Monitoring method for prediction and early warning of hidden leakage of concrete dam

Abstract

The invention relates to a monitoring method for prediction and early warning of hidden leakage of a concrete dam, and aims to solve the technical problem of lack of timely prediction and early warning of hidden leakage danger of the dam at present, and adopts the technical scheme that a long and short-term memory network model fused with an attention mechanism is constructed, and leakage data is modeled and predicted; an improved CUSUM change point detection algorithm is introduced, a sliding monitoring window and a residual threshold are set, an accumulated residual between a predicted value and a measured value is calculated, the accumulated residual is compared with the threshold after being processed by a control function, the leakage trend is recognized and early warned in advance, and a real-time meteorological sensitive factor is introduced, so that the leakage trend is accurately detected. The monitoring sensitivity is dynamically adjusted under extreme events such as rainstorm and earthquake, and the response capability to the sudden leakage risk is improved. The early warning accuracy and timeliness of the dam leakage accident can be effectively improved, time is won for dangerous case disposal, and the life and property safety of people is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of water conservancy and hydropower engineering, and particularly relates to a monitoring method for predicting and warning hidden leakage of concrete dams. Background Technique

[0002] Seepage of dams is one of the main causes of dam accidents. In recent years, due to the frequent occurrence of dam accidents caused by dam seepage, accounting for about 29.1% of dam-break accidents. Such hidden leakage danger phenomena are difficult to detect only by manual inspection or sampling monitoring methods. With the increasingly wide application of intelligent technology in the field of dam monitoring, many new algorithms can well complete the prediction of dam seepage-related data under normal conditions. However, there is currently a lack of research on the prediction and timely warning of hidden leakage danger caused by emergencies (such as floods, earthquakes) or internal structural erosion damage of dams. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and provide a monitoring method for predicting and warning hidden leakage of concrete dams.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is:

[0005] A monitoring method for predicting and warning hidden leakage of concrete dams includes the following steps:

[0006] Step 1) Arrange a seepage pressure sensor network in the potential leakage risk area of the dam body to collect seepage pressure monitoring data in real time;

[0007] Step 2) Construct a long short-term memory network model based on the attention mechanism (AT-LSTM) to perform time series modeling and prediction on the seepage pressure monitoring data, and generate seepage pressure prediction values;

[0008] Step 3) Adopt an improved CUSUM change point monitoring algorithm to calculate the residuals between the predicted values and the measured values in each sliding window period, accumulate them after being processed by the acceptance function, and compare the results with the cumulative residual threshold in real time, so as to realize the early warning of leakage trends;

[0009] Step 4) Introduce a real-time meteorological sensitive factor, and dynamically adjust the residual threshold according to rainfall and earthquake intensity to enhance the monitoring sensitivity under extreme events.

[0010] Furthermore, the arrangement area of the seepage pressure sensor network in Step 1) includes:

[0011] Dam foundation seepage pressure monitoring area: Arrange 2-3 layers of piezometers vertically along the dam foundation, with a spacing of 5-10m between each layer; Horizontally, arrange one piezometer every 10-20m according to the dam width distribution, and install seepage pressure sensors at the heel, toe, joints, and faults of the dam;

[0012] Lower and middle part layout area of the dam body: Multiple monitoring profiles are arranged along the dam height. 3 to 5 sensing points are arranged from top to bottom in each monitoring profile, and one piezometric sensor is buried in each monitoring profile.

[0013] Upstream and downstream slope areas: A number of deep piezometric sensors are symmetrically arranged inside the upstream and downstream slopes. The burial depth is set according to 1 / 3 or 1 / 2 of the dam height, combined with the arrangement inside the dam body to form a monitoring "section grid".

[0014] Furthermore, the long short-term memory network model based on the attention mechanism (AT-LSTM) in step 2) includes an input layer, a hidden layer, an attention layer, a fully connected layer, and an output layer.

[0015] The input layer is used to input multi-dimensional time series data and perform format conversion.

