An intelligent tailings dam displacement prediction and early warning method

Abstract

The present invention provides an intelligent tailings dam displacement prediction and early warning method, which belongs to the field of safety prediction and early warning technology. The method comprises the following steps: obtaining historical data of online monitoring of tailings dams and preprocessing the data; constructing a GRU main prediction model to predict the main predicted displacement; obtaining an error sequence through measured data and the main predicted displacement; decomposing the error sequence into trend term and noise term error sequences through PSO-VMD-TOPSIS, and respectively constructing GRU-based trend term and noise term error correction models; obtaining error correction values ​​and standard deviations of the trend term and noise term through MC-dropout technology; constructing a dynamic weight calculation mechanism, combining the main predicted displacement to obtain a final displacement prediction value and a confidence interval, and determining whether to issue an alarm. The method achieves high-precision real-time prediction, uncertainty quantification, and real-time early warning of tailings dam displacement, thus providing intelligent decision support for tailings dam safety status assessment and disaster early warning.

Description

Technical Field

[0001] The present invention relates to the field of public safety prediction, and in particular to an intelligent tailings dam displacement prediction and early warning method. Background Art

[0002] The safety and stability of tailings dams directly impact mine safety and production. Accidents such as landslides and dam failures can have catastrophic safety and environmental consequences. Existing tailings dams are commonly associated with safety hazards, such as long service life and increasing dam heights. Therefore, research on dam stability monitoring and safety early warning is of great engineering significance. Tailings dam displacement, a key indicator of structural safety, is a key indicator. The development of an accurate prediction model for this displacement will effectively enhance tailings dam safety early warning capabilities, provide technical support for preventing dam failures, and thus have significant engineering value.

[0003] Recently developed prediction methods based on deep learning models based on historical data can better describe the displacement characteristics of tailings dams and have attracted widespread attention in the field of tailings dam prediction. However, the evolution of tailings dam displacement exhibits significant nonlinear and time-varying characteristics. Its dynamic response is affected by multiple factors, including reservoir water level fluctuations, rainfall infiltration, and dam material creep, exhibiting a pronounced trend and a combination of randomness. This presents challenges in tailings dam prediction and early warning. On the one hand, existing methods often use independent models to predict tailings dam displacement, resulting in significant errors in their prediction performance due to a failure to fully account for randomness. On the other hand, existing methods often provide a single predicted value, failing to consider the actual displacement of the tailings dam itself, as well as the uncertainty and randomness of the predicted value. This makes it difficult to develop a reasonable warning method. Therefore, developing a more accurate tailings dam displacement error correction method, based on which prediction confidence intervals are provided, and constructing a dynamic, real-time prediction and early warning method for tailings dam displacement is crucial for the safe operation of tailings dams. Summary of the Invention

[0004] The present invention aims to provide an intelligent tailings dam displacement prediction and early warning method. The method obtains a main predicted displacement through a main prediction model; obtains an error sequence through measured data and the main predicted displacement; decomposes the error sequence into a trend term error sequence and a noise term error sequence through the PSO-VMD-TOPSIS technique, and constructs a trend term error correction model and a noise term error correction model respectively; obtains error correction values ​​and standard deviations of the trend term and noise term through the MC-dropout technique and a dynamic weight calculation mechanism; constructs a dynamic weight calculation mechanism to obtain weights of the trend term error and the noise term error, and combines the main predicted displacement to obtain a final displacement prediction value and confidence interval, and determines whether to issue an alarm. The present invention can provide reliable technical support for improving the real-time prediction accuracy of tailings dam displacement and providing effective early warning.

