基于BP神经网络的整株秸秆还田机功耗优化

Optimize on power dissipation of whole-straw returning device based on BP neural network

为解决利用回归分析法优化水稻整株秸秆还田机功耗时存在的拟合误差精度差和准确性低等缺陷,提出一种高精度和高准确性的基于BP神经网络的优化方法.本文以1ZT-210型水稻整株秸秆还田机为研究对象,选取机具前进速度,刀辊转速,刀具安装角为试验因素,还田机功耗为影响指标,以二次正交旋转组合试验数据为训练样本,获得功耗与影响因素的BP神经网络模型;并季利用提出的方法对其进行优化,获得功耗影响因素的最佳参数组合为：机具前进速度1.39 km/h,刀辊转速210 rpm,刀具安装角55°,该参数组合下还田机的最小功耗为9.21 kW.试验条件下还田机最小功耗优于回归分析法所得最小功耗10.56 kW,以BP神经网络优化结果进行验证试验,试验测得功耗值9.42 kW,与BP神经网络优化结果绝对误差为0.21 kW,相对误差为2.28%.试验结果表明：该优化方法实用性强,拟合精度高,优化结果准确稳定,为解决农业工程领域中类似优化问题提供了一种新方法.

In order to overcome the bad precision of fitted error, low veracity and other flaws when the power dissipation of whole-straw returning device was optimized by using the regression analysis method, a high precision and high veracity optimization method based on back-propagation（BP） neural network was proposed. Taking 1 ZT-210 rice straw whole straw returning device as the research objective, the testing program of three factors, five level was designed by using the orthogonal rotation method, which selected the forward velocity of device, rotate speed of knife roll and established angle of knife as experimental factors and power dissipation as influence index. The field experiment was carried out the Heilongjiang institute of agricultural mechanical engineering science according the testing program and the experimental data was obtained. The BP neural network was used to fit the experimental data, and the mathematic model of power dissipation with influence factors was established. Then, the optimal parameter combination of influence factors could be obtained by the proposed method. The optimum combination as follows： forward velocity of device is 1.39 km/h, rotate speed of knife roll is 210 rpm, and established angle is 55°, the minimum power dissipation is 9.21 kW. Comparing the result with the regression analysis method, it is better than the 10.56 kW obtained by regression analysis method. In order to check the veracity of optimization result based on BP neural network, the confirmatory experiment was carried out which selected the optimization result as testing program. The power dissipation is 9.42 kW, the absolute error is 0.21 kW and the relative error is 2.28% between the experimental result with optimization result. The confirmatory experimental was shown that the experimental result is consistent with optimization result, and the proposed method can obtain better fitting precision, higher practicability and more stable optimization result. It is a stable and feasible optimization method and offers a new method to solve the similar optimization problem in field of agriculture production.