Disaggregated Forecasting of the Total Consumption of a Given Subset of Customers
摘要
Individual electrical signals have a high volatility and are difficult to predict accurately. But, by considering the sum of some of them, the resulting signal-to-noise ratio increases and such aggregates could be useful. We propose a way to use individual data to improve the prediction of the total consumption of a given subset of customers. Clustering strategies allow to aggregate together customers of similar consumption structure, defining heterogeneous groups of similar clients. The aggregate signal of customers from each given group will be more regular and more predictable.