This　paper　analyzes　the　optimal　strategies　for　an　attacker　and　a　defender　in　an　attack-defense　game　on　a　network　consisting　of　interdependent　subnetworks.　The　defender　moves　first　and　allocates　its　resource　to　protect　the　network　nodes.　The　attacker　then　moves　and　allocates　its　resources　to　attack　the　network　nodes.　The　binary　decision　diagram　is　employed　to　obtain　all　potential　states　of　the　network　system　after　attack.　Considering　each　of　its　opponent＇s　strategies,　the　game　player　tries　to　maximize　its　own　cumulative　prospect　value.　The　backward　induction　method　is　employed　to　obtain　the　game　players＇　optimal　strategies,　respectively.　Different　resource　relationships　are　analyzed　to　testify　the　robustness　of　the　main　conclusions　and　players＇　risk　attitudes　are　also　investigated.　Numerical　examples　are　used　to　illustrate　the　analysis.