Reservoir dam monitoring and early warning method and system based on multi-source information fusion

Abstract

The invention discloses a reservoir dam monitoring and early warning method and system based on multi-source information fusion, and relates to the technical field of early warning and monitoring. The method comprises the following steps: acquiring seepage flow, displacement rate and pore water pressure data through a sensor array; generating standardized data through Kalman filtering noise reduction; extracting key monitoring parameters through grey correlation analysis; aligning a main frequency amplitude sequence with acceleration variation to generate a deformation feature vector; and when the risk reaches a threshold value, inputting logistic regression to generate a coupling early warning level and feeding back to the terminal. According to the method, through multi-source data Kalman filtering noise reduction standardization, grey correlation screening high correlation parameter construction features, Fourier transform extraction of dominant frequency and acceleration alignment, fuzzy evaluation fusion of dominant frequency attenuation and accumulated acceleration quantization deformation, logic regression based on mutation and fluctuation grading early warning, and space-time coupling, hidden danger is captured. Non-linear mapping breaks through empirical limitation, and multi-dimensional synchronization improves evaluation precision and reduces false alarm and missing alarm.

Description

Technical Field

[0001] This invention relates to the field of early warning and monitoring technology, and in particular to a method and system for monitoring and early warning of reservoir dams based on multi-source information fusion. Background Technology

[0002] The field of early warning and monitoring technology primarily involves using monitoring systems and related technologies to promptly identify potential risks and issue early warning signals, enabling effective measures to be taken to prevent catastrophic events. Core aspects of this field include meteorological monitoring, disaster early warning, and environmental monitoring technologies, which are widely applied in the construction of monitoring and early warning systems for natural disasters such as earthquakes, climate change, floods, and landslides. In water conservancy projects, early warning and monitoring technology is mainly used for stability monitoring of facilities such as dams and reservoirs to safeguard people's lives and property.

[0003] Traditional reservoir dam monitoring and early warning methods involve routine monitoring of the dam using existing equipment and technologies. This includes real-time data collection from sensors measuring water level, flow rate, soil pressure, and temperature. The data is transmitted to a monitoring center via wired or wireless means, where experts assess the dam's safety based on the collected data. Traditional methods primarily employ pressure sensor monitoring, deformation monitoring, and temperature monitoring. These methods allow for the periodic analysis of monitoring data to identify potential hazards in the dam in advance.

[0004] Traditional methods rely on manual periodic analysis of discrete parameters. Data timeliness is limited by the stability of wired transmission. Independent analysis of a single sensor ignores the synergistic effect between parameters. Fixed threshold warnings cannot respond to dynamic environmental changes. Experience-based assessments are easily affected by subjective factors. Conventional filtering is insufficient to suppress non-stationary noise. Deformation feature extraction lacks the spatiotemporal correlation between the main frequency decay and acceleration. Periodic checks miss sudden structural changes. Asynchronous data leads to biases in comprehensive analysis. Lagging warnings make it difficult to trigger emergency responses in a timely manner. Summary of the Invention

[0005] To address the technical problems of traditional methods that rely on periodic manual analysis of discrete parameters, where data timeliness is limited by the stability of wired transmission, single-sensor independent analysis ignores the synergistic effects between parameters, fixed threshold warnings cannot respond to dynamic environmental changes, experience-based assessments are susceptible to subjective interference, conventional filtering is insufficient to suppress non-stationary noise, deformation feature extraction lacks spatiotemporal correlation between dominant frequency attenuation and acceleration, periodic checks miss sudden structural changes, asynchronous data leads to comprehensive analysis bias, and delayed warnings fail to trigger timely emergency responses, this invention provides a reservoir dam monitoring and early warning method and system based on multi-source information fusion. The technical solution is as follows:

[0006] On the one hand, a method for monitoring and early warning of reservoir dams based on multi-source information fusion is provided, which includes: S1: Acquire time-series data of seepage flow, displacement rate data and pore water pressure data by sensor arrays deployed on the dam body, and use Kalman filtering algorithm to denoise the raw data to generate a standardized monitoring dataset; S2: Call the standardized monitoring dataset, input the seepage flow time series data and displacement rate data into the grey relational analysis model to calculate the correlation degree, filter parameters with a correlation degree exceeding the correlation threshold, and form a set of key monitoring parameters; S3: Based on the set of key monitoring parameters, perform discrete Fourier transform on the pore water pressure data to extract the main frequency amplitude sequence, and combine it with the acceleration change of the displacement rate data to align the timestamps and generate the dam deformation feature vector. S4: Input the dam deformation feature vector into the fuzzy comprehensive evaluation method, calculate the cumulative value of the attenuation rate of the main frequency amplitude sequence and the acceleration change, and output the deformation risk level; S5: Call the deformation risk level. When the risk level reaches the risk threshold, input the number of mutation points in the main frequency amplitude sequence and the fluctuation amplitude of the acceleration change into the logistic regression classifier to generate a coupled early warning level and feed it back to the monitoring terminal.

[0007] As a further aspect of the present invention, the standardized monitoring dataset is obtained by denoising the seepage flow time series data, displacement rate data, and pore water pressure data using the Kalman filter algorithm. The key monitoring parameter set is composed of seepage flow correlation parameters and displacement rate correlation parameters whose correlation exceeds a preset correlation threshold, selected based on a grey relational analysis model. The dam deformation feature vector is formed by combining the main frequency amplitude spectrum sequence extracted from the pore water pressure data by discrete Fourier transform with the acceleration change of the displacement rate data after timestamp alignment. The deformation risk level is generated based on the main frequency energy attenuation rate and the cumulative displacement acceleration change through a fuzzy comprehensive evaluation method. The coupling early warning level is calculated by the number of main spectrum mutation points and the acceleration standard deviation parameter through a logistic regression classifier.

[0008] As a further aspect of the present invention, the specific steps of S1 include: S101: Collect time-series data of seepage flow, displacement rate and pore water pressure by sensor arrays deployed on the dam body, establish a data index structure, adjust the consistency of time arrangement, unify the sampling period and identify missing segments, and generate the original dataset of sensor monitoring. S102: Based on the original dataset monitored by the sensor, the Kalman filter algorithm is called to set the estimation matrix and error parameters. Three types of original values ​​are input sequentially to perform filtering updates and remove noise, and a time series filtered estimation value set is obtained. The Kalman filter algorithm is optimized based on a multi-source collaborative calibration mechanism to perform unified noise reduction processing on multi-dimensional raw data of seepage, displacement and pressure. S103: Based on the time series filtered estimate set, extract multiple types of data sequences, rearrange the structure according to function, unify the units, and perform normalization processing using linear interval compression to generate a standardized monitoring dataset.

[0009] As a further aspect of the present invention, the specific steps of S2 include: S201: Call the seepage flow and displacement rate data in the standardized monitoring dataset, perform synchronous matching and spline interpolation to complete the data, identify the missing sections and set the interpolation boundary conditions, calculate the section statistical indicators, and obtain the aligned time series characteristic values; The aligned temporal feature values ​​refer to the key numerical sequences reflecting the changing trends and correlation characteristics extracted from multi-source monitoring data after time-uniform alignment and interpolation completion. S202: Based on the aligned temporal feature values, extract the difference in average values ​​of segments, perform range normalization, calculate the gray correlation coefficient using the correlation coefficient formula and summarize it to obtain the gray correlation degree between parameters. S203: Based on the gray correlation degree between the parameters, compare the correlation degree of the segments with the correlation threshold, construct a mapping table of the correspondence between parameters and segments, filter out parameters that meet the conditions and bind them to the segment positions, and generate a set of key monitoring parameters. The association threshold is a critical value used to screen the strength of the association between parameters in grey relational analysis. When the association degree exceeds the association threshold, it is judged as a significant association. Based on the modeling and analysis of measured data from three reservoirs in a certain province from 2018 to 2022, it is shown that setting the threshold range to [0.75, 0.9] can obtain the optimal classification effect.

