Method and system for inspecting defects of buried water pressure steel pipe

Abstract

The present invention provides a method and system for inspecting defects of buried water pressure steel pipes, which include the following steps: performing time series alignment and quality assessment on multi-source sensor data such as ultrasound, video and position, extracting and optimizing features through adaptive filtering and frequency smoothing technology, and adopting block compression storage based on entropy value and variability; decompressing and reconstructing features of compressed data, predicting state evolution based on multi-scale state trajectory analysis, and performing abnormality detection and classification in combination with dynamic benchmark models; constructing health feature vectors, obtaining health state distribution through nonlinear feature fusion and probability density analysis, predicting degradation trends based on environmental coupling analysis, adopting multi-level dynamic thresholds for risk rating and early warning, and finally generating maintenance priority sorting and maintenance plans. The present invention can accurately identify pipeline defects and improve detection efficiency and reliability.

Description

Technical Field

[0001] The invention relates to an inspection method, in particular to a buried water supply pressure steel pipe disease inspection method and system. Background Art

[0002] The buried water pressure steel pipe is an important infrastructure, which undertakes the core functions of urban water supply and water conservancy projects. Its safe operation is directly related to regional economic development and people's livelihood security. Due to long-term underground operation, this type of pipeline faces multiple threats such as soil corrosion, stress concentration, and material aging. Once a rupture or leakage accident occurs, the consequences are serious. Therefore, conducting disease inspection research on buried water pressure steel pipes is of great significance for timely discovering potential hidden dangers, preventing accidents, and extending the service life of pipelines.

[0003] At present, the detection of buried water pressure steel pipes mainly relies on single detection methods such as ultrasound, electromagnetics, and video, and the processing and analysis methods of detection data are relatively simple. Traditional data processing methods mostly use fixed threshold judgment and empirical rule analysis, and lack the ability to collaboratively analyze multi-source data. In terms of status assessment, it mainly relies on statistical models and deterministic analysis methods, and insufficient consideration is given to the dynamic characteristics of the pipeline degradation process and the impact of environmental factors. Early warning decisions mostly use static thresholds and fixed maintenance cycles, which are difficult to adapt to the dynamic changes in pipeline service conditions. At the same time, existing detection systems generally have problems such as low data storage efficiency and large computing resource usage.

[0004] At the specific technical level, the current detection technology has the following problems: the time series alignment accuracy of multi-source sensor data is insufficient, making it difficult to accurately reflect the instantaneous changes in pipeline status; data quality assessment lacks an adaptive mechanism and cannot effectively identify and filter unreliable data; frequency modulation features are not fully considered in the feature extraction process, resulting in the neglect of some weak disease features; state trajectory analysis fails to effectively utilize multi-scale information, affecting the ability to identify gradual diseases; health status assessment does not consider the memory effect of historical data, reducing the reliability of the assessment results; degradation trend analysis fails to accurately quantify the coupling effect of environmental factors, affecting prediction accuracy; warning threshold updates lack a dynamic optimization mechanism, which is prone to false alarms and missed alarms; maintenance decisions fail to fully consider resource constraints and skill matching issues, affecting maintenance efficiency. In addition, the existing system also has deficiencies in the hierarchical storage and rapid retrieval of feature data, affecting the real-time performance of the detection system. Summary of the invention

[0005] The purpose of the invention is to provide a method and system for inspecting defects of buried water supply pressure steel pipes, in the hope of solving at least one technical problem existing in the prior art.

[0006] Technical solution, a method for inspecting buried water supply pressure steel pipe defects, characterized in that it comprises the following steps:

[0007] Step S1: According to the original data stream collected by each sensor, preprocessing including time alignment, quality scoring, feature extraction and feature optimization is performed in sequence to obtain feature data after dimension reduction; the feature data after dimension reduction is compressed and stored in blocks according to quality level and data type to obtain compressed data blocks;

[0008] Step S2: read and decompress the compressed data block and reconstruct the features to obtain an initial feature vector; perform enhancement processing, correlation analysis and state trajectory analysis on the initial feature vector to obtain a state evolution sequence and perform anomaly detection and classification to obtain anomaly scoring and pattern classification results; then perform spatiotemporal feature fusion and optimized storage to form a pipeline state feature data packet;

[0009] Step S3, obtain and construct an initial health feature vector based on the pipeline status feature data packet, perform nonlinear feature fusion and probability density analysis on the initial health feature vector, and obtain health status distribution data; based on the health status distribution data, perform degradation trend analysis and acceleration factor calculation to obtain degradation development prediction results; perform risk level assessment and early warning strategy generation based on the degradation development prediction results, and output risk rating data and early warning instructions; based on the risk rating data, generate a maintenance priority sequence and a specific maintenance plan.

[0010] Beneficial effect: a closed-loop technical system has been formed from data collection, processing, analysis to decision-making, which has significantly improved the efficiency and reliability of pipeline inspection and maintenance. This solution is particularly suitable for large-scale water pipeline systems with high safety requirements and complex operating environments. It can effectively prevent pipeline accidents, extend the service life of pipelines, and achieve optimal management of pipeline assets. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a flow chart of the present invention.

[0012] Figure 2 It is a flow chart of step S1 of the present invention.

[0013] Figure 3 It is a flow chart of step S2 of the present invention.

[0014] Figure 4 It is a flow chart of step S3 of the present invention. DETAILED DESCRIPTION

[0015] Embodiment 1, as Figure 1 As shown, according to one aspect of the present application, a method for inspecting defects of a buried water supply penstock is provided, comprising the following steps:

[0016] Step S1: Perform time alignment and data preprocessing based on the original data streams collected by various sensors, including ultrasonic data streams, video data streams, wall thickness measurement data streams, and position data streams. Perform quality scores on the aligned data streams to obtain data streams with quality scores. Perform feature extraction and optimization on the data streams with quality scores to obtain feature data after dimensionality reduction. Compress and store the feature data in blocks according to quality level and data type to complete the data preprocessing process.

[0017] Step S2: Decompress the compressed data block output by S1 and reconstruct the features to obtain the initial feature vector. Perform enhancement processing and correlation analysis on the initial feature vector to obtain an enhanced feature group. Perform state trajectory analysis based on the enhanced feature group to obtain the state evolution sequence and prediction data. Perform anomaly detection and classification on the state evolution sequence to obtain anomaly scores and pattern classification results. Finally, perform spatiotemporal feature fusion and optimized storage to output complete pipeline state feature data.

[0018] Step S3, obtain the pipeline status feature data packet from S2, decompress the data packet, and extract the enhanced feature data, state evolution sequence, anomaly score and pattern classification results. Construct an initial health feature vector based on the extracted data. Perform nonlinear feature fusion and probability density analysis on the initial health feature vector to obtain health status distribution data. Based on the health status distribution data, perform degradation trend analysis and acceleration factor calculation to obtain degradation development prediction results. Perform risk level assessment and early warning strategy generation based on the degradation prediction results, and output risk rating data and early warning instructions. Finally, based on the risk rating data, generate maintenance priority sorting and specific maintenance plans.

[0019] In this embodiment, a complete buried water pressure steel pipe disease inspection system is constructed through innovative multi-source data fusion, state assessment and early warning decision-making technology. First, at the data processing level, the synchronization and reliability problems of data acquisition are solved through the time alignment and quality assessment of multi-source sensor data; secondly, at the state analysis level, accurate assessment of pipeline status is achieved through multi-scale feature extraction and state trajectory analysis; finally, at the decision-making level, early warning of pipeline diseases is achieved through health state distribution analysis and degradation trend prediction. The entire solution forms a closed-loop detection-assessment-early warning system, which is particularly suitable for buried pressure steel pipes, which have complex operating environments and high safety requirements. It can effectively prevent pipeline accidents, extend pipeline service life, and reduce maintenance costs.

[0020] like Figure 2 As shown, according to one aspect of the present application, step S1 is specifically:

[0021] Step S11, obtain the original data streams collected by each sensor, including ultrasonic data stream, video data stream, wall thickness measurement data stream and position data stream. Generate a master clock sequence and record the timestamp information of each data stream respectively. Calculate the offset between each data stream and the master clock sequence and generate time compensation parameters. Determine the weight coefficient according to the sensor type and use the weight coefficient to correct the time compensation parameter. Perform time alignment processing on each data stream based on the corrected compensation parameters and output a multi-source data stream with time alignment.

[0022] Step S12, receiving the timing alignment data stream output by step S11. Calculate the signal-to-noise ratio, data integrity ratio and stability ratio of each data stream respectively. Perform weighted calculation on the three ratios according to the preset weight coefficient to obtain the initial quality score. Calculate the degree of fluctuation of the quality score and generate a fluctuation penalty factor. Use the fluctuation penalty factor to correct the initial quality score to obtain the final quality score of each data stream. According to the final quality score, generate a credibility score for the data stream, and attach the credibility score to the corresponding data stream. Obtain the preset data prediction model parameters and use the prediction model to predict the data stream. Calculate the difference between the actual value and the predicted value to generate a residual sequence. Calculate the rate of change and local volatility of the residual sequence. Generate an anomaly score based on the rate of change and volatility, and attach the anomaly score to the corresponding data stream. Output a data stream with a quality score and anomaly score.

[0023] Step S13, receiving the data stream with quality score and anomaly score outputted from step S12. Extract the frequency components of the data stream and generate time-frequency spectrum data. Calculate the frequency gradient of the time-frequency spectrum data and generate frequency smoothing parameters. Construct adaptive filtering parameters according to the time-frequency spectrum amplitude and frequency smoothing parameters. Filter the data using the adaptive filtering parameters to obtain a denoised data stream. Calculate the information entropy, time-varying index and correlation between features of each feature component in the data stream. Generate a feature importance score according to the information entropy and time-varying index. Determine the feature weight coefficient based on the importance score and perform dimensionality reduction on the feature space. Output the reduced-dimensional feature data stream.

[0024] Step S14, receiving the reduced feature data stream output by step S13. Layer the feature data stream according to the data type to obtain multiple layers of feature data. Calculate the quality score and information score of each layer of feature data, and generate an inter-layer weight coefficient. Use the inter-layer weight coefficient to fuse the multiple layers of feature data to obtain a fused feature data stream. Calculate the entropy value and variability of the feature data, and divide the data into multiple data blocks according to the entropy value and variability. Calculate the quality score of each data block, and determine the compression ratio according to the quality score. Use the determined compression ratio to compress and encode the data block to obtain compressed and stored block data.

[0025] In this embodiment, the detection data of buried water pressure steel pipe is efficiently processed through multi-source data fusion and preprocessing technology. Specifically, by adopting the time alignment algorithm, the time synchronization problem of multi-source sensor data such as ultrasonic, video, wall thickness measurement and position in the acquisition process is solved. The three-dimensional quality evaluation system based on signal-to-noise ratio, data integrity and stability is innovatively introduced, and the quality score is dynamically corrected by the fluctuation penalty factor to ensure the accurate evaluation of data reliability. In the feature extraction link, adaptive filtering and frequency smoothing technology are used to effectively reduce the impact of data noise and improve the accuracy of feature extraction. By introducing a block compression strategy based on entropy and variability, the integrity of the data is guaranteed and the storage space is optimized. This multi-level data preprocessing scheme lays a reliable data foundation for subsequent pipeline disease analysis, significantly improves the data quality and processing efficiency of the entire detection system, and is particularly suitable for detection scenarios in complex environments such as buried pressure steel pipes.

[0026] like Figure 3 As shown, according to one aspect of the present application, step S2 is specifically:

[0027] Step S21, receive the compressed data block output by S1, and decompress it to obtain the initial feature vector. Calculate the time gradient data of the feature vector and obtain the historical feature data. Generate a feature enhancement coefficient based on the time gradient data and the historical feature data. Use the feature enhancement coefficient to enhance the initial feature vector to obtain an enhanced feature vector. Calculate the covariance data between each feature component in the enhanced feature vector. Obtain the rate of change data of each feature component and calculate the dynamic difference value between the features. Generate a feature association strength matrix based on the covariance data and the dynamic difference value. Set a multi-layer association threshold, group the features according to the association strength matrix, and output a hierarchical feature group.

