Accurate prediction of hip flexion cycle is critical to realize stable and consistent walking assistance using wearable pneumatic artificial muscles (PAM) to decrease physical fatigue of older adults and patients with mild motor disorders. In this study, a particle swarm optimization-long short-term memory (PSO-LSTM) algorithm is employed to predict hip flexion cycle, which is further used to control pneumatic artificial muscle assisting time. Compared to LSTM model, PSO-LSTM model demonstrates the significant improvement in fitting performance and predictive accuracy, where R-square value and root mean squared error (RMSE) of prediction results are 0.9998 and 0.0023, respectively. At other walking speeds, PSO-LSTM model can also demonstrate considerable predictive performance. Furthermore, a novel nylon-reinforced pneumatic artificial muscle is designed and fabricated, which shows a remarkable performance at response speed. Experimental results show that its response frequency can reach 2.0 Hz. Consequently, it can meet the hip flexion assisting requirement of the wearer when walking at 1.00, 1.25, and 1.50 m/s speeds. This study showcases a solution to predict hip flexion cycle for realization of effective assistance of wearable devices.