Data-driven parameter identification method for seismic damper of water conveyance channel

Abstract

The application discloses a data-driven-based parameter identification method for a water conveyance channel seismic damper, and relates to the technical field of water conveyance channel seismic dampers, and comprises the following steps: dividing acceleration data and damping force data obtained by a water conveyance channel under seismic excitation into continuous time windows, and constructing a bidirectional symmetric window group with each time window as a center, calculating the direction opposite rate and amplitude closeness between the acceleration data and the damping force data in each group, and judging whether the water conveyance channel structure appears response cancellation; in the case that the water conveyance channel structure appears response cancellation, determining the time period of response cancellation, and generating a multi-factor compression matrix based on a direction consistency factor, an amplitude offset factor and an energy residual factor. The application solves the problems of misjudgment and underestimation of the seismic damper parameters under the response cancellation condition, and realizes accurate identification and dynamic self-adaptive regulation and control of the seismic damper parameters of the water conveyance channel.

Description

Technical Field

[0001] This invention relates to the field of seismic damping devices for hydraulic aqueducts, and specifically to a data-driven method for identifying parameters of seismic damping devices for hydraulic aqueducts. Background Technology

[0002] Data-driven parameter identification of seismic dampers in hydraulic aqueducts is a novel identification technology that integrates measured structural data with data modeling methods. Its core concept is to construct a mapping relationship between excitation input and structural response by collecting dynamic measured data such as acceleration, displacement, strain, and the mechanical response of the dampers under seismic loading or vibration test conditions. This allows for the reverse identification of key damper parameters. Unlike traditional methods that rely on theoretical formulas or empirical estimations, existing methods typically employ multi-dimensional data processing and modeling mechanisms to improve the accuracy and adaptability of the identification. The implementation process generally includes the following key steps: First, the data acquisition and preprocessing stage involves deploying sensors at key locations on the aqueduct structure to acquire response data under seismic or excitation loads, and preprocessing the raw data through noise reduction and filtering. Next, the feature extraction and modeling stage utilizes system identification technology, regression analysis, and machine learning algorithms to extract characteristic parameters from the structural response that reflect changes in damper performance. Then, through model building and parameter inversion, a mathematical or intelligent model reflecting the "excitation-response-parameter" relationship is established, enabling dynamic identification of damper parameters such as damping coefficient, equivalent stiffness, and nonlinear energy dissipation characteristics. Finally, based on the identification results, working condition correlation analysis and parameter evolution evaluation are conducted to determine the evolution law of parameters with changes in working condition factors such as earthquake intensity and frequency, thus providing a scientific basis for the seismic performance evaluation, damper placement, and design optimization of the aqueduct structure under different seismic scenarios. This technology not only overcomes the uncertainties caused by model assumptions in traditional methods but also improves the matching degree of damper parameters to the actual dynamic behavior of the structure, which has significant engineering implications for improving the overall seismic energy dissipation capacity and safety assurance level of the aqueduct.

[0003] The existing technology has the following shortcomings:

[0004] In the process of data-driven identification of seismic damper parameters for hydraulic aqueducts, when the aqueduct structure undergoes complex dynamic responses under seismic excitation at a specific frequency, a cancellation phenomenon may occur where the acceleration signal and damping force signal are opposite in direction but similar in value. That is, the two types of signals are synchronous in time but almost cancel each other out numerically, forming a rare but misleading "approximately zero" response data. This phenomenon typically occurs when the damper is under strong excitation but the overall structure exhibits relative inertial motion, causing the actual effect of the damper to be neutralized by the structural response. Because the numerical amplitude of this data is extremely small during normalization, the system may misjudge it as the sensor not recording a valid response or the structure having no response, thus ignoring it as an invalid or noisy sample during the training and identification phases. Existing data-driven seismic damper parameter identification technologies for hydraulic aqueducts cannot accurately identify damper parameters based on normalized data under destructive response conditions. This is because these technologies assume all response amplitude differences originate from measurement errors or structural variations, failing to detect the loss of physical meaning due to destructive response. Consequently, key identification values ​​such as equivalent stiffness and damping coefficient are severely underestimated, resulting in dampers showing no response or performance degradation. This misidentification can lead to model bias propagation in subsequent seismic performance assessments, affecting the overall aqueduct's condition and potentially misleading engineering design and maintenance decisions, leading to structural safety hazards or resource waste.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a data-driven method for identifying the parameters of seismic dampers in hydraulic aqueducts, in order to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a data-driven method for identifying the parameters of seismic dampers in hydraulic aqueducts, specifically including the following steps:

[0008] S1. Divide the acceleration data and damping force data of the hydraulic aqueduct under seismic excitation into continuous time windows, and construct a bidirectional symmetrical window group with each time window as the center. Calculate the direction reversal rate and amplitude closeness between the acceleration data and damping force data in each group to determine whether the hydraulic aqueduct structure exhibits response cancellation.

[0009] S2. When the response cancellation occurs in the hydraulic aqueduct structure, determine the time period of response cancellation, and generate a multi-factor compression matrix based on the direction consistency factor, amplitude offset factor and energy residual factor to process the original response data in order to determine the normalized data of the hydraulic aqueduct structure under the response cancellation condition.

[0010] S3. Based on the determined normalized data, construct a set of input vectors, and calculate the direction residual term, amplitude rebound term and energy consumption delay term in sequence to generate a set of feature offset expressions.

[0011] S4. Combine the characteristic offset expression group with the normalized data to generate an identification sequence containing energy dissipation density paths, nonlinear intersection intervals and stiffness reduction points. Use the identification sequence to identify the equivalent stiffness, energy dissipation capacity and delayed response parameters of the seismic damper.

[0012] S5. Construct a bias trend regulation expression based on the changing trend of the identification sequence and the response synchronization rate of the normalized data, and adjust the factors in the bidirectional symmetric window group and the multi-factor compression matrix in real time to perform dynamic regulation of the parameter identification process.

[0013] Preferably, S1 specifically includes the following steps:

[0014] S101. The acceleration and damping force data of the hydraulic aqueduct under seismic excitation are divided into continuous time windows. The original data is segmented by setting equal time intervals so that each time window contains the same number of acceleration and damping force data, ensuring that the time series is complete, continuous and synchronous.

[0015] S102. Taking each time window as the center, extend the same number of consecutive time windows forward and backward along the time axis to construct a bidirectional symmetrical window group centered on the current time window. Then, rearrange the acceleration data and damping force data in the bidirectional symmetrical window group according to the chronological order to form a standard input data sequence for subsequent judgment.

[0016] S103. For each bidirectional symmetrical window group, compare the sign relationship between acceleration data and damping force data at the same time point, count the proportion of data points with opposite signs to the total data points, calculate the direction reversal rate, and simultaneously calculate the normalized value of amplitude difference between each pair of data points, count the proportion of them falling into the preset amplitude proximity interval, calculate the amplitude proximity degree, and determine whether the hydraulic aqueduct structure has a response cancellation situation based on the joint condition that the direction reversal rate is greater than the preset first judgment threshold and the amplitude proximity degree is greater than the preset second judgment threshold.

[0017] Preferably, S103 is as follows:

[0018] For each bidirectional symmetrical window group, the signs of acceleration data and damping force data at the same time point are compared point by point. The product sign judgment method is used to mark the data points with negative product as opposite directions. The ratio of the number of data points with opposite directions to the total number of data points in the window group is calculated to obtain the opposite direction rate.

[0019] For each pair of data points, the absolute value of the difference between the acceleration data and the damping force data is calculated. The difference value is then normalized to the [0,1] interval by the maximum amplitude normalization process. The ratio of the number of data points whose difference value falls into the preset amplitude close interval after normalization to the total number of data points is calculated to obtain the amplitude closeness.

[0020] The direction reversal rate and amplitude proximity are compared with the preset first judgment threshold and the preset second judgment threshold, respectively. If the direction reversal rate is greater than the preset first judgment threshold and the amplitude proximity is greater than the preset second judgment threshold, it is determined that the hydraulic aqueduct structure has a response cancellation situation in the bidirectional symmetrical window group.

[0021] Preferably, S2 specifically includes the following steps:

[0022] S201. In the case of response cancellation in the hydraulic aqueduct structure, based on the joint judgment condition that the direction opposition rate of acceleration data and damping force data in the bidirectional symmetrical window group is greater than the preset first judgment threshold and the amplitude proximity is greater than the preset second judgment threshold, the time window that meets the joint condition is continuously detected on the time axis, and the time windows that are adjacent in time and both meet the joint judgment condition are merged to determine the time range that forms a continuous interval, which is the time period of response cancellation.

[0023] S202. During the time period of response cancellation, extract the direction consistency factor, amplitude offset factor and energy retention factor point by point for acceleration data and damping force data. The direction consistency factor is calculated by comparing the sign consistency between the direction trend at a local time point and the average direction trend of the time period. The amplitude offset factor is calculated by the normalized value of the difference between the current data point and the median amplitude in the time period. The energy retention factor is calculated by the proportion of the energy of the data point in the total energy of the time period.

[0024] S203. The directional consistency factor, amplitude offset factor and energy residual factor are combined in chronological order to form a multi-factor compression matrix. The original response data is then weighted and compressed based on this multi-factor compression matrix to generate normalized data with a uniform numerical scale under the condition of response cancellation, which is used for subsequent identification of seismic damper parameters.

[0025] Preferably, S202 specifically refers to:

[0026] During the period of response cancellation, the local directional change trend of acceleration data and damping force data is extracted by sliding windows of fixed length. The directional trend of each window is obtained by calculating the sign change between adjacent data points in each window and comparing it with the overall directional average trend during the period of response cancellation. Data points with consistent signs are marked as consistent directional points. The directional consistency factor is the proportion of the number of consistent directional points to the total number of points.

