difference order D and the seasonal cycle s. For time series data with seasonal characteristics, it can help the model identify and adapt to seasonal patterns and improve the accuracy of prediction, making it suitable for the complex inspection environment of pumped storage power stations with obvious seasonal changes and affected by multiple factors.

[0082] Through the joint work of the parameters in the above steps, the improved LSTM can capture long-term and short-term dependencies in time series data, and control the flow of information through the gating mechanism, thereby avoiding the gradient problem of traditional RNN. In the pumped storage power station equipment failure prediction model, the setting and adjustment of these parameters will directly affect the prediction performance and accuracy of the equipment failure prediction model. By carefully designing and adjusting these parameters, the model's ability to predict potential equipment failures and maintenance needs can be improved.

[0083] Furthermore, historical data can be used to train the model, and the model performance can be optimized through cross-validation, adjusting hyperparameters, etc. At the same time, in order to improve the prediction accuracy of the model, the particle swarm optimization (PSO) algorithm can be used to optimize the key parameters of the LSTM model. The PSO algorithm can improve the performance of the model by finding the optimal solution in the parameter space. In addition, appropriate evaluation indicators such as accuracy, recall, F1 score, etc. are used to evaluate the performance of the model.

[0084] Through the above steps, a device failure prediction model based on time series analysis and neural network model can be constructed to improve the operating efficiency and reliability of pumped storage power station equipment. In the actual implementation scenario, appropriate algorithms and tools can be selected to implement this process in combination with specific business needs and data characteristics. For example, deep learning frameworks such as TensorFlow or Keras can be used to build and train LSTM models.

[0085] Further, the process proceeds to step 103: based on the multi-dimensional historical inspection data set, using the equipment failure prediction model, the potential failures and maintenance requirements of the equipment are obtained.

[0086] In one embodiment, the trained equipment failure prediction model is used to predict the current equipment status and identify potential failures and maintenance needs. A machine learning-based equipment predictive maintenance method can be used to process high-dimensional and multivariate data and extract hidden relationships in the data in complex and dynamic environments. It should be noted that as time goes by, new data will continue to be generated, and the model needs to be retrained with new data regularly to maintain the accuracy and adaptability of the model. This can be achieved through deep learning techniques, especially recurrent neural networks (RNN) and long short-term memory networks (LSTM), which have significant effects on processing time series data.

[0087] Further, the process proceeds to step 104: according to the complex inspection tasks of the drone and / or potential failures and maintenance requirements, a path planning algorithm is used to construct an intelligent path inspection model, and an intelligent path inspection planning scheme is formulated for the complex inspection tasks of the drone.

[0088] In one embodiment, the steps include:

[0089] First, the path planning algorithm of Dijkstra algorithm is adopted, and different areas of the power station are taken as nodes in the graph, the accessibility between areas is taken as edges, and the priority is taken as the weight of the edge. According to the fault prediction results of the equipment fault prediction model, the priorities are set for the inspection tasks of different areas. The specific implementation method can be:

[0090] First, the graph structure of the pumped storage power station is defined, where nodes represent different areas of the power station and edges represent the accessibility between areas. The weight of each edge can be set based on the output of the equipment failure prediction model to reflect the inspection priority of different areas.

[0091] Secondly, the weights can be set based on the output of the equipment failure prediction model. For example, if the equipment in a certain area has a higher probability of failure, then the outbound weight of that area should be lower to ensure that the inspection path prioritizes these high-risk areas. The weight setting can be expressed as:

[0092]

[0093] Among them, w ij is the weight of the edge from region i to region j, p ij is the predicted probability of failure from region i to region j. Therefore, the region with a higher probability of failure has a smaller corresponding edge weight, and thus has a higher priority in the Dijkstra algorithm.

[0094] Finally, when initializing the Dijkstra algorithm, the starting node is set to any area of ​​the power station, and the distances of all nodes are set to infinity, except for the distance of the starting node, which is set to 0. The Dijkstra algorithm is applied to find the shortest path from the starting node to all other nodes. The core steps of the algorithm include: selecting the node with the smallest distance among the unvisited nodes as the current node; updating the distance from the current node to its adjacent nodes; repeating the above steps until all nodes have been visited.

