Occurrence assessment of earth fissure based on genetic algorithms and artificial neural networks in Su-Xi-Chang land subsidence area, China

Abstract

Earth fissures in Su-Xi-Chang land subsidence area have induced massive damages to the area. The non-linear characteristic associated with the process of earth fissure formation requires non-linear method for evaluating the occurrence of the hazard. Based on quantification of influence factors on breeding the hazard, GA-ANN method, which integrates artificial neural networks (ANN) with genetic algorithms (GA), is developed for evaluating the occurrence of earth fissure hazard. Six indicators, that include the depth of bedrock burial (DBB), the degree of bedrock relief (DBR), water level (WL) (the II confined aquifer), the gradient of land subsidence (GLS), transmissivity (T) (the II confined aquifer) and the thickness of clay soil (TCS), are selected as the input patterns of the integrated approach, and danger index (DI) as the output pattern. A multilayer back-propagation neural network is trained with 30 sets of data samples including 15 sets of earth fissure samples and 15 sets of safety samples for defining the architecture of ANN. Subsequently, GA is employed by optimizing the initial weights of trained ANN by minimizing the deviation of output. The efficacy of the integrated approach is demonstrated by comparing the deviation of output from ANN and GA-ANN for 5 testing samples and the result shows that the GA-ANN method is more accurate than ANN in identifying the occurrence of earth fissure. The integrated method is applied to the assessment of earth fissure hazard in typical regions of earth fissure. According to the classification of DI, the regions are divided into four zones — danger zone, sub-danger zone, sub-safe zone and safe zone.