Some prediction models of chloride ion concentration in concrete of the transmission tower pile foundations were established based on machine learning algorithm. These models were tested through correlation coefficient, root mean square error, mean absolute error and variance ratio， and the robustness of the models were analyzed according to Monte Carlo simulation. At the same time, the models were optimized based on sea-horse optimizer. The results show that the support vector machine (SVM) model, the random forest (RF) model and the gradient boosting decision tree (GBDT) model can accurately predict the chloride ion concentration in the concrete of the transmission tower pile foundations. The correlation coefficient R2 is greater than 0.880, the root mean square error is less than 0.009, the mean absolute error is less than 0.006, and the variance ratio is greater than 0.890 for all these prediction models. According to the results of error and robustness analysis, it is recommended to prioritize the use of the GBDT model and SVM model for the prediction of chloride ion concentration in concrete. According to the optimization results, the sea-horse optimizer can significantly improve performance of model.