基于多策略改进WOA-BPNN模型的离心泵作透平性能预测  

Performance Prediction of Centrifugal Pump as Turbine Based on Multi-Strategies Improved WOA-BPNN Model

作　　者：余文进 周佩剑 顾杨帆 孟龙 温在鹏 Yu Wenjin;Zhou Peijian;Gu Yangfan;Meng Long;Wen Zaipeng(College of Metrology Measurement and Instrument,China Jiliang University,Hangzhou 310018,China;Key Laboratory of River Basin Digital Twinning of Ministry of Water Resources,China Institute of Water Resources and Hydropower Research,Beijing 100038,China)

机构地区：[1]中国计量大学计量测试与仪器学院,杭州310018 [2]中国水利水电科学研究院水利部数字孪生流域重点实验室,北京100038

出　　处：《水动力学研究与进展(A辑)》2024年第6期967-973,共7页Chinese Journal of Hydrodynamics

基　　金：水利部数字孪生流域重点实验室开放研究基金(Z0202042022);水利部重大科技项目(SKS-2022053);中国水科院基本科研业务费项目(HM0145B012021)。

摘　　要：泵作透平的性能预测对于小微型水电站的建设和余压能量回收具有重要意义。该文提出了一种基于多策略改进的鲸群优化算法(Whale Optimization Algorithm,WOA)的人工神经网络(BPNN)模型。该模型在原有WOA的基础上,引入了基于随机森林的权重归一化等多种策略,旨在提高离心泵作透平性能预测的精度和效率。该文采用改进型WOA-BPNN模型对泵作透平数据集进行训练,并与多种优化算法的预测结果对比分析,研究结果表明:改进型WOA-BPNN模型在训练过程中损失函数快速下降,均方误差值达到较低水平,且训练集与测试集R2值均能达到0.975以上,预测相对误差较小。在30次重复实验中,验证集的扬程和效率预测相对误差均保持在较低水平,满足工程实践需求。特别是在非高效工况下,改进型WOA-BPNN模型的预测相对误差显著优于其他模型,证明了其在离心泵作透平性能预测方面的可靠性和准确性。The performance prediction of pumps as turbines(PATs)is of great significance for the construction of small-scale hydropower plants and the recovery of residual pressure energy.In this paper,a multi-strategy improved Whale Optimization Algorithm(WOA)optimized Back Propagation Neural Network(BPNN)model is proposed.The model incorporates various strategies,such as weight normalization based on Random Forest,into the original WOA,aiming to enhance the accuracy and efficiency of performance prediction for centrifugal pumps operating as turbines.The improved WOA-BPNN model is trained using a PAT dataset and compared with the prediction results of various optimization algorithms.The results indicate that the loss function of the improved WOA-BPNN model decreases rapidly during training,with mean squared error values reaching a low level,and both the training and test set R2 values exceeding 0.975,showing minimal prediction error.In 30 repeated experiments,the prediction errors for water head and efficiency in the validation set remain at low levels,meeting the requirements of engineering practice.Notably,the improved WOA-BPNN model exhibits significantly better prediction accuracy in non-efficient operating conditions compared to other models,demonstrating its reliability and accuracy in PAT performance prediction.

关 键 词：泵作透平 人工神经网络 鲸群优化算法 随机森林 性能预测 

分 类 号：TH137[机械工程—机械制造及自动化]