基于AGA-BP神经网络的采空区危险性评价

针对采空区危险性评价的影响因素众多且关系复杂的特点，提出了基于AGA-BP神经网络算法评价采空区危险性。将岩体结构、地质构造、岩石抗压强度等13个影响因子作为神经网络输入，采空区危险性等级作为输出，建立一个采空区危险性评价的BP神经网络模型；采用自适应遗传算法(AGA)对BP网络的初始权值和阈值进行全局寻优，将寻优结果回代入网络中进行训练并预测得出采空区危险性等级；利用其它智能算法与该预测结果做出比较，以验证AGA-BP算法的有效性及优越性。结果表明:该算法的优化效果明显，同时在训练时间与预测精度上较其它智能算法有突出的优势，是一种在采空区危险性评价方面值得推广的新方法。

Evaluation on risk of goaf based on AGA-BP neural network

Due to the numerous influence factors with complex relationship in risk evaluation of goaf, a new method based on adaptive genetic algorithm-back propagation neural network (AGA-BP) was put forward to evaluate the risk of goaf. Firstly, a BP neural network model on risk evaluation of goaf was established, with the 13 influence factors such as rock structure, geological structure and rock compressive strength as input and the risk level of goaf as output. Then the AGA was used to search the best initial weights and threshold of BP neural network globally, and the risk degree of goaf was obtained after the searching results are substituted into the network to train and predict. Finally, the prediction result was compared with the other intelligent algorithms to verify the effectiveness and superiority of AGA-BP algorithm. The results showed that the optimization effect of AGA-BP algorithm is obvious, the algorithm has prominent advantage on the training time and prediction accuracy, and it is worth to be popularized in risk evaluation of goaf.