Reliability Analysis of a Microgrid using Dynamic Bayesian Belief Networks

Reliability analysis is made more difficult by the variability and intermittency of renewable energy sources that accompany the ageing of components. Most of the existing approaches predict the probability of generation as a single-point value or presume a predefined parametric model for it. Their drawback lies in that they rarely grasp low-probability events which are actually vital to reliability analysis. To overcome this shortcoming, this paper proposes a data-driven sequential unit commitment using the dynamic Bayesian belief networks and makes it the internal model of system operation inside the Monte Carlo simulation for reliability analysis. In each time step, the dynamic Bayesian belief network first generates the probability distributions of renewable energy generation. Then a realization of renewable energy generation is sampled as the boundary of the unit commitment of the present time step. The unit commitment is solved, and its impact of operating the generators on their ageing conditions is accumulated. The dynamic Bayesian belief network prediction and the unit commitment in sequence traverse step by step throughout the time horizon, and form one bid of the Monte Carlo simulation. The convergence of the Monte Carlo simulation is measured by a popular reliability indicator, loss of load probability. The proposed method has its effectiveness justified in its comparison with a point estimation method.