Operational Optimization of Distributed Solar PV Installations, Another Application in the Machine Learning World
With Solar PV systems becoming more accessible, and overtaking the Levelized Cost of Energy at the retail level. Distributed Energy Resources have gone from a dream to a reality in many sectors of society. As with every industry, the emergence of a disruptive technology. We observe, that there has been a continuous evolution of the industry, whereby researchers have gone all out to develop mechanisms to trade the energy in this space. Solar Electricity Generation is very dependent on weather patterns and this makes it very difficult to forecast, which adds a high degree of uncertainty to energy production, and translates to what it's known as noise.
Having said this, there are many models currently being exploited for the forecasting of energy production from PV panels. There are also many techniques that can assist in the prediction and subsequently the decisions making process of dispatching the energy. One such technique is the Auto-regressive Integrated Moving Average Model (ARIMA), then you have, Fast Fourier Transforms, Kalman Filters and Dynamic Bayesian Models. However, and from our experience, the Sticky Dirichlet Process in combination with Hidden Markov Models is the option that best captures the data, generating very accurate results. One of the many advantages of Dynamic Bayesian Modeling is that it allows to associate the data in space and time. A critical element when you are evaluating and subsequently proposing a solution to optimize systems where the data is generated in the form of a time series. We achieved this, through a what it is known as Particle filtering, and the application of the Baum Welch algorithm.
The conclusion drawn from our simulation provides additional insight in the development of products that will allow Mesh Networks and Distributed Energy Resources find avenues that will optimize the energy dispatching and reducing the disruptions caused by having high percentages of Distributed Energy Resources penetration.
