Information aggregation based trend prediction of energy structure via an improved compositional data time series forecasting model and its application

To improve the prediction accuracy of compositional data time series (CDTSs), the aggregation of compositional data was considered and applied to construct a combination forecasting model. Different from current arithmetic mean based aggregation of compositional data, the aggregation method of compositional data from the induced ordered weighted averaging (IOWA) operator was put forward. Properties of such aggregation methods are discussed. Since prediction accuracies of different individual forecasting models are diverse over time, forecasting error between the aggregated CDTSs and the original CDTS is minimized and set as an objective function of the aggregated weights. To derive the optimal weights associated to individual forecasting models, the genetic algorithm was utilized. Correspondingly, an improved time-varying combination mode and an IOWA operator based combination mode are developed. Finally, a numerical study on China's primary energy production structure is presented. The results show that the developed varying weight combination model is superior to the benchmark model in terms of prediction accuracy comparison, illustrating the feasibility and validity of the developed combination forecasting model.