Particle Swarm Optimization Based Wood Property Parameters Neural Network Modeling

Wood internal structure and its physical and mechanical properties model, as a real complex system, close ties exist between its physical parameters and mechanical properties. Also, different species have different degrees of difference. As an organic whole, different wood parameters have the unknown nonlinear relationships. This brought a certain degree of difficulty for prediction modeling and prediction accuracy improvement. Therefore, feasible and practical technology methods are needed to improve the prediction precision of wood property parameters model. This can provide an important scientific basis for the study of improved wood materials.Based on the above objects, this dissertation carried out the research mainly on the following aspects.(1) Investigate the latest developments and progress of wood parameter modeling in recent years. Based on the above information, research ideas and methods of this dissertation are presented. Thought an analysis of the physical properties of wood structural parameters, neural network modeling method for wood property parameters is applied.(2) Select Maoershan larch as tree species. Wood specimens are produced. Parameter measurement design for wood growth ring density, moisture content and the longitudinal elastic modulus is carried out to provide data preparation for the follow-up modeling.(3) The basic principles, model structure and the main steps of neural network modeling are presented. Considering some problems and difficulties in the practical application of BP (Back Propagation) algorithm, the causes and the solutions to these problems are discussed.(4) Introduction of the basic properties of the particle swarm optimization (PSO) algorithm are presented, its development and application are summarized. Based these, a combined algorithm by BP and PSO algorithm is proposed to optimize and design the neural network.(5) In order to test the validity of the method presented in this dissertation, PSO based optimization method of neural network is applied to wood property parameter modeling, which is a breakthrough in the pursuit of a single binary relation to the traditional model. The mappings of mechanical properties are achieved among wood density, moisture content and longitudinal elastic modulus. Meanwhile, material variation from the heartwood to the sapwood is reflected. Experimental and simulation results show that the method is effective.