Abstract: A short-term photovoltaic (PV) power prediction model combining Gaussian Mixture Model (GMM) integrated Kmeans clustering algorithm (referred to as GMKmeans), Black-Winged Kite Algorithm (BKA) optimized Variational Mode Decomposition (VMD), and Temporal Convolutional Network (TCN)-Bidirectional Long Short-Term Memory (BiLSTM) was proposed in response to the significant fluctuations and randomness of PV system power output due to weather conditions. The original PV data set was classified using the GMKmeans algorithm. On this basis, the BKA algorithm was introduced to optimize the selection of VMD parameters. The optimized VMD decomposed the PV power data into multiple Intrinsic Mode Functions (IMFs), allowing the model to deeply understand the complex patterns of PV power evolution over time. Finally, the TCN-BiLSTM model predicted the decomposed IMFs, and the prediction results were stacked and reconstructed to achieve high-precision prediction of PV power. Experimental results demonstrate that the model significantly improves the accuracy of PV power prediction, proving the method's effectiveness in addressing the challenges posed by the complexity of PV system output.