[0028] Step S22, receiving the hierarchical feature group output by step S21. Set multiple observation time scales and extract feature sequences at each time scale. Calculate the variability of the feature sequence at each time scale and generate a smoothing factor. Construct a multi-scale state trajectory based on the feature sequence and the smoothing factor. Perform trend analysis on the state trajectory of each time scale to obtain the prediction data at each scale. Calculate the historical error of the prediction at each scale and generate a prediction weight coefficient. Fusion the prediction data of each scale according to the prediction weight coefficient to obtain the final state evolution sequence and prediction data.

[0029] Step S23, receiving the state evolution sequence outputted in step S22. Extracting historical state data, calculating the state smoothing value, and generating a dynamic benchmark model. Calculating the deviation value between the current state and the benchmark model, and obtaining the state change rate data. Calculating the state deviation score based on the deviation value and the change rate data. Obtaining the pre-stored abnormal pattern data, calculating the matching probability between the current state and each abnormal pattern. Extracting the entropy value feature of the state sequence, and generating an uncertainty index. Calculating the final probability score of each abnormal pattern based on the matching probability and the uncertainty index, and outputting the abnormal score and pattern classification result.

[0030] Step S24, receiving the hierarchical feature group output by step S21, the state evolution sequence output by step S22, and the anomaly score and pattern classification result output by step S23. Acquire spatial position data and calculate local geometric features. Generate spatial kernel function parameters based on spatial position data and local geometric features. Perform spatiotemporal fusion processing on the hierarchical feature group using the spatial kernel function parameters to obtain spatiotemporal coupling features. Calculate the local response data of the features and extract the second-order gradient information. Enhance the spatiotemporal coupling features based on the local response data and gradient information to obtain enhanced feature data. Combine the enhanced feature data, state evolution sequence, anomaly score and pattern classification results to form a complete pipeline state feature data packet. Calculate the quality score, information content and variance value of each component in the data packet to generate a storage priority score. Perform hierarchical compression and storage on the data packet according to the storage priority score, and output the final pipeline state feature data packet.

[0031] In this embodiment, innovative feature reconstruction and state analysis techniques are used to achieve accurate assessment of pipeline status. By designing a multi-level feature enhancement algorithm, combining time gradient data and historical feature data, an adaptive feature enhancement coefficient is constructed, which significantly improves the expressiveness of the feature. The innovative introduction of a multi-scale state trajectory analysis method achieves accurate capture of pipeline state evolution by analyzing the variability of feature sequences at different time scales. An anomaly detection strategy based on a dynamic benchmark model is adopted, combined with state deviation scoring and abnormal pattern matching probability, to improve the accuracy and reliability of anomaly detection. By introducing the spatiotemporal feature fusion technology of the spatial kernel function, the correlation problem of pipeline detection data in the spatial dimension is effectively solved, providing a more comprehensive feature expression for pipeline state assessment. This multi-dimensional state analysis method is particularly suitable for detection objects such as buried pressure steel pipes with complex state evolution characteristics, and can accurately identify and predict potential pipeline diseases.

[0032] like Figure 4 According to one aspect of the present application, step S3 is specifically:

[0033] Step S31, extract enhanced feature data from the pipeline status feature data packet, obtain historical health response data, calculate the cumulative value of historical response data, and generate a memory strength parameter. Perform nonlinear mapping on the enhanced feature data according to the memory strength parameter to obtain an initial health index. Extract the information gain value, stability score and abnormal score of the feature data output by S2 to generate a feature weight coefficient. Use the feature weight coefficient to optimize the initial health index to obtain a multidimensional health index. Obtain the pre-stored basic distribution function library and extract the fluctuation characteristics of the health status data. Adjust the parameters of the basic distribution function according to the fluctuation characteristics to generate an initial probability density distribution. Calculate the uncertainty score of the health status, correct the probability density distribution, and output the final health status distribution data and the comprehensive health index.

[0034] Step S32, receiving the health status distribution data and comprehensive health index outputted in step S31. Obtain the pre-stored basic degradation pattern library and extract the change rate data of the health index. Assign weights to the basic degradation patterns according to the change rate data to obtain the initial degradation pattern. Obtain historical degradation data and calculate the time kernel function parameters. Perform evolution analysis on the initial degradation pattern using the time kernel function parameters and output the degradation trend prediction results. Obtain environmental monitoring data, including temperature, humidity, pressure and other parameters. Calculate the coupling strength value between environmental parameters and generate the environmental coupling coefficient. Construct an acceleration factor model based on the environmental coupling coefficient to obtain the initial acceleration factor data. Calculate the degree of fluctuation of the acceleration factor and generate the fluctuation correction parameter. Use the fluctuation correction parameter to correct the initial acceleration factor to obtain the final acceleration effect data. Combine the acceleration effect data with the degradation trend prediction results and output the corrected degradation development prediction results.

[0035] Step S33, receiving the health status distribution data output by step S31 and the degradation development prediction result output by step S32. Obtain the pre-stored benchmark threshold data and extract the reference health index. Calculate the deviation between the current health index and the reference value, and generate the threshold adjustment parameter in combination with the degradation development prediction result. Dynamically update the benchmark threshold according to the threshold adjustment parameter to obtain a multi-level dynamic threshold. Obtain the uncertainty assessment result and calculate the risk mapping parameter. According to the multi-level dynamic threshold and risk mapping parameters, perform a risk level assessment on the current state and output the risk level data. Obtain the pre-stored warning condition library and extract the time series data. Calculate the warning time advance and generate the time attenuation coefficient. Construct the warning condition matrix according to the risk level data and the time attenuation coefficient. Obtain the system intervention factor data and calculate the intervention weight coefficient. Use the intervention weight coefficient to optimize the warning conditions and output the final warning level and warning instructions.

[0036] Step S34, receiving the risk level data and warning level outputted from step S33. Obtain the pre-stored cost database and benefit evaluation model. Calculate the maintenance cost factor and benefit factor of each monitoring point, and generate an initial priority score. Obtain the score change rate data, and calculate the dynamic adjustment parameters. Modify the initial priority score according to the dynamic adjustment parameters to obtain the final maintenance priority ranking. Obtain the pre-stored maintenance plan library, and extract the plan feasibility data. Screen the maintenance plans according to the risk level and priority score to obtain a set of candidate plans. Calculate the benefit score, cost score, and reliability score of the candidate plans, and generate plan optimization parameters. Use the plan optimization parameters to comprehensively evaluate and rank the candidate plans, and output the final recommended maintenance plan.

[0037] In this embodiment, accurate early warning of pipeline diseases is achieved through the degradation trend analysis method based on health status distribution. A complete health status assessment system is constructed by introducing nonlinear feature fusion and probability density analysis technology. The innovative use of multimodal degradation pattern recognition algorithm, combined with the coupling analysis of environmental factors, realizes accurate modeling of pipeline degradation process. By introducing the evolutionary analysis method based on time kernel function, the accuracy of degradation trend prediction is improved. In the risk assessment link, a dynamic threshold update strategy is adopted, combined with a multi-layer warning condition matrix, to achieve accurate division of different risk levels. Through the innovative maintenance priority assessment method, cost-effectiveness, urgency and resource constraints are comprehensively considered to provide a scientific basis for pipeline maintenance decisions. This systematic health assessment and early warning scheme is particularly suitable for important infrastructure with high safety requirements such as buried pressure steel pipes, and can effectively prevent the occurrence of major accidents.

[0038] According to one aspect of the present application, step S11 is specifically:

[0039] Step S111, obtain the original data streams collected by the ultrasonic sensor, video sensor, wall thickness measurement sensor and position sensor. Extract the sampling time sequence from the ultrasonic data stream and generate a reference time axis. Generate a global clock sequence based on the reference time axis. Calculate the sampling interval for each sensor data stream and record the time interval sequence. Generate a timestamp identifier for each data stream based on the sampling interval sequence. Combine the timestamp identifier with the original data stream and output the original data stream with a timestamp.

[0040] Step S112, obtain the original data stream with timestamp output by step S111. Extract the sampling time sequence of each data stream, and calculate the time difference sequence with the reference time axis. Perform statistical analysis on the time difference sequence to obtain the time offset distribution characteristics. Calculate the initial compensation parameters according to the time offset distribution characteristics. Obtain the pre-stored sensor response characteristic data, and correct the initial compensation parameters according to the response characteristic data. Combine the corrected compensation parameters with the time difference sequence, and output the time compensation parameter set.

[0041] Step S113, obtain the time compensation parameter set output by step S112 and the original data stream with timestamp output by step S111. Extract the signal amplitude sequence of each data stream respectively and calculate the signal strength index. Generate the data stream weight coefficient according to the signal strength index. Multiply the weight coefficient with the time compensation parameter to obtain the weighted compensation parameter. Use the weighted compensation parameter to interpolate the original data stream with timestamp to generate a unified sampling sequence. Reconstruct the data stream according to the unified sampling sequence and output the multi-source data stream with time alignment.

[0042] In this embodiment, the data synchronization problem in the detection of buried water pressure steel pipes is effectively solved by the timing alignment method of multi-source sensor data. A unified time reference system is established by generating a global clock sequence based on the ultrasonic data stream as a reference. An innovative weight coefficient adjustment mechanism based on the sensor response characteristics is introduced to accurately correct the time compensation parameters, effectively eliminating the response delay differences between different sensors. An adaptive interpolation algorithm is used to reconstruct the data stream to ensure the continuity and consistency of the data. Through the dynamic evaluation of the signal strength index, real-time monitoring of data quality is achieved, ensuring the reliability of the timing alignment results. This multi-level timing alignment scheme is particularly suitable for scenarios such as buried pressure steel pipes that require collaborative detection of multiple sensors. It provides a high-quality synchronous data source for subsequent data analysis, significantly improving the accuracy and reliability of the detection results.

[0043] According to one aspect of the present application, step S12 is specifically:

[0044] Step S121, obtain the time-aligned multi-source data stream output by S113. Perform wavelet decomposition on each data stream to separate the signal component and the noise component. Calculate the energy distribution of each frequency band and generate a frequency band energy sequence. Adaptively segment the frequency band energy sequence to obtain segmented energy features. Calculate the local signal-to-noise ratio value using the segmented energy features to generate an initial signal-to-noise ratio sequence. Obtain the pre-stored noise feature template and calculate the noise similarity index. Modify the initial signal-to-noise ratio sequence according to the noise similarity index and output the modified signal-to-noise ratio sequence.

[0045] Step S122, obtain the corrected signal-to-noise ratio sequence output by S121 and the time-aligned multi-source data stream output by S113. Segment the data stream and extract data integrity features. Calculate the sampling density within the data segment and generate a density distribution sequence. Obtain the data transmission log and extract the data loss record. Mark the data missing interval according to the data loss record and generate an integrity mark sequence. Combine the density distribution sequence with the integrity mark sequence to calculate the data integrity score. Weight the signal-to-noise ratio sequence according to the integrity score and output the preliminary data quality evaluation result.

[0046] Step S123, obtain the data quality preliminary evaluation results output by S122 and the time-series aligned multi-source data stream. Calculate the sliding variance of the data stream and generate a volatility index sequence. Extract the mutation point position of the data and calculate the mutation intensity value. Construct a stability evaluation index based on the volatility index and the mutation intensity value. Obtain the pre-stored steady-state reference model and calculate the steady-state deviation sequence. Combine the steady-state deviation sequence with the stability evaluation index and output the stability score sequence.

