The　investigation　about　the　cross-correlation　among　financial　assets　has　drawn　broad　attention　recently.　Due　to　the　nonlinear　and　non-stationary　identities　of　the　financial　time　series,　e.g.,　stock　return　time　series,　the　cross-correlation　for　different　level　of　fluctuations　are　quite　important　for　both　academia　and　financial　practitioners.　Here　we　use　the　empirical　mode　decomposition　(EMD)　method　to　analyze　the　cross-correlation　structure　among　different　level　of　fluctuations　for　financial　assets.　The　correlation　based　networks　are　then　employed　to　determine　the　clustering　property　of　stock　market.　We　then　propose　several　portfolio　optimization　strategies　based　on　the　EMD　correlation-based　networks.　Using　the　topological　information　of　the　networks,　we　can　construct　some　portfolios　with　high　return　and　low　risk.　Under　two　portfolio　evaluation　frameworks,　we　prove　that　these　portfolios　have　consistently　good　performance.　(C)　2019　Elsevier　B.V.　All　rights　reserved.