The invention discloses a high-
voltage direct-current
submarine cable insulating layer transient thermoelectric risk
time sequence progressive prediction method, and aims to solve the problems that an
electric field and temperature of a
submarine cable insulating layer in a deep and far sea
wind power system are difficult to monitor in real time, and a traditional prediction method is low in efficiency or lacks physical significance. The method comprises the following steps: constructing a transient electromagnetic-thermal multi-
physics field
coupling model, and generating a
time sequence structure
data set containing a physical mechanism by taking a
wind power historical dynamic load as a boundary condition; establishing a PSO-BP neural
network model, and obtaining a high-precision prediction model through
data set training; based on a sliding data
cellular mechanism, real-time load data are fused to realize
time sequence progressive prediction, four-level risk early warning is triggered in combination with
field intensity and temperature safety coefficients, and a model is calibrated regularly by using new data. According to the method, the advantages of physical modeling and
machine learning are fused, prediction time consumption reaches the second level, efficiency and precision are considered, the defects that a pure data driving model is poor in extrapolation and a common BP neural network is prone to local optimization are overcome, risk advanced
perception is achieved, and support is provided for
submarine cable operation and maintenance.