[0027] The median of the amplitude is calculated for the acceleration and damping force data within the response cancellation time period. For each data point, the absolute value of the difference between it and the corresponding median is calculated. The normalized amplitude offset value is obtained by normalizing the maximum amplitude within the response cancellation time period. The amplitude offset factor is the normalized difference.

[0028] For each data point, calculate its instantaneous energy value and sum the energy of all data points within the response cancellation time period. Use the proportion of energy at a single point to the total energy as the energy retention factor, and apply it one-to-one with the direction consistency factor and amplitude offset factor to the construction of the subsequent multi-factor compression matrix.

[0029] Preferably, S3 is as follows:

[0030] Based on the determined normalized data, the acceleration data and damping force data are paired point by point in chronological order. The normalized acceleration data and normalized damping force data corresponding to the same time point are combined to form basic data pairs. Multiple consecutive basic data pairs are used to form input vectors. All input vectors are arranged in chronological order to construct a set of input vectors for identifying abnormal response characteristics.

[0031] For each input vector in the input vector set, the following calculations are performed in chronological order: Based on the change in the sign of the normalized acceleration and damping force data between the current input vector and the previous input vector, the difference in direction change is calculated, and this difference in direction change is used as the direction residual term; Based on the change between the normalized amplitude in the current input vector and the normalized amplitude in the previous input vector, the magnitude of the amplitude regression from the deviation state to the center state is calculated, forming the amplitude rebound term; Based on the temporal position relationship between the energy change value corresponding to the current input vector and the energy change values ​​of multiple previous input vectors, the lag ratio of the energy change in the time series is calculated, forming the energy consumption delay term;

[0032] The direction residual, amplitude rebound, and energy consumption delay terms corresponding to each input vector are combined in chronological order to form a set of expressions that correspond one-to-one with the input vector, and this set of expressions is determined as the feature offset expression group.

[0033] Preferably, S4 specifically includes the following steps:

[0034] S401. Pair the characteristic offset expression group with the normalized data according to time points, and concatenate the direction residual term, amplitude rebound term and energy consumption delay term corresponding to each time point with the normalized acceleration data and normalized damping force data of that time point to construct a combined data sequence arranged in time order.

[0035] S402. Based on the change trajectory of normalized energy information and energy consumption delay term in the combined data sequence, calculate the unit energy distribution in the time dimension to form an energy consumption density path; in the combined data sequence, screen the data intervals where the amplitude rebound term and the normalized acceleration change rate simultaneously exceed their respective preset change thresholds at multiple consecutive time points, and determine the data interval as a nonlinear cross interval; compare the change amplitude of the directional residual term and the change amplitude of the normalized stiffness response in the combined data sequence, and mark the time points where they simultaneously exceed their respective change thresholds at the same time point as stiffness decreasing points. The energy consumption density path, nonlinear cross intervals, and stiffness decreasing points together constitute the identification sequence;

[0036] S403. Calculate the energy dissipation capacity of the seismic damper based on the cumulative energy change of the energy dissipation density path in the identification sequence, calculate the delayed response parameters based on the time span and response asymmetry of the nonlinear intersection interval, and calculate the equivalent stiffness of the seismic damper based on the difference in the slope of the change between the normalized acceleration data and the normalized damping force data before and after the stiffness decrease point.

[0037] Preferably, S403 is as follows:

[0038] Based on the energy dissipation density path in the identification sequence, the energy distribution per unit time is accumulated along the time sequence, the energy increment in the continuous time period is summarized to form an energy accumulation curve, and the change amplitude of the energy accumulation curve in the corresponding time range of the identification sequence is used as the energy dissipation capacity characterization quantity of the seismic damper.

[0039] For the nonlinear cross intervals identified in the identification sequence, the start and end positions of the interval on the time axis are extracted, the corresponding time span is calculated, and the positive and negative distribution of normalized acceleration data and normalized damping force data within the interval are statistically analyzed to obtain the degree of response asymmetry. Based on the combination of the time span and the degree of response asymmetry, the delayed response parameters of the seismic damper are determined.

[0040] Using the stiffness-decreasing points in the identification sequence as boundaries, normalized acceleration data and normalized damping force data are extracted from the adjacent time periods before and after each point. The corresponding slopes between the changes in acceleration and damping force in each time period are calculated, and the differences between the slopes are compared. The difference is used as the equivalent stiffness of the seismic damper.

[0041] Preferably, S5 is as follows:

[0042] Based on the changing trend of the identification sequence, the distribution characteristics of the energy consumption density path, nonlinear intersection interval and stiffness decrease point in a continuous time period are extracted. Combined with the synchronous change ratio of acceleration data and damping force data in the normalized data, an offset trend control expression reflecting the dynamic relationship between the changing trend of the identification sequence and the response synchronization rate of the normalized data is constructed to characterize the degree of response consistency offset in the current parameter identification process.

[0043] Based on the degree of response consistency shift presented in the offset trend regulation expression, the extension length and position of the bidirectional symmetric window group on the time axis are adjusted. Based on the statistical characteristics of direction change, amplitude difference and energy ratio in the normalized data of the current identification stage, the factor weights of the direction consistency factor, amplitude shift factor and energy residual factor in the multi-factor compression matrix are updated to form an adjusted input structure that conforms to the current response state.

[0044] The real-time adjusted bidirectional symmetric window group and the updated multi-factor compression matrix are reapplied to the normalized data analysis process. The characteristic offset expression group and the identification sequence are iteratively reconstructed, and the seismic damper parameter identification process is dynamically controlled so that the identification model can continuously adapt to changes in response state.

[0045] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0046] 1. This invention, by introducing a bidirectional symmetrical window group and a multi-factor compression matrix, establishes for the first time a mechanism for identifying and utilizing the canceling response phenomenon of hydraulic aqueducts under strong earthquakes. It can accurately determine complex response states where acceleration and damping force data have opposite directions and similar amplitudes, avoiding the problem of traditional methods misjudging responses as invalid due to near-zero signals. Based on joint modeling of direction consistency factors, amplitude offset factors, and energy residual factors, it preserves subtle but structurally significant response details in the original data that are easily overlooked. During normalization, it achieves deep information compression and scale unification, providing a stable and reliable data foundation for subsequent feature extraction and parameter identification.

[0047] 2. This invention extracts key structural response features such as directional residuals, amplitude rebound, and energy consumption delay by constructing a set of feature offset expressions and an identification sequence. Combined with offset trend control expressions, it achieves dynamic control of the identification process, enabling the identification model to adapt to changes in the structural response state in real time. This improves the identification accuracy and robustness of the equivalent stiffness, energy dissipation capacity, and delayed response parameters of seismic dampers. Compared to existing technologies, this scheme not only solves the misidentification problem caused by response cancellation but also possesses dynamic adaptive capabilities. It can be widely applied to the seismic performance evaluation and structural health monitoring of hydraulic aqueducts under complex seismic conditions, providing strong technical support for improving engineering safety and operational efficiency. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0049] Figure 1 This is a flowchart illustrating the data-driven method for identifying seismic damper parameters in hydraulic aqueducts according to the present invention. Detailed Implementation

[0050] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0051] This invention provides, for example Figure 1 The data-driven method for identifying seismic damper parameters in hydraulic aqueducts, as shown, specifically includes the following steps:

[0052] S1. Divide the acceleration data and damping force data of the hydraulic aqueduct under seismic excitation into continuous time windows, and construct a bidirectional symmetrical window group with each time window as the center. Calculate the direction reversal rate and amplitude closeness between the acceleration data and damping force data in each group to determine whether the hydraulic aqueduct structure exhibits response cancellation.

[0053] In this embodiment, S1 specifically includes the following steps:

[0054] S101. The acceleration and damping force data of the hydraulic aqueduct under seismic excitation are divided into continuous time windows. The original data is segmented by setting equal time intervals so that each time window contains the same number of acceleration and damping force data, ensuring that the time series is complete, continuous and synchronous.

[0055] The acceleration and damping force data acquired from aqueducts under seismic excitation are divided into continuous time windows, which can be achieved by traversing the original time series. Specifically, the sampling frequency of the acceleration and damping force data is first determined, for example, 100 data points per second. Then, a fixed time interval, such as 0.5 seconds or 1 second, is set as the length of the time window. Subsequently, the acceleration and damping force data are divided into a series of consecutive time segments of consistent length, each segment being a time window. Within each time window, the acceleration and damping force data correspond one-to-one, forming time-synchronized data pairs. For example, if the earthquake lasts 20 seconds and the set time interval is 1 second, then 20 time windows will be formed, each containing the acceleration and damping force data for that 1 second. This method not only ensures the continuity of the data in the time dimension but also provides a unified data input format for subsequent window group construction, feature extraction, and response judgment, thus providing a clear and stable data foundation for parameter identification algorithms.

[0056] Seismic excitation refers to the external disturbance generated when a hydraulic aqueduct encounters seismic loads during actual operation or simulation testing, typically manifesting as rapid vibrations of the ground or foundation structure. Acceleration data refers to the change in velocity of the structure per unit time in various directions, recorded by sensors, reflecting the overall dynamic response of the aqueduct under seismic loading. Damping force data records the energy dissipation force generated by the seismic dampers during the process and is a core indicator for evaluating damper performance. A continuous time window refers to several time intervals arranged sequentially with fixed lengths and no overlap on the time axis, ensuring the sequentiality and stability of data processing. Setting equal time intervals is significant because it unifies the amount of data within the time window, ensuring the consistency of the dimensionality of data samples in each time segment, which is beneficial for maintaining the uniformity of the feature space during subsequent modeling and analysis. Segmenting the raw data is the process of transforming a large amount of unstructured time-series response data into structured input data, which can avoid the impact of single-point anomalies on the overall model training, while improving the robustness and interpretability of parameter identification. This operation is a prerequisite for achieving data standardization and structural behavior identification.

