基于变增益单神经元PID的秸秆旋埋还田导航系统研制

水稻田土壤松软,收割机作业后会出现残留秸秆凸起、地表坑洼等现象,导致秸秆旋埋还田作业易出现重耕、漏耕和自动驾驶路径跟踪精度差等问题。该研究基于滑移估计模型推导了拖拉机路径跟踪的前轮转角控制率,并设计了一种变增益单神经元PID导航控制器。在自主设计的电控比例液压转向系统基础上开发了秸秆旋埋还田导航系统,采用双天线RTK-GNSS获取拖拉机的实时位置和航向角信息,由变增益单神经元PID控制器根据理论转角和航向角偏差变化输出实际执行转角,实现旋埋作业自主路径跟踪。田间试验表明,作业速度为1.15m/s时,变增益单神经元PID控制器的自适应直线跟踪最大横向偏差不超过0.071 m,平均绝对偏差不超过0.031 m。与常规PID控制器相比,变增益单神经元PID控制器的最大横向偏差和平均绝对偏差控制精度分别提高了53.08%和51.72%;与单神经元PID控制器相比,最大横向偏差和平均绝对偏差控制精度分别提高了39.00%和28.21%。该研究设计的变增益单神经元PID控制器可以增强导航系统的适应性和鲁棒性,提高路径跟踪精度,适用于未来无人驾驶下的秸秆旋埋还田作业。

Development of rotary straw burying and returning navigation system based on variable-gain single-neuron PID

Soft-soil surface in a paddy field is usually left with unevenly distributed rice stalks, uncut stubbles, and machine ruts after harvesting by combines. A common treatment of rice straws is rotary burying using tractor-hitched rotary cultivators in southern China. However, misses and overlaps inevitably occur, because human tractor drivers mainly perform the current operation under the complex conditions of field surface. It is greatly urgent to develop an automatic navigation system for better operational efficiency and accuracy. In this study, a Dongfanghong LX954 tractor with a hitched rotary straw returning cultivator was taken as the research object. An automatic navigation system was developed to replace the original tractor-rotary cultivator combined one, with a novel electronic proportional control and hydraulic steering. The main hardware of the navigation system included an onboard PC, an electronic-control proportional hydraulic valve, a steering angle sensor, and a dual-antenna RTK-GNSS. The navigation software was performed on the MATLAB platform in the Windows 7 operating system. The RTK-GNSS was used to measure the real-time lateral deviations and orientation errors of the tractor. The onboard PC was used to plan the operation paths, process the GNSS measurement data, calculate the steering angles using the designed control algorithms, and finally send control commands to the steering controller. Specifically, the steering controller was used to receive the control commands through a CAN bus, thereby controlling the opening and closing of the electronic-control proportional hydraulic valve, and finally realizing automatic navigation. An estimation model of slip angle was derived using the steering angle control of the tractor, and a variable gain single-neuron PID controller was designed, in order to reduce the great slippages from the complex conditions of field surface during rotary burying of straws. The slip angles of the front-wheel and rear-wheel were added into the model to accurately calculate the control angle more suitable for the actual operation. The variable gain single-neuron PID controller output the values of steering angle by learning the previous control effects from the neuron and the adjusted gain, according to the differences between the current heading and the target heading. As such, the navigation control rapidly adapted to the subsequent soil surface in the field. MATLAB simulations were carried out to compare the control effects of a conventional, a single-neuron, and the variable gain single-neuron PID controller. The results showed that the variable gain single-neuron PID controller behaved the fastest convergence, the smallest overshoot, and the best performance of signal tracking. A road test was performed to further verify the feasibility of the designed navigation system. When the operation speed was about 2 m/s, the variable gain single-neuron PID effectively improved the performance of path tracking, in terms of speed and accuracy for both straight path and curve tracking. Furthermore, the maximum lateral deviation was 0.026 m during straight path tracking. In the curve tracking for a half circle with a diameter of 12 m, the maximum lateral deviation was 0.382 m and the mean was 0.170 m. A field experiment was performed on the rotary straw returning under the designed navigation system. In straight path tracking, the maximum lateral deviation was 0.071 m, and the mean was 0.031 m when the operation speed was 1.15 m/s. Compared with the traditional and single-neuron PID controllers, the control accuracy of the maximum error and mean absolute error were improved by 53.08% and 51.72% respectively, and compared with single neuron PID controller, the control accuracy of the maximum error and mean absolute error were improved by 39.00% and 28.21% respectively, the developed navigation controller can significantly reduce the interference from the various soft-soil surface in a paddy field, thereby enhancing the adaptability and robustness of the navigation system, particularly with the higher accuracy of path tracking.