[0010] As a further aspect of the present invention, the specific steps of S3 include: S301: Based on the raw pore water pressure data in the set of key monitoring parameters, the main frequency component is extracted using discrete Fourier transform, the amplitude of the frequency interval is screened and sorted, and a main frequency amplitude sequence is generated. S302: Call the main frequency amplitude sequence and displacement rate data to obtain the displacement rate sequence of the measuring point within the monitoring time period, calculate the acceleration change value by difference between adjacent time points, compare the acceleration change time index and amplitude timestamp, extract matching measuring point information, and generate an aligned time feature set; S303: Based on the data from multiple measurement points in the alignment time feature set, statistically analyze the time period differences and calculate the amplitude increase / decrease rate and acceleration fluctuation rate, perform linear standardization and combination processing, and generate the dam deformation feature vector.

[0011] As a further aspect of the present invention, the specific steps of S4 include: S401: Based on the dam deformation feature vector, call the main frequency amplitude sequence data, detect the main frequency amplitude change sequence over time, calculate the ratio of the main frequency amplitude difference between adjacent time nodes to the main frequency amplitude of the previous node, and generate the main frequency attenuation rate value sequence. S402: Call the main frequency attenuation rate value sequence and the acceleration change data in the dam deformation feature vector, extract the acceleration change value at each time node, calculate the total acceleration change from the start to the current time node, and generate the cumulative acceleration change value; S403: Call the cumulative change value of acceleration and the sequence of main frequency attenuation rate values, take them as input items, substitute them into the fuzzy comprehensive evaluation method, perform level assignment mapping based on the risk membership function and level discrimination threshold, and obtain the deformation risk level; The fuzzy comprehensive evaluation method transforms the main frequency attenuation rate and cumulative acceleration change into a fuzzy matrix through membership functions and weights, calculates the corresponding risk level, and realizes a multi-level fusion judgment mechanism.

[0012] As a further aspect of the present invention, the cumulative change in acceleration is calculated using the following formula: ; in, Represents the time from the start to the current time node. The cumulative change in acceleration, Representing the The change in dam acceleration at each time point relative to the previous time point. Representing the Normalized dam body acceleration changes at each time point Representing the The main frequency attenuation rate value at each time point Representing the The main frequency attenuation rate value at each time point Number the current time node.

[0013] As a further aspect of the present invention, the specific steps of S5 include: S501: Call the deformation risk level and determine whether its corresponding value reaches the risk threshold. If the condition is met, obtain the main frequency amplitude sequence in the current monitoring period, calculate the slope fluctuation trend based on its rising and falling inflection point position, extract the number of abrupt change points, monitor the acceleration change and calculate the fluctuation amplitude with the offset of adjacent time periods, and generate frequency amplitude change feature quantity. The risk threshold is a critical standard for judging whether the monitoring data has reached the level that requires early warning. If it is exceeded, the subsequent analysis and early warning mechanism will be triggered. It is determined by comparing the frequency distribution of dam displacement and pore pressure response under different monitoring conditions. S502: Based on the number of abrupt change points and the acceleration fluctuation amplitude in the frequency amplitude change feature quantity, construct a two-dimensional feature vector and input it into the logistic regression classifier. Perform linear combination and normalization processing using the regression coefficient matrix and bias term to generate a coupled early warning probability value. S503: Based on the coupled warning probability value and the predetermined coupled warning level boundary standard, divide the interval, select the preset level label according to the corresponding interval, establish the level field and monitoring terminal structure, and send the feedback instruction to obtain the coupled warning level.

[0014] As a further aspect of the present invention, the number of mutation points is calculated using the following formula: ; in, Represents the current monitoring period The number of mutation points parameter, Represents the current monitoring period The Middle The relative change in the amplitude of the dominant frequency at each inflection point The reference value representing the amplitude of the main frequency. Represents the current monitoring period The Middle The slope value of the main frequency amplitude corresponding to each inflection point Represents the current monitoring period The average slope of the dominant frequency amplitude at the inner inflection point The benchmark value representing the change in the dominant frequency slope is taken as the standard deviation of the original sample data. Representing the The confidence weight coefficients for each inflection point are calculated based on the Bayesian posterior scoring model for mutation points. It is a small positive constant used to avoid the denominator being zero. Indicates the current monitoring period The total number of inflection points detected within the system. The variance index representing the distribution of the dominant frequency energy during the monitoring period t is used as an enhancement factor.

[0015] On the other hand, a reservoir dam monitoring and early warning system based on multi-source information fusion is provided. This system is used to execute the aforementioned reservoir dam monitoring and early warning method based on multi-source information fusion. The system includes: The data fusion processing module is used to acquire time-series data of seepage flow, displacement rate data and pore water pressure data through the deployed sensor array, and to perform noise reduction processing on the raw data using the Kalman filter algorithm to generate a standardized monitoring dataset, which is then transmitted to the parameter association and filtering module. The parameter association and filtering module is used to call the standardized monitoring dataset, input the seepage flow time series data and displacement rate data into the grey relational analysis model to calculate the correlation degree, filter parameters whose correlation degree exceeds the correlation threshold, form a set of key monitoring parameters, and pass them to the deformation feature construction module. The deformation feature construction module is used to call the set of key monitoring parameters, perform discrete Fourier transform on the pore water pressure data to extract the main frequency amplitude sequence, combine the acceleration change of the displacement rate data with the timestamp alignment, generate the dam body deformation feature vector, and pass it to the risk level assessment module. The risk level assessment module is used to input the dam deformation feature vector into the fuzzy comprehensive evaluation method, calculate the cumulative value of the attenuation rate of the main frequency amplitude sequence and the acceleration change, output the deformation risk level, and transmit it to the early warning decision generation module. The early warning decision generation module is used to input the number of mutation points in the main frequency amplitude sequence and the fluctuation amplitude of the acceleration change into the logistic regression classifier when the deformation risk level reaches the risk threshold, generate a coupled early warning level, and feed the coupled early warning level back to the monitoring terminal.

[0016] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: In this embodiment of the invention, multi-source heterogeneous monitoring data (including dam seepage flow, displacement rate, and pore water pressure) are denoised using Kalman filtering to form a standardized dataset suitable for dam structural health assessment; grey relational analysis dynamically filters key parameters highly correlated with dam seepage anomalies and structural displacement changes, constructing a feature vector reflecting the stress-seepage coupling state of the dam; discrete Fourier transform extracts the dominant frequency amplitude sequence reflecting the periodic changes in pore water pressure, and timestamps it in conjunction with the acceleration change of displacement rate to accurately capture the deformation response of local dam sections under high water level or rapid adjustment conditions; fuzzy comprehensive evaluation integrates the dominant frequency attenuation rate and the cumulative displacement acceleration change, quantifying... The system optimizes the overall deformation index of the dam body to achieve a scientific assessment of the damage trend of the dam structure; a logistic regression classifier generates a dynamic early warning level reflecting the safety status of the dam structure based on the abrupt change point of the dominant frequency amplitude and the acceleration fluctuation amplitude; coupled analysis of time series data and spatial features ensures early identification of hidden dangers such as concentrated seepage and local structural instability; a nonlinear risk mapping model overcomes the limitations of traditional experience-based judgment in complex environmental conditions during reservoir operation; synchronous processing of multi-dimensional hydraulic monitoring data significantly improves the assessment accuracy under extreme floods, rapid regulation, and long-term fatigue; and a graded early warning mechanism effectively reduces the probability of false alarms and missed alarms in dam safety management, ensuring operational safety during the flood season and under sudden conditions. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the workflow of the present invention; Figure 2 This is a system flowchart of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] Please see Figure 1 This invention provides a method for monitoring and early warning of reservoir dams based on multi-source information fusion. The processing flow of this method may include the following steps: S1: Acquire time-series data of seepage flow, displacement rate data and pore water pressure data by sensor arrays deployed on the dam body, and use Kalman filtering algorithm to denoise the raw data to generate a standardized monitoring dataset; S2: Call the standardized monitoring dataset, input the time series data of seepage flow and displacement rate data into the grey relational analysis model to calculate the correlation degree, filter the parameters whose correlation degree exceeds the correlation threshold, and form a set of key monitoring parameters; S3: Based on the set of key monitoring parameters, the discrete Fourier transform of the pore water pressure data is used to extract the main frequency amplitude sequence, and the acceleration change of the displacement rate data is combined with the timestamp alignment to generate the dam deformation feature vector. S4: Input the dam deformation feature vector into the fuzzy comprehensive evaluation method, calculate the cumulative value of the attenuation rate of the main frequency amplitude sequence and the acceleration change, and output the deformation risk level; S5: Call the deformation risk level. When the risk level reaches the risk threshold, input the number of mutation points in the main frequency amplitude sequence and the fluctuation amplitude of the acceleration change into the logistic regression classifier to generate the coupling warning level and feed it back to the monitoring terminal. The standardized monitoring dataset is obtained by denoising the time-series data of seepage flow, displacement rate, and pore water pressure using the Kalman filter algorithm. The set of key monitoring parameters consists of seepage flow correlation parameters and displacement rate correlation parameters whose correlation exceeds a preset correlation threshold, selected based on a grey relational analysis model. The dam deformation feature vector is formed by combining the main frequency amplitude spectrum sequence extracted from the pore water pressure data by discrete Fourier transform with the acceleration change of the displacement rate data after timestamp alignment. The deformation risk level is generated by the fuzzy comprehensive evaluation method based on the main frequency energy attenuation rate and the cumulative displacement acceleration change. The coupling warning level is calculated by the number of main spectrum mutation points and the acceleration standard deviation parameter using a logistic regression classifier.