[0016] The hidden layer: consists of multiple LSTM cell units and is used to process time series data.

[0017] The attention layer: calculates the attention distribution at different times, dynamically assigns weights to each time point, and highlights key influencing factors.

[0018] The fully connected layer: converts data and performs local integration on the data.

[0019] The output layer: outputs the piezometric prediction result.

[0020] By introducing the attention mechanism to dynamically focus on the importance degrees of various elements at different times, ignoring irrelevant information, attaching importance to key information, and incorporating it into the memory network, the prediction performance of the model can be improved.

[0021] Furthermore, the specific steps of the early warning in step 3) are as follows:

[0022] Step 3.1) Use the AT-LSTM model in step 2) to predict the seepage data of each time node of the dam in real time, and actually measure the seepage data of this time node.

[0023] Seepage data set predicted by the model: Y pv ={y pv(i)}, i = 1, 2, 3...n;

[0024] Seepage data set actually measured: Y gt ={y gt(i)}, i = 1, 2, 3...n;

[0025] i is each monitoring time node starting from the initial monitoring point, in days.

[0026] Step 3.2) Starting from the initial monitoring point, for the actual measurement data y at each time point gt(i) and the reasonable prediction data y pv(i) to perform subtraction, and the residual value is Δy (i) :

[0027] Δy (i) = y gt(i) - y pv(i) = 1, 2, 3... n

[0028] Step 3.3) Every time a new measurement time node appears, a new round of monitoring window period T is started from Δy at this time node new : new :

[0029]

[0030] If within the specified period after this time node, the accumulated positive residual value does not reach the warning threshold S upper , the accumulated value is cleared and this window exits:

[0031]

[0032] If the accumulated residual value reaches the warning threshold S upper , then a warning report is triggered to pre-judge the development trend of the dangerous situation;

[0033]

[0034] In the formula, T is the sliding monitoring window period, S upper is the accumulated residual threshold, Δy (i) + is the positive residual, and A(t) is the acceptance function;

[0035]

[0036] Furthermore, in the said step 4), the dynamic adjustment of the residual threshold by the rainfall is as follows:

[0037] Normal rainfall (0 mm / h - 10 mm / h): Maintain the normal monitoring sensitivity of the CUSUM algorithm, and no special adjustment is required;

[0038] Moderate rainfall (10 mm / h - 50 mm / h): Such rainfall may exert a certain pressure on the dam body and increase the seepage risk. Under this condition, the accumulated residual threshold of the CUSUM algorithm is reduced by 20% - 30% to enhance the monitoring sensitivity to seepage anomalies;

[0039] Heavy rain (>50 mm / h): Heavy rain may cause significant changes in seepage, and it is necessary to significantly reduce the cumulative residual threshold of the CUSUM algorithm by 30% - 60%. In the case of floods or high water storage upstream, the data collected by the sliding window is changed from daily to hourly for analysis.

[0040] The dynamic adjustment of the residual threshold according to the earthquake intensity is as follows:

[0041] Minor earthquake (epicenter magnitude 3.0 - 4.0): Adjust the residual threshold monitored by CUSUM, reduce it by 10% - 20%, and increase the detection of minor changes;

[0042] Moderate earthquake (epicenter magnitude 4.1 - 5.0): Adjust the residual threshold monitored by CUSUM, reduce it by 30% - 50%, so as to timely capture possible structural deformations or changes in seepage paths after the earthquake.

