High-precision abnormality sensing operation and maintenance method and system based on unmanned aerial vehicle inspection

Abstract

The application discloses a high-precision anomaly perception operation and maintenance method and system based on unmanned aerial vehicle inspection, relates to the technical field of operation and maintenance management, and comprises the following steps: according to an inspection target area, multiple inspection targets are set, and features are sequentially mined to determine an inspection index set. A target unmanned aerial vehicle takes off according to a pre-designed plan through a take-off and landing charging integrated platform, monitors the inspection targets based on the inspection index set, and obtains inspection data by carrying a high-precision sensor group. An abnormal value of the data is identified and preprocessed to generate a detection data set. A multi-source heterogeneous anomaly perception algorithm model library is trained based on the inspection targets, the model library is called to perform matching analysis on the detection data set, and an anomaly perception report is generated. The report is transmitted to an operation and maintenance center, linkage operation and maintenance strategy parameters are obtained by analysis, and regional anomaly operation and maintenance control is performed. In this way, the technical effects of reducing detection difficulty and improving anomaly perception timeliness and precision are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of operation and maintenance, in particular to a high-precision abnormality perception operation and maintenance method and system based on unmanned aerial vehicle inspection. BACKGROUND

[0002] Regional anomaly detection plays a key role in security protection. In order to meet the efficient, safe and accurate operation and maintenance needs, unmanned aerial vehicle inspection can significantly improve inspection efficiency, reduce operation and maintenance costs, and reduce the safety risks of manual inspection.

[0003] The existing unmanned aerial vehicle inspection is mainly through the high-precision camera, infrared thermal imager, laser radar and other sensors carried on the unmanned aerial vehicle to collect a large amount of image, video and environmental data in real time, and detect and identify the abnormality from these data. In the face of multi-dimensional mass data, there are technical problems of great detection difficulty, low timeliness and influence on perception accuracy. SUMMARY

[0004] The present application provides a high-precision abnormality perception operation and maintenance method and system based on unmanned aerial vehicle inspection, to solve the technical problems of great detection difficulty, low timeliness and influence on perception accuracy in the prior art, and to achieve the technical effects of reducing detection difficulty, improving abnormality perception timeliness and accuracy.

[0005] In a first aspect, the present application provides a high-precision abnormality perception operation and maintenance method based on unmanned aerial vehicle inspection, wherein the method comprises:

[0006] According to the inspection target area, a plurality of inspection targets are set, and feature mining is performed on the plurality of inspection targets in turn to determine a plurality of target inspection index sets.

[0007] The target unmanned aerial vehicle takes off according to the preset inspection plan through a take-off and charging integrated platform, and inspects and monitors the plurality of inspection targets based on the plurality of target inspection index sets, wherein the target unmanned aerial vehicle carries a high-precision sensor group.

[0008] A plurality of target inspection data sets are obtained by inspection using the high-precision sensor group, and the plurality of target inspection data sets are subjected to outlier identification and data preprocessing according to the plurality of target inspection index sets, to generate a plurality of target detection data sets.

[0009] A multi-source heterogeneous abnormality perception algorithm model library is constructed based on the plurality of inspection targets, and the multi-source heterogeneous abnormality perception algorithm model library is called to perform matching analysis and processing on the plurality of target detection data sets respectively, to obtain a plurality of target abnormality perception reports.

[0010] The multiple target anomaly perception reports are transmitted to an operation and maintenance center for strategy analysis and optimization, regional linkage operation and maintenance strategy parameters are obtained, and the target inspection region is controlled based on the regional linkage operation and maintenance strategy parameters.

[0011] In a second aspect, the application further provides a high-precision anomaly perception operation and maintenance system based on unmanned aerial vehicle inspection, wherein the system comprises:

[0012] A target index extraction module is configured to set multiple inspection targets according to the target inspection region, and perform feature mining on the multiple inspection targets in sequence to determine a multiple target inspection index set.

[0013] An inspection plan execution module is configured to make a target unmanned aerial vehicle take off according to a preset inspection plan through a take-off and charging integrated platform, and perform inspection and monitoring on the multiple inspection targets based on the multiple target inspection index set, wherein the target unmanned aerial vehicle is equipped with a high-precision sensor group.

[0014] An inspection data acquisition module is configured to acquire multiple target inspection data sets by using the high-precision sensor group for inspection, perform anomaly value identification and data preprocessing on the multiple target inspection data sets according to the multiple target inspection index set, and generate multiple target detection data sets.

[0015] An anomaly perception module is configured to train and construct a multi-source heterogeneous anomaly perception algorithm model library based on the multiple inspection targets, call the multi-source heterogeneous anomaly perception algorithm model library to perform matching analysis and processing on the multiple target detection data sets respectively, and obtain multiple target anomaly perception reports.

[0016] A decision management module is configured to transmit the multiple target anomaly perception reports to an operation and maintenance center for strategy analysis and optimization, obtain regional linkage operation and maintenance strategy parameters, and control the target inspection region based on the regional linkage operation and maintenance strategy parameters.

[0017] The application discloses a high-precision anomaly perception operation and maintenance method and system based on unmanned aerial vehicle inspection, and relates to the technical field of operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 It is a flowchart of the high-precision anomaly perception operation and maintenance method based on unmanned aerial vehicle inspection.

[0019] Figure 2 It is a structural diagram of the high-precision anomaly perception operation and maintenance system based on unmanned aerial vehicle inspection.