[0024] Specifically, the steps in S1 are as follows: S101: Collect time-series data of seepage flow, displacement rate and pore water pressure by sensor arrays deployed on the dam body, establish a data index structure, adjust the consistency of time arrangement, unify the sampling period and identify missing segments, and generate the original dataset of sensor monitoring. The data collection process involves using a sensor array deployed along the dam body to acquire time-series data on seepage flow, displacement rate, and pore water pressure. First, the locations for the sensor array within the dam body need to be clearly defined. Monitoring points at different depths upstream, downstream, and central of the dam body are selected, with seepage sensors, displacement sensors, and pore water pressure sensors installed at distances of 10m, 30m, and 50m from the dam crest, respectively. During deployment, geological profiles are analyzed for different monitoring points. Soil structure, pore distribution, and groundwater channels are obtained through borehole drilling. Pore water pressure sensors are deployed in areas with concentrated pore water pressure to ensure the relevance and representativeness of the collected data. During data acquisition, sensors are numbered and time-synchronized. For example, seepage sensor A collects data every 5 minutes, displacement sensor B every 10 minutes, and pore water pressure sensor C every 2 minutes. For data sequences with different sampling periods, interpolation is used to adjust the time series, uniformly resampling the data to every 2 minutes. During this process, for any missing data segments in the original data sequence, a complete scan of the acquisition timeline is performed. The system determines whether there are consecutive gaps exceeding 5 minutes. If so, it uses linear interpolation to fill in the missing data. For example, if seepage flow data is missing at t=10 minutes, and data values ​​of 4.2 m³ / h and 4.8 m³ / h are collected at t=8 minutes and t=12 minutes respectively, the filled data value is calculated linearly: (4.2 + 4.8) / 2 = 4.5 m³ / h. For displacement rate data, if missing, the average rate of the preceding and following data points is used for filling. For example, if the rate at t=8 minutes is 0.03 mm / min, and the rate at t=12 minutes is 0.04 m / min... m / min, the rate at t=10 minutes is (0.03+0.04) / 2=0.035mm / min. For pore water pressure data, if it is missing, it is completed by linearly inferring the pressure difference between the previous and next time points. For example, if the pressure at t=8 minutes is 100kPa and the pressure at t=12 minutes is 120kPa, then the pressure at t=10 minutes is (100+120) / 2=110kPa. After the data is completed, a preliminary sensor monitoring raw dataset is generated. The three types of data, namely seepage flow, displacement rate and pore water pressure, are all aligned into a unified time series.

[0025] S102: Based on the original dataset of sensor monitoring, the Kalman filter algorithm is called to set the estimation matrix and error parameters. Three types of original values ​​are input sequentially to perform filtering updates and remove noise, and obtain a time series filtered estimation set. The Kalman filter algorithm is optimized based on a multi-source collaborative calibration mechanism to perform unified noise reduction on multi-dimensional raw data of seepage, displacement and pressure. Based on the raw dataset from sensor monitoring, data sets for three data sequences—seepage flow, displacement rate, and pore water pressure—are extracted separately. Each type of sensor data is used as an independent input sequence, and the filtering calculation process is performed sequentially according to time nodes. For seepage flow data, each time point is input one by one. The difference between the current observed value and the filtered estimate from the previous moment is calculated. If the difference exceeds a set deviation threshold, error correction is performed. Assuming the seepage flow observed at t=10 minutes is 4.5 m³ / h, and the filtered estimate from the previous moment… The measured value is 4.4 m³ / h. With a set deviation threshold of ±0.2 m³ / h, the difference |4.5 - 4.4| = 0.1 m³ / h is less than the threshold and requires no correction. The observed value of 4.5 m³ / h is directly used as the filtered value for this moment. If the difference exceeds 0.2 m³ / h, the estimated value is adjusted. The displacement rate data follows the same procedure. Assuming the observed value at t = 10 minutes is 0.035 mm / min and the filtered value at the previous moment is 0.033 mm / min, the difference |0.035 - 0.033| = 0. If the flow rate is less than the threshold ±0.005 mm / min, no correction is made. For pore water pressure data, assuming an observation value of 110 kPa at t=10 minutes and a previous filtered value of 108 kPa, the difference |110-108|=2 kPa is less than the set threshold ±5 kPa, so no correction is made. If the difference exceeds the threshold, the data is updated based on a weighted average of the current observation value and the previous filtered value. The weighting coefficient is calculated using the actual data volatility; for example, the standard deviation of the seepage flow data volatility is 0.3 m. The standard deviation of displacement rate data fluctuation is 0.005 mm / min, and the standard deviation of pore water pressure data fluctuation is 6 kPa. The weights are set to be inversely proportional to the fluctuation rate. The weight of seepage flow is set to 1 / 0.3≈3.33, the weight of displacement rate is set to 1 / 0.005=200, and the weight of pore water pressure is set to 1 / 6≈0.167. After inputting the three types of data, each time node is progressively filtered and weighted in the above calculation order to generate the filtered estimate of all time series data.

[0026] S103: Based on the time series filtered estimate set, extract multiple types of data sequences, rearrange the structure according to function, unify the units, and perform normalization processing using linear interval compression to generate a standardized monitoring dataset. Based on the time series filtered estimation set, complete time series are first extracted for three types of data: seepage flow, displacement rate, and pore water pressure. The data structure is then rearranged according to the functions of seepage monitoring, displacement monitoring, and pressure monitoring. The unit of seepage flow is uniformly converted from m³ / h to L / min using the formula: Seepage flow (L / min) = Seepage flow (m³ / h) × 1000 ÷ 60, for example, 4.5 m³ / h = 75 L / min. The unit of displacement rate remains mm / min, and the unit of pore water pressure remains kPa. After the unit conversion, the maximum and minimum values ​​of each type of data are calculated. The range for seepage flow is set to [60, 90] L / min, and the range for displacement rate is set to […]. The pore water pressure range is set to [90, 130] kPa, with a flow rate of 0.02, 0.05 mm / min. For each data point, linear interval compression normalization is performed. The normalization formula is: normalization value = (current value - minimum value) / (maximum value - minimum value). Taking a seepage flow rate of 75 L / min as an example, the normalization value = (75 - 60) / (90 - 60) = 0.5. The normalization value for a displacement rate of 0.035 mm / min is approximately (0.035 - 0.02) / (0.05 - 0.02) ≈ 0.5. The normalization value for a pore water pressure of 110 kPa is approximately (110 - 90) / (130 - 90) = 0.5. The normalization values ​​for the three types of data at each time point are calculated sequentially to generate a standardized monitoring dataset.

