The temperature of the high voltage power cable can reflect the operation status of the cable, and the prediction of the joint temperature can improve the safe operation level of the cable. Using the least squares support vector machine to establish the temperature prediction model for cable joint. Cable joint history temperature, environment temperature, environment humidity and wire core/sheath current ratio can be adapted as the input samples, the surface temperature of the cable joint for the output. Particle swarm optimization algorithm is adopted to dynamically optimize the normalized parameter and regularization parameter to improve the accuracy of prediction, and the concrete steps of the prediction method are given. A 110 kV cable joint in Shanghai is used as an example. The results prove that method can predict the temperature of cable joint with high prediction accuracy. It can also provide a reliable basis for cable temperature detection and early warning system.