Newton–Gauss curvature matrix based cDBN for online edible fungus drying prediction model

Abstract

In order to solve the problem of moisture measurement in the edible fungus production, which affect the drying process of edible fungus, here proposed the matrix based continuous belief network for on-line edible fungus drying prediction system designing. Firstly, here perform unsupervised learning process for the depth of the belief network with input signal of edible fungus acquisition, and extract the information feature of edible fungus based on continuous transmission, then realize the network weight training with conjugate gradient, after that here perform the deduction for the stability of the continuous depth of the belief network, which ensure the stability of the network output data; Then, the Newton–Gauss curvature matrix optimization method is used to replace the traditional error back propagation method, which take the local optimization of the network hidden layer weights, and realize the fast convergence and improvement of the convergence accuracy; Finally, the Lorenz function training was used to verify the validity of the matrix continuous depth belief network, and the results showed that the algorithm could improve the rate of finished products.