Imitate and optimize modern control algorithms for forestry cranes by means of artificial neural networks

Abstract

Modern hydrostatic function drives for agricultural and forestry machines require complex control algorithms. Electric controls offer significant energy and control advantages over the state of the art, such as reduced tendency to oscillate or implementation of a variable power limitation. Therefore, new algorithms are essential for sustainable optimization of future machines. The paper investigates a method to automatically transfer an existing control algorithm to an artificial neural network (ANN), which will be optimized by the Pattern Search algorithm afterwards. The method was applied to a forestry crane with an electro-hydraulic flow-on-demand control. After 41 generations of optimized parameter sets, the ANN control already shows a behavior comparable to the reference control. With this approach it is possible to transfer deterministic algorithms into stochastic algorithms with comparable transfer functions, which can then be optimized using machine learning methods.