[0047] Step S124: Obtain the data quality preliminary evaluation results output by S122 and the stability score sequence output by S123. Extract the time trend characteristics of each score sequence and generate a dynamic weight coefficient. Perform weighted fusion on each score according to the dynamic weight coefficient to obtain a comprehensive quality score. Calculate the time derivative of the quality score and generate a fluctuation penalty factor. Use the fluctuation penalty factor to correct the comprehensive quality score and output the data flow credibility score.

[0048] Step S125, obtain the data stream credibility score and time-series aligned multi-source data stream output by S124. Construct a multi-layer autoregressive prediction model to generate a predicted data sequence. Calculate the deviation between the actual data and the predicted data to obtain the original residual sequence. Adaptively segment the residual sequence and calculate the statistical features within the segment. Construct residual evaluation indicators based on the statistical features to generate anomaly candidate sequences. Obtain the pre-stored anomaly pattern library and calculate the anomaly similarity. Screen the anomaly candidate sequences based on the anomaly similarity and output the final anomaly score sequence.

[0049] Step S126, obtain the data stream credibility score output by S124 and the anomaly score sequence output by S125. Calculate the correlation coefficient of the score sequence and generate the correlation weight matrix. Perform a joint analysis on the credibility score and the anomaly score according to the correlation weight matrix to obtain an initial comprehensive score. Extract the time-varying characteristics of the score sequence and calculate the dynamic adjustment coefficient. Use the dynamic adjustment coefficient to modify the initial comprehensive score and output the data stream with the quality score and anomaly score.

[0050] In this embodiment, accurate quantification of the reliability of detection data is achieved through data quality assessment and anomaly detection methods. Multi-scale analysis of signals is performed through wavelet decomposition technology, combined with adaptive segmentation processing, to achieve accurate calculation of energy distribution in different frequency bands. The noise similarity index is innovatively introduced, and the accuracy of signal-to-noise ratio evaluation is improved by matching with pre-stored noise feature templates. A dual evaluation mechanism based on sampling density and data integrity is adopted, combined with data transmission log analysis, to achieve a comprehensive evaluation of data quality. By introducing volatility indicators and mutation intensity analysis, a complete stability evaluation system is constructed. Combined with a multi-layer autoregressive prediction model and anomaly pattern library, accurate identification of abnormal data is achieved. This systematic quality assessment scheme is particularly suitable for detection scenarios such as buried pressure steel pipes that have high data quality requirements. It can effectively filter unreliable data and improve the credibility of detection results.

[0051] According to one aspect of the present application, step S13 is specifically:

[0052] Step S131, obtain the data stream with quality score and anomaly score output by S126. Perform short-time Fourier transform on the data stream to obtain initial time-frequency spectrum data. Calculate the local peak distribution of the time-frequency spectrum and generate a peak feature sequence. Perform density clustering on the peak feature sequence to identify the main frequency components. Extract the amplitude change sequence of each frequency component and calculate the frequency modulation feature. Construct adaptive filter parameters according to the frequency modulation feature to obtain an initial filter parameter sequence.

[0053] Step S132, obtain the initial filter parameter sequence and time-frequency spectrum data output by S131. Calculate the frequency gradient of the time-frequency spectrum and generate a gradient feature map. Perform partition statistics on the gradient feature map to obtain the regional gradient distribution. Extract the time law of gradient change and generate a time-varying gradient sequence. Calculate the smoothing factor based on the time-varying gradient sequence and correct the initial filter parameters. Apply the corrected filter parameters to the time-frequency spectrum data and output the enhanced time-frequency spectrum.

[0054] Step S133, obtain the enhanced time-frequency spectrum data output by S132. Calculate the entropy value of the time-frequency spectrum at each time point to generate a time series entropy feature. Perform segmentation processing on the time series entropy feature and calculate the intra-segment statistics. Construct an information redundancy index based on the statistics and generate an initial information volume score. Obtain a pre-stored reference entropy model and calculate an entropy deviation sequence. Correct the initial information volume score according to the entropy deviation sequence and output a feature information volume sequence.

[0055] Step S134, obtain the feature information sequence and enhanced time-frequency spectrum data output by S133. Extract the time-domain variation characteristics of the time-frequency spectrum and calculate the time-varying index. Normalize the time-varying index to generate a dynamic feature sequence. Calculate the mutual information between feature sequences and construct a feature association network. Calculate the centrality index based on the feature association network and generate a feature importance sequence. Combine the feature importance sequence with the information sequence and output the feature evaluation index.

[0056] Step S135, obtain the feature evaluation index and enhanced time-frequency spectrum data output by S134. Construct a feature transformation matrix and perform an initial dimensionality reduction mapping. Calculate the feature preservation degree after dimensionality reduction and generate a dimensionality evaluation index. Optimize the transformation matrix according to the dimensionality evaluation index to obtain an improved feature mapping. Apply the improved feature mapping to the time-frequency spectrum data to generate initial low-dimensional features. Calculate the local structural features of the low-dimensional features and evaluate the degree of topological preservation. Fine-tune the low-dimensional features according to the degree of topological preservation and output the optimized dimensionality reduction feature data.

[0057] Step S136, obtain the reduced dimension feature data output by S135 and the feature evaluation index output by S134. Extract the local distribution features of the feature data and construct a feature space density map. Perform adaptive grouping based on the density map to obtain the feature group division result. Calculate the representative index of each feature group and generate a sequence of inter-group discrimination. Optimize the feature group division according to the inter-group discrimination and output the final reduced dimension feature data stream.

[0058] In this embodiment, effective dimensionality reduction of detection data is achieved through feature extraction and optimization methods based on time-frequency analysis. By using short-time Fourier transform technology combined with density clustering analysis of peak feature sequences, accurate identification of main frequency components is achieved. Frequency modulation feature analysis is innovatively introduced, and the robustness of feature extraction is improved through dynamic optimization of adaptive filtering parameters. A feature evaluation mechanism based on time series entropy and information redundancy is adopted to achieve accurate quantification of feature importance. A complete feature selection system is constructed through feature association network analysis and centrality index evaluation. Combined with local structure analysis of feature space, effective maintenance of topological relationships is achieved. This multi-dimensional feature optimization scheme is particularly suitable for detection objects such as buried pressure steel pipes with complex feature patterns. It can effectively extract key feature information, reduce data redundancy, and improve the efficiency of subsequent analysis.

[0059] According to one aspect of the present application, step S14 is specifically:

[0060] Step S141, obtain the dimension reduction feature data stream output by S136. Divide the feature sequence according to the data type to obtain the initial hierarchical data. Calculate the feature mean sequence for each layer of data to generate the hierarchical reference value. Calculate the feature deviation based on the hierarchical reference value to obtain the standardized parameter set. Use the standardized parameter set to normalize the data of each layer and output the standardized hierarchical data.

[0061] Step S142, obtain the standardized hierarchical data output by S141. Calculate the covariance matrix of each layer of data and extract the feature correlation index. Construct the inter-layer connection strength matrix based on the correlation index and generate the initial inter-layer weights. Extract the time consistency features of each layer of data and calculate the synchronization score. Modify the inter-layer weights based on the synchronization score and output the optimized inter-layer association parameters.

[0062] Step S143, obtain the inter-layer correlation parameters output by S142 and the standardized hierarchical data output by S141. Construct a hierarchical fusion network and calculate the network node weights. Perform weighted combination on the data of each layer to obtain the initial fusion result. Extract the local features of the fusion result and generate a feature consistency index. Adjust the network weights according to the feature consistency index and optimize the fusion process. Compare the optimized fusion result with the original hierarchical data and calculate the fusion loss. Fine-tune the fusion parameters based on the fusion loss and output the hierarchical fusion data.

[0063] Step S144, obtain the hierarchical fusion data output by S143. Calculate the local entropy distribution of the data and generate an entropy sequence. Segment the entropy sequence to obtain data block division points. Divide the fused data into a data block sequence according to the division points, and calculate the inter-block correlation. Optimize the data block size based on the inter-block correlation and generate the optimal block scheme. Calculate the feature statistics for each data block and output the block feature sequence.

[0064] Step S145, obtain the block feature sequence and data block sequence output by S144. Extract the frequency features of the data blocks and construct a spectral feature dictionary. Perform dictionary matching on each data block to obtain an initial coding sequence. Calculate the coding redundancy and generate compression optimization parameters. Adjust the coding strategy according to the compression optimization parameters to obtain improved compression coding. Perform error evaluation on the coding results and generate quality control indicators. Optimize the compression parameters according to the quality control indicators and output the compressed coding data.

[0065] Step S146, obtain the compressed coded data and quality control index output by S145. Construct a hierarchical storage model and calculate the storage priority. Group the compressed data according to the storage priority and generate a storage partitioning scheme. Extract the data access frequency characteristics and establish a cache strategy. Optimize the storage partition based on the cache strategy to obtain an improved storage scheme. Calculate the efficiency index of the storage scheme and generate a storage evaluation result. Adjust the storage parameters according to the storage evaluation result and output the final compressed storage data.

[0066] Step S147, obtain the compressed storage data and storage evaluation results output by S146. Construct a data index structure and generate a quick retrieval table. Perform integrity check on the compressed data and calculate the check sequence. Combine the check sequence with the compressed data to obtain a data packet structure. Extract key features of the data packet and generate metadata information. Combine the metadata information with the data packet and establish an association mapping. Integrate all information according to the association mapping and output a complete pipeline status feature data packet.

[0067] In this embodiment, efficient management of detection data is achieved through a hierarchical compression storage method. Normalization of hierarchical data is achieved through dynamic calculation of feature mean sequences and standardized parameters. The inter-layer connection strength matrix is ​​innovatively introduced, and a complete hierarchical fusion mechanism is established through time consistency feature analysis. An adaptive block strategy based on entropy distribution is adopted, combined with inter-block correlation analysis, to achieve dynamic generation of the optimal block scheme. An efficient data compression encoding system is constructed through real-time adjustment of spectral feature dictionary matching and compression optimization parameters. Combined with the hierarchical storage model and cache strategy, optimal configuration of storage resources is achieved. This systematic data management solution is particularly suitable for application scenarios such as buried pressure steel pipes that require long-term storage of large amounts of detection data. It can significantly improve storage efficiency and reduce storage costs, while ensuring rapid access and reliability of data.

[0068] According to one aspect of the present application, step S21 is specifically:

[0069] Step S211, obtain the pipeline state feature data packet output by S147. Extract the compressed feature data from the data packet and decompress it to obtain the initial feature vector. Calculate the time gradient sequence of the feature vector to generate gradient features. Extract the local extreme points of the feature data and calculate the feature fluctuation period. Segment the gradient features according to the fluctuation period to obtain a segmented gradient sequence. Obtain the historical feature sequence from the pre-stored historical database and calculate the sequence similarity. Filter the relevant historical data based on the sequence similarity and output the initial enhancement parameters.

[0070] Step S212, obtain the initial enhancement parameters and segmented gradient sequence output by S211. Calculate the cumulative distribution function of the gradient sequence and generate gradient statistical features. Extract the temporal law of gradient change and construct a dynamic threshold sequence. Filter the gradient features according to the dynamic threshold to obtain key gradient points. Calculate the spatial distribution characteristics of the key gradient points and generate the spatial weight coefficient. Combine the spatial weight coefficient with the enhancement parameter and output the optimized feature enhancement coefficient.

[0071] Step S213, obtain the feature enhancement coefficient output by S212 and the initial feature vector output by S211. Calculate the autocorrelation sequence of the feature vector and extract the periodic component. Perform frequency decomposition on the periodic component to generate a frequency feature spectrum. Extract the peak position of the frequency feature spectrum and calculate the main frequency distribution. Construct filter parameters according to the main frequency distribution and perform filtering enhancement on the feature vector. Combine the enhancement coefficient with the filtering result and output the enhanced feature vector.

