A Multi-Method Collaborative Method for Dam Monitoring Data Prediction and Early Warning

Abstract

This invention relates to a multi-method collaborative method for predicting and issuing early warnings of dam monitoring data, belonging to the field of dam safety monitoring technology. First, this invention employs a combination of comparative, graphical, and threshold methods to eliminate outliers. For missing data of varying lengths, it uses moving average interpolation and an ARIMA model to hierarchically fill in the gaps. Then, it constructs an independent prediction architecture using three models: linear regression, least squares, and support vector machine. The prediction results are obtained by dynamically weighting and fusing the inverse RMSE of each model's test set. Finally, a 95.4% confidence interval is constructed based on the average residual standard deviation of the three models. An early warning is triggered when the predicted value exceeds this interval for two or more consecutive time steps. This invention effectively solves the problems of existing technologies where single-model accuracy and robustness are difficult to balance, and early warning reliability is insufficient. It is applicable to the prediction and early warning of various types of dam monitoring data, including seepage, deformation, and stress-strain, providing scientific support for the safe operation of dams.

Description

Technical Field

[0001] This invention relates to the field of dam safety monitoring technology, specifically to a multi-method collaborative method for predicting and warning dam monitoring data. Background Technology

[0002] Dams are core water conservancy infrastructures for flood control, water supply, and power generation. Their safe operation is directly related to the safety of life and property of people downstream and the stability of the national economy. Monitoring data on dam seepage, deformation, stress, and strain exhibit strong time-series and non-linear characteristics. Abnormal changes in these data are early signs of potential safety hazards such as structural damage and seepage failure. By accurately predicting monitoring data, abnormal signs can be identified in advance, allowing time for hazard investigation and emergency response.

[0003] Current dam monitoring data prediction technologies are mainly divided into three categories: traditional statistical models have weak fitting ability for nonlinear relationships and are easily affected by outliers; machine learning models are prone to overfitting under high-dimensional features and are sensitive to data distribution; deep learning models have high training costs, poor interpretability, and insufficient generalization ability in small sample scenarios. Existing related patent technologies also have obvious defects: comparison document CN121393086A only targets the single parameter of reservoir siltation, uses a single prediction model, has poor robustness, and does not combine prediction confidence intervals for early warning; comparison document CN119578908B uses the PINN model, which has a complex structure and high computational cost, is only suitable for deformation monitoring, has a fixed early warning threshold, and lacks reliability. Summary of the Invention

[0004] In summary, existing technologies generally suffer from problems such as the inability of a single model to balance accuracy and robustness, the limited data preprocessing methods, and the lack of quantification of prediction uncertainties in early warning. There is a lack of a complete integrated prediction and early warning technology solution applicable to multiple types of monitoring data.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A multi-method collaborative method for predicting dam monitoring data includes the following steps: S1: Obtain historical monitoring data of the dam through dam safety monitoring instruments, and determine key influencing factors such as upstream and downstream water levels, rainfall, and temperature based on the type of dam monitoring points; S2: The comparison method, plotting method and threshold method are used to remove gross errors in the original monitoring data. Missing data are filled in a hierarchical manner: when there are ≤3 consecutive missing data, the moving average interpolation method is used to fill the gaps; when there are >3 consecutive missing data, the autoregressive integral moving average model is used to fill the gaps. The timestamp format of all data is unified. S3: Construct a multi-method collaborative dam monitoring data prediction model, which includes an input layer, a three-model independent prediction layer, and a fusion decision layer. The processed multi-source monitoring data is input into the input layer, and then goes through data preprocessing and feature engineering, three-model independent prediction, and dynamic weighted average fusion decision based on root mean square error. The fusion fitting result, future prediction value, and fitting effect evaluation index are output through the output layer. The three independent prediction layers are established by linear regression model, least squares model and support vector machine model respectively. The three models are trained and predicted independently using the same training set and test set, and there is no parameter sharing or data interaction between the models. The fusion decision layer calculates the root mean square error (RMSE) of each model on the test set, uses the inverse proportion of each model's RMSE as the weight of the corresponding model, and performs a weighted average of the independent predictions of the three models to obtain the fusion prediction value.

