An ancient building structure health intelligent monitoring and abnormal early warning system and method thereof

Abstract

This invention discloses an intelligent monitoring and anomaly early warning system and method for the health of ancient building structures, specifically relating to the field of ancient building structure monitoring. It addresses the problems in existing technologies, such as fixed thresholds ignoring temperature background drift, lack of end-to-end coordination and thermal path anomaly measurement, inaccurate humidity scenarios, and a lack of data-driven optimization in early warning strategies. The system collects the tilt angles and temperature sequences of both sides of the same component, fits a background line to calculate residuals based on temperature and time segments, and enables temperature and humidity joint segmentation as needed. Within a sliding time window, it selects the optimal time delay to calculate the mortise and tenon joint hysteresis consistency energy level index and the continuous same sign fraction. It calculates the temperature-induced breathing asymmetry intensity index based on temperature cycles, constructs a scenario grid based on segment labels to calculate the scenario dominance coefficient, and generates event segments using historical quantile thresholds. Finally, it updates the strategy version using cost indicators calculated from these event segments, thereby reducing false alarms and missed alarms and improving cross-scenario portability.

Description

Technical Field

[0001] This invention relates to the field of ancient building structure monitoring, and more specifically, to an intelligent monitoring and early warning system and method for the health of ancient building structures. Background Technology

[0002] The monitoring of the structural health of ancient buildings has evolved from manual inspections to a standardized model of "sensor acquisition + wireless transmission + platform display." Common parameters include tilt angle, displacement, cracks, temperature, and humidity. Some projects are overlaid with 3D models or information models for visualization, positioning, and recording. Industry practice generally adopts a process of setting fixed thresholds for each sensor parameter, triggering alarms when limits are exceeded, and requiring manual verification. A few systems introduce simple trend judgments or rule bases to generate different levels of alerts and work orders. Overall, data is used for "reading and reporting numbers," and strategies are mostly maintained by human experience, resulting in insufficient comparability and transferability across seasons, components, and sites.

[0003] Existing solutions share common gaps in key areas: First, fixed thresholds fail to differentiate between background drift caused by slow variables such as temperature and season, easily misjudging "reversible environmental responses" as abnormalities, thus masking genuine anomalies. Second, judgments rely heavily on single-point exceedances, lacking interpretable measures for the "hysteresis and homogeneity" of responses at both ends of the same component and the "loop asymmetry" of temperature rise and fall paths, making it difficult to identify early loosening of connections and thermally induced stress channel anomalies. Third, humidity often dominates structural responses during the rainy season but is not used as a context label for segmentation and learning, leading to inaccurate segmentation and threshold migration. Fourth, early warning strategies lack contextualized governance and version evaluation mechanisms for management, failing to automatically select the optimal strategy based on data-driven cost indicators under "temperature amplitude, time period, and conditional humidity" scenarios. This leads to the technical problem that this invention aims to solve: under the premise of using only the tilt angles and temperatures at both ends of the same component and introducing humidity as a context label as needed, how to first strip away the temperature background, then characterize the collaboration between the two ends and the thermal path anomaly in an interpretable way, and then learn the risk coefficient in the context space without manual weighting, and form an event judgment in conjunction with the persistent constraints, and use the judgment result for the selection and updating of the strategy version, so as to reduce false alarms and false negatives and improve cross-scenario transferability.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies in the prior art, embodiments of the present invention provide an intelligent monitoring and early warning system for the health of ancient building structures and its method. This system collects the tilt angles and temperature sequences on both sides of the same component, fits a background line to calculate residuals based on temperature and time segments, and activates temperature and humidity joint segmentation as needed. Within a sliding time window, it selects the optimal time delay to calculate the mortise and tenon joint hysteresis consistency energy level index and the continuous same sign fraction. It calculates the temperature-induced respiration asymmetry intensity index based on temperature cycles, constructs a scenario grid based on segment labels to calculate the scenario dominance coefficient, jointly determines and generates event segments using historical quantile thresholds, and updates the strategy version using cost indicators calculated from event segments. This reduces false alarms and missed alarms, thus solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for intelligent monitoring and early warning of structural health anomalies in ancient buildings, comprising the following steps: S1: Collect the tilt angle and temperature on both sides of the same structure, fit the temperature and tilt angle background in segments according to temperature and time, obtain the background line of the tilt angle on both sides and the residual of the tilt angle on both sides. If the same sign rate of the relative humidity of the tilt angle residuals on both sides of multiple consecutive windows is higher than the same sign rate of the relative temperature, then enable temperature and humidity joint segmentation and recalculate the residuals, and output the segment labels. S2: Select the optimal time delay according to the temperature monotonic segment within the sliding time window and calculate the alignment difference energy. Normalize the temperature amplitude to obtain the mortise and tenon hysteresis consistency energy level index, and at the same time calculate the continuous same sign fraction. S3: Divide the complete cycle according to the temperature extreme value, calculate the difference between the temperature rise and fall slopes and normalize it with the temperature amplitude and the residual amplitude of the tilt angle on both sides, and combine it with the hysteresis to obtain the temperature-induced breathing asymmetry intensity index. S4: Establish a scenario grid based on segment labels. When humidity is not enabled, use temperature range and time period as dimensions. When enabled, add humidity level. Perform monotonic dominance counting on the two indices to obtain the scenario dominance coefficient. S5: Combine the situation dominance coefficient and the historical quantile threshold of the continuous same sign score to generate event segments and record the time index and the peak value of the residuals on both sides of the tilt angle. S6: Calculate the cost metrics, including the risks of false triggering and omission, based on event fragments and the current strategy. If the cost decreases, update the strategy version and write it into the strategy version metadata. If the cost does not decrease, maintain it and reduce the threshold update range for the next cycle.

[0007] In a preferred embodiment, step S1 includes the following: Collect the left tilt angle sequence, right tilt angle sequence, temperature sequence, and humidity sequence of the same component. Segment the initial time axis according to the intraday time period and the direction of temperature change to ensure that the temperature sequence within the segment is monotonic. Fit the left tilt angle background line and the right tilt angle background line within each initial segment to calculate the left tilt angle residual sequence and the right tilt angle residual sequence. Calculate the sign ratio of the difference between the two tilt angle residuals and the temperature difference, as well as the sign ratio of the difference between the temperature and humidity, within a continuous sliding time window.

[0008] In a preferred embodiment, step S1 further includes the following: When the humidity-related co-sign rate is continuously higher than the temperature-related co-sign rate, joint temperature and humidity segmentation is enabled. The humidity sequence is divided into humidity level layers according to historical quantiles and segment labels are incorporated to refit the background line. The left and right tilt residual sequences are recalculated, and finally the segment label sequence, left and right tilt residual sequences are output.