[0095] Based on the output of the equipment failure prediction model, priorities are set for different areas. This can be achieved by adjusting the edge weights in the Dijkstra algorithm to ensure that high-priority areas are given priority in the inspection path.

[0096] During the inspection process, if the equipment failure prediction model provides new data, the path can be adjusted dynamically. For example, if the equipment in a certain area suddenly shows a higher risk of failure, the Dijkstra algorithm can be rerun to update the inspection path.

[0097] Furthermore, according to the priority of regional inspection tasks, the A* algorithm and TEB algorithm are used to obtain path planning in complex environments.

[0098] When planning paths in complex environments, both the A* algorithm and the TEB algorithm are effective choices, but they differ in some aspects and can complement each other. The following is a specific step and method for combining these two algorithms for path planning through an example scenario:

[0099] The A algorithm is a heuristic search algorithm that evaluates the total length of a path by combining the actual cost and the estimated cost. The evaluation function of the A algorithm is usually expressed as: f(n) = g(n) + h(n), where g(n) represents the actual cost from the starting point to the current node, and h(n) represents the estimated cost from the current node to the target node.

[0100] The above estimated cost h(n) can introduce dynamic weighting coefficients in the heuristic function to adapt to different terrain and environmental characteristics. For example, the weight of the heuristic function can be adjusted according to the complexity of different terrains, so that the algorithm pays more attention to exploration in complex terrains and more attention to efficiency in simple terrains.

[0101] In complex environments, heuristic functions can be combined with environmental perception information, such as obstacle distribution, terrain undulations, etc. The formulas include:

[0102] h(n)=d(n)+α·O(n)+β·T(n),

[0103] Among them, d(n) is the distance from node n to the target node, O(n) is the obstacle influence factor, T(n) is the terrain undulation influence factor, α and β are weight coefficients, which can be adjusted according to the actual environmental characteristics.

[0104] The TEB algorithm (Timed Elastic Band) is suitable for local path planning in dynamic environments. It optimizes the path through the concept of elastic band, making the path smoother and in line with the kinematic constraints of the drone. The TEB algorithm takes into account the speed and acceleration limits of the drone, as well as the impact of dynamic obstacles.

[0105] In order to improve the efficiency and quality of path planning, the A algorithm can be used for global path planning to generate a rough path, and then the TEB algorithm can be used to optimize the local path. This method combines the advantages of the A algorithm in global path search and the ability of the TEB algorithm in dealing with dynamic obstacles and kinematic constraints.

[0106] In actual application scenarios, especially in complex environments, unknown dynamic obstacles may be encountered. The TEB algorithm is a local path planning algorithm based on graph optimization, which has good dynamic obstacle avoidance capabilities. The TEB algorithm optimizes the path by adding vertices and edges to the graph optimization, considering the position and discrete time interval of the drone at each discrete moment, as well as constraining acceleration, speed and arrival time, and updates the path in real time to avoid obstacles. For example, when the drone encounters other dynamic obstacles caused by flying birds or bad weather during movement, the TEB algorithm can quickly replan the path to ensure that the drone can reach the target point smoothly. The experimental results show that the improved TEB algorithm can reduce the planning time of drone inspections, and has a smooth speed curve in the straight line segment without overshoot and oscillation. In addition, when encountering a curve, there is an obvious speed change process, and the planned local path is along the edge of the curve, avoiding energy loss caused by oversteering.

[0107] Therefore, through the above steps, the A* algorithm and the TEB algorithm can be effectively combined to perform efficient path planning in complex environments such as power stations, ensuring the smooth progress of power station inspection tasks.

[0108] Furthermore, by simulating the inspection path of the UAV in the path planning in the complex environment in a virtual environment, the inspection path is optimized and adjusted to cover the preset key areas.

