The invention relates to an intelligent
energy exchange collaborative prediction method based on space-time
big data, and belongs to the technical field of traffic and energy fusion. The invention provides a cross-domain collaborative space-time diagram neural
network model for solving the problem that a cross-
energy coupling relationship under a vehicle-road collaborative architecture is difficult to depict due to the fact that existing
traffic flow and
power grid load are mostly subjected to single-domain modeling based on a single
data source. The method comprises the following steps: firstly, constructing a double-
branch space-time diagram
convolution encoder of a traffic domain and an
electric energy domain, and respectively extracting deep space-time features from traffic and
power grid data; secondly, designing a hidden
space mapping and self-adaptive gating mechanism, and carrying out alignment and weighted fusion on the cross-energy characteristics; thirdly, introducing a cross attention fusion module, and adaptively modeling a time-varying
coupling relationship between the two domains; and finally, constructing a multi-task joint
loss function composed of
traffic flow prediction loss,
power grid load prediction loss and cross-domain feature consistency loss, performing joint optimization training on
model parameters, and outputting a joint prediction result of the
traffic flow and the power grid load. According to the method, traffic-power grid
coupling information is fully utilized, the accuracy and robustness of joint prediction are improved, and decision support can be provided for
traffic management and power grid dispatching optimization.