[0009] In a preferred embodiment, step S2 includes the following: Within the sliding time window, a candidate time delay set is set based on the sampling interval step size of the temperature monotonic segment. For each time delay, the right tilt angle residual sequence is shifted and the sign consistency with the left sequence is compared to calculate the same sign rate. The highest sign rate is selected as the optimal time delay. Under the optimal time delay, the absolute difference of the residuals on both sides is calculated point by point, multiplied by the sign index to obtain the alignment difference energy. The mortise and tenon hysteresis consistency energy level index is obtained by normalizing with temperature amplitude. At the same time, the ratio of the longest continuous segment length of the same sign index to the window length is calculated to obtain the continuous same sign score.

[0010] In a preferred embodiment, step S3 includes the following: Based on the temperature sequence, complete temperature cycles are divided into temperature maxima and minima. For each cycle, the rising and falling segments are extracted, and the median slope of the slope relative to the temperature is calculated to form a slope pair. The sum of the absolute values ​​of the slope differences is normalized to the temperature amplitude and the amplitude of the slope residuals on both sides to obtain the normalized slope difference value. At the same time, the approximate areas of the left and right planar trajectories are formed and normalized to take the smaller value as the lap tightness index. The geometric mean of the normalized slope difference value and the lap tightness index of multiple cycles is used to obtain the temperature-induced respiration asymmetry intensity index.

[0011] In a preferred embodiment, step S4 includes the following: A context grid is constructed based on the segmented label sequence. When temperature and humidity joint segmentation is not enabled, the temperature amplitude layer and time period layer are used as dimensions. When enabled, a humidity level layer is added. The historical sliding time window is divided into historical event samples and historical non-event samples. Within the current context grid, the current window mortise and tenon lag consistency energy level index and the temperature-induced respiration asymmetry intensity index are compared with the historical event samples. The proportion of samples that are not less than the current value is calculated as the context dominance coefficient. When there are insufficient samples, adjacent layers are merged to a preset lower limit, and the context dominance coefficient is output.

[0012] In a preferred embodiment, step S5 includes the following: The historical quantile thresholds of the situation dominance coefficient and the continuous same sign score are used as the judgment thresholds. When both are greater than or equal to the threshold, an event segment is generated. The event segment records the time index of the current sliding time window. The one with the largest absolute value of the tilt angle residual sequence on both sides is used as the peak value of the tilt angle residual on both sides and its corresponding time. The segmented label sequence is used for anomaly location and tracing.

[0013] In a preferred embodiment, step S6 includes the following: Based on event fragments, within a fixed number of sliding time windows after triggering, check whether the situation dominance coefficient and the temperature-induced respiration asymmetry intensity index drop synchronously and continuously decrease with the same sign, so as to accumulate the erroneous trigger proxy loss; within subsequent windows of untriggered windows, check whether they rise synchronously and the score rises to a high percentile, so as to accumulate the omission risk proxy loss.

[0014] In a preferred embodiment, step S6 further includes the following: The sum of the two losses is used as the cost indicator. If it decreases compared with the previous version, the policy version is updated by adjusting the quantile threshold value and the context grid configuration and writing it into the policy version metadata. Otherwise, the current policy is maintained and the threshold update magnitude of the next cycle is reduced.

[0015] A smart monitoring and early warning system for the structural health of ancient buildings includes: The background stripping module collects the tilt angle and temperature sequence of the same component on both sides, fits the background line according to temperature and time segments, and calculates the tilt angle residuals on both sides. If the same sign rate of relative humidity in continuous multi-window residuals is higher than that of relative temperature, temperature and humidity joint segmentation is enabled and the residuals are recalculated. The segmented label sequence and the tilt angle residual sequence on both sides are output. The hysteresis consistency module selects the optimal time delay within the temperature monotonic segment in the sliding time window to calculate the alignment difference energy, and obtains the mortise and tenon hysteresis consistency energy level index by normalizing the temperature amplitude. At the same time, it calculates the continuous same sign fraction and outputs the mortise and tenon hysteresis consistency energy level index and the continuous same sign fraction. The asymmetric strength module divides the complete cycle according to the temperature extreme value, calculates the difference between the temperature rise and fall slopes and normalizes it with the temperature amplitude and the residual amplitude of the tilt angle on both sides, and obtains the temperature-induced breathing asymmetric strength index by combining the lap tightness, and outputs the temperature-induced breathing asymmetric strength index. The context dominance module establishes a context grid based on the segmented label sequence. When humidity is not enabled, it uses temperature range and time period as dimensions. When enabled, it adds humidity level. It performs monotonic dominance counting on the two indices to obtain the context dominance coefficient and outputs the context dominance coefficient. The event determination module uses the situation dominance coefficient and the historical quantile threshold of the continuous same sign score to jointly determine the event, generate event segments and record the time index and the peak value of the residuals on both sides of the tilt angle, and output the event segments. The strategy optimization module calculates cost indicators with risks of false triggering and omission based on event fragments and the current strategy. If the cost decreases, the strategy version is updated and written to the strategy version metadata. If the cost does not decrease, the strategy is maintained and the threshold update range of the next cycle is reduced, and the updated strategy version or maintenance signal is output.

[0016] The technical effects and advantages of the intelligent monitoring and anomaly early warning system and method for the structural health of ancient buildings of this invention are as follows: This invention significantly reduces the misjudgment rate caused by environmental factors by stripping away the temperature background and dynamically incorporating the humidity context, thus maintaining the accuracy of anomaly warnings during the rainy season or seasonal changes.

[0017] This invention utilizes a quantitative index combining the co-hysteresis of both ends and the thermal path asymmetry with the scenario dominance coefficient to achieve risk assessment without manual weighting, thereby improving the early warning mobility across components and sites.

[0018] This invention uses event fragments to drive the calculation of cost indicators and automatically updates the strategy version, reducing false alarms and false negatives, and ensuring that the monitoring strategy is continuously optimized and adapts to long-term data accumulation. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for intelligent monitoring and early warning of structural health abnormalities in ancient buildings according to the present invention.

[0020] Figure 2 This is a schematic diagram of the structure of an intelligent monitoring and early warning system for the health of ancient building structures according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1: Figure 1 This invention presents a method for intelligent monitoring and early warning of structural health anomalies in ancient buildings, comprising: S1: Collect the tilt angle and temperature on both sides of the same structure, fit the temperature and tilt angle background in segments according to temperature and time, and obtain the background line of the tilt angle on both sides and the residual of the tilt angle on both sides. If the same sign rate of the relative humidity of the tilt angle residuals on both sides of multiple consecutive windows is higher than the same sign rate of the relative temperature, then enable temperature and humidity joint segmentation and recalculate the residuals, and output the segment labels.

[0023] S2: Select the optimal time delay according to the temperature monotonic segment within the sliding time window and calculate the alignment difference energy. Normalize the temperature amplitude to obtain the mortise and tenon hysteresis consistency energy level index, and at the same time calculate the continuous same sign fraction.

[0024] S3: Divide the complete cycle according to the extreme temperature values, calculate the difference between the temperature rise and fall slopes and normalize it with the temperature amplitude and the residual amplitude of the tilt angle on both sides, and combine it with the tightness of the hysteresis loop to obtain the intensity index of temperature-induced breathing asymmetry.