[0109] In order to ensure that the planning of complex inspection tasks is more reasonable and the tasks are completed smoothly. The inspection path of the drone can be simulated in a virtual environment and optimized according to the preset key areas. This may involve adjusting the path to avoid new obstacles or giving priority to inspecting certain high-risk areas. Through the TEB algorithm, the path can be dynamically adjusted to adapt to environmental changes while maintaining the kinematic constraints of the drone. During the path planning process, ensure that all preset key areas are covered. This may require adding additional nodes to the path or adjusting the existing path to ensure that the drone can reach these areas. In the virtual environment, the inspection progress of the drone can be monitored in real time and adjusted as needed.

[0110] By effectively simulating the path planning of drones in complex environments in the virtual environment, and optimizing the inspection path to ensure that all preset key areas are covered, this method can not only improve the inspection efficiency, but also ensure the safety of drones when performing tasks.

[0111] Further, the process proceeds to step 105: according to the intelligent path inspection planning scheme, the real-time data of the inspection is collected by using a drone, and data fusion analysis and / or equipment status diagnosis is performed; wherein, the real-time data is fused and analyzed by an artificial intelligence algorithm to detect pumped storage power station faults; based on the historical inspection data set and the real-time data, a deep learning model is used to accurately diagnose the equipment status and identify the fault type and location.

[0112] When performing UAV inspection path planning in a complex environment, inspired by the embodiments of the present invention, those skilled in the art can combine the A* algorithm and the TEB algorithm to simultaneously consider multi-sensor data and environmental change data.

[0113] First, the data from different sensors, such as LiDAR, depth cameras, etc., are integrated through multi-sensor data fusion technology to obtain accurate perception of the environment. This helps to build an environmental map and identify obstacles and traversable areas. According to the search results, a low-cost and high-precision multi-sensor information fusion system can be built to improve the integrity and accuracy of the environmental map in view of the limitations of a single sensor.

[0114] In one embodiment, the fusion analysis of the real-time data by artificial intelligence algorithm to detect pumped storage power station faults includes: collecting real-time data based on at least one sensor equipped by the drone, the sensor including a thermal imager, a high-definition camera and / or a multi-spectral sensor; wherein the real-time data includes multi-sensor data and environmental change data.

[0115] According to the real-time data collected by the sensor, the Dempster-Shafer evidence theory is used to perform data fusion, and the fused data is used to perform fault detection and diagnosis using machine learning or deep learning models.

[0116] Based on the above description of the technical solution of the present invention, the specific implementation steps of data fusion are given below:

[0117] First, the data from different sensors are integrated through multi-sensor data fusion technology. Dempster-Shafer evidence theory is an effective method for processing uncertain information. It can combine evidence (i.e. data) from different sensors to obtain more reliable conclusions. In the data fusion process, the information provided by each sensor is regarded as a basic probability assignment (BPA), which is then combined through the rules of Dempster-Shafer theory to form a global BPA, thereby obtaining a comprehensive decision basis.

[0118] Secondly, use machine learning or deep learning models to extract features from the fused data. Deep learning models, such as convolutional neural networks (CNN) or long short-term memory networks (LSTM), can automatically learn and extract key features from data without human intervention. These features can be used for subsequent fault detection and diagnosis.

[0119] Finally, a fault detection and diagnosis model is constructed. This model can automatically associate the relationship between the collected data and the health status of the machine. The model can be constructed using a deep neural network (DNN), which can adaptively learn the machine's diagnostic knowledge from the collected data instead of relying on the engineer's experience and knowledge. For example, CNN can be used to process image data to identify the appearance defects of the equipment; LSTM can be used to process time series data to identify abnormal operation of the equipment.

[0120] Furthermore, the model is trained using historical fault data and real-time monitoring data. During the training process, the model performance can be optimized using methods such as cross-validation and adjusting the learning rate. The output of the model can be a probability distribution of the fault type, which is used to identify the fault type. In addition, during the operation of the power plant, the sensor data of the equipment is collected in real time, and the trained model is used for real-time status monitoring and fault prediction. According to the output of the model, the maintenance strategy and inspection plan are dynamically adjusted.

