Dam safety monitoring method driven by earthquake monitoring data

Abstract

The invention discloses a dam safety monitoring method driven by earthquake monitoring data, and belongs to the technical field of safety monitoring of water conservancy and hydropower engineering. The method comprises the following steps: step 1, establishing an original record table; step 2, determining input characteristics as magnitude, seismic source depth, dam strong seismic monitoring point elevation, epicentral distance and azimuth angle, and determining prediction targets as peak acceleration and trigger time; step 3, adopting an adaptive neural fuzzy inference system to carry out prediction model training, respectively establishing an optimal ANFIS model, and obtaining a final peak acceleration prediction result and a final trigger time prediction result; and 4, connecting the reservoir earthquake monitoring station network with a dam safety monitoring system by using a network. The dam safety monitoring system can be started before earthquake waves reach the dam, so that deformation and the like of the dam caused by the earthquake are completely recorded, and the reliability of quick starting of the dam safety monitoring system is remarkably improved.

Description

Technical Field

[0001] The invention relates to a dam safety monitoring method driven by earthquake monitoring data, belonging to the technical field of water conservancy and hydropower engineering safety monitoring. Background Art

[0002] Dam safety monitoring is one of the conventional monitoring methods for water conservancy and hydropower projects. At present, dam safety monitoring facilities are installed on most reservoir dams, and with the advancement of monitoring technology, dam safety monitoring systems have been widely developed to achieve comprehensive monitoring and automatic collection of dam safety data. However, the current dam safety data monitoring is mainly based on periodic automatic recording.

[0003] A few dams are also equipped with strong vibration monitoring facilities. After detecting strong vibrations, these facilities will automatically start the dam safety monitoring system to collect data, so that the dam safety monitoring system can record the monitoring results of the dam after the earthquake. These results reflect the impact of the earthquake on the displacement and deformation of each monitoring point of the dam. After the earthquake, relevant staff analyzed the vibration information caused by the earthquake event at the dam monitoring point, including peak acceleration, seismic wave arrival time, etc. The peak acceleration caused by the earthquake event at the dam monitoring point is important information for evaluating the impact of the earthquake on the power station, and it is also one of the parameters that dam managers are most concerned about.

[0004] However, the operation mode of automatically starting the dam safety monitoring system to collect data after the dam strong vibration monitoring facility detects strong vibration has the following shortcomings:

[0005] First, this mode of operation may miss some key data during the time between the start of the earthquake and the activation of the dam safety monitoring system, as the dam safety monitoring system has not yet been activated, resulting in the inability to fully record the deformation process of the dam caused by the earthquake.

[0006] Secondly, the dam strong vibration monitoring equipment is only arranged on the dam and other hydraulic structures, with a narrow distribution range. After a moderate to strong earthquake occurs, it is very easy to cause the complete interruption of the dam strong earthquake monitoring data collection, and thus the dam safety monitoring system cannot be quickly started to collect data, which seriously affects the post-earthquake impact assessment of dams and hydraulic structures after a moderate to strong earthquake occurs. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a dam safety monitoring method driven by earthquake monitoring data.

[0008] The present invention is achieved through the following technical solutions:

[0009] A dam safety monitoring method driven by earthquake monitoring data comprises the following steps:

[0010] Step 1: Establish an original record table: collect and organize historical earthquake data and historical dam monitoring point strong earthquake data, and establish an original record table;

[0011] Step 2: Determine input features and prediction targets: Determine input features as magnitude, focal depth, elevation of dam strong earthquake monitoring point, epicenter distance and azimuth, and determine prediction targets as peak acceleration and trigger time;

[0012] Step 3: Obtain the prediction results of the final peak acceleration and the final trigger time: Use the adaptive neural fuzzy inference system to train the prediction model, establish the optimal ANFIS model, and obtain the final peak acceleration prediction results and the final trigger time prediction results;

[0013] Step 4: Connect the reservoir earthquake monitoring network and the dam safety monitoring system: Use the network to connect the reservoir earthquake monitoring network and the dam safety monitoring system.

