Research on Short-Term Load Forecasting Method of Power System Based on Seq2Seq-Attention Model

The load forecasting of power system has great influence on the planning, controlling, and dispatching of power grid, thus it can bring a lot of benefits and make the power system more economical. Because so many conditions can influence the load, it is a challenge for us to obtain more precise forecasting results even if on the digital age. With the rapid developing of Big Data technology and a lot of Artificial Intelligence algorithms coming, we can find some new methods that can help us improve the load forecasting accuracy. In this paper, we make some predictions with short-term load forecasting of power system based on some characteristics that the power system can easily collect, such as active power, reactive power, voltage and so on. Then a new intelligent model named Seq2Seq-Attention is established based on the theory of LSTM and the thinking of Attention function. The model can combine the data from different hours of days through different weights and express the hidden relationship of hours precisely. Therefore, we can solve the load forecasting problem of stations better with the historical multidimensional and time series featured data according to the model. Finally, a simulation model is generated and the day-ahead load forecasting of substations is verified by using the TensorFlow deep learning framework. The results can strongly prove the accuracy and practicability of the proposed model.