[0020] The reference signs are explained as follows: target index extraction module 11, inspection plan execution module 12, inspection data acquisition module 13, anomaly perception module 14, and decision management module 15. DETAILED DESCRIPTION

[0021] The technical scheme provided in the embodiment of the application is used to solve the technical problems of great detection difficulty, low timeliness and influence on perception accuracy in the prior art, and the overall idea is as follows:

[0022] Firstly, according to the inspection target area, a plurality of inspection targets are set, and feature mining is performed on the inspection targets in turn to determine a plurality of target inspection index sets; then, the target unmanned aerial vehicle takes off according to the preset inspection plan through the take-off and charging integrated platform, and inspects and monitors the inspection target based on the plurality of target inspection index sets, wherein the unmanned aerial vehicle is equipped with a high-precision sensor group; then, a plurality of target inspection data sets are obtained by using the high-precision sensor group for inspection, and a plurality of target detection data sets are generated by performing abnormal value identification and preprocessing on the data according to the target inspection index set; then, based on the inspection target, a multi-source heterogeneous abnormal perception algorithm model library is trained and constructed, and the model library is called to perform matching analysis and processing on the plurality of target detection data sets to obtain a plurality of target abnormal perception reports; then, the reports are transmitted to the operation and maintenance center for strategy analysis and optimization to obtain regional linkage operation and maintenance strategy parameters; finally, based on the regional linkage operation and maintenance strategy parameters, the inspection target area is subjected to abnormal operation and maintenance control.

[0023] The above technical solutions will be described in detail below in combination with the accompanying drawings and specific embodiments, so that the above technical solutions can be better understood. Obviously, the described embodiments are only part of the embodiments of the present application, not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments for explaining the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application. In addition, it should be noted that, for convenience of description, only parts related to the present application are shown in the drawings, not all.

[0024] Embodiment one

[0025] Figure 1 The flowchart of the present application based on the high-precision abnormal perception operation and maintenance method of unmanned aerial vehicle inspection, wherein the method comprises:

[0026] According to the inspection target area, a plurality of inspection targets are set, and feature mining is performed on the plurality of inspection targets in turn to determine a plurality of target inspection index sets.

[0027] Specifically, first, access the inspection target area and locate a plurality of inspection targets, wherein the inspection target refers to a plurality of to-be-detected objects in a specific place or range that need to be monitored and inspected, including nuclear power sources, device sources, fire risk sources, etc., which are distributed at different positions in the inspection area. The setting and selection of the inspection target are based on the abnormal perception target setting of the inspection target area.

[0028] Specifically, the target inspection index set is a set of monitoring and checking parameters generated for each inspection target, used to monitor and evaluate the state of the inspection target in real time, thereby ensuring the safety and stability of its operation. In other words, the target inspection index set includes all index dimensions that can reflect the state of the inspection target, and multiple target inspection index sets correspond to multiple inspection targets, and the target inspection index sets of inspection targets of different types, different attributes, different scales, etc. are not the same. For example, the power consumption, operating voltage, current and temperature of the device source; the temperature, smoke concentration of the fire source, etc.

[0029] In some embodiments, a plurality of target inspection index sets are determined, comprising:

[0030] The characteristics of the plurality of inspection targets are analyzed in sequence to obtain a plurality of target basic characteristic data sets, which include location, model specification, application function and historical maintenance record; the risk impact of the plurality of inspection targets is evaluated based on the plurality of target basic characteristic data sets to obtain a plurality of target risk level information; and the plurality of target feature mining depths are determined according to the plurality of target risk level information.

[0031] Based on the plurality of target feature mining depths, the plurality of target basic characteristic data sets are associated with index mining to obtain the plurality of target inspection index sets.

[0032] Specifically, first, the basic information and operating state of each inspection target are analyzed to identify the geographic or system internal location information (such as GPS coordinates, device number) of each inspection target; the model specification of the device or system is analyzed to obtain its design parameters, such as voltage, capacity, size, etc.; and the core application function of the inspection target is determined. For example, for a nuclear power device, the functions include power supply and stability guarantee; for a fire detector, the function is to monitor fire risk; the maintenance history data of the target is retrieved, including past fault repair, maintenance frequency, etc., thereby forming a basic characteristic data set of the system. The information of the basic characteristic data set helps to judge the health status and fault possibility of the device.

[0033] Specifically, according to the basic characteristic data of the target, the operating state and potential risk are evaluated, and the risk level of the target is determined, which directly affects the priority and depth of the inspection. For example, based on the location, function, and historical maintenance record, it is evaluated whether the device is currently in a risk state (maintenance interval is too long, device model is too old or load is too large), and each inspection target is assigned to different risk levels according to the risk degree. For example, high risk, medium risk and low risk.

[0034] Specifically, the feature mining depth determines the associated range and the number of indicators to be concerned when setting the inspection indicators for each inspection target. High-risk equipment needs to mine more associated features and indicators in depth to ensure that multiple associated indicators can be covered; while low-risk equipment can only focus on core indicators and basic health status.

[0035] Further, according to the determined feature mining depth, the associated indicators are mined from the target basic characteristic data set, and the final inspection indicator set is generated by analyzing the health status, performance, safety compliance, etc. of the equipment. Preferably, the correlation analysis method, principal component analysis and other methods are used to mine the associated indicators from the multiple target basic characteristic data sets, so as to obtain the multiple target inspection indicator sets that meet the feature mining depth and have the highest correlation degree.

[0036] The above method steps determine the appropriate feature mining depth through the characteristic analysis and risk assessment of the inspection target, ensuring the comprehensiveness and pertinence of the inspection indicator set. At the same time, according to different risk levels, the mining depth and the number of inspection indicators are dynamically adjusted, which improves the inspection efficiency while ensuring that high-risk equipment is fully monitored, which helps to ensure that potential problems can be found in time and measures can be taken during the inspection process.

[0037] In some implementations, obtaining the multiple target inspection indicator sets includes:

[0038] Respectively, the multiple target basic characteristic data sets are subjected to associated indicator mining to obtain multiple target associated indicator sets, and the multiple target associated indicator sets are subjected to clustering labeling to obtain multiple target classification indicator cluster sets; the multiple target classification indicator cluster sets are subjected to correlation analysis based on the multiple target basic characteristic data sets to obtain multiple indicator cluster correlation coefficient sets; the multiple target classification indicator cluster sets are subjected to primary indicator screening according to the multiple target feature mining depths and the multiple indicator cluster correlation coefficient sets to obtain multiple target feature indicator cluster sets; each feature indicator in the multiple target feature indicator cluster sets is subjected to correlation analysis and secondary indicator screening based on the multiple target feature mining depths to obtain the multiple target inspection indicator sets.

