The invention relates to the field of power
system recovery, and discloses a black-start path optimization method, which comprises the following steps of: recursively searching an initial path meeting all preset constraint conditions by taking a black-start power supply as an initial node through a
backtracking algorithm to realize full-path traversal so as to avoid missing a potential optimal solution; a graph
tensor is constructed based on node comprehensive
feature data and edge connection
feature data, a
time sequence tensor is constructed in combination with dynamic
time sequence data, spatial-temporal feature expression of a path is formed, a pre-trained
hybrid neural
network model is utilized to quickly evaluate a path comprehensive
score, a candidate path with the highest
score is screened out, and the path comprehensive
score is evaluated. And finally, the candidate path is accurately verified through electromagnetic transient
simulation, and an optimal path is determined. According to the method, path feasibility constraints are strictly considered, online path adjustment is supported to cope with dynamic changes such as
energy storage attenuation, robustness is achieved, path evaluation is accelerated through the
hybrid neural network, the
simulation verification frequency is greatly reduced, and the search efficiency is remarkably improved.