Day-ahead power load forecasting based on meteorological similar day correction and IPO-DLinear

The existing power load forecasting methods encounter significant challenges, particularly when accounting for the influence of meteorological factors on load fluctuations. Traditional methods often overlook the complex nonlinear relationship between meteorological characteristics and load, leading to reduced forecasting accuracy. A day-ahead power load forecasting model based on meteorological similar day correction (MSDC)-improved parrot optimizer (IPO)-decomposition-based linear (DLinear) is proposed. The proposed method enhances the parrot optimizer (PO) by incorporating a logistic map, adaptive mutation strategy, and spiral fluctuation search to optimize the DLinear superparameters. Periodicity and trend characteristics are extracted from the DLinear model. The load forecast value is corrected by comparing the Euclidean distance of meteorological characteristics. The resulting day-ahead power load forecasting model, IPO-DLinear-MSDC, is validated using a simulation analysis of total load data from the Zhuzhou area in Hunan from June to October 2024. The model's performance is evaluated with an average absolute percentage error (MAPE) of 4.67% and R² of 0.833, demonstrating improvements of 15.09% and 23.44%, and increases of 0.0741 and 0.1253, respectively, comparing to IPO-DLinear model and PO-DLinear model.