To improve the prediction accuracy of dissolved oxygen content in ponds, a novel long short-term memory (LSTM) optimized by an improved beetle antennae search algorithm (IBAS) was proposed. Firstly, Pearson correlation coefficient was used to obtain the linear correlation between each factor and dissolved oxygen. The key impact factors of dissolved oxygen were selected by Pearson correlation coefficient as the input feature, which can reduce the input dimension, eliminate the correlations of original variable, and improve the calculation efficiency of the model. Secondly, to balance the global search and local search, and improve the convergence speed of beetle antennae search algorithm (BAS), an IBAS with exponential decreasing strategy of attenuation factor was proposed, which linked the attenuation factor eta with the number of iterations. Finally, LSTM network was optimized by IBAS to get the best parameter combination strategy to construct a P-IBAS-LSTM prediction model between dissolved oxygen and these factors. Based on the presented model, the dissolved oxygen was predicted for an experimental pond during April 28 th to September 8 th, 2020 in the Research Center of Yixing City, Jiangsu Province. In the case of the same data, the mean squared error (MSE), root mean square error (RMSE), and the average absolute error (MAE) of the P-IBAS-LSTM were 0.6442mg2/L2, 0.8026mg/L, 0.5306mg/L, respectively. The experimental results showed that the proposed model of P-IBAS-LSTM had higher performance and stronger generalization performance when compared with common prediction models, which could meet the actual needs of predicting dissolved oxygen accurately and help farmers make decisions in ponds.