Radian Scaling: A Novel Approach to Preventing Concept Drift in Electricity Load Prediction

Artificial neural networks often suffer from concept drift when predicting the electricity load in smart cities due to seasonal trends. The most commonly used methods to solve this issue are automatically updating the model, using feature engineering to capture the seasonality trends, and using ensemble methods, making the model complex for someone to replicate. Furthermore, as the issue is mostly due to the shifted minimum and maximum values, we need a method to normalize without being bound by the training data's scale, mean, and median. We proposed radian scaling to normalize the angles between the two consecutive values for the artificial neural networks to process. Our prediction result shows high accuracy with a root-mean-square error score of 0.0021 and 0.0036 for many-to-one and many-to-many forecasting scenarios, surpassing existing methods without model modification.