[0014] The historical earthquake data in step 1 include the time of occurrence, magnitude and epicenter location, wherein the epicenter location includes its longitude, latitude and focal depth.

[0015] The historical strong earthquake data of the dam monitoring point in step 1 includes the location, triggering time and peak acceleration of the dam strong earthquake monitoring point. The location of the dam strong earthquake monitoring point includes its longitude, latitude and elevation.

[0016] In step 2, the epicenter distance d and the azimuth θ are calculated according to the following formula, and the azimuth θ is calculated with the true north direction as 0 degrees:

[0017]

[0018] Where R is the average radius of the Earth, λ 1 ,φ 1 is the longitude and latitude of the epicenter, λ 2 ,φ 2 It is the longitude and latitude of the dam strong earthquake monitoring point.

[0019] In the step 3, according to the prediction target, a sample set AD for predicting peak acceleration and a sample set TD for predicting trigger time are respectively established, and the two sample sets are respectively divided into a training set and a test set;

[0020] The method of using an adaptive neural fuzzy inference system to train a prediction model, establish an optimal ANFIS model, and obtain a final peak acceleration prediction result and a final trigger time prediction result includes the following steps:

[0021] Step 3.1, fuzzify the training set for predicting peak acceleration, and use the membership function to fuzzify the input of the training set;

[0022] Step 3.2, calculate the incentive strength of all fuzzy rules. The nodes use the form of taking the smallest, algebraic product, bounded product or strong product to calculate the membership function value of the input quantity corresponding to the i-th rule and output the result:

[0023]

[0024] Step 3.3, calculate the percentage of each fuzzy rule's incentive strength to all rule incentive strengths, and the output represents the weight of the i-th rule in all fuzzy rules:

[0025]

[0026] Step 3.4: Create an adaptive correlation function between the normalized excitation intensity and the result function. Each node in this layer corresponds to a fuzzy rule f i , the function is the output of the fuzzy rule when the membership is determined, and the number of function parameters depends on the number of input quantities. The output of the i-th node of this layer is as follows:

[0027]

[0028] Step 3.5: For each fuzzy rule f i The outputs are summed and the final ANFIS output is generated:

[0029]

[0030] Generate initial ANFIS structure;

[0031] Step 3.6: Repeat steps 3.1 to 3.5 for parameter training of the generated initial ANFIS structure to obtain a new ANFIS structure and parameters, and continuously perform rolling optimization to obtain the optimal solution;

[0032] Step 3.7: After obtaining the optimal ANFIS model and parameters, input the test set data of the predicted peak acceleration to obtain the premise parameters of the operation;

[0033] Step 3.8, using the trained ANFIS model, obtain the final peak acceleration prediction result;

[0034] Step 3.9: Perform similar processing and calculation on the training set and test set of the predicted trigger time with reference to steps 3.1 to 3.8 to obtain the final trigger time prediction result.

[0035] The sample set is divided into a training set and a test set in a ratio of 7.5:2.5.

[0036] After obtaining the optimal solution in step 3.6, the training data set is used to adjust the model parameters through forward propagation and back propagation algorithms. The goal is to minimize the prediction error. The evaluation indicators include root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 , coefficient of determination R 2 The size of the regression equation represents the degree of fit to the observed value, and its value range is between 0 and 1. The determination coefficient R 2 The closer the value of is to 1, the more accurate the membership parameter obtained by the algorithm is, that is, the predicted value is closer to the actual value. On the contrary, if the determination coefficient R 2 The closer the value is to 0, the lower the reference value of the fitted equation, that is, the lower the reliability of the obtained membership parameter.

[0037] The evaluation index is calculated using the following set of equations:

[0038]

[0039]

[0040] Where M is the total number of training data points, and are the actual baseline value and predicted value of the mth data point, Outputs the average of the data for all actual values.

