关键词: BP网络, SOM网络, 储层识别, 测井数据, 储层参数, 预测

Abstract:

Xiefengqiao anticlinal structure in southwestern margin of Jianghan Basin is a litho structural complex oil reservoir.Its oil accumulation was controlled by vertical and lateral distribution and heterogeneity of the reservoir.Two kinds of neural networks—forward network(BP) and self organizing feature mapping network(SOM)—were used in reservoir parameter simulated calculation and horizon type prediction respectively.Then chose the data of Esheng 4 and Esheng 8 Wells as training samples,used drilling and log data such as RXO,RT,GR,SP,AC,CNL and CAL as input variables to set up input/ouput mapping relation between log data and POR,So and K parameters of the reservoir,used BP network to make function approximation,used SOM network to make pattern sorting.Then two parameters RT and AC of the reservoirs were reinputted,the parameters,together with POR,So and oil saturation K outputted from BP networks were taken as characteristic parameters of the reservoir identification.The characteristic parameters were standared and trained by SOM networks so as to get model sample,then inputted the data of oil horizon that needed to be predicted,the result of oil horizon identification could be outputted from SOM networks.The validity is proved to be more than 90%.

Key words: BP network, SOM network, reservoir identification,logging data,reservoir parameter, prediction