Abstract: In order to satisfy the requirement of DTN（delay/disruption tolerant network）in different types and different differential transmission weight data，a window selection probability function optimized method was designed based on adaptive genetic algorithm.The method used red/green（important/unimportant）data error rate as fitness function to complete cross compilation and global optimization，which allowed important data could be selected in a higher rate and could be transmitted in a higher accuracy；on the other hand，unimportant data would not be lost and could be transmitted within excepted error rate range.The method was used in the extended window unequal fountain code coding of LTP protocol in DTN network transport layer；compared to the results of the traditional genetic algorithm，optimization algorithm of adaptive genetic algorithm could decrease the important data error rate and channel total error rate in deep space communication，which could satisfy the requirement of differential data transmission，decrease data retransmission caused by high error rate，increase data transmission efficiency and save channel resources.

Key words: DTN network, adaptive genetic algorithm（AGA）, function of selection window, expanding window unequal protection（EWUP）, LTP protocol