A comparative study of neuro-fuzzy systems and regression analysis in the cost estimation of offshore structures

Abstract

Cost estimating for offshore oil and gas developments comprising floating structures is commonly regarded as a deterministic process that is based on the use of norms and rules of thumb on clearly defined quantities. In most cases, this requires immense effort, a huge amount of historical data from similar projects and detailed design of the proposed development, which is usually not available in the conceptual and appraisal stages of the project. It would be valuable to have a simulation model based on stochastic functions that map a set of input parameters to a set of outputs by translating the predetermined scope to quantities and then predictive costs.This could then be used in concept selection studies, sensitivity analysis and optimisation in the front-end and development planning stages of an offshore project.This paper compares the neuro-fuzzy modelling techniques with multiple linear regression analysis in the development of cost estimating relationships and predicting the cost of offshore structures in the conceptual phase. The results are benchmarked against actual project results, comparing the accuracy, effectiveness and practicality of the both methods.This study is a part of the author’s ongoing research into the development of a probabilistic cost engineering model for offshore structures at the University of Western Australia. It is expected that this research will contribute greatly to the development of decision support systems in offshore construction cost engineering and improve confidence levels in conceptual cost estimates, thus facilitating a comprehensive front-end evaluation of offshore developments especially for marginal fields.