buds-lab/building-prediction-benchmarking

An array of open source ML models applied to long-term hourly energy prediction for institutional buildings

building-prediction-benchmarking's Issues

Make a way to visualize the model fit metrics

Come up with a way to compare all of the model fit metrics using a Python library

Include the humidity and solar radiation (if possible) from the weather file

In the same way as temperature, include the humidity and solar radiation as input variables from the weather files. I'm not sure if solar radiation is available, but maybe one of the other inputs could be used to account for certain behavior.

Implement Sci-kit Learn regression models

Here is an ML example from start to finish - it has a lot of details about the data set, the test/train data sets: https://www.kaggle.com/claytonmiller/office-building-energy-prediction-random-forest

Replicate this model using alll the regression models in this library: http://scikit-learn.org/stable/supervised_learning.html

You can do this using Python files, Jupyter notebooks, Kaggle Kernels -- how ever you'd like

Implement models from EEMeter

So far, several sci-kit learn models have been implemented to the data sets.

Now we need to apply the prediction model from this library: http://eemeter.openee.io/

Please find the overview of the analysis so far here: https://www.overleaf.com/read/stqjjzttmrmc

You can see an example of the old EEMeter package implemented here: https://github.com/buds-lab/temporal-features-for-nonres-buildings-library/blob/master/07_EEMeter%20Models%20Creation.ipynb

Create a different Jupyter notebook than the Sci-kit learn notebooks to make this prediction analysis and create all the outputs in a similar way

Include the holiday and break schedules as an input variable

I have had another freelancer to the tedious work of creating a spreadsheet of all of the holidays. I have shared this spreadsheet on slack in addition to a MASTER meta spreadsheet where we can match the buildings with their annual schedules and use that as daily input to the ML models

Use the scikit learn function `sklearn.model_selection.TimeSeriesSplit`

Use sklearn.model_selection.TimeSeriesSplit to segment the data into training and testing for cross-validation

https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.TimeSeriesSplit.html#sklearn.model_selection.TimeSeriesSplit

this will be in its own notebooks and will be a fork of the previous scenario with 9 months training and 3 months testings

Implement time-series models from stats-models

https://www.statsmodels.org/dev/tsa.html

Implement the models from M&V 2.0

https://github.com/LBNL-ETA/RMV2.0