[0121] Through the above steps, the Dempster-Shafer evidence theory can be effectively used for data fusion, combined with machine learning or deep learning models for fault detection and diagnosis, to improve the operational safety and reliability of power station equipment. It can not only improve the accuracy of fault detection, but also help to achieve predictive maintenance and improve the inspection efficiency of power stations.

[0122] In one embodiment, based on historical inspection data sets and real-time data, a deep learning model is used to accurately diagnose the equipment status and identify the fault type and location, including:

[0123] Using the deep learning model of the long short-term memory network, features are extracted from the historical inspection data set and the real-time data. The specific implementation method of extracting features has been described in the previous section and will not be repeated here.

[0124] The extracted features are used to train a classification model to identify different fault types, wherein the classification model includes a deep neural network, and the formula includes:

[0125] z=w (2) φ(w (1) x+b (1) )+b (2) ,

[0126] Among them, x is the input feature, w (1) and b (1) are the weights and biases of the first layer, w (2) and b (2) are the weights and biases of the second layer, φ is the activation function ReLU,

[0127] Construct a fault diagnosis model and output the probability distribution of fault types to identify the fault type and location.

[0128] The fault diagnosis model includes: combining the features extracted by the deep learning model and the classification model to obtain the fault type, and adding a spatial analysis layer on the basis of the classification model, and / or using a convolutional neural network (CNN) to process image data to obtain the fault location.

[0129] Select a suitable deep learning network structure, such as CNN or RNN, to build a fault diagnosis model. The input of the model is the preprocessed data, and the output is the probability distribution of the fault type. For example, CNN can be used to process image data to identify the appearance defects of the equipment; LSTM can be used to process time series data to identify the operation anomalies of the equipment. When training and optimizing the model, historical fault data can be used to train the model. During the training process, it is necessary to select a suitable loss function and optimization algorithm to minimize the prediction error of the model and improve the generalization ability of the model.

[0130] Through the trained model, a probability value can be calculated for each fault category. In a probabilistic neural network (PNN), probability evaluation is usually achieved by normalizing each row of the similarity matrix to obtain the posterior probability of each category. The decision rule is usually to select the category with the highest posterior probability as the final fault diagnosis result.

[0131] Through the above steps, a fault diagnosis model that can output the probability distribution of fault types can be constructed to identify the fault type and location, thereby improving the operating safety and reliability of power station equipment.

[0132] Further, the process proceeds to step 106: according to the results of the data fusion analysis and / or the equipment status diagnosis, an adaptive control algorithm is used for analysis to dynamically adjust the inspection task path.

[0133] In one embodiment, the adaptive control algorithm can automatically adjust the controller parameters according to the real-time performance or characteristics of the system to maintain the stability and performance of the control system. For example, the adaptive PID control algorithm can automatically adjust the parameters of the PID controller (proportional, integral, and differential parameters) to cope with the dynamic changes of the system and the changes in environmental conditions. According to the actual situation of the pumped storage power station during the complex task of drone inspection, the formula of the adaptive control algorithm is adaptively adjusted, including:

[0134]

[0135] Where u(t) is the control input at time t, e(t) is the control error between the current inspection path and the ideal inspection path at time t, and K p (e(t)) is the proportional gain of the adaptive control inspection path to the error response speed dynamically adjusted according to the error e(t), K i (e(t)) is the integral gain of the control inspection path dynamically adjusted according to the error e(t) to the historical accumulated error, K d (e(t)) is the differential gain of the control inspection path that is dynamically adjusted according to the error e(t) in response to the error change rate; the above three gains are adjusted through adaptive adjustment coefficients respectively, and the environmental perception adjustment coefficients are used when the equipment parameters change, such as temperature, pressure and flow.

