Building energy management has been a hot topic nowadays as of its importance in reducing energy wastage and carbon emissions. However, the performance of building energy consumption prediction has been stagnant due to problems such as low prediction accuracy.
With Internet of Things technology has developed rapidly, the smart sensors and smart meters can collect the energy consumption data in efficience and economic way. which the time-series machine learning can be applicable for prediction.Recently we have successfully completed a machine learning prediction project for building energy demand forecasting. The requirement is, to generate the energy demand at hourly level for Shenzhen Civic center, with the availabe data in past 24 months.
In this case study, a time-series prediction model been developed and used for the Shenzhen Civic Center energy demand forecasting.
Essentially, it is Time Series forecasting, which is a method to arrange the historical data of the forecast target into a time series, and then analyze its development trend over time, and extrapolate the future value of the forecast target.
A major advantage of using this modified LightGBM model is, it can quickly adapt to the yearly trends, the season and holiday factors in one model.
We have analyzed the data of the civic center in the past two years and conducted predictive learning to predict the hourly energy consumption for the coming three months. After comparing with the actual consumption, the result is exciting.
The machine learning energy consumption prediction model has been deployed to the cloud, provides a rolling three monthly energy consumption forecast to the building operator continuously, helps the building management team to:
- foresee the energy consumption trend
- adjust the energy equipment maintenance schedule
- reduce the energy supply disruption
With the study case, it also provides the solid base to optimize the IoT network for further smart building initiatives of Civic Center.