[0025] S4: Establish a scenario grid based on the segment labels. When humidity is not enabled, use temperature range and time period as dimensions. When enabled, add humidity level. Perform monotonic dominance counting on the two indices to obtain the scenario dominance coefficient.

[0026] S5: The event fragment is generated by combining the situation dominance coefficient and the historical quantile threshold of the continuous same sign score, and the time index and the peak value of the residuals on both sides are recorded.

[0027] S6: Calculate the cost metrics, including the risks of false triggering and omission, based on event fragments and the current strategy. If the cost decreases, update the strategy version and write it into the strategy version metadata. If the cost does not decrease, maintain it and reduce the threshold update range for the next cycle.

[0028] In the field of ancient building structural health monitoring, traditional methods often lead to misjudgments due to background drift caused by ignoring environmental factors such as temperature and humidity. This invention collects the tilt angle and temperature data of both sides of the same component, and optionally introduces humidity as a context label. It first performs segmentation and background stripping to extract residual signals that reflect structural anomalies, thereby providing a reliable basis for subsequent collaborative analysis and early warning.

[0029] The processing logic for step S1 is as follows: S1.1 Data Acquisition and Initial Time Axis Segmentation: The left-side tilt angle sequence of the same component is collected, with each time point corresponding to a left-side tilt angle value; the right-side tilt angle sequence is also collected, with each time point corresponding to a right-side tilt angle value; temperature and humidity sequences are collected simultaneously, with each time point corresponding to a temperature value and humidity value, respectively. The time axis is divided into multiple parts according to intraday periods, such as the morning period from 00:00 to 09:00, the midday period from 09:00 to 03:00, the afternoon period from 03:00 to 06:00, and the nighttime period from 06:00 to 00:00 the next day. Within each intraday period, the temperature sequence is further subdivided into segments according to the direction of temperature change, ensuring that the temperature sequence within each segment remains monotonically increasing or monotonically decreasing. If slight fluctuations occur in the temperature sequence within a segment, the ratio of the fluctuation amplitude to the overall trend slope is calculated. When this ratio is lower than a preset threshold, a monotonic straight line is fitted using the least squares method to adjust the segment boundaries until the temperature sequence within the segment meets the monotonicity requirement.

[0030] This process generates an initial segmented sequence, with each segment labeled with its start time, end time, and the direction of temperature change.

[0031] The components mentioned in this invention specifically refer to the core structural elements in ancient buildings, such as beams, columns, arches, or brackets, which are load-bearing components. These components are usually connected by mortise and tenon joints to form the skeleton system of the building. They are easily affected by environmental factors such as temperature and humidity, which can cause changes in tilt angle, displacement, or loosening of connections. Therefore, they become the key targets for monitoring and early warning to ensure the stability and long-term preservation of the overall structure of ancient buildings.

[0032] S1.2 Background line fitting and residual calculation within the segment: For each initial segment, a piecewise linear regression algorithm is applied to fit the left-side dip angle sequence and temperature sequence, with the temperature sequence as the independent variable and the left-side dip angle sequence as the dependent variable. The fitted result is a linear background line representing the expected influence of temperature on the left-side dip angle. Similarly, the same piecewise linear regression algorithm is applied to the right-side dip angle sequence and temperature sequence to fit the right-side dip angle background line. The left-side dip angle residual sequence is calculated by subtracting the value of the left-side dip angle background line at the corresponding time point from the value of the original left-side dip angle sequence at each time point. The right-side dip angle residual sequence is calculated by subtracting the value of the right-side dip angle background line at the corresponding time point from the value of the original right-side dip angle sequence at each time point. The specific implementation of the piecewise linear regression algorithm involves solving for the slope and intercept parameters for all data points within a segment, minimizing the sum of squared residuals from all data points to the fitted line. The right-side dip angle background line fitting also uses this calculation method.

[0033] This step generates left-side and right-side tilt residual sequences for subsequent humidity impact assessment.

[0034] S1.3 Calculation of the same number rate within the rolling time window and determination of humidity activation: Within multiple consecutive sliding time windows, the differences between adjacent time points of the left and right tilt angle residual sequences are first calculated. Then, the sign of the difference between the two tilt angle residuals is determined by whether they have the same sign. If both differences are positive or both are negative, the sign is positive; otherwise, it is negative. The sign of the difference between adjacent time points of the temperature sequence is calculated and compared with the signs of the differences between the two tilt angle residuals. The proportion of time points with the same sign within the window is calculated out of the total number of comparison points to obtain the temperature-related sign rate. Similarly, the sign of the difference between adjacent time points of the humidity sequence is calculated and compared with the signs of the differences between the two tilt angle residuals. The proportion of time points with the same sign within the window is calculated out of the total number of comparison points to obtain the humidity-related sign rate. If the humidity-related sign rate is higher than the temperature-related sign rate in multiple consecutive sliding time windows, the temperature and humidity joint segmentation mode is activated. The length of the sliding time window is determined based on the sampling frequency, and the sign rate is calculated by comparing sign consistency point by point and accumulating the proportion.

[0035] This judgment generates a decision on whether to enable the combined temperature and humidity segmentation.

[0036] Background line and residual recalculation under the combined temperature and humidity segmentation of S1.4: When temperature and humidity joint segmentation is enabled, the humidity sequence is divided into multiple humidity level layers based on its historical data distribution. For example, based on the ternary digits of the entire humidity data, it is divided into low humidity, medium humidity, and high humidity levels, each corresponding to a specific humidity value range. These humidity level layers are integrated as additional labels into the initial segmented sequence, forming a new segmented label sequence. Each segment label includes the original intraday time period, the direction of temperature change, and the newly added humidity level layer. For each segment with a humidity level layer, a piecewise linear regression algorithm is reapplied to the left-side tilt sequence and the temperature sequence to obtain the updated left-side tilt background line. The recalculated left-side tilt residual sequence is calculated by subtracting the value of the updated left-side tilt background line at the corresponding time point from the value of the original left-side tilt sequence. The same refitting and subtraction operations are performed on the right-side tilt sequence and the temperature sequence to obtain the updated right-side tilt background line and the recalculated right-side tilt residual sequence. The refitting calculation method is the same as the initial background line fitting, using the residual sum of squares minimization method.

[0037] This recalculation generates left-side and right-side tilt residual sequences that incorporate the humidity context.

[0038] S1.5 outputs segmented labels and the residual sequences of tilt angles on both sides: The final output is a segmented label sequence, including intraday time periods, temperature change direction, and possible humidity level labels; it also outputs left and right tilt angle residual sequences, which are the recalculated versions if temperature and humidity joint segmentation is enabled, otherwise the initial calculation versions. This output data is directly used in subsequent steps to ensure data continuity throughout the method.

