Wood thermal conductivity prediction by using support vector regression

The support vector regression (SVR) method combined with particle swarm optimization (PSO) is proposed to establish a model for predicting the thermal conductivity of timber in transverse directions (radial direction and tangential direction) based on the measuring database of thermal conductivity under different influential factors, including its density, moisture content and specific gravity. Comparing the prediction results of SVR method with those from analogism (ANA) model and BP neural network (BPNN) model, it is shown that the prediction precision is higher for SVR method by applying identical training and test samples and increase of training samples could improve the generalization ability. With the validation test by leave-one-out cross validation (LOOCV) test, maximal absolute percentage error (MPE), mean absolute error (MAE) and mean absolute percentage error(MAPE), are the smallest for the prediction of SVR method. It is suggested that SVR is an effective and powerful tool for predicting thermal conductivity of timber.

Project Supported:

Clc Number: