It is the crucial problem to accurately detect the traveling wave singularity point in fault location. Much noise is usually contained in the traveling wave signal of field data， in which case， the singularity point cannot always be detected using the conventional wavelet transform. Accordingly， a traveling wave signal processing method for singularity detection based on singularity value decomposition and wavelet transform is proposed. After the track matrix of an attractor reconstructed by time series is structured， a signal series without noise will be obtained by the optimal approximation matrix in the Frobenious norm， and the singularity point will be detected in the noise cancelled signal series. The simulation result shows that the method can maintain the singularity characteristic and accurately detect the singularity point in the noise background.