For improving the prediction accuracy of dissolved gas concentration in oil and providing theoretical basis for early fault warning and formulation of maintenance plans of transformer

in this paper a prediction model based on multi-decomposition strategy and Bayesian optimized long short-term memory neural network is proposed. Firstly

the original time series of dissolved gas in oil is decomposed by using the improved complete ensemble empirical mode decomposition with adaptive noise algorithmso as to obtain a series of subsequence components at different time scales. Then

the high frequency

IMF

1

component

which is the most difficult to predict

is decomposed by variational mode decomposition algorithm to reduce it's nonlinearity and non-stationarity further. After that

the Bayesian optimized long short-term memory neural network is used to model all the subsequence components after second decomposition and performe single step prediction and recursive multi-step prediction. Finally

the prediction results of all components are superposed and reconstructed to obtain the final single step and multi-step prediction results. The results of examples show that the method proposed in this paper can effectively improve the prediction accuracy of dissolved gas concentration in oil. The proposed method

compared with the other six comparison models

exhibits the best performance in single step and multi-step prediction and its effectiveness is fully verified.