Forecasting of Electricity Consumption Using Neural Networks in Public Hospital Buildings with Installed Smart Meters in Thailand

Thailand has promoted the national plan for energy consumption to achieve a target of the reduction of energy intensity by 30 percent in 2030. Therefore, energy conservation measures have been taken in the designated buildings particularly. For designated public hospital buildings in particular, electricity use contributes the most part of energy consumption and gradually increases every year. Therefore, this paper aims to strengthen the energy efficiency plan by forecasting electricity demand from smart meters using a recurrent neural network (RNN) model. This model is developed from the acquired dataset from over 2,100 installed smart meters. It was found that the RNN model could be used for a long-term prediction of electricity consumption with acceptable results based on the limited dataset for public hospital buildings. The study also examined the model on three scales of sample hospital buildings, which are large, medium, and small scale. The result revealed that a trend of prediction is according to actual consumption. Moreover, the accuracy of the forecast model is evaluated by the mean absolute percentage error (MAPE) and mean square error (MSE) methods, and the MAPE has lower than 20 percent.