Concept Drift and Avoiding its Negative Effects in Predictive Modeling of Failures of Electricity Production Units in Power Plants

Abstract

Ensuring the required accuracy of predictive models operating on time series is very important for industrial diagnostics systems. It is especially visible if there are a lot of models covering hundreds of devices and thousands of measurements operating under varying conditions in changing environments. In this work, we analyze the concept drift phenomenon in the context of actual measurements and predictions of the diagnostic system of boiler feed pump working in coal-fired power plants. In the practical part, we adapt algorithms and techniques operating on time series to obtain better results and reduce the negative effects of the concept drift. The results of our experiments show that the application of drift handling methods brings improvement in the effectiveness of the fault prediction process.