History matching field performance is a complex and time-consuming process yet it is a crucial step towards constructing reliable predictive models. This paper presents a novel approach where visual analytics plays a role in optimizing history matching process when used as an input to Assisted History matching applications (AHM). Silhouetted Clustering (graphical representation) is the technique used to interpret and validate consistency within clusters. This technique is using k-mean algorithm, which is an Unsupervised Algorithm.
The methodology used starts with identifying the optimal number of clusters of reservoir model grid cells using Silhouetted Clustering. This was done by discovering the reservoir geo-bodies and classifying them based on patterns of static rock properties. Once the optimal number of clusters is found, sensitivity are then formulated to run on all generated clusters maintaining a property multiplier value throughout the runs. That is to detect the most effective clusters on matching field historical performance. The last step is performing sensitivity runs on the most effective group of clusters using different ranges of the property multipliers until a satisfactory match is found.
The new methodology shows a significant improvement in the matching quality of the reservoir model as a result of the cohesive clusters. Those unified clusters helped to, easily, pinpoint the candidate cluster for property modifier. This greatly improved the quality of history matching the reservoir without affecting its geo-bodies nature. The main challenge is to decide on the right number of clusters that mimics the natural grouping and distribution of the reservoir properties. The optimal number of clusters is selected based on silhouette average value, clusters' density and well separation. Therefore, Silhouetted Clustering was used to validate the cohesion within the reservoir clusters in minutes.
The novelty in this approach is introducing Silhouetted Clustering, a visual analytics technique, to optimize AHM process where we can ensure graphically how well the grid cells classified within its cluster. Moreover, this technique provides the confidence while working with the clusters knowing that the natural pattern of the reservoir is honored. The integration between Artificial Intelligence and Visual Analytics has greatly contributed to improve the overall quality of simulation models in less time.