[0072] Step S214, obtain the enhanced feature vector output by S213. Calculate the mutual covariance matrix between the feature components and generate the initial correlation matrix. Extract the time change rate of the features and construct a dynamic change sequence. Calculate the synchronization index between the change sequences and generate a time series correlation matrix. Perform feature decomposition on the correlation matrix and extract the main correlation mode. Calculate the dynamic difference value between the features according to the correlation mode and output the feature correlation strength data.

[0073] Step S215, obtain the feature association strength data output by S214. Construct a feature association network and calculate the network node degree distribution. Extract the community structure of the network and generate an initial grouping scheme. Calculate the aggregation index of the features within the group and construct the inter-group distance matrix. Optimize the grouping scheme based on the inter-group distance to obtain an improved feature grouping. Calculate the representative index for each feature group and output a hierarchical feature organization structure.

[0074] Step S216, obtain the hierarchical feature organization structure output by S215 and the enhanced feature vector output by S213. Extract the statistical moments of each layer of features and generate hierarchical descriptors. Calculate the information gain index between layers and construct a hierarchical importance sequence. Perform weighted combination of the features of each layer according to the importance sequence to obtain a multi-scale feature representation. Calculate the completeness index of the feature representation and generate a feature evaluation result. Optimize the feature representation according to the evaluation result and output the final hierarchical feature group.

[0075] In this embodiment, accurate reconstruction of pipeline state features is achieved through a state analysis method based on multi-dimensional feature enhancement. An adaptive feature enhancement mechanism is constructed through time gradient sequence analysis and fluctuation period recognition, combined with similarity matching of historical feature sequences. The gradient statistical features and dynamic threshold sequences are innovatively introduced, and the optimization adjustment of the feature enhancement coefficient is achieved through spatial distribution analysis of key gradient points. A complete periodic component extraction system is established by using autocorrelation sequence analysis and frequency decomposition technology, combined with the main frequency distribution characteristics. Accurate characterization of feature hierarchical relationships is achieved through feature association network analysis and community structure recognition. This multi-level feature enhancement scheme is particularly suitable for detection objects with complex state characteristics such as buried pressure steel pipes. It can effectively improve the expression ability of features, provide a more reliable feature basis for subsequent state evaluation, and significantly improve the accuracy of state recognition.

[0076] According to one aspect of the present application, step S21 is specifically:

[0077] Step S221, obtain the hierarchical feature group output by S216. Construct multiple time scale sequences, including hourly, daily and weekly scales. Calculate the sliding mean of the feature sequence for each time scale to generate trend baseline data. Extract the deviation between the feature sequence and the baseline and calculate the fluctuation feature sequence. Determine the optimal observation window length based on the fluctuation characteristics and generate a window parameter set. Extract local statistical features for each observation window and output a multi-scale observation sequence.

[0078] Step S222, obtain the multi-scale observation sequence output by S221. Calculate the autocorrelation function of each scale sequence and extract the periodic features. Perform spectral analysis on the periodic features to generate a frequency feature map. Extract the main components of the frequency features and build a periodic pattern library. Calculate the phase characteristics of the sequence according to the periodic pattern and generate a phase difference distribution. Determine the scale correlation strength based on the phase difference distribution and output the scale correlation parameter.

[0079] Step S223, obtain the scale association parameters output by S222 and the multi-scale observation sequence output by S221. Calculate the local coefficient of variation of the observation sequence and generate a variation index. Extract the mutation point position of the sequence and construct a mutation feature sequence. Perform density analysis on the mutation feature and generate a mutation pattern spectrum. Calculate the stability index of the sequence based on the mutation pattern spectrum and output the state variation feature.

[0080] Step S224, obtain the state variation characteristics output by S223 and the scale association parameters output by S222. Construct a state mapping function and calculate the initial state trajectory. Extract the geometric features of the trajectory and generate a curvature sequence. Segment the curvature sequence and construct the trajectory segmentation points. Divide the trajectory into multiple subsequences according to the segmentation points and calculate the subsequence characteristics. Optimize the state mapping based on the subsequence characteristics and output the optimized state trajectory.

[0081] Step S225, obtain the optimized state trajectory output by S224. Extract the evolution direction characteristics of the trajectory and generate a trend vector. Calculate the acceleration characteristics of the trajectory and construct a change rate sequence. Perform distribution analysis on the change rate sequence and generate a rate mode spectrum. Construct an evolution prediction model based on the rate mode spectrum to obtain prediction results at each scale. Calculate the confidence interval of the prediction result and output multi-scale prediction data.

[0082] Step S226, obtain the multi-scale prediction data output by S225. Calculate the historical prediction error sequence and generate error statistical features. Extract the time-varying features of the error and construct a dynamic weight sequence. Perform weighted fusion on the prediction results of each scale to obtain an initial fusion prediction. Calculate the uncertainty index of the fusion prediction and generate a credibility score. Optimize the fusion result according to the credibility score and output the final state evolution sequence.

[0083] Step S227, obtain the state evolution sequence output by S226 and the optimized state trajectory output by S224. Extract the key turning points of the evolution sequence and construct a state transition diagram. Calculate the topological features of the transition diagram and generate a structural feature sequence. Perform cluster analysis on the structural features to obtain a typical evolution pattern. Segmentally annotate the state trajectory according to the typical pattern and generate an evolution stage sequence. Combine the evolution stage sequence with the state evolution sequence and output a complete state trajectory analysis result.

[0084] In this embodiment, dynamic tracking of pipeline state evolution is achieved through a multi-scale state trajectory analysis method. A complete state observation system is established by constructing multi-layer time scale sequences at the hourly, daily and weekly levels, combined with the dynamic extraction of trend baseline data. The innovative introduction of periodic characteristic spectrum analysis and phase difference distribution characteristics, and the precise calculation of scale correlation strength, achieves effective fusion of state information at different time scales. The state variation feature analysis and mutation pattern spectrum recognition are adopted, combined with trajectory segmentation and geometric feature extraction, to construct a system state evolution evaluation mechanism. Accurate prediction of state development trends is achieved through rate pattern spectrum analysis and evolution prediction models. This multi-dimensional state analysis scheme is particularly suitable for detection objects such as buried pressure steel pipes with complex dynamic characteristics. It can accurately capture the laws of state changes and warn of potential risks in advance.

[0085] According to one aspect of the present application, step S23 is specifically:

[0086] Step S231, obtain the state trajectory analysis result output by S227. Extract the statistical moment sequence of the state data and generate the benchmark feature vector. Perform piecewise smoothing on the state data and calculate the smoothing coefficient sequence. Adjust the state data according to the smoothing coefficient to obtain the initial benchmark sequence. Extract the historical state records from the pre-stored historical database and calculate the historical statistical features. Combine the historical statistical features with the initial benchmark sequence to generate the dynamic benchmark model parameters. Construct the state benchmark according to the benchmark model parameters and output the benchmark state sequence.

[0087] Step S232, obtain the reference state sequence output by S231 and the state trajectory analysis result output by S227. Calculate the distance matrix between the state data and the reference sequence to generate a deviation sequence. Extract the rate of change characteristics of the state data and construct a velocity vector field. Perform divergence analysis on the velocity vector field to obtain a divergence sequence. Correct the deviation value according to the divergence sequence to generate an initial deviation score. Calculate the time accumulation value of the deviation score and output the state deviation score sequence.

[0088] Step S233, obtain the state deviation score sequence output by S232. Extract the local extreme value points of the deviation score and construct a feature point sequence. Calculate the spatiotemporal distribution characteristics of the feature points and generate a clustering index. Perform threshold analysis on the clustering index to obtain the initial anomaly label. Extract the neighborhood features of the anomaly label and calculate the spatial correlation. Optimize the anomaly label according to the spatial correlation and output the anomaly candidate sequence.

[0089] Step S234, obtain the abnormal candidate sequence output by S233. Extract standard pattern features from the pre-stored abnormal pattern library and construct a reference template set. Calculate the matching matrix between the abnormal candidate and the template to generate a similarity sequence. Extract the principal component features of the similarity sequence and construct a pattern mapping function. Calculate the probability of the abnormal type according to the pattern mapping to obtain the initial classification result. Calculate the confidence index for the classification result and output the abnormal pattern probability distribution.

[0090] Step S235, obtain the abnormal mode probability distribution output by S234 and the state deviation score sequence output by S232. Calculate the entropy value characteristics of the state sequence and generate the uncertainty index. Extract the time evolution characteristics of the probability distribution and construct the dynamic feature sequence. Perform stability analysis on the dynamic features to obtain the reliability score. Modify the probability distribution according to the reliability score to generate the optimized abnormal probability. Combine the uncertainty index with the abnormal probability and output the modified abnormal score.

[0091] Step S236, obtain the modified anomaly score output by S235 and the probability distribution of the anomaly pattern output by S234. Construct a probability transfer matrix and calculate the state transition characteristics. Extract the time series pattern of the transition sequence and generate evolution law data. Perform frequency analysis on the evolution law to obtain the main pattern sequence. Construct a prediction function based on the main pattern and calculate the anomaly development trend. Combine the anomaly development trend with the current anomaly score and output the anomaly warning indicator.

[0092] Step S237, obtain the abnormal warning indicators output by S236, the modified abnormal score and abnormal pattern probability distribution output by S235. Calculate the cumulative impact of various abnormalities and generate risk measurement indicators. Extract the risk propagation path characteristics and build an impact network model. Perform importance analysis on the impact network to obtain a sequence of key nodes. Calculate the system vulnerability based on the state distribution of key nodes and generate risk assessment results. Combine the risk assessment results with the abnormal warning indicators and output the final abnormal score and pattern classification results.

[0093] In this embodiment, accurate identification of abnormal pipeline states is achieved through an anomaly detection method based on a dynamic benchmark model. A complete anomaly assessment system is constructed by calculating state smoothing values ​​and analyzing deviation scores, combined with state change rate data. Innovatively introducing abnormal pattern matching and entropy feature analysis, reliable discrimination of abnormal states is achieved through dynamic evaluation of uncertainty indicators. By adopting probability transfer matrix analysis and state transition feature extraction, combined with evolution law data, a system abnormal development prediction mechanism is established. Accurate assessment of system vulnerability is achieved through risk measurement indicators and impact network models. This multi-level anomaly detection scheme is particularly suitable for detection objects with high safety requirements such as buried pressure steel pipes, and can detect potential faults in a timely manner and prevent major accidents.

[0094] According to one aspect of the present application, step S24 is specifically:

[0095] Step S241, obtain the anomaly score and pattern classification results output by S237 and the hierarchical feature group output by S216. Extract the spatial coordinate data of each monitoring point and calculate the distance matrix between points. Perform cluster analysis on the distance matrix to generate a spatial partition sequence. Calculate the feature distribution parameters within the partition and build a local feature model. Extract geometric feature parameters based on the local feature model and generate a spatial feature description. Combine the spatial feature description with the original feature and output feature data with spatial attributes.

[0096] Step S242, obtain the feature data with spatial attributes output by S241. Calculate the spatial autocorrelation function of the feature data and generate a correlation length sequence. Extract the spatial variation gradient of the feature and construct the gradient field data. Perform divergence analysis on the gradient field to obtain the spatial variation feature. Construct the spatial kernel function parameters according to the variation feature and generate the kernel function matrix. Perform convolution operation on the kernel function matrix and the feature data to output the spatial smoothing feature.

[0097] Step S243, obtain the spatial smoothing feature outputted by S242. Calculate the time accumulation value of the feature data and generate a time series feature sequence. Extract the periodic component of the time series feature and construct a periodic feature library. Perform phase analysis on the periodic feature to obtain a phase difference sequence. Construct a time coupling function based on the phase difference and generate a coupling coefficient matrix. Combine the coupling coefficient matrix with the spatial smoothing feature and output the spatiotemporal coupling data.

[0098] Step S244, obtain the spatiotemporal coupling data output by S243. Calculate the local response intensity of the data and generate a response feature map. Extract the main direction of the response feature and construct the direction field data. Perform density analysis on the direction field to obtain the characteristic flow field. Calculate the second-order derivative of the data according to the characteristic flow field to generate the curvature feature. Fuse the curvature feature with the spatiotemporal coupling data and output the enhanced feature data.