[0027] Table 1: Example Table of Normalized Data As shown in Table 1, the units for seepage flow, displacement rate and pore water pressure have been unified and normalized, and a standardized monitoring dataset has been generated.

[0028] Specifically, the steps of S2 are as follows: S201: Call the seepage flow and displacement rate data in the standardized monitoring dataset, perform synchronous matching and spline interpolation to complete the data, identify the missing sections and set the interpolation boundary conditions, calculate the section statistical indicators, and obtain the aligned time series characteristic values; The system retrieves seepage flow and displacement rate data from a standardized monitoring dataset. Specifically, it extracts the two key parameters, seepage flow (L / min) and displacement rate (mm / d), point-by-point and performs joint comparison using the same timestamp. For data with timestamp differences, it matches the nearest neighbor time points based on the time series of seepage flow and displacement rate. For example, if a monitoring point records displacement rate on July 1st and 4th, 2021, but seepage flow only on July 2nd and 5th, then spline interpolation is used to fill in the missing displacement rate value on July 2nd on the time axis, thus achieving synchronous data matching. During interpolation, a cubic spline function is used to construct a local polynomial curve, generating an estimated value for the missing point between two known points. For instance, if the seepage flow is 35.2 L / min and 38.6 L / min on day i and i+1, and the displacement rate is 0.45 mm / d and 0.48 mm / d, then the interpolation point can be calculated from the cubic spline to obtain an estimated seepage flow of approximately 36 L / min on the intermediate day. The flow rate was 8 L / min, and the displacement rate was 0.465 mm / d. This data was marked as interpolated and distinguished from the original measurement value in subsequent statistics. For data sequences with missing measurement segments exceeding 3 days in the original data, boundary interpolation conditions were set, allowing spline estimation only within a maximum span of 2 days. The excess portion was marked as an incomplete missing item and removed from the statistical indicators. Subsequently, multiple data segments were statistically analyzed in groups of 7 days. The statistical items included the maximum value, minimum value, mean, standard deviation, and variation range. For example, the displacement rate in a certain week was 0.52 mm / d, the minimum value was 0.44 mm / d, the mean was 0.48 mm / d, the standard deviation was 0.025 mm / d, and the variation range was 0.08 mm / d. Finally, based on the aligned time series, the paired data of seepage flow and displacement rate were summarized at the end of each period as the time series characteristic value of that period. For example, the characteristic value of the first period was (36.8, 0.465), and the characteristic value of the second period was (38.2, 0.482), etc.

[0029] S202: Based on the aligned temporal feature values, extract the difference in the average value of the segments, perform range normalization, calculate the gray correlation coefficient using the correlation coefficient formula and summarize it to obtain the gray correlation degree between parameters. Based on the aligned time-series characteristic values, the seepage flow rate and displacement rate within each cycle are first matched according to the cycle number. Then, the difference between the cycle mean of seepage flow rate and the cycle mean of displacement rate is calculated. For example, in cycle 1, the average seepage flow rate is 36.8 L / min and the average displacement rate is 0.465 mm / d, while in cycle 2 it is 38.2 L / min and 0.482 mm / d. The difference is calculated as (38.2-36.8) / 36.8≈0.038. Next, the range of the cycle difference values ​​is normalized. The processing assumes the maximum value of the periodic seepage flow difference is 0.092 and the minimum value is 0.012. The normalized result for period 1 is (0.038-0.012) / (0.092-0.012)≈0.325, yielding the normalized seepage flow difference sequence. The displacement rate is processed in the same way. Then, the grey relational coefficient calculation method is used to evaluate the correlation. Specifically, for each group of periodic seepage flow and displacement rate difference sequences, the absolute difference ratio and minimum-maximum control strategy are applied to calculate the correlation coefficient γ. The formula is: ; in, Let be the absolute value of the normalized difference in the i-th group and the k-th period. , , To determine the correlation coefficient, we take 0.5. Assuming the maximum difference is 0.52 and the minimum is 0.12 within 5 periods, and the difference for the 3rd period is 0.28, then the correlation coefficient for the 3rd period is: ; The overall grey relational degree is obtained by summing the arithmetic mean of the periodic correlation coefficients within 5 periods. For example, if the correlation coefficients of the 5 periods are 0.7037, 0.7682, 0.8124, 0.7391, and 0.7816, then the grey relational degree is 0.761.

[0030] Table 2: Key Period Differences and Correlation Coefficients As shown in Table 2, the summary results of the correlation coefficients corresponding to the periods constitute a gray correlation sequence between parameters, which is used for subsequent filtering.

[0031] S203: Based on the gray correlation between parameters, compare the correlation between segments and the correlation threshold, construct a mapping table of parameter-segment correspondence, filter parameters that meet the conditions and bind them to segment locations, and generate a set of key monitoring parameters. The association threshold is a critical value used to screen the strength of the association between parameters in grey relational analysis. When the association degree exceeds the association threshold, it is judged as a significant association. Based on the modeling and analysis of measured data of three reservoirs in a certain province from 2018 to 2022, it is shown that setting the threshold range to [0.75, 0.9] can obtain the optimal classification effect. Based on the grey relational degree between parameters, the grey relational coefficient of each monitoring parameter is compared with that of the corresponding period. First, the correlation threshold range is set to [0.75, 0.9]. When the grey relational coefficient of a parameter is higher than 0.75, it is determined that it has a correlation in the corresponding segment. For example, the correlation coefficient between seepage flow and displacement rate in period 3 is 0.8124, which is within the threshold range. Therefore, the seepage flow in period 3 is determined to be a key parameter. If two consecutive periods exceed this threshold, the parameter is set as strongly correlated within the corresponding segment. If both period 3 and period 5 exceed the threshold, further records are made. A one-to-one mapping table between cycle number and parameter ID is constructed, forming a parameter segment mapping matrix. The mapping format is (cycle number, parameter ID, grey relational coefficient), such as (3, seepage flow #A1, 0.8124), (5, seepage flow #A1, 0.7816). This is used to filter the set of monitoring parameters that meet the strong correlation criteria, which will be the focus of subsequent monitoring. Finally, a set of key monitoring parameters is generated, whose structure includes three items: parameter number, corresponding segment number, and grey relational coefficient value. At the same time, a label field is formed in the system database for subsequent model calls.

[0032] The results indicate that, within the set threshold range [0.75, 0.9], parameters with a grey relational coefficient higher than 0.75 are identified as key monitoring parameters in their corresponding segments and are included in the final monitoring set.

[0033] Specifically, the steps of S3 are as follows: S301: Based on the raw pore water pressure data in the key monitoring parameter set, the dominant frequency component is extracted using discrete Fourier transform, the amplitude of the frequency interval is screened and sorted, and a dominant frequency amplitude sequence is generated. Based on the raw pore water pressure data from the key monitoring parameter set, the monitoring data first needs to be formatted to extract the pore water pressure value sequence collected at different times for the corresponding monitoring points. For example, in a hydraulic dam project, a certain monitoring point records the pore water pressure once per hour for 72 consecutive hours, and the obtained sequence is as follows: The sequence, measured in kPa, is used as a time series input. A Discrete Fourier Transform (DFT) is then performed, converting the time-domain signal into a frequency-domain signal. The sampling frequency needs to be set during this conversion process. and number of sampling points If measured in hours, , Then the frequency domain resolution is The modulus of the transformed complex number is taken to obtain the amplitude corresponding to each frequency. By filtering the frequency range, for example, the range from 0.05Hz to 0.2Hz, corresponding to frequency indices 4 to 14, the frequency components within this frequency band are arranged in descending order of amplitude by comparing the amplitudes. This sequence is denoted as the main frequency amplitude sequence, such as: kPa; Instead of using a maximum value threshold for selection, the dominant frequency is selected by sorting by amplitude and retaining the top n items. The selection of n is based on the empirical setting of 5 items according to the fluctuation intensity of the monitoring cycle. Frequency items with amplitudes lower than 20% of the overall average are removed. That is, if the average amplitude is 12 kPa, then the frequency items with amplitudes greater than 14.4 kPa are retained. Finally, the dominant frequency amplitude sequence is used as a frequency domain characteristic index to represent the main frequency band of pore water pressure change at the monitoring point.

