神经网络在油田动态预测方面的应用

常规的动态预测方法因适应的开发阶段和范围不同,在应用过程中各有其局限性.BP网络则能克服这些缺点,不仅能描述油田开发的整个过程,而且还可以考虑单一变量和多变量影响因素,把能影响动态预测指标的各种因素自行组织起来,加以训练和学习,建立起广义的、精确的动态预测模型.在对各种BP网络改进方法进行全面综合研究的基础上,总结并提出了自组织优化学习因子方法,进一步增强了BP网络的自适应性能.同时依据BP网络的特点,提出了多变量预测技术,并将该技术应用到油藏数值模拟和油田开发规划之中,取得了较好的成果,为神经网络广泛应用于油田开发找到了突破口.

THE APPLICATION OF AN ARTIFICIAL NEURAL NETWORKS TO PREDICT PERFORMANCE PARAMETERS OF RESERVOIR

The regular dynamic prediction methods have respectively shown deficiencies because their applied development stages are different. However,BP networks,which may overcome that restriction can not only represent the whole development process,but also include signal variable and multiple variables that may influence the prediction. It can sleforganize all sorts of factors that have an effect on the behavior prediction,further selftrain and self-study to establish the comprehensive accurate model for dynamic prediction. Based on the various improved methods of BP networks,the methods of selforganized and optimized study factor are presented,which can enhance the adaptive capacity of BP networks. A prediction technique of multiple variables is put forward. Finally good effects are gained by applying this technology to numerical simulation of reservoir and oil field development plans as a beginning of the wide applications of neural networks technique to the area of oil field development.