[0099] Step S245, obtain the enhanced feature data output by S244, the anomaly score and pattern classification results output by S237. Calculate the quality score of the feature data and generate a quality evaluation sequence. Extract the information entropy value of the feature and construct an information content index. Calculate the variance distribution of the feature data to obtain a stability score. Generate a storage priority based on the quality evaluation, information content and stability score, and output a feature priority sequence.

[0100] Step S246, obtain the feature priority sequence output by S245 and the enhanced feature data output by S244. Construct a hierarchical storage strategy and generate a storage level division. Compress and encode data of different levels to obtain a compressed data stream. Calculate the reconstruction error of the compressed data and generate an error evaluation index. Optimize the compression parameters according to the error evaluation and output the compressed storage data.

[0101] Step S247, obtain the compressed storage data output by S246, the anomaly score and pattern classification results output by S237. Construct a data index structure and generate a quick retrieval table. Calculate the integrity check code of the data and construct a verification sequence. Combine the verification sequence with the compressed data to obtain a data packet structure. Extract the key descriptors of the data packet and generate metadata information. Integrate the metadata information with the data packet and output a complete pipeline status feature data packet.

[0102] In this embodiment, a full-dimensional analysis of pipeline status characteristics is achieved through the spatiotemporal feature fusion method. A complete spatial feature extraction system is established through dynamic optimization of spatial kernel function parameters combined with local geometric feature analysis. The spatiotemporal coupling features and local response data analysis are innovatively introduced, and accurate enhancement of features is achieved through the extraction of second-order gradient information. A comprehensive evaluation of quality scores, information volume, and variance values ​​is adopted, combined with storage priority scoring, to construct a systematic data management mechanism. Efficient management of feature data is achieved through a hierarchical compression storage strategy. This multi-dimensional feature fusion solution is particularly suitable for detection objects such as buried pressure steel pipes with complex spatial distribution characteristics. It can fully grasp the pipeline status information and improve detection accuracy.

[0103] According to one aspect of the present application, step S31 is specifically:

[0104] Step S311, obtain the pipeline status feature data packet output by S247. Extract the compressed data from the data packet and decompress it to obtain the complete feature data. Obtain the historical health response record from the pre-stored historical database and calculate the response accumulation value. Time-weight the response accumulation value to generate the historical impact coefficient. Extract the time series features of the feature data and construct the time series feature vector. Modify the time series features according to the historical impact coefficient and output the initial time series features.

[0105] Step S312, obtain the initial time series features output by S311. Calculate the nonlinear correlation of the feature sequence and generate a correlation matrix. Extract the eigenvector of the correlation matrix and construct a mapping basis function. Expand the feature data by basis functions to obtain a feature component sequence. Calculate the weight coefficient of the component sequence and generate nonlinear mapping parameters. Use the mapping parameters to transform the feature data and output nonlinear mapping features.

[0106] Step S313, obtain the nonlinear mapping features output by S312. Calculate the entropy increase rate of the feature data and generate an information gain sequence. Extract the stability index of the feature and construct a fluctuation evaluation vector. Perform time-varying analysis on the feature sequence to obtain dynamic feature indexes. Generate feature weights based on information gain and stability indexes and output weighted feature sequences.

[0107] Step S314: Obtain the weighted feature sequence output by S313 and the anomaly score output by S247. Calculate the multi-scale entropy of the feature sequence and generate a complexity index. Extract the time distribution of the anomaly score and construct a risk weight sequence. Perform risk weighting on the feature data to obtain risk-adjusted features. Normalize the risk-adjusted features according to the complexity index and output the normalized health index.

[0108] Step S315, obtain the normalized health index output by S314. Extract the distribution function template from the pre-stored basic distribution library and calculate the parameter estimation value. Perform kernel density estimation on the health index and generate a probability density curve. Calculate the skewness and kurtosis of the probability density and construct a shape feature vector. Optimize the distribution parameters according to the shape characteristics and output the initial probability distribution.

[0109] Step S316, obtain the initial probability distribution outputted by S315. Calculate the fluctuation characteristics of the distribution and generate the instability index. Extract the time-varying characteristics of the distribution and construct the evolution feature sequence. Perform trend analysis on the evolution feature to obtain the development trend. Modify the probability distribution according to the instability index and the development trend, and output the modified probability distribution.

[0110] Step S317, obtain the modified probability distribution output by S316. Calculate the uncertainty index of the distribution and generate a confidence interval sequence. Extract typical values ​​of the health state and construct a reference state set. Perform reliability assessment on the current state and obtain a reliability index. Calculate the comprehensive health index based on the reliability index and confidence interval, and output the health state distribution data.

[0111] In this embodiment, an accurate assessment of the health status of the pipeline is achieved through a health feature construction method based on memory strength. An adaptive feature mapping mechanism is constructed through the cumulative analysis of historical health response data combined with the dynamic adjustment of memory strength parameters. The information gain value and stability score analysis are innovatively introduced, and the accurate quantification of health indicators is achieved through the optimization configuration of feature weight coefficients. A complete probability density distribution evaluation system is established by using a basic distribution function library and fluctuation feature analysis combined with uncertainty scoring. A comprehensive assessment of the health status of the pipeline is achieved through the combined analysis of multi-dimensional health indicators. This multi-level health assessment scheme is particularly suitable for detection objects such as buried pressure steel pipes with complex degradation characteristics. It can accurately reflect the health status of the pipeline and provide a reliable basis for maintenance decisions. This method effectively solves the problems of insufficient use of historical data and low accuracy of health status assessment in traditional health assessment methods.

[0112] According to one aspect of the present application, step S32 is specifically:

[0113] Step S321, obtain the health status distribution data output by S317. Perform wavelet decomposition on the health status data to generate a multi-scale coefficient sequence. Extract the statistical characteristics of each scale coefficient and construct a feature vector group. Obtain standard pattern data from the pre-stored basic degradation pattern library and calculate the pattern matching degree. Classify the status data according to the matching degree and generate an initial degradation pattern sequence. Combine the degradation pattern sequence with the health index change rate and output the basic degradation characteristics.

[0114] Step S322, obtain the basic degradation characteristics output by S321. Calculate the time first-order difference of the degradation characteristics to generate a change rate sequence. Extract the local extreme points of the change rate and construct a turning point sequence. Perform cluster analysis on the turning point sequence to obtain the key stage division. Calculate the stage characteristic parameters according to the key stage and generate the degradation stage characteristics. Combine the degradation stage characteristics with the change rate sequence to output the segmented degradation pattern.

[0115] Step S323, obtain the segmented degradation pattern output by S322. Perform kernel density estimation on the degradation data to generate a density distribution curve. Extract the peak position of the density curve and construct a pattern center sequence. Calculate the migration trajectory of the pattern center to obtain the evolution path characteristics. Construct a time kernel function based on the evolution path characteristics to generate a time weight sequence. Combine the time weight sequence with the degradation pattern to output the time series degradation characteristics.

[0116] Step S324, obtain the time series degradation characteristics output by S323. Extract the temperature sequence in the environmental monitoring data and calculate the temperature change characteristics. Obtain the pressure monitoring data and generate the pressure fluctuation sequence. Extract the humidity monitoring data and construct the humidity distribution characteristics. Calculate the comprehensive environmental index based on the temperature, pressure and humidity data and output the environmental characteristic sequence.

[0117] Step S325: Obtain the environmental feature sequence output by S324. Calculate the mutual correlation coefficients between environmental parameters and generate a correlation matrix. Extract the eigenvalues ​​of the correlation matrix and construct a feature vector group. Perform principal component analysis on the environmental features to obtain the main influencing factors. Construct an environmental coupling function based on the influencing factors and generate a coupling strength sequence. Combine the coupling strength sequence with the environmental features and output the environmental action characteristics.

[0118] Step S326, obtain the environmental action characteristics output by S325 and the time series degradation characteristics output by S323. Calculate the response relationship between the degradation rate and the environmental factor and generate a response coefficient matrix. Extract the time-varying characteristics of the response coefficient and build a dynamic response model. Optimize the parameters of the response model to obtain the acceleration factor sequence. Correct the degradation characteristics according to the acceleration factor and generate a corrected degradation model. Calculate the fluctuation degree of the correction model and output the acceleration effect data.

[0119] Step S327, obtain the acceleration effect data and the modified degradation mode outputted by S326. Construct a degradation trend prediction model and generate an initial prediction sequence. Calculate the error distribution of the prediction sequence and construct a confidence interval. Perform a reliability assessment on the prediction results and obtain a credibility index. Optimize the prediction sequence according to the credibility index and generate a final prediction result. Combine the prediction result with the acceleration effect data and output the degradation development prediction result.

[0120] In this embodiment, accurate modeling of the pipeline degradation process is achieved through a trend analysis method based on multimodal degradation pattern recognition. An adaptive degradation pattern recognition mechanism is constructed through dynamic matching of the basic degradation pattern library and combined with change rate data analysis. The time kernel function and environmental coupling coefficient analysis are innovatively introduced, and accurate prediction of the degradation process is achieved through real-time optimization of the acceleration factor model. A complete degradation trend assessment system is established by using fluctuation correction parameters and acceleration effect data analysis combined with multiple verification mechanisms. Accurate quantification of the impact of external factors is achieved through comprehensive analysis of environmental monitoring data. This multi-dimensional degradation analysis scheme is particularly suitable for buried pressure steel pipes, which are detection objects affected by multiple environmental factors. It can accurately predict the development trend of degradation and provide a scientific basis for preventive maintenance. This method effectively solves the problem that traditional degradation analysis methods do not consider environmental factors enough and have low prediction accuracy.

[0121] According to one aspect of the present application, step S33 is specifically:

[0122] Step S331, obtain the health status distribution data output by S317 and the degradation development prediction result output by S327. Extract the standard threshold data from the pre-stored benchmark threshold library and calculate the reference health index. Standardize the health status data to generate a normalized health sequence. Calculate the deviation between the current health index and the reference value and construct a deviation sequence. Calculate the adjustment factor based on the deviation sequence and the degradation prediction data and output the threshold adjustment parameter.

[0123] Step S332, obtain the threshold adjustment parameters output by S331. Extract historical warning records and calculate the warning threshold distribution. Perform hierarchical clustering on the threshold distribution to generate a multi-level threshold sequence. Dynamically update the thresholds at each level according to the adjustment parameters and construct a threshold update function. Apply the update function to the baseline threshold and output a dynamic threshold sequence.

[0124] Step S333, obtain the dynamic threshold sequence output by S332 and the degradation development prediction result output by S327. Calculate the uncertainty index of the prediction result and generate a confidence sequence. Extract the mutation characteristics of the prediction data and construct the risk sensitivity index. Construct a risk mapping function based on the uncertainty index and sensitivity and generate risk conversion parameters. Use the risk conversion parameters to map the prediction data and output the risk mapping result.

[0125] Step S334: Obtain the risk mapping result output by S333 and the dynamic threshold sequence output by S332. Calculate the distance between the risk value and each level of threshold and generate a risk measurement sequence. Extract the time series characteristics of risk evolution and construct trend indicators. Evaluate the current state based on the risk measurement and trend indicators and generate a preliminary risk level judgment result. Verify the reliability of the preliminary judgment result and output the risk level data.

[0126] Step S335, obtain the risk level data output by S334. Extract standard warning rules from the pre-stored warning condition library and calculate the rule matching degree. Perform time series analysis on the warning rules to generate a time-related sequence. Calculate the distribution of warning lead time and construct a time decay function. Modify the warning rules according to the time decay function and output the warning condition sequence.

[0127] Step S336: Obtain the warning condition sequence output by S335 and the risk level data output by S334. Extract the system operation state parameters and calculate the state feature vector. Classify and analyze the state features to generate a state pattern sequence. Extract intervention factors based on the state pattern and construct an intervention weight function. Use the intervention weight to optimize the warning conditions and output the warning strategy plan.