[0034] S302: Call the main frequency amplitude sequence and displacement rate data to obtain the displacement rate sequence of the measuring point within the monitoring period, calculate the acceleration change value of the data difference between adjacent time points, compare the acceleration change time index with the amplitude timestamp, extract the matching measuring point information, and generate an aligned time feature set; To retrieve the aforementioned main frequency amplitude sequence and displacement rate data, their time axes must first be synchronized. If the pore water pressure frequency characteristic timestamp is from t1 to t72, the corresponding displacement rate sequence of the measuring point must be extracted within the same time range. For example, the displacement rate sequence recorded by measuring point A on the dam body is... If the velocity is mm / h, then the velocity values ​​at two adjacent time points are differentially processed to obtain the acceleration change sequence as follows: mm / h², where the difference between any two points is calculated as ,For example The time points after differentiation (e.g., t2~t72) are matched with the time points corresponding to the aforementioned main frequency amplitude. That is, the timestamps corresponding to significant acceleration changes (greater than 0.005 mm / h²) are extracted. If the main frequency amplitude has a high value point within that timestamp (e.g., the maximum value of the main frequency sequence in the aforementioned example is 18.3 kPa, corresponding to t50), then it is considered that there is a matching relationship at that time. The measurement point information and its corresponding time are recorded to form an aligned time feature set. The above matching operation is completed by combining multiple measurement points (e.g., measurement points A, B, and C) to finally form a matching index set of measurement point acceleration and main frequency response within a time period.

[0035] S303: Based on the data from multiple measuring points in the alignment time feature set, statistically analyze the time period differences and calculate the amplitude increase / decrease rate and acceleration fluctuation rate, perform linear standardization and combination processing, and generate the dam deformation feature vector; Based on data from multiple measurement points in the alignment time feature set, the time difference between each pair of matching times is first calculated, and the time deviation range is calculated in units of t. For example, if the matching times of measurement points A and B are t45 and t48 respectively, with a difference of 3 hours, the average of the time differences between measurement points is calculated. Secondly, obtain the amplitude increase / decrease rate and acceleration volatility rate, amplitude increase / decrease rate Can be defined as If the amplitude of the dominant frequency at a certain measuring point increases from 12.1 kPa to 18.3 kPa, then acceleration volatility Can be defined as If the acceleration sequence is mm / h², then The amplitude increase / decrease rate and acceleration volatility are linearly standardized using the normalization formula: ; Within a range of multiple measurement points, a unified transformation is performed using maximum and minimum values. If the amplitude increase / decrease rate is between 0.2 and 0.6, the standardized value is... Finally, the standardized amplitude change and acceleration fluctuation are combined as two-dimensional vectors to form the dam deformation feature vector of each measuring point. The feature vectors of the measuring points are summarized to form the overall deformation feature set of the dam.

[0036] Table 3: Example Table of Deformation Feature Calculation Table 3 shows the calculation process and results of deformation characteristics at three monitoring points. All calculations were performed using actual sampled data, with standardized intervals of [0.2, 0.6] and [0.4, 1.6]. The resulting deformation feature vectors are used in subsequent dam health monitoring and displacement prediction tasks.

[0037] Specifically, the steps of S4 are as follows: S401: Based on the dam deformation feature vector, call the main frequency amplitude sequence data, detect the main frequency amplitude change sequence over time, calculate the ratio of the main frequency amplitude difference between adjacent time nodes to the main frequency amplitude of the previous node, and generate the main frequency attenuation rate value sequence. Based on the dam deformation feature vector, the dominant frequency amplitude sequence data for each monitoring time point is first retrieved from the original monitoring data. This data can be obtained from the spectral information collected by seismic wave sensors deployed on the dam. For example, assuming four monitoring points are deployed on the dam with a sampling frequency of 100Hz, the dominant frequency amplitude sequence obtained within a certain time period is... Then, the main frequency amplitude is processed according to the time sequence, and the difference between the main frequency amplitudes at adjacent time points is calculated, that is, the operations are performed sequentially: , , Then, the ratio of the difference to the amplitude of the dominant frequency of the previous node is calculated and expressed as: , , Then, the absolute value of the ratio is used as the initial basis for calculating the main frequency attenuation rate, ultimately forming a sequence of main frequency attenuation rate values, for example, obtaining... In actual operation, a reasonable threshold should be set to determine whether a certain attenuation rate is within an abnormal range. If an empirical threshold range is used, the normal range could be defined as follows: The range of mild abnormalities is The moderately abnormal range is The severely abnormal range is In the current example, the third time point has fallen into the moderate anomaly range. During dam deformation monitoring, when the dominant frequency attenuation rate enters a slightly abnormal or worse range during a certain period, the monitoring system will automatically increase the sampling frequency for that period, changing from once every 10 minutes to once every 1 minute. If it enters the moderate or severe anomaly range, the dam microseismic monitoring module will be activated in parallel, and an inspection instruction will be sent to on-site maintenance personnel, requiring them to complete the visual inspection of the dam arch and upstream and downstream shoulders within 12 hours to prevent potential structural hazards. This process achieves a quantitative expression of the dominant frequency change trend at each node through item-by-item comparison and ratio calculation of the dominant frequency amplitude.

[0038] S402: Call the main frequency attenuation rate value sequence and the acceleration change data in the dam deformation feature vector, extract the acceleration change value at each time node, calculate the total acceleration change from the start to the current time node, and generate the cumulative acceleration change value; Call the main frequency decay rate value sequence The acceleration change data in the dam deformation feature vector is assumed to correspond to the acceleration change sequence as follows: At the same time, the absolute acceleration value at each time point is obtained. The values ​​have been normalized to the reference acceleration. The cumulative change in acceleration is calculated using the formula: ; in, Represents the time from the start to the current time node. The cumulative change in acceleration, Representing the The change in dam acceleration at each time point relative to the previous time point. Representing the Normalized dam body acceleration changes at each time point Representing the The main frequency attenuation rate value at each time point Representing the The main frequency attenuation rate value at each time point Number the current time node.

[0039] The calculation formula is: ; The specific calculations are as follows: Item 1 ( ,at this time (Default initial value is 0): ; Item 2 ( ): ; Item 3 ( ): ; Adding the three terms together, we get: ; This result indicates that the cumulative change in acceleration up to the third time point is This value can be used as input for subsequent fuzzy judgment. A cumulative change in acceleration exceeding 0.04 m / s² is considered a "high dynamic disturbance" signal. The reservoir management system will record the external environmental information (such as rainfall and seismic station data) for that period in the background and notify the dam structure analysis department to conduct dynamic response simulation. If the cumulative change is accompanied by an increase in the dominant frequency attenuation rate, dual monitoring will be adopted—increased sampling in the next monitoring cycle and additional pore water pressure sensors will be deployed in the dam toe area to track changes in internal seepage.

