云南省洱海灌区水稻智能灌溉决策模型研究  

Study on Smart Irrigation Decision-making for Paddy Rice in Erhai Irrigation District,Yunnan Province

作　　者：周梦林 陈士彪 赵学银 林恩 崔远来[1] 李宇琪 罗玉峰[1] 陈梦婷[1,3] ZHOU Meng-lin;CHEN Shi-biao;ZHAO Xue-yin;LIN En;CUI Yuan-lai;LI Yu-qi;LUO Yu-feng;CHEN Meng-ting(State Key Laboratory of Water Resources Engineering and Management,Wuhan University,Wuhan 430072,China;Dali Bai Autonomous Prefecture Institute of Survey and Design of Water Conservancy and Hydropower,Dali 671014,Yunnan Province,China;Guangdong Research Institute of Water Resources and Hydropower,Guangzhou 510610,China)

机构地区：[1]武汉大学水资源工程与调度全国重点实验室,武汉430072 [2]大理白族自治州水利水电勘测设计研究院,云南大理671014 [3]广东省水利水电科学研究院,广州510610

出　　处：《节水灌溉》2024年第5期52-58,65,共8页Water Saving Irrigation

基　　金：国家自然科学基金项目(52379046);广东省水利科技创新项目(2020-08)。

摘　　要：提高有效降雨利用率,是节约灌溉用水的重要途径之一。为进一步提高稻田降雨利用率,基于水量平衡原理和作物水分生产函数,结合强化学习方法,构建考虑未来降雨的智能灌溉决策模型。收集了大理站点2012-2020年的实测气象数据和天气预报数据,对智能灌溉决策模型进行训练,将模型应用于云南省洱海灌区。结果表明:洱海地区天气预报存在一定的空报率和漏报率,TS评分较高,降雨预报质量较高。与常规灌溉决策相比,采用强化学习方法的智能灌溉决策,平均每年可以减少灌溉次数0.2次,节约灌水量6.5 mm,节水率为6.0%,提高降雨利用率3.0%,减少排水量6.2 mm,且未造成产量损失。因此,采用智能灌溉决策能在考虑未来天气情况下有效提高降雨利用率,节约灌溉用水,且不造成减产。Improving the effective rainfall utilization rate is one of the main ways to save irrigation water in rice cultivation.To further improve the rainfall utilization rate,a reinforcement learning method for irrigation decisions considering weather forecasting is used to construct a water balance model and rice crop water production function to simulate the real environment and rice growth of farmland.The measured meteorological data and weather forecast data of Dali station from 2012 to 2020 were collected to train the intelligent irrigation decision-making model,and the model was applied to Erhai irrigation district in Yunnan Province.According to the experimental research and model training,the daily rainfall forecasting performance of Erhai irrigation district was acceptable and that potentially uncertain rainfall exists for learning and utilization.Compared with conventional irrigation decision-making,the use of reinforcement learning irrigation decision-making can reduce the irrigation frequency by 0.2 times per year,save irrigation water by 6.5 mm,with the save water rate of 6.0%.Moreover,it increases rainfall utilization rate by 3.0%,and reduces drainage by 6.2 mm with no production loss caused.Therefore,smart irrigation decision-making can effectively improve rainfall utilization and save irrigation water while considering production.

关 键 词：智能灌溉决策 强化学习方法 有效降雨 短期天气预报 

分 类 号：TV93[水利工程—水利水电工程]