[0043] Strong earthquake (epicenter magnitude >5.0): Adjust the residual threshold monitored by CUSUM, reduce it by 50% - 80%, and change the data collected by the sliding window from daily to hourly for analysis.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] 1. The present invention constructs a long short-term memory network model integrating an attention mechanism to model and predict leakage data, and introduces an improved CUSUM change point detection algorithm. By setting a sliding monitoring window and a residual threshold, the cumulative residual between the predicted value and the measured value is calculated, and after being processed by a control function, it is compared with the threshold to realize the early identification and warning of the leakage trend;

[0046] 2. Based on the traditional CUSUM (cumulative sum) algorithm, the present invention proposes a time series residual-driven improved CUSUM algorithm for dam leakage risk identification. This algorithm integrates a sliding window mechanism and an intelligent prediction model (AT-LSTM) to realize the early warning of abnormal seepage trends;

[0047] 3. The present invention introduces real-time meteorological sensitive factors to dynamically adjust the monitoring sensitivity under extreme events such as heavy rain and earthquakes, and improves the response ability to sudden leakage risks;

[0048] 4. The present invention can effectively improve the warning accuracy and timeliness of dam leakage accidents, gain time for emergency handling of dangerous situations, and ensure the safety of people's lives and property. Description of the Drawings

[0049] Figure 1 It is a schematic diagram of the sensor layout structure of the present invention;

[0050] Figure 2Schematic diagram of the process of the present invention;

[0051] Figure 3 Schematic diagram of the framework of the AT-LSTM network model of the present invention;

[0052] Figure 4 Schematic diagram of the principle of the improved CUSUM algorithm;

[0053] Figure 5 Demonstration diagram of the improved CUSUM early warning process of the embodiment of the present invention;

[0054] Figure 6 Diagram of the sensitivity analysis results of parameter combinations of the embodiment of the present invention; Detailed implementation manners

[0055] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0056] As Figure 2 shown, a monitoring method for predicting and warning the concealed leakage of a concrete dam includes the following steps:

[0057] Step 1) Arrange a piezometric sensor network in the potential leakage risk area of the dam body, and collect piezometric monitoring data in real time;

[0058] As Figure 1 shown, piezometric sensors are reasonably arranged in the key areas of the dam body and the dam foundation (including the heel of the dam, the toe of the dam, the middle section, and the upstream and downstream slopes) to form a sensing network that comprehensively covers in space and continuously monitors in time, providing high-frequency and high-reliability real-time piezometric data input for the algorithm, and supporting the rapid identification and early warning judgment of abnormal dam leakage. The specific arrangement method is as follows:

[0059] Piezometric monitoring area of the dam foundation: Arrange 2 to 3 layers of piezometers vertically along the dam foundation, with a spacing of 5 to 10 m between each layer; arrange one piezometric sensor every 10 to 20 m horizontally according to the dam width, and piezometric sensors are arranged at the heel of the dam, the toe of the dam, joints, and faults to monitor whether there is an abnormal pressure increase trend in the seepage channels of the dam foundation;

[0060] Arrangement area in the middle and lower parts of the dam body: Arrange a monitoring section every 1 / 4 of the dam length along the dam height. Each monitoring section is provided with 3 to 5 sensing points from top to bottom, and a piezometric sensor is buried in each monitoring section. The burial depth is selected according to the design section or key parts (such as construction joints and deformation joints) to cover the horizontal and vertical piezometric changes of the main body of the dam and reflect the structural risks;

[0061] Upstream and downstream slope areas: Symmetrically arrange several deep piezometric sensors inside the upstream and downstream slopes; the burial depth is set according to 1 / 3 or 1 / 2 of the dam height, combined with the arrangement inside the dam body to form a monitoring "section grid" to judge the change of the water head seepage path and the development trend of the abnormal seepage path.

[0062] The osmotic pressure sensor adopts a vibrating wire type or fiber optic type osmometer, has the ability of waterproof and anti-corrosion, and supports long-term stable operation and remote data acquisition and transmission.

[0063] All sensors are connected to the upper computer system through the data acquisition module; support RS485 or fiber optic communication protocol; the real-time acquisition frequency is once per hour, and the dynamic sampling frequency can be set according to the algorithm requirements.

[0064] The data is automatically input into the AT-LSTM prediction module and triggers the CUSUM monitoring mechanism.

