Reservoir bank slope short-term deformation trend IKHM-EFS-TimeMixer prediction method considering prediction interval hydro meteorological change

Abstract

The invention discloses a reservoir bank slope short-term deformation trend IKHM-EFS-TimeMixer prediction method considering a prediction interval hydro meteorological change. The method comprises the following steps: 1) collecting a reservoir bank slope historical deformation sequence and carrying out clustering distinguishing; 2) an IKHM algorithm is initialized, and IKHM clustering is carried out on the reservoir bank slope variable factor sequences with similar characteristics; 3) constructing a clustering feature analysis matrix of the variable factor sequence; and 4) carrying out EWT decomposition on the displacement sequences in the clustering feature analysis matrix, removing noise sequences determined in each decomposed sequence, and reconstructing each group of displacement sequences. And 5) based on the updated clustering feature analysis matrix and considering the hydro meteorological change of the prediction interval, applying a TimeMixer model to predict the deformation trend of the reservoir bank slope in any step length in the short-term future. According to the method, the clustering result with obvious global characteristics is obtained through optimization clustering, the superposed noise item and the feature item after deformation sequence clustering are distinguished, and the hydro-meteorological change of a certain prediction interval is considered, so that the method has better performance in the aspect of predicting the deformation trend of the bank slope in a short period.

Description

Technical Field

[0001] The invention relates to the technical field of reservoir bank slope deformation prediction, in particular to an IKHM-EFS-TimeMixer prediction method for reservoir bank slope short-term deformation trend considering hydrological and meteorological changes in a prediction interval. Background Art

[0002] Landslides caused by slope instability in reservoir areas are one of the most common geological disasters. The deformation of the slope before the landslide usually occurs in a slow process, and some of its damage characteristics, such as deformation rate and deformation amplitude, can be mastered through existing monitoring technology. Therefore, slope displacement prediction has become an economical and effective way to reduce risks.

[0003] The deformation and displacement prediction of reservoir bank slopes is of great significance in the early warning of reservoir bank landslide disasters. Deep learning, as a nonlinear prediction method, can process historical deformation sequences and predict the deformation trend of future slopes by learning the characteristics of deformation monitoring data under the coupling of multiple mutable factors based on historical displacement monitoring results. However, long-term monitoring inevitably generates massive monitoring data, including deformation monitoring results under the action of various mutable factors, as well as abnormal values ​​under the conditions of aging and failure of monitoring instruments. How to effectively process long-term deformation sequences in the context of massive monitoring data and predict the short-term deformation trend of reservoir bank slopes using deep learning methods is the problem to be solved by the present invention. At present, there is no research in the publicly published literature and patents that applies the IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slopes that takes into account hydrological and meteorological changes in the prediction interval to the field of reservoir bank slope deformation prediction technology. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an IKHM-EFS-TimeMixer prediction method for the short-term deformation trend of the reservoir bank slope taking into account the hydrological and meteorological changes in the prediction interval, effectively deal with the long-term deformation sequence of the reservoir bank slope, and use a deep learning method to predict the short-term deformation trend of the reservoir bank slope in different prediction intervals. A IKHM-EFS-TimeMixer prediction method for the short-term deformation trend of the reservoir bank slope taking into account the hydrological and meteorological changes in the prediction interval is proposed.

[0005] To achieve the above object, the present invention adopts the following technical solution: an IKHM-EFS-TimeMixer prediction method for the short-term deformation trend of the reservoir bank slope considering the hydrological and meteorological changes in the prediction interval, comprising the following steps:

[0006] Step 1: Collect the historical deformation sequence of the reservoir bank slope with a time dimension of T and m reservoir bank slope change factors, cluster and distinguish the historical deformation sequence, set the upper limit of clustering uln_cluster equal to the number of reservoir bank slope change factors m, and calculate the prediction accuracy comprehensive evaluation function MF (MAE, MSE, R 2 )Determine the best number of clusters best_num_cluster;

[0007] Step 2: Initialize the IKHM algorithm and cluster the reservoir bank slope mutation factor sequences with similar characteristics, set the number of cluster centers n equal to the optimal cluster number best_num_cluster, and determine the cluster centers;

