Steel structure intelligent monitoring and adjusting method and system

Abstract

The invention discloses an intelligent monitoring and adjusting method and system for a steel structure, and the method comprises the steps: deploying a plurality of types of sensors at key parts of the steel structure, continuously collecting various types of physical quantity data, recognizing potential abnormal events in real time through a dynamic threshold detection mechanism, and responding to the abnormality of a certain sensor. The system automatically associates and extracts the time sequence data of other sensors associated with the spatial position of the sensor before and after the exception to form a multi-dimensional and time-space coupled data slice, and then performs time-space feature extraction on the data to extract the mode representation of the complex exception state; and based on this, accurate matching and diagnosis of the causes of the abnormal events are realized. Through the mode, the intelligent sensing and analysis capability of the health state of the complex, multi-source and multi-dimensional steel structure is improved, solid data support and decision basis are provided for subsequent scientific adjustment measures, and the operation safety and management intelligence level of large-scale steel structure engineering are better guaranteed.

Description

Technical Field

[0001] The present application relates to the field of intelligent monitoring and adjustment of steel structures, and more specifically, to a method and system for intelligent monitoring and adjustment of steel structures. Background Art

[0002] With the continuous expansion of modern construction and the increasing complexity of its structural forms, steel structures have been widely used in large public buildings, bridges, and industrial plants. While steel structures offer advantages such as high strength, lightweight, and a short construction period, they are susceptible to interference from various factors during their service life, including load fluctuations, environmental impacts, and fluctuations in construction quality. This makes it difficult to monitor the health of the structure in real time. Failure to detect and adjust anomalies in a timely manner can easily lead to safety accidents or economic losses. Therefore, developing an efficient and intelligent steel structure monitoring and adjustment solution is crucial for ensuring project safety, extending structural lifespan, and enhancing intelligent management.

[0003] At present, although some technical solutions for steel structure health monitoring have been proposed, such as threshold alarm systems based on single sensor data or methods relying on simple statistical analysis, these solutions generally have problems such as low intelligence, insufficient ability to recognize complex abnormal patterns, and lack of ability to trace the root causes of abnormalities. Specifically, existing methods can often only detect sensor data exceeding the limit, but cannot effectively distinguish between noise interference and real abnormal events. It is even more difficult to extract deep correlation features from multi-dimensional and multi-source data, and there is a lack of systematic understanding of the mechanism of abnormal state formation. In addition, in actual engineering applications, due to the complexity and variability of the construction process, it is difficult to accurately locate and diagnose the cause of the abnormality by relying solely on data from a single time point or a single sensor for analysis, which affects the scientific nature and effectiveness of subsequent adjustment measures.

[0004] Therefore, an optimized intelligent monitoring and adjustment method for steel structures is expected. Summary of the Invention

[0005] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a method and system for intelligent monitoring and adjustment of steel structures, which continuously collects various types of physical quantity data by deploying multiple types of sensors at key locations of steel structures, and uses a dynamic threshold detection mechanism to identify potential abnormal events in real time. In response to an abnormality in a certain sensor, the system automatically associates and extracts time series data of other sensors associated with its spatial position in the time periods before and after the abnormality, forming multi-dimensional, spatiotemporal coupled data slices, and then performs spatiotemporal feature extraction on these data to extract pattern representations of complex abnormal states, and based on this, achieves accurate matching and diagnosis of the causes of abnormal events. In this way, the intelligent perception and analysis capabilities of the health status of complex, multi-source, and multi-dimensional steel structures are improved, providing solid data support and decision-making basis for subsequent scientific adjustment measures, and better ensuring the operational safety and intelligent management level of large-scale steel structure projects.

[0006] According to one aspect of the present application, a method for intelligent monitoring and adjustment of a steel structure is provided, comprising:

[0007] Continuously monitor the sensor data of each sensor deployed on the steel structure, and perform abnormal event detection on the sensor data of each sensor based on dynamic threshold detection;

[0008] In response to detecting an abnormal event occurring at a certain sensor, automatically extracting time series data of all sensors related to the sensor within a preset time period before and after the abnormal event to obtain spatiotemporal data slices;

[0009] The spatiotemporal data slices are sorted according to the time dimension and sensor sample dimension to obtain a set of sensor data time series;

[0010] Performing spatiotemporal feature extraction on a collection of sensor data time series to obtain a representation vector of the abnormal state sensor data pattern;

[0011] Based on the occurrence timestamp of the anomaly, a list of anomaly-related construction activities is extracted from the structured construction log dataset;

[0012] The abnormality-related construction activity list and the abnormal state sensor data pattern representation vector are input into the pre-built abnormal pattern-root cause knowledge base for pattern matching to obtain a diagnosis list.

[0013] According to another aspect of the present application, a steel structure intelligent monitoring and adjustment system is provided, comprising:

[0014] An abnormal event detection module is used to continuously monitor the sensor data of each sensor deployed on the steel structure and perform abnormal event detection on the sensor data of each sensor based on dynamic threshold detection;

[0015] a spatiotemporal data slice extraction module for automatically extracting time series data of all sensors related to the sensor within a preset time period before and after the occurrence of the abnormal event in response to detecting an abnormal event occurring at a certain sensor to obtain a spatiotemporal data slice;

[0016] A data sorting module is used to sort the spatiotemporal data slices according to the time dimension and the sensor sample dimension to obtain a set of sensor data time series;

[0017] A spatiotemporal feature extraction module is used to extract spatiotemporal features from a set of sensor data time series to obtain a representation vector of an abnormal state sensor data pattern;

[0018] The exception-related construction activity list extraction module is used to extract the exception-related construction activity list from the structured construction log dataset based on the occurrence timestamp of the exception time;

[0019] The diagnosis result generation module is used to input the abnormality-related construction activity list and the abnormal state sensor data pattern representation vector into the pre-built abnormal pattern-root cause knowledge base for pattern matching to obtain a diagnosis list.

