To　improve　the　forecasting　accuracy　of　spatial　panel　data　with　the　spatial　autocorrelation,　the　significance　of　spatial　autocorrelation　by　Moran　coefficient　is　tested　by　Z-statistic.　The　panel　data　is　considered　as　a　set　of　time-series　data,　and　the　genetic　algorithm　back-propagation　(GABP)　neural　network　forecasting　model　is　established.　And　the　panel　data　is　considered　as　a　set　of　cross-sectional　data,　Kriging　algorithm　forecasting　model　is　established.　Then　the　combination　forecasting　model　of　panel　data　is　established　by　the　results　of　two　models,　and　weighted　by　the　new　approach　of　information　entropy.　An　empirical　study　is　carried　out　with　some　counties＇　lever　of　township　in　Fujian　2007,　China.　The　result　shows　that　the　combination　forecasting　model　of　spatial　panel　data　based　on　information　entropy　is　the　most　effective　one.