[0128] Step S337, obtain the early warning strategy program outputted by S336. Calculate the execution cost of the early warning action and generate a cost evaluation sequence. Extract the evaluation index of the early warning effect and construct a benefit score. Perform multi-objective optimization on the early warning strategy to obtain an optimized early warning program. Generate early warning execution instructions according to the optimized program, and output the early warning level and early warning instructions.

[0129] Step S338, obtain the warning level and warning instructions output by S337, and the risk level data output by S334. Construct a warning response evaluation model and calculate the response feature sequence. Extract historical warning effect data and generate effect evaluation indicators. Evaluate the performance of the warning system and obtain the system reliability index. Optimize the warning parameters according to the reliability index and generate improvement suggestions. Integrate the improvement suggestions with the warning plan and output the final risk rating data and warning instructions.

[0130] In this embodiment, accurate assessment and timely warning of pipeline risks are achieved through a multi-level dynamic threshold warning method. An adaptive threshold adjustment system is constructed through a dynamic update mechanism of the benchmark threshold, combined with health index deviation analysis. The risk mapping parameters and warning time advance analysis are innovatively introduced, and the precise formulation of the warning strategy is achieved through the optimal configuration of the intervention weight coefficient. A complete risk assessment system is established by using the warning condition matrix and system intervention factor analysis, combined with a multi-layer verification mechanism. Accurate division of risk levels is achieved through dynamic adjustment of the warning level. This multi-level warning scheme is particularly suitable for detection objects with high safety requirements such as buried pressure steel pipes, and can detect potential risks in a timely manner and prevent accidents. This method effectively solves the problems of delayed warning and high false alarm rate of traditional warning methods.

[0131] According to one aspect of the present application, step S34 is specifically:

[0132] Step S341, obtain the risk rating data and warning instructions output by S338. Obtain maintenance cost parameters from the pre-stored cost database and calculate the basic cost vector. Extract equipment operation time data and generate time loss coefficient. Quantitatively evaluate the maintenance difficulty and construct a difficulty index sequence. Calculate the comprehensive cost factor based on the cost parameter, time coefficient and difficulty index, and output the cost evaluation sequence.

[0133] Step S342, obtain the cost evaluation sequence output by S341. Extract the evaluation function from the pre-stored benefit model library and calculate the direct benefit index. Extract the equipment life extension data after maintenance and generate the life gain sequence. Perform reliability analysis on the maintenance quality and construct quality assurance parameters. Calculate the comprehensive benefit factor based on the benefit index and reliability parameter and output the benefit evaluation sequence.

[0134] Step S343: Obtain the cost evaluation sequence output by S341 and the benefit evaluation sequence output by S342. Calculate the cost-benefit ratio and generate the initial priority index. Extract the historical maintenance records and construct the experience weight sequence. Evaluate the maintenance urgency and generate the urgency coefficient. Calculate the comprehensive priority score based on the priority index and the urgency coefficient and output the initial priority sequence.

[0135] Step S344, obtain the initial priority sequence output by S343. Calculate the changing trend of the score and generate a trend feature sequence. Extract the fluctuation characteristics of the score and construct a stability index. Perform a time series analysis on the score to obtain a dynamic feature vector. Optimize the priority score according to the trend characteristics and stability index, and output a modified priority sequence.

[0136] Step S345, obtain the correction priority sequence output by S344. Extract the standard solution template from the maintenance solution library and calculate the solution matching degree. Decompose the maintenance operation and generate an operation sequence diagram. Extract the dependency relationship between operations and construct an association matrix. Optimize the operation sequence according to the association matrix and output the initial maintenance solution.

[0137] Step S346, obtain the initial maintenance plan output by S345. Extract the equipment operating environment data and calculate the environmental constraints. Perform availability analysis on maintenance resources and generate resource restriction sequences. Extract personnel skill distribution data and build a skill matching matrix. Filter feasible solutions according to various constraints and output a set of candidate solutions.

[0138] Step S347, obtain the candidate solution set outputted by S346. Calculate the solution execution efficiency index and generate an efficiency score sequence. Extract solution reliability data and construct a risk assessment matrix. Evaluate the adaptability of the solution and obtain the adaptability index. Sort the candidate solutions according to the multi-dimensional evaluation results and output the solution optimization sequence.

[0139] Step S348: Obtain the solution optimization sequence output by S347 and the revised priority sequence output by S344. Extract historical execution feedback data and calculate the success rate index. Verify the solution implementation conditions and generate a feasibility score. Construct solution adjustment rules and obtain an optimization parameter set. Adjust the maintenance solution according to the optimization parameters and generate a final maintenance plan. Combine the maintenance plan with the priority sequence and output the maintenance priority ranking and recommended maintenance solution.

[0140] In this embodiment, the optimization formulation of pipeline maintenance strategies is achieved through a maintenance decision-making method based on multi-dimensional evaluation. A scientific priority evaluation mechanism is constructed through a comprehensive analysis of cost factors and benefit factors, combined with real-time optimization of dynamic adjustment parameters. The feasibility evaluation of the scheme and the resource constraint analysis are innovatively introduced, and the accurate screening of maintenance schemes is achieved through the dynamic configuration of the skill matching matrix. A complete scheme evaluation system is established by adopting efficiency scoring, risk assessment and adaptability indicator analysis, combined with multiple verification mechanisms. Through the comprehensive analysis of historical execution feedback data, continuous optimization of maintenance schemes is achieved. This multi-dimensional decision-making scheme is particularly suitable for detection objects such as buried pressure steel pipes that are difficult to maintain. It can formulate the optimal maintenance strategy, improve maintenance efficiency, and reduce maintenance costs. This method effectively solves the problem that traditional maintenance decision-making methods do not consider resource constraints enough and have low maintenance efficiency.

[0141] Example 2: The implementation process is as follows

[0142] Time base establishment: Create a master clock sequence T = {t1, t2, ..., t n}, the sampling frequency is 1000Hz; assign an independent timestamp sequence ST_i={st_i1,st_i2,...,st_i m}

[0143] Cross-modal data synchronization correction: Calculate the time offset of each sensor data stream: ΔT_i(k)=μ_i·∫[w_i(t)·|ST_i(t) is T(t)|]dt; where: μ_i is the sensor type characteristic coefficient (0<μ_i<1); w_i(t) is the time-varying weight function; k is the segment number;

[0144] A dynamic compensation mechanism is established: T_comp(i,k)=T(k)+ΔT_i(k)·exp(-λ·|dΔT_i / dt|); where: λ is the delay attenuation coefficient; dΔT_i / dt is the time offset change rate.

[0145] Signal quality assessment: Construct quality scoring function: Q(d)=α·SNR(d)+β·Comp(d)+γ·Stab(d)

[0146] Where: d is the data stream; SNR is the signal-to-noise ratio evaluation function; Comp is the integrity evaluation function; Stab is the stability evaluation function; α, β, γ are weight coefficients and α+β+γ=1;

[0147] Calculate the data stream credibility: R(d)=Q(d)·[1 is η·V(Q(d))]; where: V(Q(d)) is the variance of the quality score; η is the fluctuation penalty coefficient (0<η<1).

[0148] Data flow anomaly detection, including:

[0149] Construct the residual sequence: ε(t)=d(t) is P(d(t)|d(t-1),...,d(tk)); where: P(·) is the k-order prediction function; k is the size of the lookback window;

[0150] Anomaly score calculation: A(t)=|ε(t)|·exp(ρ·|dε / dt|)·[1+σ·V(ε)];

[0151] Where: ρ is the rate of change weight; σ is the volatility weight; V(ε) is the local variance;

[0152] Data denoising and enhancement, including: Building adaptive filters:

[0153] H(ω,t)=[1+θ·|S(ω,t)|²]·exp(-φ·|dS / dω|);

[0154] Where: S(ω,t) is the time spectrum; θ is the enhancement coefficient; φ is the smoothing coefficient; dS / dω is the frequency gradient;

[0155] Signal reconstruction: d'(t)=IFFT{H(ω,t)·FFT[d(t)]}.

[0156] Calculate feature importance: I(f)=E·[1+ψ·|df / dt|]·[1 is κ·C(f,F)];

[0157] Where: E is the information entropy; ψ is the time-varying weight; κ is the redundancy weight; C(f,F) is the mutual correlation coefficient between feature f and feature set F.

[0158] Construct a compressed mapping: M:X→Y,dim(Y) <dim(X);Y=G[X·W(I)];

[0159] Where: W(I) is the weight matrix based on importance; G is the nonlinear mapping function.

[0160] Construct a multi-level fusion structure: D_l=F_l{D_{l-1},N_l(d_l)};

[0161] Where: D_l is the fusion result of the lth layer; F_l is the fusion function; N_l is the normalization function; d_l is the original data of the current layer.

[0162] Calculate the inter-layer weight: W_l=softmax[ω·Q(D_l)+(1-ω)·I(D_l)];

[0163] Where: Q(D_l) is the quality score; I(D_l) is the information score; ω is the balance factor (0<ω<1).

[0164] Data partitioning strategy: B_i=Split(D,δ·H(D_i)+(1-δ)·V(D_i))

[0165] Where: H(D_i) is the block-level entropy; V(D_i) is the data variability; δ is the block weight coefficient;

[0166] Compression coding: C(B_i)=E[B_i·Z(Q(B_i))];

[0167] Where: E is the encoding function; Z is the compression rate adjustment function; Q(B_i) is the block quality score.

[0168] Output: time-aligned multimodal data stream; data quality score matrix; feature matrix after dimensionality reduction; compressed and stored block data.

[0169] Based on the reduced dimension feature matrix, construct the initial feature vector: F(t)=[f1(t),f2(t),...,fn(t)];

[0170] Where fi(t) is the i-th basic characteristic component;

[0171] Calculate the progressive enhancement feature: F'(t)=F(t)+α·LA F(t)+β·∫[w(τ)·F(t-τ)]dτ;

[0172] Where: α is the gradient enhancement coefficient; β is the historical information weight; w(τ) is the time decay function. LA is the Laplace operator.

[0173] Feature correlation analysis, including: constructing a feature correlation strength matrix: C(i,j,t)=λ·|cov(fi,fj)|·exp(-η·|dfi / dt is dfj / dt|);

[0174] Among them: λ is the basic correlation coefficient; η is the dynamic difference penalty factor;

[0175] Generate a hierarchical feature group: G_k={fi|C(i,j,t)>θk, any j∈G_k}; where θk is the k-th layer association threshold.

[0176] Construct a multi-scale observation window: W_s(t)={F'(tk·s)|k=0,1,...,m};

[0177] Where: s is the scale parameter; m is the window length;

[0178] Calculate the state trajectory: T(t,s)=Φ[W_s(t)]·exp(-λ·V[W_s(t)]);

[0179] Where: Φ is the trajectory mapping function; V is the local variability; λ is the smoothing factor.

[0180] Construct a multi-layer predictor: P_l(t+Δt)=ψ_l[T(t,s_l)]+ω_l·∫[K_l(τ)·dT / dτ]dτ;

[0181] Among them: ψ_l is the prediction function of the lth layer; ω_l is the dynamic weight; K_l is the kernel function.

[0182] Fusion of multi-layer prediction results: P(t+Δt)=∑[μ_l·P_l(t+Δt)]·[1 is ε·E_l(t)];

[0183] Among them: μ_l is the layer weight; E_l is the historical prediction error; ε is the error penalty coefficient.

[0184] State deviation detection, building a dynamic benchmark model: B(t)=γ·B(t-1)+(1-γ)·M[T(t,s)];

[0185] Where: γ is the historical weight (0<γ<1); M is the smoothing operator;

[0186] Calculate the deviation score: D(t) = ||T(t,s) is B(t)||·[1+ρ·|dT / dt|];

[0187] Where: ρ is the rate of change weight; ||·|| is the weighted norm.

