The effective prediction of the urban events can provide the decision support for the government to avoid, control or mitigate the relevant social risks. Firstly, a conditional strength function based on integral derivative method is proposed to improve the precision of sequence prediction. Secondly, a temporal point process model based on recurrent neural network and cumulative hazard function is constructed. The nonlinear dependence of historical events is captured by recurrent neural network, and the cumulative hazard function is obtained by fully connected network. Finally, the representative synthetic data sets and real data sets are selected to compare the performance of several models. The experimental results show that the proposed method can better predict the time series of urban events, and is superior to the traditional temporal point processes in the aspects of mean absolute error and mean negative log-likelihood, which exhibits the superiority of the model.