[0008] Step 3: Based on the cluster centers, construct a clustering feature analysis matrix of mutagenic factor sequences with n dimensions (T×(m+1));

[0009] Step 4: De-noise the displacement sequences of the clustering feature analysis matrices of the n variable factor sequences. First, perform EWT decomposition on them. Set the number of sequence decompositions equal to the number of reservoir bank slope variable factors m. The displacement sequence of each group of feature analysis matrices is decomposed to obtain a displacement sequence decomposition matrix, which contains m decomposition sequences mra. The fuzzy entropy value and SMSE value of each decomposition sequence mra are calculated. The mra decomposition sequence with a higher fuzzy entropy value and a lower SMSE value in each decomposition sequence mra is judged as a noise sequence and removed to reconstruct the displacement sequence of each group of feature analysis matrices.

[0010] Step 5: Based on the displacement sequence of the clustering feature analysis matrix of each group of mutation factor sequences and considering the hydrological and meteorological changes in the prediction interval, a multi-input deformation prediction data set is established, and a multi-step prediction model for short-term deformation of the reservoir bank slope is constructed. The TimeMixer model is used to predict the deformation amplitude of the reservoir bank slope in the short term in any step length in the future.

[0011] In a preferred embodiment, the acquisition time dimension in step 1 is T, and the historical deformation sequence data[:,:] of the bank slope with m bank slope mutation factors is (T×(m+1)) First, set the slice ratio and slice range to (0, 1), and obtain the historical deformation sequence training set data_train[:,:] and the historical deformation sequence test set data_test[:,:]:

[0012] data_train[: , :]=data[: , :]*slice (1)

[0013] data_test[: , :]=data[data[: , :]*slice.shape[0]: , :] (2)

[0014] Then cluster the historical deformation sequence and set the upper limit of clustering uln_cluster equal to the number of bank slope mutation factors m:

[0015] uln_cluster=data_train[: , :].shape[1]-1 (3)

[0016] The performance results of the historical deformation sequence under different cluster numbers (i, 1 to the upper limit of cluster number uln_cluster) are analyzed, including the mean absolute error MAE and its maximum value MAE.max, the mean square error MSE and its maximum value MSE.max, the absolute coefficient R 2 and its maximum value R 2 .max, to calculate the prediction accuracy comprehensive evaluation function MF (MAE, MSE, R 2 ) Determine the best number of clusters best_num_cluster, and its evaluation function is specifically expressed as follows:

[0017]

[0018] In a preferred embodiment, step 2 initializes the IKHM algorithm and clusters the reservoir bank slope mutation factor sequences with similar characteristics, sets the number of cluster centers n equal to the optimal cluster number best_num_cluster, and determines n cluster centers cluster_center[1,:m+1], including the following steps:

[0019] Step 21: Set the maximum number of iterations iter max , initialize IKHM algorithm parameters, initialize loop update instruction G 1 =0;

[0020] Step 22: Set the loop update instruction G 2 =G 3 =0;

[0021] Step 23: Perform IKHM clustering:

[0022] Step 24: G 1 =G 1 +1.If G 1 <itermax, then return to step 24;

[0023] Step 25: Get n cluster centers cluster_center[1,:m+1].

[0024] In a preferred embodiment, the step 23 specifically includes:

[0025] Step 231: Use the Sobol-AOA algorithm to update the cluster center, initialize j=best_num_cluster initial cluster centers with the Sobol sequence as the AOA initial population, calculate the fitness values ​​of all initial cluster centers in the population and sort them, select to enter the exploration or development stage through the mathematical optimizer acceleration function, and expand the global and local search for the optimal cluster center;

[0026] Step 232: G 2 =G 2 +1; If G 2 <3; then return to step 231;

[0027] Step 233: The current position of the particle is used as the initial cluster center and updated using the KHM algorithm, characterized in that all cluster centers obtained in steps 231-232 are used to calculate the affiliation relationship and corresponding weight between each data point and the initial cluster center, and each cluster center is updated by its affiliation, weight, and data point, and the update is repeated until the KHM iteration number set by the IKHM algorithm parameters is reached or the cluster center does not change;

[0028] Step 234: G 3 =G 3 +1; If G 3 <7, then return to step 233.