[0039] Step S1 uses the left tilt angle sequence, right tilt angle sequence, temperature sequence, and humidity sequence of the same component as input. First, the initial time axis is segmented according to the intraday time period and the direction of temperature change to ensure the monotonicity of temperature within the segment. Then, within each segment, a piecewise linear regression algorithm is applied to fit the background line of the left tilt angle and the background line of the right tilt angle, and the residual sequences of the left tilt angle and the right tilt angle are calculated. Next, within the rolling time window, the same sign rate of the difference signs of the residuals of the two tilt angles with the difference signs of temperature and humidity is calculated. If the humidity-related same sign rate is continuously higher than the temperature-related same sign rate, joint temperature and humidity segmentation is enabled. The humidity sequence is divided into humidity level layers according to historical quantiles and integrated with the segment labels. The background line is refitted and the residual sequence is recalculated. Finally, the segment label sequence, as well as the left tilt angle residual sequence and the right tilt angle residual sequence, are output, thereby effectively stripping away the temperature background and dynamically integrating the humidity context, providing a clean residual signal to support subsequent anomaly early warning analysis.

[0040] In the structural health monitoring of ancient buildings, step S1 has already output segmented label sequences, left tilt angle residual sequences, and right tilt angle residual sequences by segmenting and peeling off the temperature background and optionally incorporating the humidity context. These data provide a clean foundation for identifying collaborative anomalies at both ends of the component. In step S2, the focus is further on the temperature monotonic segment within the sliding time window to quantify the hysteretic consistency and continuous unidirectionality of the tilt angle residuals on both sides, thereby characterizing the early loosening features of the mortise and tenon joint and improving the interpretability and early warning accuracy of the anomalies.

[0041] The processing logic for step S2 is as follows: S2.1 Sliding Time Window and Candidate Delay Set Settings: Based on the segmented label sequence output in step S1, temperature monotonic segments are located within a unified sliding time window. These segments correspond to the monotonically increasing or decreasing parts of the temperature sequence. The length of the sliding time window is determined based on the sampling frequency and the component response period, typically covering multiple complete temperature monotonic segments. A candidate time delay set is established using the sampling interval as the step size, for example, incrementing sequentially from zero sampling interval to the maximum time delay value, forming a discrete time delay list. This list is used for subsequent evaluation of the time offset of the right-side tilt residual sequence relative to the left-side tilt residual sequence. This setting process ensures that the candidate time delays cover the possible lag range, providing comprehensive alternatives for selecting the optimal time delay. The candidate time delay set is used to compare the sign consistency of the residuals on both sides.

[0042] Based on the selected latency, each option needs to be tested to find the one that maximizes sign consistency. This optimization is achieved by calculating the same sign rate.

[0043] S2.2 Optimal Delay Selection: For each delay value in the candidate delay set, the right-side tilt residual sequence is shifted by that delay, i.e., the entire sequence is moved backward by the corresponding number of sampling points to align it with the left-side tilt residual sequence. Then, within a sliding time window, the signs of the difference between the shifted right-side tilt residual sequence and the left-side tilt residual sequence are compared point-by-point. The sign is determined by the sign of the difference between the residual values ​​at adjacent time points; if they match, they are recorded as having the same sign. The proportion of time points with the same sign within the statistical window is used as the same-sign rate, and the delay value with the highest same-sign rate is selected as the optimal delay. This selection process uses an exhaustive search algorithm to traverse all candidates to ensure that the globally optimal delay is found, reflecting the actual lag in the response at both ends of the component. The optimal delay is used for further alignment calculations.

[0044] After applying optimal latency, the difference after alignment needs to be quantified, and the combination of absolute difference and same sign index should be introduced to evaluate the energy level.

[0045] S2.3 Alignment difference calculation: Under optimal time delay, the right-side tilt residual sequence is time-shifted to align with the left-side tilt residual sequence. The absolute difference between the two sequences is calculated point-by-point and multiplied by a sign index, where the sign index is 1 if the difference is of the same sign, and 0 otherwise. These products are averaged over the entire sliding time window to obtain the alignment difference energy, which characterizes the average bias strength of the aligned residuals and contributes only when the signs are consistent. This calculation captures the cooperative bias after hysteresis correction, providing a quantitative basis for subsequent normalization. The alignment difference energy has been obtained; normalization is now ready.

[0046] Alignment difference needs to take into account the scale of temperature change, and the index is standardized by normalizing the temperature amplitude.

[0047] S2.4 Mortise and tenon joint hysteresis uniformity energy level index normalization: The temperature amplitude within the sliding time window is calculated, which is the difference between the maximum and minimum values ​​of the temperature sequence. The alignment difference is then divided by this temperature amplitude to normalize the result, yielding the mortise and tenon hysteresis consistency energy level index. This index is comparable across different temperature fluctuations; lower values ​​indicate better consistency and reflect the stability of the mortise and tenon connection. This normalization process ensures the index is unaffected by temperature scale and is suitable for cross-seasonal comparisons. The mortise and tenon hysteresis consistency energy level index is used in conjunction with subsequent indices.

[0048] In addition to the energy level index, persistence also needs to be assessed, which can be supplemented by parallel computing of consecutive segments with the same number.

[0049] S2.5 Calculation of consecutive fractions with the same sign: Under optimal time-delay alignment, sequences of indices with the same sign are statistically analyzed chronologically, i.e., segments with consecutive 1s. The length of the longest consecutive segment with the same sign is identified and divided by the total length of the sliding time window to obtain a persistence score. This score quantifies the persistence of residual sign consistency; high values ​​indicate stable coordination, while low values ​​suggest potential discontinuities. This calculation is implemented using a sequence scanning algorithm to capture the continuity characteristics of the time dimension. The calculation of the persistence score supplements the temporal dimension of the index.

[0050] S2.6 Output Mortise and Tenon Hysteresis Consistent Energy Level Index and Continuity Same Sign Fraction: The mortise and tenon joint hysteresis consistency energy level index and the continuous same sign score are used directly in steps S3 and S4 as outputs, ensuring that the data flows to subsequent anomaly indices and contextual analysis. Interpretable hysteresis and same-sign measures are provided, improving the early identification capability of loose connections.

[0051] Through the processing in step S2 above, this invention accurately selects the optimal time delay within the sliding time window and calculates the alignment difference and the continuous same sign fraction, thereby realizing a quantitative description of the synergy of the tilt angle residuals at both ends of the component, reducing the limitations of single-point judgment, and providing key index support for contextualized early warning.

[0052] In the field of ancient building structural health monitoring, step S1 has already stripped away the temperature background and output the left tilt angle residual sequence, the right tilt angle residual sequence, and the segmented label sequence. Step S2 quantifies the mortise and tenon hysteresis consistency energy level index and the continuous same sign fraction within the sliding time window. These provide collaborative measurements for anomaly warning. In step S3, the extreme values ​​of the temperature sequence are further used to divide the complete cycle, and the slope difference and loop tightness are calculated for the temperature rise and fall path, thereby generating the temperature-induced breathing asymmetric intensity index, accurately characterizing the thermally induced stress channel anomaly, and improving the interpretability and cross-scenario adaptability of the warning.

[0053] Step S3 includes the following: S3.1 Temperature Cycle Division: Based on the temperature sequence, the sequence is divided into complete temperature cycles using temperature maxima and minima as cutoff points. Each cycle starts from a minimum, rises to a maximum, and then cools back to the next minimum, ensuring that the cycles cover continuous temperature rises and falls without overlap. This division is performed within a sliding time window, identifying all complete cycles covered by the window; the number of cycles depends on the window length and the frequency of temperature fluctuations. After this process, multiple complete temperature cycle sequences are obtained for subsequent path analysis.