[0007] A multi-method collaborative early warning method for dam monitoring data includes the following steps: T1: Obtain the fusion fitting results and future prediction values ​​output by the multi-method collaborative dam monitoring data prediction method; T2: Calculate the residual standard deviations between the predicted and measured values ​​for the linear regression model, least squares model, and support vector machine model, respectively, and take the arithmetic mean of the three residual standard deviations as the mean residual standard deviation. A confidence interval for the predicted values ​​was constructed using a 95.4% confidence level. T3: Establish an early warning linkage mechanism to compare the future predicted value of each time step with the upper and lower boundaries of the corresponding confidence interval in real time. When the triggering conditions are met, an alarm is automatically triggered and a structured alarm message is generated. At the same time, the predicted value and confidence interval for the next n days are visualized and a time series diagram containing the prediction curve and the confidence interval boundary is generated.

[0008] Furthermore, the dam safety monitoring instruments in step S1 include seepage, deformation, stress-strain, environmental, and auxiliary instruments; seepage instruments include piezometers, manometers, and flow meters; deformation instruments include inclinometers, convergence meters, total stations, and GPS; stress-strain instruments include strain gauges, rebar gauges, and pressure cells; environmental instruments include rain gauges, water level gauges, and thermo-hygrometers; and auxiliary instruments include crack gauges and seismographs.

[0009] Furthermore, the criteria for removing gross errors in step S2 are as follows: the data is considered gross and removed when all three of the following conditions are met: The monitored value at a certain moment exceeds the historical reasonable threshold range for that monitoring point; The deviation between the monitored value and the monitored values ​​at adjacent time points exceeds a preset threshold; There were no abnormal mutations in the key influencing factors at the corresponding time point.

[0010] Furthermore, in step S2, the sliding window radius of the moving average interpolation method is 3 to 7 days, and the total window size is 2k+1 data points, where k is the sliding window radius, calculated using the following formula: ; In the formula, For the missing values ​​at time point t, interpolation estimates The known observations at time point t+i; The autoregressive integral moving average model is ARIMA(p,d,q), where p is the order of the autoregressive term, d is the difference order, and q is the order of the moving average term.

[0011] Furthermore, in step S3, the input layer receives multi-source monitoring data aligned to timestamps, including deformation values, stress values, seepage values, water levels, rainfall, and air temperature. The data preprocessing and feature engineering layers perform secondary verification of missing values, feature standardization or normalization, and construct a sliding time window dataset for multi-step prediction modeling.

[0012] Furthermore, the formula for calculating the weights of the fusion decision layer in step S3 is as follows: ; In the formula, Let be the weight of the i-th model, and the sum of the weights of all models is 1.000; The formula for calculating the fused output prediction value is: ; In the formula, To fuse the predicted output values, , , These are the weights for the least squares model, the support vector machine model, and the linear regression model, respectively. , , These are the independent predicted values ​​for the corresponding models.

[0013] Furthermore, the formula for calculating the boundary of the confidence interval in step T2 is: ; In the formula, 1 represents the fused future prediction value output by the multi-method collaborative dam monitoring data prediction method, and 2 represents the Z-value corresponding to the 95.4% confidence level under the standard normal distribution.

[0014] Furthermore, in step T3, the warning trigger condition is that the predicted value exceeds the confidence interval for two or more consecutive time steps; the structured alarm information includes the measurement point number, the predicted value exceeding the confidence interval, the corresponding confidence interval range, the trigger time, and the risk level; The risk levels are divided into yellow (mildly abnormal), orange (moderately abnormal), and red (severely abnormal).

