Forecasting of Long-term Electricity Consumption in China: A Combined Approach based on KPCA and Linear Regression Model

In this chapter, we demonstrate that the integrated model of Kernel Principal Component Analysis (KPCA) and Linear Regression (LR) outperforms other methods, thereby offering a novel and feasible approach for long-term electricity consumption prediction. We propose a combined model of KPCA and LR for this purpose. Despite a limited sample size, the model can accurately forecast the temporal changes in total electricity consumption, boasting high interpretability and practical utility. We employed KPCA to reduce the complexity of the original data, then input the dimensionally reduced data into a Backpropagation Neural Network (BPNN) and other models, yielding optimal model results. Visualization of the three principal components derived through KPCA revealed that the first principal component represents the long-term growth of electricity consumption, while the other two components represent the long-term fluctuation of electricity consumption. Additionally, population features, price features, and industrial structure features also contribute to the increase in China's electricity consumption, albeit in a more fluctuating manner. Lastly, we predict that China's total societal electricity consumption will reach 1.83 trillion KWH by 2035, a forecast more optimistic than that of Oxford experts and consistent with China's victory in combating COVID-19. The model holds an optimistic view of China's future economic prospects, aligning with China's rapid economic growth and its comprehensive victory in the fight against COVID-19.