[0054] Based on the complete temperature cycle, it is necessary to separate the rise and fall segments to assess asymmetry, and refine the calculation by extracting specific segments.

[0055] S3.2 Extraction of heating and cooling sections and calculation of slope: For each complete temperature cycle, the heating segment (the temperature increase from minimum to maximum) and the cooling segment (the temperature decrease from maximum to the next minimum) are extracted. Within the heating segment, the median slope of the left-side residual sequence relative to the temperature sequence and the median slope of the right-side residual sequence relative to the temperature sequence are calculated to form heating slope pairs. Similarly, within the cooling segment, the median slopes of the left and right sides are calculated to form cooling slope pairs. The median slope is obtained using a median algorithm and reflects the typical response rate of the residuals to temperature. This extraction and calculation ensures that the path characteristics of each cycle are captured independently. After this step, the heating slope pairs and cooling slope pairs are generated, ready for difference evaluation.

[0056] After the slope is calculated, the difference needs to be quantified, and the asymmetry needs to be standardized by combining amplitude normalization.

[0057] S3.3 Slope difference normalization value calculation: For each complete temperature cycle, the absolute difference between the slopes on the left and right sides of the rising slope is taken, and then added to the absolute difference between the slopes on the left and right sides of the falling slope, resulting in the sum of the absolute values ​​of the total slope differences. Next, the temperature amplitude of the cycle is calculated (maximum minus minimum), and the amplitudes of the two side slope residuals are calculated (maximum minus minimum of the left and right side slope residual sequences within the cycle). The sum of the absolute values ​​of the total slope differences is divided by the product of the temperature amplitude and the amplitudes of the two side slope residuals to achieve normalization, yielding a normalized slope difference value. This value quantifies the slope asymmetry intensity of the rising and falling paths. This calculation process is applied to all cycles within the sliding time window to ensure that the values ​​are dimensionless and comparable. The normalized slope difference value for each cycle is used in conjunction with tightness.

[0058] In addition to the slope, the path closure also needs to be evaluated, and the area can be approximated by the planar trajectory.

[0059] S3.4 Formation of Planar Trajectories and Approximation of Area: Within each complete temperature cycle, the temperature sequence is plotted on the x-axis, and the left-side tilt angle residual sequence on the y-axis, forming a left-side planar trajectory. Similarly, the temperature sequence and the right-side tilt angle residual sequence form a right-side planar trajectory. The trajectory points are connected chronologically, and the areas of the left and right planar trajectories are approximated using the trapezoidal quadrature formula, which estimates the area by integrating the closed regions of the trajectories piecewise. This formation and approximation process captures the loop response pattern of the residuals to temperature. The calculated trajectory areas are used for compactness indices.

[0060] After obtaining the trajectory area, it needs to be normalized and compared, and the minimum compactness should be selected to highlight the asymmetry.

[0061] S3.5 loop tightness index calculation: For each complete temperature cycle, the left-side trajectory tightness is obtained by dividing the area of ​​the left-side planar trajectory by the product of the cycle's temperature amplitude and the left-side tilt angle residual amplitude; the right-side trajectory tightness is calculated similarly. Then, the smaller value between the left-side and right-side trajectory tightness is taken as the loop tightness index for that cycle. This index reflects the tightness of the loop path; a low value indicates asymmetric expansion. This calculation ensures index standardization, facilitating multi-cycle aggregation. After this process, the loop tightness index for each cycle is ready.

[0062] Once all cyclic indicators are available, they need to be aggregated to generate an index, and a geometric mean is used to balance the contributions of multiple cycles.

[0063] S3.6 Calculation of the intensity index of temperature-induced respiration asymmetry: Over several complete temperature cycles covered by the sliding time window, the geometric mean of all slope difference normalization values ​​and corresponding loop tightness indices is calculated, i.e., the geometric mean of the slope difference normalization value and loop tightness index for each cycle. Then, the overall geometric mean of these means is calculated to obtain the thermo-induced breathing asymmetry intensity index. This index integrates the asymmetry of path slope and tightness, and a higher value indicates thermo-induced anomalies.

[0064] The temperature-induced breathing asymmetry strength index is used to quantify the degree of asymmetry in the tilt angle residual response path of ancient building components during temperature rise and fall cycles. It reflects the abnormal characteristics of the thermally induced stress channel, including the difference in the slope of heating and cooling and the tightness of the trajectory loop, to identify potential early loosening of connections or uneven distribution of thermal stress. The larger the index value, the higher the asymmetry strength, indicating a significant risk of abnormal structural response; the smaller the value, the more symmetrical the response path, and the relatively stable health of the structure.

[0065] S3.7 Output Thermo-Induced Respiratory Asymmetry Intensity Index: The thermally induced breathing asymmetry intensity index is used as the output directly in step S4 to ensure coordination with the mortise and tenon hysteresis consistent energy level index, forming a complete anomaly measurement. This output marks the end of step S3 and provides a quantitative description of the thermally induced path.

[0066] Through the processing in step S3 above, this invention uses temperature cycling to accurately calculate the slope difference normalization value and hysteresis tightness index, and aggregates them into a temperature-induced breathing asymmetry intensity index, which effectively identifies thermally induced stress channel anomalies, reduces false negatives, and enhances the transferability of the method.

[0067] In the field of ancient building structural health monitoring, step S1 has stripped the temperature background and output segmented label sequences and residual sequences. Steps S2 and S3 respectively generate the mortise and tenon hysteresis consistency energy level index, the continuous same sign fraction, and the temperature-induced breathing asymmetry intensity index. These indices quantify component coordination and thermal anomalies. In step S4, a scenario grid is constructed based on the segmented label sequences, the historical window is classified as a sample, and the risk coefficient, i.e. the scenario dominance coefficient, is learned by monotonically dominant counting within the current scenario grid. It can adapt to different environmental scenarios without manual weighting, thereby improving the transferability and accuracy of early warning.

[0068] Step S4 includes the following: S4.1 Contextual Grid Construction: Based on the segmented label sequence, it is determined whether temperature and humidity joint segmentation is enabled. If not enabled, a context grid is constructed using a temperature amplitude layer and a time period layer. The temperature amplitude layer is divided into multiple layers based on the quantiles of historical temperature amplitudes, and the time period layer corresponds to fixed categories such as morning, noon, afternoon, and night. If enabled, a humidity level layer is added, which is also divided into multiple layers based on the quantiles of historical humidity data. The context grid forms a multi-dimensional grid structure, with the number of layers in each dimension determined by the data distribution, typically three to five layers to balance granularity and coverage. The current sliding time window and historical windows are mapped to corresponding grid coordinates through their segmented label sequences.

