Time Series Prediction of Mining Subsidence Based on Genetic Algorithm Neural Network

Abstract

In order to find out the dynamics law of underground coal mining subsidence, BP neural network was used for time series prediction. First, genetic algorithm was used to optimize the initial network weight to overcome the inherent defects of BP neural network, then train the initial BP neural network with samples and a time series prediction model was established. A railway bridge observing station in a mining area of HeBei was shown as example to describe the method for time series prediction using genetic algorithm BP neural network (GA-BP). The maximum absolute error of forecast value is 14% and the maximum relative error is 15mm, results show that the forecast results fit for the measured values perfectly. The initial network weight can be selected effectively to use BP neural network for mining subsidence time series prediction and avoid the network falling into local minimum, and the network forecasting performance can be improved effectively. The research provides a new method for dynamic mining subsidence prediction.