DefinePK

Abstract

In these rapidly changing times of smart cities, a smart use of energy has become a financial rescue buoy. The forecasting of electricity loads is crucial for the stability of the grid, for resource allocation; however, it also becomes more and more important in the context of integrating wind and solar energy into the grid. Nonetheless, precise prediction is difficult when energy usage changes differently due to the variability of weather, human activity, and renewable generation. Some traditional statistical models, including linear regression and autoregressive type approaches, often fail to model the non-linear and multi-dimensional information underlying the data, resulting in a suboptimal forecasting performance. To address these limitations, this paper applies state-of-the-art machine learning and time series techniques to improve the forecasting accuracy of electricity load. Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GB), Long-Short Term Memory (LSTM), the Facebook Prophet, Extreme Gradient Boosting (XGBoost), and Linear Regression are used for prediction. Taking advantage of an extensive electricity consumption dataset as well as time series characteristics and context features, we obtain better forecasting with the proposed model. RF was the best performing among all the models with Lowest MAE=0.021, Extremely low RMSE = 0.014, and Highest R2=0.982 (Almost perfect). These results validate the potential of advanced machine learning based models to provide data capitalism-driven energy management solutions for smart cities.