Solving the optimal diet problem necessarily involves estimating the daily requirements in positive and negative nutrients. Most approaches proposed in the literature are based on standard nominal estimates, which may cause shortages in some nutrients and overdoses in others. The approach proposed in this paper consists in personalizing these needs based on an intelligent system. In the beginning, we present the needs derived from the recommendations of experts in the field of nutrition in trapezoidal numbers. Based on this model, we generate a vast database. The latter is used to educate a deep learning neural network, the architecture of which we optimize by the fuzzy genetic algorithm method in the way of adopting a customized regulation term. Our system estimates nutrient requirements based only on gender and age. These estimations are integrated into a mathematical model obtained in our previous work. Then we again use the fuzzy genetic algorithm to draw up personalized diets. The proposed system has demonstrated a very high capacity to predict the needs of different individuals and has allowed the drawing up of very high-quality diets.