Abstract:

In order to solve the problems of inaccurate prediction of rice moisture content， large hysteresis and low coefficient of determination during the drying process， the response surface methodology （RSM） was used to analyze the main factors affecting the rice moisture content during drying. Genetic algorithm （GA） was used to optimize the traditional BP （back propagation） neural network model， and a prediction model was established based on RSM-GA， which could accurately predict the moisture content of grain during drying. The results showed that the hot air temperature， grain layer temperature and ambient relative humidity had significant effects on the change of moisture content during drying process of rice. The hot air temperature， grain layer temperature and ambient relative humidity were used as the input layers of the prediction model， the rice moisture content was the output layer. The optimal number of intermediate hidden layers of the prediction model was determined to be 10 through empirical formulas， and the neuron structure of the prediction model established was 3-10-1. During model training， the optimal performance was at the 15 th time， the minimum root mean square error was 0.621 84×10^-3， and the optimal Matlab simulation test setting parameters were obtained. When the iterating was to 200 generations， the fitness value was stabilized at 0.019 5. After optimized by genetic algorithm， the coefficient of determination of the prediction model was 0.980， which was 5% higher than the traditional model； the root mean square error was 0.009， which was 17% lower than the traditional model. In summary， the performance of optimized neural network model was improved， which provided a reference for subsequent control strategy research.

Key words: rice, response surface method, moisture content prediction, genetic algorithm, BP neural network

摘要：

为解决稻谷在干燥过程中含水率预测不准确、滞后性较大以及决定系数不高等问题，采用响应面（response surface methodology，RSM）法分析干燥过程中影响稻谷含水率的主要因素，利用遗传算法（genetic algorithm，GA）对传统BP（back propagation）神经网络模型进行优化，欲建立一种基于RSM-GA的稻谷含水率预测模型。结果表明，热风温度、粮层温度、环境相对湿度在干燥过程中对稻谷含水率影响显著。以热风温度、粮层温度、环境相对湿度为预测模型的输入层，稻谷含水率为输出层，通过经验公式确定预测模型的最优中间隐含层数为10，由此建立预测模型的神经元结构为3-10-1。在进行模型训练时，最优性能表现在第15次，最小均方根误差为0.621 84×10^-3，得到最优的Matlab仿真试验设置参数，当迭代至200代时，适应度值在0.019 5时趋于稳定。经过遗传算法优化后的预测模型决定系数为0.980，较传统模型提高5%；均方根误差为0.009，降低17%。综上，优化后的神经网络模型性能提高，为后续控制策略研究提供参考。

关键词: 稻谷, 响应面法, 含水率预测, 遗传算法, BP神经网络