基于PSR和PSO的区域地下水埋深ELM预测模型

PSR and PSO-based ELM prediction model of regional groundwater buried depth

针对传统区域地下水埋深预测方法精度不高问题,提出一种基于相空间重构(PSR)、粒子群算法(PSO)的极限学习机(ELM)的非线性预测模型。首先利用C-C法对地下水埋深原始时序数据进行相空间重构(PSR),然后利用PSO-ELM对地下水埋深进行预测。将模型应用于中国黑龙江省红兴隆管理局红旗岭农场地下水埋深预测,结果表明:该模型取得了较好的预测效果,后验差比值C为0.074,小误差频率p为1,相对均方误差E1为6.36%,拟合准确率E2达到92.66%,试预报效果指标E3达到95.80%;与PSR-ELM、PSR-RBF等模型相比,PSR-PSO-ELM在试预报方面可使RMSE分别降低49%和70%,使误差区间分别降低28.2%和68.6%,证明PSO能够有效改善ELM模型的预测性能;分析了气候因素和人类活动对当地地下水埋深动态变化的影响。

Aiming at the problem of lower precision of the conventional method for predicting the buried depth of regional groundwater,a nonlinear prediction model based on empirical mode decomposition（ EMD）,phase space reconstruction（ PSR） and（ PSO） extreme learning machine（ ELM） of particle swarm optimization is proposed herein. At first,the phase space reconstruction（ PSR） is made on the original time series data of the buried depth of groundwater with C-C method,and then the buried depth of groundwater is predicted with PSO-ELM. This model is applied to the prediction of the groundwater depth of a farm in Heilongjiang Province,from which the result shows that better prediction effect is obtained from the model with the posterior difference ratio C = 0. 074,small error frequency p = 1 and relative mean square error E1= 6. 36%,while the fitting accuracy rate E2 reaches to 92. 66 and the test prediction effect index E3 reaches to 92. 66%. Compared with the models of PSR-ELM,PSR-RBF,etc.,the root-mean-square errors（ RMSE） can be lowered by 49% and 70% respectively by PSR-PSO-ELM in the aspect of the test prediction,which makes the error intervals be lowered by 28. 2% and 68. 6% respectively,thus demonstrates that PSO can effectively improve the prediction performance of ELM model,by which the impacts from climate factors and human activities on the dynamic change of the buried depth of the local groundwater are analyzed as well.