Type-2 fuzzy logic system applications for power systems

Abstract

In the move towards ubiquitous information & communications technology, an opportunity for further optimisation of the power system as a whole has arisen. Nonetheless, the fast growth of intermittent generation concurrently with markets deregulation is driving a need for timely algorithms that can derive value from these new data sources. Type-2 fuzzy logic systems can offer approximate solutions to these computationally hard tasks by expressing non-linear relationships in a more flexible fashion. This thesis explores how type-2 fuzzy logic systems can provide solutions to two of these challenging power system problems; short-term load forecasting and voltage control in distribution networks. On one hand, time-series forecasting is a key input for economic secure power systems as there are many tasks that require a precise determination of the future short-term load (e.g. unit commitment or security assessment among others), but also when dealing with electricity as commodity. As a consequence, short-term load forecasting becomes essential for energy stakeholders and any inaccuracy can be directly translated into their financial performance. All these is reflected in current power systems literature trends where a significant number of papers cover the subject. Extending the existing literature, this work focuses in how these should be implemented from beginning to end to bring to light their predictive performance. Following this research direction, this thesis introduces a novel framework to automatically design type-2 fuzzy logic systems. On the other hand, the low-carbon economy is pushing the grid status even closer to its operational limits. Distribution networks are becoming active systems with power flows and voltages defined not only by load, but also by generation. As consequence, even if it is not yet absolutely clear how power systems will evolve in the long-term, all plausible future scenarios claim for real-time algorithms that can provide near optimal solutions to this challenging mixed-integer non-linear problem. Aligned with research and industry efforts, this thesis introduces a scalable implementation to tackle this task in divide-and-conquer fashion