Abstract:

Constrained by factors such as quality of seism ic data，lithology and structures，single seism ic attribute can only be used to predict reservoirs to a certain extent and there aremultiple possibilities.Through calibrating seism ic attributeswith well data，seism ic attribute fusion techniques can correlate oil/gas potentialwith seism ic attributes.For oil/gas potential prediction of reservoirs，mathematics-based multi-attribute fusion can avoid the ambiguity of single attribute reservoir interpretation.Back propagation（BP）neural network is very good at nonlinear fitting，but it is easy to get a localm inimum without convergence，influencing the accuracy of prediction.To solve this problem，particle swarm is adopted first to optim ize the network weight and threshold，and BP neural network is then used to differentiate reservoirs and nonreservoirs.The results are satisfactory.

Key words: particle swarm optimization, BP neural network, seismic attribute fusion technique, reservoir prediction