[0005] The present invention is achieved in that:

[0006] The technical solution to achieve the purpose of the present invention is: an intelligent tailings dam displacement prediction and early warning method, comprising the following steps:

[0007] Step 1: Obtain online monitoring historical data of the tailings dam, including displacement, rainfall, and reservoir water level data;

[0008] Step 2: Data preprocessing, including processing the displacement data using the adaptive exponentially weighted moving average (EWMA) method and independently normalizing the rainfall, reservoir water level and processed displacement data;

[0009] Step 3: Set the parameters of the prediction model;

[0010] The prediction model includes a main prediction model, a trend term error correction model and a noise term error correction model;

[0011] Step 4: Construct a main prediction model framework for tailings dam displacement based on a gated recurrent unit (GRU) neural network and train the main prediction model based on normalized data;

[0012] Step 5: Predict the main predicted displacement through the main prediction model, calculate the error between the main predicted displacement result and the monitoring result, and construct the error sequence;

[0013] Step 6: Optimize the hyperparameters of the variational mode decomposition (VMD)-topsis method (TOPSIS) model using the particle swarm optimization (PSO) algorithm. Decompose the error sequence based on the optimized VMD-TOPSIS model to obtain the trend term error sequence and the noise term error sequence.

[0014] Step 7: Construct a gated recurrent unit neural network (GRU) model framework with a random dropout layer (dropout), use the trend term error sequence to construct a trend term error correction model, and use the noise term error sequence to construct a noise term error correction model;

[0015] Step 8: Using Monte Carlo random dropout sampling (MC dropout), we can get the mean of the trend term correction based on the trend term error correction model. m 1 and standard deviation s 1. Obtain the mean value of the noise term correction based on the noise term error correction model m 2 and standard deviation s 2;

[0016] Step 9: Calculate the final weight of the trend item based on the dynamic weight calculation mechanism W q and the final weight of the noise term Wz ;

[0017] Step 10: Calculate the final displacement prediction value and confidence interval, and denormalize;

[0018] Step 11: After obtaining the next round of monitoring data, if the monitored displacement value exceeds the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be abnormal and a displacement alarm is issued; if the monitored displacement value does not exceed the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be normal; then, determine whether to continue the prediction. If so, normalize the next round of data, return to step 5 and execute the loop, otherwise terminate the calculation.

[0019] Furthermore, in step 3, the parameters of the prediction model include the window size of the main prediction model, the window size of the trend term error correction model and the noise term error correction model, the noise tolerance t , Monte Carlo sampling times T、 Mode number of variational mode decomposition K The value range and penalty factor α The value range of , the value range of sample entropy weight, the value range of energy ratio weight, and the value range of correlation weight.

[0020] Furthermore, in step six, the variational mode decomposition-approximate ideal solution sorting method model is a combination model that combines the variational mode decomposition method and the approximate ideal solution sorting method, wherein the modal number of the variational mode decomposition is K , penalty factor α , sample entropy weight of the sorting method approaching the ideal solution W S , energy ratio weight W E , relevance weight W R All of them are obtained through particle swarm optimization algorithm, and the optimization range is selected as 3≤ K ≤8,1000≤ α ≤3000, 0.1≤ W S ≤0.6,0.1≤ W E ≤0.6,0.1≤ W R ≤0.6;

[0021] The model realizes the mode optimization and reconstruction of the error sequence through the following steps:

[0022] Step S61: Decompose the error sequence into K modalities;

[0023] Step S62: Calculate the sample entropy of each modality SK , energy ratio E K And the correlation coefficient with the original error R K ， And after normalization, we get S K1 、 E K1 、 R K1 ;

[0024] Step S63: According to the optimized weight W S 、 W E 、 W R Calculate the comprehensive score of each modality C K :

[0025] C K = W S × S K1 + W E × E K1 + W R × R K1

[0026] The modes with a comprehensive score greater than the average score are selected as the optimal subset, and the optimal mode subsets are added together to obtain the trend term error sequence; the remaining modes are added together to obtain the noise term error sequence.

[0027] Furthermore, in step eight, the Monte Carlo random drop sampling step includes:

[0028] S81: Trend term error correction model T forward propagation, keeping the random dropout layer activated during each propagation, and obtaining T Trend item correction amount;

[0029] S82: Noise term error correction model T forward propagation, keeping the random dropout layer activated during each propagation, and obtaining T Noise correction amount;

[0030] S83: Calculate the T The mean of the trend corrections m 1 and standard deviation s 1;

[0031] S84: Calculate the T The mean of the noise corrections m 2 and standard deviation s 2;

[0032] in, T The value range is 50~200 times.

