The purpose of night sky brightness prediction model was established for prevention and evaluation of urban light pollution.The BP neural network algorithm based on time series theory was introduced, owing to the nonlinear distribution of light pollution night sky brightness.The principle of neural network was discussed and the prediction model based on time series theory was established.The test data as the training sample was trained by improved BP algorithm with MATLAB and the error were analyzed.When the Number of Hidden Neurons was five and the training function is trainlm, the maximal absolute error can reach 0.003 6 cd/m2 and the maximal relative error 2.361 4%.Running results are consistent with the experiment data and the relation between the target vectors is better.