摘要
降水量的预报精度对以其为直接或间接补给源的地下水资源评价具有重要的影响。将时间序列方法与随机过程离散状态的马尔可夫链理论相结合 ,提出了时间序列马尔可夫模型以预报大气降水量。模型根据降水序列的特征 ,采用一维非平稳时间序列模型进行预测 ,预测结果总体效果较好 ,但峰值点处误差较大。为了提高模型对波动性较大随机变量的预报精度 ,利用随机过程马尔可夫概率状态转移矩阵预报方法对其预测值进行二次拟合。实例计算表明 :时间序列马尔可夫模型预报效果良好 ,预报精度明显高于单一的时间序列模型精度。该结果拓宽了时间序列预报模型的应用范围 。
Prediction accuracy of precipitation has an important influence on evaluation of groundwater resources that is recharged directly or indirectly by the precipitation. In former literature, many authors usually adopted the frequency of precipitation in the past to predict the precipitation in the future, and brought it into the models which evaluated the quantity and/or quality of groundwater. But this method is rather conservative. In this paper, we combine time series method with the dispersed Markov Chains theory of stochastic process, and present a Markov Model based the time series analysis for predicting the precipitation. In the course of modeling, first we select one dimension nonstationary time series model to predict the precipitation in light of characteristics of the precipitation series. The result shows that the time series prediction is feasible as a whole, but there exists bigger errors when predicting the variables on the tops of the curves. In order to improve the prediction accuracy of the model, especially to the data with stronger fluctuation, we use the method of Markov′s state change probability matrix to fit them again. Then we attain the fitting values. To test the time series Markov model, we apply it to predict the precipitation in the section of Xuzhou in Jiangsu Province as an example. Results show that the time series Markov Model is efficient, and it has higher accuracy than that of the single model of time series. It enlarges applied scope of the time series model, and it is of important practical values and theoretical magnificence to the evaluation of groundwater resources that is finally recharged by the precipitation.
出处
《地理科学》
CSCD
北大核心
2001年第4期350-353,共4页
Scientia Geographica Sinica
基金
国家自然科学基金资助项目 (49772 162 )
博士点基金项目 (19990 2 842 1)