[0029] In a preferred embodiment, step 3 constructs n mutagenic factor sequence clustering feature analysis matrices i_cluster_mat[:T / n,:m+1] with dimensions ((T / n)×(m+1)) according to n cluster centers cluster_center[1,:m+1], where i={0,1,2,...,n}.

[0030] 6. The IKHM-EFS-TimeMixer prediction method for the short-term deformation trend of the reservoir bank slope considering the hydrological and meteorological changes in the prediction interval according to claim 1 is characterized in that the displacement sequences of the clustering feature analysis matrix of the n mutable factor sequences described in step 4 are denoised respectively, and the i-th group of displacement sequences is expressed as:

[0031] i_displacement[:T / n,1]=i_cluster_mat[:T / n,m+1],i={0,1,2,...,n}(5)

[0032] Perform EWT decomposition on the displacement sequence i_displacement[:T / n,1] of each set of characteristic analysis matrices, set the number of sequence decompositions num_mra equal to the number of reservoir bank slope variable factors m, and obtain a displacement sequence decomposition matrix mra_mat[:T / n,:m] by decomposing each set of displacement sequences:

[0033] num_mra=mra_mat[ : T / n , : m].shape[1] (6)

[0034] Calculate the fuzzy entropy FuzzyEn value of num_mra mra decomposition sequences of each group of displacement sequences:

[0035] FuzzyEn_mra[j]=EH.FuzzEn(mra_mat[:, j]), j={0,1,2,...,m}(7)

[0036] Calculate the SMSE values ​​of num_mra mra decomposition sequences for each group of displacement sequences:

[0037] SMSE_mra[j]=mra_mat[i,j]**2+SMSE_mra[0,j],

[0038] i={0,1,2,...,mra_mat[:].shape [0]},j={0,1,2,...,m} (8)

[0039] The noise sequences of the mra decomposition sequences with higher fuzzy entropy values ​​and lower SMSE values ​​in each mra decomposition sequence are eliminated, and the displacement sequence of each group of feature analysis matrices is reconstructed.

[0040] In a preferred embodiment, step 5 is based on each group of variable factor sequence clustering feature analysis matrix and considers the changes in hydrological and meteorological characteristics in the prediction interval to establish a measured water level Measured rainfall Measured temperature Measured aging (δ θ ) factors and displacement sequence δ 位移序列[t,t+a] , and the water level forecast obtained from hydrological and meteorological forecasts (δ HP ), forecast rainfall (δ RP ), predicted temperature (δ TP )'s multi-input deformation prediction dataset data_predict[:T / n,:m+3a+1+a], including the following steps:

[0041] Step 51: Set the step length of short-term prediction to a days, define the prediction interval [t, t+a], collect hydrological and meteorological forecast data within the prediction interval, and obtain the forecast water level at each t+i moment Precipitation forecast Forecast Temperature i = {1, 2, ..., a};

[0042] Step 52: Build including measured water level Measured rainfall Measured temperature Aging (δ θ ) factors and the displacement sequence (δ 位移序列[t,t+a] ), and the water level forecast obtained from hydro-meteorological forecasts Precipitation forecast Forecast Temperature Factored multi-input deformation prediction dataset data_predict[:T / n,:m+3a+1+a]:

[0043]

[0044]

[0045] Step 53: n multi-input deformation prediction data sets data_predict[:T / n,:m+3a+1+a] are constructed from the above measured and predicted water levels, rainfall, temperature, time factor and displacement sequences, and the model output mode is set to multi-input multi-output. The TimeMixer model is used for prediction and the deformation prediction matrix Y_predict[:T,1+a] of the reservoir bank slope in the short term of a step length days is output in the form of multiple outputs;

[0046] Step 54: Take the average value of each row of the deformation prediction matrix Y_predict[:T,1+a] of the reservoir bank slope for a short-term step length of a days, Average.Y_predict[i,1+a], i={1,2,...,T} as the short-term deformation prediction value of the reservoir bank slope and output it.