[0015] Furthermore, when applied to the monitoring of total seepage downstream of a rockfill dam with a concrete panel, total seepage monitoring points are set up downstream of the dam, and seepage flow data are collected using a measuring weir, with a monitoring frequency of once per day; daily cumulative rainfall, daily average temperature, and upstream and downstream water level data are collected simultaneously, and all monitoring data are uniformly stored in CSV format, including timestamp, monitoring point number, flow value, rainfall, temperature, upstream water level, and downstream water level fields; During data preprocessing, the Z-score standardization method was used to unify the dimensions of all features, and a sliding time window dataset with an input window of 7 days and an output window of 7 days was constructed for seepage prediction in the next 7 days.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. A triple gross error removal standard combining comparison, plotting, and thresholding methods, along with simultaneous verification of key influencing factors, can accurately identify abnormal data caused by instrument malfunctions and environmental interference, significantly improving the removal accuracy. A tiered missing data imputation strategy is adopted for missing data of different lengths. When there are ≤3 consecutive missing data, moving average interpolation is used to ensure processing efficiency, and when there are >3 consecutive missing data, the ARIMA model is used to ensure interpolation accuracy, providing high-quality standardized basic data for subsequent predictions. 2. Construct an independent prediction architecture with three models: linear regression, least squares, and support vector machine. The three models are trained independently on the same dataset without parameter sharing, giving full play to the complementary advantages of different models: linear regression and least squares capture the linear trend of the data, while support vector machine fits nonlinear relationships; the fusion decision layer dynamically assigns weights based on the inverse RMSE of each model's test set, so that the model with the smaller prediction error receives higher weights. Compared with a single model, the prediction accuracy is improved by more than 30%, and it can still maintain stable performance when there is noise in the data or sudden changes in working conditions, making it suitable for small sample scenarios commonly seen in engineering sites. 3. A 95.4% confidence interval is constructed based on the average residual standard deviation of three independent models, which quantifies the uncertainty in the prediction process. Compared with the fixed threshold warning of existing technologies, it can better reflect the normal fluctuation range of monitoring data. The trigger condition of the predicted value exceeding the confidence interval for two or more consecutive time steps is adopted, which effectively avoids false alarms caused by instantaneous data fluctuations, while ensuring the timely detection of real anomalies, and significantly improving the reliability of the warning. 4. Through a preprocessing step that combines chart analysis and statistical thresholds, outlier data is removed and valid measurement points are selected to ensure the reliability of input data; the model adopts a reasonable activation function and parameter initialization strategy, which has strong noise resistance and can adapt to actual engineering scenarios with fluctuations in monitoring data. 5. The technical solution is applicable to various types of dams, such as concrete dams, rockfill dams with concrete panels, and earth-rock dams. It supports the prediction and early warning of multiple types of time-series monitoring data, including seepage, deformation, stress, and strain. The overall process is highly automated and can be directly integrated into existing intelligent dam operation and maintenance systems without the need for large-scale modifications to existing monitoring equipment. The standardized evaluation indicators and visualized prediction curves output by the solution facilitate managers to quickly grasp the dam's operating status and provide a scientific basis for safety decisions. Attached Figure Description

[0017] Figure 1 Figure 1 shows a comparison of flow rates before and after gross error removal, as presented in this invention. in Figure 1 (a) is the flow chart before gross error removal. Figure 1 (b) is the flow diagram after gross error removal; Figure 2 This is a flowchart of the ARIMA model gap-filling process of the present invention; Figure 3 This is a flow chart after data processing for WE_DB1 in this invention; Figure 4 A schematic diagram illustrating the setting of confidence intervals for the fusion layer of this invention; Figure 5. Prediction curve of the predicted value for the next n days according to the present invention. Detailed Implementation

[0018] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment takes seepage monitoring of a panel rockfill dam as an example to illustrate the technical solution of the present invention in detail, but the scope of protection of the present invention is not limited thereto.

