The invention discloses a power network alarm space feature analysis method based on an improved entropy weight method. The method comprises the following specific implementation steps: S1, dividing an alarm transaction set by adopting a sliding time window algorithm; S2, carrying out weighting processing on the alarm data in each alarm transaction set by adopting an improved entropy weight methodin combination with the actual topology of the network; S3, further dividing the alarm transaction set into subsets with spatial characteristics according to the weights. Through the above processing, each alarm transaction set with time characteristics is further divided into subsets with strict correlation in space. According to the invention, the alarm transaction set is divided by adopting asliding window algorithm. Massive alarm information is analyzed from a small granularity. The analysis efficiency is improved, an entropy weight method is optimized according to the characteristics ofalarm data in a power grid. The optimized entropy weight method is used for weighting each alarm data, and finally analysis is started from equipment sending an alarm with the maximum weight, so thatthe alarm space feature extraction efficiency is improved.