基于降水量的白洋淀最低水位预测研究

为了每年能够提前预测未来10月份至翌年4月份的白洋淀最低水位，指导水资源管理和防灾减灾，利用白洋淀区域多个气象站点雨季不同时段的平均降水量和白洋淀水位资料，采取历史资料回归和机器学习方法，对白洋淀水位随区域降水量的变化规律进行了分析，研究建立了一种利用当年雨季平均降水量和当年雨季前水位预测当年雨季后至翌年雨季前白洋淀最低水位的方法。通过建立最低水位预测模型对已有的数据进行验证，发现所建的模型模拟和预测的结果误差在5%以下，精度较高。根据2018年5月白洋淀水位和白洋淀区域2018年7-8月、7-9月平均降水量预测的2018年10月至2019年4月白洋淀的最低水位分别为8.52 m和8.38 m，根据最新监测的2018年10月至2019年2月20日的最低水位实况，预测误差在4%以下，预测精度较高。因此，所建模型能够提前3~7个月动态预测白洋淀即将出现的最低水位，可为提高雄安新区区域水资源综合管理水平、统筹分配区域水资源、合理安排补水等提供科学依据。

Prediction of minimum water level in Baiyangdian Lake based on precipitation

The planning outline of Xiong''an New Area clearly states that the proportion of blue-green space in the Xiong''an New Area will be stable at 70% in the future. Baiyangdian Lake is the largest wetland and water body in the Xiong''an New Area. Understanding and predicting changes in this water body is of great significance to ensure the production of domestic water and ecological security of the Xiong''an New Area. In order to predict the lowest water level of Baiyangdian Lake from October to April of the next year, and to guide water resources management and disaster prevention and mitigation of the Xiong''an New Area, a historical data regression method and a machine learning method were used to analyze the water level variation in the Baiyangdian Lake with different regional precipitation. It was found that the lowest water level of Baiyangdian Lake from October to April was highly correlated with both the mean precipitation in the current rainy season and the water level in May of that year. Based on this result, a method was established to predict the lowest water level in Baiyangdian Lake from the current rainy season until the next rainy season by using the precipitation data in the current rainy season and the water level before the current rainy season. According to the precipitation and water level data in Baiyangdian Lake from 2001 to 2017, the prediction model for the lowest water level was established. Furthermore, gradient descent algorithm was adopted for machine learning and training. After model verification, the lowest water level prediction model was developed based on the average rainfall from July to August/July to September and the water level in May as the prediction factors. The model was verified using the collected data; the fit of the model was acceptable as the simulated and predicted result errors were both below 5%. According to the Baiyangdian Lake water level in May 2018 and the average precipitation data of July to August/July to September 2018, the lowest water level of Baiyangdian Lake from October 2018 to April 2019 would be between 8.52 m and 8.38 m, which was higher than the lowest ecological water level of Baiyangdian Lake. According to the latest minimum water level from October 2018 to February 20, 2019, the model predicted result errors were below 4%. Thus, our model can predict the lowest water level of Baiyangdian Lake 3-7 months in advance, which provides a significant scientific basis to improve the comprehensive management and the rational allocation of water resources in the Xiong''an New Area.