Integrating LSTMs with Online Density Estimation for the Probabilistic Forecast of Energy Consumption
In machine learning applications in the energy sector, it is often necessary to have both highly accurate predictions and information about the probabilities of certain scenarios to occur. We address this challenge by integrating and combining long short-term memory networks (LSTMs) and online density estimation into a real-time data streaming architecture of an energy trader. The online density estimation is done in the MiDEO framework, which estimates joint densities of data streams based on ensembles of chains of Hoeffding trees. One attractive feature of the solution is that queries can be sent to the here-called forecast-based point density estimators (FPDE) to derive information from a compact representation of two data streams, leading to a new perspective to the problem. The experiments indicate promising application possibilities of FPDE, including but not limited to the estimation of uncertainties, early model evaluation and the simulation of alternative scenarios.
KeywordsDensity estimation Data stream Energy consumption
- 1.Zhu, L., Laptev, N.: Deep and confident prediction for time series at uber. In: 2017 IEEE International Conference on Data Mining Workshops (ICDMW), pp. 103–110 (2017)Google Scholar
- 2.Dai, H., Kozareva, Z., Dai, B., Smola, A., Song, L.: Learning steady-states of iterative algorithms over graphs. In: International Conference on Machine Learning, pp. 1114–1122 (2018)Google Scholar
- 3.Lin, F., Beadon, M., Dixit, H.D., Vunnam, G., Desai, A., Sankar, S.: Hardware remediation at scale. In: 2018 48th Annual IEEE/IFIP International Conference on Dependable Systems and Networks Workshops (DSN-W), pp. 14–17. IEEE (2018)Google Scholar
- 5.Berriel, R., Teixeira Lopes, A., Rodrigues, A., Varejao, F., Oliveira-Santos, T.: Monthly energy consumption forecast: a deep learning approach. In: 2017 International Joint Conference on Neural Networks (IJCNN), pp. 4283–4290 (2017)Google Scholar