[0039] Specifically, first, the indicator set related to the target operation, performance, safety, etc. is mined from the target basic characteristic data set, and output as multiple target associated indicator sets. These indicators are important data sources for target monitoring.

[0040] Specifically, each inspection target includes different multiple types of associated indicators, such as physical parameters (e.g., temperature, pressure, vibration), performance parameters (e.g., current, voltage, throughput), environmental parameters (e.g., temperature, humidity, light), and the like. Therefore, the distribution performs intra-set clustering division on the obtained multiple target associated indicator sets to obtain multiple target classification indicator cluster sets, so as to more clearly organize and manage indicators of different categories.

[0041] Further, a suitable statistical method is selected to calculate the correlation coefficient between each indicator cluster, such as Pearson correlation coefficient, Spearman correlation coefficient, and Kendall correlation coefficient, etc. For example, first, the data in each pair of indicator clusters is paired. Then, the correlation coefficient between each pair of indicator clusters is calculated using the selected correlation calculation method, and the calculation result is recorded and stored to form a correlation coefficient set of multiple indicator clusters.

[0042] Specifically, first, the screening range is determined according to the feature mining depth, and different inspection targets set different indicator screening ranges according to their risk levels and feature mining depths. The screening range of a high-risk target is wider, and a low-risk target selects fewer core indicators. Then, for each inspection target, the corresponding indicator clusters are sorted from high to low according to the correlation coefficient, and the indicator clusters with the highest correlation are selected according to the feature mining depth. Each inspection target finally forms a feature indicator cluster set. For example, if the feature mining depth of a high-risk inspection target is 3, the top 3 indicator clusters with the highest correlation coefficient are selected from the serialized target classification indicator cluster set as the first-level indicator screening result.

[0043] Further, according to the feature mining depth, the multiple indicators in the target feature indicator cluster set are respectively screened according to the correlation coefficient. In other words, the indicator item with the highest abnormal correlation is selected from the multiple indicator clusters of the first-level indicator screening result, so as to obtain multiple target inspection indicator sets.

[0044] Through the above method steps, multi-level correlation analysis and screening are performed to generate inspection indicator sets covering the health status, performance, and environmental impact of the target, ensuring accurate and efficient inspection process, fully reflecting the running state of the inspection target while avoiding data redundancy, which helps to reduce the difficulty of inspection collection and data analysis.

[0045] The target unmanned aerial vehicle takes off according to the preset inspection plan through the take-off and charging integrated platform, and inspects and monitors the multiple inspection targets based on the multiple target inspection indicator sets, wherein the target unmanned aerial vehicle is equipped with a high-precision sensor group.

[0046] Specifically, the take-off, landing and charging integrated platform is used for take-off, landing and charging of the unmanned aerial vehicle. The target unmanned aerial vehicle is equipped with a high-precision sensor group for inspection and monitoring. The high-precision sensor group includes a high-precision camera, an infrared sensor, a laser radar, a gas sensor, etc. The specific configuration of the high-precision sensor group is determined according to the monitoring target of the target unmanned aerial vehicle.

[0047] Specifically, the preset inspection plan includes information such as inspection time, inspection route and inspection target. The unmanned aerial vehicle flies according to the preset inspection route and monitors the inspection target according to a plurality of target inspection index sets through the equipped high-precision sensor group.

[0048] Specifically, after completing the inspection task, the target unmanned aerial vehicle returns to the take-off, landing and charging integrated platform according to the preset return route and automatically lands on the take-off, landing and charging integrated platform. The platform charges the unmanned aerial vehicle for the next inspection task.

[0049] A plurality of target inspection data sets are obtained by using the high-precision sensor group for inspection. The plurality of target inspection data sets are subjected to outlier identification and data preprocessing according to the plurality of target inspection index sets to generate a plurality of target detection data sets.

[0050] Specifically, a plurality of target inspection data sets of a plurality of inspection targets are inspected and collected by the high-precision sensor group configured on the target unmanned aerial vehicle. The data types included in the plurality of target inspection data sets correspond to the plurality of target inspection index sets. For example, the target inspection data set can include temperature data, image data, position data, environmental air data, etc.

[0051] Specifically, the data collection frequency of the sensors in the high-precision sensor group is set according to the characteristics and requirements of the inspection target. For high-risk or rapidly changing targets, the collection frequency should be set higher; for stable running targets, the collection frequency can be appropriately reduced.

[0052] In some embodiments, generating a plurality of target detection data sets includes:

[0053] According to the plurality of target inspection index sets, a plurality of target data anomaly identification factors are set, including data consistency, noise data and range rationality. Based on the plurality of target data anomaly identification factors, outlier identification is performed on the plurality of target inspection data sets to obtain a plurality of target anomaly data sets. According to the plurality of target data anomaly identification factors, a plurality of target anomaly data preprocessing programs are obtained by mapping. Based on the plurality of target anomaly data preprocessing programs, data cleaning is performed on the plurality of target anomaly data sets to obtain a plurality of available target data sets. Based on a sensor experience error factor, the plurality of available target data sets are feedback corrected to generate the plurality of target detection data sets.

[0054] Specifically, the data anomaly identification factor is used to identify and determine the abnormal situation that may exist in the inspection data set, including data consistency (whether the data is consistent), noise data (whether the data is affected by noise), and range rationality (whether the data deviates from the normal range).