[0057] S102. Taking each time window as the center, extend the same number of consecutive time windows forward and backward along the time axis to construct a bidirectional symmetrical window group centered on the current time window. Then, rearrange the acceleration data and damping force data in the bidirectional symmetrical window group according to the chronological order to form a standard input data sequence for subsequent judgment.

[0058] Expanding each time window by the same number of consecutive time windows before and after it along the time axis is designed to capture the dynamic response characteristics before and after the current time period, thereby enhancing the sensitivity to transient and trend changes in parameter identification. In implementation, an expansion range parameter can be set, for example, expanding by two time windows to the left and two to the right. Each time window's center will then combine with its two preceding and two following windows to form a bidirectional symmetrical window group containing five time windows. For example, if the current time window is the 10th window, the corresponding bidirectional symmetrical window group consists of windows 8 to 12. Within this window group, all acceleration and damping force data are connected sequentially from front to back, forming a continuous data stream with a unified timestamp, allowing this data group to be directly input into subsequent data processing flows used for identification or judgment. This approach ensures that the data referenced by each identification and judgment unit not only includes the current response state but also considers its temporal causes and consequences, making parameter identification more complete and robust. For example, when analyzing whether there is a cancellation of response in the aqueduct response, a single time window may not be sufficient to identify the trend, while a two-way symmetrical window group can identify complex behaviors such as fluctuation continuation and response reversal, thereby improving the accuracy of identification.

[0059] The timeline refers to the forward and backward time segments within the currently selected time window, encompassing potential lag or advance effects during response evolution. A bidirectional symmetrical window group places the current time window in the center, with an equal number of consecutive time windows added to its left and right, forming a time data set with the center as the axis of symmetry. This structure is beneficial for capturing the symmetry and directionality of signal changes. The standard input data sequence used for subsequent judgment refers to a unified input format formed by concatenating all acceleration and damping force data within the window group in chronological order. Its function is to provide a consistent and temporally continuous data foundation for subsequent processing operations such as parameter identification, feature extraction, or response determination. This standard input data sequence eliminates breaks between time segments, allowing data processing algorithms to analyze the changing trends, mutual influences, and response patterns within multiple windows in an integrated manner, thereby improving the reliability of the identification results. Overall, this process achieves expansion and integration of data in the temporal dimension, providing a prerequisite for in-depth analysis of complex dynamic behaviors.

[0060] S103. For each bidirectional symmetrical window group, compare the sign relationship between acceleration data and damping force data at the same time point, count the proportion of data points with opposite signs to the total data points, calculate the direction reversal rate, and simultaneously calculate the normalized value of amplitude difference between each pair of data points, count the proportion of them falling into the preset amplitude proximity interval, calculate the amplitude proximity degree, and determine whether the hydraulic aqueduct structure has a response cancellation situation based on the joint condition that the direction reversal rate is greater than the preset first judgment threshold and the amplitude proximity degree is greater than the preset second judgment threshold.

[0061] The reason for comparing the sign relationship between acceleration and damping force data at the same time point for each bidirectional symmetrical window group, and calculating the direction opposition rate and amplitude similarity, is that a special but highly misleading response phenomenon exists in the data-driven identification of seismic dampers for hydraulic aqueducts: response cancellation. This phenomenon mainly manifests as strong numerical changes in acceleration and damping force data at specific seismic frequencies, but with opposite directions and similar amplitudes, ultimately forming a "nearly zero" signal illusion. If this type of signal is not specifically identified and eliminated, the system will misjudge it as no structural response or ineffective damper, severely interfering with the parameter identification results. Therefore, by constructing bidirectional symmetrical window groups and performing fine-grained analysis of response data in different time periods, this special cancellation behavior can be captured and quantified. The direction opposition rate measures the degree of directional opposition, and the amplitude similarity measures the degree of numerical convergence. The joint judgment mechanism of the two can ensure the accuracy of identification while eliminating the interference of accidental noise or local disturbances, thus establishing a clear and reliable data foundation for subsequent parameter identification. This processing method addresses the problem of lost physical meaning at the signal level and is a key step in improving the stability and engineering applicability of recognition models.

[0062] In this embodiment, S103 specifically refers to:

[0063] For each bidirectional symmetrical window group, the signs of acceleration data and damping force data at the same time point are compared point by point. The product sign judgment method is used to mark the data points with negative product as opposite directions. The ratio of the number of data points with opposite directions to the total number of data points in the window group is calculated to obtain the opposite direction rate.

[0064] For each bidirectional symmetrical window group, the signs of acceleration and damping force data at the same time point are compared point by point to analyze their directional consistency during the dynamic response process. This is achieved by extracting the acceleration and damping force data at each time point and determining whether their signs are the same. Specifically, this is done by multiplying the two data points; if the product is negative, the directions are opposite; if the product is positive or zero, the directions are the same or one of them is zero. This method is called the product sign determination method, and its advantages are simple calculation, adaptability to continuous data processing, and rapid identification of the relative directional relationship between signals. The direction inversion rate is obtained by counting the number of data points with negative products at all time points and calculating the ratio with the total number of data points in the window group. The direction inversion rate reflects whether the acceleration and damping force data are in a state of directional opposition for most of the time period. For example, if there are 100 data points in a bidirectional symmetrical window group, and the two types of data have opposite directions at 80 time points, then the direction inversion rate is 0.8. If this ratio exceeds a preset threshold, it can be preliminarily determined that the current window group exhibits a strong directional opposition trend, which may be a precursor to response cancellation. This process not only effectively uncovers the inverse coupling relationship between acceleration and damping force but also provides fundamental characteristic indicators for subsequent identification of abnormal response patterns. Among the various technical characteristics, the positive and negative signs of acceleration and damping force data at the same time point reflect the directional characteristics of their physical behavior. Point-by-point comparison ensures the precision of the calculation granularity, the product sign judgment method provides a unified and generalizable judgment logic, and the directional opposition rate provides a quantitative basis for the entire identification process.

[0065] For each pair of data points, the absolute value of the difference between the acceleration data and the damping force data is calculated. The difference value is then normalized to the [0,1] interval by the maximum amplitude normalization process. The ratio of the number of data points whose difference value falls into the preset amplitude close interval after normalization to the total number of data points is calculated to obtain the amplitude closeness.

[0066] Calculating the absolute value of the amplitude difference between acceleration and damping force data for each data point pair is to measure their numerical closeness and avoid the canceling effect caused by positive and negative signs. Specifically, at the same time point, the values ​​of acceleration and damping force data are taken, and the absolute value of their numerical difference is calculated as the amplitude difference between them. For example, if the acceleration value is 2.1 and the damping force value is -2.3 at a certain time point, the difference is 4.4, and the absolute value is 4.4. Since the original data values ​​at each time point have different ranges, maximum amplitude normalization must be performed to map all difference values ​​to a standardized [0,1] interval, avoiding interference from different orders of magnitude. Maximum amplitude normalization can be achieved by extracting the maximum value of acceleration and damping force data in the current window group and using it as the normalization benchmark. Each amplitude difference is then divided by this maximum value to form a relative difference. The preset amplitude proximity interval is a small range of values ​​set manually to define which normalized difference values ​​can be considered "close," such as 0 to 0.1. When the normalized difference falls within this interval, the acceleration data and damping force data at that point are considered numerically close. The number of data points falling within this interval across all normalized differences is counted, and the ratio is calculated to the total number of data points within the window group; this yields the amplitude closeness. Amplitude closeness reflects whether acceleration and damping force show a high degree of similarity in numerical amplitude throughout the analysis period. If most point pairs meet the closeness condition, it indicates a potential trend of numerical cancellation between the two types of data, which becomes an important basis for judging response cancellation. In the technical features, the absolute value of the amplitude difference reflects the numerical distance; maximum amplitude normalization ensures scale uniformity; standardization to the [0,1] interval provides a normalized unified comparison benchmark; the preset amplitude closeness interval defines the meaning of numerical "closeness"; and amplitude closeness provides a reliable quantitative indicator for subsequent response cancellation judgment.

[0067] The direction reversal rate and amplitude proximity are compared with the preset first judgment threshold and the preset second judgment threshold, respectively. If the direction reversal rate is greater than the preset first judgment threshold and the amplitude proximity is greater than the preset second judgment threshold, it is determined that the hydraulic aqueduct structure has a response cancellation situation in the bidirectional symmetrical window group.

[0068] The comparison of the direction reversal rate and amplitude similarity with preset first and second judgment thresholds is used to quantitatively determine whether the hydraulic aqueduct structure exhibits response cancellation within a specific time period. The specific implementation is as follows: First, two thresholds are set. The first judgment threshold sets the minimum requirement for the direction reversal rate, for example, 0.7, indicating that at least 70% of the data points are in opposite directions. The second judgment threshold sets the minimum standard for amplitude similarity, for example, 0.8, indicating that at least 80% of the data points are numerically similar. The analyzed direction reversal rate and amplitude similarity are compared with the two thresholds respectively. Only when both indicators exceed their respective thresholds is a response cancellation considered to have occurred. This "dual-indicator joint judgment" logic design effectively avoids misjudgments caused by a single feature. For example, although the directions are opposite, the numerical differences are large, or the values ​​are similar but the directions are mostly consistent; these will not be misidentified as response cancellation. For instance, in a window group, if the direction reversal rate is 0.82 and the amplitude similarity is 0.85, both exceeding the preset thresholds, then a response cancellation phenomenon is determined to exist in that window group. The preset first judgment threshold is a control line for the degree of directional opposition, while the second judgment threshold is a restriction standard for the degree of numerical approximation. The combination of the two can achieve accurate identification of rare but significant cancellation cases. This process is implemented through logical comparison, without the need for complex models, and features high computational efficiency and strong adaptability, laying a clear data screening foundation for subsequent identification processes.

