A Methodology for Creating Sequential Multi-Period Base- Case Scenarios for Large Data Sets

Abstract

Key performance indicators in engineering problems include but are not limited to financial, operational, management and environmental factors, which are significantly affected by aspects such as seasonality, fouling, economic climate, production rates, supply and demand. The search for an optimal solution to a problem must take into consideration this variability, otherwise running the risk of critical dimensioning or cost estimation errors. Testing solutions using full data sets covering large periods of time can be a computational challenge, and the analysis of results complicated. For the feasibility of such a study, it is therefore necessary to reduce the large data sets to a number of base case scenarios, which simultaneously reduce the number of data points to be handled while still representing the variability of the system. A novel method is therefore developed to address this problem. This method offers a way of designing an index of sequential periods common to each production level, which when averaged accurately represent periods of nominal values for each level. The method exploits a multi-objective evolutionary algorithm, minimising the standard deviation of the base cases compared to the real data as well as respecting crucial null value periods. Null value periods are typically found in turnarounds or supply and demand problems and are usually incorrectly represented in other methods. Lastly, the resulting base cases are sequential periods, which is important when dealing with scheduling, shutdown or storage problems. The method is tested using anonymised data and is compared to previously existing methods, with results showing improvement in the performance of the base cases with respect to the objective functions.