基于改进ROA-LSTM时间序列的长江短期流量预测  

Short-Term Forecast of Yangtze River Discharge Based on Improved ROA-LSTM Time Series

作　　者：刘恒[1] 范洋 王聪[1] 丘仲锋 LIU Heng;FAN Yang;WANG Cong;QIU Zhongfeng(School of Electronic&Information Engineering,Nanjing University of Information Science&Technology,Nanjing Jiangsu 210044,China)

机构地区：[1]南京信息工程大学电子与信息工程学院,江苏南京210044

出　　处：《传感技术学报》2024年第12期2064-2070,共7页Chinese Journal of Sensors and Actuators

基　　金：国家重点研发计划项目(2019YFC1804704)。

摘　　要：为了更准确地预测长江流量的短期时序变化,克服传统LSTM模型在时间序列预测中参数选择困难和易陷入局部最优解的问题,通过将WOA算法与SFO算法改进的ROA优化算法与注意力机制相结合,构建了ROA优化算法与LSTM模型相结合的时间序列预测组合模型ROA-LSTM。将该模型的预测结果与声层析系统的实测长江流量数据进行对比分析,在三日以内的短期预测中,该模型相比传统RNN模型预测准确度提升2~3倍,并在流量变化波动趋势和峰值预测方面表现更为出色。To enhance the accuracy of short-term predictions regarding the temporal variations in Yangtze River flow,and to address chal-lenges related to parameter selection as well as the propensity for traditional LSTM models to converge on local optimum solutions in time series forecasting,a hybrid model known as ROA-LSTM is developed.This model integrates an improved ROA optimization algorithm with an attention mechanism alongside the WOA and SFO algorithm,effectively combining the ROA optimization technique with the LSTM framework.The predictive performance of the proposed model is evaluated against actual Yangtze River flow data obtained from an acoustic tomography system.In terms of short-term forecasts spanning three days or less,the proposed model demonstrates a 2-3 fold increase in prediction accuracy compared to conventional RNN models.Furthermore,it exhibits superior capabilities in both trends forecasting in flow changes and peak predictions.

关 键 词：深度学习 短期流量预测 改进ROA-LSTM 注意力机制 参数优化 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]