关键词: 水淹层, 自动识别, 过程神经网络, 判别模型, 测井资料, 数据转换

Abstract: A method based on process neural network for automatically identifying water-flooded formations in middle or late period of oilfield development was proposed.The process neural network consists of some process neurons and normal non-time variable neurons.It is a continuous neural network based on certain topological structures.The inputs and weights of process neural network may be the continuous process functions.The process neural network can automatically extract the forms and amplitude values of logging curves for small layers and combine the distinctions of many curves to generate the output type of water-flooded formation.In consideration of the practical log data being the discrete data of equal separation sampling changed with logging depth,a new method based on discrete Walsh conversion was used to convert the log data.The direct input from primordial log data to network was realized.The types of water-flooded formations in a region were determined, according to the analysis data of cored holes and interpretation results of experts.Because the sedimentary micro-facies and cycle types of small layer have some influences on water-flooding status of oil-bearing formation,the standard pattern database includes water flooding type of the oil-bearing formation,which assumes the sedimentary characteristics of formation.The process neural network identification model has good steadiness and strong generalization ability and is adaptable to identify the water-flooded formation. Practical process-ing of data obtained from seven wells in the north of Saertu of Daqing Oilfield shows a good result.

Key words: water-flooded formation, automatic identification, process neural network, discriminating model, log data, data conversion