Short-Term Electric Load Forecasting Using Artificial Neural Networks
Abstract
Short-term forecasting of power consumption is an important tool for decision makers in the energy sector. Power load forecasting is a challenging multi-step ahead time series forecasting problem since the power load is dependent on a number of different factors, e.g. temperature, time of day, time of week and recent power consumption. The goal of this master thesis was to develop a model that predicts the power consumption in Nord-Trøndelag for each hour of the next day. Several different models based on artificial neural networks were developed and tested on a historical data set consisting of hourly observations of power loads and temperatures in Nord-Trøndelag from 2011 to 2017. The data set was provided by Nord-Trøndelag Elektrisitetsverk (NTE). The performance of the models was compared to NTE's current model and several other types of models that are commonly used for short-term load forecasting. The final proposed model is an ensemble average of the two best performing multilayer perceptrons tested and a time-varying linear regression model that uses Kalman filtering for weight estimation. The proposed model is very efficient in terms of time usage and a large improvement compared to NTE's current model.