[0040] S403: Call the sequence of cumulative acceleration change value and main frequency attenuation rate value, take the two as input items, substitute them into the fuzzy comprehensive evaluation method, perform level assignment mapping based on risk membership function and level discrimination threshold, and obtain deformation risk level; The fuzzy comprehensive evaluation method transforms the main frequency attenuation rate and the cumulative change in acceleration into a fuzzy matrix through membership functions and weights, calculates the corresponding risk level, and realizes a multi-level fusion judgment mechanism. Call the above cumulative change in acceleration value With the sequence of main frequency attenuation rate values Select the current frequency attenuation rate value respectively and cumulative change value As input for risk identification, according to the fuzzy comprehensive evaluation method, it is input into the membership function for normalization processing, for example, ... Risk level intervals are divided according to the following membership functions: Normal: [0, 0.02], membership degree is 1; Mild anomaly: (0.02, 0.04], membership degree decreases linearly to 0; Moderate anomaly: (0.04, 0.06], membership degree increasing from 0 to 1; Severe anomaly: (0.06, ∞), membership degree is 1; for It falls within the moderately abnormal range, and the initial membership degree is set to 0.085; at the same time, it is set... The judgment interval is: Normal: [0, 0.02], Mild abnormality: (0.02, 0.04], Moderate abnormality: (0.04, 0.06], Severe abnormality: (0.06, ∞); but For the corresponding moderately abnormal interval, its membership degree is set to 0.445, and the weights are respectively... (Frequency attenuation rate) (Cumulative acceleration value), the comprehensive evaluation value is calculated using the weighted average method: ; Based on the discrimination threshold setting: :normal; Mild abnormality; Moderately abnormal; Serious abnormality; Then the current value The assessment result is "mildly abnormal", which is the deformation risk level at the current moment.

[0041] A mild anomaly level will trigger a dam area inspection order, requiring monitoring personnel to conduct foot measurements and visual inspections of the abnormal area within 6 hours, and increasing the data collection frequency of relevant monitoring points (e.g., from once every 10 minutes to once every 2 minutes) to prevent further deterioration of the deformation. A normal level will maintain the original monitoring frequency without additional action; a moderate anomaly level will activate on-site monitoring; and a severe anomaly level will trigger an emergency plan, including flood discharge or temporary reinforcement measures.

[0042] Table 4: Time Series Calculation Parameters for Monitoring Points Table 4 lists the parameters of the main frequency and acceleration changes at different time points and their corresponding risk level membership. The t3 node was ultimately determined to be at the mild anomaly level.

[0043] Specifically, the steps of S5 are as follows: S501: Call the deformation risk level and determine whether its corresponding value reaches the risk threshold. If the condition is met, obtain the main frequency amplitude sequence in the current monitoring period, calculate the slope fluctuation trend based on its rising and falling inflection point position, extract the number of abrupt change points, monitor the acceleration change and calculate the fluctuation amplitude with the offset of adjacent time periods, and generate frequency amplitude change feature quantity. The risk threshold is a critical standard for judging whether the monitoring data has reached the level that requires early warning. If it is exceeded, the subsequent analysis and early warning mechanism will be triggered. It is determined by comparing the frequency distribution of dam displacement and pore pressure response under different monitoring conditions. During the process of calling the deformation risk level, the risk level code of the current monitoring area must first be read by the front-end sensor. The system automatically converts the level into a risk-corresponding numerical threshold range. If the value is higher than the set risk trigger threshold, for example, the preset risk threshold is set to 0.65. When the real-time risk level mapping value obtained by the sensor reaches 0.72, the system determines that the alarm trigger condition has been met. At this time, the frequency amplitude data analysis stage begins. The system retrieves the dominant frequency amplitude time series data of the current monitoring period t from the monitoring database. This data is generated from the surface vibration signal after FFT transformation. Assuming that the sampling frequency of this period is 100Hz, the dominant frequency amplitude sequence is as follows: ; The sequence is scanned using a sliding window method. Inflection points are determined based on the switching points of the first derivative's direction of change. Points 3, 5, and 7 in the sequence are identified as inflection points, and their indices are marked. Furthermore, the slope is calculated between each pair of adjacent inflection points. For example, the slope between points 3 and 5 is... The system traverses the interval to calculate its slope value set. Then, the average of the above slope values ​​is calculated to obtain... The original slope standard deviation of the sample dataset was set to [value]. variance index Calculated from the squared deviation of the amplitude energy sequence, if the amplitude energy sequence is: ; Its mean is 0.0608, thus yielding The number of mutation points is calculated using the following formula: ; in, Represents the current monitoring period The number of mutation points parameter, Represents the current monitoring period The Middle The relative change in the amplitude of the dominant frequency at each inflection point The reference value representing the amplitude of the main frequency. Represents the current monitoring period The Middle The slope value of the main frequency amplitude corresponding to each inflection point Represents the current monitoring period The average slope of the dominant frequency amplitude at the inner inflection point The benchmark value representing the change in the dominant frequency slope is taken as the standard deviation of the original sample data. Representing the The confidence weight coefficients for each inflection point are calculated based on the Bayesian posterior scoring model for mutation points. It is a small positive constant used to avoid the denominator being zero. Indicates the current monitoring period The total number of inflection points detected within the system. The variance index representing the distribution of the dominant frequency energy during the monitoring period t is used as an enhancement factor.

[0044] Perform calculations, where Divide the amplitude difference between each inflection point and the previous inflection point by the reference amplitude. Therefore, assuming the reference amplitude , Let the amplitude at the third inflection point be 0.25 and the amplitude at the second inflection point be 0.21, then we have Confidence weight Based on the posterior risk rate of this type of mutation point in historical monitoring, if the system gives a confidence level of 0.83 for this point through Bayesian posterior analysis, then the contribution of the third inflection point is: ; Similarly, the contribution values ​​of other inflection points are calculated to finally obtain the parameter for the number of abrupt change points: ; This indicates that the total contribution value of mutation points during the monitoring period is 0.479. If the preset early warning threshold for the number of mutation points is 0.35, this value exceeds the threshold and can be judged as a risk state.

[0045] When the number of mutation points exceeds a threshold, the system immediately determines whether the mutation originates from structural anomalies or external disturbances (such as local earthquakes or flood impacts). Dam personnel will conduct inspections of key areas (such as the arch crown section, expansion joints, and spillway) every two hours and utilize infrared thermal imagers to detect areas of structural stress concentration in order to quickly identify potential hazards.

[0046] Table 5: Examples of Mutation Point Feature Extraction Table 5 lists the magnitude change, slope value, weighting coefficient, and contribution to the total abrupt change value for each inflection point.

[0047] S502: Based on the number of abrupt change points and acceleration fluctuation amplitude in the frequency amplitude change feature, construct a two-dimensional feature vector and input it into the logistic regression classifier. Perform linear combination and normalization processing using the regression coefficient matrix and bias term to generate a coupled early warning probability value. Based on the number of abrupt changes in the frequency amplitude variation characteristic and the acceleration fluctuation amplitude, a two-dimensional feature vector is constructed. The system needs to extract the parameter of the number of abrupt changes from the previous step. Combined with acceleration fluctuation amplitude parameters to form an eigenvector Among them, the acceleration fluctuation amplitude It is calculated from the difference between the maximum and minimum acceleration during the current monitoring period, for example, the main frequency amplitude sequence is... The maximum value is 0.31, the minimum value is 0.18, and the difference is... Therefore, the feature vector is The vector is then input into the logistic regression classifier model, and the dot product of the feature terms and the trained regression coefficient vector is calculated. Assume the regression coefficient matrix is... If the bias term is -1.08, then the linear combination is: ; This value is then normalized using the sigmoid function to obtain the coupling warning probability value: ; The coupling warning probability value corresponding to the current feature combination is 0.450.

[0048] At the dam safety monitoring and command center, when the coupled early warning probability value is between 0.4 and 0.6, it falls under the "median risk" state. On-site operators will increase the synchronous reading of acceleration and crack sensors and upload the data to the dispatch center in real time. Simultaneously, additional nighttime drone patrols will be arranged for this dam section to ensure the detection of cracks or seepage under invisible light conditions, and the image data will be archived as a basis for subsequent risk trend analysis.

