Utilizing Block Characters for Real-time Load Forecasting in Energy Internet

With the development of Energy Internet (EI), its load forecast becomes more important and necessary. Although the load of EI has some temporal periodic features, its inherent uncertainties in different types of loads make accurate prediction more difficult. Sometimes even the typical neural network may produce apparent load forecast errors and take non-negligible computing time. In EI load forecast with one step further real-time prediction, we apply the threshold factor of block characters into dimension determination of ELM’s (Extreme Learning Machine) training model size. Through corresponding simulations, the proposed algorithm shows more accurate results than ELM used alone, and also than ESN (Echo State Network), which exhibits large application potentials.