In this paper,the junction temperature of high-power LED array of th e LED 120W double inlet and outlet jet impingement water cooling system developed by the research group is predicted by combining t he genetic algorithm (GA) with BP neural network. Taking advantage of genetic algorithm to optimize the weights and threshold of B P network and BP algorithm for training the network can reduce the shortcomings of local minimum value and slow convergence spe ed of using BP network alone.And in the training process,in order to make the network output have space a long enough on ge netic algorithms for data processing,we make some improvements to the original data normalized to [0.050.95].The collected data is studied,trained and forecast by the model and the results show that the model can reflect the junction temperature of LED better.T he prediction accuracy is improved by 14.14% using the GA-optimized BP neural n etwork than that using the LM-optimized BP neural network.and the predicted effect is more stable.The junction temperature prediction model o f the BP neural network combined with genetic algorithm is more able to grasp the initiative of the heat dissipation structure design than the traditional measurement,and it has high practical value to extend the life of the high power LED.