Short-term electrical load demand forecasting using artificial neural networks for off-grid distributed generation applications

The rapid transformation in the power sector with the move to renewables unlocks many applications. One of the application areas is off-grid distributed generation, where solar photovoltaics-energy storage systems are used to power remote facilities with small electrical loads. Since energy storage costs dominate the overall system costs, effective utilization of energy storage is essential. To reduce life cycle costing, and to maximize the system reliability, requires not only accurate system design but also optimum operational efficiency. To run the plant at optimum levels requires effective power flow and a load management mechanism to sustain power demand effectively. For off-grid solar photovoltaics-energy storage systems, both power supply and power demand behaviors are independent and nonlinear, which creates operational and modelling complexities. In this paper, only one side of the equation, the prediction of power demand, will be studied by using Artificial Neural Networks (ANN). As a case study, remote oil & gas facility electrical load is considered, using several multi-layer neural network prediction models to evaluate forecasted load demand accuracy. We find that CNNARX2 cascaded nonlinear autoregressive neural network model 2, utilizing 15-min load demand data, provides the best fit to the load with 0.9778 R-Value and 5.49% Root mean squared error (RMSE).