Abstract:When predicting and estimating reservoirs,we usually use kinds of seismic attributions which are related to reservoir prediction.Those seismic attributions extracted from kinds of methods include abundant geologic information,but some attributions are likely to correlate with one another,which results in repetition and redundancies of information.So the infinite increase of the number of attribution brings bad effects on reservoir prediction.To special problem,picking the best attribution subset from the whole seismic attributions is necessary,that is the optimisation of seismic attributions,whose purpose is from numerous seismic attributions to choose small number of ones that are the most closely associated with and the most sensitive to the research target,and then we may carry out inversion of reservoir parameters of the target(such as porosity,content of mud,thickness of reservoir etc.)by using optimized seismic attributions.This paper mainly discusses the GA-BP method combining the Genetic Ahgorithm of seismic attributions optimisation with BP neural network,and we resulted in better geologic effect after applying it to the 3-D seismic data in LJF block of Dagang Oilfield.