[0055] Specifically, the detection results of different sensors are compared to determine whether the data is significantly inconsistent. For example, by using statistical indicators such as variance and standard deviation, it is determined whether the data of multiple temperature sensors has significant differences. By setting a noise identification algorithm, extreme values or unreasonable fluctuations that appear in the inspection data set for a short time are identified, and noise data is filtered out. The reasonable value range of each type of data is set, and the range rationality is defined. If a certain data exceeds the reasonable value range, it is determined to be abnormal. Furthermore, the identified abnormal data is sorted and saved as an abnormal data set for subsequent data processing and analysis.

[0056] Further, according to different abnormality identification factors, the corresponding preprocessing program is automatically mapped. For example, the preprocessing of noise data may include outlier rejection and wavelet denoising, while the preprocessing of data consistency includes data smoothing and correction based on interpolation methods. Based on the characteristics of the abnormal data set, suitable data cleaning programs are generated, which helps to ensure that the abnormal data can be restored to a reasonable state or eliminated after processing.

[0057] Specifically, the sensor may produce cumulative errors during long-term use, and then the historical data accumulated experience error factor is used to further correct the data. For example, first, based on the long-term operation history data of the sensor, the deviation law of the sensor is calculated (such as the temperature sensor may measure a higher value in a high temperature environment). Then, the experience error factor is used to adjust each data in the target data set by a certain percentage to achieve feedback correction and further improve the accuracy of the data.

[0058] The above method steps use multiple identification factors (such as consistency, noise, and range rationality) to identify abnormalities, ensuring that abnormal values in the data can be quickly and accurately found, and abnormal data can be specifically eliminated or corrected, ensuring the quality of the cleaned data. Through the feedback correction of the sensor experience error factor, the accuracy of the data is further improved, ensuring that the final generated detection data set can be used for accurate decision-making and analysis.

[0059] Based on the plurality of inspection targets, a multi-source heterogeneous anomaly perception algorithm model library is constructed and trained, and the multi-source heterogeneous anomaly perception algorithm model library is called to respectively match and analyze the plurality of target detection data sets, and a plurality of target anomaly perception reports are obtained.

[0060] In some embodiments, a multi-source heterogeneous anomaly perception algorithm model library is constructed, including:

[0061] Based on the plurality of target inspection index sets, multi-source heterogeneous data collection is performed to obtain a plurality of target multi-source heterogeneous anomaly data sets; anomaly perception feature extraction is performed on the plurality of target multi-source heterogeneous anomaly data sets to obtain a plurality of target anomaly perception data feature sets; feature algorithm selection and anomaly model training are performed on the plurality of target multi-source heterogeneous anomaly data sets according to the plurality of target anomaly perception data feature sets to obtain a plurality of target anomaly recognition model sets; and the plurality of target anomaly recognition model sets are integrated based on the plurality of inspection targets to construct the multi-source heterogeneous anomaly perception algorithm model library.

[0062] Specifically, first, the plurality of target inspection index sets are taken as index targets of sample data, historical inspection data is parsed, and a plurality of target multi-source heterogeneous anomaly data sets are obtained, wherein the historical inspection data includes inspection logs, anomaly lists, and corresponding anomaly handling records of the plurality of inspection targets.

[0063] Specifically, the anomaly perception features are features extracted from the multi-source heterogeneous anomaly data sets and highly related to the abnormal state of the target device or system. By performing anomaly perception feature extraction on the plurality of target multi-source heterogeneous anomaly data sets, the data part contributing to anomaly monitoring in the original sample data is retained, data refinement of the plurality of target multi-source heterogeneous anomaly data sets is achieved, the anomaly perception data feature sets obtained have high data quality, which helps to improve the efficiency of the subsequent training process and avoid overfitting.

[0064] Specifically, according to the feature sets of different inspection targets, a suitable feature extraction algorithm is selected to extract key features that best reflect the abnormal state. For example, a convolutional neural network (CNN) is used to process image data, an LSTM is used to process time series data, and a PointNet is used to process three-dimensional point cloud data. Optionally, before feature extraction, the data is preprocessed, such as standardization and normalization, to ensure that data from different sources can be processed under the same framework.

[0065] Specifically, the anomaly perception feature sets are used to train anomaly recognition models. During the training process, the model learns the anomaly features of different inspection targets to obtain the anomaly recognition ability for different inspection targets. In other words, by learning the mapping relationship between multiple features in the anomaly perception feature set and the corresponding anomaly situation, the ability to judge the anomaly situation is obtained. For example, the model can identify the type, level, and range of the anomaly.

[0066] Further, according to the target types of the plurality of inspection targets, the abnormality recognition models of different inspection targets are integrated to form a unified model library. Each model in the multi-source heterogeneous abnormality perception algorithm model library provides specific abnormality recognition functions according to different inspection targets and data sources. By constructing an abnormality perception algorithm model library that can process multi-source heterogeneous data and adapt to different inspection targets, it is helpful to quickly call the target abnormality recognition model when processing different inspection tasks or inspection targets.

[0067] In some implementations, a plurality of target multi-source heterogeneous abnormality data sets are acquired, including:

[0068] The area inspection abnormality database is mined, first-order association collection is performed in the area inspection abnormality database based on the plurality of target inspection index sets, and a plurality of first-order association index data sets are obtained. Based on the plurality of first-order association index data sets, second-order association collection is performed in the area inspection abnormality database, and a plurality of second-order association index data sets are obtained. Until the N-1 order index data is traversed, where 2≤N≤4, based on the plurality of first-order association index data sets, the plurality of second-order association index data sets, and the N-1 order index data, the plurality of target multi-source heterogeneous abnormality data sets are combined.

[0069] Specifically, first, the historical inspection data in the area is collected and sorted, including normal data and abnormal data, to construct an area inspection abnormality database. Then, based on the target inspection index set, first-order association collection is performed in the area inspection abnormality database, correlation analysis, association rule mining and other technologies are used to identify abnormal data directly related to the target inspection index, and a plurality of first-order association index data sets directly related to the target index are acquired. The first-order association index data set is a collection of abnormal indication indexes that are strongly or indirectly related to the inspection target.

