Bayesian separation of wind power generation signals

One of most challenging and important tasks for electricity grid operators and utility companies is to predict and estimate the precise energy consumption and generation of individual households which have their own decentralized production system. This is a under-determined source separation problem since only the difference between energy production and consumption in the micro-generation system is visible. Therefore, we present a latent variable model with a polynomial regression form for the separation and then the model is used by several statistical algorithms to explore the underlying energy consumption and production from the differenced signals. In order to efficiently find global optima of the hidden variables of the model, we develop a source separation algorithm based on the Integrated Nested Laplace Approximation (INLA).