[0049] S503: Based on the coupled warning probability value and the established coupled warning level boundary standard, the interval is divided, and the preset level label is selected according to the corresponding interval. The level field and monitoring terminal structure are established and the feedback instruction is sent to obtain the coupled warning level. Receive and process the obtained coupling early warning probability value Next, the system first needs to divide the value into intervals. The associated level division criteria have been set through previous data training and expert evaluation. The specific boundary criteria are as follows: The interval is defined as a Level 1 warning, indicating extremely low risk and no additional measures are required; The interval is defined as a Level II warning, indicating a low risk, but requiring enhanced monitoring; then... The interval is defined as a Level 3 warning, indicating a high risk and the need to prepare countermeasures; finally, The interval is defined as a Level 4 warning, indicating an extremely high risk that requires immediate action.

[0050] By using boundary markers, the current risk situation can be quickly assessed, allowing for the development of corresponding monitoring and emergency response strategies. For example, if the system issues a Level 1 warning, monitoring personnel can choose to continue daily monitoring and data recording; for a Level 2 warning, monitoring personnel need to frequently check the on-site status and report the situation regularly; while Level 3 or 4 alarms require the activation of emergency plans, notification of relevant departments, and on-site assessment.

[0051] In actual operation, when the system determines a Level II warning, the dam monitoring center immediately increases the sampling rate of the accelerometer in that area (e.g., from once per minute to once every 20 seconds) and arranges drones to conduct image surveys of the dam surface and both sides of the dam. If the warning is determined to be Level III, the water conservancy and flood control department is notified to deploy response plans in advance, and temporary stress gauges are added to the dam to track deformation trends in real time. If the warning is Level IV, the emergency flood discharge plan is directly triggered, and an evacuation warning is issued to downstream residential areas. This ensures that the judgment results are highly integrated with engineering actions, forming a closed loop of "calculation results - risk level - emergency measures".

[0052] In this embodiment of the invention, multi-source heterogeneous monitoring data (including dam seepage flow, displacement rate, and pore water pressure) are denoised using Kalman filtering to form a standardized dataset suitable for dam structural health assessment; grey relational analysis dynamically filters key parameters highly correlated with dam seepage anomalies and structural displacement changes, constructing a feature vector reflecting the stress-seepage coupling state of the dam; discrete Fourier transform extracts the dominant frequency amplitude sequence reflecting the periodic changes in pore water pressure, and timestamps it in conjunction with the acceleration change of displacement rate to accurately capture the deformation response of local dam sections under high water level or rapid adjustment conditions; fuzzy comprehensive evaluation integrates the dominant frequency attenuation rate and the cumulative displacement acceleration change, quantifying... The system optimizes the overall deformation index of the dam body to achieve a scientific assessment of the damage trend of the dam structure; a logistic regression classifier generates a dynamic early warning level reflecting the safety status of the dam structure based on the abrupt change point of the dominant frequency amplitude and the acceleration fluctuation amplitude; coupled analysis of time series data and spatial features ensures early identification of hidden dangers such as concentrated seepage and local structural instability; a nonlinear risk mapping model overcomes the limitations of traditional experience-based judgment in complex environmental conditions during reservoir operation; synchronous processing of multi-dimensional hydraulic monitoring data significantly improves the assessment accuracy under extreme floods, rapid regulation, and long-term fatigue; and a graded early warning mechanism effectively reduces the probability of false alarms and missed alarms in dam safety management, ensuring operational safety during the flood season and under sudden conditions.

[0053] Furthermore, such as Figure 2 As shown in the figure, this embodiment of the invention also provides a reservoir dam monitoring and early warning system based on multi-source information fusion, the system comprising: The data fusion processing module is used to acquire time-series data of seepage flow, displacement rate and pore water pressure through a high-precision piezoelectric seepage sensor array, laser displacement meter and vibrating wire pore water pressure meter. It is connected to the data acquisition unit through RS485 bus and uses Kalman filtering algorithm to reduce noise in the raw data to generate a standardized monitoring dataset, which is then transmitted to the parameter association and filtering module. The parameter association and filtering module is used to call the standardized monitoring dataset, input the seepage flow time series data and displacement rate data into the grey relational analysis model to calculate the correlation degree, and interact with the data through the Ethernet TCP protocol. It filters parameters whose correlation degree exceeds the correlation threshold, forms a set of key monitoring parameters, and passes them to the deformation feature construction module. The deformation feature construction module is used to call the key monitoring parameter set, perform discrete Fourier transform on the pore water pressure data to extract the main frequency amplitude sequence, combine the acceleration change of the displacement rate data for timestamp alignment, transmit data with the risk level assessment module through the fiber optic communication interface, generate the dam body deformation feature vector, and transmit it to the risk level assessment module. The risk level assessment module is used to input the dam deformation feature vector into the fuzzy comprehensive evaluation method, calculate the cumulative value of the attenuation rate of the main frequency amplitude sequence and the change in acceleration, set the evaluation index weights according to monitoring needs, output the deformation risk level, and transmit it to the early warning decision generation module. The early warning decision generation module is used to input the number of mutation points in the main frequency amplitude sequence and the fluctuation amplitude of the acceleration change into the logistic regression classifier when the deformation risk level reaches the risk threshold, generate a coupled early warning level, and feed the coupled early warning level back to the Web visualization platform of the monitoring terminal through the wireless communication module.

[0054] For ease of explanation, Figure 2 Only the main components of the system are shown. The system of this embodiment can be used to perform... Figure 1 The technical solutions of the method embodiments shown are similar in principle and in effect, and will not be described again here.

[0055] In an exemplary embodiment, the present invention also provides an electronic device, the electronic device comprising: processor; The memory stores computer-readable instructions, which, when loaded and executed by the processor, implement the steps of the reservoir dam monitoring and early warning method based on multi-source information fusion as described above.

[0056] In an exemplary embodiment, the present invention also provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the steps of the reservoir dam monitoring and early warning method based on multi-source information fusion described above. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc.

[0057] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0058] The use of terms such as "an embodiment," "an embodiment," "an exemplary embodiment," and "some embodiments" in the specification indicates that the described embodiment may include a specific feature, structure, or characteristic, but not every embodiment necessarily includes that specific feature, structure, or characteristic. Furthermore, when a specific feature, structure, or characteristic is described in connection with an embodiment, implementing such a feature, structure, or characteristic in conjunction with other embodiments (whether explicitly described or not) should be within the knowledge of those skilled in the art.

[0059] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0060] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0061] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers 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 the present invention.

[0062] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus 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 device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0063] 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.

[0064] In addition, the functional units in the various embodiments of the present invention 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.

[0065] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0067] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for monitoring and early warning of reservoir dams based on multi-source information fusion, characterized in that, Includes the following steps: S1: Acquire time-series data of seepage flow, displacement rate data and pore water pressure data by sensor arrays deployed on the dam body, and use Kalman filtering algorithm to denoise the raw data to generate a standardized monitoring dataset; S2: Call the standardized monitoring dataset, input the seepage flow time series data and displacement rate data into the grey relational analysis model to calculate the correlation degree, filter parameters with a correlation degree exceeding the correlation threshold, and form a set of key monitoring parameters; S3: Based on the set of key monitoring parameters, perform discrete Fourier transform on the pore water pressure data to extract the main frequency amplitude sequence, and combine it with the acceleration change of the displacement rate data to align the timestamps and generate the dam deformation feature vector. S4: Input the dam deformation feature vector into the fuzzy comprehensive evaluation method, calculate the cumulative value of the attenuation rate of the main frequency amplitude sequence and the acceleration change, and output the deformation risk level; S5: Call the deformation risk level. When the risk level reaches the risk threshold, input the number of mutation points in the main frequency amplitude sequence and the fluctuation amplitude of the acceleration change into the logistic regression classifier to generate a coupled early warning level and feed it back to the monitoring terminal.

2. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 1, characterized in that, The standardized monitoring dataset is obtained by denoising the seepage flow time series data, displacement rate data, and pore water pressure data using the Kalman filter algorithm. The set of key monitoring parameters is composed of seepage flow correlation parameters and displacement rate correlation parameters whose correlation exceeds a preset correlation threshold, selected based on a grey relational analysis model. The dam deformation feature vector is formed by aligning the main frequency amplitude spectrum sequence extracted from the pore water pressure data using discrete Fourier transform with the acceleration change of the displacement rate data using timestamps. The deformation risk level is generated based on the main frequency energy attenuation rate and the cumulative displacement acceleration change using a fuzzy comprehensive evaluation method. The coupling warning level is calculated by using a logistic regression classifier based on the number of main spectrum mutation points and the acceleration standard deviation parameter.

3. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 1, characterized in that, The specific steps of S1 include: S101: Collect time-series data of seepage flow, displacement rate and pore water pressure by sensor arrays deployed on the dam body, establish a data index structure, adjust the consistency of time arrangement, unify the sampling period and identify missing segments, and generate the original dataset of sensor monitoring. S102: Based on the original dataset monitored by the sensor, the Kalman filter algorithm is called to set the estimation matrix and error parameters. Three types of original values ​​are input sequentially to perform filtering updates and remove noise, and a time series filtered estimation value set is obtained. S103: Based on the time series filtered estimate set, extract multiple types of data sequences, rearrange the structure according to function, unify the units, and perform normalization processing using linear interval compression to generate a standardized monitoring dataset.

4. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 3, characterized in that, The specific steps of S2 include: S201: Call the seepage flow and displacement rate data in the standardized monitoring dataset, perform synchronous matching and spline interpolation to complete the data, identify the missing sections and set the interpolation boundary conditions, calculate the section statistical indicators, and obtain the aligned time series characteristic values; S202: Based on the aligned temporal feature values, extract the difference in average values ​​of segments, perform range normalization, calculate the gray correlation coefficient using the correlation coefficient formula and summarize it to obtain the gray correlation degree between parameters. S203: Based on the gray correlation degree between the parameters, compare the correlation degree of the segments with the correlation threshold, construct a mapping table of the correspondence between parameters and segments, filter out parameters that meet the conditions and bind them to the segment positions, and generate a set of key monitoring parameters.

5. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 4, characterized in that, The specific steps of S3 include: S301: Based on the raw pore water pressure data in the set of key monitoring parameters, the main frequency component is extracted using discrete Fourier transform, the amplitude of the frequency interval is screened and sorted, and a main frequency amplitude sequence is generated. S302: Call the main frequency amplitude sequence and displacement rate data to obtain the displacement rate sequence of the measuring point within the monitoring time period, calculate the acceleration change value by difference between adjacent time points, compare the acceleration change time index and amplitude timestamp, extract matching measuring point information, and generate an aligned time feature set; S303: Based on the data from multiple measurement points in the alignment time feature set, statistically analyze the time period differences and calculate the amplitude increase / decrease rate and acceleration fluctuation rate, perform linear standardization and combination processing, and generate the dam deformation feature vector.

6. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 5, characterized in that, The specific steps of S4 include: S401: Based on the dam deformation feature vector, call the main frequency amplitude sequence data, detect the main frequency amplitude change sequence over time, calculate the ratio of the main frequency amplitude difference between adjacent time nodes to the main frequency amplitude of the previous node, and generate the main frequency attenuation rate value sequence. S402: Call the main frequency attenuation rate value sequence and the acceleration change data in the dam deformation feature vector, extract the acceleration change value at each time node, calculate the total acceleration change from the start to the current time node, and generate the cumulative acceleration change value; S403: Call the cumulative acceleration change value and the main frequency attenuation rate value sequence, take them as input items, substitute them into the fuzzy comprehensive evaluation method, and perform level assignment mapping based on the risk membership function and level discrimination threshold to obtain the deformation risk level.

7. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 6, characterized in that, The cumulative change in acceleration is calculated using the following formula: ； in, Represents the time from the start to the current time node. The cumulative change in acceleration, Representing the The change in dam acceleration at each time point compared to the previous time point. Representing the Normalized dam body acceleration changes at each time point Representing the The main frequency attenuation rate value at each time point Representing the The main frequency attenuation rate value at each time point Number the current time node.

8. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 6, characterized in that, The specific steps of S5 include: S501: Call the deformation risk level and determine whether its corresponding value reaches the risk threshold. If the condition is met, obtain the main frequency amplitude sequence in the current monitoring period, calculate the slope fluctuation trend based on its rising and falling inflection point position, extract the number of abrupt change points, monitor the acceleration change and calculate the fluctuation amplitude with the offset of adjacent time periods, and generate frequency amplitude change feature quantity. S502: Based on the number of abrupt change points and the acceleration fluctuation amplitude in the frequency amplitude change feature quantity, construct a two-dimensional feature vector and input it into the logistic regression classifier. Perform linear combination and normalization processing using the regression coefficient matrix and bias term to generate a coupled early warning probability value. S503: Based on the coupled warning probability value and the predetermined coupled warning level boundary standard, divide the interval, select the preset level label according to the corresponding interval, establish the level field and monitoring terminal structure, and send the feedback instruction to obtain the coupled warning level.

9. The reservoir dam monitoring and early warning method based on multi-source information fusion according to claim 8, characterized in that, The number of mutation points is calculated using the following formula: ； in, Represents the current monitoring period The number of mutation points parameter, Represents the current monitoring period The Middle The corresponding change in the amplitude of the dominant frequency at each inflection point The reference value representing the amplitude of the main frequency. Represents the current monitoring period The Middle The slope value of the main frequency amplitude corresponding to each inflection point Represents the current monitoring period The average slope of the dominant frequency amplitude at the inner inflection point The benchmark value representing the change in the dominant frequency slope is taken as the standard deviation of the original sample data. Representing the The confidence weight coefficients for each inflection point are calculated based on the Bayesian posterior scoring model for mutation points. It is a positive constant, used to avoid the denominator being zero. Indicates the current monitoring period The total number of inflection points detected internally. The variance index representing the distribution of the dominant frequency energy during the monitoring period t is used as an enhancement factor.

10. A reservoir dam monitoring and early warning system based on multi-source information fusion, characterized in that, The system is used to implement the reservoir dam monitoring and early warning method based on multi-source information fusion as described in any one of claims 1-9, and the system includes: The data fusion processing module is used to acquire time-series data of seepage flow, displacement rate data and pore water pressure data through the deployed sensor array, and to perform noise reduction processing on the raw data using the Kalman filter algorithm to generate a standardized monitoring dataset, which is then transmitted to the parameter association and filtering module. The parameter association and filtering module is used to call the standardized monitoring dataset, input the seepage flow time series data and displacement rate data into the grey relational analysis model to calculate the correlation degree, filter parameters whose correlation degree exceeds the correlation threshold, form a set of key monitoring parameters, and pass them to the deformation feature construction module. The deformation feature construction module is used to call the set of key monitoring parameters, perform discrete Fourier transform on the pore water pressure data to extract the main frequency amplitude sequence, combine the acceleration change of the displacement rate data with the timestamp alignment, generate the dam body deformation feature vector, and pass it to the risk level assessment module. The risk level assessment module is used to input the dam deformation feature vector into the fuzzy comprehensive evaluation method, calculate the cumulative value of the attenuation rate of the main frequency amplitude sequence and the acceleration change, output the deformation risk level, and transmit it to the early warning decision generation module. The early warning decision generation module is used to input the number of mutation points in the main frequency amplitude sequence and the fluctuation amplitude of the acceleration change into the logistic regression classifier when the deformation risk level reaches the risk threshold, generate a coupled early warning level, and feed the coupled early warning level back to the monitoring terminal.

Images