[0070] Specifically, based on the plurality of first-order association index data sets, second-order association collection is performed in the area inspection abnormality database to obtain a plurality of second-order association index data sets related to the first-order association index data sets. The second-order association index data set is a collection of abnormal indication indexes that are weakly or indirectly related to the inspection target.

[0071] Further, the above process is repeated, each time taking the previous first-order correlation indicator dataset as the basis, to perform the next-order correlation collection, gradually expanding the range of the correlation dataset for correlation collection, until N-1 order, combining multiple first-order correlation indicator datasets, multiple second-order correlation indicator datasets, and up to N-1 order indicator datasets to form multiple target multi-source heterogeneous abnormal data sets. Wherein N is greater than or equal to 2 and less than or equal to 4, that is, the correlation collection is performed at least four times and at most twice. By gradually expanding the range of the correlation dataset, the amount of data collection is increased, providing more data support for the training of the model, and thus helping to improve the accuracy and robustness of the model.

[0072] The multiple target abnormal perception reports are transmitted to an operation and maintenance center for strategy analysis and optimization, to obtain regional linkage operation and maintenance strategy parameters, and to perform abnormal operation and maintenance management and control on the target area based on the regional linkage operation and maintenance strategy parameters.

[0073] Specifically, the multiple target abnormal perception reports generated during the inspection process are transmitted to the operation and maintenance center in real time, and each abnormal perception report contains target abnormal feature data such as temperature abnormality, humidity abnormality, image abnormality, etc. Wherein, the operation and maintenance center is used for operation and maintenance management of the target area, and the operation and maintenance center contains all abnormal handling plans and operation and maintenance strategies of multiple inspection targets in the target area.

[0074] In some embodiments, obtaining the regional linkage operation and maintenance strategy parameters comprises:

[0075] Based on the multiple target abnormal perception reports, feature matching is performed with the abnormal operation and maintenance strategy library respectively, to obtain multiple target applicable abnormal operation and maintenance strategies; according to the multiple target applicable abnormal operation and maintenance strategies, multiple abnormal operation and maintenance strategy spaces are constructed, and parameter analysis is performed on the multiple target abnormal perception reports according to the multiple abnormal operation and maintenance strategy spaces, to obtain multiple target operation and maintenance strategy parameter thresholds; parameter global optimization is performed in the multiple target operation and maintenance strategy parameter thresholds according to a preset search step, to obtain multiple target operation and maintenance strategy parameters; and the multiple target operation and maintenance strategy parameters are integrated according to the unmanned aerial vehicle inspection sequence, to obtain the regional linkage operation and maintenance strategy parameters.

[0076] Specifically, the operation and maintenance center is configured with an abnormal operation and maintenance strategy library, which contains operation and maintenance strategies corresponding to various abnormal situations. Further, based on the multiple target abnormal perception reports, feature matching is performed with the abnormal operation and maintenance strategy library respectively, to obtain multiple target applicable abnormal operation and maintenance strategies. For example, based on a rule matching algorithm, the abnormal features in the perception report are matched with the feature templates in the strategy library to identify the most suitable operation and maintenance strategy, such as the top K operation and maintenance strategies with the highest activation frequency, preferably, K is determined based on the risk level of the inspection target.

[0077] Specifically, according to the abnormal operation and maintenance strategy applicable to multiple targets, a strategy space is generated for each target anomaly perception report, multiple abnormal operation and maintenance strategy spaces are generated, wherein the abnormal operation and maintenance strategy space contains all commonly used operation and maintenance strategies and parameter ranges. Then, the parameter range of each operation and maintenance strategy is analyzed, the upper and lower limit values of each parameter are determined, and a parameter threshold set is formed. The parameter threshold set quantitatively represents the application range of multiple commonly used operation and maintenance strategies, which helps to accurately select the appropriate operation and maintenance strategy.

[0078] Further, an optimization algorithm (such as genetic algorithm, particle swarm optimization, etc.) is used to perform global search within the above multiple target operation and maintenance strategy parameter thresholds according to a preset search step length to find the optimal operation and maintenance strategy parameter combination. Preferably, the preset search step length is an adaptive step length,

[0079] Finally, based on the inspection sequence and target position, the operation and maintenance strategy parameters of each target are linked and integrated to form a whole regional linkage operation and maintenance strategy, and the output is a regional linkage operation and maintenance strategy parameter. The regional linkage operation and maintenance strategy parameter contains the execution sequence, path and multiple operation and maintenance strategies of the operation and maintenance task. Based on the linkage integration of the inspection sequence and the target position, the execution of the operation and maintenance task is carried out on the basis of the reliable inspection route, which further ensures the reliability of the operation and maintenance task.

[0080] The above method steps obtain the optimal multiple operation and maintenance strategy parameters through strategy analysis and global optimization, improve the operation and maintenance efficiency. Based on the feature matching of the abnormal perception report and the operation and maintenance strategy library, the accuracy of the operation and maintenance strategy is ensured. Through linkage integration, the scheduling of unmanned aerial vehicles and other operation and maintenance equipment is optimized, the resource utilization rate is improved, and the reliability of the inspection task is ensured.

[0081] In summary, the high-precision abnormal perception operation and maintenance method based on unmanned aerial vehicle inspection provided by the present application has the following technical effects:

[0082] By setting multiple inspection targets according to the inspection target area, and performing feature mining on the inspection targets in turn to determine multiple target inspection indicator sets; the target unmanned aerial vehicle takes off according to the preset inspection plan through the take-off and charging integrated platform, and inspects and monitors the inspection targets based on the multiple target inspection indicator sets, wherein the target unmanned aerial vehicle is equipped with a high-precision sensor group; the high-precision sensor group is used to collect inspection data and obtain multiple target inspection data sets, and the data sets are subjected to outlier identification and data preprocessing according to the inspection indicator sets to generate multiple target detection data sets; a multi-source heterogeneous anomaly perception algorithm model library is constructed based on the inspection targets, the model library is called to perform matching analysis and processing on the multiple target detection data sets to generate multiple target anomaly perception reports; the multiple target anomaly perception reports are transmitted to the operation and maintenance center for strategy analysis and optimization, regional linkage operation and maintenance strategy parameters are obtained, and the inspection target area is subjected to abnormal operation and maintenance control according to the parameters. Thus, the technical effects of reducing detection difficulty and improving anomaly perception timeliness and accuracy are achieved.

