多模型的油藏模拟自动历史拟合方法研究

传统油气藏数值模拟通常仅建立单个随机地质模型，采用人工历史拟合方式获得符合油藏动态的地质模型并用于方案预测。但由于地质资料相对稀少且地层非均质性的客观事实存在，历史拟合问题必然存在多解性，所获得的单个地质模型无法保证准确反映地下真实情况。本次研究中首先利用静态地质资料生产大量随机实现，使用PCA降维方法减少模型数据量，再利用聚类方法挑选出多个特征各异的实现作为初始模型，采用基于SPSA算法的自动历史拟合方法，获得多个符合油藏动态却又包含不同特征的历史拟合模型。结果表明，多模型能够更接近地下的真实情况，产生的预测结果不再是单一动态曲线，而是具有多种开发可能性的一系列曲线，这样使预测更为科学可靠。

Research on Automatic History Matching Method Based on Multi Models

In traditional reservoir numerical simulation, only a single random geological model is established, and the artificial history fitting method is used to obtain the geological model that conforms to reservoir production history and is used for project prediction. However, due to the relative scarcity of geological data and the heterogeneity of reservoir, the historical fitting problem must have multiple solutions, and the single geological model obtained cannot guarantee the accurate reflection of the real underground situation. In the numerical simulation study in this research, static geological data are used to produce a large number of random implementations, PCA dimensionality reduction method is used to reduce the amount of model data, and then the clustering method is used to select a number of implementations with different characteristics as the initial model. The automatic history fitting method based on SPSA algorithm was used to obtain several historical fitting models which conform to reservoir dynamics but contain different characteristics. Multi models can reflect the real underground situation more completely, and the prediction result is no longer a single dynamic curve, but a series of curves with multiple development possibilities, making the prediction more scientific and reliable.