To effectively prevent the risk posed by the randomness and volatility of photovoltaic (PV) power generation on the safe and stable operation of the power grid, an innovative combined predicted method composing of a similar day selection method based on the combined weight method (CWM) and comprehensive similar distance (CSD), a sequence decomposition and recombination method based on variational mode decomposition (VMD), improved whale optimization algorithm (IWOA), singular spectrum analysis (SSA) and fuzzy entropy (FE), as well as a IWOA-convolutional neural network-long short-term memory (CNN-LSTM) prediction model was developed for establishing a high-precision integrated forecasting model of PV power output. Firstly, the key meteorological factors affecting PV power output were extracted by aid of the Pearson correlation coefficient (PCC) method. Secondly, a combined method named CWM, which integrates the analytic hierarchy process (AHP) and entropy weight method (EWM), was employed to assign the weight coefficients to corresponding meteorological factors. Then, the weighted meteorological factors were incorporated into the CSD calculation process for generating a high-quality training sample set with the meteorological characteristics similar to those of the forecasted day. Next, the IWOA algorithm based on Tent chaotic mapping, variable spiral update and adaptive weight dynamic adjustment mechanism was designed. A brand-new combination of IWOA-VMD, SSA and FE was applied to achieve effective decomposition and recombination of original power sequence. Finally, a high-precision PV power output forecasting model based on IWOA-CNN-LSTM was constructed. Experimental results on a PV power station in Yunnan demonstrated that the proposed model outperforms other benchmark models in terms of accuracy and stability, demonstrating its potential for practical application in PV power prediction.