摘要
为探索深度学习算法在水库调度领域的应用,利用网络爬虫技术,收集了溪洛渡水电站的调度运行数据,基于RNN、LSTM、GRU3种循环神经网络,学习电站现有调度规则,构建了溪洛渡水库的出流量预测模型,并探究不同参数设定对模型精度和计算速度的影响,对比了3种模型的模拟性能,分析了影响水库调度的主要因素。研究结果表明,隐层数、训练批量、迭代次数、隐层节点数和批量值是影响模型精度和计算速度的主要参数;3种模型具备良好的学习能力,能够根据水库的历史调度数据,学习应对不同场景的调度规则,生成出流方案,可为调度决策方案的制定提供参考依据。
To explore the application of deep learning algorithm in the field of reservoir operations,web crawler technique was used to collect operation data of Xiluodu Hydropower Station.The operation rules of this power station were obtained by recurrent neural network(RNN),long short-term memory(LSTM),gated recurrent unit(GRU).And then the prediction model of outflow for Xiluodu reservoir was established,and the impact of parameters on the model of accuracy and computing speed was discussed.The simulation performance of three models was compared,and the main factors for influencing reservoir operation were analyzed.The results show that the hidden layers,batch sizes,iterations and hidden nodes directly affect the precision and runtimes of three models;The three models have good learning ability,and can learn the operation rules for different scenarios based on the historical operation data,and generate the outflow scheme,which can provide a reference for formulating operation decision plan.
作者
汤正阳
张迪
林俊强
刘毅
彭期冬
尚毅梓
TANG Zheng-yang;ZHANG Di;LIN Jun-qiang;LIU Yi;PENG Qi-dong;SHANG Yi-zi(Hubei Key Laboratory of Intelligent Yangtze and Hydroelectric Science,China Yangtze Power Co.,Ltd.,Yichang 443000,China;State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin,China Institute of Water Resources and Hydropower Research,Yichang 443000,China)
出处
《水电能源科学》
北大核心
2021年第5期83-86,70,共5页
Water Resources and Power
基金
中国长江电力股份有限公司科技项目(2418020003)
流域水循环模拟与调控国家重点实验室自主研究课题(SKL2020ZY10)
江苏省“333工程”科研项目(BRA2019245)。
关键词
水库调度
出流量预测
循环神经网络
长短期记忆网络
门限循环单元网络
reservoir operation
outflow prediction
recurrent neural network
long short-term memory network
gated recurrent unit
