Abstract:

Objective: The present study utilized an artificial neural network model to forecast the moisture content variation during the drying process of phenolic resin impregnated heat-treated bamboo bundles (PHB), elucidating the impact of drying temperature, drying time, paving method, and initial moisture content on the moisture variation during the drying process. The findings provide a fundamental reference for achieving high-quality and efficient drying of PHB. Method: The measured data of moisture content during the drying process of PHB samples was utilized to create a dataset, with input variables including drying temperature, drying time, paving method, and initial moisture content. The output variable was the moisture content during the drying process. Subsequently, this dataset was divided into three sets: the training set, consisting of 308 data points (70% of the total data); the validation set, comprising 66 data points (15% of the total data); and the test set, containing 66 data points (15% of the total data). The particle swarm optimization (PSO) algorithm was employed to optimize the initial weights and thresholds of back propagation ropagation (BP) neural network, thereby constructing a PSO-BP neural network prediction model. This model has been verified and analyzed. Result: The PSO-BP neural network model exhibited robust predictive capabilities. In the test set, it achieved a coefficient of determination (R²) of 0.98, a mean square error (MSE) of 1.27, a mean absolute error (MAE) of 3.73, and a residual predictive deviation (RPD) of 7.96. Compared to the BP neural network, the PSO-BP neural network demonstrated an improvement in R² and RPD by 6.53% and 110.2%, respectively, while reducing MSE and MAE by 54.0% and 71.86%. The model verification demonstrated that the moisture content variation during the drying process of PHB was primarily influenced by the drying temperature and paving method. Both factors have a significant impact on the predictive accuracy of the PSO-BP neural network model. Optimal prediction performance was achieved when using a drying temperature of 60 ℃, regardless of the four different paving methods employed, resulting in R² values exceeding 0.969 and MSE staying below 3. Employing three layers of paving yielded superior outcomes under various drying temperature conditions, with R² values surpassing 0.99 and MSE remaining below 2. Additionally, neither drying time nor initial moisture content significantly affected the predictive accuracy of the model. Conclusion: The PSO-BP neural network model exhibited remarkable accuracy in predicting the moisture content during the drying process of PHB samples. It effectively addresses issues such as significant prediction errors and slow convergence rates that are commonly encountered with traditional BP neural networks, thereby providing valuable technical support for the high-quality and efficient drying of PHB.

Key words: phenolic resin impregnated heat-treated bamboo bundles (PHB), drying, moisture content, particle swarm optimization (PSO) algorithm, back propagation (BP), neural network