Abstract: The method of design and implementation for parallel genetic algorithm is based on thick grain, with the multi-Agent combined with genetic evolution technology, to optimize the seismic attribution selection. A multi-Agent system includes two kinds of Agents:件叔ent and MAgent, they can exchange the useful information which can represent the current situation of evolution. The methods arc benefit to improve the performance of parallel genetic algorithm, and to raise the searching efficiency of the parallel genetic algorithm. The projecting feature in the parallel module is less communicating overhead. It was used in the optimization of oil exploration attribute selection and the result is good.

Key words: Agent, Attribute optimization, Uenetic algorithm, Parallel