[0083] Embodiment two

[0084] Figure 2 is a structural schematic diagram of the high-precision anomaly perception operation and maintenance system based on unmanned aerial vehicle inspection of the present application. For example, Figure 1 The flowchart of the high-precision anomaly perception operation and maintenance method based on unmanned aerial vehicle inspection of the present application can be implemented by the structure as shown in Figure 2

[0085] Based on the same idea as the high-precision anomaly perception operation and maintenance method based on unmanned aerial vehicle inspection in the embodiments, the present application also provides a high-precision anomaly perception operation and maintenance system based on unmanned aerial vehicle inspection, which comprises:

[0086] A target indicator extraction module 11 is configured to set multiple inspection targets according to the inspection target area, and perform feature mining on the multiple inspection targets in turn to determine multiple target inspection indicator sets.

[0087] An inspection plan execution module 12 is configured to make the target unmanned aerial vehicle take off according to the preset inspection plan through the take-off and charging integrated platform, and inspect and monitor the multiple inspection targets based on the multiple target inspection indicator sets, wherein the target unmanned aerial vehicle is equipped with a high-precision sensor group.

[0088] An inspection data collection module 13 is configured to use the high-precision sensor group to collect multiple target inspection data sets, perform outlier identification and data preprocessing on the multiple target inspection data sets according to the multiple target inspection indicator sets, and generate multiple target detection data sets.

[0089] ​An anomaly perception module 14 is configured to train a multi-source heterogeneous anomaly perception algorithm model library based on the plurality of inspection targets, call the multi-source heterogeneous anomaly perception algorithm model library to perform matching analysis and processing on the plurality of target detection data sets respectively, and obtain a plurality of target anomaly perception reports.

[0090] A decision management module 15 is configured to transmit the plurality of target anomaly perception reports to an operation and maintenance center to perform strategy analysis and optimization, obtain regional linkage operation and maintenance strategy parameters, and perform anomaly operation and maintenance management and control on the inspection target region based on the regional linkage operation and maintenance strategy parameters.

[0091] The target index extraction module 11 includes:

[0092] A characteristic analysis and basic characteristic data set acquisition unit is configured to perform characteristic analysis on the plurality of inspection targets in sequence, and acquire a plurality of target basic characteristic data sets, wherein the plurality of target basic characteristic data sets include position, model specification, application function, and historical maintenance record.

[0093] A risk impact assessment and risk level information determination unit is configured to perform risk impact assessment on the plurality of inspection targets based on the plurality of target basic characteristic data sets, and obtain a plurality of target risk level information.

[0094] A feature mining depth determination unit is configured to determine a plurality of target feature mining depths according to the plurality of target risk level information.

[0095] An associated index mining and inspection index set generation unit is configured to perform associated index mining on the plurality of target basic characteristic data sets based on the plurality of target feature mining depths respectively, and obtain the plurality of target inspection index sets.

[0096] In some implementations, the associated index mining and inspection index set generation unit in the target index extraction module 11 includes:

[0097] An associated index mining and clustering labeling unit is configured to perform associated index mining on the plurality of target basic characteristic data sets respectively, obtain a plurality of target associated index sets, and perform clustering labeling on the plurality of target associated index sets, to obtain a plurality of target classification index cluster sets.

[0098] A classification index cluster correlation analysis unit is configured to perform correlation analysis on the plurality of target classification index cluster sets based on the plurality of target basic characteristic data sets, and obtain a plurality of index cluster correlation coefficient sets.

[0099] A primary index screening unit is configured to perform primary index screening on the plurality of target classification index cluster sets according to the plurality of target feature mining depths and the plurality of index cluster correlation coefficient sets, and obtain a plurality of target feature index cluster sets.

[0100] The characteristic index correlation analysis and secondary index screening unit is configured to perform correlation analysis and secondary index screening on each characteristic index in the target characteristic index cluster set based on the plurality of target characteristics to obtain the plurality of target inspection indexes.

[0101] In some embodiments, the inspection data acquisition module 13 includes:

[0102] The abnormality identification factor setting and target data abnormality identification unit is configured to set a plurality of target data abnormality identification factors according to the plurality of target inspection indexes, wherein the plurality of target data abnormality identification factors include data consistency, noise data, and range rationality.

[0103] The abnormal value identification and abnormal data set acquisition unit is configured to identify abnormal values in the plurality of target inspection data sets based on the plurality of target data abnormality identification factors to obtain a plurality of target abnormal data sets.

[0104] The abnormal data preprocessing program mapping and data cleaning unit is configured to map a plurality of target abnormal data preprocessing programs according to the plurality of target data abnormality identification factors, clean the plurality of target abnormal data sets based on the plurality of target abnormal data preprocessing programs, and obtain a plurality of available target data sets.

[0105] The sensor experience error correction and target detection data set generation unit is configured to correct the plurality of available target data sets based on a sensor experience error factor to generate a plurality of target detection data sets.

[0106] In some embodiments, the anomaly perception module 14 includes:

[0107] The multi-source heterogeneous data acquisition and abnormal data set acquisition unit is configured to acquire a plurality of target multi-source heterogeneous abnormal data sets based on the plurality of target inspection indexes.

[0108] The anomaly perception feature extraction unit is configured to extract anomaly perception features from the plurality of target multi-source heterogeneous abnormal data sets to obtain a plurality of target anomaly perception data feature sets.

