Hybrid intelligent technique for voltage/VAR control in power systems

Modern power systems are far from immune from voltage collapse, and examples abound over the past decade. This is due to the fact the current generation of automatic emergency and operational Volt/VAr control systems can be ineffective and unreliable for some cases. The artificial intelligence (AI) techniques allow developing new automatic intelligent Volt/VAr control systems, which will mitigate negative consequences caused by the human factor and the shortcomings of traditional control systems. This study presents a hybrid Volt/VAr control system using an AI-based technology. The system consists of two components: a centralised deep learning control system (DeepCS) for preventive security control and a decentralised multi-agent control system (MACS) for emergency control. The DeepCS includes two deep neural networks – a deep multilayer perceptron to recognise reactive power injections and a long short-term memory network to predict system state after control actions. The MACS is a decentralised automatic Volt/VAr control that involves determining the time of critical overload and switching to the load shedding procedure. The proposed approach is highly efficient for various scenarios of two (IEEE 6, IEEE 118) test systems and one real 33-bus Bodaibo subsystem.