[0041] In step 3.8, after an earthquake of a certain magnitude occurs at a certain location, its peak acceleration at the strong earthquake monitoring points of the dams of the target power station is obtained through machine learning; in step 3.9, the time relationship between the propagation of the seismic wave from the epicenter to the strong earthquake monitoring points of the dams of the target power station is obtained through machine learning.

[0042] The reservoir seismic monitoring network in step 4 shares data with national seismic stations in the vicinity of the reservoir. After an earthquake occurs, earthquake information is obtained through the reservoir seismic monitoring network, and the ANFIS model established in step 3 is used to quickly estimate the time it takes for the seismic wave to reach the dam and the peak acceleration of the earthquake at the dam's strong earthquake monitoring point. The estimated value is then pushed to the dam safety monitoring platform for comparison with the start threshold to determine whether to start the dam safety monitoring system for data collection.

[0043] The beneficial effects of the present invention are:

[0044] 1. Directly carry out machine learning on the parameters of historical earthquake data, including epicenter location, magnitude, earthquake occurrence time, and vibration parameter information of historical earthquakes at the strong earthquake monitoring points of each dam of the target power station, including the location, trigger time and peak acceleration of the strong earthquake monitoring points of the dam, to form a prediction model of earthquake information and vibration parameters of the strong earthquake monitoring points of the dam. After the earthquake occurs, the prediction model can be used to quickly and automatically predict the maximum peak acceleration caused by the earthquake at the strong earthquake monitoring points of each dam of the target power station, and judge whether to start the dam safety monitoring system based on this. For earthquakes far away from the dam, the dam safety monitoring system can be started before the seismic wave reaches the dam, so as to achieve a complete record of the deformation caused by the earthquake on the dam.

[0045] 2. The reservoir earthquake monitoring network is arranged around the reservoir. Since the reservoir earthquake monitoring network shares data with the national earthquake stations in the vicinity of the reservoir, it has a wide coverage. After a moderate to strong earthquake occurs, the real-time data transmission of all monitoring points of the reservoir earthquake monitoring network will generally not be interrupted. After an earthquake occurs, the reservoir earthquake monitoring network can take advantage of its wide coverage to ensure that the dam safety monitoring system can be quickly started for data collection, significantly improving the reliability of the rapid start-up of the dam safety monitoring system; ensuring that the post-earthquake impact assessment of dams and hydraulic structures after a moderate to strong earthquake can be carried out normally. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of the present invention;

[0047] Figure 2 A comparison diagram of the peak acceleration predicted value and the actual value of the present invention;

[0048] Figure 3 It is a comparison chart of the trigger time prediction value and the actual value of the present invention. DETAILED DESCRIPTION

[0049] The technical solution of the present invention is further described below, but the scope of protection claimed is not limited to the description.

[0050] like Figures 1 to 3 As shown, the dam safety monitoring method driven by earthquake monitoring data described in the present invention comprises the following steps:

[0051] Step 1: Create an original record table: Collect and organize historical earthquake data within the monitoring range and historical dam monitoring point strong earthquake data, and create an original record table.

[0052] The historical earthquake data in step 1 include the time of occurrence, magnitude and epicenter location, wherein the epicenter location includes its longitude, latitude and focal depth.

[0053] The historical strong earthquake data of the dam monitoring point in step 1 includes the location, triggering time and peak acceleration of the dam strong earthquake monitoring point. The location of the dam strong earthquake monitoring point includes its longitude, latitude and elevation.

[0054] The original record table established is shown in Table 1.

[0055] Table 1 Original record table

[0056]

[0057]

[0058] Step 2: Determine input features and prediction targets: Based on the original record table and earthquake propagation theory analysis, determine the input features as magnitude, focal depth, elevation of the dam strong earthquake monitoring point, epicenter distance and azimuth, and determine the prediction targets as peak acceleration and trigger time.