[0069] S2. When the response cancellation occurs in the hydraulic aqueduct structure, determine the time period of response cancellation, and generate a multi-factor compression matrix based on the direction consistency factor, amplitude offset factor and energy residual factor to process the original response data in order to determine the normalized data of the hydraulic aqueduct structure under the response cancellation condition.

[0070] In this embodiment, S2 specifically includes the following steps:

[0071] S201. In the case of response cancellation in the hydraulic aqueduct structure, based on the joint judgment condition that the direction opposition rate of acceleration data and damping force data in the bidirectional symmetrical window group is greater than the preset first judgment threshold and the amplitude proximity is greater than the preset second judgment threshold, the time window that meets the joint condition is continuously detected on the time axis, and the time windows that are adjacent in time and both meet the joint judgment condition are merged to determine the time range that forms a continuous interval, which is the time period of response cancellation.

[0072] After a hydraulic aqueduct structure is determined to exhibit response cancellation, it is necessary to further clarify the duration of this phenomenon over time to avoid mistaking sporadic, occasional anomalies for genuine response cancellation. In implementation, firstly, based on the calculation results of a bidirectional symmetrical window group, the direction reversal rate and amplitude proximity are obtained for each time window, and these are compared with preset first and second judgment thresholds. Only when both the direction reversal rate and amplitude proximity exceed their respective thresholds within the same time window is the time window marked as exhibiting response cancellation characteristics. Subsequently, continuity checks are performed on these marked time windows along the time axis, i.e., checking whether these time windows are adjacent in time sequence or whether the interval is less than a predetermined allowable range. If multiple time windows meeting the conditions are consecutively arranged on the time axis, these time windows are merged into a single time interval, which is defined as the time period of response cancellation. For example, if the 5th, 6th and 7th time windows all simultaneously meet the criteria for opposite direction rate and similarity of amplitude, then these three consecutive windows are merged, and their corresponding time ranges constitute a complete response cancellation time period. Window that appears alone and does not form a continuity will not be included in this time period.

[0073] Continuity detection refers to the sequential relationship analysis of multiple time windows that meet the judgment criteria in the time dimension. The core is to determine whether these time windows are continuously or approximately continuously distributed on the time axis, thereby distinguishing between persistent response cancellation and instantaneous abnormal fluctuations. Continuity detection avoids misjudgments caused by noise interference or instantaneous abrupt changes, ensuring the temporal stability and consistency of the identified response cancellation. The time period of response cancellation refers to a time interval in which acceleration data and damping force data simultaneously exhibit opposite directions and similar amplitudes at most time points. This time period reflects the actual physical process of damping and structural inertia canceling each other out in the structural dynamic response. Limiting response cancellation to a continuous time period facilitates the subsequent extraction of direction consistency factors, amplitude offset factors, and energy residual factors based on this time period, thus ensuring that normalization processing and parameter identification are both based on a physically consistent time range. This approach enhances the reliability of data judgment in the time dimension and is an important foundation for achieving accurate parameter identification.

[0074] S202. During the time period of response cancellation, extract the direction consistency factor, amplitude offset factor and energy retention factor point by point for acceleration data and damping force data. The direction consistency factor is calculated by comparing the sign consistency between the direction trend at a local time point and the average direction trend of the time period. The amplitude offset factor is calculated by the normalized value of the difference between the current data point and the median amplitude in the time period. The energy retention factor is calculated by the proportion of the energy of the data point in the total energy of the time period.

[0075] S203. The directional consistency factor, amplitude offset factor and energy residual factor are combined in chronological order to form a multi-factor compression matrix. The original response data is then weighted and compressed based on this multi-factor compression matrix to generate normalized data with a uniform numerical scale under the condition of response cancellation, which is used for subsequent identification of seismic damper parameters.

[0076] In the case of response cancellation, to avoid abnormal data interfering with the subsequent identification of seismic damper parameters, the original response data needs to be processed specifically. First, the directional consistency factor, amplitude offset factor, and energy residual factor are arranged point-by-point in chronological order, forming a three-dimensional factor array. Each time point corresponds to three values, reflecting the intensity of expression in direction, amplitude, and energy characteristics at that point. By combining these three types of factors according to a weighting strategy, a matrix-structured weighted factor set is generated, which is the multi-factor compression matrix. Each matrix element corresponds to the comprehensive importance weight of a time point. Subsequently, the original acceleration data and damping force data are multiplied by the corresponding matrix weights to perform numerical compression. In this way, the parts of the data that have a real physical contribution to the response are retained or amplified, while noise components or data affected by response cancellation are compressed or weakened, thereby achieving effective information filtering.

[0077] In this process, the multi-factor compression matrix is ​​a key component, integrating information from multiple physical dimensions. This allows data of different properties to be uniformly mapped to a weighted expression space within the same timeframe. The raw response data refers to unprocessed acceleration and damping force signals, which are prone to extremely small amplitudes or unstable signs when responses cancel each other out. Weighted compression processing, through matrix operations, suppresses ineffective features and enhances effective features, achieving accurate reconstruction of the physical phenomena. The processed data is unified to a stable range on a numerical scale, facilitating subsequent model processing. This normalized data with a unified numerical scale not only preserves the information of the true physical response but also eliminates misleading factors caused by response cancellation, forming the core foundation for improving the accuracy of seismic damper parameter identification.

[0078] In this embodiment, S202 specifically refers to:

[0079] During the period of response cancellation, the local directional change trend of acceleration data and damping force data is extracted by sliding windows of fixed length. The directional trend of each window is obtained by calculating the sign change between adjacent data points in each window and comparing it with the overall directional average trend during the period of response cancellation. Data points with consistent signs are marked as consistent directional points. The directional consistency factor is the proportion of the number of consistent directional points to the total number of points.

[0080] To analyze the directional changes of acceleration and damping force data in different local regions during the response cancellation period, the raw data within this period needs to be locally divided into fixed-length sliding windows. Each sliding window contains several consecutive time points. By setting a constant number of data points, the granularity of analysis for each window is controlled, allowing the window to slide point by point on the time axis to cover the entire response cancellation period. Within each sliding window, the sign relationship between adjacent data points is compared to determine the directional trend of the data within the current window. For example, in a sliding window containing five data points, if the data sequence shows alternating positive and negative signs, it indicates a drastic fluctuation in the local directional trend; if the sign changes are minimal, it indicates a relatively stable directional trend. Subsequently, the directional trend within each sliding window is compared with the overall average directional trend of the response cancellation period. The overall average directional trend is obtained by statistically analyzing the directional changes among all data points over the entire period and serves as a benchmark for measuring the global trend. When the directional trend of a sliding window matches the sign of the overall average directional trend, the data points within that window are marked as consistent directional points. Finally, the proportion of all consistent directional points within the response cancellation period is calculated to obtain the directional consistency factor. A higher value for this factor indicates a stronger consistency between the local and overall directional trends, thus allowing for a higher confidence weight to be assigned to the data segment during subsequent normalization. This method effectively identifies which data regions exhibit stability in directional characteristics, providing crucial support for feature compression in response-cancelling scenarios.

[0081] The median of the amplitude is calculated for the acceleration and damping force data within the response cancellation time period. For each data point, the absolute value of the difference between it and the corresponding median is calculated. The normalized amplitude offset value is obtained by normalizing the maximum amplitude within the response cancellation time period. The amplitude offset factor is the normalized difference.

[0082] To measure the dispersion of each data point relative to the overall amplitude distribution within the response cancellation period, amplitude offset calculations need to be performed separately for acceleration and damping force data. First, all acceleration data within the response cancellation period are sorted, and the data in the middle position is taken as the median amplitude. The same processing is performed on the damping force data. The median is insensitive to outliers and can represent the central trend of the data. Next, for each data point, the absolute value of the difference between it and the corresponding amplitude median is calculated, reflecting the degree of deviation of that data point from the median level. To eliminate the influence of different units and dimensions, the absolute value of the difference is further normalized using the maximum amplitude within the period, mapping all offset values ​​to a uniform numerical range, so that the results are not affected by the original data scale. For example, if the maximum amplitude is a peak acceleration or damping force value, the difference between each data point and the median will be compressed to between zero and one. The normalized offset value is the normalized amplitude offset value; the larger this value, the more the amplitude at that point deviates from the overall trend. Finally, using this normalized value as an amplitude offset factor, anomaly labels are established point-by-point in the amplitude distribution of the data. In this way, it is possible to identify which points in the response cancellation data have a small impact on the identification results and should have their weight reduced, and which points have large deviations but are physically significant and should have their influence strengthened, thereby achieving more accurate data compression and feature extraction in subsequent multi-factor compression.

[0083] For each data point, calculate its instantaneous energy value and sum the energy of all data points within the response cancellation time period. Use the proportion of energy at a single point to the total energy as the energy retention factor, and apply it one-to-one with the direction consistency factor and amplitude offset factor to the construction of the subsequent multi-factor compression matrix.

