Short-Term Load Forecasting by FOS-ELM with Re-Learning Method

This article presents a method for short-term load forecasting by Fully Online Sequential Extreme Learning Machine (FOS-ELM) model. The model can learn incrementally without the need for initial training samples. The model is suitable for use in locations where historical load data is not available to train the model. This article also presents the improvement of the forecasting accuracy of the FOS-ELM model by letting the FOS-ELM model learn from each newly received sample more than once, called “re-learning”. A comparative experiment between the FOS-ELM model without re-learning and the FOS-ELM model with re-learning found that the re-learning method can reduce the error in load forecasting. This is because the model can quickly adapt to the trends of new data. But after the fifth round of re-learning, the forecasting error value could no longer decrease. This is because the model has an over-fit with the latest learned data. In other words, re-learning allows the model to have lower forecasting errors, but the number of re-learning times must be chosen appropriately.