The invention belongs to the field of computer vision and deep learning, particularly relates to a human skeleton behavior recognition method, system and device based on a graph convolutional neural network, which aims to solve the problem that the human skeleton behavior recognition result based on the graph convolutional neural network is low in precision. The method comprises the following steps of acquiring a skeleton video frame and normalizing the skeleton video frame; constructing a human joint natural connection graph corresponding to each frame of graph; learning the unnatural connection edge to obtain a human body joint connection diagram; allocating a weight value to each edge of the human body joint connection diagram; carrying out graph convolution operation to obtain the spatial information of the skeleton sequence; and carrying out convolution operation on the time dimension to obtain the behavior category of the skeleton sequence. According to the method, the natural connection edge can learn the basic human body behavior characteristics, the non-natural connection edge can learn the additional behavior characteristics, a graph is formed by the natural connection edge and the non-natural connection edge together, the human body motion information can be more fully represented, and the recognition performance is improved.