An improved algorithm for traffic anomaly detection in large-scale networks has been proposed to solve the current problem of detecting massive data and categorizing inhomogeneous data traffic. The algorithm combines the FCM algorithm and GRNN, using the FCM algorithm to cluster the data traffic samples, and then using the GRNN to train the convolution of the sample points closest to the FCM cluster center and iteratively update them until a stable cluster center is obtained; MFOA is introduced to perform parameter tuning of the FCM-GRNN, and the global optimization-seeking property of the MFOA and the three-dimensional spatial search method are utilized to Iterative optimization to find the optimal Spread value; using KDDCUP99 dataset for the test, it is concluded that the detection rate of the proposed algorithm is 91.36%, and the false detection rate is 1.154%, and the proposed algorithm has a better ability to detect abnormal traffic.