Inflow Forecasting Based On Principal Component Analysis and Long Short Term Memory

Hydropower is a major form of energy production in the Nordic power market. Hence, high accuracy inflow forecasting plays an essential role in predicting the power price. This paper proposed PCA-LSTM prediction modelthat combines machine learning technology, e.g., long short-term memory (LSTM), with the data-driven approach, e.g., principal component analysis (PCA). Based on the proposed approaches and the available inflow and hydrological variables, three different scenarios are conducted to validate the performance. Three performance criteria, i.e., Mean Relative Error (MRE), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), are applied to evaluate the model performance. The results show that the prediction accuracy is increased after introducing the hydro-metrological data into the model. Furthermore, the two-steps approach based on the inflow data and hydro-metrological data has the best performance. This implies that it is better to apply the data-driven approach for the hydrological data before implementing the machine learning technology.