[0084] To measure the contribution of each data point to the system's energy performance during the response cancellation period, instantaneous energy calculations need to be performed on both acceleration and damping force data. The instantaneous energy value refers to the physical energy intensity exhibited by acceleration or damping force at a specific point in time, typically measured by squared data amplitude at that point. This squared approach avoids the cancellation effect of positive and negative signs, more accurately reflecting the energy distribution of the data. After calculating the instantaneous energy values ​​for all data points within the entire response cancellation period, these values ​​are summed to obtain the total energy value for that period. Subsequently, the ratio of the instantaneous energy value of each data point to the total energy value is calculated to obtain the proportion of that point in the energy distribution; this proportion is the energy retention factor. A higher energy retention factor indicates a more prominent energy level at that point, representing a more significant response to the damper's action. Based on this, the direction consistency factor, amplitude offset factor, and energy retention factor can be correlated one-to-one, constructing a complete characteristic sequence. The significance of this is that even in an abnormal state where the response cancels out, data points with strong physical meaning can be retained through energy features, while redundant data is compressed. This provides a complete and accurate energy dimension input for the multi-factor compression matrix, improving the reliability and robustness of subsequent identification stages.

[0085] S3. Based on the determined normalized data, construct a set of input vectors, and calculate the direction residual term, amplitude rebound term and energy consumption delay term in sequence to generate a set of feature offset expressions.

[0086] In this embodiment, S3 specifically refers to:

[0087] Based on the determined normalized data, the acceleration data and damping force data are paired point by point in chronological order. The normalized acceleration data and normalized damping force data corresponding to the same time point are combined to form basic data pairs. Multiple consecutive basic data pairs are used to form input vectors. All input vectors are arranged in chronological order to construct a set of input vectors for identifying abnormal response characteristics.

[0088] After the normalized data is determined, it is necessary to pair the normalized acceleration data at each moment with the corresponding normalized damping force data, forming a set of basic data pairs, using time as the main thread. Specifically, the two data channels can be synchronously traversed in time sequence, and the two normalized values ​​at each moment can be packaged and combined into a two-dimensional vector to form a basic data pair. Subsequently, sliding combination is performed on multiple consecutive basic data pairs, for example, every 10 consecutive data pairs form an input vector, and these input vectors are arranged sequentially in time to form a complete set of input vectors. This constructed set of input vectors can simultaneously retain the local dynamic characteristics of the signal in time, direction, and amplitude, which can be used to identify abnormal response characteristics such as abrupt changes in direction, amplitude rebound, or energy hysteresis. This method can focus on extracting local behaviors in the time series, making feature analysis discriminative and sensitive, and improving the accuracy of parameter identification.

[0089] Normalized acceleration data and normalized damping force data refer to two types of response signals that have undergone normalization processing at a unified numerical scale. Their numerical ranges are standardized to the same interval to eliminate inconsistencies in the physical magnitude or units of the original signals. A basic data pair is a pair of normalized data points at the same time point, representing the dual-channel response state at that moment. An input vector is a composite vector body composed of a set of continuous basic data pairs, reflecting the joint evolution characteristics of the signal within a local time interval. The set of input vectors used to identify abnormal response features is a series of input vector sequences organized in chronological order. This set provides the contextual continuity and spatial distribution input basis for feature term calculation and is the core input source for subsequent feature offset modeling and parameter identification.

[0090] For each input vector in the input vector set, the following calculations are performed in chronological order: Based on the change in the sign of the normalized acceleration and damping force data between the current input vector and the previous input vector, the difference in direction change is calculated, and this difference in direction change is used as the direction residual term; Based on the change between the normalized amplitude in the current input vector and the normalized amplitude in the previous input vector, the magnitude of the amplitude regression from the deviation state to the center state is calculated, forming the amplitude rebound term; Based on the temporal position relationship between the energy change value corresponding to the current input vector and the energy change values ​​of multiple previous input vectors, the lag ratio of the energy change in the time series is calculated, forming the energy consumption delay term;

[0091] After the input vector set is constructed, each input vector needs to be analyzed individually and compared with its previous input vector in time to extract three types of features for modeling abnormal response behavior. First, in the calculation of the directional residual term, the normalized acceleration and normalized damping force data from the current and previous input vectors are extracted to determine their directional signs, and the trends of change in direction are compared, such as a change from positive to negative or vice versa, recording abrupt changes in sign. The difference is calculated and accumulated into a directional change difference, which is the directional residual term, used to measure the inconsistency of structural response direction switching. Second, in the calculation of the amplitude rebound term, the normalized amplitude from the current and previous input vectors is extracted, the difference is calculated, and it is determined whether the trend of change is regressing from the deviation from the center value to the center value. If a regression trend exists, the change in amplitude is used as the amplitude rebound term to characterize the vibration convergence behavior caused by the damping effect. Finally, in the calculation of the energy consumption delay term, the time distribution position between the energy change value corresponding to the current input vector and the energy change values ​​in its previous multiple input vectors is analyzed to determine whether the current energy release lags behind the previous stress accumulation. If there is a significant delay, its lag ratio in the time series is calculated to form the energy consumption delay term, thereby identifying the damper response lag phenomenon.

[0092] The current input vector and the previous input vector refer to two adjacent input vectors within a continuous time segment. Each segment contains a time interval combination of normalized acceleration and damping force data, reflecting the structure's response characteristics within a local time interval. The change in direction sign refers to the pattern of positive and negative signs of the acceleration and damping force values ​​in the two vectors, revealing abrupt changes in the signal's directional control. The difference in direction change is calculated by comparing the consistency of the directions of corresponding data points in the two vectors, serving as a numerical indicator of discontinuous directional response. The amplitude rebound term quantifies the regression phenomenon by showing the amplitude gradually approaching the average or median value; its core is to measure the system's ability to converge from a high-response state to a stable state. The energy change value represents the instantaneous energy release degree expressed by each input vector within its time segment, while the energy consumption delay term extracts the amplitude and location of the delay by comparing the temporal distribution of the energy release peaks, reflecting the damper's time offset in the response. The combination of these three features provides a high-dimensional input for identifying structural nonlinear responses, improving the accuracy and adaptability of parameter identification.

[0093] The direction residual, amplitude rebound, and energy consumption delay terms corresponding to each input vector are combined in chronological order to form a set of expressions that correspond one-to-one with the input vector, and this set of expressions is determined as the feature offset expression group.

[0094] To extract anomalous response features from the input vector set that can be used for subsequent seismic damper parameter identification, the directional residual, amplitude rebound, and energy dissipation delay terms calculated for each input vector need to be combined in chronological order. Specifically, the three feature values ​​corresponding to each input vector are integrated sequentially according to their arrangement on the time axis. The three values ​​at the same time point are packaged into a set of feature expressions. Each set of feature expressions consists of three parts: the directional residual, amplitude rebound, and energy dissipation delay, forming a complete response offset feature description unit. As the input vector sequence progresses, multiple such feature expression units can be continuously generated, ultimately combined in chronological order into a continuous set of expressions. This entire set of expressions constitutes the feature offset expression set, representing the nonlinear offset behavior trajectory of the structure over a continuous time range. Through this set of expressions, the anomalous patterns, energy dissipation lag, and stiffness rebound of the response in local time can be tracked, providing a data-driven high-dimensional feature foundation for subsequent damper parameter identification.

[0095] The feature offset expression set is a collection of response anomaly expression units mapped from the input vector set. Its core components include three types of feature terms: directional residual, amplitude rebound, and energy consumption delay. Each type of feature term relies on dynamic calculation of the normalized response data and the time series context. The directional residual term reflects the degree of directional inconsistency in the instantaneous response of the structure, revealing the phenomenon of frequent changes in excitation direction. The amplitude rebound term quantifies the trend of the structural response regressing to a steady state and is a key indicator for measuring the vibration recovery capability of the system. The energy consumption delay term identifies the time-series lag in energy release behavior, indicating whether the damper's energy dissipation response is timely and effective. These three types of features converge into a complete expression unit in the time dimension, corresponding one-to-one with the input vectors to form the dynamic feature trajectory of the structural response over multiple time windows. This expression set not only preserves the structural regularity of the original response data but also enhances the ability to identify nonlinear changing behaviors.

[0096] S4. Combine the characteristic offset expression group with the normalized data to generate an identification sequence containing energy dissipation density paths, nonlinear intersection intervals and stiffness reduction points. Use the identification sequence to identify the equivalent stiffness, energy dissipation capacity and delayed response parameters of the seismic damper.

[0097] In this embodiment, S4 specifically includes the following steps:

[0098] S401. Pair the characteristic offset expression group with the normalized data according to time points, and concatenate the direction residual term, amplitude rebound term and energy consumption delay term corresponding to each time point with the normalized acceleration data and normalized damping force data of that time point to construct a combined data sequence arranged in time order.

[0099] When processing the feature offset expression set and normalized data, the directional residual term, amplitude rebound term, and energy consumption delay term at each time point can be extracted using time as an index. These are then paired with the normalized acceleration and damping force data at the same time point, forming a joint data vector containing multidimensional information. This joint data vector is typically constructed through array concatenation, field mapping, or vector expansion to ensure a one-to-one correspondence between the response characteristics at each time point and the physical data. To maintain temporal continuity, all constructed joint data vectors need to be arranged chronologically according to the timeline of the original response data, ultimately forming a complete combined data sequence. For example, if at a certain time t1, the normalized acceleration is 0.6, the normalized damping force is 0.5, the directional residual is 0.2, the amplitude rebound is 0.4, and the energy consumption delay is 0.1, then the combined data vector at that time point is [0.6, 0.5, 0.2, 0.4, 0.1], and it is sequentially combined with the combined data at other time points to form a sequence for subsequent identification calculations.

