改进粒子群优化GRU网络的储粮温度预测方法  

Grain Storage Temperature Prediction Method Based on GRU Network Optimized by Improved Particle Swarm Optimization

作　　者：蒋思玮 孙妍 陈静[1] 袁昕[1] 宋雪桦[1] JIANG Siwei;SUN Yan;CHEN Jing;YUAN Xin;SONG Xuehua(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212013)

机构地区：[1]江苏大学计算机科学与通信工程学院,镇江212013

出　　处：《计算机与数字工程》2023年第5期1036-1041,1156,共7页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61902156);江苏省自然科学基金项目(编号:BK20180860);镇江市重大专项成果转化(编号:ZD2019004)资助。

摘　　要：粮仓温度是判断储粮安全的重要指标,预测储粮温度的变化是储粮安全监测和预警的有效手段。论文提出一种基于门控循环单元的储粮温度预测方法,该方法构建两层GRU网络和全连接层,引入了非线性惯性因子和自适应学习因子的粒子群算法优化神经网络的初始权重,在模型中加入Dropout算法和RMSProp优化器训练网络参数。采用实验仓的传感器数据训练和测试模型,实验结果表明论文提出的IPSO-GRU模型预测值与实际值的均方根误差为0.078,与GRU网络、LSTM网络、BP网络对比误差分别减小13%、16%、74%,论文模型能很好地拟合储粮温度的变化。The temperature of granary is an important index to judge the safety of grain storage.Predicting the change of grain storage temperature is an effective way to monitor and warn the safety of grain storage.In this paper,a method for predicting grain temperature storage based on GRU is proposed,which constructs a two-layer GRU network and a fully connected layer.Particle swarm optimization with nonlinear inertial factor and adaptive learning factor are adopted to optimize the initial weight of the neural network.Dropout algorithm and RMSProp optimizer are added into the model to train network parameters.The experimental results show that the RMS error between the predicted value of IPSO-GRU model and the actual value is 0.078.The error is reduced by 13%,16%and 74%compared with GRU network,LSTM network and BP network.The model in this paper can well fit the grain temperature change curve.

关 键 词：门控循环单元 粒子群算法 神经网络 粮食温度预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]