Estimating the Value of Information Using Bayesian Optimization with Gaussian Process Surrogate Models - An Application to Failure Rates at Offshore Wind Farms

Abstract

Finding operation and maintenance (O&M) strategies that increase the profitability of an offshore wind farm is essential in order to be competitive with other sources of renewable energy. An O&M strategy is characterized by a set of decision variables, such as vessel fleet and the number of available technicians. The presence of uncertain variables that influence the profitability of an O&M strategy, such as varying weather conditions, makes the search for good strategies difficult. Information about the uncertain variables will help to find good strategies, but such knowledge often comes with a price. It is therefore of great interest to find out whether buying information is a worthwhile investment.

By utilizing a simulation tool, the profitability of different O&M strategies can be investigated. The set of possible strategies are, however, large so that only a small subset may be explored. Bayesian optimization with a Gaussian process surrogate model is therefore used to find favorable O&M strategies in a high dimensional input space. After obtaining these strategies, the value of information for an interrelated uncertain variable was estimated. The search for favorable O&M strategies and estimation of the value of information for the failure rate for a wind turbine component category is demonstrated in a relevant case from the offshore wind industry. The strategies identified from the Bayesian optimization appear reasonable and the estimated value of information suggests that information gathering could be worthwhile, depending on the price of the information.