[0019] Example 1: A multi-method collaborative method for predicting dam monitoring data: This embodiment focuses on the total seepage monitoring point WE_DB1 downstream of a rockfill dam with concrete face. The prediction method of this invention is used to process and predict the seepage flow data. The specific steps are as follows: S1: Data Collection and Determination of Key Influencing Factors A total seepage monitoring point WE_DB1 was set up behind the rockfill dam. Seepage flow data was collected using a measuring weir. The monitoring frequency was once a day, and the data time span was from January 1, 2020 to November 23, 2025. Based on engineering experience and data correlation analysis, the key influencing factors for seepage downstream of the dam were determined to be upstream and downstream water levels, daily cumulative rainfall, and daily average temperature. Rain gauges, temperature sensors, and water level gauges were simultaneously deployed around the dam to collect corresponding environmental data, with monitoring frequency consistent with the seepage data. All monitoring data is stored in CSV format, containing seven fields: timestamp, monitoring point number, flow rate, rainfall, temperature, upstream water level, and downstream water level. The flow rate is measured in liters per second, rainfall in millimeters, temperature in degrees Celsius, and water level in meters, ensuring data traceability. S2: Data Preprocessing 2.1 Gross Error Removal: Gross errors are removed by combining the comparison method, the plotting method, and the threshold method. The judgment criteria are that three conditions must be met simultaneously: First, the monitored value exceeds the reasonable threshold range of the historical data of the measuring point; Second, the deviation from the monitored value at the previous and next adjacent time points exceeds the preset threshold; Third, there are no abnormal changes in the key influencing factors at the corresponding time point. Taking March 26, 2020 as an example, the flow rate measured at the WE_DB1 measuring point was 114 liters per second, while the measurements on the adjacent days of March 25 and March 27, 2020, were 38.9 liters per second and 37.4 liters per second, respectively. The historical maximum flow rate at this measuring point was 89.3 liters per second, and the rainfall on that day was only 0.2 mm, with the upstream water level showing a downward trend. Therefore, this data was determined to be a gross error caused by instrument malfunction and was removed. A comparison of the flow rate process lines before and after gross error removal is provided below. Figure 1 As shown; 2.2 Missing data imputation: A hierarchical imputation strategy is used to handle missing data. For a single consecutive missing data point after removing outliers on March 26, 2020, imputation was performed using the moving average interpolation method. The moving window radius was set to 1, and the total window size was 3 days. The calculated imputation value was 38.15 liters per second. For four consecutive missing data points from July 23 to July 26, 2020, an autoregressive integral moving average (ARIMA) model was used for imputation. The model parameters, determined by the AIC criterion, consisted of a second-order autoregressive term, a first-order differencing term, and a first-order moving average term, i.e., ARIMA(2,1,1). The specific imputation process is as follows... Figure 2 As shown, the resulting fill values ​​were 77.68 liters per second, 75.73 liters per second, 73.84 liters per second, and 72.00 liters per second, respectively. The timestamp format for all data is unified to year-month-day, and the processed continuous traffic data is as follows: Figure 3 As shown; S3: Construction and Operation of Multi-Method Collaborative Prediction Models 3.1 Model Structure Construction: Construct a multi-method collaborative prediction model that includes an input layer, a data preprocessing and feature engineering layer, a three-model independent prediction layer, a fusion decision layer, and an output layer; 3.2 Input Layer and Feature Engineering: The input layer receives multi-source data aligned to timestamps, including flow rate, upstream water level, downstream water level, rainfall, and temperature; The data preprocessing and feature engineering layer completes three tasks in sequence: First, it performs secondary verification of missing values ​​in the data; second, it uses the Z-score standardization method to unify the dimensions of all features; third, it constructs a sliding time window dataset, with an input window length of 7 days and an output window length of 7 days, for predicting seepage flow rate in the next 7 days. 3.3 Independent Training of Three Models: The dataset was divided into training and test sets according to time. The training set covered January 1, 2020 to December 31, 2024, and the test set covered January 1, 2025 to November 23, 2025. A linear regression model, a least squares model, and a support vector machine model were established separately. All three models were trained independently using the same training and test sets, with no parameter sharing or data interaction between them. 3.4 Dynamic weighted fusion decision: The root mean square error of the three models was calculated on the test set. The root mean square error of the least squares model was 6.039572543, the root mean square error of the support vector machine model was 0.047141343, and the root mean square error of the linear regression model was 10.12204534. The weights are calculated based on the reciprocal of the root mean square error of each model, using the following formula: ; In the formula, Let be the weights of the i-th model. Let be the root mean square error of the i-th model, and the sum of the weights of all models is 1.000; Substituting the values, we get the following weights: Least Squares model weights are 0.007709331, Support Vector Machine model weights are 0.987690703, and Linear Regression model weights are 0.004599966. The formula for calculating the fused output prediction value is: ; In the formula, To fuse the predicted output values, , , These are the independent predicted values ​​for the least squares model, the support vector machine model, and the linear regression model, respectively. 3.5 Output layer results, output layer outputs three types of results: First, the evaluation indicators for the fit effect include the coefficient of determination R² of 0.9998, the root mean square error of 0.047, and the mean absolute error of 0.7852. Second, the predicted seepage flow rates for the next 7 days from November 24, 2025 to November 30, 2025, including the independent predictions from each model and the fused output predictions, are shown in the table below:

[0020] Third, the 7-day forecast curve, such as Figure 5 As shown.

