發(fā)布時間：2018-10-20 15:54

【摘要】：遞歸神經(jīng)網(wǎng)絡(luò)(RNN)模型近年來在許多任務(wù)上表現(xiàn)出了優(yōu)良的性能。運用具有長短期記憶(LSTM)單元的遞歸神經(jīng)網(wǎng)絡(luò)構(gòu)建模型和通過時間反向傳播(BPTT)算法更新網(wǎng)絡(luò)權(quán)重解決長期降雨量的預(yù)測問題,較好地解決了高維數(shù)、非線性和局部極小問題。選取了前饋神經(jīng)網(wǎng)絡(luò)模型(FNN)、小波神經(jīng)網(wǎng)絡(luò)(WNN)模型和整合移動平均自回歸(ARIMA)模型3種模型進(jìn)行驗證比較。仿真結(jié)果表明,遞歸神經(jīng)網(wǎng)絡(luò)模型優(yōu)于其他模型,訓(xùn)練結(jié)果與實際值接近,預(yù)測精度較高。預(yù)測結(jié)果為農(nóng)業(yè)用水管理、合理制定灌溉制度提供了重要的科學(xué)依據(jù)。
[Abstract]:The recursive neural network (RNN) model has shown excellent performance in many tasks in recent years. Using the recurrent neural network with long and short memory (LSTM) unit to construct the model and renew the network weight through the time back propagation (BPTT) algorithm to solve the long-term rainfall prediction problem, the problems of high dimension, nonlinear and local minimization are well solved. The feedforward neural network model (FNN), wavelet neural network (WNN) model and the integrated moving average autoregressive (ARIMA) model are selected for verification and comparison. The simulation results show that the recurrent neural network model is superior to other models, the training results are close to the actual values, and the prediction accuracy is high. The predicted results provide an important scientific basis for the management of agricultural water use and the rational formulation of irrigation systems.
【作者單位】： 西安交通大學(xué)機(jī)械制造系統(tǒng)工程國家重點實驗室;
【基金】：“十三五”國家重點研發(fā)計劃(2016YFC0400202)
【分類號】：S161.6;TP183
，

本文編號：2283634