基于神经网络参数优化的热失控状态温度预测研究

为提高夹杂热失控现象的微波干燥褐煤过程中神经网络预测温度精度,提出一种基于二次滤波及粒子群寻优的神经网络参数优化算法。该方法先引入小波分析对训练数据进行软阈值滤波处理,使温度数据在描述变化趋势的同时突出非平稳特征,而后使用粒子群算法寻找该趋势特征对应的神经网络最优的隐层节点数、学习率及最佳训练次数的组合,最后在预测中使用前向均值阈值滤波处理输入数据配合该最优网络进行预测。实验结果表明,该方法能同时提高热失控和非热失控状态下温度预测精度,使预测平均绝对误差下降59.2%。

Temperature prediction study base on parameter optimization ofSneural network in thermal runaway state

In order to improve theSprecision ofSneural network prediction temperature ofSmicrowaveSdrying of ligniteSprocess inclusionSof thermal runaway.SProposing aStwo filtering and particle swarmSoptimizationSalgorithm based onSneural networkSparameter optimization.SThe method firstSintroducedSwavelet analysis toStrainSdata bySsoft threshold filteringSprocessing,Smake the dataSto describe theSchange trend andSstand out theSnon-stationaryScharacteristics at the same time,Sand thenSuse theSparticle swarm algorithm to find theSoptimalSfeature ofSneural network: the hidden layer nodeSnumber,SlearningSrate and the bestStrainingStimes,SfinallySusedSForward meanSthreshold filtering deal with input dataStoSprediction withSthe optimal network.SThe experimental results show that,Sthis method canSalsoSimprove the thermal runawaySand non-thermal runawaySconditionStemperatureSprediction accuracy, theSprediction ofSthe mean absoluteSerror is decreased by 59.2%.