Multi-model Automatic Sifting Methodology in Load Forecasting

摘要

The diversity of models is an important issue in load forecasting. To improve the forecasting precision, it is necessary to distinguish between better models and bad ones. Unfortunately, this task is very difficult. This paper proposes a novel multi-model automatic sifting methodology to solve this problem. In the new algorithm, the odds-matrix method is used to calculate the weight of each model, which reflects the 'optimality' of an individual forecasting model. Thus, the efficacy of each model can be differentiated by evaluating probability distribution function of the weights. Numerical studies show that this method is satisfying in improving forecasting precision.