A Study on the Optimal Design of Reinforced Learning-Driven Personalized Physical Training Strategies in Physical Education Instruction

Reinforcement learning is applied to recommender systems to balance the relationship between new and existing items and improve the accuracy of recommended items. In this paper, a personalized physical training strategy recommendation model combining LSTM and reinforcement learning is proposed to perform physical training strategy recommendation to optimize the physical education process. A SOM neural network is utilized to segment the physical fitness data of different students. On this basis, the recommendation model utilizes the LSTM long and short-term interest acquisition module to obtain user’s real-time preference and convert the sequence processing problem into a Markov decision process. Adding high and low scoring decision-making actions to the pseudo-twin network, delayed rewards were obtained, and noisy interaction records were removed. The SOM neural network clustering method obtained the characteristics of each class of students’ physical fitness, which paved the way for the recommendation of the subsequent individualized physical fitness training strategies. The recommendation model in this paper obtained Ave_RMSE and Ave_MAE values that outperformed other algorithms on two different datasets. The Ave_RMSE and Ave_MAE values are 56.68% and 68.70% higher than KNNWithMeans on the mL-1m dataset, respectively. There are similarities and dissimilarities between physical training strategies with higher recommendations and experimental subjects. The superiority of the recommendation model based on LSTM and reinforcement learning in physical education optimization has been demonstrated.