[0047] Compared with the prior art, the present invention has the following beneficial effects: the present invention obtains clustering results with obvious global characteristics by optimizing clustering, distinguishes the superimposed noise terms and characteristic terms after clustering the deformation sequence, and considers the hydrological and meteorological changes in a certain prediction interval, and has better performance in predicting the deformation trend of the reservoir bank slope in the short term. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A method flow chart of a preferred embodiment of the present invention;

[0049] Figure 2 The historical measured values ​​of a certain reservoir bank slope deformation monitoring point in the preferred embodiment of the present invention;

[0050] Figure 3This is the grouping clustering effect when the number of cluster centers is 2 in the preferred embodiment of the present invention;

[0051] Figure 4 It is an EWT decomposition sequence diagram of a preferred embodiment of the present invention;

[0052] Figure 5 It is a schematic diagram of the deformation sequence before and after removing the noise sequence according to the fuzzy entropy and SMSE value in the preferred embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0054] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0055] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0056] The technical problem to be solved by the invention is to effectively process the long-term deformation sequence of the bank slope under the background of massive monitoring data, and to use the deep learning method to predict the short-term deformation trend of the bank slope in different prediction intervals. A IKHM-EFS-TimeMixer prediction method for the short-term deformation trend of the bank slope that takes into account the hydrological and meteorological changes in the prediction interval is proposed. First, the historical deformation sequence of the bank slope is collected and clustered, the upper limit of the clustering number is set and the optimal number of clusters is determined; secondly, the IKHM algorithm is initialized and the bank slope mutation factor sequence with similar characteristics is clustered by IKHM to determine the number of clusters and the cluster center; then, according to the cluster center, the cluster feature analysis matrix of the mutation factor sequence is constructed; then, the displacement sequence in the cluster feature analysis matrix is ​​decomposed by EWT, the noise sequence determined in each decomposition sequence is eliminated, and each group of displacement sequences is reconstructed. Finally, based on the updated cluster feature analysis matrix and considering the hydrological and meteorological changes in the prediction interval, the TimeMixer model is used to predict the deformation trend of the bank slope in the short term in the future at any step length. As shown in the attached figure Figure 1-5 As shown. It mainly includes the following 5 steps:

[0057] Step 1: Collect the historical deformation sequence of the reservoir bank slope with a time dimension of T and m reservoir bank slope deformation factors, as shown in the attached figure. Figure 2As shown in the figure, the historical deformation sequence is clustered and distinguished, the upper limit of clustering number uln_cluster is set equal to the number of bank slope change factors m, and the prediction accuracy comprehensive evaluation function MF (MAE, MSE, R 2 )Determine the best number of clusters best_num_cluster;

[0058] Step 2: Initialize the IKHM algorithm and cluster the reservoir bank slope mutation factor sequences with similar characteristics, set the number of cluster centers n equal to the optimal cluster number best_num_cluster, and determine the cluster centers;

[0059] Step 3: Based on the cluster centers, construct a clustering feature analysis matrix of mutagenic factor sequences with n dimensions (T×(m+1));

[0060] Step 4: De-noise the displacement sequences of the clustering feature analysis matrices of the n variable factor sequences. First, perform EWT decomposition on them. Set the number of sequence decompositions equal to the number of reservoir bank slope variable factors m. The displacement sequence of each group of feature analysis matrices is decomposed to obtain a displacement sequence decomposition matrix, which contains m decomposition sequences mra. The fuzzy entropy value and SMSE value of each decomposition sequence mra are calculated. The mra decomposition sequence with a higher fuzzy entropy value and a lower SMSE value in each decomposition sequence mra is judged as a noise sequence and removed to reconstruct the displacement sequence of each group of feature analysis matrices.

[0061] Step 5: Based on the displacement sequence of the clustering feature analysis matrix of each group of mutation factor sequences and considering the hydrological and meteorological changes in the prediction interval, a multi-input deformation prediction data set is established, and a multi-step prediction model for short-term deformation of the reservoir bank slope is constructed. The TimeMixer model is used to predict the deformation amplitude of the reservoir bank slope in the short term in any step length in the future.

