根据实测冰情数据分析发现，可将冰盖厚度演变过程作为预测头道拐站的开河日期的主要依据，同时还应考虑封冻期气温、流量等对冰盖厚度的持续性和累积性影响。据此提出了一种应用数据挖掘技术和LSSVM进行头道拐站开河日期预测的新方法。应用LSSVM模型对头道拐站2010年、2011年和2012年开河日期的预测结果表明，可在封冻期内任一冰盖厚度测量日期利用上述方法对该站的开河日期进行预测，有效延长了预见期，且在3月6日前的预测值均满足许可误差合格率的要求。根据LSSVM模型预测误差呈波动性变化的特点，提出了预测开河日期的均值法，可使开河日期预测精度得到显著提高。

According to the analysis of measured ice data, the paper pointed out that the evolution of the ice sheet thickness can be used as the main basis for forecasting the break-up date at the TouDaoguai station along the Yellow River, and the persistent and cumulative effects of temperature and flow during the freezing period are also important factors which affect the timing of break-up date. A new method to forecast the break-up date using the data mining technology (DM) and least square support machine (LSSVM) was presented in this paper. The LSSVM model was used to predict the break-up dates at the TouDaoguai station in 2010, 2011, and 2012, and the results showed that the method can predict the break-up date on any date of ice sheet thickness observation during the frozen period, which can expand the forecast period effectively. Furthermore, the predicted values of break-up date before March 6th met the requirements of the allowable prediction error rate. An average value method of forecasting the break-up date was proposed according to the characteristics of volatility change in the prediction error produced by the LSSVM model, which can improve the prediction accuracy significantly.

水利部公益性行业科研专项项目(201201080）;中央级公益性科研院所基本科研业务费专项资金项目(HKY-JBYW-2010-06)