Prepared by Manisa Pipattanasomporn
During the past several years, I have been involved in smart building research. I would like to create this set of tutorials to share the knowledge that I learned, related to building-level data analytics, building-level load forecasting and more.
This page was created in December 2019, and contents will be added as they become available!
Work with CU-BEMS Data: Chamchuri 5 building: This notebook explores the CU-BEMS dataset of Chamchuri 5 building. Chamchuri 5 is a seven-story academic office building located at Chulalongkorn University, Thailand. Recorded data are: power consumption (kW) of individual AC units, lighting and plug loads in each zone, on each floor of the building, together with corresponding indoor temperature (deg C), humidity (%) and illuminance (lux) in each zone.
Working with time-series data Python is powerful to handle time series data. In this tutorial, we will learn: how to import time-series data, how to deal with missing time-series data, how to filter a selected time range, how to visualize time-series data and how to resample time-series data.
Load forecasting: Lookback forecasting: This is the simplest method for load forecasting. Let’s see what are the resulting error metrics (RMSE, MAPE) using this method.
Load forecasting: Multiple linear regression (MLR): One of the most basic ML is multiple linear regression. In this tutorial, we will use MLR to perform hour-ahead load forecasting when Lt-1, t_outdoor and others are inputs to the forecasting model. The LinearRegression model from sklearn.linear_model was used.
Load forecasting: Autoregression model (AR): Autoregression is a linear regression model that uses lagged variables as inputs to predict the value at the next time step. This tutorial uses AR model from statsmodels.tsa.ar_model to automatically chooses optimal lag values for the model.
Load forecasting: Artifical Neural Network (ANN): We will use ANN model (keras library) to perform hour-ahead load forecasting.
Load forecasting: LSTM and BiLSTM: We will use LSTM and BiLSTM (keras library) for hour-ahead load forecasting.
For those who would like to review commonly used Python codes/libraries for data analytics, please refer to ReviewBasicPython.ipynb.