[0100] Pairing refers to matching multiple feature parameters from different sources but with temporal correlation under a unified time index, ensuring that data from all dimensions are fused on a simultaneous time basis. Data stitching refers to merging multiple values ​​from different indicators at each time point into a new multidimensional data structure, enabling previously separate physical quantities and feature quantities to collaboratively characterize response behavior. A combined data sequence arranged in chronological order is a continuous data stream structure that retains the joint response information at each time point, possessing traceability and sequence consistency. This structure provides a unified input data foundation for subsequent identification of energy-consuming paths, nonlinear behavior, and stiffness degradation, serving as a crucial bridge connecting earlier multi-factor analysis with later identification models.

[0101] S402. Based on the change trajectory of normalized energy information and energy consumption delay term in the combined data sequence, calculate the unit energy distribution in the time dimension to form an energy consumption density path; in the combined data sequence, screen the data intervals where the amplitude rebound term and the normalized acceleration change rate simultaneously exceed their respective preset change thresholds at multiple consecutive time points, and determine the data interval as a nonlinear cross interval; compare the change amplitude of the directional residual term and the change amplitude of the normalized stiffness response in the combined data sequence, and mark the time points where they simultaneously exceed their respective change thresholds at the same time point as stiffness decreasing points. The energy consumption density path, nonlinear cross intervals, and stiffness decreasing points together constitute the identification sequence;

[0102] In the combined data sequence, normalized energy information and energy consumption delay term are key data reflecting the energy release characteristics of the system. By linking the energy consumption delay term and normalized energy at different time points, the degree of energy change per unit time at each time point can be calculated, and these can be integrated in chronological order to construct an energy density distribution sequence per unit time. This distribution can be used to identify the accumulation and release characteristics during the energy release process, ultimately forming a continuous energy consumption density path to characterize the actual performance of the damper in energy dissipation. Furthermore, the amplitude rebound term and normalized acceleration change rate are jointly monitored in the combined data sequence. By comparing the numerical changes of the two at multiple consecutive time points, time periods that simultaneously exceed their respective preset change thresholds are selected, and these time periods are marked as nonlinear cross-intervals where the structural response behavior exhibits significant nonlinear characteristics. To identify changes in structural stiffness, the change amplitudes of the directional residual term and normalized stiffness response also need to be analyzed synchronously. When the change amplitudes of both indicators at the same time point exceed their respective change thresholds, that time point can be determined as a stiffness decrease point. These three identification results together constitute a complete identification sequence, providing a precise location basis for the subsequent extraction of seismic damper parameters.

[0103] In the combined data sequence, normalized energy information reflects the absolute magnitude and trend of energy release under a unified numerical scale, while the energy dissipation delay term describes the relative lag in energy release. The joint trajectory of these two terms forms the core foundation for identifying energy behavior patterns. The energy dissipation density path is a curve structure that continuously records the energy distribution per unit time in the time dimension, used to quantify the energy dissipation capacity of the damper. Preset change thresholds refer to numerical limits set for different characteristic indicators, used to filter abnormal response events or critical behavior intervals. Nonlinear crossover intervals are time periods where abrupt changes occur simultaneously in multiple characteristic dimensions, indicating that the damper or structure exhibits coupled nonlinear behavior during this period. Corresponding change thresholds are sensitivity standards set for different indicators; for example, directional residuals and stiffness response changes each have independent upper limits of change. Stiffness decrease points are time nodes where structural stiffness characteristics undergo abrupt changes, often corresponding to locations where material or connecting component performance degrades. The identification sequence, as a time series collection summarizing energy dissipation, nonlinearity, and stiffness change characteristics, possesses high integration and serves as an important carrier connecting response characteristics with physical parameter identification.

[0104] S403. Calculate the energy dissipation capacity of the seismic damper based on the cumulative energy change of the energy dissipation density path in the identification sequence, calculate the delayed response parameters based on the time span and response asymmetry of the nonlinear intersection interval, and calculate the equivalent stiffness of the seismic damper based on the difference in the slope of the change between the normalized acceleration data and the normalized damping force data before and after the stiffness decrease point.

[0105] By calculating the energy dissipation capacity of seismic dampers based on the cumulative energy changes along the energy density path in the identified sequence, calculating the delayed response parameters based on the time span and response asymmetry of the nonlinear intersection intervals, and calculating the equivalent stiffness of the seismic dampers based on the difference in the slope of the normalized acceleration data and normalized damping force data before and after the stiffness reduction point, the actual working performance of seismic dampers during seismic excitation can be comprehensively characterized from three core dimensions: energy release, response delay, and stiffness degradation. This method can not only quantify the energy dissipation efficiency of the damper for the input seismic motion, but also reveal whether there are response offsets and structural hysteresis phenomena under strong earthquakes. Furthermore, by comparing the stiffness before and after the earthquake, it can assess whether the mechanical performance degrades during operation. The above three parameters are independently observable and mutually corroborative, enabling the construction of a precise identification framework for seismic damper performance from different perspectives. This provides a scientific, comprehensive, and quantifiable basis for subsequent structural health assessments, damper selection, and optimized control.

[0106] In this embodiment, S403 specifically refers to:

[0107] Based on the energy dissipation density path in the identification sequence, the energy distribution per unit time is accumulated along the time sequence, the energy increment in the continuous time period is summarized to form an energy accumulation curve, and the change amplitude of the energy accumulation curve in the corresponding time range of the identification sequence is used as the energy dissipation capacity characterization quantity of the seismic damper.

[0108] When processing the energy dissipation density path in the identification sequence, the energy distribution value per unit time at each time point is first extracted in chronological order. These energy distribution values ​​are then accumulated by summing the accumulated result of the previous time point with the unit energy value at the current time point, progressively forming an energy accumulation curve that grows over time. This curve records the total energy release from the initial time to any given time point, providing a direct representation of the energy dissipation capability of the seismic damper. In practical implementation, a fixed time sampling interval can be set, and the energy consumption data at each moment can be accumulated sequentially. The maximum-minimum difference of the curve over the entire identification time interval is used as the numerical representation of the energy release capability. This method effectively reflects the actual absorption and dissipation level of the damper under seismic disturbances.

[0109] Cumulative processing is a computational method that integrates energy change information from a time series to identify the overall trend of energy distribution. Energy increment refers to the unit energy change between two consecutive time points and is a fundamental building block for constructing the energy accumulation curve. The energy accumulation curve is a curve that continuously increases or stabilizes over time; its shape and slope directly reflect the rhythm and intensity of energy input and dissipation. The time range corresponding to the identified sequence refers to the time period involved in the analysis, typically including continuous time segments where response cancellation occurs. The magnitude of change refers to the total energy change between the start and end time points of the energy accumulation curve, usually expressed as the difference between the end and start points. Energy dissipation capacity is a numerical indicator used to quantitatively reflect the overall energy dissipation performance of the damper; its magnitude determines the efficiency and strength of the damper's control over the structural response. This method allows for the precise capture of the damper's core performance parameters, providing a basis for subsequent structural performance evaluation and damper design optimization.

[0110] For the nonlinear cross intervals identified in the identification sequence, the start and end positions of the interval on the time axis are extracted, the corresponding time span is calculated, and the positive and negative distribution of normalized acceleration data and normalized damping force data within the interval are statistically analyzed to obtain the degree of response asymmetry. Based on the combination of the time span and the degree of response asymmetry, the delayed response parameters of the seismic damper are determined.

[0111] In the identification sequence, when processing the identified nonlinear intersection intervals, it is first necessary to determine the start and end positions of the interval on the time axis and calculate its time span accordingly. Subsequently, normalized acceleration and normalized damping force data are extracted within this time period. By statistically analyzing the proportion of positive and negative values ​​within this time range, it is possible to identify whether there is a significant shift in the sign distribution of the data, thereby measuring the degree of difference in the system's response in the positive and negative directions, i.e., the degree of response asymmetry. For example, if within a nonlinear intersection interval, acceleration is mainly positive while damping force is mainly negative, it indicates a directional shift in the system response, potentially indicating hysteresis. By jointly considering the statistically obtained time span and the degree of response asymmetry, the extent to which the damping response lags behind the changes in structural inertial force can be quantified, thereby calculating the delayed response parameters of the seismic damper.

[0112] Nonlinear crossover intervals are data intervals selected from combined data sequences where the rate of change of acceleration and the magnitude rebound term both exceed their respective change thresholds at multiple consecutive time points. These intervals reflect the typical stage of a damper entering a nonlinear response state. Response asymmetry is used to assess the degree of imbalance between normalized acceleration data and normalized damping force data in the positive and negative directions, often measured by methods such as sign ratio difference or statistical skewness. Delayed response parameters are quantitative indicators describing the response lag of seismic dampers, reflecting the important performance characteristic of the damper's ability to provide timely reverse control force during earthquakes. By jointly considering the time span and response asymmetry, not only can the duration of the damper entering the nonlinear stage be identified, but also the specific degree of its response lag, which helps to accurately assess its time-dependent performance and seismic control capability.

[0113] Using the stiffness-decreasing points in the identification sequence as boundaries, normalized acceleration data and normalized damping force data are extracted from the adjacent time periods before and after each point. The corresponding slopes between the changes in acceleration and damping force in each time period are calculated, and the differences between the slopes are compared. The difference is used as the equivalent stiffness of the seismic damper.