[0020] Compared with the existing technology, the present application provides a method and system for intelligent monitoring and adjustment of steel structures. By deploying multiple types of sensors at key locations of steel structures, the system continuously collects various types of physical quantity data and uses a dynamic threshold detection mechanism to identify potential abnormal events in real time. In response to an abnormality in a certain sensor, the system automatically associates and extracts time series data of other sensors associated with its spatial position in the time periods before and after the abnormality, forming multi-dimensional, spatiotemporal coupled data slices. Subsequently, spatiotemporal feature extraction is performed on these data to extract pattern representations of complex abnormal states, and based on this, accurate matching and diagnosis of the causes of abnormal events are achieved. In this way, the intelligent perception and analysis capabilities of the health status of complex, multi-source, and multi-dimensional steel structures are improved, providing solid data support and decision-making basis for subsequent scientific adjustment measures, and better ensuring the operational safety and intelligent management level of large-scale steel structure projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above and other purposes, features, and advantages of the present application will become more apparent through a more detailed description of the embodiments of the present application in conjunction with the accompanying drawings. The accompanying drawings are intended to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] Figure 1 Flowchart of a method for intelligent monitoring and adjustment of steel structures according to an embodiment of the present application;

[0023] Figure 2 Schematic diagram of data flow of the steel structure intelligent monitoring and adjustment method according to an embodiment of the present application;

[0024] Figure 3 4 is a block diagram of a steel structure intelligent monitoring and adjustment system according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0026] As used in this application and the claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not intended to refer to the singular but may include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0027] Although the present application makes various references to certain modules in the system according to embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are illustrative only, and different aspects of the system and method can use different modules.

[0028] Flowcharts are used in this application to illustrate the operations performed by the systems according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the various steps may be processed in reverse order or simultaneously, as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0029] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described herein.

[0030] In the technical solution of this application, a method for intelligent monitoring and adjustment of steel structures is proposed. Figure 1 Flowchart of the intelligent monitoring and adjustment method for steel structures according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the steel structure intelligent monitoring and adjustment method according to the embodiment of the present application. Figure 1 and Figure 2As shown, the intelligent monitoring and adjustment method for steel structures according to the embodiment of the present application includes the following steps: S1, continuously monitoring the sensor data of each sensor deployed on the steel structure, and performing abnormal event detection on the sensor data of each sensor based on dynamic threshold detection; S2, in response to detecting an abnormal event of a certain sensor, automatically extracting the time series data of all sensors related to the sensor within a preset time period before and after the abnormal event to obtain spatiotemporal data slices; S3, sorting the spatiotemporal data slices according to the time dimension and the sensor sample dimension to obtain a set of sensor data time series; S4, performing spatiotemporal feature extraction on the set of sensor data time series to obtain an abnormal state sensor data pattern representation vector; S5, extracting a list of abnormal related construction activities from a structured construction log data set based on the timestamp of the abnormal time; S6, inputting the list of abnormal related construction activities and the abnormal state sensor data pattern representation vector into a pre-built abnormal pattern-root cause knowledge base for pattern matching to obtain a diagnosis list.

[0031] In particular, the S1 continuously monitors the sensor data of each sensor deployed on the steel structure, and performs abnormal event detection on the sensor data of each sensor based on dynamic threshold detection. In response to the reading of the sensor data of each sensor exceeding the limit or the abnormal rate of change, it is determined that an abnormal event has occurred in the sensor. It should be understood that the steel structure will be affected by multiple factors such as load changes, ambient temperature fluctuations, and differences in construction quality during actual service, resulting in its health status being in dynamic change at all times. Traditional methods that rely solely on a single threshold or static analysis are often difficult to adapt to such complex and changeable working conditions, and are prone to false alarms or omissions. Therefore, through the continuous collection of sensor data and the setting of dynamic thresholds, the degree of deviation between the current state of the structure and the historical operating rules can be reflected in real time, effectively distinguishing normal fluctuations from real anomalies.

[0032] It is worth mentioning that the sensor data includes but is not limited to parameters such as strain, acceleration, temperature and displacement. Specifically, strain data can directly reflect the stress and damage of the steel structure and is an important indicator for judging load anomalies; acceleration data is used to capture vibration characteristics and is extremely sensitive to fatigue damage and sudden events; temperature data helps to eliminate non-structural interference caused by environmental changes and improve detection accuracy; displacement data can accurately reveal the overall or local deformation of the structure, providing key support for safety assessment. In the technical solution of this application, by continuously monitoring the sensor data of each sensor deployed on the steel structure and detecting abnormal events based on the dynamic threshold detection of the sensor data of each sensor, early warning signals such as reading exceeding the limit or abnormal change rate can be captured in time, achieving a rapid response to potential risks and providing a reliable basis for subsequent in-depth data analysis and tracing the root cause of the anomaly. These multi-source and multi-dimensional data are comprehensively analyzed through the dynamic threshold mechanism, enabling the system to flexibly adapt to the health monitoring needs under different working conditions, greatly improving the accuracy and timeliness of anomaly detection, thereby effectively supporting the safe operation and intelligent maintenance management of the steel structure throughout its life cycle.

[0033] Specifically, in step S2, in response to detecting an abnormal event at a particular sensor, time series data from all sensors related to that sensor within a preset time period before and after the abnormal event is automatically extracted to obtain a spatiotemporal data slice. It should be understood that steel structures have complex spatial relationships and variable force paths. A sensor anomaly is often not just a local phenomenon but may be closely related to changes in data from surrounding structural units or other sensors. Single-point data alone cannot reveal the true anomaly and its scope. Therefore, it is necessary to determine and extract data from all related sensors based on sensor layout information, combined with their spatial locations and interconnected relationships, to fully capture the full picture of changes in the structural health state. Therefore, to form multi-dimensional, spatiotemporally coupled data slices and provide a rich data foundation for subsequent deep time series feature analysis and complex pattern recognition, the technical solution of this application, in response to detecting an abnormal event at a particular sensor, time series data from all sensors related to that sensor within a preset time period before and after the abnormal event is automatically extracted to obtain a spatiotemporal data slice. All sensors related to that sensor are determined based on sensor layout information, which describes the sensors' spatial locations within the steel structure and their interconnected relationships. This allows for effective differentiation between isolated noise interference and true structural anomalies, and by comparing the linked changes in data from multiple related points over the same period, more accurately locates the source of the anomaly and its propagation path. This significantly improves the accuracy and reliability of the system's anomaly diagnosis under complex operating conditions.

