Development of a Forecasting Module based on Tensorflow for use in Energy Management Systems

Abstract--The use of Energy Management Systems (EMSs) allows obtaining remarkable advantages for both end-users of electrical energy and grid operators. These systems can take advantage of a suitable forecasting of load demand and meteoclimatic variables tied to power generation. In facts, the forecasting ability enables a more effective planning of the power allocation. The aim of this paper is the development of a forecasting module that can be interfaced to EMSs to deliver a 24h ahead forecasting. The module is based on a suitable Artificial Neural Network (ANN), namely the nonlinear autoregressive with exogenous input (NARX) ANN. Such an ANN has been implemented using Tensorflow library and writing Python code. It has been trained using a public solar irradiance dataset, and several tests have been performed to assess its performance with different numbers of output units, hidden layers, and neurons per hidden layer. The obtained results show that the obtained forecasting module has good performance and is suitable for embedded implementation and online operation to support EMSs.