Wind energy forecasting using R : Data Science for Internet of Things Project

Participants from the Data Science for Internet of Things course are working on some excellent projects.

Here is a great example -

Created by Vaijayanti – and other collaborators from the course

Project name: 4Wind

Domain:  Renewables/Wind energy

Problem statement: To make day-ahead forecast for wind power (10 minute interval data)

Data: collected from around 5000 turbines

Duration: She has been working for more than a year and still ongoing

Programming Approach : Using R for data science. To evaluate various models: ARIMA, Holt-Winters forecast, neural networks ACF and PACF plots reveal 35 lags matter. Hence the data is recoded to have 35 columns of lags for both wind speed and wind power.

R packages: forecast, neuralnet, nnet

Challenges: Training neural networks has become heavy on a PC. Hence, experimenting with multiple models including Distributed(Spark) and AWS, Azure etc. 35 lags were based on ACF and PACF analysis. However, this number could be different. More analysis might yield different lags as predictors. Also, we could use not just consecutive lags but also the samples a day before and year before are also predictors.

UI: Using the Shiny package 

Publication/Open source methodology: Contribution to Data Science for IoT methodology and to forthcoming book on Data Science for IoT

Image source: wikipedia