CN-122021749-A - Strip mine slope deformation prediction method and system based on SA-PSO-MLP fusion algorithm

CN122021749ACN 122021749 ACN122021749 ACN 122021749ACN-122021749-A

Abstract

The invention discloses a strip mine slope deformation prediction method based on an SA-PSO-MLP fusion algorithm, which comprises the following steps of 1, data preprocessing, 2, global searching of initial weights and thresholds of a multi-layer perceptron MLP model by using a particle swarm optimization algorithm PSO, and 3, refining of global optimal solutions obtained by searching of the particle swarm optimization algorithm PSO by using a simulated annealing algorithm SA to obtain weights and thresholds of an optimal multi-layer perceptron MLP network. The invention also provides a strip mine slope deformation prediction system based on the SA-PSO-MLP fusion algorithm. The invention combines the simulated annealing algorithm with the particle swarm optimization algorithm, fully plays the global optimizing capability of PSO and the local fine searching capability of SA, effectively solves the problems of easy sinking into local optimization and strong parameter sensitivity of the traditional model, has the advantages of high prediction precision, high convergence speed and strong robustness, and has good application effect on predicting the deformation of the strip mine side slope and similar geologic bodies.

Inventors

MIAO HAIBIN
JI RIGALA
WANG ZHICHAO
HE SHUAI
HAN MENG
ZHU WENDE
WANG DAN
YU JIANGHAO
YUAN RUIYANG
ZHANG YANG

Assignees

中煤科工集团沈阳研究院有限公司
煤炭科学研究总院

Dates

Publication Date: 20260512
Application Date: 20251229

Claims (10)

1. The strip mine slope deformation prediction method based on SA-PSO-MLP is characterized by comprising the following steps of: Step 1, data preprocessing: processing original displacement data monitored by a strip mine GNSS, wherein the processing comprises outlier detection, missing value filling and normalization to obtain preprocessing data, and dividing the preprocessing data into a training set and a testing set to be input as an MLP model; step 2, performing global search on initial weights and thresholds of the multi-layer perceptron MLP model by using a particle swarm optimization algorithm PSO, and specifically comprising the following steps: Step 201, setting particle swarm size Maximum number of iterations Inertial weighting ; Step 202, taking initial weight and threshold of the MLP model as particles in a particle swarm, and initializing the positions and the speeds of the particles into random number sequences respectively; Step 203, for each particle, calculating an fitness value of the current position of the particle, and evaluating each particle by the fitness value to obtain a global optimal position: If the fitness value is better than the historical optimal fitness value of the particle, taking the current fitness value as a new individual optimal position p i , and updating the current position x i as an individual optimal position; if the fitness value is better than the current global optimal position g, taking the fitness value as a new global optimal solution, and recording the current position of the particle as a new global optimal position; otherwise, go to step 204; Step 204, updating the speed and position of each particle; And step 3, refining a global optimal solution obtained by searching a particle swarm optimization algorithm PSO by using a simulated annealing algorithm SA to obtain a weight and a threshold of an optimal multi-layer perceptron MLP network, wherein the method specifically comprises the following steps of: step 301, given an initial temperature Inputting the initial solution obtained in the step 204; Step 302, according to the current temperature Generating a new solution and calculating the new solution And the change amount deltae of the fitness value of the current solution: ; Wherein, delta E is the adaptation value change of the new solution and the current solution, E new is the adaptation value of the new solution, E old is the adaptation value of the current solution; step 303, judging whether Δe is smaller than 0, if Δe is smaller than 0, accepting a new solution, and proceeding to step 306; If ΔE is greater than or equal to 0, then step 304 is entered; step 304, judging probability Whether or not greater than rand, rand represents the generation of a random number from the interval [0,1 ]; wherein the probability P , Is the current temperature; if P > rand, then accept the new solution and go to step 306; if P is less than or equal to rand, returning to the step 302; step 306, judging whether the termination condition is satisfied, wherein the termination condition is that the temperature is reduced to a certain minimum value Or the preset maximum iteration times T are reached, if the maximum iteration times T are met, the optimal solution is output, namely the weight and the threshold of the multi-layer perceptron MLP network; if not, a temperature decay is performed and the process returns to step 302 to continue the iteration.
2. The method for predicting slope deformation of strip mine based on SA-PSO-MLP as set forth in claim 1, wherein the outlier detection in the step 1 adopts a Z-score method, and the specific calculation formula is: ; where Z is the normalized data value, and if |Z| >3, the point is considered as an outlier, X is a data point, μ is the mean of the data, and σ is the standard deviation.
3. The method for predicting the slope deformation of the strip mine based on SA-PSO-MLP as set forth in claim 1, wherein the missing values in the step 1 are complemented by cubic spline interpolation.
4. The strip mine slope deformation prediction method based on SA-PSO-MLP as set forth in claim 1, wherein the inertia weight w adopts an adaptive update strategy: ; Wherein w max 、w min is the maximum inertial weight and the minimum inertial weight, T is the current iteration number, and T is the maximum iteration number.
5. The method for predicting slope deformation of strip mine based on SA-PSO-MLP according to claim 1, wherein the MLP model comprises an input layer, two hidden layers and an output layer, the hidden layer activation function is a ReLU, and the output layer is a linear function.
6. The method for predicting slope deformation of strip mine based on SA-PSO-MLP according to claim 1, wherein the data preprocessing in the step 1 generates samples by a time window method, the window length is 7 days, and the 8 th day displacement value is predicted by using the data of the first 7 days.
7. The method for predicting the slope deformation of the strip mine based on SA-PSO-MLP according to claim 1, wherein the fitness value is obtained by calculating a fitness function F, and the fitness function F adopts a weighted combination form of a mean square error and a decision coefficient, and the calculation formula is as follows: ; Wherein, the The value range is [0,1] for the weight coefficient; the MSE is the mean square error and, Wherein y m is the true value of the mth sample; n is the number of samples; R 2 is a coefficient of determination, ; Wherein, the As an average of the true values, 。
8. The method for predicting slope deformation of strip mine based on SA-PSO-MLP as set forth in claim 1, wherein in said step 204, the update formula of the speed and the position is: ; ; Wherein t is the iteration number, Is the velocity of the ith particle at generation t; the position of the ith particle in the generation t, p i as the optimal position of the individual, g as the global optimal position, c 1 、c 2 as the learning factor, and r 1 、r 2 as the random number in the interval of [0,1 ].
9. The method for predicting slope deformation of strip mine based on SA-PSO-MLP as set forth in claim 1, wherein the temperature decay in step 306 employs an exponential cooling strategy: ; Wherein, the The value range is that the temperature reduction coefficient is ; Is the current temperature.
10. A strip mine slope deformation prediction system based on an SA-PSO-MLP fusion algorithm, characterized by being configured to perform the strip mine slope deformation prediction method based on the SA-PSO-MLP fusion algorithm according to any one of claims 1 to 9.

