A Bayesian Hidden Markov Model for Identifying Exceptional Events in Electricity Distribution

Hidden Markov Models (HMM) describe the relationship between an observed process {Yn}n>0 and an underlying and unobserved process {Xn}n≥0, that is assumed to be a Markov chain and whose realization Xk governs the distribution of the corresponding Yk (Cappé et al., 2005). The Italian regulatory mechanism for quality of service in electricity distribution applies penalties and rewards to distribution companies based on the number and duration of interruptions of service per consumer, separating continuity data into normal and exceptional categories (Fumagalli et al., 2006). Regulatory Order 172/07 issued by the Italian Regulatory Authority for Electricity and Gas (AEEG) identifies exceptional events as those 6-hours time periods when the number of faults of service is larger than a threshold determined, for each company and Italian province, by a statistical method based on the analysis of the distribution of the number of faults for that company in that province during a given three year time span (AEEG, 2007). We aim at supporting the AEEG identification method, by analyzing a large dataset provided by AEEG and consisting of the recorded hourly number of faults for different companies and different administrative districts in Italy in the three years time period 2004 2006. The statistical approach proposed here is based on the idea that the hourly number of faults depends on the status of the “global” system involved in electricity distribution; moreover, it is assumed that when the system is in an exceptional operating status a large number of interruptions protracting in time occurs. These considerations naturally leads to an HMM.