ABSTRACT: today, methodologies for advanced control and optimisation (expert systems, ARM or neural predictors, process on-line simulators) are surely useful for a wide fraction of industrial requirements, but they have serious limitations in many real field applications. One of the most serious problems for some innovative methods based on learning is that they are built on fixed optimisation rules and that they do not care about the evolution of the system during its life (i.e. not controlled variables or constraints), in other words they cannot adapt to a changing environment. Moreover the learning stage is generally difficult for data lacking and development activities for process optimisation require deep knowledge of the specific process.

The ideas proposed in this work are aimed to curry out a new approach, based on artificial life (alife) environments, for on-line adaptive optimisation of complex processes. The basic features of the proposed approach are: no intensive modelling (continuos learning directly from measurements) and capability to follow the system evolution (adaptation to environmental changes). The basic idea consists in the reversal of the concept of the knowledge based systems (KS). During the construction of a KS, the knowledge of the operators is verbally transferred to the KS builder. In our proposal, the process knowledge is not verbally transferred, but it is directly developed by the system through measurements observations. The driving process is the dynamic building of a model on the basis of the observations of the effects that the regulation actions have on the system performance. The essence of this approach could be synthesised by the following sentence: "not control rules but autonomous structures able to dynamically adapt and generate optimised-control rules".

We tested the proposed methodology on a very well known and widely studied dynamical system : the Chua’s circuit. It has the advantage to be a pretty simple system to implement and it has a wide range of chaotic regimes depending on the value of the parameters defining the circuit. On this testbed we defined a fitness function to maximise. Experimentation concerned the on-line optimisation of Chua’s circuit in different regimes. We suppose not to know the equations describing the circuit, so we have been working on the rough signal generated by a simulator of the circuit and our system builds an internal model by implementing a continuos learning. To test the adaptive capability of the alife environment we considered one parameter affected by unknown changes and then we let the alife environment to try to adapt itself to the new condition. Preliminary results are very promising and they show the system is able to dynamically adapt to slow environmental changes by recovering the optimal condition.

Keywords: Artificial Life, Adaptive Systems, On-line Optimisation, Evolutionary Computation, Continuos Learning, Dynamical Systems, Chua’s Circuit