[0062] Take a bank slope deformation monitoring point as an example. The historical monitoring values ​​of this point are as follows: Figure 2 As shown, now follow the method flow ( Figure 1 ) for a detailed introduction and description:

[0063] Step 1: Collect the historical deformation sequence data[:8201,:7] of the reservoir bank slope with a time dimension of 8201 and 4 factors causing the reservoir bank slope change (including 3 items of 1st, 2nd and 3rd power measured water level factors, 1 item of measured rainfall factor, 1 item of measured temperature factor, 2 items of time effect factors, a total of 7 items), cluster and distinguish the historical deformation sequence, set the upper limit of clustering uln_cluster equal to the number of factors causing the reservoir bank slope change m = 4, and calculate the prediction accuracy comprehensive evaluation function MF (MAE, MSE, R 2 ) Determine the best number of clusters best_num_cluster, including the following steps:

[0064] First, the historical deformation sequence of the reservoir bank slope is sliced, the slice ratio slice range is set to (0, 1), and the slice ratio is 0.8. The historical deformation sequence training set data_train[:,:7] and the historical deformation sequence test set data_test[:,:7] are obtained:

[0065] data_train[:,:7]=data[:8201,:7]*0.8 (1)

[0066] data_test[:,:7]=data[data[:8201,:7]*0.8.shape[0]:,:7] (2)

[0067] Then cluster the historical deformation sequence and set the upper limit of clustering uln_cluster equal to the number of bank slope mutation factors m:

[0068] uln_cluster=m (3)

[0069] The calculated upper limit of clustering number uln_cluster is equal to the number of reservoir bank slope variable factors m=4.

[0070] The performance results of the historical deformation sequence under different cluster numbers (i, 1 to the upper limit of cluster number uln_cluster = 4) are analyzed, including the mean absolute error MAE and its maximum value MAE.max, the mean square error MSE and its maximum value MSE.max, the absolute coefficient R2 and its maximum value R 2 .max, to calculate the prediction accuracy comprehensive evaluation function MF (MAE, MSE, R 2 ) Determine the best cluster number best_num_cluster=2, and its evaluation function is specifically expressed as follows:

[0071]

[0072]

[0073] Step 2 initializes the IKHM algorithm and clusters the reservoir bank slope mutation factor sequences with similar characteristics, sets the number of cluster centers n equal to the optimal number of clusters best_num_cluster = 2, and determines two cluster centers cluster_center[1,:7], including the following steps:

[0074] Step 21: Set the maximum number of iterations iter max , initialize IKHM algorithm parameters, initialize loop update instruction G 1 =0;

[0075] Step 22: Set the loop update instruction G 2 =G 3 =0;

[0076] Step 23: Perform IKHM clustering:

[0077] Step 231: Use the Sobol-AOA algorithm to update the cluster center, characterized in that the Sobol sequence is used to initialize j=best_num_cluster=2 initial cluster centers as the AOA initial population, the fitness values ​​of all initial cluster centers in the population are calculated and sorted, and the mathematical optimizer acceleration function is used to select the exploration or development stage to conduct global and local searches for the optimal cluster center.

[0078] Step 232: G 2 =G 2 +1. If G 2 <3. Then return to Step 3.1.

[0079] Step 233: The current position of the particle is used as the initial cluster center and updated using the KHM algorithm, characterized in that all cluster centers obtained in steps 231-232 are used to calculate the affiliation and corresponding weight of each data point, and each cluster center is updated by its affiliation, weight, and data point, and the update is repeated until the KHM iteration number set by the IKHM algorithm parameters is reached or the cluster center does not change.

[0080] Step 234: G 3 =G 3 +1. If G 3 <7, then return to step 233.

[0081] Step 24: G 1 =G 1 +1.If G 1 <itermax, then return to step 24.

[0082] Step 25: Get two cluster centers cluster_center[1,:8].

