基于ARIMA-GRU模型的地面沉降预测方法研究

Land subsidence prediction method based on ARIMA-GRU model

由于社会经济的快速发展和城市规模的迅速扩大,地面沉降已成为我国许多城市面临的一个突出问题,有效的地面沉降预测可以更好地协调城市发展。以北京市通州区为研究区域,结合一种组合模型ARIMA-GRU进行地面沉降预测方法研究。把2005—2015年的年度水准点高程数据作为基准值,运用方差倒数法确定组合模型最优解的权重比例,并采用线性模型和非线性模型的组合方法预测2016—2019年的水准点高程。运用2个单独模型和ARIMA-GRU组合模型进行未来数据趋势变化的预测,并对MSE、RMSE、MAPE模型指标值进行量化分析,和单一ARIMA和GRU模型相比,MSE分别降低了76.84和9.5,RMSE分别降低了3.58和0.52,MAPE分别降低了0.13和0.01。组合模型预测结果的MSE、RMSE、MAPE,预测效果更好。建模实验结果表明,该方法能预测地面沉降变化并实现有效预测,验证了模型的可行性。

Due to the rapid development of social economy and the extensive expansion of urban scale,land subsidence has become a prominent problem for many cities in China.Effective land subsidence prediction can better coordinate urban de-velopment.In this paper,Tongzhou District of Beijing is taken as the research area,and a combined model ARIMA-GRU is used to study the land subsidence prediction method.Firstly,taking the annual leveling point elevation data from 2005 to 2015 as the reference values,the inverse variance method is used to determine the weight proportion of the optimal solu-tion of the combined model.And the combination method of linear model and nonlinear model are used to predict the level-ing point elevation from 2016 to 2019.Then,two separate models and ARIMA-GRU combined model are used to predict the future data trend change,and the index values of MSE,RMSE and MAPE are quantitatively analyzed.Compared with the single ARIMA and GRU model,MSE is reduced by 76.84 and 9.5,RMSE by 3.58 and 0.52,and MAPE by 0.13 and 0.01 re-spectively.The MSE,RMSE and MAPE of the combined model prediction results have better prediction effect.The modeling experiment shows that this method can better predict land subsidence change in an effectively way and produces a model with proved higher feasibility.