[0065] By reasonably arranging osmotic pressure sensors in the key areas of the dam body and foundation (including the heel of the dam, the toe of the dam, the middle section, and the upstream and downstream slopes), a sensing network that comprehensively covers in space and continuously monitors in time is formed, providing high-frequency and high-reliability real-time osmotic pressure data input for the algorithm, and supporting the rapid identification and early warning judgment of abnormal dam leakage.

[0066] Step 2) Construct a long short-term memory network model based on the attention mechanism (AT-LSTM) to perform time series modeling and prediction on the osmotic pressure monitoring data, and generate osmotic pressure prediction values.

[0067] Compared with traditional neural networks, the long short-term memory neural network LSTM can effectively solve the forgetting problem and can stretch the time unit of the processed time series data. Its core idea is to construct cell units, selectively remember important information, filter out noise information, and reduce the memory burden. The structure includes three gates, an input gate, an output gate, and a forget gate. The output vector of the structure is divided into a current state vector and an output vector. The inputs of the structure are the state vector of the previous moment unit, the output vector of the previous moment unit, and the input vector of the current moment.

[0068] The calculation formula for each cell unit is as follows:

[0069] h (t) =o t ·tanh(C (t) )

[0070]

[0071] Where W is the weight matrix; b is the bias value vector; the σ (sigmoid) function takes values between 0 and 1. When multiplying matrices, those elements with a value of 0 can be erased, which is equivalent to selectively forgetting part of the memory and becoming the forget gate; the tang function takes values between -1 and 1 and is used to summarize information.

[0072] Attention Mechanism: The LSTM is good at processing time series data and capturing dependencies in the sequence. However, it gives the same attention to all input information, which affects the prediction accuracy. Therefore, the attention mechanism is introduced. It is like installing a focusing device in data processing, which can assign different weights according to the importance of each factor to the leakage situation, thus highlighting the key factors.

[0073] The attention mechanism consists of three input vectors, namely Q (Query), K (Key), and V (Value).

[0074] Among them: Q is the query variable, used to obtain the correlation with other variables; K is the key vector, reflecting the degree of association between the corresponding input element and the query vector; V is the value vector, which is a vector containing actual data information.

[0075] The operation steps are as follows:

[0076] Data encoding: Convert the input data into corresponding <Key, Value> key-value pairs through appropriate encoding;

[0077] Calculate the correlation: Calculate the correlation between the query vector and each key vector, and obtain the attention weight corresponding to each input element after normalization processing;

[0078] Information weighted aggregation: According to the weight coefficient W i Perform weighted summation on each value vector.

[0079] Attention weight coefficient W:

[0080] W = softmax(QK T )

[0081] The output after being processed by the attention mechanism:

[0082] Attention(Q, K, V) = W · V = softmax(QK T ) · V

[0083] As Figure 3 shown, the long short-term memory network model based on the attention mechanism (AT-LSTM) includes an input layer, a hidden layer, an attention layer, a fully connected layer, and an output layer;

[0084] The input layer is used to input multi-dimensional time series data and perform format conversion;

[0085] The hidden layer consists of multiple LSTM cell units and is used to process time series data;

[0086] The attention layer calculates the attention distribution at different times, dynamically assigns weights to each time point, and highlights the key influencing factors;

[0087] The fully connected layer converts the data and performs local integration on the data;

[0088] The output layer outputs the seepage pressure prediction result.

[0089] By introducing the attention mechanism, the importance degree of various elements at different times is dynamically concerned, irrelevant information is ignored, key information is emphasized, and after incorporating the memory network, the prediction performance of the model can be improved.

