基于GA-GRNN的地表下沉系数预测方法研究

将遗传算法(GA)和广义回归神经网络(GRNN)方法进行融合,采用GA算法搜寻最优的GRNN光滑因子,简要分析了地表下沉系数的影响因素,建立了基于GA-GRNN的地表下沉系数预测模型。以我国典型观测站的数据资料作为学习和测试样本,将预测结果与实测值进行比较。结果表明:采用GA-GRNN模型预测地表下沉系数能够综合考虑诸多的地质采矿因素,预测结果与实测值得最大相对误差仅为5.44%,完全满足现场工程的需要,为今后预测地表下沉系数提出了一种新的方法。

Study on the prediction method of surface subsidence coefficient using GA-GRNN

A new method by combining Genetic Algorithm (GA) and Generalized Regression Neural Network (GRNN) is presented. In this method, the GA is used to optimize the smoothing parameter of GRNN. An intelligent prediction model for surface subsidence coefficient using this hybrid GA-GRNN algorithm is constructed based on the analysis of impact factors. Typical data of surface moving observation stations is used as learning and test samples. Comparison analysis is made between predicted values generated by GA-GRNN method and measured values. Results indicate that this model could make use of multi-factor. The maximum relative error between the predicted results and the measured is only 5.44%, which is fully meet the needs of field engineering. A new approach for the future prediction of surface subsidence coefficient is proposed.