[0188] Abnormal pattern classification, including: Building a hierarchical probability model:

[0189] P(A_k|D,T)=π_k·L(D,T|A_k) / ∑[π_j·L(D,T|A_j)];

[0190] Where: A_k is the kth abnormal pattern; π_k is the prior probability; L is the likelihood function.

[0191] Abnormality assessment: S(A_k,t)=P(A_k|D,T)·[1+φ·H(D,T)];

[0192] Where: φ is the uncertainty weight; H is the information entropy.

[0193] Feature enhancement and optimization, including spatiotemporal feature fusion and feature enhancement:

[0194] Fusion of spatiotemporal features to construct spatiotemporal coupling features: F_st(x,t)=F'(t)oK_s(x)·[1+δ·G(x,t)]

[0195] Where: o is the convolution operation; K_s is the spatial kernel function; G is the local geometric feature; δ is the coupling strength.

[0196] Feature enhancement: F_e(x,t)=F_st(x,t)+χ·LA 2 F_st(x,t)·R(x,t);

[0197] Where: χ is the enhancement coefficient; LA 2 is the Laplace operator; R is the local response function.

[0198] Feature optimized storage, including calculation of storage priority and hierarchical storage strategy: Calculation of storage priority: P_store(f)=Q(f)·[1+ν·I(f)]·exp(-σ·V(f)); where: Q is the quality score; I is the amount of information; V is the variance; ν,σ are weight coefficients;

[0199] Hierarchical storage strategy: Store(f)=Compress[f,L(P_store(f))]; where: L is the compression level function; Compress is the adaptive compression function.

[0200] Output data: enhanced feature matrix F_e; state trajectory sequence T; anomaly score sequence D; anomaly pattern probability distribution P(A_k).

[0201] Based on the enhanced feature matrix F_e, a multidimensional health index is constructed: H(t)=Ω[F_e(t)]·[1+κ·∫R(τ)dτ];

[0202] Where: Ω is the nonlinear mapping function; R(τ) is the historical response function; κ is the memory strength coefficient.

[0203] Dynamic optimization of feature weights: W(t)=softmax{ζ·I(F_e)+η·S(F_e)+μ·D(t)};

[0204] Where: I(F_e) is the information gain; S(F_e) is the stability score; D(t) is the anomaly score from S2; ζ, η, μ are balance coefficients.

[0205] Construct health status distribution: P(h|H)=∑[π_i·φ_i(h|θ_i)]·exp(-λ·V[H]);

[0206] Among them: φ_i is the basic distribution function; θ_i is the distribution parameter; π_i is the mixing weight; V[H] is the volatility measure.

[0207] Calculate the comprehensive health index: HI(t)=∫h·P(h|H)dh·[1 is ε·E(t)];

[0208] Where: E(t) is the uncertainty evaluation function; ε is the uncertainty penalty coefficient.

[0209] Construct degradation mode library: M(t)=∑[α_k·Ψ_k(t)]·exp(-β·|dHI / dt|);

[0210] Among them: Ψ_k is the basic degradation mode function; α_k is the mode weight; β is the change rate penalty coefficient;

[0211] Deterioration trend extraction: T(t)=M(t)+γ·∫[K(τ)·M(t-τ)]dτ;

[0212] Among them: K(τ) is the time kernel function; γ is the historical impact weight.

[0213] Construct the acceleration factor model: A(t)=Φ[E_1(t),E_2(t),...,E_n(t)]·[1+ρ·C(t)];

[0214] Where: E_i(t) is the environmental factor; Φ is the nonlinear mapping function; C(t) is the coupling strength; ρ is the coupling coefficient.

[0215] Calculate the acceleration effect: R(t)=dT / dt·[1+σ·A(t)]·exp(-ω·V[A]);

[0216] Where: σ is the acceleration factor; V[A] is the volatility of the acceleration factor; ω is the volatility penalty coefficient.

[0217] Construct a dynamic threshold system: Θ(t)=Θ_0+δ·[HI(t) is HI_ref]·[1+ν·R(t)];

[0218] Among them: Θ_0 is the benchmark threshold vector; HI_ref is the reference health index; δ is the adjustment coefficient; ν is the risk sensitivity.

[0219] Risk level calculation: RL(t)=f[HI(t),T(t),A(t)]·[1 is μ·U(t)];

[0220] Where: f is the risk mapping function; U(t) is the uncertainty function; μ is the uncertainty weight.

[0221] Construct the warning condition matrix: W(t)=[w_ij(t)]; w_ij(t)=ψ[RL_i(t),Θ_j(t)]·exp(-η·Δt);

[0222] Where: ψ is the conditional evaluation function; Δt is the time advance; η is the time decay coefficient.

[0223] Determination of warning level: AL(t)=argmax{P[W(t)|S(t)]·[1+λ·I(t)]};

[0224] Where: S(t) is the system state vector; I(t) is the intervention factor; λ is the intervention weight.

[0225] Construct priority score: PS(i,t)=χ·RL_i(t)+ξ·C_i(t)+ζ·E_i(t);

[0226] Among them: C_i(t) is the cost factor; E_i(t) is the benefit factor; χ, ξ, ζ are weight coefficients.

[0227] Dynamic sorting strategy: Rank(i,t)=PS(i,t)·[1+τ·ΔPS_i(t)]·exp(-φ·U_i);

[0228] Where: ΔPS_i(t) is the rate of change of score; U_i is the uncertainty measure; τ, φ are adjustment parameters.

[0229] Construct solution library mapping: M(s,t)=Γ[RL(t),PS(t)]·[1+ω·F(s)];

[0230] Where: Γ is the scheme mapping function; F(s) is the feasibility score; ω is the feasibility weight.

[0231] Optimization scheme selection: S*(t)=argmax{B[s,t]·[1 is ε·C(s)]·R(s)};

[0232] Where: B[s,t] is the scheme benefit function; C(s) is the cost function; R(s) is the reliability score; ε is the cost weight.

[0233] Output: health index sequence HI(t); risk level assessment RL(t); warning level sequence AL(t); maintenance priority ranking Rank(i,t); recommended maintenance plan S*(t).

[0234] Example 3: Other steps refer to Example 1 and Example 2, with the following differences:

[0235] Time alignment model: TC(i) = ∑(w[i] * (t[i] is t[ref])) * R[i]; where: TC(i) is the time compensation value of the i-th sensor; w[i] is the weight coefficient of the i-th sensor (0 < w[i] ≤ 1); t[i] is the sampling timestamp of the i-th sensor; t[ref] is the reference time series; R[i] is the sensor response characteristic correction coefficient.

[0236] Data quality scoring model: Q = α * SNR + β * Comp + γ * Stab; SNR = 10 * log10(Ps / Pn); Comp = Nd / Nt; Stab = 1 for σ / μ; where: Q is the comprehensive quality score; α, β, γ are weight coefficients, and α + β + γ = 1; SNR is the signal-to-noise ratio; Comp is the data integrity ratio; Stab is the stability ratio; Ps is the signal power; Pn is the noise power; Nd is the number of valid data points; Nt is the total number of data points; σ is the standard deviation; μ is the mean.

[0237] Feature extraction model: F(t, ω) = ∫x(τ)h(τ - t)e -jωτ dτ; K(ω) = LA F(t, ω) / LA ω; AF(t, ω) = F(t, ω) * H(K(ω)); where: F(t, ω) is the time-frequency spectrum; x(τ) is the original signal; h(τ) is the window function; K(ω) is the frequency gradient; AF(t, ω) is the time-frequency spectrum after adaptive filtering; H(·) is the adaptive filtering function;

[0238] Feature enhancement model: E(f) = f + λ * LA f * M(h); M(h) = exp(-||h||² / 2σ²); λ = α * (1 for exp(-var(LA f))); LA is the Laplace operator.

[0239] where: E(f) is the enhanced feature; f is the original feature; LA f is the time gradient; M(h) is the memory intensity function; h is the historical feature sequence; λ is the enhancement coefficient; σ is the scale parameter; var(·) is the variance operator. LA is the Laplace operator.

[0240] State trajectory prediction model: S(t + τ) = ∑w[i] * S[i](t + τ); w[i] = exp(-ε[i]² / σ[i]²); ε[i] = ||S[i](t) is S_actual(t)||;

[0241] where: S(t + τ) is the state prediction value after τ time; w[i] is the prediction weight of the i-th scale; S[i](t) is the prediction value of the i-th scale; ε[i] is the historical prediction error; σ[i] is the scale parameter; S_actual(t) is the actual state value.

[0242] Anomaly detection model: AD(t)=||S(t) is B(t)||*V(t)*P(m); B(t)=α*B(t-1)+(1-α)*S(t); V(t)=dS(t) / dt; P(m)=exp(-D(S(t),M));

[0243] Where: AD(t) is the anomaly score; S(t) is the current state; B(t) is the dynamic baseline value; V(t) is the state change rate; P(m) is the probability of abnormal pattern matching; α is the smoothing coefficient (0<α<1); D(·,·) is the distance measure between the state and the abnormal pattern; M is the pre-stored abnormal pattern set.

[0244] Health status assessment model: H(t)=∑w[i]*f[i](t)*R(h); R(h)=exp(-λ*t)*∑h(τ) / N;

[0245] w[i]=IG[i]*ST[i]; where: H(t) is the health index; f[i](t) is the i-th characteristic component; R(h) is the memory strength parameter; h(τ) is the historical health response sequence; w[i] is the feature weight; IG[i] is the information gain value; ST[i] is the stability score; λ is the decay coefficient; N is the length of historical data.

[0246] Degradation trend prediction model: D(t+Δt)=D(t)*AF(E)*(1+G(t)); AF(E)=∏(1+β[i]*E[i]); G(t)=α*(dD / dt)+(1-α)*G(t-1); where: D(t) is the degree of degradation at time t; AF(E) is the environmental acceleration factor; E[i] is the i-th environmental parameter; β[i] is the environmental parameter weight; G(t) is the degradation growth rate; α is the smoothing coefficient.

[0247] Risk assessment and early warning model: R(t)=W(t)*(1+δ(t))*exp(ΔH(t) / H_ref); W(t)=1 is exp(-λ*t); δ(t)=||D(t) is D_pred(t)|| / D_pred(t);

[0248] Where: R(t) is the risk level; W(t) is the time decay function; δ(t) is the prediction deviation rate; ΔH(t) is the change in health index; H_ref is the reference health value; D_pred(t) is the predicted degradation degree; λ is the time decay coefficient.

[0249] Maintenance priority evaluation model: P=(C_b*E_f) / (C_m*T_r)*U_f; E_f=∑w[i]*B[i] / C[i];

[0250] U_f=1 is exp(-R*T_d); where: P is the maintenance priority score; C_b is the baseline cost factor; E_f is the benefit factor; C_m is the maintenance cost; T_r is the time required for maintenance; U_f is the urgency coefficient; B[i] is the i-th benefit indicator; C[i] is the i-th cost indicator; w[i] is the weight coefficient; R is the risk level; T_d is the delay time.

[0251] Spatiotemporal feature fusion model, F_st(x,t)=K_s(x)*K_t(t)*F(x,t); K_s(x)=exp(-||x is x_0||² / σ_s²); K_t(t)=exp(-|t is t_0|² / σ_t²); where: F_st(x,t) is the spatiotemporal fusion feature; K_s(x) is the spatial kernel function; K_t(t) is the temporal kernel function; F(x,t) is the original feature; x_0, t_0 are the reference position and time point; σ_s, σ_t are the spatial and temporal scale parameters.

[0252] It should be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.

