Prediction Algorithm Of MC Nearly FSP And Adaptive Weighted Fusion Method Of Temperature

Wood is renewable and recyclable resource. With excellent performance, friendliness to the environment and low energy consumption, it is world's recognized the green material. Wood drying is an important segment for the guarantee and improvement of the wood's quality and promotion of utilization rate of the wood. Lumber MC is one of the key parameters in drying process and quality, while wood's FSP is an important turning point of its performance.Because electrical measurement method is an equivalent resistance which measures ion mass migration in the moisture of wood and inorganic like gray, etc. FSP moisture content is free water finished evaporating and adsorption still stays in the lumen, this interval's wood dc features is unsteady. Experimental observation shows measuring MC of lumber FSP near section by electricity measurement will appear measurements'sudden deviation from the true value phenomenon, namely "blind spots". To further improve the precision entire range MC in timber drying process, the solution of the lumber FSP moisture content problems is needed.This paper researches statistical theory and the SVM model, expounds machine learning problem, experience and structural risk minimization principle, machine learning VC dimension basic theory, analysises the basic principle of SVM and expatiates the regression theory. Then it introduces the basic principle of BP neural network, topology structure, and the mapping relationship, analyses the training algorithm of BP neural network and its ideas.Based on research in lumber MC detection principle, this paper puts forward to conduct training model to content of the measured data through BP neural network and the SVM method, then forecasts MC near the FSP. Simulation results show that the BP neural network uses small sample data, and often appears "adapter" phenomenon when training; while using large sample data, generalization ability is strong in training, which can achieve accurate prediction effect. SVM forecasting processes higher precision than the BP neural network, and can be realized with a few sample data, solving the MC prediction near lumber FSP problem.Stove temperature is important control variable in drying process, due to the interference from the high temperature, humidity and dry environment, and fan relays shifting, temperature sensor detection accuracy is reduced.This paper proposes two level fusion methods in space and time based on adaptive weighted which requires no prior probability distribution knowledge in measurement data, with simple programming, little calculation, could effectively remove the errors in temperature sensor and improve precision of plume temperature.