Day ahead Power Demand Forecasting for Hybrid Power at the Edge

We describe the investigation and testing of univariate forecasting techniques on IoT hardware for application at "the Edge" using power demand forecasting. An evaluation of common forecasting techniques is presented, tested using the Morocco Buildings Electricity ConsumptionDatasets. An architecture is describedfor the Edge system that would enable 1-day forward forecasts of power demand for use in provisioning power in a hybrid power system. Several of the configurations examined in this study performed comparably with current trends in forecasting methods and are suitable for this application at the Edge, providing a balance of performance and accuracy. A Long Short-Term Memory (LSTM) Neural Network configuration provided the most effective balance of performance, accuracy and simplicity of deployment that is desirable for an application at the Edge.