[0069] S4.2 Historical Window Sample Division: The accumulated historical sliding time windows are categorized by checking whether event fragments are generated in step S5: if the window triggers an event fragment, it is classified as a historical event sample; if not, it is classified as a historical non-event sample. The historical window stores its mortise-and-tenon hysteresis uniformity energy level index, temperature-induced respiration asymmetry intensity index, and corresponding situational grid coordinates. The accumulation process is continuously updated with monitoring to ensure that the samples reflect long-term data patterns. Historical event samples are used for subsequent calculations, while historical non-event samples are retained for potential expansion.

[0070] S4.3 Comparison of current window with historical event samples and dominance count: Determine the current sliding time window's position within the context grid and obtain its mortise-and-tenon hysteresis uniformity energy level index and thermo-induced respiration asymmetry intensity index. For each historical event sample within this grid, check if its mortise-and-tenon hysteresis uniformity energy level index is not less than its current value, and simultaneously check if its thermo-induced respiration asymmetry intensity index is not less than its current value. If both conditions are met, it is recorded as a dominance. Accumulate all dominance counts and divide by the total number of historical event samples within the grid to calculate the dominance count ratio as the context dominance coefficient.

[0071] S4.4 Merging adjacent layers when samples are insufficient: When the total number of historical event samples within the current context cell is below a preset lower limit, adjacent layers are identified sequentially. First, adjacent layers above and below are merged at the temperature amplitude layer, then at the time period layer, and finally at the humidity level layer. If applicable, the dominance count of adjacent layers is added to the total number of samples in the current calculation. Merging stops when the total sample size reaches the preset lower limit, with priority based on dimensional correlation to minimize context bias. The updated dominance count ratio is used as the context dominance coefficient and calculated using the same formula.

[0072] Dimensional correlation refers to the inherent strength of the relationship between different dimensions of a context grid. For example, the temperature amplitude layer and the humidity level layer typically have a high correlation because both are influenced by seasonal environments, while the time period layer and the temperature amplitude layer have a relatively low correlation, mainly dominated by intraday cycles. To minimize contextual bias, i.e., to ensure that the merged samples retain the semantic consistency of the original context as much as possible without introducing irrelevant noise, the merging order prioritizes adjacent layers of the most correlated dimensions. For example, first merge the layers above and below the temperature amplitude layer, then the humidity level layer, and finally the time period layer. This prioritization design assesses correlation by quantifying the Pearson correlation coefficient or mutual information between dimensions and arranges the merging paths in descending order during implementation, thereby maintaining the accuracy of the risk coefficient and the cross-scenario transferability of the method.

[0073] S4.5 Output Context Dominance Coefficient: The situation dominance coefficient is output as a factor for joint determination in step S5.

[0074] The situation dominance coefficient represents the risk level of the current sliding time window relative to historical event samples within a specific situational grid. It is quantified by the proportion of times the mortise-and-tenon hysteresis uniformity level index and the temperature-induced respiration asymmetry intensity index are both not less than the historical event sample values, thus assessing the relative severity of anomalous events. A larger coefficient indicates a higher dominance ratio of the current window in historical anomalous situations, suggesting a higher risk of structural anomalies; a smaller value indicates a lower dominance ratio and relatively stable structural health.

[0075] Step S4 constructs a scenario grid based on the segmented label sequence. When temperature and humidity joint segmentation is not enabled, the temperature amplitude layer and time period layer are used as dimensions. When enabled, a humidity level layer is added. The historical sliding time window is divided into historical event samples and historical non-event samples. Within the current scenario grid, the mortise and tenon lag consistency energy level index and the temperature-induced respiration asymmetry intensity index of the current window are compared with the historical event samples one by one. The proportion of the number of times that are not less than the current value is counted as the scenario dominance coefficient. If the sample is insufficient, adjacent layers are merged until a preset lower limit is reached, thereby realizing data-driven risk quantification, supporting subsequent event judgment and improving the accuracy of early warning.

[0076] In the field of ancient building structural health monitoring, steps S1 to S3 have generated key indices and residual sequences, and step S4 calculates the situation dominance coefficient through situation grids. These provide a basis for risk quantification. However, in step S5 of this invention, the situation dominance coefficient and the historical quantile threshold of the continuous same sign score are used for joint judgment to generate event fragments and record core information, thereby realizing the continuous constraint and traceability of anomalies, significantly reducing false alarms and missed alarms and enhancing the reliability of early warning.

[0077] Step S5 includes the following: S5.1 Historical quantile threshold calculation: Based on the accumulated historical context dominance coefficient sequence and the historical continuous same-sign score sequence, each sequence is sorted. Then, a specific quantile position is selected as the historical quantile threshold. This threshold is calculated using a quantile algorithm and represents the positions in the sequence that are above a certain proportion of historical values. The historical sequence is extracted from all previous sliding time windows to ensure that the threshold dynamically reflects the data evolution trend. This calculation is performed before each current sliding time window, and the threshold is used as a decision threshold. The formula is: ,in Historical quantile threshold, The sorted sequence, Quantity ratio, For sequence length, For the floor function, the subscript indicates the first floor. Each element.

[0078] S5.2 Joint Judgment: The system retrieves the situation dominance coefficient and the persistent same-sign score for the current sliding time window. It then compares the situation dominance coefficient with its historical percentile threshold, and the persistent same-sign score with its historical percentile threshold. If both conditions are met, the system is deemed an anomaly and event fragment generation is triggered; otherwise, it is considered normal and no event is triggered. The determination uses a logical AND operation, activating only when both the risk coefficient and the persistence score exceed the historical thresholds, ensuring the rigor and situational adaptability of the anomaly determination.

[0079] S5.3 Event Fragment Generation and Recording: When the joint decision passes, an event segment is generated. This segment includes the time index of the current sliding time window, i.e., the start and end times. Within the window, the absolute values ​​of the left and right tilt angle residual sequences are calculated point-by-point. The largest absolute value is selected as the peak value of the two tilt angle residuals, and the precise time corresponding to this peak value is recorded. Simultaneously, the current segment label sequence is attached for tracing and locating abnormal situations. The event segment, as an independent recording unit, stores all elements for subsequent cost index calculation and manual verification in step S6.

[0080] Step S5 uses the scenario dominance coefficient and the historical quantile threshold of the continuous same sign score as the judgment threshold. When both meet the threshold conditions, an event segment is generated. This segment records the time index of the current sliding time window, the peak value of the residual sequence of the two sides of the tilt angle and its corresponding time, and the segment label sequence, thereby realizing the joint identification and tracing of abnormal events, supporting subsequent strategy optimization and reducing false alarms and false negatives.

[0081] Steps S1 to S5 have generated event fragments through data processing and judgment. These fragments capture abnormal signals and support risk assessment. In step S6, the event fragments and the current strategy are used to calculate the cost indicators of false triggering and omission risks. By comparing, the automatic update or maintenance of the strategy version is driven, realizing contextualized governance and version evaluation, significantly reducing manual intervention and improving the cross-scenario optimization capability of the early warning strategy.

