Traffic congestion appears with different shapes and patterns that may evolve quickly over time. Static spectral clustering techniques are unable to manage these traffic variations. This paper proposes an evolutionary spectral clustering algorithm that partitions the time-varying heterogeneous network into connected homogeneous regions. The complexity of the algorithm is simplified by computing similarities in a way to obtain a sparse matrix. Next, the evolutionary spectral clustering algorithm is applied on roads speeds in order to obtain clusters results that fit the current traffic state while simultaneously not deviate from previous histories. Experimental results on real city traffic network architecture demonstrate the superiority of the proposed evolutionary spectral clustering algorithm in robustness and effectiveness when compared with the static clustering method.