Pedestrian flow early-stage prediction method based on space-time diagram convolution

The invention discloses a pedestrian flow early-stage prediction method based on space-time diagram convolution. The method comprises the steps of obtaining a training sample of pedestrian flow, constructing an encoder-decoder structure employing diffusion convolution of an undirected graph, constructing a controller for controlling the step length of historical data, and carrying out iterative training on the early-stage prediction model based on the space-time diagram convolution to obtain a pedestrian flow early-stage prediction result. Real-time data samples collected by sensors are input into the model, sample features are extracted through an encoder, a decoder is initialized through output of the encoder, a controller judges whether the current moment meets the condition of the optimal step length or not according to the hidden state finally output by the decoder, if yes, a final predicted value is obtained through a full connection layer, namely the number of pedestrians passing through each sensor ar