Electric company resource planners face a range of important uncertainties when they develop future generation and transmission capacity expansion plans. Some of these uncertainties include future input fuel prices (e.g., natural gas), expected future wholesale power prices,cost and performance characteristics for potential new power generation and transmission assets and evolving regulatory requirements.
Most current resource planning analyses are based on deterministic modeling approaches that use a fixed set of input values and the key uncertainties typically are assessed exogenously using scenario analyses and/or sensitivity analyses. Since both the context for resource planning decisions and the future values of key input variables are characterized by deep uncertainty, it is important to explore how stochastic capacity expansion modeling approaches impact modeling outcomes and how this approach compares to deterministic modeling approaches.
This EPRI technical update report highlights the differences between deterministic and stochastic capacity expansion modelingapproaches and compares the results of using both approaches. For this analysis, we used the John Hopkins Stochastic Multi-stage Integrated Network Expansion (JHSMINE) model to incorporate a combination of scenarios as input variables to the model rather than a single set of fixed values. JHSMINE is astochastic capacity expansion model that maximizes or minimizes the objective function of a resource plan to a single solution, as a function of multiple-input decisions. Results show that the investment cost of plans using a stochastic model is generally higher than the deterministic model. However, when we iteratively ran the results from these models through a distribution of potential future fuel prices, the mean of the realized output costs of stochasticmodel-based solutions were lower than the deterministic model-based solution.
This technical document also provides details on high-performance computers, advanced desktop processors, and virtual machines that can be used for stochastic modeling. We also compare the computational performance of the stochastic model on an advanced desktop processor versus a commonly used laptop computer as a function of number of stochastic scenarios and representative hours in a year.