In recent years,real time precise point positioning (PPP) technique has become a hot topic in the field of satellite positioning.As for PPP,real time precise satellite coordinates and clock errors are indispensable.Among the products,final ephemeris cannot be available in real time; ultra-rapid ephemeris cannot meet the cm-level positioning accuracy requirements and the data stream of real time clock error correction service often has the problem of service interruption or incomplete information.Therefore,the research has important significance to forecast real time precise satellite clock error.At present,almost all the methods of satellite clock error modeling and forecasting are based on mathematic algorithm which use polynomial,grey model and neural network algorithm to establish mathematical model and don’t consider the effect of physical factors on satellite clock.However,frequency stability of satellite clock is influenced by acceleration of satellite and the instantaneous clock error is easily affected by these factors.In view of this,this paper puts forward a comprehensive consideration of satellite acceleration and final clock error products,using neural network to build a clock error prediction model.Experimental GNSS satellite clock error data processing results show that this method can improve the prediction accuracy of GNSS satellite clock error modeling and satisfy the demand of real time PPP.