[0059] In step 2, the epicenter distance d and the azimuth θ are calculated according to the following Haversine formula, and the azimuth θ is calculated with the true north direction as 0 degrees:

[0060]

[0061] Where R is the average radius of the Earth, λ 1 ,φ 1 is the longitude and latitude of the epicenter, λ 2 ,φ 2 is the longitude and latitude of the dam strong earthquake monitoring point. R is a known quantity, the longitude λ of the epicenter 1 , latitude φ 1 , the longitude λ of the dam strong earthquake monitoring point 2 , latitude φ 2 , obtained from the original record table. The input features are sorted as shown in Table 2.

[0062] Table 2 Input feature table

[0063]

[0064]

[0065] Step 3: Obtain the prediction results of the final peak acceleration and the final trigger time: Use the adaptive neural fuzzy inference system to train the prediction model, establish the optimal ANFIS model, and obtain the final peak acceleration prediction results and the final trigger time prediction results.

[0066] In step 3, according to the prediction target, a sample set of predicted peak acceleration is established respectively. AD and the sample set of predicted trigger time TD, and divide the two sample sets into training set and test set respectively. Establish a sample set for predicting peak acceleration AD As shown in Table 3, a sample set of predicted trigger time is established TD As shown in Table 4. Among them, the sample set of predicted peak acceleration AD A total of 920 records or samples are included, and the sample set for predicting trigger time TD There are 716 records or samples in total. The two sample sets are divided into training set and test set in the ratio of 7.5:2.5.

[0067] Table 3 Sample set AD for predicting peak acceleration

[0068] Serial number Magnitude Focal depth Epicenter distance Azimuth Elevation Peak acceleration 1 1.6 10 15.88056 182.06 888 0.393 2 0.8 13 23.81217 70.87 414 1.589 3 1.5 6 4.7665 54.43 414 1.987 4 1.1 5 9.81706 89.49 888 0.897 5 3 4 21.85878 84.88 640 4.037 6 1 7 15.11873 96.84 888 0.32 7 1.9 7 18.35108 80.75 592 0.57 8 0.8 9 6.7574 137.26 1310 4.387 9 0.7 8 6.51134 133.96 1215 0.358 10 1.7 6 20.606 93.35 435 0.236 …… …… …… …… …… …… …… 920 0.9 5 12.711 60.94 1328 0.038

[0069] Table 4 Sample set TD for predicting trigger time

[0070]

[0071]

[0072] Adaptive neuro-fuzzy inference system is used to train the prediction model and establish the optimal ANFIS model. The structure of ANFIS consists of five layers, from input to output, namely fuzzification layer, rule layer, standardization layer, defuzzification layer and output layer. The method for obtaining the final peak acceleration prediction result and the final trigger time prediction result includes the following steps:

[0073] Step 3.1, fuzzification layer: Adopt adaptive neural fuzzy inference system to train prediction model, fuzzify the training set of predicted peak acceleration, use membership function to fuzzify the input of training set, the node function is membership function, and output fuzzy set membership function value O 1,i It is expressed as:

[0074]

[0075] Where x, y, and z are the inputs of node i; A i , B i-n , C i-2n are fuzzy sets related to input x, y and z respectively; O 1,i represents the membership function value after fuzzification; is the triangle membership function, defined as:

[0076]

[0077]

[0078] where a i , bi and c i It is the relevant parameter of the triangle membership function and is the premise parameter. Parameter a i and c i Corresponding to the left and right vertices of the lower part of the triangle, parameter b i Corresponding to the vertex of the upper part of the triangle, that is, parameter a i and c i Determine the left and right "feet" of the triangle, and parameter b i Determine the "peak" of the triangle. The advantage of the triangle membership function is that it can effectively suppress the difference before and after training and better preserve the original data information.

