基于IPSO-BP神经网络的高含沙水体对鱼类影响预测方法  

A prediction method for the impact of hyper-concentrated flow on fishes basedon the IPSO-BP neural network

作　　者：李晓晨 白音包力皋[1] 李向东 许凤冉[1] 穆祥鹏[1] 董志强 LI Xiaochen;Baiyinbaoligao;LI Xiangdong;XU Fengran;MU Xiangpeng;DONG Zhiqiang(State Key Laboratory of Simulation and Regulation of Water Cyclein River Bain,China Institute of Water Resources and Hydropower Research,Beijing 100038,China;Qinghai Institute of Water Conservancy and Hydro-electric power Design Co.,Ltd,Xining 810001,China)

机构地区：[1]中国水利水电科学研究院流域水循环模拟与调控国家重点实验室,北京100038 [2]青海省水利水电勘测规划设计研究院有限公司,青海西宁810001

出　　处：《水利学报》2023年第3期291-301,共11页Journal of Hydraulic Engineering

基　　金：青海省基础研究计划项目(2021-ZJ-759)。

摘　　要：水库进行水力排沙时,高含沙水流过程可能会对鱼类等水生动物产生负面影响,其量化评估方法研究较为薄弱。为了预测和评估水库排沙过程对下游鱼类的影响,本文利用黄河花斑裸鲤和鲤鱼在高含沙水体中生存特性研究的实验数据,综合考虑含沙量和粒径、溶解氧、暴露时间、水温等因子对鱼类生存的影响,建立了基于IPSO-BP神经网络的高含沙水体对鱼类致死影响预测方法,对目标鱼类死亡率的预测误差小于6%。本文使用了与BP神经网络紧密耦合并引入动态参数和变异扰动的IPSO算法,较BP和PSO-BP神经网络预测能力更佳,相比国内外已有的Stress Index(SI)、Severity of Ill Effect(SEV)和多元拟合方法预测精度得到显著提升。分析表明,本文提出的预测方法能够考虑高含沙水体中鱼类生存受多环境因子联合制约,且多因子之间存在复杂关联的情况,可为评估高含沙水流过程对水生态的影响提供新的方法。The hyper-concentrated flow process may cause negative impacts on fishes and other aquatic animals during reservoir sediment flushing.Nevertheless,there is a lack of studies on corresponding quantitative assessment methods for the degree of impact of the hyper-concentrated flow.In order to predict and evaluate the impact of reservoir sediment discharge processes on downstream fish,this paper uses experimental data from the study of survival characteristics of Yellow River Gymnocypris Eckloni and Cyprinus Carpio in hyper-concentrated flow and established an IPSO-BP neural network-based method for predicting the impact on fish mortality,and takes into account the effects of suspended sediment concentration,median particle size,dissolved oxygen,exposure time,water temperature,and other factors on fish survival.The prediction error of the target fish mortality is less than 6%.In this paper,the IPSO algorithm,which is closely coupled with BP neural network and introduces dynamic parameters and variational perturbations,has better prediction ability than BP and PSO-BP neural networks,and its accuracy is significantly improved when compared to existing Stress Index(SI),Severity of Ill Effect(SEV),and multivariate fitting evaluation methods at home and abroad.The analysis demonstrates that the prediction method proposed in this paper can account for the situation in which fish mortality in hyper-concentrated flow is governed by a combination of multiple environmental factors with complex correlations among multiple factors.This paper provides a new method for assessing the impact of hyper-concentrated flow on fishes.

关 键 词：IPSO-BP神经网络 高含沙水流 鱼类 致死率 预测方法 

分 类 号：TV145[水利工程—水力学及河流动力学] TP183[自动化与计算机技术—控制理论与控制工程]