Improved Modeling of Unit Commitment Decisions under Uncertainty

The massive integration of variable and limitedly predictable electricity generation from renewable energy sources (RES) leads to so-called balancing costs (a.o. integration costs). Therefore, system operators are continuously seeking novel sources of operational flexibility and improved methods to size, procure and deploy the associated operational reserves. With these challenges is mind, we develop an operational modeling framework to study the impact of stochastic RES-based generation, novel flexibility providers and demand response on day-to-day electricity generation system operation. This framework consists of a statistical characterization of the uncertainty, which provides input for the generation of scenarios and reserve sizing techniques, which in turn are representations of the uncertainty in so-called unit commitment (UC) models. Historically, these models were typically deterministic in nature, which may make them poorly suited to study and operate power systems with high shares of limitedly predictable RES-based generation. Therefore, the first objective set forth in this dissertation relates to the improvement of existing and the development of novel UC models considering imperfect RES-based electricity generation forecasts, with a focus on the cost-efficiency, reliability and computational effort associated with their solutions. We study and improve the performance of three UC models found in the scientific literature: a deterministic UC model, a stochastic UC model and an improved interval UC model. A deterministic UC model tackles uncertainty by considering reserve constraints, which trigger -- if properly designed -- sufficient scheduled capacity to absorb forecast errors during real-time operation. The cost-efficiency of the deterministic model is improved by (1) the explicit consideration of energy storage-based reserve provision and (2) state-of-the-art probabilistic reserve sizing techniques, based on a novel distributional characterization of the uncertainty on RES-based generation forecasts. Stochastic models on the other hand employ a direct scenario-based representation of the uncertainty, which in theory leads to more cost-efficient UC schedules. The solution stability and bias of the resulting UC schedules have drastically been improved through the development of a dedicated scenario reduction technique. The improved interval UC model attempts to reduce the conservatism of the deterministic UC model by improving the realism of the ramping requirements imposed on the scheduled reserves. Nevertheless, the presented qualitative and quantitative analysis showed that the deterministic and improved interval models yield sub-optimal, overly conservative UC schedules, albeit at a low computational cost. We will attribute this sub-optimal behavior to the inability of these models to account for the expected deployment cost of the scheduled reserves. In contrast, the stochastic UC model results in cost-optimal UC schedules if the selected scenarios are sufficiently representative of the uncertainty at hand, but at an extremely high computational cost. In pursuit of a UC model that combines the cost-optimality of the stochastic model with the computational effort of the deterministic model, we develop two novel UC formulations: a hybrid deterministic-stochastic and a probabilistic UC model. Both formulations include, although through a different approach, an approximation of the expected deployment cost of the scheduled reserves. The hybrid UC model combines a probabilistic reserve requirement and a limited set of scenarios. This model allows approximating the stable solution of the stochastic UC model, but at a computational cost that is an order of magnitude lower than that of the original stochastic problem. The probabilistic model on the other hand is characterized by calculation times similar to that of the deterministic model, but allows for significant cost reductions through the internalization of the reserve sizing problem. The second objective addressed in this dissertation relates to the study of the arbitrage and regulation services an activated demand side may offer. In particular, we study the system value of demand response with electric heating systems. To this end, an integrated model is proposed, which entails the inclusion of a physical demand side model, sufficient to represent the operational flexibility available in limitedly controllable residential electric heating systems, in a probabilistic UC model. In a numerical case study, inspired by the Belgian power system, demand response-based arbitrage and regulation services are shown to contribute significantly to the minimization of the balancing cost associated with imperfect RES-based generation forecasts. The presented models and techniques can be used to assess the impact of uncertainty on reasonably large electric power systems, as illustrated in the last chapter of this dissertation. Independent system operators may use these models to optimize their UC decisions taking into account the uncertainty in their system. The integrated model may lead to adequate estimates of the system value of DR. In addition, demand aggregators may use the presented approach to optimize the scheduling and operation of DR-adherent loads.status: publishe