LSTMs for Energy Forecasting

From REU@MU
Revision as of 13:33, 4 August 2017 by Cnapun (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Researcher: Nikil Pancha Mentor: Dr. Richard Povinelli

Description

Natural gas and electricity demand forecasting are important because companies need to know how much gas/electricity to purchase in advance. If too little is purchased, then the utility might need to purchase energy on the spot market, which inflates prices excessively. On the other hand, if too much is purchased, the company will have unused energy that likely will go to waste, and in the case of natural gas, they may face penalties for their overpurchasing.

Long Short-Term Memory networks (LSTMs) are a form of Artificial Neural Network (ANN), and in particular a type of Recurrent Neural Network (RNN). RNNs are a class of neural network that operate on sequences, taking a cell state and input vector as an input to the next timestep, in contrast to a traditional ANN which does not take into account any sequential structure. Because time series are sequential by nature, the RNN is a natural structure to use in time series forecasting.

Milestones and Goals

5 June Don't give up
12 June Complete literature serach
19 June Have data gathered
26 June Have learned framework (Tensorflow and Keras most likely)
3 July Have preliminary models
10 July Draft paper
17 July Second round of models and review draft
24 July Deal with errors, finish poster
31 July Finish presentation and paper