Abstract: To address the modeling problems of complex hysteresis characteristics of flexible joints in lightweight industrial robots, an improved TCN (temporal convolutional network) hysteresis model for robot flexible joints based on phase lead compensation is proposed to describe the hysteresis characteristics of flexible joints between motor drive current and torsional angle. On the basis of restoring the joint hysteresis characteristic data after removing noise interference through Kalman filtering, research focuses on the following an improved TCN hysteresis model is designed, which incorporates dilated convolutions with different factors into each branch of the model to address the significant error issue at extremities points in the TCN neural network hysteresis model. In response to the residual errors caused by phase lag in the improved TCN hysteresis model, a phase lead component is designed and concatenated with the TCN neural network model, thereby constructing an improved TCN neural network hysteresis model with phase lead compensation and further enhancing the accuracy of the hysteresis model. Compared with the improved TCN hysteresis model and classical PI hysteresis model, the experimental results have verified the proposed robot joint hysteresis model with high accuracy.