[0136] The above adjustment formula takes into account the real-time change of the inspection path control error, the historical accumulated error, and the error change rate, and dynamically adjusts the parameters according to the real-time operating parameters of the power plant equipment to adapt to environmental changes and improve control performance. In this way, the adaptive control algorithm can respond more flexibly to various complex situations in the inspection tasks of power plant equipment.

[0137] Combined with the fault detection and diagnosis results, the adaptive control algorithm is used to dynamically adjust the inspection path. For example, if the equipment in a certain area is predicted to have a high risk of failure, the adaptive control algorithm can increase the inspection frequency or priority of that area. During the inspection process, the operating status of the equipment and environmental changes are monitored in real time, and this data is fed back into the fault detection and diagnosis model. The adaptive control algorithm can dynamically adjust the inspection path based on this real-time data to ensure that the inspection task can adapt to changes in the environment.

[0138] Further, the process proceeds to step 107: collecting the results of the inspection tasks, using big data analysis to analyze the results, and improving the intelligent path inspection task planning scheme of the intelligent path inspection model.

[0139] In one embodiment, the inspection task planning scheme of the improved intelligent path inspection model includes: re-running the Dijkstra algorithm to update the inspection path. For example, when a device in a certain area fails or the prediction model updates the failure probability, the Dijkstra algorithm is re-run to update the inspection path.

[0140] In one embodiment, the planning method for complex drone inspection tasks of pumped-storage power stations provided by the present invention also includes: using expert assisted decision-making to adjust the planning of complex drone inspection tasks; wherein, expert assisted decision-making includes using AR devices to present data analysis results and key information to inspection personnel, providing real-time auxiliary planning support.

[0141] Although multiple embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art may conceive of many changes, modifications, and alternatives without departing from the thought and spirit of the present invention. It should be understood that in the process of practicing the present invention, various alternatives to the embodiments of the present invention described herein may be adopted. The appended claims are intended to define the scope of protection of the present invention, and therefore cover equivalents or alternatives within the scope of these claims.

Claims

1. A planning method for complex tasks of UAV inspection of pumped storage power stations, characterized in that: include: Based on the historical inspection data of the pumped storage power station, key indicators and patterns are obtained by identifying the historical inspection data through big data analysis, and a multi-dimensional historical inspection data set is constructed based on the key indicators and patterns; Based on the historical maintenance records and real-time monitoring data of the pumped storage power station, an equipment failure prediction model is constructed through time series and neural network models; According to the multi-dimensional historical inspection data set, using the equipment failure prediction model, potential failures and maintenance requirements are obtained; According to the complex inspection tasks of the drone and / or potential faults and maintenance requirements, a path planning algorithm is used to build an intelligent path inspection model, and an intelligent path inspection planning scheme is formulated for the complex inspection tasks of the drone; According to the intelligent path inspection planning scheme, the real-time inspection data is collected by drones to perform data fusion analysis and / or equipment status diagnosis; wherein the real-time data is fused and analyzed by artificial intelligence algorithms to detect pumped storage power station faults; based on the historical inspection data set and the real-time data, the deep learning model is used to accurately diagnose the equipment status and identify the fault type and location; According to the results of the data fusion analysis and / or equipment status diagnosis, an adaptive control algorithm is used to analyze and dynamically adjust the inspection task path; Collect the results of inspection tasks, analyze the results through big data analysis, and improve the intelligent path inspection planning scheme of the intelligent path inspection model.

2. The planning method according to claim 1, characterized in that: The planning method further comprises: Use experts to assist decision making and adjust the planning of complex drone inspection tasks; The expert-assisted decision-making includes using AR devices to present data analysis results and key information to inspection personnel, providing real-time auxiliary planning.