[0021] Example 2: A multi-method collaborative early warning method for dam monitoring data: This embodiment is based on the prediction results of Embodiment 1, and uses the early warning method of the present invention to conduct early warning of dam seepage safety. The specific steps are as follows: T1: Obtain the prediction result: The fusion fitting results and the predicted seepage flow rate for the next 7 days from November 24, 2025 to November 30, 2025 were obtained from the prediction model output of Example 1. T2: Constructing the prediction confidence interval: Calculate the residual standard deviation between the predicted and measured values ​​for each of the three models using the following formula: ; In the formula, These are the historical measured seepage values ​​for the WE_DB1 measuring point. These are the predicted values ​​from the corresponding model. This refers to the sample size of historical monitoring data. Substituting the data, we calculated that the residual standard deviation of the least squares model is 6.04, the residual standard deviation of the support vector machine model is 0.05, and the residual standard deviation of the linear regression model is 10.12. The arithmetic mean of the three residual standard deviations is taken as the mean residual standard deviation, and the mean residual standard deviation is calculated to be approximately 5.40. A confidence interval is constructed using a 95.4% confidence level, and the boundary calculation formula is as follows: ; Substituting the mean residual standard deviation value, we get: ; Taking the fusion output prediction of 86.70 liters per second on November 24, 2025 as an example, its corresponding 95.4% confidence interval has a lower limit of 75.9 liters per second and an upper limit of 97.7 liters per second. The correspondence between the fusion prediction and the confidence interval is as follows: Figure 4 As shown; T3: Early Warning Linkage and Visualization; Establish an early warning linkage mechanism to compare the predicted value at each time step with the upper and lower boundaries of the corresponding confidence interval in real time. The early warning is triggered when the predicted value at two or more consecutive time steps exceeds the confidence interval. Suppose the system detects a predicted value of 98.50 liters per second on November 26, 2025, which exceeds the upper limit of the confidence interval for that day, 97.7 liters per second; and a predicted value of 99.20 liters per second on November 27, 2025, which again exceeds the upper limit of the confidence interval for that day, 97.9 liters per second. If the trigger condition of two consecutive time step predicted values ​​exceeding the confidence interval is met, the system will automatically trigger a yellow alert, corresponding to a mild anomaly risk level, and generate the following structured alarm information: Alarm number: ALERT-20251124-001; Measurement point number: WE_DB1; Anomaly prediction: 99.20 liters per second; Confidence interval range: lower limit 76.3 liters per second, upper limit 97.9 liters per second; Trigger time: 08:00 on November 24, 2025; Risk level: Yellow; Meanwhile, ECharts is used to generate a time series graph containing the prediction curve and confidence interval boundaries, and abnormal time points are highlighted in red and pushed to the dam safety management platform.

[0022] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-method collaborative method for predicting dam monitoring data, characterized in that, Includes the following steps: S1: Obtain historical monitoring data of the dam through dam safety monitoring instruments, and determine key influencing factors such as upstream and downstream water levels, rainfall, and temperature based on the type of dam monitoring points; S2: The comparison method, plotting method and threshold method are used to remove gross errors in the original monitoring data. Missing data are filled in a hierarchical manner: when there are ≤3 consecutive missing data, the moving average interpolation method is used to fill the gaps; when there are >3 consecutive missing data, the autoregressive integral moving average model is used to fill the gaps. The timestamp format of all data is unified. S3: Construct a multi-method collaborative dam monitoring data prediction model, which includes an input layer, a three-model independent prediction layer, and a fusion decision layer. Processed multi-source monitoring data is input into the input layer, sequentially undergoing data preprocessing and feature engineering, three-model independent prediction, and dynamic weighted average fusion decision based on root mean square error. The output layer outputs the fused fitting results, future predicted values, and fitting performance evaluation metrics. The three independent prediction layers are established by linear regression model, least squares model and support vector machine model respectively. The three models are trained and predicted independently using the same training set and test set, and there is no parameter sharing or data interaction between the models. The fusion decision layer calculates the root mean square error (RMSE) of each model on the test set, uses the inverse proportion of each model's RMSE as the weight of the corresponding model, and performs a weighted average of the independent predictions of the three models to obtain the fusion prediction value.

2. A multi-method collaborative early warning method for dam monitoring data, characterized in that, Includes the following steps: T1: Obtain the fusion fitting results and future prediction values ​​output by the multi-method collaborative dam monitoring data prediction method; T2: Calculate the residual standard deviations between the predicted and measured values ​​for the linear regression model, least squares model, and support vector machine model, respectively, and take the arithmetic mean of the three residual standard deviations as the mean residual standard deviation. A confidence interval for the predicted values ​​was constructed using a 95.4% confidence level. T3: Establish an early warning linkage mechanism to compare the future predicted value of each time step with the upper and lower boundaries of the corresponding confidence interval in real time. When the triggering conditions are met, an alarm is automatically triggered and a structured alarm message is generated. At the same time, the predicted value and confidence interval for the next n days are visualized and a time series diagram containing the prediction curve and the confidence interval boundary is generated.