[0082] S6.1 Calculation of Losses Due to False Triggering of Agent: Based on the event fragment generated in step S5, within a fixed number of sliding time windows after its triggering, the monitoring checks whether the situation dominance coefficient and the temperature-induced respiratory asymmetry intensity index synchronously and rapidly decline, and simultaneously checks whether the continuous same-sign score significantly decreases. If these conditions are met, a false trigger cost is accumulated, which represents the proxy loss of the false alarm. The fixed number of sliding time windows is determined by the monitoring period and is usually several consecutive windows to capture short-term recovery trends. This calculation process evaluates the index changes window by window, using a differential analysis algorithm to determine decline and reduction, ensuring that the loss is accumulated only when the environmental response is reversible. After this step is completed, the false trigger proxy loss value is obtained and used to compose the total cost.

[0083] In addition to the calculation of false triggers, it is also necessary to assess the potential risks of untriggered windows and quantify the omissions through subsequent window monitoring.

[0084] S6.2 Calculation of Loss Due to Omission Risk: For sliding time windows of untriggered event segments, observe whether the situation dominance coefficient and the temperature-induced respiration asymmetry intensity index rise synchronously within subsequent windows, accompanied by a sustained increase in the same-sign score to their respective high quantiles. If this condition is met, an omission cost is accumulated, representing the proxy loss for ignoring the anomaly. The high quantile is determined using a quantile algorithm based on historical sequences, reflecting the threshold for extreme anomalies. This calculation process scans a fixed number of subsequent windows, using trend analysis algorithms to detect increases and rises, ensuring that the loss is accumulated only when a true anomaly manifests. After this processing, the omission risk proxy loss value is obtained and combined with the false triggering.

[0085] After all losses are accumulated, a total metric needs to be formed and compared. The decision to update is made by simply summing the results and comparing them with the version numbers.

[0086] S6.3 Cost Indicator Calculation and Strategy Version Decision: The sum of the losses from false triggering and omission risks is used as the cost metric. This metric is compared with the cost metric of the previous strategy version. If the current cost metric decreases, the strategy version is updated, i.e., the quantile threshold value and scenario grid configuration are adjusted, and the strategy version metadata, including the strategy version number, effective time, quantile threshold value, and scenario grid configuration, is written. If the cost metric does not decrease, the current strategy is maintained, and the quantile threshold update magnitude for the next cycle is reduced. This reduction is achieved by multiplying by a preset decreasing factor, which is less than a certain asymptotic stability boundary. This decision-making process uses a direct comparison algorithm to ensure that updates only occur when performance improves, and the reduction magnitude is based on a preset decreasing rate. After this step, the strategy version is optimized or stabilized.

[0087] The specific adjustment and configuration process for updating the strategy version when the cost indicator decreases is as follows: First, evaluate the current quantile threshold value, usually represented as a list of quantile ratio values, for example, initially [0.8, 0.9], corresponding to the thresholds for the situation dominance coefficient and the continuous same sign score; the adjustment logic is to multiply each quantile threshold value by a decreasing factor, such as 0.95, to lower the threshold and thus improve the sensitivity of anomaly detection, avoid missing potential risks, and minimize further false alarms. Adjusting the situation grid configuration involves increasing the number of grid dimensions, for example, by adding 1 to the original number of dimensions, to refine the granularity of the temperature amplitude layer, time period layer, or humidity level layer, and improve situation discrimination capability. The specific calculation logic is: new quantile threshold value = current quantile threshold value × 0.95, new situation grid configuration dimension number = current dimension number + 1.

[0088] For example, assuming the current quantile threshold is [0.8, 0.9] and the context grid configuration dimension is 2, when the cost metric decreases, the updated quantile threshold is [0.76, 0.855] and the context grid configuration dimension is 3. This adjustment ensures that the strategy version is optimized incrementally, based on data-driven rather than human experience.

[0089] Through the processing in step S6 above, this invention calculates the cost indicators of false triggering and omission based on event fragments, and drives the update of strategy version and metadata based on this data, thereby realizing the automatic optimization of the early warning strategy, significantly reducing false alarms and omissions and enhancing cross-scenario portability.

[0090] Example 2: Figure 2 This invention discloses an intelligent monitoring and early warning system for the structural health of ancient buildings, comprising: The background stripping module collects the tilt angle and temperature sequences of both sides of the same component, fits the background line according to temperature and time segments, and calculates the tilt angle residuals on both sides. If the same sign rate of relative humidity in the residuals of consecutive multi-windows is higher than that of relative temperature, the temperature and humidity joint segmentation is enabled and the residuals are recalculated, and the segmented label sequence and the tilt angle residual sequence on both sides are output.

[0091] The hysteresis consistency module selects the optimal time delay within the sliding time window according to the temperature monotonic segment to calculate the alignment difference energy, and obtains the mortise and tenon hysteresis consistency energy level index by normalizing the temperature amplitude. At the same time, it calculates the continuous same sign fraction and outputs the mortise and tenon hysteresis consistency energy level index and the continuous same sign fraction.

[0092] The asymmetric intensity module divides the complete cycle according to the temperature extreme value, calculates the difference between the temperature rise and fall slopes and normalizes it with the temperature amplitude and the residual amplitude of the tilt angle on both sides, and obtains the temperature-induced breathing asymmetric intensity index by combining the lap tightness, and outputs the temperature-induced breathing asymmetric intensity index.

[0093] The context dominance module establishes a context grid based on the segmented label sequence. When humidity is not enabled, it uses temperature range and time period as dimensions. When enabled, it adds humidity level. It performs monotonic dominance counting on the two indices to obtain the context dominance coefficient and outputs the context dominance coefficient.

[0094] The event determination module uses the situation dominance coefficient and the historical quantile threshold of the continuous same sign score to jointly determine the event, generate event segments, record the time index and the peak value of the residuals on both sides of the tilt angle, and output the event segments.

[0095] The strategy optimization module calculates cost indicators with risks of false triggering and omission based on event fragments and the current strategy. If the cost decreases, the strategy version is updated and written to the strategy version metadata. If the cost does not decrease, the strategy is maintained and the threshold update range of the next cycle is reduced, and the updated strategy version or maintenance signal is output.

[0096] Specifically, the above description is only a preferred embodiment of this application and is not intended to limit this application.

[0097] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0098] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for intelligent monitoring and early warning of structural health anomalies in ancient buildings, characterized in that, Including the following steps: S1: Collect the tilt angle and temperature on both sides of the same structure, fit the temperature and tilt angle background in segments according to temperature and time, obtain the background line of the tilt angle on both sides and the residual of the tilt angle on both sides. If the same sign rate of the relative humidity of the tilt angle residuals on both sides of multiple consecutive windows is higher than the same sign rate of the relative temperature, then enable temperature and humidity joint segmentation and recalculate the residuals, and output the segment labels. S2: Select the optimal time delay according to the temperature monotonic segment within the sliding time window and calculate the alignment difference energy. Normalize the temperature amplitude to obtain the mortise and tenon hysteresis consistency energy level index, and at the same time calculate the continuous same sign fraction. S3: Divide the complete cycle according to the temperature extreme value, calculate the difference between the temperature rise and fall slopes and normalize it with the temperature amplitude and the residual amplitude of the tilt angle on both sides, and combine it with the hysteresis to obtain the temperature-induced breathing asymmetry intensity index. S4: Establish a scenario grid based on segment labels. When humidity is not enabled, use temperature range and time period as dimensions. When enabled, add humidity level. Perform monotonic dominance counting on the two indices to obtain the scenario dominance coefficient. S5: Combine the situation dominance coefficient and the historical quantile threshold of the continuous same sign score to generate event segments and record the time index and the peak value of the residuals on both sides of the tilt angle. S6: Calculate the cost metrics, including the risks of false triggering and omission, based on event fragments and the current strategy. If the cost decreases, update the strategy version and write it into the strategy version metadata. If the cost does not decrease, maintain it and reduce the threshold update range for the next cycle.

2. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 1, characterized in that, Step S1 includes the following: Collect the left tilt angle sequence, right tilt angle sequence, temperature sequence, and humidity sequence of the same component. Segment the initial time axis according to the intraday time period and the direction of temperature change to ensure that the temperature sequence within the segment is monotonic. Fit the left tilt angle background line and the right tilt angle background line within each initial segment to calculate the left tilt angle residual sequence and the right tilt angle residual sequence. Calculate the sign ratio of the difference between the two tilt angle residuals and the temperature difference, as well as the sign ratio of the difference between the temperature and humidity, within a continuous sliding time window.

3. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 2, characterized in that, Step S1 also includes the following: When the humidity-related co-sign rate is continuously higher than the temperature-related co-sign rate, joint temperature and humidity segmentation is enabled. The humidity sequence is divided into humidity level layers according to historical quantiles and segment labels are incorporated to refit the background line. The left and right tilt residual sequences are recalculated, and finally the segment label sequence, left and right tilt residual sequences are output.

4. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 3, characterized in that, Step S2 includes the following: Within the sliding time window, a candidate time delay set is set based on the sampling interval step size of the temperature monotonic segment. For each time delay, the right tilt angle residual sequence is shifted and the sign consistency with the left sequence is compared to calculate the same sign rate. The highest sign rate is selected as the optimal time delay. Under the optimal time delay, the absolute difference of the residuals on both sides is calculated point by point, multiplied by the sign index to obtain the alignment difference energy. The mortise and tenon hysteresis consistency energy level index is obtained by normalizing with temperature amplitude. At the same time, the ratio of the longest continuous segment length of the same sign index to the window length is calculated to obtain the continuous same sign score.

5. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 4, characterized in that, Step S3 includes the following: Based on the temperature sequence, complete temperature cycles are divided into temperature maxima and minima. For each cycle, the rising and falling segments are extracted, and the median slope of the slope relative to the temperature is calculated to form a slope pair. The sum of the absolute values ​​of the slope differences is normalized to the temperature amplitude and the amplitude of the slope residuals on both sides to obtain the normalized slope difference value. At the same time, the approximate areas of the left and right planar trajectories are formed and normalized to take the smaller value as the lap tightness index. The geometric mean of the normalized slope difference value and the lap tightness index of multiple cycles is used to obtain the temperature-induced respiration asymmetry intensity index.

6. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 5, characterized in that, Step S4 includes the following: A context grid is constructed based on the segmented label sequence. When temperature and humidity joint segmentation is not enabled, the temperature amplitude layer and time period layer are used as dimensions. When enabled, a humidity level layer is added. The historical sliding time window is divided into historical event samples and historical non-event samples. Within the current context grid, the current window mortise and tenon lag consistency energy level index and the temperature-induced respiration asymmetry intensity index are compared with the historical event samples. The proportion of samples that are not less than the current value is calculated as the context dominance coefficient. When there are insufficient samples, adjacent layers are merged to a preset lower limit, and the context dominance coefficient is output.

7. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 6, characterized in that, Step S5 includes the following: The historical quantile thresholds of the situation dominance coefficient and the continuous same sign score are used as the judgment thresholds. When both are greater than or equal to the threshold, an event segment is generated. The event segment records the time index of the current sliding time window. The one with the largest absolute value of the tilt angle residual sequence on both sides is used as the peak value of the tilt angle residual on both sides and its corresponding time. The segmented label sequence is used for anomaly location and tracing.

8. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 7, characterized in that, Step S6 includes the following: Based on event fragments, within a fixed number of sliding time windows after triggering, check whether the situation dominance coefficient and the temperature-induced respiration asymmetry intensity index drop synchronously and continuously decrease with the same sign, so as to accumulate the erroneous trigger proxy loss; within subsequent windows of untriggered windows, check whether they rise synchronously and the score rises to a high percentile, so as to accumulate the omission risk proxy loss.

9. The method for intelligent monitoring and early warning of structural health anomalies in ancient buildings according to claim 8, characterized in that, Step S6 also includes the following: The sum of the two losses is used as the cost indicator. If it decreases compared with the previous version, the policy version is updated by adjusting the quantile threshold value and the context grid configuration and writing it into the policy version metadata. Otherwise, the current policy is maintained and the threshold update magnitude of the next cycle is reduced.

10. A smart monitoring and early warning system for the structural health of ancient buildings, used to implement the smart monitoring and early warning method for the structural health of ancient buildings as described in any one of claims 1-9, characterized in that, include: The background stripping module collects the tilt angle and temperature sequence of the same component on both sides, fits the background line according to temperature and time segments, and calculates the tilt angle residuals on both sides. If the same sign rate of relative humidity in continuous multi-window residuals is higher than that of relative temperature, temperature and humidity joint segmentation is enabled and the residuals are recalculated. The segmented label sequence and the tilt angle residual sequence on both sides are output. The hysteresis consistency module selects the optimal time delay within the temperature monotonic segment in the sliding time window to calculate the alignment difference energy, and obtains the mortise and tenon hysteresis consistency energy level index by normalizing the temperature amplitude. At the same time, it calculates the continuous same sign fraction and outputs the mortise and tenon hysteresis consistency energy level index and the continuous same sign fraction. The asymmetric strength module divides the complete cycle according to the temperature extreme value, calculates the difference between the temperature rise and fall slopes and normalizes it with the temperature amplitude and the residual amplitude of the tilt angle on both sides, and obtains the temperature-induced breathing asymmetric strength index by combining the lap tightness, and outputs the temperature-induced breathing asymmetric strength index. The context dominance module establishes a context grid based on the segmented label sequence. When humidity is not enabled, it uses temperature range and time period as dimensions. When enabled, it adds humidity level. It performs monotonic dominance counting on the two indices to obtain the context dominance coefficient and outputs the context dominance coefficient. The event determination module uses the situation dominance coefficient and the historical quantile threshold of the continuous same sign score to jointly determine the event, generate event segments and record the time index and the peak value of the residuals on both sides of the tilt angle, and output the event segments. The strategy optimization module calculates cost indicators with risks of false triggering and omission based on event fragments and the current strategy. If the cost decreases, the strategy version is updated and written to the strategy version metadata. If the cost does not decrease, the strategy is maintained and the threshold update range of the next cycle is reduced, and the updated strategy version or maintenance signal is output.

Images