Research On Prediction Algorithm Of Rice Drying Process And Quality Change Based On Glass Transition

The majority of China’s rice-growing regions have a mild environment with high humidity,and the rice has a high moisture content after harvest.To prepare it for long-term storage,it must be dried.The traditional drying method greatly affects the rice’s quality.In order to minimize the quality loss of dried rice,researchers currently use neural network models to simulate the drying process,anticipate the drying outcomes,and optimize the drying process.In order to adjust the drying settings,the neural network model is frequently used to anticipate changes in moisture content and material quality.When the data is sufficient,the model’s prediction error is minimal.The prediction model’s error may rise as the drying process is more complicated or when there are other contributing factors.A wide range of techniques can be employed to optimize the model structure,lower prediction errors,and increase accuracy.Currently,there are numerous neural network models that can be utilized to create prediction models.The quality of the dried rice is more manageable the smaller the prediction error.Therefore,creating a neural network prediction model with less inaccuracy is quite important.First,a rubber-state split-range variable temperature drying process was presented.It was based on the theory of the glass transition and the simulation of heat and moisture transmission during the drying of rice.In order to compare the benefits and drawbacks of the three drying techniques,the constant temperature and variable temperature drying tests were conducted.Various neural network prediction models are established towards the end.To choose the neural network prediction model with the best performance,a large number of experimental data are used to train and test the prediction model based on the rubber state variable temperature drying process.These are the primary research findings and contents::(1)A constant temperature drying test was conducted to obtain rice with various moisture contents in order to explore the glass transition of rice during constant temperature drying.A thermal conductivity meter was used to measure the heat and moisture transfer parameters,which were then entered into the COMSOL Multiphysics 6.0 program.Rice was dried using a heat and moisture transfer model at 45,55,and 65 degrees Celsius.The glass transition process of rice was predicted based on changes in temperature and moisture in various rice parts during the simulation process,which provided a theoretical foundation for determining the drying parameters in the first stage of the rubber state variable temperature drying process.The simulation findings demonstrate that most of the test rice starch molecules can be converted from a glass state to a rubber state after 30 minutes of drying at65℃constant temperature.(2)The rubber state variable temperature drying test with varied heating rates was constructed based on the simulation findings of the COMSOL Multiphysics 6.0 program.The constant temperature and glass state variable temperature drying tests were also performed,and the results of all three tests were compared.The findings shown that,despite the glassy variable temperature drying process’slow drying pace,it could significantly lower the rate of rice cracking and guarantee the grain’s quality after drying.The cracking rate in the rubber state split-range variable temperature drying process was decreased by roughly64.7%compared to the constant temperature drying process at 60-65℃,and the rate of whole milled rice was raised by almost 112.5%.The protein and amylose content of rice before and after drying were similar to that of the glass state variable temperature drying,however the drying rate was higher and the drying period was shorter.(3)Six distinct neural network prediction models are created using the MATLAB program.The training data(90%)and test data(10%)for the neural network prediction model were gathered,and they relate to the drying temperature,crack rate,wet basis moisture content,and protein mass fraction of rice during the rubber state variable temperature drying process.The output layer included the crack rate,wet basis moisture content,and protein mass fraction.The input layer included the drying temperature change and initial quality parameters.The findings suggested that the LSTM neural network prediction model could be used to anticipate the drying of rice,but that its prediction error was higher than that of the BP neural network model.By including an optimization technique,the BP neural network model’s prediction error may be decreased and the fitting degree R~2can be kept between 0.96 and 0.99.