Abstract: In thermal power units, the operation efficiency of the cold end system directly affects the economy of the whole unit. Through the optimization of the operation mode of the cold end system, the net power output of the unit can be maximized. Neural network have been introduced into the comprehensive optimization of thermal power unit cold end system, and the cold end optimization strategy is proposed. Taking specific thermal power generation unit as the object, the optimization mode and operation principle of the cold end system are analyzed, and the coupling system consisting of core equipment such as circulating pump, condenser and turbine is established. A power plant example is used to analyze and verify the model, and a neural network is introduced to solve the problem of parameter confirmation of multi-input and multi-output model. The final prediction model is obtained. The results show that the prediction error of the model is less than 1% and the prediction effect is good. Based on this prediction model, genetic algorithm is used to find the optimal circulation pump operation combination, which provides a reference for the optimization of cold end operation of thermal power units.