Regional Energy Analyst (REA)

Regional Energy Analyst, the first data-driven software for the analysis of the future energy consumption of buildings across sectors, cities and regions.

The first version of this software package includes a model for forecasting the energy demand of buildings across 96 cities in the united states under different scenarios of climate change.

The tool is built on a hierarchical bayesian regression model and a Deep neural network of half a million buildings surveyed between 2010-2011.

The tool includes two models.

1. A Hierarchical Bayesian Neural Netowork
2. A Wide and Deep Neural Network.

The tool includes 96 cities including all capitals of state.

The tool includes forecasts for 3 scenarios of Climate Change.

1. Family A1B
2. Familiy B1
3. Family A1

It Includes a sub-set of the original training and testing database. It Includes a sub-set of the original database for predictions. This should be enough for testing the approach.

Repository contents

• databases: It includes databases for training, testing and doing predictions with the models.
• models: it includes scripts to define, carry out inference, check performance and make predictions with a hierarchical and a deep NN model.
• results: It stores the results of the inference, performance and prediction processes.
• analysis: It includes scripts to generate tables and plots for further analysis. (the tables and plots are stored in there)

Installation for the Hierarchical Bayesian model

1. install anaconda distribution for python 3 and python 3
2. install pymc3, theano, scikitlearn and pandas. Probably you need to install more, so python will let you know in case we are missing it.

Installation for the Neural Network

1. install tensorflow, I advise to use a GPU to run it quickly.

Step0. preprocess the data (sorry only available for the authors)

1. run the script data_processing/IPCC_scenarios_cleaner.py
2. run the script data_processing/enthalpy_calculation.py
3. run the script data_processing/split_enthalpy_by_period.py.py
4. run the script data_processing/training_and_testing_database.py
5. run the script data_processing/prediction_database.py

Step 1. Configure the script

• open the excel file configuration.xlsx/test_cities and indicate the names of the cities to evaluate.
• open the script configuration.py and indicate the paths to the datasets and the configuration.xlsx. Do this step only if you have an alternative database to that one provided in the repository.

Step 2. Define and run the hierachical model

• Open the script models/hierarchical/1_definition_and_inference.py
• Check the input configurations at the end of the script. A name of the model will be infered from these inputs.
• Run the script
• The results are stored in results/hierarchical/inference/ [model name].pkl

Step 3. Check performance of the hierachical model

• Open the script models/hierarchical/2_performance_check.py
• Indicate the name of the model to use (e.g., log_log_all_standard_1000)
• Run the script
• The results are stored in results/hierarchical/performance/ [model name].csv

Step 4. run some predictions of the hierachical model

• Open the script models/hierarchical/3_predictions.py
• Indicate the name of the model to use (e.g., log_log_all_standard_1000)
• Run the script
• The results are stored in results/hierarchical/predictions/ [model name]/[city].csv

Step 5. Create tables and plots for analysis

• Run the script analysis/coefficients_hierarchical_model.py to create a PDF of the regression coefficients inferred with the hierarchical model.
• Run the script analysis/data_pair_plot.py to create a pairplot of the input training or test database.
• Run the script analysis/IPCC-scenario_plots.py to create a box_plot of the heating degree days of all IPCC scenarios.
• Run the script analysis/plots_predictions.py to create a box_plot of the energy consumption predicted for the IPCC sceanrios. NOTE: You must specify the hierarchical model name in the inputs at the end of each script.