The invention discloses a
transformer fault diagnosis method based on
deep learning, and belongs to the field of
transformer fault diagnosis. The method comprises the following steps: firstly, carrying out duplicate removal and abnormal value
processing on concentration data of fault characteristic gases H2, CH4, C2H2, C2H4 and C2H6 collected by an
analysis method of dissolved gas in oil, fillingmissing values by using a
random forest method, and then carrying out normalization
processing on the data to form a
training set sample and a
test set sample; establishing a three-layer stack type sparse
noise reduction auto-
encoder model, and
rewriting a
cross entropy loss function in a traditional classification model into a Focal
loss function; according to the method, hyper-parameters are determined through class sample weights, white
Gaussian noise is added into input, an auto-
encoder is made to fully extract effective features, and therefore an effective
feature extraction model is obtained, and a Softmax classifier is used for outputting a diagnosis result of the model. Compared with the existing methods such as a three-
ratio method, an SVM and a BP neural network, the transformerfault diagnosis method provided by the invention has good diagnosis performance, and the accuracy of
transformer fault diagnosis is effectively improved.