[0079] At present, there are two major types of machine learning models: large models and small models. Generally speaking, large models require large-scale data for training to obtain the best prediction model, otherwise overfitting will occur, resulting in a decrease in prediction accuracy; while small models only require small-scale data for training. Since the historical data collected at present is relatively small, a small model is considered as a prediction model, but the stability of the small model is not enough. Therefore, in order to ensure the prediction accuracy or promotion ability of the prediction model, the present invention uses an adaptive neural fuzzy inference system to train the prediction model. Theoretical analysis and experimental tests have proved that the peak acceleration prediction method based on the adaptive neural fuzzy inference system also has good prediction results on small-scale data.

[0080] Step 3.2, fuzzy reasoning layer (rule layer): Calculate the incentive strength of all fuzzy rules. The nodes use the form of taking the smallest, algebraic product, bounded product or strong product to calculate the membership function value of the input quantity corresponding to the i-th rule and output the result:

[0081]

[0082] In the formula, w i It represents the excitation intensity of the i-th rule, that is, the arithmetic product of all input signals.

[0083] Step 3.3, standardization layer: calculate the percentage of each fuzzy rule's incentive strength to all rule incentive strengths, and its output represents the weight of the i-th rule in all fuzzy rules:

[0084]

[0085] In the formula, Represents the normalized incentive intensity of the i-th rule.

[0086] Step 3.4, defuzzification layer: Create an adaptive correlation function between the normalized excitation intensity and the result function. Each node in this layer corresponds to a fuzzy rule f i, the function is the output of the fuzzy rule when the membership is determined. It is generally a linear function, and the number of function parameters depends on the number of input quantities. The output of the i-th node of this layer is as follows:

[0087]

[0088] In the formula, m i 、p i ,q i and r i is the result parameter of the ith node, usually called the consequent parameter. Each node in this layer is connected to its own normalization node and receives the initial input x, y, and z.

[0089] Step 3.5, output layer: for each fuzzy rule f i The outputs are summed and the final ANFIS output is generated:

[0090]

[0091] Generate the initial ANFIS structure.

[0092] Step 3.6: Repeat steps 3.1 to 3.5 for parameter training on the generated initial ANFIS structure to obtain a new ANFIS structure and parameters, and continuously perform rolling optimization to obtain the optimal solution.

[0093] After obtaining the optimal solution in step 3.6, the training data set is used to adjust the model parameters through forward propagation and back propagation algorithms. The goal is to minimize the prediction error. The evaluation indicators include root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 , coefficient of determination R 2 The size of the regression equation represents the degree of fit to the observed value, and its value range is between 0 and 1. The determination coefficient R 2 The closer the value of is to 1, the more accurate the membership parameter obtained by the algorithm is, that is, the predicted value is closer to the actual value. On the contrary, if the determination coefficient R 2 The closer the value is to 0, the lower the reference value of the fitted equation, that is, the lower the reliability of the obtained membership parameter.

[0094] The evaluation index is calculated using the following set of equations:

[0095]

[0096] Where M is the total number of training data points, and are the actual baseline value and predicted value of the mth data point, Outputs the average of the data for all actual values.

[0097] Step 3.7: After obtaining the optimal ANFIS model and parameters, input the test set data for predicting peak acceleration to obtain the prerequisite parameters for the operation.

[0098] Step 3.8: Use the trained ANFIS model to obtain the final peak acceleration prediction result. Predict the 230 samples in the test set and compare them with the measured values. Figure 2 shown.

[0099] Compare Figure 2 The predicted values ​​and measured values ​​shown show that, although the fitting accuracy of individual maxima is not high due to the small number of model parameters and sample data, the overall prediction accuracy is good, and the determination coefficient reaches 0.9531. As the number of sample data collected increases and the complexity of the prediction model increases, a very high prediction accuracy can be obtained. The present invention proves that the application of the ANFIS model can effectively predict the peak acceleration and solve the problem of predicting the vibration information of the monitoring point under the conditions of missing earthquake data and strong earthquake records.

