The invention discloses an RBM-based comprehensive feature extraction method and a device for the log information of an electric power information system. The method comprises the following steps of acquiring the log information of each equipment in a current system; for each type of logs r, constructing an initialization restricted boltzmann machine neural network RBMr and carrying out training,wherein the input data is the data of logs r; optimizing the network weight according to the fast learning method for the comparison degree, and obtaining a stable RBMr; constructing a restricted boltzmann machine neural network RBMcom used for extracting comprehensive features, initializing the RBMcom, splicing the output data of a RBMcom hidden layer and forming a data sequence; training the RBMcom and obtaining the stable RBMcom; based on a well trained restricted boltzmann machine neural network set, inputting various types of log data, and obtaining comprehensive feature data, wherein thedimension of the data is C. By means of the method and the device, the problem of the log data heterogeneity can be solved, and the dimensionality reduction is achieved. Meanwhile, integrated features after fusion are obtained. The prediction speed and the prediction accuracy of the safety situation of the electric power information system can be effectively improved.