中文摘要：

油田剩余油分布预测被国内外石油领域专家公认为世界难题，目前，其预测准确率低的根源在于或者只考虑部分客观证据、或者只考虑部分主观证据，导致对剩余油分布水淹类型等特征分类准确率低、可靠性差。所以，如何对来自多专业领域不同层次的全部客观证据及领域专家长期积累的主观证据进行融合．成为剩余油分布研究的核心问题。文章通过BP神经网络联合模型与两级D-S证据推理模型的优势互补进行主客观证据融合，实现了剩余油分布多属性特征的准确分类。提出了将BP神经网络分类结果的可信度及专家系统推理结论的可信度作为D—S证据推理模型输入证据基本概率赋值的有效方法。为各类多源信息融合系统的研究和工程实现提供了示例、途径和有益的经验。

英文摘要：

Oilfield remaining oil distribution forecast＇ are called world-level difficult problems by oil domain specialists in the world.The root of low forecast correctness are only consider objective evidences or subjective evidence,so the forecast results still exist larger limit,it result in low accurate rate,low reliability,slow run speed,low automation degree to identify the classification characteristics（e.g.flood type of remaining oil） and to compute quantitative parameters.So, how to fuse all objective evidences and subjective evidences is a key problem to research remaining oil distribution,the objective evidences come from many different specialty domains in different levels,the subjective evidences come from domain specialists.In this paper,a new model is proposed,which the model integrated BP neural networks combination models and two-level D-S evidence reasoning models,the model utilized the merits of two kind models and avoided their demerits,the exact classification is implemented about remaining oil distribution about many characteristics.A ef fective method is proposed,namely,the classification output reliability of each BP network and the reasoning result reliability of each domain fuzzy expert system are regarded as basic probability assignment of input evidence in D-S evidence reasoning model.The research results in this paper provided examples,approaches and useful experiences to the research of multiple sources information fusion system in different types.