Intelligent Farmer Agent for Multi-agent Ecological Simulations Optimization

Abstract

Abstract. This paper presents the development of a bivalve farmer agent interacting with a realistic ecological simulation system. The purpose of the farmer agent is to determine the best combinations of bivalve seeding areas in a large region, maximizing the production without exceeding the total allowed seeding area. A system based on simulated annealing, tabu search, genetic algorithms and reinforcement learning, was developed to minimize the number of iterations required to unravel a semi-optimum solution by using customizable tactics. The farmer agent is part of a multi-agent system where several agents, representing human interaction with the coastal ecosystems, communicate with a realistic simulator developed especially for aquatic ecological simulations. The experiments performed revealed promising results in the field of optimization techniques and multi-agent systems applied to ecological simulations. The results obtained open many other possible uses of the simulation architecture with applications in industrial and ecological management problems, towards sustainable development.

Sistema Inteligente para Auxílio na Seleção e Execução de Tarefas num Jogo Social

“…O trabalho futuro passa pela melhoria da arquitetura do sistema mult i-agente utilizando o trabalho realizado em [19][20][21][22][23][24][25], e a melhoria da linguagem Farmlang [12,13,26]. A criação de u ma interface natural para permit ir u m jogo mais fácil [27][28][29] e a utilização de metodologias de aprendizage m [30][31][32][33] para a criação de perfis do utilizador [34][35] são também melhorias possíveis.…”

ECOSIMNET: A Framework For Ecological Simulations

“…A previous bivalve farmer agent was developed by (Cruz et al 2007) with the objective of finding the best combinations of locations to seed and harvest bivalve species within a delimited area of exploitation, using the previous model, towards bivalve production maximization. In that work the farmer agent had to choose the best 5 cells (corresponding each to a 500m x 500m area) within a large region area of 88 admissible cells to explore oysters -corresponding to more than 39 millions of possible combinations.…”

“…• Repeat the experiments made by (Cruz et al 2007) using 1 and using 3 simulators, and compare the results obtained in terms of temporal savings and quality of the final solutions achieved;…”

“…The experiments used the algorithms implemented by (Cruz et al 2007), controlled by a simulated annealing algorithm (Kirkpatrick et al 1983) modified to support parallelisation and with more flexibility to control the number of simulations in each experimental round.…”

“…The results of the development of an intelligent bivalve farmer agent to find out the best combination of bivalve seeding areas, within an allowed area of exploitation, were reported in (Cruz et al 2007) and showed promising guidelines. One of the constraints pointed out in the conclusions of that study was the time consumed by each simulation run.…”