[0114] At the stiffness-decreasing point identified in the sequence, this time point can be used as a boundary. A fixed-length segment of normalized acceleration and normalized damping force data is extracted from the time series before and after this point. Point-by-point paired analysis is then performed on the changes in acceleration and damping force within the two time periods. Within each time period, the changes in acceleration and damping force over time are calculated using time as the horizontal axis, and the slope of these changes is determined through linear fitting. This slope represents the correspondence between acceleration and damping force and is typically used to describe the stiffness level of a structure's response to an input seismic motion. Next, the slope values ​​obtained from the two time periods before and after the stiffness-decreasing point are compared. If the slope of the later period is significantly lower than that of the earlier period, it indicates that the damper's response capability decreases after this point. Therefore, the difference in slope before and after the point is used as an equivalent stiffness evaluation index for the seismic damper to determine whether its stiffness has changed due to energy dissipation or material nonlinear effects.

[0115] The slopes corresponding to the changes in acceleration and damping force over different time periods are derived by observing the response rate relationship of normalized response data within a certain time window. A higher slope indicates stronger response stiffness. The difference in slopes before and after the event reflects the evolution trend of the structure's response capability. If the slope decreases significantly in the later stage, it may mean that the damper's stiffness has decreased during continuous operation, possibly related to equipment aging, performance degradation, or hysteresis energy dissipation mechanisms. Equivalent stiffness is a summary parameter of the damper's actual response capability throughout the entire seismic response process, comprehensively reflecting the structure's true deformation control and recovery capabilities during an earthquake. The stiffness change information obtained by comparing the slopes before and after the event can not only help determine the current performance status of the damper but also provide a basis for subsequent maintenance decisions and parameter resetting.

[0116] S5. Construct a bias trend regulation expression based on the changing trend of the identification sequence and the response synchronization rate of the normalized data, and adjust the factors in the bidirectional symmetric window group and the multi-factor compression matrix in real time to perform dynamic regulation of the parameter identification process.

[0117] In this embodiment, S5 specifically refers to:

[0118] Based on the changing trend of the identification sequence, the distribution characteristics of the energy consumption density path, nonlinear intersection interval and stiffness decrease point in a continuous time period are extracted. Combined with the synchronous change ratio of acceleration data and damping force data in the normalized data, an offset trend control expression reflecting the dynamic relationship between the changing trend of the identification sequence and the response synchronization rate of the normalized data is constructed to characterize the degree of response consistency offset in the current parameter identification process.

[0119] In the process of identifying seismic damper parameters, it is necessary to extract the changing trends of three key features in the identification sequence—energy dissipation density path, nonlinear intersection intervals, and stiffness reduction points—on the time axis. This includes analyzing whether the distribution exhibits trends such as concentration, drift, increase, or decrease, and calculating the gradient and characteristic frequency of these changes in the time direction. Simultaneously, the synchronous changes of normalized acceleration data and normalized damping force data within the same time period need to be quantified. Specifically, this can be achieved by comparing the trend of the sign and amplitude changes of the two data sequences at each time point, and calculating the ratio of the number of synchronous changes to the total number of comparisons to obtain the response synchronization rate. The trend information of the identified sequence and the response synchronization rate are then input into the dynamic analysis module to construct a deviation trend control expression. This expression reflects the coupling relationship between the changing trend of the identified sequence and the degree of response synchronization, thereby dynamically determining whether the current data characteristics have deviated from the response consistency region assumed by the preset model.

[0120] The changing trend of the identification sequence refers to the change in the rate of energy accumulation per unit time in the energy consumption density path, the fluctuation of the distribution density and span of the nonlinear intersection interval on the time axis, and whether the stiffness decrease points are concentrated forward or backward. These changing directions together constitute the changing direction of the distribution characteristics. The response synchronization rate in the normalized data reflects the degree to which acceleration and damping force maintain coordinated changes in time sequence, and is an important indicator for measuring the consistency of input data. The offset trend control expression, as a mathematical mapping structure, integrates the above two types of information. Numerically, it can be used to guide subsequent analysis modules to determine whether the current response is in a stable identification range. If there is a large deviation, the parameter extraction strategy needs to be dynamically adjusted to improve the accuracy and adaptability of the identification process.

[0121] Based on the degree of response consistency shift presented in the offset trend regulation expression, the extension length and position of the bidirectional symmetric window group on the time axis are adjusted. Based on the statistical characteristics of direction change, amplitude difference and energy ratio in the normalized data of the current identification stage, the factor weights of the direction consistency factor, amplitude shift factor and energy residual factor in the multi-factor compression matrix are updated to form an adjusted input structure that conforms to the current response state.

[0122] When the offset trend adjustment expression shows a high degree of response consistency offset, it indicates that there is a temporal positioning error or feature mismatch between the response features of the current normalized data and the recognition sequence. To achieve higher precision parameter recognition, it is necessary to dynamically adjust the bidirectional symmetrical window group on the time axis. Specifically, this involves repositioning the start and end points of each time window to adapt to new changes in the recognition features. The implementation method is as follows: based on the direction of the offset trend change in the current time period, the time window is shifted and expanded by a certain step. Based on each adjusted window, the direction inversion rate and amplitude similarity of the data within the window are recalculated to determine whether the window adjustment has improved response consistency. This mechanism allows for flexible correction of the judgment interval for response cancellation periods, thereby enhancing the stability and accuracy of the recognition.

[0123] Simultaneously, to improve the ability of the multi-factor compression matrix to represent the current response state, it is necessary to update the weights of the direction consistency factor, amplitude offset factor, and energy residual factor based on the statistical characteristics extracted from the normalized data of the current stage. Direction change statistics can be obtained by statistically analyzing the proportion of consistent and inconsistent signs of normalized acceleration and damping force over a certain time period; amplitude difference statistics are constructed based on the mean or variance of amplitude offset at each time point; and energy ratio statistics are summarized by the ratio of instantaneous energy to total energy. Using these statistical characteristics as a basis, the weights of the original factors in the multi-factor compression matrix are proportionally adjusted. The updated weight structure will more accurately match the characteristic form of the current dynamic response, thereby improving the ability of the normalized data to represent the true response state and constructing an adjusted input structure that better reflects actual working conditions.

[0124] The real-time adjusted bidirectional symmetric window group and the updated multi-factor compression matrix are reapplied to the normalized data analysis process. The characteristic offset expression group and the identification sequence are iteratively reconstructed, and the seismic damper parameter identification process is dynamically controlled so that the identification model can continuously adapt to changes in response state.

[0125] To achieve dynamic control of the seismic damper parameter identification process, the real-time adjusted bidirectional symmetrical window group and the updated multi-factor compression matrix need to be reintroduced into the normalized data analysis process. Specifically, firstly, based on the adjusted time window, a new round of sliding extraction is performed on the acceleration and damping force data to reconstruct the window group data set used to determine the response cancellation state. Then, the updated multi-factor compression matrix is ​​used to perform a second weighted compression on these data sets to obtain new normalized data. Based on this normalized data, a new set of input vectors is generated sequentially, and the directional residual term, amplitude rebound term, and energy consumption delay term are recalculated to form a new set of characteristic offset expressions. Simultaneously, a new identification sequence is constructed by reassembling the characteristic offset expression set and the normalized data. The entire process iteratively reconstructs the original analysis results, enabling the identification process to continuously absorb the latest changes in response characteristics, achieving dynamic correction of the identification model and optimization of the parameter identification path.

[0126] In this process, the bidirectional symmetrical window group undertakes the core task of time positioning, controlling the start and end boundaries of data analysis; the direction consistency factor, amplitude offset factor, and energy residual factor in the multi-factor compression matrix redefine the contribution value of each data point in the normalization process through adjusted weights; the normalized data is a standardized response input formed under the action of a specific compression factor, serving as the basic input source for feature analysis; the feature offset expression group is a set of feature mapping structures dynamically generated based on the normalized data, used to characterize the response characteristics in terms of direction, amplitude, and energy; the identification sequence is a structure built on the correlation between feature expression and temporal continuity, used to calibrate key indicators such as energy dissipation capacity, delayed response, and stiffness change. Through continuous iterative reconstruction, adaptive adjustment of the parameter identification path can be achieved, thereby improving the accuracy and robustness of the seismic damper identification model under seismic conditions.

