Intrigued by the sheer ingraspable possibilities in biological systems on the one side and the application of natural computation (artificial neural netword, genetic algorithms) methods to biologically oriented problems on the other side, we were motivated to apply soft computing methods to the simulation of niche evolution of populations and behavior of biological systems.
Especially groups of algae in aquatic environment drew our attention. In our simulation we focus on a subset known or observed behaviors. Abstracting from these behaviors, we envisioned a multi agent system (MAS) where single agents are modelled as reactive agents, i.e. stimulus response oriented reaction. Reactions of the agents are modeled by means of an ANN with a new kind of transfer function -- cubic splines transfer functions. Evolution of out steady state system is modeled by a GA with multiple chromosome encoding.
Basically our system is split into two static components and one dynamic component. The first static component models the environment of the agents, where parameters like light (intensity , distribution and periodicity) and substrate (amount, distribution and periodicity) may be set. So far we have not considred other parameters like temperature. The three dimensional world is mapped onto a two dimensional representation with boundaries, we call it aquarium. The user of our system may set the parameters free at will.
The second component is related to the individual agent. Agents are modelled by means of an ANN which receives a certain number of sensor values as inputs from their environment. Additionally the internal energy of an agent is fed into the ANN. The output of the ANN, i.e. the reaction of the agent, corresponds to the steering behavior and results in changing of the location of the agent or a stand still behavior. Nonmonotonic transformations of the output of a neuron in the ANN is achieved through so-called cubic-splines transfer functions.
The last component determines the dynamic behavior of the population. A GA is utilized to model the evolution of the polpulation in our steady state system. The ANNs of the agents are encoded in multiple chromosomes, where sexual/asexual reproduction generates additional individuals. Agents may evaporate through dissimination of their energy -- each actions requires a certain amount of energy.
At the time of writing we are mainly concerned with establishing the parameter ranges for the following experiment. We investigate the evolution of specialized subpopulations with predominance for external stimuli, i.e. light or substrate. We research niche building and specialization of the different metabolic pathways. In our model preferences for light or substrate are encoded into the genotypes of the agents. During the course of evolution we suspect to observe a grouping of the population into three groups. The first group specializes on transforming light into internal energy. The second group evolves preferences for substrate and an intermediate group reacts to both external influences.