452419 Extensive Sensitivity Analysis and Stochastic Global Optimization for Renewable Energy Businesses Under Operational Level Uncertainties
By processing biomass in biochemical or thermochemical pathways, renewable energy facilities are able to produce fuels, value-added chemicals, heat and bioelectricity. First and second generation biorefineries are already operating worldwide, and process systems engineering has contributed over the past few years in addressing the complexity of modelling and decision making to optimize and make cost-effective renewable energy projects. Usually, the optimization of renewable energy businesses considers a deterministic design approach in which all the model parameters are assumed to be known from literature. However, during conceptual design there is lack of information and reliability, especially in novel or recently developed processes. When ideality supposed, external factors that might affect the process behavior and the project profitability are mistreated. Underestimating these limitations may lead into non-optimal designs and generate extra expenses or difficulties during startup and operation. Technological risk management should be considered during decision making since it permits to select optimal conditions when uncertainties are taken into account. Therefore, the global optimization problem should be addressed considering risk management at technological level in initial stages. It will allow managers to have a conservative and reliable method for making and supporting critical decisions. To identify significant uncertainty sources in the process, a sensitivity analysis of the parameters is required.
Our proposed framework is composed of different sections. First, a global sensitivity analysis based on Sobol’s method is performed to identify the significant sources of uncertainty in the kinetic parameters involved in the biological reactions of biofuels and value-added chemicals production. The method is computationally efficient since it averages the results of the evaluated functions and increases the original size of the sample. The first and total sensitivity indices are obtained by calculating the total variance and the individual variances generated when each parameter changes separately. One of the main characteristics of the current work is to explore the efficiency of two different approaches for selecting significant uncertain parameters: 1) evaluating all parameters simultaneously, 2) evaluating parameters independently per product pathway. Then, the identified uncertain parameters are utilized to optimize the profitability of the plant under uncertainty. A stochastic global optimization technique (ParLMSRBF-R), similar to a metaheuristic method, is used to maximize the cash flow after tax of a renewable energy business while considering uncertainty and risk management. In this iterative optimization approach the decision variables are selected intelligently with the help of a Radial Basis Function as the response surface or surrogate model. Finally, the optimal points obtained are compared statistically with optimal values from other methods, e.g. Monte-Carlo based optimization. To evaluate the efficacy of this novel decision-making framework, a hypothetical multiproduct lignocellulosic biorefinery is used as a case study.
This optimization framework appears to be more effective than other optimization procedures, e.g. Monte-Carlo based optimization. The algorithm searches for the optimal operating conditions using a fitted and continuously updating Radial Basis Function without losing the clarity of the actual mechanism of the process. The algorithm selects candidate points to be evaluated near the current best solution neighbourhood achieving exploration and exploitation in the solution domain. In terms of computational effort, the optimization algorithm employed requires less expensive function evaluations to reach global optima. In fact, ParLMSRBF-R achieves optima with less function evaluations in comparison to Monte Carlo-based optimization approach. In addition, the objective value improves in contrast to the best value achieved with Monte-Carlo based optimization.