Prediction Of Remaining Life Of Power Transformer Based On Improved Hidden Markov Mode

In the context of China’s "carbon peak and carbon neutrality" goals,the development of the electric power industry has attracted significant attention.Transformers,as key equipment for energy conversion and transmission,hold an important position.Given that transformers are complex systems with varying states,a malfunction can result in a widespread impact.Therefore,timely evaluation and prediction of the remaining lifespan of transformers is essential in devising effective maintenance strategies,reducing the occurrence of failures,and enhancing the reliability of power system operations.Most existing research on transformer maintenance relies on expert experience and tends to classify transformer operation as either normal or faulty states,overlooking the multiple possible changes that may occur during operation.As a result,real-time monitoring and maintenance scheduling based on the actual state cannot be realized.This article focuses on studying the state changes of transformers and utilizes historical operating data to carry out the following research:Firstly,based on the aging law and aging state parameters of transformers,the operating state space of transformers is constructed,with defects being introduced and the operating state being classified into five states: normal operation,minor defect state,major defect state,emergency defect state,and faulty state.In addition,based on the defect transfer rate and failure rate,a remaining lifespan prediction model for transformers is constructed.Secondly,a comprehensive transformer remaining life prediction model was constructed based on the Dissolved Gas Analysis(DGA)and Hidden Markov Model(HMM).By combining the different operating states of the transformer with the hidden states in the HMM,the model parameters were trained to achieve state evaluation.To address the issue of the HMM’s tendency to produce local extremes,the Particle Swarm Optimization(PSO)algorithm was introduced,leading to the development of the PSO-HMM model,which effectively improved the accuracy of state evaluation.In addition,a proportional hazard model was used to predict the remaining life of the transformer.The proportional hazard rate model selected a fault rate function based on the temperature rise aging model and a connection function based on DGA data,with parameter estimation performed using maximum likelihood estimation.This method provided a comprehensive and objective reflection of the transformer’s condition.Finally,However,due to the complex and varied operating conditions of transformers,it is difficult to reflect their actual operating conditions using only a single DGA data.To make the prediction model more applicable,it is necessary to consider other factors that may affect the transformer’s lifespan.On the basis of DGA information,multiple types of data,including preventive tests and inspections,were comprehensively considered.Because these data were collected at different time scales,the data from the defect record form for power grid transformers were used as input.Then,a time-varying defect transition rate was constructed to replace the static fault rate.Based on this defect transition rate and Markov transition matrix,a transformer remaining life prediction model was constructed,and the reliability function and remaining life prediction value of the transformer were solved according to the predicted defect transition rate.Furthermore,the impact of the rollback correction model was also considered.Through case analysis,it was demonstrated that this method provides a more realistic prediction of remaining life,thereby improving the accuracy of transformer remaining life prediction.