[0083] Step 3: Based on n cluster centers cluster_center[1,:8], construct a 2-dimensional (4100×8) mutagenic factor sequence clustering feature analysis matrix i_cluster_mat[:4100,:8], where i={0,1,2,...,n}. The grouping and clustering effect when the number of cluster centers is 2 is shown in the attached figure. Figure 3 As shown;

[0084] In step 4, the displacement sequences of the n mutagenic factor sequence clustering feature analysis matrices are subjected to denoising respectively, and the i-th group of displacement sequences is expressed as:

[0085] i_displacement[: 4100, 1]=i_cluster_mat[: 4100, 8],i={0,1,2,...,n} (5)

[0086] Perform EWT decomposition on the displacement sequence i_displacement[:4100,1] of each set of characteristic analysis matrix, set the number of sequence decompositions num_mra equal to the number of reservoir bank slope variable factors m=4, and obtain a displacement sequence decomposition matrix mra_mat[:4100,:4] through decomposition. The decomposition results are shown in the attached Figure 4 :

[0087] num_mra=mra_mat[ : T / n , : m].shape[1]=4 (6)

[0088] Calculate the fuzzy entropy FuzzyEn value of num_mra mra decomposition sequences of each group of displacement sequences:

[0089] FuzzyEn_mra[j]=EH.FuzzEn(mra_mat[:, j]), j={0,1,2,...,m} (7)

[0090] Calculate the SMSE values ​​of num_mra mra decomposition sequences for each group of displacement sequences:

[0091]

[0092] The noise sequences of the mra decomposition sequences with higher fuzzy entropy values ​​and lower SMSE values ​​in each mra decomposition sequence are eliminated, and the displacement sequence of each set of feature analysis matrices is reconstructed. The reconstruction results are shown in the attached Figure 5 .

[0093] Step 5: Based on the clustering characteristic analysis matrix of each group of variable factor sequences and considering the changes in hydrological and meteorological characteristics in the prediction interval, establish a model including the measured water level Measured rainfall Measured temperature Measured aging (δ θ ) factors and displacement sequence δ 位移序列[t,t+a] , and the water level forecast obtained from hydrological and meteorological forecasts (δ HP ), forecast rainfall (δ RP ), predicted temperature (δ TP)'s multi-input deformation prediction dataset data_predict[:4100,:4+9+1+3], including the following steps:

[0094] Step 51: Set the step length of short-term prediction to 3 days, define the prediction interval [t, t+3], collect hydrological and meteorological forecast data within the prediction interval, and obtain the predicted water level at each t+i moment Precipitation forecast Forecast Temperature i={1,2,3}.

[0095] Step 52: Build including measured water level Measured rainfall Measured temperature Aging (δ θ ) factors and the displacement sequence (δ 位移序列[t,t+a] ), and the water level forecast obtained from hydro-meteorological forecasts Precipitation forecast Forecast Temperature Factored multi-input deformation prediction dataset data_predict[:4100,:4+9+1+3]:

[0096]

[0097] Step 53: Based on the two multi-input deformation prediction data sets data_predict[:4100,:4+9+1+3] constructed from the above measured and predicted water levels, rainfall, temperature, time factor and displacement sequences, set the model output mode to multi-input multi-output, apply the TimeMixer model to predict and output the short-term three-step-day deformation prediction matrix Y_predict[:8201,1+3] of the reservoir bank slope in the form of multi-output.

[0098] Step 54: Take the average value of each row of the short-term 3-step-day deformation prediction matrix Y_predict[:8201,1+3] of the reservoir bank slope Average.Y_predict[i,1+3], i={1,2,...,8201} as the short-term deformation prediction value of the reservoir bank slope and output it.

