楸木板材干燥含水率时间序列仿真分析与预测

Prediction and Simulation Analysis on Time Serials of Juglans mandshurica Lumber Moisture Content During Conventional Kiln Drying

对楸木板材常规窑干过程中含水率时间序列进行仿真分析,研究干燥过程中木材含水率的实时监控和预测。首先分析楸木板材干燥含水率随时间的变化特征,并通过三阶多项式对其进行拟合分析,然后利用人工神经网络构建模型模拟仿真分析,并利用构建的模型对木材干燥含水率进行预测。结果表明,常规干燥木材含水率呈非线性下降趋势,多项式拟合构建含水率与时间的定标函数为y=60.7155-0.0595x-0.0004256x2+0.0000006496x3,函数拟合仿真与实测值误差较大;构建拓扑结构为6×8×1的BP神经网络模型对含水率时间序列仿真结果最大误差为2.87%,利用训练好的网络模型预测含水率,平均误差为3.63%,最大误差为11.15%,最小误差为0.04%,25个样本中3个样本预测误差超过10%,其余22个样本的预测误差都小于9%;木材干燥含水率时间序列仿真分析和预测局域较大误差发生在含水率波动较大的区域,可以利用构建的BP神经网络对楸木干燥含水率进一步预测。

In this study,the time series of moisture content in Juglans mandshurica lumber during drying process were simulated and analyzed,and the real-time monitoring and prediction of moisture content in wood drying process were studied.Firstly,the change characteristics of the lumber moisture with time were analyzed,and fitted by the third-order polynomial.Then,the artificial neural network was employed to construct the model for simulation and analysis,and the model was used to predict the wood drying moisture content.The results showed that the moisture content of conventional drying wood presented a non-linear downward trend.The calibration function of water content and time constructed by polynomial fitting was y=60.7155-0.0595x-0.0004256x2+0.0000000006496x3,and there existed a large error between function fitting simulation and measured value.The maximum error of the simulation results was 2.87%obtained by constructing BP neural network model with the topology structure of 6×8×1.By the trained BP network model to predict water content,the average error was 3.63%,the maximum and minimum errors were 11.15%and 0.04%.The prediction errors of 3 samples in 25 samples were more than 10%,the prediction errors of the remaining 22 samples were less than 9%.Moreover,local large errors occurred in the region with large fluctuation of water content.The BP neural network can be used to predict the drying moisture content of J.mandshurica lumber in one step.