[0127] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0128] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0129] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0130] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0131] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0132] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0133] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A data-driven method for identifying parameters of seismic dampers in hydraulic aqueducts, characterized in that, Specifically, the following steps are included: S1. Divide the acceleration data and damping force data of the hydraulic aqueduct under seismic excitation into continuous time windows, and construct a bidirectional symmetrical window group with each time window as the center. Calculate the direction reversal rate and amplitude closeness between the acceleration data and damping force data in each group to determine whether the hydraulic aqueduct structure exhibits response cancellation. S2. When the response cancellation occurs in the hydraulic aqueduct structure, determine the time period of response cancellation, and generate a multi-factor compression matrix based on the direction consistency factor, amplitude offset factor and energy residual factor to process the original response data in order to determine the normalized data of the hydraulic aqueduct structure under the response cancellation condition. S2 specifically includes the following steps: S201. In the case of response cancellation in the hydraulic aqueduct structure, based on the joint judgment condition that the direction opposite rate of acceleration data and damping force data in the bidirectional symmetrical window group is greater than the preset first judgment threshold and the amplitude closeness is greater than the preset second judgment threshold, the time window that meets the joint judgment condition is continuously detected on the time axis, and the time windows that are adjacent in time and both meet the joint judgment condition are merged to determine the time range that forms a continuous interval, which is the time period of response cancellation. S202. During the time period of response cancellation, extract the direction consistency factor, amplitude offset factor and energy retention factor point by point for acceleration data and damping force data. The direction consistency factor is calculated by comparing the sign consistency between the direction trend at a local time point and the average direction trend of the time period. The amplitude offset factor is calculated by the normalized value of the difference between the current data point and the median amplitude in the time period. The energy retention factor is calculated by the proportion of the energy of the data point in the total energy of the time period. S203. Combine the direction consistency factor, amplitude offset factor and energy residual factor in time order to form a multi-factor compression matrix, and perform weighted compression processing on the original response data based on the multi-factor compression matrix to generate normalized data with a uniform numerical scale under the condition of response cancellation, which is used for subsequent identification of seismic damper parameters. S3. Based on the determined normalized data, construct a set of input vectors, and calculate the direction residual term, amplitude rebound term and energy consumption delay term in sequence to generate a set of feature offset expressions. S4. Combine the characteristic offset expression group with the normalized data to generate an identification sequence containing energy dissipation density paths, nonlinear intersection intervals and stiffness reduction points. Use the identification sequence to identify the equivalent stiffness, energy dissipation capacity and delayed response parameters of the seismic damper. S4 specifically includes the following steps: S401. Pair the characteristic offset expression group with the normalized data according to time points, and concatenate the direction residual term, amplitude rebound term and energy consumption delay term corresponding to each time point with the normalized acceleration data and normalized damping force data of that time point to construct a combined data sequence arranged in time order. S402. Based on the change trajectory of normalized energy information and energy consumption delay term in the combined data sequence, calculate the unit energy distribution in the time dimension to form an energy consumption density path; in the combined data sequence, screen the data intervals where the amplitude rebound term and the normalized acceleration change rate simultaneously exceed their respective preset change thresholds at multiple consecutive time points, and determine the data interval as a nonlinear cross interval; compare the change amplitude of the directional residual term and the change amplitude of the normalized stiffness response in the combined data sequence, and mark the time points where they simultaneously exceed their respective change thresholds at the same time point as stiffness decreasing points. The energy consumption density path, nonlinear cross intervals, and stiffness decreasing points together constitute the identification sequence; S403. Calculate the energy dissipation capacity of the seismic damper based on the cumulative energy change of the energy dissipation density path in the identification sequence, calculate the delayed response parameters based on the time span and response asymmetry of the nonlinear intersection interval, and calculate the equivalent stiffness of the seismic damper based on the difference in the slope of the change between the normalized acceleration data and the normalized damping force data before and after the stiffness decrease point. S5. Construct a bias trend regulation expression based on the changing trend of the identification sequence and the response synchronization rate of the normalized data, and adjust the factors in the bidirectional symmetric window group and the multi-factor compression matrix in real time to perform dynamic regulation of the parameter identification process.

2. The data-driven method for identifying seismic damper parameters in hydraulic aqueducts according to claim 1, characterized in that, S1 specifically includes the following steps: S101. The acceleration and damping force data of the hydraulic aqueduct under seismic excitation are divided into continuous time windows. The original data is segmented by setting equal time intervals so that each time window contains the same number of acceleration and damping force data, ensuring that the time series is complete, continuous and synchronous. S102. Taking each time window as the center, extend the same number of consecutive time windows forward and backward along the time axis to construct a bidirectional symmetrical window group centered on the current time window. Then, rearrange the acceleration data and damping force data in the bidirectional symmetrical window group according to the chronological order to form a standard input data sequence for subsequent judgment. S103. For each bidirectional symmetrical window group, compare the sign relationship between acceleration data and damping force data at the same time point, count the proportion of data points with opposite signs to the total data points, calculate the direction reversal rate, and simultaneously calculate the normalized value of amplitude difference between each pair of data points, count the proportion of them falling into the preset amplitude proximity interval, calculate the amplitude proximity degree, and determine whether the hydraulic aqueduct structure has a response cancellation situation based on the joint condition that the direction reversal rate is greater than the preset first judgment threshold and the amplitude proximity degree is greater than the preset second judgment threshold.

3. The data-driven method for identifying seismic damper parameters in hydraulic aqueducts according to claim 2, characterized in that, S103 specifically refers to: For each bidirectional symmetrical window group, the signs of acceleration data and damping force data at the same time point are compared point by point. The product sign judgment method is used to mark the data points with negative product as opposite directions. The ratio of the number of data points with opposite directions to the total number of data points in the window group is calculated to obtain the opposite direction rate. For each pair of data points, the absolute value of the difference between the acceleration data and the damping force data is calculated. The difference value is then normalized to the [0,1] interval by the maximum amplitude normalization process. The ratio of the number of data points whose difference value falls into the preset amplitude close interval after normalization to the total number of data points is calculated to obtain the amplitude closeness. The direction reversal rate and amplitude proximity are compared with the preset first judgment threshold and the preset second judgment threshold, respectively. If the direction reversal rate is greater than the preset first judgment threshold and the amplitude proximity is greater than the preset second judgment threshold, it is determined that the hydraulic aqueduct structure has a response cancellation situation in the bidirectional symmetrical window group.

4. The data-driven method for identifying seismic damper parameters in hydraulic aqueducts according to claim 1, characterized in that, S202 specifically refers to: During the period of response cancellation, the local directional change trend of acceleration data and damping force data is extracted by sliding windows of fixed length. The directional trend of each window is obtained by calculating the sign change between adjacent data points in each window and comparing it with the overall directional average trend during the period of response cancellation. Data points with consistent signs are marked as consistent directional points. The directional consistency factor is the proportion of the number of consistent directional points to the total number of points. The median of the amplitude is calculated for the acceleration and damping force data within the response cancellation time period. For each data point, the absolute value of the difference between it and the corresponding median is calculated. The normalized amplitude offset value is obtained by normalizing the maximum amplitude within the response cancellation time period. The amplitude offset factor is the normalized difference. For each data point, calculate its instantaneous energy value and sum the energy of all data points within the response cancellation time period. Use the proportion of energy at a single point to the total energy as the energy retention factor, and apply it one-to-one with the direction consistency factor and amplitude offset factor to the construction of the subsequent multi-factor compression matrix.

5. The data-driven method for identifying seismic damper parameters in hydraulic aqueducts according to claim 1, characterized in that, S3 specifically refers to: Based on the determined normalized data, the acceleration data and damping force data are paired point by point in chronological order. The normalized acceleration data and normalized damping force data corresponding to the same time point are combined to form basic data pairs. Multiple consecutive basic data pairs are used to form input vectors. All input vectors are arranged in chronological order to construct a set of input vectors for identifying abnormal response characteristics. For each input vector in the input vector set, the following calculations are performed in chronological order: Based on the change in the direction sign of the normalized acceleration data and normalized damping force data in the current input vector and the previous input vector, the difference in direction change is calculated, and this difference in direction change is used as the direction residual term; Based on the change in the normalized amplitude in the current input vector and the normalized amplitude in the previous input vector, the magnitude of the amplitude regression from the deviation state to the center state is calculated, forming the amplitude rebound term; Based on the temporal positional relationship between the energy change value corresponding to the current input vector and the energy change values ​​of multiple previous input vectors, the lag ratio of energy change in the time series is calculated to form an energy consumption delay term. The direction residual, amplitude rebound, and energy consumption delay terms corresponding to each input vector are combined in chronological order to form a set of expressions that correspond one-to-one with the input vector, and this set of expressions is determined as the feature offset expression group.

6. The data-driven method for identifying seismic damper parameters in hydraulic aqueducts according to claim 1, characterized in that, S403 specifically refers to: Based on the energy dissipation density path in the identification sequence, the energy distribution per unit time is accumulated along the time sequence, the energy increment in the continuous time period is summarized to form an energy accumulation curve, and the change amplitude of the energy accumulation curve in the corresponding time range of the identification sequence is used as the energy dissipation capacity characterization quantity of the seismic damper. For the nonlinear cross intervals identified in the identification sequence, the start and end positions of the interval on the time axis are extracted, the corresponding time span is calculated, and the positive and negative distribution of normalized acceleration data and normalized damping force data within the interval are statistically analyzed to obtain the degree of response asymmetry. Based on the combination of the time span and the degree of response asymmetry, the delayed response parameters of the seismic damper are determined. Using the stiffness-decreasing points in the identification sequence as boundaries, normalized acceleration data and normalized damping force data are extracted from the adjacent time periods before and after each point. The corresponding slopes between the changes in acceleration and damping force in each time period are calculated, and the differences between the slopes are compared. The difference is used as the equivalent stiffness of the seismic damper.

7. The data-driven method for identifying seismic damper parameters in hydraulic aqueducts according to claim 1, characterized in that, S5 specifically refers to: Based on the changing trend of the identification sequence, the distribution characteristics of the energy consumption density path, nonlinear intersection interval and stiffness decrease point in a continuous time period are extracted. Combined with the synchronous change ratio of acceleration data and damping force data in the normalized data, an offset trend control expression reflecting the dynamic relationship between the changing trend of the identification sequence and the response synchronization rate of the normalized data is constructed to characterize the degree of response consistency offset in the current parameter identification process. Based on the degree of response consistency shift presented in the offset trend regulation expression, the extension length and position of the bidirectional symmetric window group on the time axis are adjusted. Based on the statistical characteristics of direction change, amplitude difference and energy ratio in the normalized data of the current identification stage, the factor weights of the direction consistency factor, amplitude shift factor and energy residual factor in the multi-factor compression matrix are updated to form an adjusted input structure that conforms to the current response state. The real-time adjusted bidirectional symmetric window group and the updated multi-factor compression matrix are reapplied to the normalized data analysis process. The characteristic offset expression group and the identification sequence are iteratively reconstructed, and the seismic damper parameter identification process is dynamically controlled so that the identification model can continuously adapt to changes in response state.

Images