Abstract: Accurate transformer heavy and overload event forecasting improves power supply dependability and safety. The existing transformer heavy and overload warning strategies must be more accurate and attainable. They cannot manage temporal covariate shift (TCS) or dynamic transformer rating (DTR), which is the main issue. This paper presents a week-ahead heavy and overload warning method to address the above challenges. Informer is used to create a week-ahead load forecasting model. Domain adaption (DA) models reduce the distribution gap between historical and future load data. The data distribution changes with time; hence, an evolutionary updating model is proposed. A transformer cooling mode-based hot-spot temperature (HST) model and an assessment model for transformer load capacity are proposed. Finally, a DTR-based heavy and overload warning method is proposed. The proposed method is validated using the measured data collected from a power supply bureau in western China. The simulation results demonstrate a maximum improvement of 32.55% in load forecasting accuracy, and the suggested method is universal and outperforms the existing methods.