The contribution of distribution systems to overall customer reliability is quite significant. There is a growing concern in distribution-system-reliability assessment due to the improvement of customer requirement. Based on analyzing the characteristic of the urban distribution feeders, this dissertation proposes a new network model and three failure analysis techniques for distribution system reliability evaluation. Probabilistic distribution features of various system indices are discussed through zone simulation methods. The effect of time-varying component failure rates on system reliability is also analyzed. To identify the weak points in the system, a Bayesian network model is built for distribution system reliability re-assessment. First the dissertation defines a zone-network diagram for distribution system reliability evaluation based on feeder partition method. Load zone based technique, failure zone based technique and zone-network based technique are proposed for failure effect analysis. Taking the advantage of zone-network diagram, propositions can be deduced for each technique and the average value of reliability indices can be quickly obtained by analytical methods. Considering the uncertain load, interval models are used for system reliability evaluation taking the partial loss of continuity (PLOC) events into account, which is possible to perform practical engineering analysis.Then zone-based time sequential simulation methods and zone-based non-sequential simulation methods are presented. Using these approaches we discuss the effects of network configuration, convergence criteria and sampling methods on convergence speed. Probabilistic distributions of system indices are also studied. Results from different component models or zone-node models are compared. Considering the time-varying nature and uncertainty of system parameters, we adopt time-dependent model to describe the effects of weather conditions, aging and maintenance on component failure rates. Time-varying restoration time and load are modeled using bounded random variables. The stochastic thinning technique of non-homogeneous Poisson process is applied to sample the component state durations, which improves the simulation efficiency and accuracy.A dual isomorphic Bayesian network model for medium voltage distribution system reliability re-assessment is also proposed. The zone-network diagram and the zone-based failure analysis method are used to construct standard building blocks in the model. This technique provides a zone-based analysis, which reduces the computation burden in conventional component-by-component method and simplifies the structure of Bayesian network model. Posterior outage probability indices as well as outage frequency indices are obtained in two layers. Weak points are identified during backward propagation for what-if studies. The model provides useful tools for distribution system reliability re-assessment. |