In view of the common small-sample data and nonlinear characteristics in energy systems, the randomness of algorithm optimization cannot guarantee the stability of modeling results. This study aims to propose a repeatable grey multivariable prediction framework to improve the repeatability of the model itself.
Firstly, based on the traditional multivariate discrete power model, the power parameters of characteristic behavior sequences and linear correction terms are introduced. Secondly, the Monte Carlo mean method is incorporated into the modeling framework of the model. By conducting Monte Carlo simulations on four optimization algorithms with the same number of iterations and initial conditions, the ideal algorithm with the strongest stability is screened out. Finally, through comparisons with other prediction models in two case studies, the effectiveness and superiority of the new model under the optimal algorithm are fully verified.
The newly proposed model not only effectively identifies the nonlinear characteristics of energy consumption systems but also overcomes the strong randomness issue in optimization algorithms, serving as an effective approach to enhance model repeatability and stability. In case studies, the modeling results of the new model for energy consumption in China and Jiangsu show a high degree of consistency with actual data, exhibiting MAPE of 0.14 and 0.37%, respectively. These results significantly outperform other models.
This paper presents a scientific and effective prediction method, aiming to handle the commonly encountered nonlinear characteristic data in reality. The energy consumption forecasting results in this paper can provide scientific references for China’s energy structure adjustment, thereby gradually achieving sustainable development.
This paper introduces power parameters of characteristic behavior sequences and linear correction terms to fully identify the nonlinear characteristics of all sequences and the internal influences of the system. Furthermore, this paper optimizes model parameters by selecting the most stable algorithm through comparisons of multiple algorithms and adopts the idea of Monte Carlo simulation to take the mean value, processing the mean values of multiple modeling results to improve the repeatability and robustness of the model.
