Session-based recommendation has received extensive attention in recent years and has been widely applied to various scenarios in the real world, and next item recommendation is one of the most popular subfields of session-based recommendation. Benchmark metrics for verifying the quality of next item recommendation include hit rate, average inverse rank and normalized folding cumulative gain, etc. However, most existing methods tend to use cross-entropy loss to train recommendation models, with insufficient consideration of the impact of evaluation metrics, and model training is difficult to directly improve the actual evaluation metrics, resulting in inconsistency between optimization goals and ranking metrics, and causing suboptimal recommendation quality of the recommendation system. In order to solve the above problems, the method of directly optimizing evaluation metrics has been widely used in the field of ranking learning in recent years, and the existing methods perform single metric optimization for MRR or NDCG metrics. We propose a model of session recommendation method with multiple evaluation metrics constraints, aiming to alleviate the problem of inconsistency between the optimization objective and evaluation metrics to optimize the next item recommendation. In order to narrow the gap between the objective function and the evaluation metrics and better use the recommendation model to complete the next item recommendation, the model integrates the HR, MRR and NDCG metrics, and constrains the optimization direction of the cross-entropy loss function in the form of regularization terms. Comprehensive experiments on two real-world datasets validate the effectiveness of our proposed approach