Ethylene production plays an important role in the petrochemical industry. Due to the high dimensionality of industrial data, the established data models are complex and contain a large number of hyperparameters. The selection of different hyperparameters of the neural network model has a great influence on the prediction accuracy. Therefore, a long short-term memory (LSTM) integrating Snake Optimizer algorithm (SOA) is proposed to achieve accurate prediction of ethylene production in complex chemical production processes. LSTM extracts the temporal dynamic interaction feature to predict ethylene production in a complex ethylene production process. Then, the SOA optimizes the hyperparameters of the LSTM as optimization variables including number of hidden layer nodes, batch size and training times, which can adaptively adjust the moving step size according to the current search state and target state of the solution vector. The root mean squared error (RMSE) is token as the optimization target of the SOA to obtain the global optimal solution of LSTM hyperparameters. Compared with back propagation (BP) neural network, extreme learning machine (ELM), and LSTM in the ethylene data, the SOA-LSTM achieve the state-of-the-art results. The RMSE, the MAPE and the MAE of SOA-LSTM are 4847.60, 4.10%, and 3593.54 respectively. Through the accurate prediction of ethylene production in the complex ethylene production process, its production efficiency can be effectively improved and carbon emissions can be reduced.