Claims

1. A method for inspecting buried water supply penstock for defects, characterized in that: The steps include: Step S1: According to the original data stream collected by each sensor, preprocessing including time alignment, quality scoring, feature extraction and feature optimization is performed in sequence to obtain feature data after dimensionality reduction; The feature data after dimension reduction is compressed and stored in blocks according to quality level and data type to obtain compressed data blocks; Step S2, reading and decompressing the compressed data block and reconstructing the features to obtain an initial feature vector; The initial feature vector is enhanced, correlated and analyzed, and the state trajectory is analyzed to obtain the state evolution sequence and perform anomaly detection and classification to obtain anomaly scores and pattern classification results. Then, spatiotemporal feature fusion and optimized storage are performed to form a pipeline state feature data packet. Step S3, obtain and construct an initial health feature vector based on the pipeline state feature data packet, perform nonlinear feature fusion and probability density analysis on the initial health feature vector, and obtain health state distribution data; based on the health state distribution data, perform degradation trend analysis and acceleration factor calculation to obtain degradation development prediction results; perform risk level assessment and early warning strategy generation according to the degradation development prediction results, and output risk rating data and early warning instructions; based on the risk rating data, generate a maintenance priority sequence and a specific maintenance plan; Among them, the spatiotemporal features are fused and optimized for storage to form a pipeline state feature data packet, specifically: Generate spatial kernel function parameters based on spatial position data and local geometric features; use spatial kernel function parameters to perform spatiotemporal fusion processing on hierarchical feature groups to obtain spatiotemporal coupling features; calculate local response data of features and extract second-order gradient information; enhance spatiotemporal coupling features based on local response data and gradient information to obtain enhanced feature data; combine enhanced feature data, state evolution sequence, anomaly score and pattern classification results to form a complete pipeline state feature data packet; calculate the quality score, information content and variance value of each component in the data packet to generate a storage priority score; perform hierarchical compression and storage on the data packet according to the storage priority score, and output the final pipeline state feature data packet.

2. The method for inspecting buried water supply penstock defects according to claim 1 is characterized in that: The step S1 is specifically as follows: Step S11, obtaining the original data streams collected by each sensor, including ultrasonic data stream, video data stream, wall thickness measurement data stream and position data stream; generating a master clock sequence, and recording the timestamp information of each data stream respectively; calculating the offset between each original data stream and the master clock sequence, and generating a time compensation parameter; determining a weight coefficient according to the sensor type, and using the weight coefficient to correct the time compensation parameter; performing time alignment processing on each data stream according to the corrected compensation parameter, and forming a time-aligned multi-source data stream; Step S12, receiving time-aligned multi-source data streams, and calculating the signal-to-noise ratio, data integrity ratio and stability ratio of each data stream respectively; The three ratios are weighted according to the preset weight coefficients to obtain the initial quality score; Calculate the degree of fluctuation of the quality score and generate a fluctuation penalty factor; The initial quality score is corrected using the fluctuation penalty factor to obtain the final quality score of each data stream; Generate a credibility score for the data flow based on the final quality score, and attach the credibility score to the corresponding data flow; Obtain preset data prediction model parameters and use the prediction model to predict the data flow; Calculate the difference between the actual value and the predicted value to generate a residual sequence; calculate the rate of change and local volatility of the residual sequence; generate anomaly scores based on the rate of change and volatility, and attach the anomaly scores to the corresponding data streams; Get a data stream with quality scores and anomaly scores; Step S13, reading a data stream with a quality score and anomaly score; extracting frequency components of the data stream to generate time-frequency spectrum data; calculating the frequency gradient of the time-frequency spectrum data to generate frequency smoothing parameters; constructing adaptive filtering parameters according to the time-frequency spectrum amplitude and the frequency smoothing parameters; filtering the data using the adaptive filtering parameters to obtain a denoised data stream; Calculate the information entropy, time-varying index and correlation between features of each feature component in the data stream; generate feature importance scores based on the information entropy and time-varying index; determine the feature weight coefficient based on the importance score, and perform dimensionality reduction on the feature space; obtain the feature data stream after dimensionality reduction; Step S14, reading the feature data stream after dimensionality reduction, stratifying the feature data stream according to the data type to obtain multi-layer feature data; calculating the quality score and information score of each layer of feature data, generating inter-layer weight coefficients; using the inter-layer weight coefficients to fuse the multi-layer feature data to obtain a fused feature data stream; Calculate the entropy value and variability of the feature data, and divide the data into multiple data blocks according to the entropy value and variability; calculate the quality score of each data block, and determine the compression ratio according to the quality score; The data block is compressed and encoded using a determined compression ratio to obtain compressed and stored block data, which is recorded as a compressed data block.

3. The method for inspecting buried water supply penstock defects according to claim 2 is characterized in that: The step S2 is specifically as follows: Step S21, read the compressed data block, and perform decompression processing to obtain an initial feature vector; Calculate the time gradient data of the feature vector and obtain the historical feature data; generate the feature enhancement coefficient based on the time gradient data and the historical feature data; The initial feature vector is enhanced by using the feature enhancement coefficient to obtain an enhanced feature vector; the covariance data between the feature components in the enhanced feature vector is calculated; the change rate data of each feature component is obtained, and the dynamic difference value between the features is calculated; the feature correlation strength matrix is ​​generated according to the covariance data and the dynamic difference value; Set multi-layer correlation thresholds, group features according to the correlation strength matrix, and output hierarchical feature groups; Step S22, reading the hierarchical feature group; Set multiple observation time scales and extract feature sequences at each time scale; Calculate the characteristic sequence variability at each time scale and generate a smoothing factor; Construct multi-scale state trajectories based on feature sequences and smoothing factors; Perform trend analysis on the state trajectory of each time scale to obtain the prediction data at each scale; Calculate the historical errors of predictions at each scale and generate prediction weight coefficients; fuse the prediction data at each scale according to the prediction weight coefficients to obtain the final state evolution sequence and prediction data; Step S23, read the final state evolution sequence, extract historical state data, calculate the state smoothing value, generate a dynamic benchmark model; calculate the deviation value between the current state and the benchmark model, and obtain the state change rate data; Calculate the state deviation score based on the deviation value and change rate data; obtain the pre-stored abnormal pattern data and calculate the matching probability between the current state and each abnormal pattern; extract the entropy value characteristics of the state sequence and generate uncertainty indicators; calculate the final probability score of each abnormal pattern based on the matching probability and uncertainty indicators, and output the abnormal score and pattern classification results; based on the comparative analysis of the predicted data and the current state data, cyclically optimize the sensitivity of abnormal detection; Step S24, receiving the hierarchical feature group, state evolution sequence, anomaly score and pattern classification result; obtaining spatial position data, and calculating local geometric features.

4. The method for inspecting buried water supply penstock defects according to claim 1 is characterized in that: The step S3 is specifically as follows: Step S31, extracting enhanced feature data from the pipeline state feature data packet, obtaining historical health response data, calculating the cumulative value of the historical response data, and generating a memory strength parameter; performing nonlinear mapping on the enhanced feature data according to the memory strength parameter to obtain an initial health indicator; Extract the information gain value, stability score and anomaly score of the feature data and generate the feature weight coefficient; Optimize the initial health index using the feature weight coefficient to obtain a multidimensional health index; obtain the pre-stored basic distribution function library and extract the fluctuation characteristics of the health status data; adjust the parameters of the basic distribution function according to the fluctuation characteristics to generate the initial probability density distribution; calculate the uncertainty score of the health status, correct the probability density distribution, and output the final health status distribution data and comprehensive health index; Step S32, reading the health status distribution data and the comprehensive health index; obtaining the pre-stored basic degradation mode library, extracting the change rate data of the health index; weighting the basic degradation mode according to the change rate data to obtain the initial degradation mode; Obtain historical degradation data and calculate time kernel function parameters; use time kernel function parameters to perform evolution analysis on the initial degradation mode and output degradation trend prediction results; Obtain environmental monitoring data, including parameters such as temperature, humidity, and pressure; Calculate the coupling strength value between environmental parameters to generate the environmental coupling coefficient; construct the acceleration factor model according to the environmental coupling coefficient to obtain the initial acceleration factor data; Calculate the fluctuation degree of the acceleration factor and generate the fluctuation correction parameter; The initial acceleration factor is corrected using the fluctuation correction parameter to obtain the final acceleration effect data; Combine the acceleration effect data with the degradation trend prediction results to output the revised degradation development prediction results; compare the prediction data with the actual degradation data, and optimize the identification of degradation patterns and the calculation of acceleration factors based on the comparison results; Step S33, obtaining health status distribution data and degradation development prediction results; obtaining pre-stored benchmark threshold data, and extracting a reference health index; Calculate the deviation between the current health index and the reference value, and generate the threshold adjustment parameters based on the deterioration development prediction results; Dynamically update the reference threshold according to the threshold adjustment parameter to obtain a multi-level dynamic threshold; Obtain uncertainty assessment results and calculate risk mapping parameters; According to the multi-level dynamic thresholds and risk mapping parameters, the risk level of the current status is evaluated and the risk level data is output; Obtain the pre-stored warning condition library and extract time series data; Calculate the advance amount of warning time and generate the time decay coefficient; Construct an early warning condition matrix based on risk level data and time decay coefficient; Obtain system intervention factor data and calculate intervention weight coefficients; use intervention weight coefficients to optimize warning conditions and output the final warning level and warning instructions; Step S34: read the risk level data and warning level, obtain the pre-stored cost database and benefit evaluation model; calculate the maintenance cost factor and benefit factor of each monitoring point, and generate an initial priority score; Obtain the score change rate data and calculate the dynamic adjustment parameters; modify the initial priority score according to the dynamic adjustment parameters to obtain the final maintenance priority ranking; Obtain the pre-stored maintenance plan library and extract plan feasibility data; screen the maintenance plans according to the risk level and priority score to obtain a set of candidate plans; calculate the benefit score, cost score and reliability score of the candidate plans and generate plan optimization parameters; The candidate solutions are comprehensively evaluated and ranked using solution optimization parameters, and the final recommended maintenance solution is output.

5. The method for inspecting buried water supply penstock defects according to claim 2 is characterized in that: The step S11 is specifically as follows: Step S111, reading the original data streams collected by the ultrasonic sensor, the video sensor, the wall thickness measurement sensor and the position sensor; extracting the sampling time sequence from the ultrasonic data stream to generate a reference time axis; generating a global clock sequence based on the reference time axis; Calculate the sampling interval for each sensor data stream and record the time interval sequence; Generate a timestamp identifier for each data stream according to the sampling interval sequence; Combine the timestamp identifier with the original data stream, and output the original data stream with the timestamp; Step S112, obtaining the original data stream with the timestamp; extracting the sampling time sequence of each data stream, calculating the time difference sequence with the reference time axis; performing statistical analysis on the time difference sequence to obtain the time offset distribution characteristics; Calculate initial compensation parameters according to time offset distribution characteristics; obtain pre-stored sensor response characteristic data, and correct the initial compensation parameters according to the response characteristic data; combine the corrected compensation parameters with the time difference sequence, and output a time compensation parameter set; Step S113, reading the time compensation parameter set and the original data stream with the timestamp; The signal amplitude sequence of each data stream is extracted respectively, and the signal strength index is calculated; the data stream weight coefficient is generated according to the signal strength index; the weight coefficient is multiplied by the time compensation parameter to obtain the weighted compensation parameter; the original data stream with timestamp is interpolated using the weighted compensation parameter to generate a unified sampling sequence; the data stream is reconstructed according to the unified sampling sequence, and a multi-source data stream with time alignment is output.

6. Buried water supply penstock disease inspection system, characterized by: include: A mobile platform, including a mobile track, a laser ranging device, a video acquisition device and an obstacle removal device; A data processing unit includes receiving and processing data acquired by the laser ranging device and the video acquisition device; Wherein, the data processing unit includes: at least one processor; and, a memory communicatively connected to at least one of the processors; wherein, The memory stores instructions that can be executed by the processor, and the instructions are used to be executed by the processor to implement the buried water supply pressure steel pipe defect inspection method described in any one of claims 1 to 5.

Images