3. The planning method according to claim 1, characterized in that: Based on the historical inspection data of the pumped storage power station, identifying the historical inspection data through big data analysis to obtain key indicators and patterns, and constructing a multi-dimensional historical inspection data set based on the key indicators and patterns includes: Preprocessing historical inspection data, including data cleaning, standardization and formatting; Use big data statistical algorithms or machine learning algorithms to identify key indicators related to fault prediction in pre-processed historical inspection data and construct fault correlation features; Using clustering algorithm to classify the device states to identify the device state feature vectors under different device states; Using an association rule learning algorithm to obtain an association relationship between the fault association feature and the device state feature vector, and identifying a combination of device components that frequently fail; Analyzing the time series data in the historical inspection data to identify fault periodicity patterns and trend changes; According to the preset data model, create fact tables and dimension tables to build a multi-dimensional historical inspection data set to store data.

4. The planning method according to claim 3, characterized in that: The creating of fact tables and dimension tables according to the expected data model to construct a multi-dimensional historical inspection data set to store data includes: storing measurement data in the fact table and storing detailed information of dimensions in the dimension table; In the data warehouse, a multi-dimensional historical inspection data set is constructed through data aggregation and indexing based on predefined dimensions and metrics.

5. The planning method according to claim 1, characterized in that: The equipment failure prediction model is constructed based on the historical maintenance records and real-time monitoring data of the pumped storage power station through time series analysis and neural network model, including: Extract features relevant to equipment failure prediction, Use time series analysis algorithms to analyze historical equipment operation data to identify potential failure modes; Using the recurrent neural network LSTM, we build an equipment failure prediction model, including: In the forget gate of the LSTM unit, a sigmoid activation function is used to output a value between 0 and 1; At the input gate of the LSTM unit, calculate how much new information to update to the unit state at the current time step; The cell state is updated by combining the results of the forget gate and the input gate; At the output gate of the LSTM unit, determining whether the information in the cell state is passed to the hidden state at the next moment or is used as the model output at the current moment; The final hidden state h is obtained by element-wise multiplication of the value of the cell state after the tanh function. t .

6. The planning method according to claim 1, characterized in that: According to the complex inspection tasks of the UAV and / or the potential failure and maintenance requirements of the equipment, a path planning algorithm is used to construct an intelligent path inspection model to formulate an intelligent path inspection plan for the complex inspection tasks of the UAV, including: A path planning algorithm based on the Dijkstra algorithm is adopted, different areas of the pumped storage power station are taken as nodes in a graph, the accessibility between the areas is taken as edges, and the priority is taken as the weight of the edge. According to the fault prediction results of the equipment fault prediction model, priorities are set for inspection tasks in different areas. According to the priority of regional inspection tasks, the A* algorithm and TEB algorithm are used to obtain path planning in complex environments; By simulating the inspection path of the UAV in the path planning in the complex environment in a virtual environment, the inspection path is optimized and adjusted to cover the preset key areas.

7. The planning method according to claim 1, characterized in that: The method of fusing and analyzing the real-time data by an artificial intelligence algorithm to detect a pumped storage power station fault includes: Collecting real-time data according to at least one sensor equipped by the drone, wherein the sensor includes a thermal imager, a high-definition camera, and a multi-spectral sensor; wherein the real-time data includes multi-sensor data and environmental change data; According to the real-time data collected by the sensor, the Dempster-Shafer evidence theory is used to perform data fusion; Through the fused data, machine learning or deep learning models are used for fault detection and diagnosis.

8. The planning method according to claim 1, characterized in that: The method of using a deep learning model to accurately diagnose the equipment status and identify the fault type and location based on the historical inspection data set and the real-time data includes: Using a deep learning model of a long short-term memory network, device status features are extracted from historical inspection data sets and the real-time data; Training a classification model using the device status features to identify different types of device failures, wherein the classification model includes a deep neural network; Construct a fault diagnosis model and output the probability distribution of fault types to identify the fault type and location; The fault diagnosis model includes: combining the features extracted by the deep learning model and the classification model to obtain the fault type, and adding a spatial analysis layer on the basis of the classification model, using a convolutional neural network (CNN) to process image data to obtain the fault location.

9. The planning method according to claim 1, characterized in that: The intelligent path inspection planning scheme of the improved intelligent path inspection model includes: re-running the Dijkstra algorithm to update the inspection path.

Images