[0034] In particular, S3 organizes the spatiotemporal data slices according to the time dimension and sensor sample dimension to obtain a set of sensor data time series. It should be understood that the data collected by steel structure sensors is typically multidimensional and complex, covering continuous observation records from multiple sensors at different time points. Direct analysis of unorganized spatiotemporal data slices, which contain a large amount of raw, multi-source, and spatiotemporally intertwined data, will result in high computational complexity, difficulty in feature extraction, and difficulty in capturing temporal patterns. Therefore, in the technical solution of the present application, the spatiotemporal data slices are organized according to the time dimension and sensor sample dimension to obtain a set of sensor data time series. The resulting set of sensor data time series not only fully reflects the dynamic response changes of each relevant sensor before and after the abnormal event, but also reflects the interactive relationship between sensors in time and space, helping to capture cross-sensor correlation features and abnormal propagation patterns. In this way, the ability to understand temporal dependencies and spatial correlations is significantly improved, ensuring the accuracy and reliability of anomaly identification, and ultimately laying a solid data foundation for subsequent abnormal pattern aggregation analysis and root cause diagnosis.

[0035] In particular, S4 performs spatiotemporal feature extraction on the set of sensor data time series to obtain an abnormal state sensor data pattern representation vector. In an embodiment of the present application, first, a bidirectional LSTM model is used to extract temporal features from each sensor data time series in the set of sensor data time series to obtain a set of sensor data temporal pattern feature encoding vectors. It should be understood that abnormal events in steel structures are often not caused by data mutations at a single time point, but rather by the evolution of temporal patterns formed by load accumulation, material fatigue, or long-term effects of environmental factors. Traditional unidirectional LSTMs can only capture the influence of historical time steps. By simultaneously learning historical and future information from both the forward and reverse directions of the time series, bidirectional LSTMs can more comprehensively reveal the potential temporal patterns and change trends in the time series, effectively avoiding feature loss and misjudgment caused by information fragmentation. Therefore, in the technical solution of the present application, a bidirectional LSTM model is used to extract temporal features from each sensor data time series in the set of sensor data time series to obtain a set of sensor data temporal pattern feature encoding vectors. Specifically, steel structure sensor data includes multi-dimensional physical quantities such as strain, acceleration, temperature, and displacement. Changes in these quantities are not only influenced by previous operating conditions but may also reflect subsequent environmental or loading changes. The bidirectional LSTM's ability to correlate states from moment to moment greatly enhances the ability to express the temporal features of abnormal states. Temporal feature extraction based on the bidirectional LSTM model enhances the robustness and intelligence of the entire intelligent monitoring approach under complex operating conditions, enabling the system to accurately capture subtle and complex abnormal changes.

[0036] Furthermore, a temporal feature spatial aggregation analysis is performed on the set of sensor data temporal pattern feature encoding vectors to obtain an abnormal state sensor data pattern representation vector. It should be understood that the essence of abnormal steel structure states is often jointly characterized by the dynamic patterns of multi-sensor spatiotemporal coupling. Although the bidirectional LSTM has extracted the local evolution patterns of each sensor's temporal pattern data, as a mechanical transmission system, the propagation of abnormalities in steel structures has significant spatial correlation and time delay. For example, a stress anomaly at a node may trigger a chain reaction at adjacent nodes within a subsequent time window. A single sensor feature encoding cannot fully capture this cross-temporal and spatial synergistic effect. Simple feature splicing or statistical aggregation methods in the prior art easily ignore the spatial topological relationships of the sensor network and the dynamic characteristics of anomaly propagation, resulting in key correlation features being submerged in noise. Therefore, to construct a high-dimensional representation that can simultaneously encode the global structural state ground state and the local abnormal excited state, the technical solution of this application performs a temporal feature spatial aggregation analysis on the set of sensor data temporal pattern feature encoding vectors to obtain an abnormal state sensor data pattern representation vector. Specifically, abnormal events in steel structures usually manifest as the deviation from the overall mechanical equilibrium state (ground state characteristics) and the superposition of abnormal fluctuations in specific node groups (excited state characteristics).

[0037] Specifically, clustering and self-attention mechanisms are used to fuse global commonalities (such as the ground state of the overall structural stress distribution) with local dynamic excitation features (such as abnormal vibration modes at specific nodes), forming a composite representation vector that reflects the overall healthy ground state of the steel structure while highlighting the significance of local abnormal excited states. In this process, K-means clustering is used to extract common primitives from sensor temporal features. The self-attention mechanism is then used to model the global correlations between these primitives, forming an aggregate encoding vector of collective ground state features that represents the overall health of the structure. Specifically, a geometric boundary optimization technique based on an induced metric performs spatial topological calibration on individual feature vectors, calculates their excited state deviations from the ground state features, and enhances the contributions of key abnormal nodes through significance weight modulation. For example, in a local buckling event in a truss structure, the ground state features reflect the overall load distribution, while the excited state features highlight the abnormal deformation patterns at the buckling node and its adjacent supports. The superposition of the two accurately characterizes the mechanical path of buckling. By imitating the principle of wave function superposition, high-dimensional spatiotemporal features are compressed into low-dimensional representations with higher information density, so that subsequent root cause diagnosis can not only identify systemic risks (such as construction load exceeding the limit) based on global ground state information, but also capture local hidden dangers (such as weld cracking) through excited state significance. In actual engineering scenarios with frequent construction interference and complex noise, the directionality of abnormal pattern representation to the real damage mechanism of the structure is improved, avoiding the misjudgment problem of traditional methods mistakenly identifying construction vibration noise as structural damage due to ignoring spatiotemporal correlation.

[0038] Specifically, first, the set of sensor data temporal pattern feature encoding vectors is input into the K-Means clustering network to obtain K sensor data temporal pattern cluster center encoding vectors. It should be understood that although the bidirectional LSTM has extracted the temporal dynamic characteristics of each sensor, the steel structure is a mechanical conduction network, and its abnormal events are usually manifested as coordinated changes of multiple sensor data within a specific time and space range. For example, structural deformation caused by excessive load often forms similar data drift patterns in multiple associated sensors on the mechanical conduction path. Therefore, in order to structurally summarize the scattered temporal feature encoding vectors and extract the ground-state features that characterize the common behavior of the system, in the technical solution of the present application, the set of sensor data temporal pattern feature encoding vectors is input into the K-Means clustering network to obtain K sensor data temporal pattern cluster center encoding vectors.

[0039] Through K-Means clustering, it is possible to automatically identify subsets of sensor data with similar evolutionary patterns from high-dimensional feature spaces, thereby capturing potential structural anomaly primitives. Specifically, in steel structure scenarios, the time series data collected by the sensor network contains both steady-state fluctuations under normal operating conditions (such as thermal expansion and contraction caused by day and night temperature differences) and implicit characteristics of the embryonic stage of abnormal events. K-Means iteratively optimizes the intra-cluster distance and classifies sensor feature vectors with similar evolutionary patterns into the same cluster. The cluster center encoding vector essentially reflects the centroid of a certain type of common behavior pattern. For example, in a large steel truss structure, sensors located on the same force path may show a synchronous stress increase trend when an anomaly occurs. The clustering process can automatically discover such spatial correlation-driven pattern commonalities, laying the foundation for the subsequent construction of global ground-state features.

