ing at the problems that the linear dimension reduction method of three-dimensional (3D) fluorescence spectra of algae is not ideal and the model recognition accuracy is low, a classification model is constructed by using local linear embedding (LLE) algorithm to reduce the dimension and using the golden sine algorithm (Gold-SA) to optimize the support vector machine (SVM). The 3D fluorescence spectrum data of algae after dimension reduction by LLE algorithm is used as the input of SVM, and other two dimension reduction methods are compared. The results show that LLE algorithm has the best dimension reduction effect and the highest recognition accuracy. In order to further improve category recognition ability, the Gold-SA is used to optimize SVM and establish a Gold-SA-SVM model, and the other four classification models are compared. The results show that the classification recognition accuracy, precision, recall rate, and F1 score of the Gold-SA-SVM model are significantly improved, and the method can accurately realize the classification of Aureococcus anophagefferens, Chlorella, and Synechococcus elongatus, providing an effective reference for the research of brown tide.