[0109] The feature algorithm selection and abnormal model training unit is configured to select feature algorithms and train abnormal models for the plurality of target multi-source heterogeneous abnormal data sets according to the plurality of target anomaly perception data feature sets to obtain a plurality of target abnormal identification model sets.

[0110] The abnormal identification model integration unit is configured to integrate the plurality of target abnormal identification model sets based on the plurality of inspection targets to construct the multi-source heterogeneous anomaly perception algorithm model library.

[0111] In some implementations, the multi-source heterogeneous data collection and abnormal data set acquisition unit in the anomaly perception module 1 comprises:

[0112] An anomaly database mining and correlation collection unit is configured to mine an area inspection anomaly database, perform first-order correlation collection in the area inspection anomaly database based on the plurality of target inspection index sets, and obtain a plurality of first-order correlation index data sets.

[0113] A second-order correlation collection unit is configured to perform second-order correlation collection in the area inspection anomaly database based on the plurality of first-order correlation index data sets, and obtain a plurality of second-order correlation index data sets.

[0114] A multi-order correlation collection unit is configured to perform correlation collection up to N-1 order index data, where 2≤N≤4, based on the plurality of first-order correlation index data sets, the plurality of second-order correlation index data sets, and up to the N-1 order index data, to combine the plurality of target multi-source heterogeneous anomaly data sets.

[0115] In some embodiments, the decision management module 15 comprises:

[0116] An anomaly perception report matching and operation and maintenance strategy acquisition unit is configured to perform feature matching between the plurality of target anomaly perception reports and an anomaly operation and maintenance strategy library respectively, and obtain a plurality of target applicable anomaly operation and maintenance strategies.

[0117] An operation and maintenance strategy construction and parameter analysis unit is configured to construct a plurality of anomaly operation and maintenance strategy spaces according to the plurality of target applicable anomaly operation and maintenance strategies, and perform parameter analysis on the plurality of target anomaly perception reports according to the plurality of anomaly operation and maintenance strategy spaces, to obtain a plurality of target operation and maintenance strategy parameter thresholds.

[0118] A parameter global optimization unit is configured to perform parameter global optimization in the plurality of target operation and maintenance strategy parameter thresholds respectively according to a preset search step, to obtain a plurality of target operation and maintenance strategy parameters.

[0119] A linkage integration and regional linkage operation and maintenance strategy generation unit is configured to perform linkage integration on the plurality of target operation and maintenance strategy parameters according to a UAV inspection sequence, to obtain the regional linkage operation and maintenance strategy parameters.

[0120] It should be understood that the embodiments mentioned in the specification focus on their differences from other embodiments, and the specific embodiments in the first embodiment are also applicable to the high-precision anomaly perception operation and maintenance system based on UAV inspection described in the second embodiment. For the sake of brevity of the specification, no further expansion is made here.

[0121] It should be understood that the embodiments disclosed herein and the foregoing description thereof are merely exemplary in nature and, thus, that various changes in the details thereof can be implemented by those skilled in the art without departing from the spirit and scope of the present application. Such changes are intended to fall within the scope of the present application as defined by the appended claims.

