基于遗传BP神经网络的核桃破裂功预测模型

Rupture energy prediction model for walnut shell breaking based on genetic BP neural network

针对核桃壳破裂所需机械能易受核桃含水率、加载速度和体积级别等多种因素影响,提出一种核桃壳破裂功预测方法。以南疆地区温185核桃为研究对象,选择核桃含水率(4%、6%、8%、10%)、加载速度(100、200、300、400 mm/min)和横径级别(1、2、3、4级)3个因素作为BP神经网络模型的输入量,利用遗传算法优化神经网络的权值与阈值,建立温185核桃破壳破裂功的遗传BP神经网络预测模型。结果表明:遗传BP神经网络模型能较好表达温185核桃破壳破裂功与主控因素之间的非线性关系,预测结果与实测值之间的平均绝对百分比误差为0.035,测试样本的网络输出值与网络目标值的相关系数达0.92488,模型预测效果较佳。研究结果为温185核桃破壳取仁加工过程的在线监控提供参考依据。

Traditional operations of walnut harvest and breaking shell seriously affect the machining quality and efficiency in walnut processing. With continuous exploring areas and increasing output of walnut, developing deep-processing technology is of extremely vital significance to walnut industry. Walnut shell breaking is an important stage of walnut industrialization process. Mechanical force is widely used to obtain a large number of broken walnut kernels in walnut industrialization process. The properties of volume size, shell thickness and texture characteristics of walnuts greatly affect the process of obtaining the kernels. Walnut shell stress and deformation depends on the contents of H2O, volume of size and loading speed during breaking shell. The fruits of Wen 185 sorted by diameter （Divide 4 grades, reference walnut processing standards in Hetian） and H2O content （4%, 6%, 8% and 10%） were compressed by microcomputer-controlled machine using different loading rate （100, 200, 300 and 400 mm/min）. Meanwhile, force-deformation curves were analyzed and rupture energy were calculated. It was important to predict the walnut shell rupture energy for improving the design and development of walnut processing equipments. The back-propagating （BP） artificial neural network was an effective prediction model, which highlighted the characteristics of fast, accurate and better adaptability. However, the BP had the deficiencies of insufficient network global search ability, slow convergence and local optimum iteration. The remedy patterns of genetic algorithm that performed global searching would optimize the weights and thresholds in BP network, and thereby improve the accuracy of predictions. For Wen 185 walnut in southern Xinjiang, the H2O content, compression speeds, and transverse diameter were considered as the basic characteristic parameters for BP neural networks models. Genetic algorithm was used to optimize the weights and bias of BP neural work. Optimized BP neural network was applied to predict the rupture energy of walnut shell breaking. The genetic BP prediction neural network model was trained and tested with the experimental data collected from rupture energy. The results showed that the errors between predicted and tested results were small, and there was non-linear relationship between rupture energy and main controlling factors in the model which resulted from the genetic BP network. The correlation coefficient of the network output value between samples and BP network was 0.92488. The optimized BP neural network model had a stronger ability for nonlinear approach, which actually reflected the nonlinear relationship between the rupture energy of walnut shell breaking and main controlling factors. The predicted results from the genetic BP network were better than the back-propagating artificial neural network. Therefore, the genetic BP network is an effective method used for prediction of the rupture energy of walnut shell.