[0090] Step 3) Use an improved CUSUM change point detection algorithm, as Figure 4 shown, calculate the residuals between the predicted values and the measured values within each sliding window period, accumulate them after being processed by the acceptance function, and compare the results with the cumulative residual threshold in real time, so as to realize the early warning of the leakage trend. The specific steps are as follows:

[0091] Step 3.1) Use the AT-LSTM model in step 2) to predict the seepage data of each time node of the dam in real time, and actually measure the seepage data of this time node;

[0092] Seepage data set predicted by the model: Y pv ={y pv(i)}, i = 1, 2, 3... n;

[0093] Seepage data set actually measured: Y gt ={y gt(i)}, i = 1, 2, 3... n;

[0094] i is each monitoring time node starting from the initial monitoring point, in days;

[0095] Step 3.2) Starting from the initial monitoring point, subtract the actual measurement data y gt(i) from the reasonable prediction data y pv(i) , and the residual value is Δy (i) :

[0096] Δy (i) =y gt(i) -y pv(i) =1, 2, 3... n

[0097] Step 3.3) Every time a new measurement time node appears, a new round of monitoring window period T new is started from this time node Δy new :

[0098]

[0099] If within the specified period after this time node, the cumulative value of positive residuals is collected and does not reach the warning threshold S upper , the cumulative value is cleared and this window exits:

[0100]

[0101] If the cumulative residual value reaches the warning threshold S upper , a warning report is triggered to preliminarily judge the development trend of the danger;

[0102]

[0103] In the formula, T is the sliding monitoring window period, S upper is the cumulative residual threshold, Δy (i) + is the positive residual, and A(t) is the acceptance function;

[0104]

[0105] Step 4) Introduce a real-time meteorological sensitivity factor, and dynamically adjust the residual threshold according to rainfall and earthquake intensity to enhance the monitoring sensitivity under extreme events;

[0106] Rainfall is the most direct meteorological factor affecting dam seepage. Especially heavy rain may cause phenomena such as increased surface water flow of the dam, changes in seepage channels, and later mountain floods. The dynamic adjustment of the residual threshold by rainfall is as follows:

[0107] Normal rainfall (0 mm / h - 10 mm / h): Maintain the normal monitoring sensitivity of the CUSUM algorithm without special adjustment;

[0108] Moderate rainfall (10 mm / h - 50 mm / h): Such rainfall may exert a certain pressure on the dam and increase the risk of seepage. Under this condition, the cumulative residual threshold of the CUSUM algorithm is reduced by 20% - 30% to enhance the monitoring sensitivity to seepage anomalies;

[0109] Heavy rain (>50 mm / h): Heavy rain may cause large-scale seepage changes, and it is necessary to significantly reduce the cumulative residual threshold of the CUSUM algorithm by 30% - 60%. In the case of floods or upstream high-level water storage, etc., the data collected by the sliding window is changed from daily to hourly for analysis;

[0110] The impact of earthquake intensity on the dam structure may cause cracks, deformations or changes in seepage paths. The dynamic adjustment of the residual threshold by the earthquake intensity is as follows:

[0111] Small earthquake (epicenter magnitude 3.0 - 4.0): Adjust the residual threshold for CUSUM monitoring, reducing it by 10% - 20% to increase the detection of minor changes;

[0112] Moderate earthquake (epicenter magnitude 4.1 - 5.0): Adjust the residual threshold for CUSUM monitoring, reducing it by 30% - 50% to promptly capture possible structural deformations or changes in seepage paths after an earthquake;

[0113] Strong earthquake (epicenter magnitude > 5.0): Adjust the residual threshold for CUSUM monitoring, reducing it by 50% - 80%, and change the data collection of the sliding window from daily to hourly for analysis.

[0114] In step 3) of the present invention, based on the traditional CUSUM (Cumulative Sum) algorithm, a time - series residual - driven improved CUSUM algorithm for dam leakage risk identification is proposed. This algorithm integrates a sliding window mechanism and an intelligent prediction model (AT - LSTM) to achieve early warning of abnormal seepage trends. The specific improvements are as follows:

[0115] 1. Introduction of the residual - driven prediction mechanism

[0116] The traditional CUSUM algorithm relies on the statistical change trend of fixed monitoring data for anomaly detection and is easily affected by data non - stationarity. In the present invention, an LSTM model based on the attention mechanism (AT - LSTM) is first used to predict the seepage value of the dam at the next moment in real - time, and then the difference between the predicted value and the real - time measured value is calculated to form a dynamic residual sequence.

