Mental health management of college students is a crucial aspect of college student management. To this end, this paper extracts the characteristic sequences of consumption behaviors, including the number of consumption per week, the number of times of use breakfast, the number of times of deviating from the normal time of using breakfast and the number of times fetching water from the daily consumption records of students. Then, the best model MLSTM-FCN is selected from the different existing time series classification models. Aiming at the problem that the FCN branch has a small sensory field and can only run in one direction in the network structure of the MLSTM-FCN two-branch splicing, it is proposed to use a Transformer to replace the FCN branch so as to put forward a Trans-LSTM students’ psychological abnormality early warning model based on the classification of time series. Model. According to the model comparison, the Trans-LSTM fusion model F1 has the highest mean value. The value is 0.8245. Finally, in the psychological intervention based on ideology and politics for students with psychological problems identified by the early warning model, the results show that the individual’s pressure t=8.163, df=6, p<0.001, and the social environment pressure t=9.872, df=6, p<0.001, indicating that the subject’s mental anomalies can be classified based on the time series. 0.001, indicating a significant difference between before and after the intervention for the subjects.