ABSTRACT



    This thesis addresses the resolution of two problems relating directly to fault diagnosis in electric power transmission networks through artificial intelligence techniques, more specifically those implemented by means of artificial neural networks (ANN’s). 
     
    The first problem derives from the complexities of diagnosing simple and multiple faults, due to the enormous number of alarms generated by the SCADA system during an event, many of which are not directly related to the faulty component, and to the topological growth of the electric power system itself.  

     The second problem, in some cases, arises from the loss of relevant information from the SCADA system (state of switches and/or primary relays) resulting in the emission of an unreliable diagnosis.

     With respect to the first problem, a design methodology is proposed for fault diagnosis using generic neural structures, one for each type of component comprising the network (transmission line, transformer, bus bar), taking into account only the alarms from switch and primary relay states, and/or back-up states of each component. The method proposed provides a diagnosis for both simple and multiple faults, regardless of the number of alarms generated or the size of the network. Moreover, it does not require a network configurator and can be implemented by control center operators.

      For the second problem, the proposal comprises a design methodology with the use of artificial neuron structures taking into consideration analogical signals and frequency spectrums from typical current and voltage faults in a transmission line, which are obtained from simulations. The method proposed provides a diagnosis of the transmission line, combined with the diagnosis emitted previously (logic state of switches and relays), thereby ensuring the reliability of the final diagnosis.
    The diagnostic methodology proposed herein is currently being applied in the electric power transmission network throughout the urban area of the city of Merida, Yucatan, Mexico, with satisfactory results.  

     In addition, a planning system has been developed through the implementation of a neural structure comprising several neural modules, which takes into consideration the cost function of each generator as well as the overload restriction in other components of the system, as a consequence of the liberation of a faulty component. The proposed planning system provides optimal re-dispatch of the generators that are on line, in order to avoid overloading the remaining components. The planning system has been tested on the IEEE 30-Bus experimental electric power system.