LSTM-Based Multi-scale Model for Wind Speed Forecasting
Wind speed forecasting is crucial for the penetration of wind energy sources in electrical systems, since accurate wind speed forecasts directly translates into accurate wind power predictions. A framework called Multi-scale RNNs specifically addresses the issue of learning long term dependencies in RNNs. Following that approach, we devised a LSTM-based Multi-scale model that learns to build different temporal scales from the original wind speed series that are then used as input for multiple LSTMs, whose final internal states are used to forecast wind speed future values. Results from two real wind speed datasets from northern Chile show that this approach outperforms the standard LSTM and its capable of working with very long input series without overfitting, while being computationally efficient regarding training times.
KeywordsWind speed forecasting Long Short-Term Memory Multi-scale recurrent networks
This work was supported in part by Fondecyt Grant 1170123, Basal Project FB0821 and the PIIC Research Project, DGIIP-UTFSM (Chile).
- 1.Hallas, M., Dorffner, G.: A comparative study on feedforward and recurrent neural networks in time series prediction using gradient descent learning (1998)Google Scholar
- 8.Pascanu, R., Mikolov, T., Bengio, Y.: On the difficulty of training recurrent neural networks. In International Conference on Machine Learning, pp. 1310–1318, February 2013Google Scholar
- 9.Chung, J., Ahn, S., Bengio, Y.: Hierarchical multiscale recurrent neural networks. arXiv preprint arXiv:1609.01704 (2016)
- 10.Koutnik, J., Greff, K., Gomez, F., Schmidhuber, J.: A clockwork RNN. In: International Conference on Machine Learning, pp. 1863–1871, January 2014Google Scholar