[0040] In a specific example of the present application, a set of sensor data temporal pattern feature encoding vectors is input into a K-Means clustering network to obtain K sensor data temporal pattern cluster center encoding vectors using the following formula; wherein, the formula is:

[0041] X={x1,x2,...,x i ,...,x n}∈R n×d

[0042] C=K_Means(X)={c1,c2,...,c k}

[0043] Where X is the set of characteristic encoding vectors of the sensor data temporal pattern, x1, x2, x i ,x nare the first, second, i-th and n-th sensor data temporal pattern feature coding vectors in the set of the sensor data temporal pattern feature coding vectors, n represents the number of sensor data temporal pattern feature coding vectors, d represents the dimension of the sensor data temporal pattern feature coding vector, K_Means(·) represents the K-Means clustering function, C is the K sensor data temporal pattern clustering center coding vector, c1, c2, c k They are respectively the first, second and kth sensor data time series pattern cluster center encoding vectors among the K sensor data time series pattern cluster center encoding vectors.

[0044] Next, the K sensor data temporal pattern cluster center encoding vectors are subjected to collective ground state feature aggregation based on the self-attention mechanism to obtain the sensor data temporal pattern collective ground state feature aggregation encoding vector. It should be understood that although K-Means clustering has extracted the common primitives of local temporal features, the mechanical conduction characteristics between steel components require that the system must capture potential correlations across cluster centers. For example, sensor clusters located on different force paths may exhibit asynchronous but causally related strain fluctuations due to the same load abnormality event, and a simple cluster center cannot characterize such dynamic coupling relationships across spatial domains. Therefore, in order to construct a global ground state feature vector that can characterize the overall health status of the steel structure, in the technical solution of the present application, the K sensor data temporal pattern cluster center encoding vectors are subjected to collective ground state feature aggregation based on the self-attention mechanism to obtain the sensor data temporal pattern collective ground state feature aggregation encoding vector.

[0045] Among them, the self-attention mechanism can model the implicit topological associations of stress wave propagation paths in steel structures by establishing a fully connected association network between each cluster center, breaking through the limitations of local clustering vision. The self-attention mechanism performs contextual enhancement on K cluster centers, allowing each cluster center to incorporate the structural state information carried by other cluster centers while retaining its own pattern characteristics. For example, in a large spatial grid structure, the sensor cluster centers of the roof node and the support column node may represent the steady-state characteristics of different subsystems respectively, and the self-attention weight matrix can automatically identify the load transfer dependency between the two, fusing the scattered local primitives into a ground state encoding vector reflecting the overall mechanical equilibrium state. This global perspective of the ground state characteristics can effectively distinguish between systemic fluctuations under normal working conditions (such as global vibration caused by wind load) and abnormal disturbances caused by local damage.

[0046] In a specific example of the present application, the following formula is used to perform collective ground state feature aggregation based on the self-attention mechanism on the K sensor data temporal pattern cluster center encoding vectors to obtain the sensor data temporal pattern collective ground state feature aggregation encoding vector; wherein, the formula is:

[0047] Q=CW Q

[0048] K=CW k

[0049] V=CW

[0050]

[0051] g=layerNorm(C+Attention(Q,K,V))

[0052] Among them, W Q 、W k and W v represents the linear transformation matrix, Q represents the query matrix, K represents the key matrix, V represents the value matrix, Attention(·) represents the self-attention mechanism, Softmax(·) represents the Softmax function, layerNorm(·) represents the normalization function, and g is the ground state feature aggregation encoding vector of the sensor data temporal pattern set.

[0053] Then, based on the sensor data time series pattern set ground state feature aggregation coding vector, the excited state coding vector of each sensor data time series pattern feature coding vector in the set of sensor data time series pattern feature coding vectors relative to the sensor data time series pattern set ground state feature aggregation coding vector is calculated to obtain a set of sensor data time series pattern feature compensation excited state coding vectors. In an embodiment of the application, first, based on the sensor data time series pattern set ground state feature aggregation coding vector, the excited state coding vector of each sensor data time series pattern feature coding vector in the set of sensor data time series pattern feature coding vectors relative to the sensor data time series pattern set ground state feature aggregation coding vector is calculated to obtain a set of initial sensor data time series pattern feature compensation excited state coding vectors. It should be understood that the essence of abnormal events in steel structures is usually manifested as a combined effect of overall mechanical state deviation and local node-specific disturbance. Although the global ground state feature has captured the steady-state characteristics of the structural system, hidden damage (such as weld microcracks or bolt loosening) often only causes abnormal fluctuations that deviate from the ground state at specific sensor nodes. For example, in the health monitoring of large steel bridges, the overall thermal expansion and contraction effects caused by changes in ambient temperature will be characterized by the ground state characteristics as systematic strain fluctuations, while the strain concentration caused by local fatigue cracks needs to be specifically captured through excited state coding. Because traditional methods do not decouple global and local features, such local anomalies are easily submerged in overall fluctuations. Therefore, in order to explicitly separate the common ground state components and specific excited state components in the sensor timing characteristics, a dual-channel analysis framework for anomaly detection is constructed. In the technical solution of the present application, based on the ground state feature aggregation coding vector of the sensor data timing pattern set, the excited state coding vector of each sensor data timing pattern feature coding vector in the set of sensor data timing pattern feature coding vectors is calculated relative to the ground state feature aggregation coding vector of the sensor data timing pattern set to obtain a set of initial sensor data timing pattern feature compensation excited state coding vectors.

[0054] Here, by calculating the difference between each sensor's time-series encoding vector and its ground-state characteristics, the system can quantify the degree of deviation of each sensor node from the overall health state of the structure. This spatially correlated excited-state feature distribution can not only effectively distinguish random noise from real structural damage (such as distinguishing global excited-state modes caused by transient vibrations of lifting equipment from localized persistent excited-state modes caused by crack propagation), but can also reverse-infer the root cause of the anomaly through the spatial propagation path of the excited-state vector, providing a physically interpretable characteristic basis for construction quality tracing, significantly enhancing the reliability of health status diagnosis and decision-making support capabilities for complex steel structure systems.

