Study Of Moisture Forecasting Model Based On GA-BP Neural Network Of Thin-Layer In Drying Process Of Alfalfa

Drying process of alfalfa, as it is a biological material, is complex and difficult to control, check and do real-time measurement of its moisture content with the existing apparatus and methods. However, the application of neural networks in simulation modeling for the thin-layer drying process of Alfalfa and predicting its moisture content can be very helpful for the design optimization of the Afalfa drying system as well as references collection for the automatic control of the system.This article is mainly focused on the following studies:1. Define three influencing factors to the moisture content in the Alfalfa drying based on a single factor Alfalfa drying experiment: hot wind temperature, hot wind speed, and the initial moisture content of Alfalfa; then identify the input and output factors for the modeling according to a three factors and four levels orthogonal experiment.2. Build a nonlinear multiple regression model according to the data collected from the orthogonal experiment, and this model is then demonstrated incapable to precisely predict the moisture content of Alfalfa.3. Apply BP neural networks in modeling for the thin-layer drying process of Alfalfa, which provides a solution to the precise modeling issue that cannot be worked out with traditional mathematical methods, and gets results of the best fitting goodness up to 0.99117, sample mean-square error at 0.685437, maximum absolute error at 9.4581% and mean absolute percentage error at 16.88%. According to these results, this model can offer more precise predictions than the multiple regression model, yet sometimes also with bigger errors.4. Introduce the method of genetic algorithm into determining the initial weights of a neural network, and propose the new method of thin-layer drying modeling of Alfalfa with GA-BP neural networks, which has solved the flaw that, because of improper application of the initial weights, the BP neural networks tend to show minimal data partly. The final results are: the best fitting goodness is up to 0.99813, sample mean-square error is 0.652233, maximum absolute error is 4.6847% and mean absolute percentage error is 3.19%. This method is proved capable to improve precision in network studies and predictions, and get better fitting goodness between predicted and measured data in the experimental conditions, thus get more precise predictions of the moisture content in the Alfalfa drying process.