In order to improve the accuracy of photovoltaic power generation power prediction, the standard dung beetle optimization algorithm（DBO）was improved by using three strategies：Bernoulli mapping, the spiral update mechanism of whale optimization algorithm （WOA）and the optimal individual adaptive t-distribution. Through verification on 8 standard test functions, the results show that the improved algorithm has significant improvements in convergence speed and optimization ability. Furthermore, the improved dung beetle optimization algorithm was used to optimize the long short-term memory network model（IDBO-LSTM）for photovoltaic power generation power prediction, and compared with six other models. The prediction results show that IDBO-LSTM exhibits better prediction performance under 3 different weather types than other models. Compared with the DBO-LSTM model, the average absolute error rate（MAPE）of IDBOLSTM decreased by 0.08%, 3.51%, 4.02%, respectively.