Description

Strip mine slope deformation prediction method and system based on SA-PSO-MLP fusion algorithm Technical Field The invention belongs to the technical field of mine safety monitoring and geological disaster prediction, and particularly relates to a strip mine slope deformation prediction method and system based on an SA-PSO-MLP fusion algorithm. Background The instability of the slope of the strip mine is one of main dangerous sources for the safe production of the mine. The existing slope deformation prediction technology mainly comprises a finite element method based on a mechanical model, a statistical regression model and a prediction algorithm based on machine learning. The traditional mechanical model depends on complex geological parameters, so that the dynamic change process of the surface deformation is difficult to reflect in real time, and the statistical method such as a gray prediction model GM (1, 1) has lower precision when the sample is insufficient or the noise is larger. In recent years, neural networks and intelligent optimization algorithms are widely used in the field of slope monitoring. For example, the methods such as GA-BP, PSO-MLP, SSA-BP and the like improve model accuracy by optimizing initial weights of the network, but the problems of local optimization, high parameter sensitivity and the like are common. Especially under the complex geological conditions of the strip mine, the prediction stability and generalization performance of the algorithms are still insufficient. Therefore, how to construct a slope deformation prediction model with global optimizing capability and local convergence accuracy so as to improve prediction accuracy and model robustness becomes an important research direction in the field. Disclosure of Invention In order to solve the technical problems, the invention aims to provide a strip mine slope deformation prediction method and system based on an SA-PSO-MLP fusion algorithm, wherein SA is a simulated annealing algorithm, PSO is particle swarm optimization, and MLP is a multi-layer perceptron. The invention utilizes a simulated annealing algorithm to improve the searching capability of particle swarm optimization, and combines a multi-layer perceptron to realize high-precision prediction of the slope displacement of the strip mine. In order to achieve the above purpose, the invention provides a strip mine slope deformation prediction method based on SA-PSO-MLP, which comprises the following steps: Step 1, data preprocessing: processing original displacement data monitored by a strip mine GNSS, wherein the processing comprises outlier detection, missing value filling and normalization to obtain preprocessing data, and dividing the preprocessing data into a training set and a testing set to be input as an MLP model; step 2, performing global search on initial weights and thresholds of the multi-layer perceptron MLP model by using a particle swarm optimization algorithm PSO, and specifically comprising the following steps: Step 201, setting particle swarm size Maximum number of iterationsInertial weighting; Step 202, taking initial weight and threshold of the MLP model as particles in a particle swarm, and initializing the positions and the speeds of the particles into random number sequences respectively; Step 203, for each particle, calculating an fitness value of the current position of the particle, and evaluating each particle by the fitness value to obtain a global optimal position: If the fitness value is better than the historical optimal fitness value of the particle, taking the current fitness value as a new individual optimal position p i, and updating the current position x i as an individual optimal position; if the fitness value is better than the current global optimal position g, taking the fitness value as a new global optimal solution, and recording the current position of the particle as a new global optimal position; otherwise, go to step 204; Step 204, updating the speed and position of each particle; And step 3, refining a global optimal solution obtained by searching a particle swarm optimization algorithm PSO by using a simulated annealing algorithm SA to obtain a weight and a threshold of an optimal multi-layer perceptron MLP network, wherein the method specifically comprises the following steps of: step 301, given an initial temperature Inputting the initial solution obtained in the step 204; Step 302, according to the current temperature Generating a new solution and calculating the new solutionAnd the change amount deltae of the fitness value of the current solution: ; Wherein, delta E is the adaptation value change of the new solution and the current solution, E new is the adaptation value of the new solution, E old is the adaptation value of the current solution; step 303, judging whether Δe is smaller than 0, if Δe is smaller than 0, accepting a new solution, and proceeding to step 306; If ΔE is greater than or equal to 0, the