Claims

1. The IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slope considering hydrological and meteorological changes in the prediction interval is characterized by: The following steps are involved: Step 1: Collect the historical deformation sequence of the reservoir bank slope with a time dimension of T and m reservoir bank slope change factors, cluster and distinguish the historical deformation sequence, set the upper limit of clustering uln_cluster equal to the number of reservoir bank slope change factors m, and calculate the prediction accuracy comprehensive evaluation function MF (MAE, MSE, R 2 )Determine the best number of clusters best_num_cluster; Step 2: Initialize the IKHM algorithm and cluster the reservoir bank slope mutation factor sequences with similar characteristics, set the number of cluster centers n equal to the optimal cluster number best_num_cluster, and determine the cluster centers; Step 3: Based on the cluster centers, construct a clustering feature analysis matrix of mutagenic factor sequences with n dimensions (T×(m+1)); Step 4: De-noise the displacement sequences of the clustering feature analysis matrices of the n variable factor sequences. First, perform EWT decomposition on them. Set the number of sequence decompositions equal to the number of reservoir bank slope variable factors m. The displacement sequence of each group of feature analysis matrices is decomposed to obtain a displacement sequence decomposition matrix, which contains m decomposition sequences mra. The fuzzy entropy value and SMSE value of each decomposition sequence mra are calculated. The mra decomposition sequence with a higher fuzzy entropy value and a lower SMSE value in each decomposition sequence mra is judged as a noise sequence and removed to reconstruct the displacement sequence of each group of feature analysis matrices. Step 5: Based on the displacement sequence of the clustering feature analysis matrix of each group of mutation factor sequences and considering the hydrological and meteorological changes in the prediction interval, a multi-input deformation prediction data set is established, and a multi-step prediction model for short-term deformation of the reservoir bank slope is constructed. The TimeMixer model is used to predict the deformation amplitude of the reservoir bank slope in the short term in any step length in the future.

2. The IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slope considering hydrological and meteorological changes in the prediction interval according to claim 1, characterized in that: The acquisition time dimension in step 1 is T, and the historical deformation sequence data[:,:] of the bank slope with m bank slope deformation factors is (T×(m+1)) First, set the slice ratio and slice range to (0, 1), and obtain the historical deformation sequence training set data_train[:,:] and the historical deformation sequence test set data_test[:,:]: data_train[: , :]=data[: , :]*slice (1) data_test[: , :]=data[data[: , :]*slice.shape[0]: , :] (2) Then cluster the historical deformation sequence and set the upper limit of clustering uln_cluster equal to the number of bank slope mutation factors m: uln_cluster=data_train[: , :].shape[1]-1 (3) The performance results of the historical deformation sequence under different cluster numbers (i, 1 to the upper limit of cluster number uln_cluster) are analyzed, including the mean absolute error MAE and its maximum value MAE.max, the mean square error MSE and its maximum value MSE.max, the absolute coefficient R 2 and its maximum value R 2 .max, to calculate the prediction accuracy comprehensive evaluation function MF (MAE, MSE, R 2 ) Determine the best number of clusters best_num_cluster, and its evaluation function is specifically expressed as follows:

3. The IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slope considering hydrological and meteorological changes in the prediction interval according to claim 1, characterized in that: Step 2 initializes the IKHM algorithm and clusters the reservoir bank slope mutation factor sequences with similar characteristics, sets the number of cluster centers n equal to the optimal number of clusters best_num_cluster, and determines n cluster centers cluster_center[1,:m+1], including the following steps: Step 21: Set the maximum number of iterations iter max , initialize IKHM algorithm parameters, initialize loop update instruction G1 = 0; Step 22: Set the loop update instruction G2=G3=0; Step 23: Perform IKHM clustering: Step 24: G1 = G1 + 1. If G1 < itermax, return to step 24; Step 25: Get n cluster centers cluster_center[1,:m+1].

4. The IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slope considering hydrological and meteorological changes in the prediction interval according to claim 3, characterized in that: The step 23 specifically includes: Step 231: Use the Sobol-AOA algorithm to update the cluster center, initialize j=best_num_cluster initial cluster centers with the Sobol sequence as the AOA initial population, calculate the fitness values ​​of all initial cluster centers in the population and sort them, select to enter the exploration or development stage through the mathematical optimizer acceleration function, and expand the global and local search for the optimal cluster center; Step 232: G2=G2+1; If G2<3; then return to step 231; Step 233: The current position of the particle is used as the initial cluster center and updated using the KHM algorithm, characterized in that all cluster centers obtained in steps 231-232 are used to calculate the affiliation relationship and corresponding weight between each data point and the initial cluster center, and each cluster center is updated by its affiliation, weight, and data point, and the update is repeated until the KHM iteration number set by the IKHM algorithm parameters is reached or the cluster center does not change; Step 234 : G3=G3+1; If G3<7, return to step 233 .