[0033] Furthermore, in step nine, the dynamic weight calculation mechanism is specifically as follows:

[0034] S91: Calculate the absolute mean of trend item corrections | m 1| and the absolute mean of the noise correction| m 2|;

[0035] S92: Calculate the contribution of trend items w 1=| m 1| / (| m 1|+| m 2|+1×10 -6 ), the noise term contribution w 2=1- w 1;

[0036] S93: Calculate the confidence level of trend items c 1=1 / ( s 1+1×10 -6 ) and the noise term confidence c 2=1 / ( s 2+10 -6 ),in s 1 and s 2 are the standard deviations of the trend term and noise term respectively;

[0037] S94: Calculate the comprehensive weight: the final weight of the trend item W q = w 1× c 1 / ( w 1× c 1+ w 2× c 2), the final weight of the noise term W z =1- W q .

[0038] Furthermore, in step 10, the final displacement prediction value and confidence interval, the final displacement prediction value refers to the mean of the final displacement prediction value m 总 , the confidence interval includes the upper limit and the lower limit of the confidence interval, and the calculation method is:

[0039] S101: Final displacement prediction value mean m 总 = m 1× W q + m 2× W z ;

[0040] S102: Standard deviation of final displacement prediction value s 总 =( s 1 2 × W q 2 + s 2 2 × W z 2 ) 1 / 2 ;

[0041] S103: Upper limit of confidence interval = main prediction displacement + m 总 +1.96× s 总 ;

[0042] S104: Confidence interval lower limit = main prediction displacement + m 总 -1.96× s 总 .

[0043] The beneficial effects of the present invention are as follows: the present invention provides an intelligent real-time prediction and early warning method for tailings dam displacement. Among them, the VMD-TOPSIS model optimized by PSO can adapt to the hyperparameters of VMD and TOPSIS, and more accurately separate the trend term and the noise term; constructing a GRU model for the trend term and the noise term respectively can make more accurate predictions of the trend term and the noise term; after combining the Monte Carlo sampling and the dropout mechanism, not only the displacement prediction value can be obtained, but also the probability distribution of the displacement prediction value can be obtained; the dynamic weight calculation mechanism can adaptively allocate the importance of the trend term and the noise term, so as to more accurately obtain the mean and confidence interval of the final displacement prediction value. In summary, the method of the present invention can not only make the model have better prediction accuracy, but also obtain the prediction confidence interval, providing more reliable technical support for the real-time prediction and early warning of tailings dam displacement. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0045] Figure 1 This is a flow chart of an intelligent tailings dam displacement prediction and early warning method provided by an embodiment of the present invention;

[0046] Figure 2 Schematic diagram of the final displacement prediction result (including confidence interval) of the tailings dam provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in combination with the implementation cases and drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention for which protection is sought, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0048] The method of the present invention is described below by taking a specific case of real-time prediction of landslide displacement as an example.

[0049] like Figure 1 , an intelligent tailings dam displacement prediction and early warning method, comprising the following steps:

[0050] Step 1: Obtain online monitoring historical data of the tailings dam, including displacement, rainfall, and reservoir water level data;

[0051] Step 2: Data preprocessing, including processing the displacement data using the adaptive exponentially weighted moving average (EWMA) method and independently normalizing the rainfall, reservoir water level and processed displacement data;

[0052] Step 3: Set the parameters of the prediction model;

[0053] The prediction model includes a main prediction model, a trend term error correction model and a noise term error correction model;

[0054] Step 4: Construct a main prediction model framework for tailings dam displacement based on a gated recurrent unit (GRU) neural network and train the main prediction model based on normalized data;

[0055] Step 5: Predict the main predicted displacement through the main prediction model, calculate the error between the main predicted displacement result and the monitoring result, and construct the error sequence;

[0056] Step 6: Optimize the hyperparameters of the variational mode decomposition (VMD)-topsis method (TOPSIS) model using the particle swarm optimization (PSO) algorithm. Decompose the error sequence based on the optimized VMD-TOPSIS model to obtain the trend term error sequence and the noise term error sequence.

