Wind power variability model Part I - Foundations

A novel technique to account wind variability is presented based on Markov Chains and classification of observations. This model describes the power system status through combination of cases or “snapshots of the network” obtained from the clustering of observations. The occurrence and variability of each case is modeled by a Markov Chain. This approach models the non-linear conventional behavior of the farms but also events that rarely happens but that they have a high impact in system reliability and stability (such as sudden disconnection of generators due to grid perturbations, swift change in wind during storms, etc). This model requires running just as many power flows as states has the system and it allows to derive easily and rigorously the probability of events. Moreover, the regulation of spinning reserve or reactive control can be easily optimized using Markov Decision Processes. The use of network snapshots allows to use a full model of the grid (instead of linear models). Intermediate cases are interpolated using fuzzy clustering, reducing the required number of states for a given accuracy. To explain adequately the foundations and to show the potential applications of this approach, this work has been divided in three parts. In this part, the theoretical foundations and an overview of the method are presented. The second part shows the estimation of Markov Parameters for a system with three wind farms. The third part illustrates the stochastic power flow of the three wind farms and introduces the possible optimization through Markov Decision Processes.