Welcome to Individual Household Power Consumption Forecasting Homepage!

Data description

Five households power consumption data is investigated. The data was originally collected by UK-DALE project, consisting of disaggregated data for one year (2015). We merge the data into samples with time interval at 5 mins, and perform time series forecasting on the univariate data, which is the aggregate power consumption for the five households. Among the five households, households 1, 3, 5 involve more activities and human interference, whereas the rest of houses are less active. The power consumption curves for house 2 and 4 are more stable, more suitable for classic forecasting methods, such as ARIMA.

Pre-processed data download:

Household 1

Household 2

Household 3

Household 4

Household 5

Forecasting results using CNN-LSTM

The proposed CNN-LSTM is a hybrid deep learning neural network that can deal with volatile time seies data. The prediction results of our method for houses 1, 3 and 5 are shown below:

Prediction results for Household1:

Prediction results for Household3:

Prediction results for Household5:


The source codes of the proposed CNN-LSTM framework is freely available here.

Pre-requist for system installation:



Back to Yan Ke's Homepage.