[0055] In a specific example of the present application, the excited state coding vector of each sensor data timing pattern feature coding vector in the set of sensor data timing pattern feature coding vectors relative to the ground state feature aggregation coding vector of the sensor data timing pattern set is calculated using the following formula to obtain a set of initial sensor data timing pattern feature compensation excited state coding vectors; wherein, the formula is:

[0056] e i =Sigmoid(W e [x i ;g]+b e )⊙(x i -g)

[0057] Where Sigmoid(·) represents the Sigmoid function, W e represents the weight matrix, b e represents the bias term, [·;·] represents vector concatenation, ⊙ represents point multiplication by position, e i is x i The corresponding initial sensor data timing pattern characteristic compensates the excited state coding vector.

[0058] Next, each sensor data temporal pattern feature encoding vector in the set of sensor data temporal pattern feature encoding vectors is subjected to geometric boundary optimization based on an induced metric to obtain a set of optimized sensor data temporal pattern feature encoding vectors. Furthermore, based on the set of optimized sensor data temporal pattern feature encoding vectors, the set of initial sensor data temporal pattern feature compensation excited state encoding vectors is modified to obtain a set of sensor data temporal pattern feature compensation excited state encoding vectors. It should be understood that although the encoding vectors of sensor temporal features extracted through a bidirectional LSTM can represent temporal dynamics, when the excited state encoding is directly calculated, redundant noise and irrelevant features in the high-dimensional space may distort the representation of abnormal patterns. Therefore, in a preferred example of the present application, first, each sensor data timing pattern feature coding vector in the set of sensor data timing pattern feature coding vectors is subjected to geometric boundary optimization based on the induced metric to obtain a set of optimized sensor data timing pattern feature coding vectors; then, based on the set of optimized sensor data timing pattern feature coding vectors, the set of initial sensor data timing pattern feature compensation excited state coding vectors is corrected to obtain a set of sensor data timing pattern feature compensation excited state coding vectors.

[0059] That is, by introducing the geometric boundary constraints of the induced metric, the high-dimensional time series features are mapped to a low-dimensional manifold space that is compatible with the physical topology of the steel structure, thereby separating real anomalies from noise interference. Specifically, the induced metric embeds the spatial connection topology of the sensor nodes (such as the mechanical conduction path of adjacent components) into the feature optimization process by defining the geometric mapping relationship between the ground state space and the original feature space. For example, in the health monitoring of steel bridges, the strain anomaly of a certain support node will be transmitted along the longitudinal direction of the main beam. The geometric boundary optimization constrains the projection direction of the feature vector in the low-dimensional space, so that the excited state encoding can reflect the gradient attenuation pattern of such anomalies along the structural force transmission path, rather than the random fluctuation of isolated nodes. The set of optimized sensor data time series pattern feature encoding vectors can effectively suppress the global feature offset caused by ambient temperature change or transient load by correcting the geometric distribution of the initial excited state encoding, while enhancing the sensitivity to local persistent anomalies (such as weld fatigue cracks). This geometric optimization mechanism based on structural physical properties provides feature inputs that combine spatial topological correlation and temporal evolution laws for subsequent root cause tracing, significantly improving the accuracy and interpretability of abnormality diagnosis in complex steel structure systems.

[0060] In this example, the set of initial sensor data timing pattern feature compensation excited state coding vectors is modified by the following formula to obtain the set of sensor data timing pattern feature compensation excited state coding vectors: wherein the formula is;

[0061] e i =Sigmoid(W e [x i ;g]+b e )⊙(x i -g)

[0062] ξ={e1,e2,...,e i ,...,e n}

[0063]

[0064] e′ i =x′ i ⊙e i

[0065] Where ξ is the set of the initial sensor data temporal pattern characteristic compensation excited state coding vectors, e1, e2, e i ,e n are the first, second, i-th and n-th initial sensor data timing pattern feature compensation excited state coding vectors in the set of the initial sensor data timing pattern feature compensation excited state coding vectors, ω represents the modulation coefficient, x′ iTo optimize the sensor data temporal pattern feature encoding vector, e′ i The i-th sensor data timing pattern feature compensation excited state coding vector in the set of sensor data timing pattern feature compensation excited state coding vectors is obtained.

[0066] Then, the excited state significance modulation weight factor of each sensor data temporal pattern feature compensation excited state coding vector in the set of sensor data temporal pattern feature compensation excited state coding vectors is calculated to obtain a set of sensor data temporal pattern feature compensation excited state significance modulation weight factors. It should be understood that the abnormal features in the sensor data often overlap with the noise interference in the excited state coding space. The complex environmental loads (such as wind vibration, temperature change) and construction activities (such as hoisting vibration) of the steel structure will cause frequent instantaneous fluctuations in the sensor data. Although these fluctuations appear as characteristics deviating from the ground state in the excited state coding, their physical meaning may only represent normal working condition disturbances rather than real structural damage. Therefore, in order to construct a dynamic feature importance evaluation mechanism, by quantifying the abnormal contribution of each excited state coding vector, the intelligent distinction between abnormal signals and noise is achieved. In the technical solution of the present application, the excited state significance modulation weight factor of each sensor data temporal pattern feature compensation excited state coding vector in the set of sensor data temporal pattern feature compensation excited state coding vectors is calculated to obtain a set of sensor data temporal pattern feature compensation excited state significance modulation weight factors.

[0067] Specifically, the sensor data temporal pattern feature compensation excited state significance modulation weight factor is dynamically calculated based on the spatiotemporal consistency, evolutionary persistence, and spatial topological correlation of the excited state encoding vector. In one example, in the monitoring of industrial plant trusses, the excited state encoding generated by a weld crack at a node will show the characteristics of increasing over time and spreading along the force transmission path of the component, and its weight factor will be dynamically increased; while the transient fluctuations of isolated nodes caused by temporary construction equipment are given low weights due to the lack of spatiotemporal correlation, and are thus effectively suppressed in the subsequent feature aggregation stage. This intelligent weighting mechanism not only reduces the false alarm rate, but also provides feature input with spatiotemporal topological explanatory power for root cause tracing, thereby improving the decision-making reliability of the steel structure health monitoring system.

