Hourly Day-Ahead Power Forecasting for PV Plant Based on Bidirectional LSTM

A novel hourly day-ahead power forecasting approach for PV plant based on bidirectional LSTM is proposed in this paper. Firstly, after analyzing the periodic characteristics of PV plant daily power curves, we employ K-means to cluster days into different types of weather according to the irradiance index. Then, a bidirectional Long Short-Term Memory (LSTM) is presented to build forecasting models for each type of weather in four seasons. An empirical study on a real dataset shows that the proposed method can effectively use multivariate time series information to predict the power for PV plants and obtain better performance than Autoregressive Integrated Moving Average model (ARIMA), Extreme Learning Machine (ELM) and Multilayer Perceptron (MLP).