SIMULATED NEUROCONTROL OF AN AUTOGENOUS MILL WITH EVOLUTIONARY REINFORCEMENT LEARNING J.W. de V. Groenewald 1 , C. Aldrich 1, *, J.J. Eksteen 1 , A.v.E. Conradie 1 and L.P. Coetzer 2 1 Department of Process Engineering, University of Stellenbosch, Private Bag X1, Matieland 7602, South Africa, Fax +27(21)8082059, *E-mail: ca1@sun.ac.za (corresponding author) 2 Anglo Platinum Management Services, Anglo Platinum, South Africa. Abstract: In this investigation the development of nonlinear control system for an autogenous mill was considered. A symbiotic adaptive neuroevolution algorithm was used in conjunction with a dynamic multilayer perceptron model fitted to actual plant data to evolve neurocontrol systems. Simulation studies established the potential of the approach, which yielded satisfactory results, despite having had to learn from a model that covered part of the state space only. Copyright © 2007 IFAC Keywords: Neural Networks, Nonlinearity, Neural Control, Time Series Analysis, System Identification. 1. BACKGROUND Autogenous mills operate on the same principles as ball mills, except that grinding is accomplished though the impact of larger pieces of material on smaller pieces of equal density. For successful autogenous milling, it is thus essential that the top size range of the ore is capable of generating a grinding load able to crush the finer portion of the feed, as well as its own progeny. At the same time, the large lumps in the feed should not be so durable that it cannot be reduced at a rate equal to that at which feed is entering the mill. The principal variables that affect the control of an autogenous mill are ore feed rate, water addition rate, ore feed hardness, and ore feed size analysis. Ore feed rate and water addition rate are variables that may be deliberately manipulated in order to control the process. In contrast, ore feed hardness and feed particle size distribution are disturbance variables, causing major problems when attempting to run a stable mill under optimal conditions (Napier-Munn et al., 1999). Although grinding circuits exhibit nonlinear dynamic behaviour, controller design has largely been invest- tigated from a linear controller perspective (e.g. PID control) that requires the use of linear process models. As rigorous grinding circuit models are invariably nonlinear, the nonlinear models are typically linea- rized in the vicinity of a predetermined economically optimal operating region. Although these linear MIMO control strategies have been applied with suc- cess in industrial grinding circuits (Herbst and Rajamani, 1979), linearization becomes ineffective in the presence of severe process nonlinearities. In this paper, it is shown by way of simulation studies that an advanced control scheme based on evolutio- nary reinforcement learning from a dynamic model fitted to historic process data can lead to significantly improved control of an industrial autogenous mill. The paper is organized as follows. In section 2, the principles of evolutionary reinforcement learning are considered. In section 3, identification of an industrial mill is explained and in section 4, the results of the simulation study are presented, followed by some concluding remarks in section 5.