A hybrid prediction model of photovoltaic power system based on AP, ISSA-based VMD, CLKAN and error correction

Abstract

The accurate prediction of photovoltaic (PV) power output is crucial for optimal energy management. However, PV power generation systems are influenced by various meteorological factors, resulting in the fluctuation and intermittency issues in their output power. To enhance the prediction accuracy of PV power, this study proposes a novel hybrid model, the Convolutional Neural Network-Long Short-Term Memory-Kolmogorov-Arnold networks (CLKAN), integrated with the Improved Sparrow Search Algorithm (ISSA), Variational Mode Decomposition (VMD), and Error Correction (EC). The key contributions of this research are as follows: (i) The Iterative Chaotic Map with Infinite Collapses (ICMIC) method and Quantum Rotation Gate (QRG) are applied for the first time to enhance the SSA optimization algorithm. (ii) By leveraging CNN-LSTM for input sequence feature extraction, the KAN network with the B-spline basis function is innovatively utilized to connect the CLKAN model. Experimental analysis conducted on eight typical days across different seasons in two weathers at a PV power station in Yunnan Province, China, shows that the proposed model achieves higher prediction accuracy and reduced the amount of computation. For the instance in summer of sunny weather, the proposed model achieved the best performance with MAE, SMAPE, RMSE, R², FLOPs, Par values of 0.22 MW, 0.88 %, 0.28 MW, 99.91 %, 63,136 and 37,120 respectively, demonstrating its superior performance. Furthermore, the application of the CLKAN model, based on AP and ISVMD, for PV power prediction at stations in Yunnan and Gansu, China, highlights the model's robustness across various spatial and temporal scales.