In order to make healthy and personalized diets conveniently, a dietary recommendation system for Chinese cuisine is proposed. Firstly, the ResNet-18 neural network model that can classify three meals was trained through the self-built database, and the three-meal confidence of 2167 Chinese dishes was obtained as the basis for recommendation. Then, the optimization model was constructed, where the information of nutrients, ingredients and categories of dishes were abstracted as decision variables, and the objectives of nutrition and eating habits were set as constraints. The optimal solution of the recommended dishes and their grams was calculated by using the adaptive Gurobi optimizer. Experiments show that the system can achieve the set goals at three levels: at the nutrient level, it can meet the intake targets of 19 nutrients for healthy users or 30 chronic patients; At the level of food ingredients, it can ensure the food diversity, complete the selection of favorite dishes and the avoidance of taboo ingredients; At the dish level, it can realize the function of pairing dishes and recommending three meals respectively.