Since the early to mid 1980s much of the effort in power systems analysis has turned away from the methodology of formal mathematical modelling which came from the fields of operations research, control theory and numerical analysis to the less rigorous techniques of artificial intelligence (AI). Today the main AI techniques found in power systems applications are those utilising the logic and knowledge representations of expert systems, fuzzy systems, artificial neural networks (ANN) and, more recently, evolutionary computing. These techniques will be outlined in this chapter and the power system applications indicated.

From the preliminary results, the SIMIAN architecture appears to be a workable and efficient system for power system modelling. Future work will investigate the effects of a scaling in the size of the network. The IEEE 30-Bus network is unrealistically small compared to typical distribution planning studies of 5000 nodes. The dynamic behaviour capability of the SIMIAN system was found to allow easy extension, whilst maintaining the object-oriented nature of the architecture. Integration of a GUI (graphical user interface) for browsing model parameters and topological representation of a network were both implemented using the dynamic function, and state machine functionality. The use of a common mechanism not only simplified the design but also enabled the resultant GUI classes to be constructed in a highly generic manner and as completely independent application objects. In porting the 'applications' to disparate hardware platforms, such as PCs, this latter ability requires that only this one portion need be ported - the object server application may remain on a workstation. In addition, the automatic updating of information between applications and the OODBMS was completed seamlessly through the event handling system. From this experience, it appears that more complex applications, for instance SCADA could be 'bolted-on' in a comparable manner.

In this chapter a real-time alarm handling and fault analysis (AHFA) expert system is described , this being seen as more than simply an alarm processor. AHFA is able to perform its diagnoses directly in real-time without recourse either to situation specific heuristics or to model-based 'generate and test'. The key is a novel form of abstraction, also described in this chapter. Also described here is an adaptive form of AHFA based on a learning classifier system (LCS), and this uses a fault simulator to direct the entire fault diagnosis. Unlike a conventional expert system, rules are not programmed into the LCS, although that is quite possible. On the contrary, the LCS is trained by example. The diagnostics are then adapted through genetic algorithms (GAs), which directly operate on the rule strings.This chapter is intended to give an overview of the problem area of alarm analysis for power system transmission, along with a couple of approaches to help tackle the problems using computer-based technology.

The reliable operation of large power systems with small stability margins is highly dependent on control systems and protection devices. Progress in the field of microprocessor systems and demanding requirements in respect of the performance of protective relays are the reasons for digital device applications to power system protection. The superiority of numeric protection over its analogue alternatives is attributed to such factors as accurate extraction of the fundamental voltage and current components through filtering, functional benefits resulting from multi-processor design and extensive self-monitoring, etc. However, all these reasons have not led to a major impact on speed, sensitivity and selectivity of primary protective relays, and the gains are only marginal; this is so because conventional digital relays still rely on deterministic signal models and a heuristic approach for decision making, so that only a fraction of the information contained within voltage and current signals as well as knowledge about the plant to be protected is used.

In this chapter the structured development of KBS for decision support roles in power system operation, control and monitoring will be considered. Three techniques, 'classical' KBS, CBR and MBR will be considered as potential complementary routes towards the implementation of decision support system functionality. Case studies of the application of these techniques in practical power systems applications will be described. While the use of these techniques has proven successful in isolation, the potential exists to combine them together with appropriate analytical support within an integrated decision support system. The chapter concludes with a discussion of a potential implementational framework for such a system, and the scope it provides for more flexible and user responsive decision support.

In this chapter we outline a generic neuro-expert system (GENUES) architecture for hybrid reasoning. The architecture consists of five phases, namely, decomposition phase, control phase, decision phase, preprocessing phase, and postprocessing phase. The architecture is particularly applicable in time critical diagnostic/classification domains and data intensive domains in general. We describe the application of GENUES in a real time alarm processing system in a power system control centre.

Reyrolle Protection have carried out research in conjunction with Bath University into applying adaptive techniques to autoreclose schemes and have produced an algorithm based on an artificial neural network which can recognise when it is 'safe to reclose' and when it is 'unsafe to reclose'. This algorithm is based on examination of the induced voltage on the faulted phase and by applying pattern recognition techniques determines when the secondary arc extinguishes. Significant operational advantages can now be realised using this technology resulting in changes to existing operational philosophy. Conventional autoreclose relays applied to the system have followed the philosophy of 'reclose to restore the system', but a progression from this philosophy to 'reclose only if safe to do so' can now be made using this adaptive approach. With this adaptive technique the main requirement remains to protect the investment i.e. the system, by reducing damaging shocks and voltage dips and maintaining continuity of supply.