[0068] In a specific example of the present application, the excited-state significance modulation weight factor of each sensor data timing pattern feature compensation excited-state coding vector in the set of sensor data timing pattern feature compensation excited-state coding vectors is calculated using the following formula to obtain a set of sensor data timing pattern feature compensation excited-state significance modulation weight factors; wherein, the formula is:

[0069]

[0070] A={α1,α2,...,αi ,...,α n}

[0071] Where exp(·) represents the exponential function, w T represents the excited state saliency modulation weight matrix, ReLU(·) represents the ReLU function, W s represents the fully connected layer, α i for e i The corresponding sensor data temporal pattern feature compensation excited state significant modulation weight factor, A is the set of sensor data temporal pattern feature compensation excited state significant modulation weight factors, α1, α2, α i ,α n They are respectively the 1st, 2nd, i-th and n-th sensor data timing pattern feature compensation excited state significant modulation weight factors in the set of sensor data timing pattern feature compensation excited state significant modulation weight factors.

[0072] Furthermore, based on a set of sensor data temporal pattern feature compensation excited state significance modulation weight factors, the set of sensor data temporal pattern feature compensation excited state code vectors is weighted modulated to obtain a set of sensor data temporal pattern feature compensation excited state significant code vectors. It should be understood that abnormal events in steel structures often coexist with complex environmental disturbances. The excited state codes in the sensor data may reflect real component damage (such as weld cracking or loose bolts) or may contain transient interference signals such as construction vibration and temperature fluctuations. Traditional methods lack a feature importance assessment mechanism and have difficulty distinguishing between these two essentially different patterns. Therefore, to construct a dynamic feature enhancement mechanism, the excited state codes are adaptively strengthened or suppressed by weight factors. In the technical solution of the present application, based on a set of sensor data temporal pattern feature compensation excited state significance modulation weight factors, the set of sensor data temporal pattern feature compensation excited state code vectors is weighted modulated to obtain a set of sensor data temporal pattern feature compensation excited state significant code vectors. Through the weighted modulated significant code vectors, the system can focus on abnormal patterns with structural mechanical significance. This feature enhancement mechanism not only reduces the false alarm rate caused by environmental noise, but also provides high-fidelity input data for subsequent root cause analysis by retaining the spatiotemporal trajectory characteristics of anomaly propagation (such as the diffusion pattern of damage along the force transmission path).

[0073] In a specific example of the present application, the set of sensor data temporal pattern feature compensation excited state coding vectors is weighted modulated using the following formula to obtain a set of sensor data temporal pattern feature compensation excited state saliency coding vectors; wherein, the formula is:

[0074]

[0075]

[0076] in, for e i and α i The corresponding sensor data temporal pattern characteristic compensation excited state significant encoding vector, is a set of significant coding vectors for compensating excited states for the temporal pattern feature of the sensor data, They are respectively the 1st, 2nd, i-th and n-th sensor data timing pattern feature compensation excited state significant coding vectors in the set of the sensor data timing pattern feature compensation excited state significant coding vectors.

[0077] Subsequently, the set of significant coding vectors of the sensor data temporal pattern feature compensation excited state and the aggregate coding vector of the sensor data temporal pattern set ground state features are input into the feature gain superposition network to obtain the abnormal state sensor data pattern representation vector. It should be understood that the complete representation of the abnormal state of the steel structure needs to capture the global steady-state characteristics and local abnormal disturbances of the system at the same time. The mechanical properties of the steel structure determine that its health state is affected by both the overall load distribution and the chain reaction caused by local node damage. If only relying on a single feature dimension, it is easy to cause the coupling characteristics of the overall deformation and local damage to be lost, resulting in the inability to distinguish between systemic load exceeding the limit and hidden component damage during subsequent root cause tracing. Therefore, in order to construct a multi-scale fusion feature representation vector, in the technical solution of the present application, the set of significant coding vectors of the sensor data temporal pattern feature compensation excited state and the aggregate coding vector of the sensor data temporal pattern set ground state features are input into the feature gain superposition network to obtain the abnormal state sensor data pattern representation vector.

[0078] Specifically, the ground-state feature aggregation encoding vector of the sensor data temporal pattern aggregate carries a steady-state portrait of the structural system (e.g., the temperature-stress response pattern under normal operating conditions), while the excited-state significant encoding vector of the sensor data temporal pattern feature compensation focuses on abnormal events that deviate from the steady state (e.g., progressive strain shifts caused by weld cracks). The feature gain superposition network, through a dynamic weight allocation mechanism, fuses these two into a composite representation that combines wide-area coverage with local sensitivity. For example, in bridge health monitoring, the superimposed feature vector can reflect both the overall vibration mode of the entire bridge under traffic loads (dominated by the ground state) and the localized cable tension anomaly caused by corrosion in a specific hanger anchorage area (dominated by the excited state), providing a multi-dimensional basis for damage localization. This multi-scale feature fusion mechanism not only improves the ability to analyze complex anomaly patterns but also, by preserving the spatiotemporal trajectory characteristics of anomaly propagation (e.g., the diffusion gradient of damage along the force transmission path), provides a decision-making basis for structural health assessment that is both physically meaningful and statistically significant, thereby reducing the risk of underreporting and improving the targeting of maintenance measures.

[0079] In a specific example of the present application, the set of sensor data temporal pattern feature compensation excited state significant coding vectors and the sensor data temporal pattern set ground state feature aggregation coding vector are input into the feature gain superposition network to obtain the abnormal state sensor data pattern representation vector according to the following formula; wherein, the formula is:

[0080]

[0081] Where MLP(·) represents a multi-layer perceptron, v f is a representation vector of the abnormal state sensor data pattern.

[0082] In particular, the S5 extracts a list of abnormal related construction activities from the structured construction log data set based on the timestamp of the occurrence of the abnormal time. It should be understood that the occurrence of abnormal conditions in steel structures is often closely related to specific operations during the construction process. Different construction activities such as welding, hoisting, and reinforcement may have different types and degrees of impact on the structure. By utilizing the time nodes of abnormal events, the system can automatically match and retrieve relevant construction records of the corresponding time period in the construction log to ensure that the associated construction activity information is highly timely and targeted. Therefore, in order to achieve the spatiotemporal association of data, in the technical solution of the present application, a list of abnormal related construction activities is extracted from the structured construction log data set based on the timestamp of the occurrence of the abnormal time.

[0083] Specifically, when a sensor detects an anomaly, the system quickly identifies all construction activities within the time window that may have affected the anomaly by matching the anomaly's timestamp, creating a list of related construction activities. By integrating this data with construction activity data, it provides more comprehensive context for subsequent adjustment decisions, guiding the development of effective and rational intervention measures. This ensures that the entire building steel structure monitoring and adjustment system not only possesses high sensitivity in anomaly detection but also demonstrates strong assurance and support capabilities in anomaly tracking and solution development.

