Automation and Optimization of Working Speed and Depth in Agricultural Soil Tillage with a Model Predictive Control based on Machine Learning

While facing environmental challenges due to climate change, the need for optimization and automation of agricultural tasks is increasing. Furthermore, costs and the lack of qualified personnel require efficient and highly automated control systems for agricultural machinery.
Therefore, this work addresses these challenges by optimizing the working speed of a tractor and soil tillage implement combination to maintain efficient operating points during high power demands.
A system was developed that predicts a suitable working speed based on a draft force and traction model in combination with the usage of a neural network for fuel rate prediction. The machine operator is able to customize optimization parameters such as fuel efficiency, performance or total costs depending on the individual needs and situation. These parameters lead to a reward function to value the machines state. Based on these objectives the network is able to predict the system state for various potential target speeds and evaluate their optimization parameters to select the most promising target speed. This target speed gets received by the tractor and leads to a new machine state.
... mehr
The fuel rate prediction network is trained on previously collected training data. Using different methods, for example transfer learning, the network can be adapted easily to different sizes and types of tractors. As the draft force models are based on equations, they can be changed to adapt to turning and no-turning soil tillage.
To maintain a sufficient working quality and simplify online parametrization of draft force requirements, the implement working depth is automatically adjusted based on active Lidar measurements. The adjustments take the working conditions and agricultural requirements into account.
The system was validated during field measurements on different locations with various customized optimization parameters. The results show a suitable reaction to changing operating conditions.