Claims

1. A high-precision anomaly perception operation and maintenance method based on unmanned aerial vehicle inspection, characterized in that, The method comprises: According to the inspection target area, a plurality of inspection targets are set, and feature mining is performed on the plurality of inspection targets in turn to determine a plurality of target inspection indicator sets; The target unmanned aerial vehicle takes off according to a preset inspection plan through a take-off and charging integrated platform, and inspects and monitors the plurality of inspection targets based on the plurality of target inspection indicator sets, wherein the target unmanned aerial vehicle carries a high-precision sensor group; A plurality of target inspection data sets are obtained by using the high-precision sensor group for inspection, anomaly value identification and data preprocessing are performed on the plurality of target inspection data sets according to the plurality of target inspection indicator sets, and a plurality of target detection data sets are generated; A multi-source heterogeneous anomaly perception algorithm model library is constructed based on the plurality of inspection targets, the multi-source heterogeneous anomaly perception algorithm model library is called to perform matching analysis and processing on the plurality of target detection data sets respectively, and a plurality of target anomaly perception reports are obtained; The plurality of target anomaly perception reports are transmitted to an operation and maintenance center for strategy analysis and optimization, regional linkage operation and maintenance strategy parameters are obtained, and the inspection target area is controlled based on the regional linkage operation and maintenance strategy parameters. The determination of the plurality of target inspection indicator sets comprises: Characteristic analysis is performed on the plurality of inspection targets in turn to obtain a plurality of target basic characteristic data sets, the plurality of target basic characteristic data sets include position, model specification, application function and historical maintenance record; Risk impact assessment is performed on the plurality of inspection targets based on the plurality of target basic characteristic data sets to obtain a plurality of target risk level information; According to the plurality of target risk level information, a plurality of target feature mining depths are determined; Based on the plurality of target feature mining depths, associated indicator mining is performed on the plurality of target basic characteristic data sets respectively to obtain the plurality of target inspection indicator sets. 2.The high-precision anomaly perception operation and maintenance method based on UAV inspection of claim 1, wherein, The plurality of target inspection indicator sets are obtained, comprising: Associated indicator mining is performed on the plurality of target basic characteristic data sets respectively to obtain a plurality of target associated indicator sets, and clustering labeling is performed on the plurality of target associated indicator sets to obtain a plurality of target classification indicator cluster sets; Correlation analysis is performed on the plurality of target classification indicator cluster sets based on the plurality of target basic characteristic data sets to obtain a plurality of indicator cluster correlation coefficient sets; According to the plurality of target feature mining depths and the plurality of indicator cluster correlation coefficient sets, first-level indicator screening is performed on the plurality of target classification indicator cluster sets to obtain a plurality of target feature indicator cluster sets; Based on the plurality of target feature mining depths, correlation analysis and second-level indicator screening are performed on each feature indicator in the plurality of target feature indicator cluster sets to obtain the plurality of target inspection indicator sets. 3.The high-precision anomaly perception operation and maintenance method based on UAV inspection of claim 1, wherein, The plurality of target detection data sets are generated, comprising: According to the plurality of target inspection indicator sets, a plurality of target data anomaly identification factors are set, the plurality of target data anomaly identification factors include data consistency, noise data and range rationality; Based on the plurality of target data anomaly identification factors, anomaly value identification is performed on the plurality of target inspection data sets to obtain a plurality of target anomaly data sets; According to the plurality of target data anomaly identification factors, a plurality of target anomaly data preprocessing programs are obtained by mapping, and the plurality of target anomaly data sets are cleaned by data based on the plurality of target anomaly data preprocessing programs, to obtain a plurality of available target data sets; Based on the sensor experience error factor, the plurality of available target data sets are feedback corrected to generate the plurality of target detection data sets. 4.The high-precision anomaly perception operation and maintenance method based on UAV inspection of claim 1, wherein, The construction of the multi-source heterogeneous anomaly perception algorithm model library includes: Based on the plurality of target inspection index sets, a plurality of target multi-source heterogeneous anomaly data sets are acquired by multi-source heterogeneous data collection; Anomaly perception feature extraction is performed on the plurality of target multi-source heterogeneous anomaly data sets to obtain a plurality of target anomaly perception data feature sets; According to the plurality of target anomaly perception data feature sets, feature algorithm selection and anomaly model training are performed on the plurality of target multi-source heterogeneous anomaly data sets to obtain a plurality of target anomaly identification model sets; Based on the plurality of inspection targets, the plurality of target anomaly identification model sets are integrated to construct the multi-source heterogeneous anomaly perception algorithm model library. 5.The high-precision anomaly perception operation and maintenance method based on UAV inspection according to claim 4, wherein, The acquisition of the plurality of target multi-source heterogeneous anomaly data sets includes: A regional inspection anomaly database is mined and acquired, and based on the plurality of target inspection index sets, first-order association collection is performed in the regional inspection anomaly database to obtain a plurality of first-order association index data sets; Based on the plurality of first-order association index data sets, second-order association collection is performed in the regional inspection anomaly database to obtain a plurality of second-order association index data sets; Until the N-1 order index data is traversed and collected, wherein 2≤N≤4, based on the plurality of first-order association index data sets, the plurality of second-order association index data sets, and the N-1 order index data, the plurality of target multi-source heterogeneous anomaly data sets are combined. 6.The high-precision anomaly perception operation and maintenance method based on UAV inspection according to claim 1, wherein, The acquisition of the regional linkage operation and maintenance strategy parameter includes: Based on the plurality of target anomaly perception reports, feature matching is performed with the anomaly operation and maintenance strategy library respectively to obtain a plurality of target applicable anomaly operation and maintenance strategies; According to the plurality of target applicable anomaly operation and maintenance strategies, a plurality of anomaly operation and maintenance strategy spaces are constructed, and parameter analysis is performed on the plurality of target anomaly perception reports according to the plurality of anomaly operation and maintenance strategy spaces to obtain a plurality of target operation and maintenance strategy parameter thresholds; According to a preset search step, parameter global optimization is performed in the plurality of target operation and maintenance strategy parameter thresholds respectively to obtain a plurality of target operation and maintenance strategy parameters; The plurality of target operation and maintenance strategy parameters are integrated according to the unmanned aerial vehicle inspection sequence to obtain the regional linkage operation and maintenance strategy parameter.

7. A high-precision anomaly perception operation and maintenance system based on unmanned aerial vehicle inspection, characterized in that, The system is used to execute the high-precision anomaly perception operation and maintenance method based on unmanned aerial vehicle inspection according to any one of claims 1-6, and the system includes: A target index extraction module, the target index extraction module is used to set a plurality of inspection targets according to an inspection target region, and perform feature mining on the plurality of inspection targets in turn to determine a plurality of target inspection index sets; The inspection plan execution module is configured to make the target UAV take off according to a preset inspection plan through the take-off and landing charging integrated platform, and perform inspection and monitoring on the plurality of inspection targets based on the plurality of target inspection index sets, wherein the target UAV is equipped with a high-precision sensor group; The inspection data acquisition module is configured to acquire a plurality of target inspection data sets by performing inspection using the high-precision sensor group, perform outlier identification and data preprocessing on the plurality of target inspection data sets according to the plurality of target inspection index sets, and generate a plurality of target detection data sets; The anomaly perception module is configured to train and construct a multi-source heterogeneous anomaly perception algorithm model library based on the plurality of inspection targets, call the multi-source heterogeneous anomaly perception algorithm model library to perform matching analysis and processing on the plurality of target detection data sets respectively, and obtain a plurality of target anomaly perception reports; The decision management module is configured to transmit the plurality of target anomaly perception reports to an operation and maintenance center for strategy analysis and optimization, obtain regional linkage operation and maintenance strategy parameters, and perform abnormal operation and maintenance control on the inspection target region based on the regional linkage operation and maintenance strategy parameters. The target index extraction module includes: The characteristic analysis and basic characteristic data set acquisition unit is configured to perform characteristic analysis on the plurality of inspection targets in sequence, and acquire a plurality of target basic characteristic data sets, wherein the plurality of target basic characteristic data sets include position, model specification, application function, and historical maintenance record; The risk impact assessment and risk level information determination unit is configured to perform risk impact assessment on the plurality of inspection targets based on the plurality of target basic characteristic data sets, and obtain a plurality of target risk level information; The feature mining depth determination unit is configured to determine a plurality of target feature mining depths according to the plurality of target risk level information; The correlation index mining and inspection index set generation unit is configured to perform correlation index mining on the plurality of target basic characteristic data sets based on the plurality of target feature mining depths, and obtain the plurality of target inspection index sets.

Images