[0057] Step 7: Construct a gated recurrent unit neural network (GRU) model framework with a random dropout layer (dropout), use the trend term error sequence to construct a trend term error correction model, and use the noise term error sequence to construct a noise term error correction model;

[0058] Step 8: Using Monte Carlo random dropout sampling (MC dropout), we can get the mean of the trend term correction based on the trend term error correction model. m 1 and standard deviation s 1. Obtain the mean value of the noise term correction based on the noise term error correction model m 2 and standard deviation s 2;

[0059] Step 9: Calculate the final weight of the trend item based on the dynamic weight calculation mechanism W q and the final weight of the noise term W z ;

[0060] Step 10: Calculate the final displacement prediction value and confidence interval, and denormalize;

[0061] Step 11: After obtaining the next round of monitoring data, if the monitored displacement value exceeds the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be abnormal and a displacement alarm is issued; if the monitored displacement value does not exceed the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be normal; then, determine whether to continue the prediction. If so, normalize the next round of data, return to step 5 and execute the loop, otherwise terminate the calculation.

[0062] Furthermore, in step 3, the parameters of the prediction model include the window size of the main prediction model, the window size of the trend term error correction model and the noise term error correction model, the error correction start cycle, T W , noise margin t, Monte Carlo sampling times T、 Mode number of variational mode decomposition K The value range and penalty factor α The value range of , the value range of sample entropy weight, the value range of energy ratio weight, and the value range of correlation weight.

[0063] Furthermore, in step six, the variational mode decomposition-approximate ideal solution sorting method model is a combination model that combines the variational mode decomposition method and the approximate ideal solution sorting method, wherein the modal number of the variational mode decomposition is K , penalty factor α , sample entropy weight of the sorting method approaching the ideal solution W S , energy ratio weight W E , relevance weight W R All of them are obtained through particle swarm optimization algorithm, and the optimization range is selected as 3≤ K ≤8,1000≤ α ≤3000, 0.1≤ W S ≤0.6,0.1≤ W E ≤0.6,0.1≤ W R ≤0.6;

[0064] The model realizes the mode optimization and reconstruction of the error sequence through the following steps:

[0065] Step S61: Decompose the error sequence into K modalities;

[0066] Step S62: Calculate the sample entropy of each modality S K , energy ratio E K And the correlation coefficient with the original error R K ， And after normalization, we get S K1 、 E K1 、 R K1 ;

[0067] Step S63: According to the optimized weight W S 、 W E 、 W R Calculate the comprehensive score of each modalityC K :

[0068] C K = W S × S K1 + W E × E K1 + W R × R K1

[0069] The modes with a comprehensive score greater than the average score are selected as the optimal subset, and the optimal mode subsets are added together to obtain the trend term error sequence; the remaining modes are added together to obtain the noise term error sequence.

[0070] Furthermore, in step eight, the Monte Carlo random drop sampling step includes:

[0071] S81: Perform T forward propagations on the trend term error correction model, keeping the random dropout layer activated during each propagation, and obtain T trend term corrections;

[0072] S82: Perform T forward propagations on the noise error correction model, keeping the random dropout layer activated during each propagation, and obtain T noise correction values;

[0073] S83: Calculate the mean μ1 and standard deviation σ1 of the T trend term corrections;

[0074] S84: Calculate the T The mean of the noise corrections m 2 and standard deviation s 2;

[0075] in, T The value range is 50~200 times.

[0076] Furthermore, in step nine, the dynamic weight calculation mechanism is specifically as follows:

[0077] S91: Calculate the absolute mean of trend item corrections | m 1| and the absolute mean of the noise correction| m 2|;

[0078] S92: Calculate the contribution of trend items w 1=| m 1| / (| m 1|+| m 2|+1×10 -6), the noise term contribution w 2=1- w 1;

[0079] S93: Calculate the confidence level of trend items c 1=1 / ( s 1+1×10 -6 ) and the noise term confidence c 2=1 / ( s 2+10 -6 ),in s 1 and s 2 are the standard deviations of the trend term and noise term respectively;

[0080] S94: Calculate the comprehensive weight: the final weight of the trend item W q = w 1× c 1 / ( w 1× c 1+ w 2× c 2), the final weight of the noise term W z =1- W q .