[0084] Specifically, in S6, the list of abnormal construction activities and the representation vectors of abnormal sensor data patterns are input into a pre-built abnormal pattern-root cause knowledge base for pattern matching to generate a diagnosis list. It should be understood that the occurrence of abnormal conditions in steel structures often involves both the physical response patterns of the structure itself and is closely related to the specific operations and environmental conditions of the construction activities. Relying solely on sensor data cannot fully reveal the causes of abnormalities. The rich operational information in construction logs provides the context and possible triggers for the abnormalities. By combining dynamically captured abnormal sensor data patterns with construction activity information into a knowledge base system, deep fusion and intelligent matching of multi-dimensional, multi-source information can be achieved, building a bridge between structural response and construction behavior. The abnormal pattern-root cause knowledge base, based on a wealth of historical case studies, experimental data, and expert experience, integrates multiple abnormal patterns and corresponding root cause information in a structured manner, enabling accurate identification and classification of new abnormal data. This approach enhances the processing capabilities of complex multi-source data, enables the coordinated interpretation of construction behavior and structural status, and significantly reduces the risk of misdiagnosis.

[0085] It's worth noting that after obtaining the diagnostic list, the "intelligent monitoring and adjustment" system for steel structures enters the decision-making and execution phase, transforming the aforementioned complex data analysis and root cause diagnosis results into specific, scientific, and efficient adjustment measures, achieving closed-loop management of the structural health status. During implementation, the system first intelligently analyzes the anomaly type, impact range, and corresponding root cause in the diagnostic list. For example, if the diagnostic results indicate a localized strain anomaly in a certain area due to fluctuations in welding quality, the system automatically retrieves historical construction activities, material batches, and the spatiotemporal response patterns of relevant sensors in that area to comprehensively determine the severity and development trend of the anomaly. Then, based on typical cases and expert rules matched to the knowledge base, the system automatically recommends optimal adjustment strategies. These strategies can range from targeted engineering interventions, such as reinforcing specific nodes, re-welding, or replacing damaged components, to dynamic adjustments at the operation and maintenance level, such as optimizing load distribution, temporarily restricting local use, or adjusting environmental parameters (such as temperature compensation). At the same time, for false alarms or tolerable small fluctuations caused by non-structural noise, the system recommends continued observation and periodic review rather than blind intervention, thus avoiding wasted resources. Furthermore, throughout the entire process, all adjustments and their effect data are transmitted back in real time and incorporated into the monitoring database, providing data support for subsequent health status assessments and continuous optimization of the knowledge base. This not only greatly improves the operational safety and management efficiency of large-scale steel structure projects, but also drives the construction industry towards digitalization and intelligentization.

[0086] In summary, the intelligent monitoring and adjustment method for steel structures according to the embodiment of the present application is explained. It continuously collects various physical quantity data by deploying multiple types of sensors at key locations of steel structures, and utilizes a dynamic threshold detection mechanism to identify potential abnormal events in real time. In response to an abnormality in a certain sensor, the system automatically associates and extracts time series data of other sensors associated with its spatial position in the time periods before and after the abnormality, forming multi-dimensional, spatiotemporal coupled data slices. Subsequently, spatiotemporal feature extraction is performed on these data to extract pattern representations of complex abnormal states, and based on this, accurate matching and diagnosis of the causes of abnormal events are achieved. In this way, the intelligent perception and analysis capabilities of the health status of complex, multi-source, and multi-dimensional steel structures are improved, providing solid data support and decision-making basis for subsequent scientific adjustment measures, and better ensuring the operational safety and intelligent management level of large-scale steel structure projects.

[0087] Furthermore, a steel structure intelligent monitoring and adjustment system is also provided.

[0088] Figure 3 FIG is a block diagram of a steel structure intelligent monitoring and adjustment system according to an embodiment of the present application. Figure 3 As shown, the steel structure intelligent monitoring and adjustment system 300 according to the embodiment of the present application includes: an abnormal event detection module 310, which is used to continuously monitor the sensor data of each sensor deployed on the steel structure, and perform abnormal event detection on the sensor data of each sensor based on dynamic threshold detection; a spatiotemporal data slice extraction module 320, which is used to automatically extract the time series data of all sensors related to the sensor within a preset time period before and after the abnormal event in response to detecting an abnormal event of a certain sensor to obtain a spatiotemporal data slice; a data sorting module 330, which is used to sort the spatiotemporal data slice according to the time dimension and the sensor sample dimension to obtain a set of sensor data time series; a spatiotemporal feature extraction module 340, which is used to extract spatiotemporal features from the set of sensor data time series to obtain an abnormal state sensor data pattern representation vector; an abnormality-related construction activity list extraction module 350, which is used to extract a list of abnormality-related construction activities from a structured construction log data set based on the occurrence timestamp of the abnormal time; and a diagnosis result generation module 360, which is used to input the abnormality-related construction activity list and the abnormal state sensor data pattern representation vector into a pre-built abnormal pattern-root cause knowledge base for pattern matching to obtain a diagnosis list.

[0089] As described above, the intelligent monitoring and adjustment system 300 for steel structures according to the embodiments of the present application can be implemented in various wireless terminals, such as a server equipped with an intelligent monitoring and adjustment algorithm for steel structures. In one possible implementation, the intelligent monitoring and adjustment system 300 for steel structures according to the embodiments of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the intelligent monitoring and adjustment system 300 for steel structures can be a software module in the operating system of the wireless terminal, or can be an application developed specifically for the wireless terminal; of course, the intelligent monitoring and adjustment system 300 for steel structures can also be one of the many hardware modules of the wireless terminal.

[0090] Alternatively, in another example, the steel structure intelligent monitoring and adjustment system 300 and the wireless terminal may also be separate devices, and the steel structure intelligent monitoring and adjustment system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0091] While various embodiments of the present disclosure have been described above, the above descriptions are illustrative, non-exhaustive, and not intended to be limiting of the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A steel structure intelligent monitoring and adjustment method, characterized in that: include: Continuously monitor the sensor data of each sensor deployed on the steel structure, and perform abnormal event detection on the sensor data of each sensor based on dynamic threshold detection; In response to detecting an abnormal event occurring at a certain sensor, automatically extracting time series data of all sensors related to the sensor within a preset time period before and after the abnormal event to obtain spatiotemporal data slices; The spatiotemporal data slices are sorted according to the time dimension and sensor sample dimension to obtain a set of sensor data time series; Performing spatiotemporal feature extraction on a collection of sensor data time series to obtain a representation vector of the abnormal state sensor data pattern; Based on the occurrence timestamp of the anomaly, a list of anomaly-related construction activities is extracted from the structured construction log dataset; The abnormality-related construction activity list and the abnormal state sensor data pattern representation vector are input into the pre-built abnormal pattern-root cause knowledge base for pattern matching to obtain a diagnosis list.