[0117] 2. Setting of the sliding window period

[0118] Whenever a new measurement time point arrives, the system automatically starts a new sliding monitoring window from the current time point and continuously monitors the subsequent residual fluctuation trend within this window.

[0119] This mechanism realizes real - time dynamic monitoring and multi - cycle coverage, enabling the early warning mechanism to have continuity and immediacy and being able to respond promptly to new signs of danger.

[0120] 3. Positive residual accumulation mechanism

[0121] This algorithm focuses on the cumulative total of positive residuals and uses an acceptance function to perform non - linear amplification and screening on positive residuals. This mechanism aims to improve the ability to identify the trend of increasing abnormal leakage and avoid false triggering caused by false alarms or noise interference.

[0122] Table 1 Main differences and improvements between the improved CUSUM algorithm and the traditional CUSUM algorithm

[0123]

[0124] As Figure 5 shown, combining accident examples, this paper tests the warning effect of the LSTM-CUSUM model by simulating seepage data when a leakage danger occurs. It is found that the algorithm combined with LSTM predicted data can effectively identify the gradually changing trend and early warn of the mutation risk. The hidden leakage sign starts to appear on the 60th day of the simulation. CUSUM determines the mutation trend around the 72nd day through residual accumulation, about eight days earlier than the ordinary outlier detection.

[0125] As Figure 6 shown, different monitoring periods and cumulative residual thresholds are selected within a reasonable range for combination: (T, S upper , to observe the response of this algorithm;

[0126] C: [85, 2.0] determines the mutation trend around the 70th day, C: [90, 2.1] determines the mutation trend around the 71st day, C: [95, 2.2] determines the mutation trend around the 73rd day, C: [100, 2.3] determines the mutation trend around the 75th day, C: [105, 2.4] determines the mutation trend around the 76th day, C: [110, 2.5] determines the mutation trend around the 78th day, all earlier than the conventional outlier detection at 80 days. Therefore, different parameter combinations selected under this simulated leakage danger can achieve the effect of early warning, indicating that the model has a certain selection space for parameter selection.

Claims

1. A monitoring method for predicting and warning of hidden leakage in concrete dams, characterized in that, It includes the following steps: Step 1) Arrange a seepage pressure sensor network in the potential seepage risk area of the dam body to collect seepage pressure monitoring data in real time; Step 2) Construct a long short-term memory network model based on the attention mechanism (AT-LSTM) to perform time series modeling and prediction on the seepage pressure monitoring data and generate seepage pressure prediction values; Step 3) Adopt an improved CUSUM change point monitoring algorithm to calculate the residuals between the predicted values and the measured values within each sliding window period, accumulate them after being processed by the acceptance function, and compare the results with the cumulative residual threshold in real time, so as to achieve early warning of the seepage trend; Step 4) Introduce real-time meteorological sensitive factors, and dynamically adjust the residual threshold according to rainfall and seismic intensity to enhance the monitoring sensitivity under extreme events.

2. The monitoring method for predicting and warning the hidden leakage of a concrete dam according to claim 1, characterized in that, The arrangement area of the seepage pressure sensor network in the said Step 1) includes: Dam foundation seepage pressure monitoring area: Arrange 2 - 3 layers of piezometers vertically along the dam foundation, with a spacing of 5 - 10m between each layer; Horizontally, arrange one every 10 - 20m according to the dam width distribution, and arrange seepage pressure sensors at the heel, toe, joints, and faults of the dam; Arrangement area in the middle and lower parts of the dam body: Arrange multiple monitoring profiles along the dam height, with 3 - 5 sensing points arranged from top to bottom in each monitoring profile, and bury one seepage pressure sensor in each monitoring profile; Upstream and downstream slope areas: Symmetrically arrange several deep seepage pressure sensors inside the upstream and downstream slopes; The burial depth is set according to 1 / 3 or 1 / 2 of the dam height, combined with the arrangement inside the dam body to form a monitoring "section grid".