[0081] Furthermore, in step 10, the final displacement prediction value and confidence interval, the final displacement prediction value refers to the mean of the final displacement prediction value m 总 , the confidence interval includes the upper limit and the lower limit of the confidence interval, and the calculation method is:

[0082] S101: Final displacement prediction value mean m 总 = m 1× W q + m 2× W z ;

[0083] S102: Standard deviation of final displacement prediction value s 总 =( s 1 2 × W q 2 + s 2 2 × W z 2 ) 1 / 2 ;

[0084] S103: Upper limit of confidence interval = main prediction displacement + m 总 +1.96× s 总 ;

[0085] S104: Confidence interval lower limit = main prediction displacement + m 总 -1.96× s 总 .

[0086] The experimental data comes from the online monitoring data of a tailings dam. A total of 2500 sets of monitoring sample data were obtained, of which the first 2140 sets of data were used to build the initial model, and the last 360 sets of data were used for prediction.

[0087] The specific implementation process is as follows:

[0088] like Figure 1 , an intelligent tailings dam displacement prediction and early warning method, comprising the following steps:

[0089] Step 1: Obtain online monitoring historical data of the tailings dam, including displacement, rainfall, and reservoir water level data;

[0090] Step 2: Data preprocessing, including processing the displacement data using the adaptive exponentially weighted moving average (EWMA) method and independently normalizing the rainfall, reservoir water level and processed displacement data;

[0091] Step 3: Set the parameters of the prediction model;

[0092] The prediction model includes a main prediction model, a trend term error correction model and a noise term error correction model;

[0093] The window size of the main prediction model, the trend error correction model and the noise error correction model are all 120, and the noise tolerance is t =0, Monte Carlo sampling times T =50 、 Mode number of variational mode decomposition K The value range is [3,8], the penalty factor α The value range of is [1000,3000], the value range of sample entropy weight, the value range of energy ratio weight, and the value range of correlation weight are all [0.1, 0.6].

[0094] Step 4: Construct a main prediction model framework for tailings dam displacement based on a gated recurrent unit (GRU) neural network and train the main prediction model based on normalized data;

[0095] Step 5: Predict the main predicted displacement through the main prediction model, calculate the error between the main predicted displacement result and the monitoring result, and construct the error sequence;

[0096] Step 6: Optimize the hyperparameters of the variational mode decomposition (VMD)-topsis method (TOPSIS) model using the particle swarm optimization (PSO) algorithm. Decompose the error sequence based on the optimized VMD-TOPSIS model to obtain the trend term error sequence and the noise term error sequence.

[0097] Step 7: Construct a gated recurrent unit neural network (GRU) model framework with a random dropout layer (dropout), use the trend term error sequence to construct a trend term error correction model, and use the noise term error sequence to construct a noise term error correction model;

[0098] Step 8: Using Monte Carlo random dropout sampling (MC dropout), we can get the mean of the trend term correction based on the trend term error correction model. m 1 and standard deviation s 1. Obtain the mean value of the noise term correction based on the noise term error correction model m 2 and standard deviation s 2;

[0099] Step 9: Calculate the final weight of the trend item based on the dynamic weight calculation mechanism W q and the final weight of the noise term W z ;

[0100] Step 10: Calculate the final displacement prediction value and confidence interval, and denormalize;

[0101] Step 11: After obtaining the next round of monitoring data, if the monitored displacement value exceeds the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be abnormal and a displacement alarm is issued; if the monitored displacement value does not exceed the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be normal; then, determine whether to continue the prediction. If so, normalize the next round of data, return to step 5 and execute the loop, otherwise terminate the calculation.

