Research And Application On Prediction Of Power Transformer Fault Based On Stream Data Mining

In the 21 st century, with the development of China's economy, the improvement of people's material living standards and rising utilization of electrical equipment, the demand for power resources continue to increase. Thus, in recent years, the scale of China's power grid system and the scale of delivery of power resources are also rising,resulting in a load of increasing electrical equipment. The power transformer plays a vital role in the course of transmission power resources and it is also the fault-prone equipment. When the power transformer failure, it causes power outages throughout the region, that not only affect people's lives, but also may affect industrial production,resulting in a significant impact on the economy. So the power transformer fault prediction is a very meaningful work. Based on the above situation, this paper starts on the study from four aspects.Firstly, the data for the characteristics of power transformer were studied.Analyses found that the power transformer was multi-dimensional data stream timing data, by introducing the idea of time granularity to establish the appropriate timing for power transformers multi-dimensional model. Based on this model, further study had been cast about the relationship between various dimensions and the structure of multi-granularity time from the perspective of the prediction.The study provided a theoretical basis based on a timing stream data analysis.Secondly, the research on integrated learning power transformer fault prediction algorithm model was studied. In this paper, an integrated learning ideas had been promoted to avoid a single machine learning limitations and one-sidedness. By introducing the thought of integrated learning, the failure prediction model integrated learning was established. This model with one-dimensional time series data stream and a comprehensive analysis could give an exact forecast. This article studied the multi-dimensional time series fault prediction based on data flow that provided an ideological foundation.Thirdly, the multiple classifiers competitive strategy was study. The power transformer was multi-dimensional data stream timing data and the integratedlearning thought only a dimensional data to predict each time. It may occur the process of data to predict all dimensions in two different predictions sets, and failure prediction results must be unique, so you must use a fault prediction mechanism to obtain only prediction results in the two forecast results. In this study, power transformer fault prediction solutions was provided.Finally, in order to improve power transformer fault prediction effect,multi-granularity of time thinking and a variety of data processing means for the raw data preprocessing were study. Comparing experimental results, it demonstrates the superiority of integrated learning power transformer fault prediction model algorithm and the applicability of multiple classifiers competitive strategy.