Research on Power Load Forecasting Technology based on Time Series

Abstract

Multivariate, multi-step power load forecasting represents a classic complex time series modeling challenge, with its core difficulty lying in simultaneously capturing long-term dependencies and local non-stationary fluctuations. In this paper, the calculations of MSE and MAE are performed using standardized data scales. Consequently, the experimental results presented in subsequent chapters represent dimensionless errors on a standardized scale, rather than absolute errors measured in physical units such as megawatts. Thus, subtle numerical differences among models under this evaluation protocol remain clearly comparable.