The air quality in children's living spaces has a significant impact on their health, and ultrafine particulate matter (PM2.5), as one of the main pollutants in the air, poses a particularly prominent threat to children's respiratory and immune systems. Therefore, it important to conduct health risk assessment on ultrafine particle exercise in children's living spaces. In this article, an intelligent risk assessment model of ultrafine atmospheric particles in children's sports space in residential areas based on probabilistic neural network (PNN) is studied. This model utilizes probabilistic neural networks to monitor and predict ultrafine particles in residential spaces, evaluating their potential risks to children's health. By collecting relevant data such as particulate matter concentration and air quality index in residential spaces, and using probabilistic neural networks for training and prediction, accurate assessment of the health risks of ultrafine particulate matter during exercise can be achieved. This model can be applied to densely populated places such as families and schools, providing parents and relevant institutions with scientific risk warning and prevention measures to ensure the health and safety of children. Probabilistic neural networks can learn and simulate complex nonlinear relationships, making the model more accurate in predicting the concentration of ultrafine particles and health risks. So as to construct a method system of health risk assessment of atmospheric particles, and the model is heavy metal elements in atmospheric fine particles, and corresponding improvement measures are put forward according to the assessment results. For circulatory system diseases, the degree of harm of atmospheric temperature and fine particles to children's health is basically close, and with the continuous decline of temperature, the synergistic effect of the two is more obvious. The concentration of atmospheric ultrafine particles significantly affects the health risk of children's sports activities space. All the experiments in this article are quite good. Among them, the average accuracy of the PNN health risk identification model is 0.840, and the average recall rate is 0.837. The health risk assessment model in this article strengthens the correlation of data information through scientific methods, grasps the basic information of children's sports activity spaces in residential areas, and formulates targeted sports health risk assessments for children's sports activity spaces in residential areas. Compared with the comparison method, the response speed is significantly improved, with an accuracy rate of over 90%.