3. The multi-method collaborative dam monitoring data prediction method according to claim 1, characterized in that, In step S1, the dam safety monitoring instruments include seepage, deformation, stress-strain, and environmental instruments such as rain gauges, water level gauges, and thermometers / hygrometers; auxiliary instruments include crack gauges and seismographs. Deformation-type instruments include inclinometers, convergence meters, total stations, and GPS devices; Stress-strain instruments include strain gauges, rebar gauges, and pressure cells; Environmental instruments include rain gauges, water level gauges, and thermometers / hygrometers; auxiliary instruments include crack gauges and seismographs.

4. The multi-method collaborative dam monitoring data prediction method according to claim 1, characterized in that, The criteria for removing gross errors in step S2 are as follows: if all three of the following conditions are met, the data is determined to be a gross error and removed: 4.1 The monitored value at a certain moment exceeds the historical reasonable threshold range for that monitoring point; 4.2 The deviation between the monitored value and the monitored values ​​at adjacent time points exceeds the preset threshold; 4.

3. No abnormal mutations were observed in the key influencing factors at the corresponding time point.

5. The multi-method collaborative dam monitoring data prediction method according to claim 1, characterized in that, In step S2, the sliding window radius of the moving average interpolation method is 3 to 7 days, and the total window size is 2k+1 data points, where k is the sliding window radius. The calculation formula is as follows: ； In the formula, For the missing values ​​at time point t, interpolation estimates The known observations at time point t+i; The autoregressive integral moving average model is ARIMA(p,d,q), where p is the order of the autoregressive term, d is the difference order, and q is the order of the moving average term.

6. The multi-method collaborative dam monitoring data prediction method according to claim 1, characterized in that, In step S3, the input layer receives multi-source monitoring data aligned to timestamps, including deformation values, stress values, seepage values, water levels, rainfall, and air temperature. The data preprocessing and feature engineering layers perform secondary verification of missing values, feature standardization or normalization, and construct a sliding time window dataset for multi-step prediction modeling.

7. The multi-method collaborative dam monitoring data prediction method according to claim 1, characterized in that, The formula for calculating the weights of the fusion decision layer in step S3 is as follows: ； In the formula, Let be the weight of the i-th model, and the sum of the weights of all models is 1.000; The formula for calculating the fused output prediction value is: ； In the formula, To fuse the predicted output values, , , These are the weights for the least squares model, the support vector machine model, and the linear regression model, respectively. , , These are the independent predicted values ​​for the corresponding models.

8. The multi-method collaborative dam monitoring data early warning method according to claim 2, characterized in that, The formula for calculating the boundaries of the confidence interval in step T2 is: ； In the formula, 1 represents the fused future prediction value output by the multi-method collaborative dam monitoring data prediction method, and 2 represents the Z-value corresponding to the 95.4% confidence level under the standard normal distribution.

9. The multi-method collaborative dam monitoring data early warning method according to claim 2, characterized in that, In step T3, the warning trigger condition is that the predicted value exceeds the confidence interval for two or more consecutive time steps; the structured alarm information includes the measurement point number, the predicted value exceeding the confidence interval, the corresponding confidence interval range, the trigger time, and the risk level; The risk levels are divided into yellow (mildly abnormal), orange (moderately abnormal), and red (severely abnormal).

10. The multi-method collaborative dam monitoring data prediction method according to claim 1, characterized in that, When applied to the monitoring of total seepage downstream of a rockfill dam with a concrete panel, total seepage monitoring points are set up downstream of the dam, and seepage flow data are collected using a measuring weir. The monitoring frequency is once per day. The system simultaneously collects daily cumulative rainfall, daily average temperature, and upstream and downstream water level data. All monitoring data are stored in CSV format, including timestamp, monitoring point number, flow rate, rainfall, temperature, upstream water level, and downstream water level fields. During data preprocessing, the Z-score standardization method was used to unify the dimensions of all features, and a sliding time window dataset with an input window of 7 days and an output window of 7 days was constructed for seepage prediction in the next 7 days.

Images