3. A monitoring method for predicting and warning of hidden leakage in concrete dams according to claim 1, characterized in that, The long short-term memory network model based on the attention mechanism (AT-LSTM) in the said Step 2) includes an input layer, a hidden layer, an attention layer, a fully connected layer, and an output layer; The input layer is used to input multi-dimensional time series data and perform format conversion; The hidden layer: Consists of multiple LSTM cell units and is used to process time series data; The attention layer: Calculates the attention distribution at different times, dynamically assigns weights to each time point, and highlights key influencing factors; The fully connected layer: Converts data and performs local integration on the data; The output layer: Outputs the seepage pressure prediction result; By introducing the attention mechanism to dynamically focus on the importance degrees of various elements at different times, ignoring irrelevant information, attaching importance to key information, and incorporating it into the memory network, the prediction performance of the model can be improved.

4. The monitoring method for predicting and warning of hidden leakage in concrete dams according to claim 1, characterized in that, The specific steps of the early warning in the said Step 3) are as follows: Step 3.1) Use the AT-LSTM model in Step 2) to predict the seepage data of each time node of the dam in real time, and actually measure the seepage data of this time node; Model-predicted seepage dataset: Y pv ={y pv(i)}, i = 1, 2, 3...n; Actual measured seepage data set: Y gt ={y gt(i)},i = 1, 2, 3...n; i is each monitoring time node starting from the initial monitoring point, in days; Step 3.2) Starting from the initial monitoring point, for the actual measurement data y at each time point gt(i) and the reasonably predicted data y pv(i) perform a difference operation, and the residual value is Δy (i) : Δy (i) = y gt(i) - y pv(i) where i = 1, 2, 3...n Step 3.3) Whenever a new measurement time node appears, start a new monitoring window period T from this time node Δy new and new : If within the specified period after this time node, the cumulative value of positive residuals is collected and does not reach the warning threshold S upper , the cumulative value is cleared and this window exits: If the cumulative residual value reaches the warning threshold S upper , a warning report is triggered to preliminarily determine the development trend of the danger where T is the sliding monitoring window period, and S upper is the cumulative residual threshold, and Δy (i) + is the positive residual, and A(t) is the acceptance function; 5. The monitoring method for predicting and warning of concealed leakage in concrete dams according to claim 1, characterized in that, In the said Step 4), the dynamic adjustment of the residual threshold by the rainfall is as follows: Normal rainfall (0mm / h - 10mm / h): Maintain the normal monitoring sensitivity of the CUSUM algorithm without special adjustment; Moderate rainfall (10mm / h - 50mm / h): Such rainfall may generate a certain pressure on the dam body, increasing the seepage risk. Under this condition, the cumulative residual threshold of the CUSUM algorithm is reduced by 20% - 30% to enhance the monitoring sensitivity to seepage anomalies; Heavy rain (>50 mm / h): Heavy rain may cause significant changes in seepage, and it is necessary to significantly reduce the cumulative residual threshold of the CUSUM algorithm by 30% - 60%. In case of floods or upstream high-level water storage, etc., the data collected by the sliding window should be changed from daily collection to hourly collection for analysis; The dynamic adjustment of the residual threshold according to the earthquake intensity is as follows: Minor earthquake (epicenter magnitude 3.0 - 4.0): Adjust the residual threshold monitored by CUSUM, reduce it by 10% - 20%, and increase the detection of minor changes; Moderate earthquake (epicenter magnitude 4.1 - 5.0): Adjust the residual threshold monitored by CUSUM, reduce it by 30% - 50%, in order to promptly capture possible structural deformations or changes in seepage paths after the earthquake; Strong earthquake (epicenter magnitude >5.0): Adjust the residual threshold monitored by CUSUM, reduce it by 50% - 80%, and change the data collected by the sliding window from daily collection to hourly collection for analysis.

Images