论文部分内容阅读
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。
Conventional dynamic prediction methods have their limitations in the application process due to different development stages and scopes of adaptation. BP network can overcome these shortcomings, not only can describe the whole process of oilfield development, but also can consider single variables and multivariate influencing factors, which can affect the dynamic forecasting indicators of various factors to organize themselves, to be trained and learn to establish Generalized and accurate dynamic prediction model. Based on a comprehensive and comprehensive study of various BP network improvement methods, a self-organizing optimization learning factor method is summarized and proposed to further enhance the adaptive performance of the BP network. At the same time, based on the characteristics of BP network, a multivariable prediction technique is proposed. The technique is applied to reservoir numerical simulation and oilfield development planning. Good results have been obtained, finding a breakthrough for the neural network to be widely used in oilfield development.