2. The steel structure intelligent monitoring and adjustment method according to claim 1, characterized in that: Abnormal event detection is performed on the sensor data of each sensor based on dynamic threshold detection, including: in response to a reading exceeding a limit or an abnormal change rate of the sensor data of each sensor, determining that an abnormal event occurs in the sensor.

3. The steel structure intelligent monitoring and adjustment method according to claim 2, characterized in that: All sensors related to the sensor are determined based on the sensor layout information, where the sensor layout information describes the spatial positions of the sensors on the steel structure and their interconnected relationships.

4. The steel structure intelligent monitoring and adjustment method according to claim 3, characterized in that: Perform spatiotemporal feature extraction on the set of sensor data time series to obtain the abnormal state sensor data pattern representation vector, including: Performing time series feature extraction based on a bidirectional LSTM model on each sensor data time series in the set of sensor data time series to obtain a set of sensor data time series pattern feature encoding vectors; A temporal feature space aggregation analysis is performed on the set of sensor data temporal pattern feature encoding vectors to obtain an abnormal state sensor data pattern representation vector.

5. The steel structure intelligent monitoring and adjustment method according to claim 4, characterized in that: Performing time series feature space aggregation analysis on the set of sensor data time series pattern feature encoding vectors to obtain abnormal state sensor data pattern representation vectors, including: Performing cluster analysis and set ground state feature aggregation on the set of sensor data time series pattern feature encoding vectors to obtain the sensor data time series pattern set ground state feature aggregation encoding vector; Based on the base state feature aggregation coding vector of the sensor data time series pattern set, the set of sensor data time series pattern feature coding vectors is subjected to feature gain aggregation based on excited state significance modulation to obtain an abnormal state sensor data pattern representation vector.

6. The steel structure intelligent monitoring and adjustment method according to claim 5, characterized in that: Cluster analysis and set base state feature aggregation are performed on the set of sensor data time series pattern feature encoding vectors to obtain the sensor data time series pattern set base state feature aggregation encoding vector, including: Input the set of sensor data temporal pattern feature encoding vectors into the K-Means clustering network to obtain K sensor data temporal pattern cluster center encoding vectors; The collective ground state feature aggregation of the K sensor data temporal pattern cluster center encoding vectors is performed based on the self-attention mechanism to obtain the sensor data temporal pattern collective ground state feature aggregation encoding vector.

7. The steel structure intelligent monitoring and adjustment method according to claim 6, characterized in that: Based on the base state feature aggregation coding vector of the sensor data time series pattern set, the set of sensor data time series pattern feature coding vectors is subjected to feature gain aggregation based on excited state saliency modulation to obtain an abnormal state sensor data pattern representation vector, including: Based on the base state feature aggregation coding vector of the sensor data timing pattern set, calculating the excited state coding vector of each sensor data timing pattern feature coding vector in the set of sensor data timing pattern feature coding vectors relative to the base state feature aggregation coding vector of the sensor data timing pattern set to obtain a set of sensor data timing pattern feature compensation excited state coding vectors; Calculating an excited-state significance modulation weight factor of each sensor data temporal pattern feature compensation excited-state coding vector in the set of sensor data temporal pattern feature compensation excited-state coding vectors to obtain a set of sensor data temporal pattern feature compensation excited-state significance modulation weight factors; Based on a set of sensor data timing pattern feature compensation excited state saliency modulation weight factors, weighted modulating a set of sensor data timing pattern feature compensation excited state coding vectors to obtain a set of sensor data timing pattern feature compensation excited state saliency coding vectors; The set of sensor data temporal pattern feature compensation excited state significant coding vectors and the sensor data temporal pattern set ground state feature aggregation coding vector are input into the feature gain superposition network to obtain the abnormal state sensor data pattern representation vector.

8. The steel structure intelligent monitoring and adjustment method according to claim 7, characterized in that: Based on the base state feature aggregation coding vector of the sensor data time series pattern set, calculating the excited state coding vector of each sensor data time series pattern feature coding vector in the set of sensor data time series pattern feature coding vectors relative to the base state feature aggregation coding vector of the sensor data time series pattern set to obtain a set of sensor data time series pattern feature compensation excited state coding vectors, including: Based on the base state feature aggregation coding vector of the sensor data timing pattern set, calculating the excited state coding vector of each sensor data timing pattern feature coding vector in the set of sensor data timing pattern feature coding vectors relative to the base state feature aggregation coding vector of the sensor data timing pattern set to obtain a set of initial sensor data timing pattern feature compensation excited state coding vectors; performing geometric boundary optimization based on an induced metric on each sensor data time series pattern feature encoding vector in the set of sensor data time series pattern feature encoding vectors to obtain a set of optimized sensor data time series pattern feature encoding vectors; Based on the set of optimized sensor data timing pattern feature coding vectors, the set of initial sensor data timing pattern feature compensation excited state coding vectors is corrected to obtain a set of sensor data timing pattern feature compensation excited state coding vectors.

9. A steel structure intelligent monitoring and adjustment system, characterized in that: include: An abnormal event detection module is used to continuously monitor the sensor data of each sensor deployed on the steel structure and perform abnormal event detection on the sensor data of each sensor based on dynamic threshold detection; a spatiotemporal data slice extraction module for automatically extracting time series data of all sensors related to the sensor within a preset time period before and after the occurrence of the abnormal event in response to detecting an abnormal event occurring at a certain sensor to obtain a spatiotemporal data slice; A data sorting module is used to sort the spatiotemporal data slices according to the time dimension and the sensor sample dimension to obtain a set of sensor data time series; A spatiotemporal feature extraction module is used to extract spatiotemporal features from a set of sensor data time series to obtain a representation vector of an abnormal state sensor data pattern; The exception-related construction activity list extraction module is used to extract the exception-related construction activity list from the structured construction log dataset based on the occurrence timestamp of the exception time; The diagnosis result generation module is used to input the abnormality-related construction activity list and the abnormal state sensor data pattern representation vector into the pre-built abnormal pattern-root cause knowledge base for pattern matching to obtain a diagnosis list.

Images