Publications

This paper followed a probabilistic modeling workflow to incorporate reservoir uncertainty by inputting a range of values for each parameter instead of one deterministic value. This workflow aims to find several equiprobable history-matched models to capture the probable reservoir conditions and then find the production forecast range. Artificial intelligence algorithms, i.e., Particle Swarm Optimization (PSO) and k-means clustering, are utilized to assist during the modeling. PSO is one of the optimization methods that aims to find the minimum objective function. In reservoir modeling, the optimization method is used to minimize the objective function of the difference between the model’s calculation and the historical data. In the probabilistic approach, our objective is to find several unique history-matched model variants from the range of inputted data. However, the optimization algorithm such as the PSO tends to converge or be trapped into one solution and may ignore other probable variants. This paper proposes additional procedures to capture the uncertainty range better. To capture more variants, the PSO is modified by introducing a process that reruns the PSO but shrinks the search area. This new shrunk area is chosen based on the density of variants in the whole search area from the previous run. The new search area improves PSO’s probability of finding the missed variants. Another improved process is adding a hierarchical k-means clustering method to filter similar variants. By introducing this process, all variants are clustered into several groups. Some representatives from each group are picked to be carried over into the forecast step. The improved workflow is applied first to the analytic functions and then to the actual oil fields. This study shows that introducing sequential shrinking in Particle Swarm Optimization (PSO) and hierarchical clustering procedures can improve the capability of the probabilistic method to capture the range of production forecast.