排灌机械工程学报2025，Vol.43Issue(10)：981-989,9.DOI:10.3969/j.issn.1674-8530.24.0004
基于机器学习的泵喷尾部喷流非定常预测
Unsteady prediction of pump-jet tail jet flow based on machine learning
摘要
Abstract
To address the challenges of large computational scale,time-consuming analysis,and high costs associated with unsteady flow field calculations in pump-jet thrusters under navigation conditions,a flow field prediction framework was developed using two machine learning methods:recurrent neural network(RNN)and long short term memory(LSTM).Time-series data of the pump-jet tail jet flow field at a navigation speed of v=2.91 kn were obtained through numerical methods.Sample sets were constructed,and the prediction framework was trained to perform unsteady periodic predictions of the pump-jet tail jet flow.The results demonstrate that the determination coefficients(R2)of both predic-tion frameworks exceed 0.995 0,with mean absolute errors(MAE)below 0.079 0 and root mean square errors(RMSE)under 0.089 0.The computational time per training epoch does not exceed 40 s for the RNN model and 115 s for the LSTM model,respectively.The machine learning approach based on RNN and LSTM neural network prediction models proves capable of accurately and efficiently pre-dicting the complex unsteady flow characteristics and spatial distribution patterns of the pump-jet tail jet flow.Furthermore,the two models exhibit distinct predictive accuracies for different variables during long-term and short-term forecasting.The RNN model demonstrates superior efficiency and higher accuracy for short-term predictions,while the LSTM model exhibits enhanced predictive capa-bility for long-term forecasting.关键词
泵喷推进器/循环神经网络/长短期记忆网络/流场预测/回归/喷流Key words
pump-jet thruster/recurrent neural network/long short-term memory/flow field prediction/regression/jet flow分类
交通工程引用本文复制引用
郭荣,罗鑫,杨兴,王雪晗..基于机器学习的泵喷尾部喷流非定常预测[J].排灌机械工程学报,2025,43(10):981-989,9.基金项目
国家自然科学基金资助项目(52009050) （52009050）
中国博士科学基金资助项目(2020M673537) （2020M673537）
兰州理工大学红柳优青项目 （）