[0102] Figure 2 The comparison between the predicted results and the actual values ​​is shown. It can be seen that the present invention can not only accurately predict the displacement of the tailings dam, but also all the measured values ​​are within the predicted confidence interval, indicating that the prediction effect is good and no alarm occurs. At the same time, in order to further verify the advantages of the model, the following models are calculated and discussed respectively:

[0103] Model I: Only the GRU-MC-Dropout main model is used without error correction model;

[0104] Model II: GRU-single-error MC-Dropout correction model (all errors are considered as trend terms);

[0105] Model III: GRU-VMD double error correction model;

[0106] Model IV: GRU-PSO-VMD dual error correction model;

[0107] Model V: This paper’s model;

[0108] First, the performance of uncertainty analysis of each model is evaluated using the prediction interval coverage probability (PICP), coverage width criterion (CWC), center of interval deviation (MPICD), and average interval score (AIS), as shown in Table 1:

[0109]

[0110] As shown in Table 1, although all models achieve full coverage of PICP, the proposed method (Model V) achieves the best overall performance with CWC = 0.181, MPICD = 0.209, and AIS = 1.611. This demonstrates that the proposed method significantly improves prediction accuracy and interval positioning capabilities while ensuring reliable coverage. It can provide accurate and reliable prediction intervals for tailings dam safety warnings, effectively supporting engineering risk decision-making.

[0111] Then, the mean absolute error (MAE), absolute percentage error (MAPE), root mean square error (RMSE) and goodness of fit R were used to calculate the final predicted displacement mean. 2 A total of four indicators are used to quantitatively evaluate each model. The evaluation results of the prediction mean of each model are shown in Table 2:

[0112]

[0113] As shown in Table 2, the method of the present invention has the following advantages: MAE, MAPE, RMSE and goodness of fit R 2 The four indicators are all the best values ​​among all the methods, which fully demonstrates the advantages of the model in predicting tailings dam displacement. It shows that the results of the present invention can effectively improve the accuracy of tailings dam displacement prediction and provide a basis for achieving tailings dam safety management and preventing major tailings dam failure accidents.

[0114] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. An intelligent tailings dam displacement prediction and early warning method, characterized in that: The following steps are involved: Step 1: Obtain online monitoring historical data of the tailings dam, including displacement, rainfall, and reservoir water level data; Step 2: Data preprocessing, including processing the displacement data using the adaptive exponentially weighted moving average method, and independently normalizing the rainfall, reservoir water level, and processed displacement data; Step 3: Set the parameters of the prediction model; The prediction model includes a main prediction model, a trend term error correction model and a noise term error correction model; Step 4: Construct a main prediction model framework for tailings dam displacement based on a gated recurrent unit neural network and train the main prediction model based on the normalized data; Step 5: Predict the main predicted displacement through the main prediction model, calculate the error between the main predicted displacement result and the monitoring result, and construct the error sequence; Step 6: Optimize the hyperparameters of the VMD-APS model using the particle swarm optimization algorithm. Decompose the error sequence based on the optimized VMD-APS model to obtain the trend error sequence and the noise error sequence. Step 7: Construct a gated recurrent unit neural network model framework with a random dropout layer, use the trend term error sequence to construct a trend term error correction model, and use the noise term error sequence to construct a noise term error correction model; Step 8: Obtain the mean of the trend term correction based on the trend term error correction model through Monte Carlo random dropout sampling μ 1 and standard deviation σ 1. Obtain the mean value of the noise term correction based on the noise term error correction model μ 2 and standard deviation σ 2; Step 9: Calculate the final weight of the trend item based on the dynamic weight calculation mechanism W q and the final weight of the noise term W z ; Step 10: Calculate the final displacement prediction value and confidence interval, and denormalize; Step 11: After obtaining the next round of monitoring data, if the monitored displacement value exceeds the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be abnormal and a displacement alarm is issued; if the monitored displacement value does not exceed the upper and lower limits of the 95% confidence interval of the predicted value, the displacement trend is determined to be normal; Then, determine whether to continue the prediction. If so, normalize the next round of data, return to step 5 and execute the loop. Otherwise, terminate the calculation.

