The invention provides an electric load mode extraction method and system, a computer device and a medium. The method comprises the following steps: acquiring a user electric power consumption data sequence at a preset sampling time interval to obtain an N-dimensional array; Wherein N is a positive integer; Obtaining the neighborhood radius by using the preset mathematical relationship between thedistance between the N-dimensional arrays and the neighborhood radius; Calculating the number of neighborhood arrays of the N-dimensional array by utilizing the neighborhood radius, so as to obtain the MinPts by utilizing the preset mathematical relationship between the number of neighborhood arrays and the MinPts; Utilizing the neighborhood radius and the MinPts to classify the N-dimensional array corresponding to each user by a density-based spatial clustering algorithm; Average the same dimension of the same kind of N-dimensional array to get the corresponding average load array. The improved clustering algorithm does not need to provide a specific clustering number in advance, can cluster according to the distribution density of the data itself, and the application range is wide an application is flexible.