基于动态模糊神经网络的水电站来水预测方法  

Short⁃term inflow water prediction of hydropower stationsbased on dynamic fuzzy neural network

作　　者：彭放 郭金婷 柳玉兰 王立志 PENG Fang;GUO Jingting;LIU Yulan;WANG Lizhi(Guodian Dadu River Drainage Area Hydroelectricity Development Co.,Ltd.,Chengdu 610041,China)

机构地区：[1]国能大渡河流域水电开发有限公司,四川成都610041

出　　处：《电子设计工程》2025年第6期76-80,共5页Electronic Design Engineering

摘　　要：针对水库来水的波动性和随机性直接影响水位预测准确性的问题,基于T-S型模糊系统提出一种新的动态模糊神经网络进行短期(未来1 h)来水预测。采用FCM算法对输入空间划分产生规则,同时通过引入重构误差确定规则数目,从而构建紧凑的网络结构和完成规则前件参数学习;采用独特的结构单元作为后件层,使用梯度下降法给出后件参数调整方法。以瀑布沟入库流量作为实验数据进行验证,通过与BP、LSTM和ANFIS预测模型对比表明本文方法的预测精度更好,且预测稳定性更佳。For the problem of accurately prediction of water flow caused by the fluctuation and randomness of inflow water in hydropower stations,a new dynamic fuzzy neural network based on T-S fuzzy system is proposed to predict short-term(within 1 hour)inflow water.Using the FCM algorithm to partition the input space,a compact network structure is constructed by introducing reconstruction errors and completing the learning of antecedents parameter.With the unique structural elements as consequents of the rules,and through gradient descent method to provide consequents parameter adjustment methods.The proposed method was verified by using the inflow flow of Pubugou as experiment data,and compared with BP、LSTM and ANFIS prediction models.The results showed that the proposed method had better prediction accuracy and better prediction stability.

关 键 词：水电站 来水预测 准确度 模糊神经网络 

分 类 号：TN0[电子电信—物理电子学]