APPLYING OF THE PARTIAL LEAST SQUARES REGRESSION AND NEURAL NETWORK MODEL TO DISTINGUISHING RESEVOIR AND PREDICTING PRODUCT IN THE GASFIELD ——An case study from the MA5_1 Member in the central gasfield of ShangXi-GanSu-NingXia basin

It is difficult to accurately distinguish gas, water and dry layer in tight reservoir and predict product.This paper proposes a partial least squares regression and neural network model(PLSNN) to solve the problem. The basic thought is pre-disposing auto variable by using the partial least square regression and forming BP model of distinguishing reservoir and predicting product by using VLBP and AMOBP.92 samples of 19 testing wells from the MA5_1 Member in the central gasfield of Shangxi-Gansu-Ningxia basin were selected from which 14 characteristic parameters were extracted and 5 main characteristic parameters of Rlld,Δt, kh, KΦ_S and EE were obtained by the partial least squares regression.BP model was established by using the 5 characteristic parameters as input variable and making the reservoir type and product level as output variable. The result indicated that the accuracy rate is 100% and the equel error is 50% lower than the traditional three BP network, and the model has fast rate of convergence and high precision.Therefore, PLSNN is a new method for distinguishing tight resevoir.