Abstract
With the increasing use of solar thermal energy systems and small-scale photovoltaic (PV) energy generation, there is a need to develop intelligent controllers to manage efficiently the energy generated by these systems. Ideally, these intelligent controllers will be able to predict the availability and magnitude of the solar resource and energy demand to plan in advance for periods when the solar resource is low or when energy demand is high.
This work demonstrates that it is possible to deliver a viable energy management strategy for a small-scale photovoltaic-battery-grid (PBG) system that is capable of coordinating the energy flows among the different energy sources. In doing so, the problems of modelling and control of the energy distribution for the PBG system using model predictive control (MPC) were addressed.
In devising this strategy, this work developed a nonlinear autoregressive recurrent neural network with exogenous inputs (NARX) and demonstrated that such a network was able to forecast both the energy demand for a typical New Zealand house, and the solar irradiation levels across a number of different locations (within New Zealand). Following on from this it was shown that using artificial neural network (ANN) based solar radiation and energy demand forecasts as measured disturbances allowed the MPC to plan for periods of low sunshine or high-energy demand.
The performance of the overall ANN informed model predictive control system was verified using simulation results and compared with the open-loop optimal control approach. The results showed that, for a typical Auckland weekly household load of 355 kWh, the MPC approach imported approximately 110 kWh less energy from the grid than the optimal control approach. Furthermore, for the one-week period examined, the MPC approach managed to export over 15 kWh more energy to the grid than a controller based on the optimal control approach using the same PV production and energy demand data with the same objective function and constraints.