5. The IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slope considering hydrological and meteorological changes in the prediction interval according to claim 1, characterized in that: Step 3: Based on the n cluster centers cluster_center[1,:m+1], construct n mutagenic factor sequence clustering feature analysis matrices i_cluster_mat[:T / n,:m+1] with dimensions ((T / n)×(m+1)), where i={0,1,2,...,n}.

6. The IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slope considering hydrological and meteorological changes in the prediction interval according to claim 1, characterized in that: In step 4, the displacement sequences of the n mutagenic factor sequence clustering feature analysis matrices are subjected to denoising respectively, and the i-th group of displacement sequences is expressed as: i_displacement[:T / n,1]=i_cluster_mat[:T / n,m+1],i={0,1,2,...,n} (5) Perform EWT decomposition on the displacement sequence i_displacement[:T / n,1] of each set of characteristic analysis matrices, set the number of sequence decompositions num_mra equal to the number of reservoir bank slope variable factors m, and obtain a displacement sequence decomposition matrix mra_mat[:T / n,:m] by decomposing each set of displacement sequences: num_mra=mra_mat[ : T / n , : m].shape[1] (6) Calculate the fuzzy entropy FuzzyEn value of num_mra mra decomposition sequences of each group of displacement sequences: FuzzyEn_mra[j]=EH.FuzzEn(mra_mat[:, j]), j={0,1,2,...,m}(7) Calculate the SMSE values ​​of num_mra mra decomposition sequences for each group of displacement sequences: SMSE_mra[j]=mra_mat[i,j]**2+SMSE_mra[0,j], i={0,1,2,...,mra_mat[:].shape [0]},j={0,1,2,...,m} (8) The noise sequences of the mra decomposition sequences with higher fuzzy entropy values ​​and lower SMSE values ​​in each mra decomposition sequence are eliminated, and the displacement sequence of each group of feature analysis matrices is reconstructed.

7. The IKHM-EFS-TimeMixer prediction method for short-term deformation trend of reservoir bank slope considering hydrological and meteorological changes in the prediction interval according to claim 1, characterized in that: Step 5: Based on the clustering characteristic analysis matrix of each group of variable factor sequences and considering the changes in hydrological and meteorological characteristics in the prediction interval, establish a model including the measured water level Measured rainfall Measured temperature Measured aging (δ θ ) factors and displacement sequence δ 位移序列[t,t+a] , and the water level forecast obtained from hydrological and meteorological forecasts (δ HP ), forecast rainfall (δ RP ), predicted temperature (δ TP )'s multi-input deformation prediction dataset data_predict[:T / n,:m+3a+1+a], including the following steps: Step 51: Set the step length of short-term prediction to a days, define the prediction interval [t, t+a], collect hydrological and meteorological forecast data within the prediction interval, and obtain the forecast water level at each t+i moment Precipitation forecast Forecast Temperature i = {1, 2, ..., a}; Step 52: Build including measured water level Measured rainfall Measured temperature (δ Tt ), aging (δ θ ) factors and the displacement sequence (δ 位移序列[t,t+a] ), and the water level forecast obtained from hydro-meteorological forecasts Precipitation forecast Forecast Temperature Factored multi-input deformation prediction dataset data_predict[:T / n,:m+3a+1+a]: Step 53: n multi-input deformation prediction data sets data_predict[:T / n,:m+3a+1+a] are constructed from the above measured and predicted water levels, rainfall, temperature, time factor and displacement sequences, and the model output mode is set to multi-input multi-output. The TimeMixer model is used for prediction and the deformation prediction matrix Y_predict[:T,1+a] of the reservoir bank slope in the short term of a step length days is output in the form of multiple outputs; Step 54: Take the average value of each row of the deformation prediction matrix Y_predict[:T,1+a] of the reservoir bank slope for a short-term step length of a days, Average.Y_predict[i,1+a], i={1,2,...,T} as the short-term deformation prediction value of the reservoir bank slope and output it.

Images