Optimized photovoltaic power generation prediction model based on the improved northern goshawk algorithm

Accurate photovoltaic (PV) power forecasting is fundamental to optimizing PV system control. To address the limited accuracy of current PV power forecasting methods, this work proposes a novel hybrid forecasting model based on variational mode decomposition-improved northern goshawk optimization-long short-term memory (VMD-INGO-LSTM). To eliminate the volatile characteristics inherent in the PV data, the original data are first decomposed using VMD. An improved northern goshawk optimization (INGO) algorithm is subsequently employed to optimize the parameters of the LSTM network. The INGO algorithm incorporates three key enhancements to the standard northern goshawk optimization algorithm: an inverse learning strategy, Gaussian mutation, and greedy selection. Specifically, the optimization process is implemented sequentially, first employing the inverse learning strategy, then conducting Gaussian mutation, and finally executing the greedy selection. This order ensures proper algorithm iteration and optimization. The INGO algorithm optimizes the LSTM by tuning its hidden layer neuron number and the length of the time series window. The resulting VMD-INGO-LSTM model enables accurate prediction of future PV power output. The model performance was validated using irradiance and power time series data, with a 1-hour resolution, from a PV power plant located in Hefei, China, collected from 2023 to 2024. The results demonstrate a substantial improvement in overall prediction accuracy. These findings indicate that compared with the other models, the proposed model can effectively increase the PV power forecasting accuracy.