Reliability-driven clustering methodology for probabilistic forecast of environmental conditions in power electronics applications

Abstract

Modern power systems are today characterized by an increasing utilization of the power electronics. An accurate predictions of the power electronics reliability and lifetime are crucial to avoid unforeseen power outages and maintenance costs. One of the critical parts of the lifetime prediction is the representation of the environmental conditions in terms of mission profile. In current long-term system planning, the forecast of the generation and load define the mission profile of power electronics. Moreover, the forecast methods are not suitable for prediction of the environmental conditions for extended prediction horizons and high temporal granularity. Hence, a new reliability-driven method for probabilistic forecast is proposed. The method incorporates the reliability of power converters within the clustering procedure. In such a way, it is assured that the predicted mission profiles will result in a lifetime prediction with a sufficient accuracy. The analysis results indicate a superior accuracy compared with the commonly used mission profile-based procedure for reliability and lifetime estimation. Thus, the method can be used for a variety of power electronics applications, where accurate prediction of power electronics lifetime can assist in a more optimized and cost-effective system design as well as operation.

Keywords

• Mission profile
• Clustering
• Probabilistic forecast
• Power electronics
• Reliability
• Long-term system planning