In recent years, our Marine economy has been rapidly developed, and the proportion of our national economy has been increasing year by year, but it faces deep structural imbalance, mainly because of the high energy consumption industry in our Marine economy, Marine ranching is one of them. Only through the study of energy consumption can we improve energy efficiency and management level, and then reduce energy consumption. Therefore, the energy consumption research of Marine ranches is imminent. For a long time, the lack of research on the prediction of energy consumption of Marine ranching has been restricting the scientific management of Marine ranching and the improvement of energy consumption efficiency. To solve this problem, a time series prediction model based on generative adversarial network (GAN) and Long Term and Short Term memory network (LSTM) was proposed for the prediction of energy consumption in Marine ranching. Firstly, the generative adversarial network is used to enhance the data set and generate new data, so as to improve the measurement accuracy. Then, the long-term and short-term memory neural network is used to predict the future energy consumption. Finally, in order to verify the validity of the model, the data set of a Marine ranching was used for experiments and compared with the SVR model and LSTM model. Experimental results show that compared with other models, GAN-LSTM has higher accuracy in predicting data, and the data predicted by GAN- LSTM is often more real and accurate than that by traditional models.