The invention discloses a soft
package battery fault detection method and
system based on an adaptive neural network, and the method comprises the steps: constructing a battery temperature
spatial distribution model considering thermal anomaly and sensor faults, carrying out the homogenization of non-homogeneous boundary conditions through employing an auxiliary function
decomposition method, and building an ODE dimension reduction
system in combination with a
Galerkin method; the method comprises the following steps: acquiring surface temperature, current and
terminal voltage data of a battery, designing stable parameters through a
linear matrix inequality, estimating a nonlinear thermal generation item by adopting a neural network observer with a self-adaptive weight matrix, and constructing an error
system; a lumped residual error evaluation mechanism is introduced, a dynamic fault threshold value is generated based on a
kernel density estimation technology, and online detection is completed by comparing a residual error evaluation value of an output
estimation error with the threshold value in real time. According to the method, limitation of a traditional lumped parameter model is broken through, distributed thermal fault detection of the two-dimensional soft
package battery can be achieved only through a small number of thermocouples, the hardware cost is greatly reduced while precision is guaranteed, and a new technical path is provided for
power battery safety management.