地下水位预测模型对比分析研究

Comparative Analysis of Groundwater Level Prediction Models

为更好地研究地下水位预测模型,选用灰色GM(1,1)、叠加的马尔科夫链和BP神经网络3种模型,选取2007-2010年长春市的地下水位资料进行地下水位预测研究,对比分析了3种模型的预测结果以及其适用情况。结果表明,3种预测方法的平均绝对误差均小于10%,在一定程度上表明3种模型均具备一定可信度,其中叠加的马尔科夫链模型的误差相对较大,另外两种则相差不大。灰色GM(1,1)模型适用于把握数据的大体变化趋势;叠加马尔科夫链模型适用于对相对稳定的数据的预测;BP神经网络模型需要基于相对较多的数据进行有弹回地校正模拟,且预测的结果相对拟合程度更好,预测结果更理想。

In order to study the models of groundwater level, three models, including gray GM （1,1）, superimposed Markov chains and BP neural network, were selected to predict groundwater level in Changchun City base on the data from 2007 to 2010. The predict results of three models were compared and its applicability was analyzed. The results showed that the average absolute error of three prediction models were less than 10%, indicating that the three models had a certain credibility; the error of superimposed Markov chain model was relatively large, while the other two were less; the Grey GM（1,1） model was applicable to grasp the general trend of data; Superposition Markov chain model was suitable for relatively stable data prediction; BP neural network model need to base on relative more data to correct the simulation result, and its predicted results had better regression level.