A hybrid medium-long-term photovoltaic power forecasting model based on symplectic geometry mode decomposition and Mamba

Accurate forecasting of photovoltaic (PV) power plays a pivotal role in the economic dispatch and optimal functioning of power systems. However, the inherent volatility and uncertainty of PV power generation often lead to substantial errors in its forecasting. To address the issue, a novel hybrid model based on symplectic geometry mode decomposition (SGMD) and Mamba, referred to as SGMD-Mamba, is proposed for medium-long-term PV power forecasting in this paper. Firstly, SGMD adaptively decomposes the original PV power series into multiple stable and noise-reduced components. This decomposition preserves the key characteristics of the original signal while mitigating the impact of noise, thereby enhancing the accuracy of subsequent modeling processes. Secondly, Mamba is introduced for the first time into the PV forecasting domain based on the state-space model (SSM) framework. Mamba leverages the framework to integrate the advantages of sequential processing and global information handling, effectively capturing the trends of each stationary PV power subsequence. Finally, the final forecast value is derived by aggregating the predicted results of all subsequences generated by Mamba. The combination of SGMD and Mamba not only addresses the inherent limitations in a single model when dealing with complex PV power sequences, but also ensures high accuracy and efficiency in PV power forecasting. Experimental results demonstrate that SGMD-Mamba achieves RMSE, MAE, MAPE and R 2 values of 0.1348, 0.0911, 0.4590, and 0.9431 respectively, surpassing several existing methodologies and exhibiting outstanding medium-long-term PV forecasting capability. • A novel hybrid method based on SGMD and Mamba is proposed for PV power forecasting. • SGMD is used to decompose PV power series to reduce the impact of noise. • Mamba is applied for the first time in PV power forecasting to enhance accuracy.