2. The intelligent tailings dam displacement prediction and early warning method according to claim 1, characterized in that: In step 3, the parameters of the prediction model include the window size of the main prediction model, the window size of the trend term error correction model and the noise term error correction model, the noise tolerance τ , Monte Carlo sampling times T、 Mode number of variational mode decomposition K The value range and penalty factor α The value range of , the value range of sample entropy weight, the value range of energy ratio weight, and the value range of correlation weight.

3. The intelligent tailings dam displacement prediction and early warning method according to claim 1, characterized in that: In step 6, the variational mode decomposition-approximate ideal solution sorting method model is a combination model that combines the variational mode decomposition method and the approximate ideal solution sorting method, wherein the mode number of the variational mode decomposition is K , penalty factor α , sample entropy weight of the sorting method approaching the ideal solution W S , energy ratio weight W E , relevance weight W R All of them are obtained through particle swarm optimization algorithm, and the optimization range is selected as 3≤ K ≤8,1000≤ α ≤3000, 0.1≤ W S ≤0.6,0.1≤ W E ≤0.6,0.1≤ W R ≤0.6; The model realizes the mode optimization and reconstruction of the error sequence through the following steps: Step S61: Decompose the error sequence into K modalities; Step S62: Calculate the sample entropy of each modality S K , energy ratio E K And the correlation coefficient with the original error R K ， After normalization, we get S K1 、 E K1 、 R K1 ; Step S63: According to the optimized weight W S 、 W E 、 W R Calculate the comprehensive score of each modality C K : C K = W S × S K1 + W E × E K1 + W R × R K1 Select the modes with comprehensive scores greater than the average score as the optimal subset, add the optimal mode subsets together to obtain the trend term error sequence; The remaining modes are added together to obtain the noise term error sequence.

4. The intelligent tailings dam displacement prediction and early warning method according to claim 1, characterized in that: In step eight, the Monte Carlo random dropout sampling includes the following steps: S81: Trend term error correction model T forward propagation, keeping the random dropout layer activated during each propagation, and obtaining T Trend item correction amount; S82: Noise term error correction model T forward propagation, keeping the random dropout layer activated during each propagation, and obtaining T Noise correction amount; S83: Calculate the T The mean of the trend corrections μ 1 and standard deviation σ 1; S84: Calculate the T The mean of the noise corrections μ 2 and standard deviation σ 2; in, T The value range is 50~200 times.

5. The intelligent tailings dam displacement prediction and early warning method according to claim 1, characterized in that: In step nine, the dynamic weight calculation mechanism is specifically as follows: S91: Calculate the absolute mean of trend item corrections | μ 1| and the absolute mean of the noise correction| μ 2|; S92: Calculate the contribution of trend items w 1=| μ 1| / (| μ 1|+| μ 2|+1×10 -6 ), the noise term contribution w 2=1- w 1; S93: Calculate the confidence level of trend items c 1=1 / ( σ 1+1×10 -6 ) and the noise term confidence c 2=1 / ( σ 2+10 -6 ),in σ 1 and σ 2 are the standard deviations of the trend term and noise term respectively; S94: Calculate the comprehensive weight: the final weight of the trend item W q = w 1× c 1 / ( w 1× c 1+ w 2× c 2), the final weight of the noise term W z =1- W q .

6. The intelligent tailings dam displacement prediction and early warning method according to claim 1, characterized in that: In step 10, the final displacement prediction value and confidence interval, the final displacement prediction value refers to the mean of the final displacement prediction value μ 总 , the confidence interval includes the upper limit and the lower limit of the confidence interval, and the calculation method is: S101: Final displacement prediction value mean μ 总 = μ 1× W q + μ 2× W z ; S102: Standard deviation of final displacement prediction value σ 总 =( σ 1 2 × W q 2 + σ 2 2 × W z 2 ) 1 / 2 ; S103: Upper limit of confidence interval = main prediction displacement + μ 总 +1.96× σ 总 ; S104: Confidence interval lower limit = main prediction displacement + μ 总 -1.96× σ 总 .

Images