Research On Rapid Detection Method Of Formaldehyde Emission From Wood-based Panels Based On Small Sample Learning

Wood-based panels and other wood materials are widely used in indoor decoration,and the release of formaldehyde from these materials is one of the major contributors to indoor air pollution.Therefore,it is necessary to detect the formaldehyde emission from wood-based panels.The 1m3climatic chamber method,established as the arbitration standard in the national regulations,produces authoritative test results.However,it is characterized by a lengthy testing period,demands a high level of expertise in operation,and exhibits relatively low efficiency.On the other hand,emerging electrochemical sensor methods offer advantages such as high sensitivity,ease of operation,and rapid detection.The purpose of this thesis is to utilize machine learning algorithms to perform regression modeling on the detection results of the two methods mentioned above,in order to achieve a detection accuracy for the sensor method close to that of the 1m~3 climate chamber method.This aims to save manual inspection time,enhance detection efficiency,and achieve rapid detection of formaldehyde in wood-based panels.These mainly contents of this thesis are listed as follows.First of all,a climate chamber with controlled temperature,relative humidity,and ventilation rate is utilized for formaldehyde detection.Formaldehyde emissions from wood-based panel samples are simultaneously detected using a spectrophotometer and an electrochemical sensor.Data from the electrochemical sensor are preprocessed,and feature extraction and kernel principal component analysis are performed to serve as inputs for the model,with spectrophotometer results used as the model outputs.This data preparation process serves as the foundation for the subsequent modeling work.Secondly,for improving the accuracy of regression modeling with a small sample dataset,the support vector regression(SVR)based on statistical learning theory is adopted.An improved grey wolf optimizer algorithm(LGWO)is proposed to optimize the model’s parameters and further enhance the model’s generalization performance and accuracy.The LGWO algorithm improves the standard grey wolf algorithm in three aspects:(1)using an good point set theory to determine the initial population to increase population diversity;(2)introducing a non-linear convergence factor to balance the global and local search abilities of the population;(3)combining the Levy flight algorithm to enhance global optimization,and using an inertia weight strategy based on fitness to improve the optimization accuracy.The LGWO-SVR combination model is compared with other swarm intelligence algorithms to validate the effectiveness of the LGWO algorithm in improving the performance of the SVR model for predicting formaldehyde emissions from wood-based panels with a small sample dataset.Finally,to address the instability of single LGWO-SVR models in regression accuracy with small sample data,an Ada Boost.R2 ensemble learning strategy is adopted.The single LGWO-SVR model is used as a weak learner,and multiple weak learners are combined into a strong regression model through ensemble learning.The experimental results show that the Ada Boost.R2-LGWO-SVR model achieves stable fitting degrees of around 0.95 in multiple cross-validation experiments,and both regression accuracy and robustness are improved.In comparison experiments with different models,the proposed model outperforms the single LGWO-SVR,BPNN,and ELM models in all error metrics,validating the effectiveness of the proposed model for predicting formaldehyde emissions from wood-based panels.Moreover,compared with the 1m3climate chamber method,the proposed method based on the model eliminates manual operations such as solution preparation,gas sampling,chemical reactions,and concentration quantification,saving 1.75 hours per test,approximately 63.6%of the testing time,providing an effective solution for rapid detection of formaldehyde emissions from wood-based panels.