Study On Groundwater Prediction Method And Its Application In Northern Large-Scale Irrigation District

Groundwater is important irrigation source for the large northern irrigation districts. Groundwater dynamitic is not only external behavior of natural factors and human activity, but also an important signal to evaluate the rational development and utilization of groundwater resources. Consequently, research on prediction of groundwater dynamitic in large northern irrigation has both theoretical and practical value for the rational development and utilization of groundwater resources and sustainable development of irrigation agriculture.According to analysis of groundwater characteristics and type classification, quantitative and qualitative analysis combined with modeling were applied to study methodology to predict groundwater dynamitic and application. Main conclusions were drawn as follows:(1) Based on literature review on the advantage, disadvantage and applicable conditions of regular groundwater dynamics prediction methods, and also taking into main influencing factors and dynamic characteristics, wavelet de-noising based partial least squares regression model and chaos models based on support vector machine were introduced into groundwater dynamics prediction. The characteristics and applicability were discussed according to the specific case study.(2) According to modeling principle and methods of wavelet de-noising based partial least squares regression model, models about Yuanshang, Yuanxia irrigation districts in Baojixia irrigation district and monitoring well 534 of Wuwei Basin in Shiyang river basin were constructed, with the mean square error of 0.77, 0.90 and 0.25, indicating a high fitting and prediction accuracy. Comparing the model results with multi-linear regression model and main component regression model, it showed that wavelet de-noising based partial least squares regression model is better than these two methods. The reason is that it is able to solve the multiple correlations and improve the stability, fitting and prediction accuracy of models.(3) Chaos models based on support vector machine were applied to three monitoring wells of Wuwei basin and Yuanshang, Yuanxia irrigation districts to predict groundwater dynamitic. The results showed that the prediction accuracy were 0.98, 0.92, 0.86, 0.75 and 0.82 respectively. Comparing the results with one-rank local-region model of chaotic, it showed that chaos models based on support vector machine has a better prediction accuracy because it can explore the useful information of singe variable series with generalization ability and it also solves the local optimization. As a consequent, it will improve the prediction accuracy of groundwater table. In addition, the results showed that the groundwater table series have obvious chaotic components, which have significant impact on groundwater dynamics variation.(4)Wavelet de-noising based partial least squares regression model has no high demand for data. When the samples are limited, it still has satisfactory prediction results and the model can reflect the influencing degree of all factors on the groundwater dynamics. It is good for short term of groundwater table prediction. However, chaos models based on support vector machine gain significant advantage when the study area has long term groundwater table monitoring data in simulation and prediction. It can explore impact of other factors influencing a single variable on the single variable series and consequently can be used for the middle and long term groundwater table prediction. As a conclusion, both of these two methods have application values in the large irrigation districts.