Energy Consumption Prediction via Data Decomposition

The patent segments the energy consumption time series into multiple components (trend component, seasonal component, and residual component) using decomposition facilities. This segmentation allows the prediction function to be trained on simpler, decomposed data rather than complex raw data, significantly reducing training time while improving prediction precision. The trend component captures long-term patterns, the seasonal component captures periodic variations, and the residual component captures random fluctuations, enabling more efficient and accurate prediction.

The patent performs preliminary decomposition of the energy consumption data into trend and seasonal components before prediction. This preliminary action transforms the raw data into a format that enables both accurate retrospective analysis and real-time forecasting. By pre-processing the data to separate systematic patterns from random variations, the system can quickly generate real-time predictions without performing complex analysis at prediction time, thus providing both accuracy and real-time capability.

The patent changes the parameters of the training data by transforming raw energy consumption data into decomposed components (trend and seasonal). This parameter transformation allows the use of limited historical data more effectively - instead of trying to extract all patterns from raw data, the system trains on specifically transformed parameters that capture the essential behavior. This reduces the amount of historical data needed while improving prediction precision, as the decomposed parameters provide clearer signals for training the prediction function.