[0100] Step 3.9: Perform similar processing and calculation on the training set and test set of the predicted trigger time as in Steps 3.1 to 3.8 to obtain the final trigger time prediction result. Predict 179 samples in the test set and compare them with the measured values, such as Figure 3 shown.

[0101] Compare Figure 3 From the predicted values ​​and measured values ​​shown, it can be found that the fitting accuracy for individual maximum values ​​is not high, but the overall prediction accuracy is good, and the determination coefficient reaches 0.96882.

[0102] In step 3.8, after an earthquake of a certain magnitude occurs at a certain location, its peak acceleration at the strong earthquake monitoring points of the dams of the target power station is obtained through machine learning, and there is no need to analyze and process the strong earthquake monitoring data of the dam; in step 3.9, the time relationship between the propagation of seismic waves from the epicenter to the strong earthquake monitoring points of the dams of the target power station is obtained through machine learning, which does not rely on the actual underground structure velocity model.

[0103] Step 4: Connect the reservoir earthquake monitoring network and the dam safety monitoring system: Use the network to connect the reservoir earthquake monitoring network and the dam safety monitoring system.

[0104] The reservoir earthquake monitoring network in step 4 shares data with the national earthquake stations in the vicinity of the reservoir. After an earthquake occurs, earthquake information is obtained through the reservoir earthquake monitoring network, and the ANFIS model established in step 3 is used to quickly estimate the time when the seismic wave reaches the dam and the peak acceleration of the earthquake at the strong earthquake monitoring point of the dam, and the estimated value is pushed to the dam safety monitoring platform for comparison with the start threshold to determine whether to start the dam safety monitoring system for data collection. The reservoir earthquake monitoring network is arranged around the reservoir. Since the reservoir earthquake monitoring network shares data with the national earthquake stations in the vicinity of the reservoir, it has a wide coverage. After a moderate to strong earthquake occurs, it generally does not cause the real-time data transmission of all monitoring points of the reservoir earthquake monitoring network to be interrupted. After an earthquake occurs, the reservoir earthquake monitoring network can take advantage of its wide coverage to quickly obtain earthquake information, including the epicenter location, earthquake time, and earthquake magnitude.

Claims

1. A dam safety monitoring method driven by earthquake monitoring data, characterized in that: The following steps are involved: Step 1: Establish an original record table: collect and organize historical earthquake data and historical dam monitoring point strong earthquake data, and establish an original record table; Step 2: Determine input features and prediction targets: Determine input features as magnitude, focal depth, elevation of dam strong earthquake monitoring point, epicenter distance and azimuth, and determine prediction targets as peak acceleration and trigger time; Step 3: Obtain the prediction results of the final peak acceleration and the final trigger time: Use the adaptive neural fuzzy inference system to train the prediction model, establish the optimal ANFIS model, and obtain the final peak acceleration prediction results and the final trigger time prediction results; Step 4: Connect the reservoir earthquake monitoring network and the dam safety monitoring system: Use the network to connect the reservoir earthquake monitoring network and the dam safety monitoring system.

2. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 1, characterized in that: The historical earthquake data in step 1 include the time of occurrence, magnitude and epicenter location, wherein the epicenter location includes its longitude, latitude and focal depth.

3. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 1, characterized in that: The historical strong earthquake data of the dam monitoring point in step 1 includes the location, triggering time and peak acceleration of the dam strong earthquake monitoring point. The location of the dam strong earthquake monitoring point includes its longitude, latitude and elevation.

4. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 1, characterized in that: In step 2, the epicenter distance d and the azimuth θ are calculated according to the following formula, and the azimuth θ is calculated with the true north direction as 0 degrees: Where R is the average radius of the earth, λ1 and φ1 are the longitude and latitude of the epicenter, and λ2 and φ2 are the longitude and latitude of the dam strong earthquake monitoring point.

5. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 1, characterized in that: In the step 3, according to the prediction target, a sample set AD for predicting peak acceleration and a sample set TD for predicting trigger time are respectively established, and the two sample sets are respectively divided into a training set and a test set; The method of using an adaptive neural fuzzy inference system to train a prediction model, establish an optimal ANFIS model, and obtain a final peak acceleration prediction result and a final trigger time prediction result includes the following steps: Step 3.1, fuzzify the training set for predicting peak acceleration, and use the membership function to fuzzify the input of the training set; Step 3.2, calculate the incentive strength of all fuzzy rules. The nodes use the form of taking the smallest, algebraic product, bounded product or strong product to calculate the membership function value of the input quantity corresponding to the i-th rule and output the result: The 2,i =w i =μ Ai (x)m Bi (y)m Ci (z),i=1,2,…,n, Step 3.3, calculate the percentage of each fuzzy rule's incentive strength to all rule incentive strengths, and the output represents the weight of the i-th rule in all fuzzy rules: Step 3.4: Create an adaptive correlation function between the normalized excitation intensity and the result function. Each node in this layer corresponds to a fuzzy rule f i , the function is the output of the fuzzy rule when the membership is determined, and the number of function parameters depends on the number of input quantities. The output of the i-th node of this layer is as follows: Step 3.5: For each fuzzy rule f i The outputs are summed and the final ANFIS output is generated: Generate initial ANFIS structure; Step 3.6: Repeat steps 3.1 to 3.5 for parameter training of the generated initial ANFIS structure to obtain a new ANFIS structure and parameters, and continuously perform rolling optimization to obtain the optimal solution; Step 3.7: After obtaining the optimal ANFIS model and parameters, input the test set data of the predicted peak acceleration to obtain the premise parameters of the operation; Step 3.8, using the trained ANFIS model, obtain the final peak acceleration prediction result; Step 3.9: Perform similar processing and calculation on the training set and test set of the predicted trigger time with reference to steps 3.1 to 3.8 to obtain the final trigger time prediction result.

6. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 5, characterized in that: The sample set is divided into a training set and a test set in a ratio of 7.5:2.

5.

7. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 5, characterized in that: After obtaining the optimal solution in step 3.6, the training data set is used to adjust the model parameters through forward propagation and back propagation algorithms. The goal is to minimize the prediction error. The evaluation indicators include root mean square error RMSE, mean absolute error MAE and determination coefficient R 2 , coefficient of determination R 2 The size of the regression equation represents the degree of fit to the observed value, and its value range is between 0 and 1. The determination coefficient R 2 The closer the value of is to 1, the more accurate the membership parameter obtained by the algorithm is, that is, the predicted value is closer to the actual value. On the contrary, if the determination coefficient R 2 The closer the value is to 0, the lower the reference value of the fitted equation, that is, the lower the reliability of the obtained membership parameter.

8. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 7, characterized in that: The evaluation index is calculated using the following equation: Where M is the total number of training data points, and are the actual baseline value and predicted value of the mth data point, Outputs the average of the data for all actual values.

9. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 5, characterized in that: In step 3.8, after an earthquake of a certain magnitude occurs at a certain location, its peak acceleration at the strong earthquake monitoring points of the dams of the target power station is obtained through machine learning; in step 3.9, the time relationship between the propagation of the seismic wave from the epicenter to the strong earthquake monitoring points of the dams of the target power station is obtained through machine learning.

10. The dam safety monitoring method driven by earthquake monitoring data as claimed in claim 1, characterized in that: The reservoir seismic monitoring network in step 4 shares data with national seismic stations in the vicinity of the reservoir. After an earthquake occurs, earthquake information is obtained through the reservoir seismic monitoring network, and the ANFIS model established in step 3 is used to quickly estimate the time it takes for the seismic wave to reach the dam and the peak acceleration of the earthquake at the dam's strong earthquake monitoring point. The estimated value is then pushed to the dam safety monitoring platform for comparison with the start threshold to determine whether to start the dam safety monitoring system for data collection.

Images