The invention discloses an online optimization control method for a tram. The method comprises the steps of: collecting and recording actual operating parameters of the tram line, and clustering parameter data to obtain a line parameter sample; using the neural network to train the line parameter sample based on line expectation requirements, and outputting the control parameters for ensuring thepunctuality and the comfort of the train; adjusting the train operating parameters in real time according to the control parameters, and simultaneously performing energy recovery braking. The method can bring the best riding experience under the premise of ensuring the punctuality, and greatly optimizes the driving comfort and the punctuality of the tram.