In auxiliary training of youth running, traditional methods often find it difficult to accurately capture the complex posture changes during the movement. To solve this problem, this study proposes a deep learning algorithm framework that combines convolutional neural networks (CNN) and long short-term memory networks (LSTM), and optimizes it by integrating attention modules to enhance the model’s ability to capture key action details and improve recognition accuracy. The study first builds a human posture motion model to capture the posture changes of adolescents during running. Then, a deep learning model is designed and implemented, which uses CNN to extract spatial features and LSTM to process time series data to capture the spatiotemporal characteristics of running. On this basis, the attention module is further integrated to enhance the model’s ability to capture key action details by learning important areas in the image space and key points in the time series. Experimental results show that the CNN-LSTM model integrated with the attention module shows significant results in assisting adolescent running training. The completion time of one kilometer running has decreased among teenagers who use this model to assist training. The average completion time is 165s-338s, while it is 244s-369s under the traditional method. In addition, teenagers using the model recovered from fatigue an average of 81.7 s faster than those using traditional methods, showing the advantages of deep learning models in auxiliary training.