为确保双级振动精密排种器工作时在充种均匀的前提下实现连续播种, 设计智能定量供种系统.为提高定量供种精度, 基于BP神经网络对勺式外槽轮供种装置建立定量供种预测模型, 建立隐层结点数为6的神经网络模型.BP网络训练结果表明, 当网络模型训练步数为71步时, 网络的均方误差为4.61×10-5, 小于设定值5×10-5;采用16个理论供种模型样本与测试样本进行BP网络测试, 结果表明, 基于BP神经网络预测模型仿真得到的预测值相对误差较小, 其精度高于理论供种模型的精度, 且神经网络相对误差均小于5％, 获得的样本误差平方和为5.59×10-4, 小于设定目标值8×10-4, 满足预先设定要求;最后, 利用建立的定量供种预测模型, 对4种不同千粒重的超级稻种子进行仿真, 得到振幅分别为0、5、10、15μm下的排种轮转速与供种量关系, 该研究结果可为确定定量供种器的工作参数提供理论依据.%In order to realize the continuous seeding of duble-vibrating precision seed meter under the premise of uniform filling, the intelligent quantitative supply seed system was designed.In order to improve the precision of quantitative supply seed, a prediction model of quantitative supply seed was established based on BP neural network for the spoon-type outer groove wheel seeding device.After sample data preprocessing and network initialization, a neural network model with hidden node number 6 was established, and then BP network training was performed.The results showed that when the network model training step reached 71 steps, the mean square error of the network was 4.61×10-5, less than the set value of 5×10-5, which met the requirements.For the established network model, a total of 16 test samples were tested in 4 groups.The results showed that the relative error of the predicted values based on the BP neural network prediction model was smaller, and the accuracy was higher than the theoretical model, and the relative error of the neural network was less than 5％, the obtained square error of the sample error was 5.59×10-4, which was less than the set target value of 8×10-4, which satisfied the preset requirement.Finally, using the established quantitative seeding prediction model, four different 1 000-grain super rice seeds were simulated to obtain the relationship between the seed wheel rotation speed and the supply seed quantity with amplitudes of 0, 5, 10 and 15 μm.The research results can provide a basis for determining the working parameters of the quantitative seeder.
展开▼
