Firstly, the concepts of the traveling wave entropy and the feature function oftraveling wave entropy were defined. Then the statistic characters of the traveling waveentropy feature function, mean value and variance were analyzed after the zero-ordercomponent of the traveling wave of online cable was selected to serve as the observedobject. Finally, the new recognition algorithm of minimum risk neural network was pre-sented. The simulation experiments show that the recognitions of the early fault states canbe completed correctly by using the proposed recognition algorithm. The classes of cablefaults include in 1-phase ground faults, and the 2-phase short circuit faults or ground faults,and the 3-phase short circuit faults or ground faults, open circuit. The fault resistancerange is 1×10-1-1×109Ω.