This paper proposes a novel method for estimating lower limb muscle forces using a CNN–LSTM neural network combined with a wearable sensor system. The method involves developing a wearable sensor system to collect joint angles and angular velocities during walking, and using these data as inputs for the neural network model. The muscle forces calculated using OpenSim based on the Static Optimization (SO) method were used as the standard values to train the neural network model. Four lower limb muscles of the left leg, including gluteus maximus (GM), rectus femoris (RF), gastrocnemius (GAST), and soleus (SOL), were selected as the study subjects. The results showed that the CNN–LSTM model performed better than standard CNN and LSTM in muscle force estimation under slow, medium, and fast walking speeds. The average correlation coefficients between true and estimated values of the four muscle forces were 0.9801, 0.9829, and 0.9809, respectively. The model demonstrated good robustness and generalization, performing well even when the estimated object was not included in the training data. The method provides a convenient way for estimating muscle forces, which can assist in quantitative analysis of human motion and muscle injury. The model establishes the relationship between joint kinematic signals and muscle forces during walking, offering a more efficient and convenient alternative to the SO method used in OpenSim for muscle force calculation. The study also evaluated the performance of the CNN–LSTM model in different scenarios, including intrasession and intersession tests, showing that the model had minimal impact from individual differences and strong universality. The results indicate that the proposed method is robust and generalizable, with potential applications in gait evaluation and rehabilitation equipment development.This paper proposes a novel method for estimating lower limb muscle forces using a CNN–LSTM neural network combined with a wearable sensor system. The method involves developing a wearable sensor system to collect joint angles and angular velocities during walking, and using these data as inputs for the neural network model. The muscle forces calculated using OpenSim based on the Static Optimization (SO) method were used as the standard values to train the neural network model. Four lower limb muscles of the left leg, including gluteus maximus (GM), rectus femoris (RF), gastrocnemius (GAST), and soleus (SOL), were selected as the study subjects. The results showed that the CNN–LSTM model performed better than standard CNN and LSTM in muscle force estimation under slow, medium, and fast walking speeds. The average correlation coefficients between true and estimated values of the four muscle forces were 0.9801, 0.9829, and 0.9809, respectively. The model demonstrated good robustness and generalization, performing well even when the estimated object was not included in the training data. The method provides a convenient way for estimating muscle forces, which can assist in quantitative analysis of human motion and muscle injury. The model establishes the relationship between joint kinematic signals and muscle forces during walking, offering a more efficient and convenient alternative to the SO method used in OpenSim for muscle force calculation. The study also evaluated the performance of the CNN–LSTM model in different scenarios, including intrasession and intersession tests, showing that the model had minimal impact from individual differences and strong universality. The results indicate that the proposed method is robust and generalizable, with potential applications in gait evaluation and rehabilitation equipment development.