摘要
为了建立预测烤烟钾氯比的BP神经网络,以影响烤烟钾氯比的可度量因素作为网络输入,并通过经验方法选取合适的网络参数,采用粒子群算法优化神经网络的初始权值和阈值,建立最佳的神经网络预测模型。结果表明:优化后的神经网络,测试样本中预测值与期望值的总相关系数为0.97155,提高了13.77%,误差在[-1,1]的样本占86.67%。该模型拟合能力较好,有一定的预测能力,泛化能力得到了明显提高,有助于评价烟叶的燃烧性和品质。
An optimized model of BP neural network for predicting the ratio of potassium to chlorine （K/Cl ratio） in flue-cured tobacco was established by using the measurable factors influencing K/C1 ratio as network inputs, selecting appropriate network parameters with empirical method, and optimizing initial weights and thresholds of neural network with particle swarm algorithm. The results showed that the total correlation coefficient between the prediction value of the test sample and the expected value was 0.97155, which increased by 13.77%, and the samples with the error within [-1, 1] accounted for 86.67%. This model features good fitting ability, certain predictive ability, and significantly improved generalization ability, it is helpful to the evaluation of smoking quality and combustibility of tobacco leaves.
出处
《烟草科技》
EI
CAS
北大核心
2014年第6期49-53,共5页
Tobacco Science & Technology
基金
曲靖烟草公司项目"烟草工商研共建原料与品牌协同发展机制(ESTB)项目"(110200801036)
"曲靖市‘天香’特色优质烟叶研究与开发"(09YN025)
关键词
烤烟
钾氯比
BP神经网络
粒子群
初始权值
初始阈值
Flue-cured tobacco
Ratio of potassium to chlorine
BP neural network
Particle swarm
Initialweight
Initial threshold
