Model for prediction of surface subsidence coefficient in backfilled coal mining areas based on genetic algorithm and BP neural network

Abstract

In order to effectively predict surface subsidence in backfilled coal mining areas, a prediction model is developed based on an improved genetic algorithm (GA) with optimized selection, crossover and mutation operators, and an improved back-propagation neural network (BPNN). The GA-BPNN model uses training and testing samples from mobile surface observation station data in backfilled coal mining areas in China as well as carefully-selected input variables. A comparison of calculated results from the GA-BPNN model and conventional BPNN model with measured data shows that the GA-BPNN model not only allows faster and more accurate learning, but also produces more reliable prediction results. It also has higher fitting precision and greater generalization abilities than conventional BPNN model, thus it is a new theoretical approach to the surface subsidence calculation in backfilled coal mining areas.