Debiased probabilistic reliability assessment of integrated heat-electricity systems with deep learning uncertainty

Fast growing uncertainties in integrated heat-electricity systems (IHESs) raise a desire for reliability assessment in a shorter period. This paper proposes a debiased gated recurrent unit (GRU)-based model for the short-term probabilistic reliability assessment (PRA) of IHESs. First, numerous uncertainties in IHES, such as electric load, thermal load, heat pump generation, photovoltaic panel generation, etc., are considered to train a GRU-based PRA model. Different from conventional methods, the proposed method can provide the probabilistic distribution information of energy not supplied (ENS). Second, a cost-sensitive method based on kernel density estimation is proposed to correct the assessment tendency induced by the imbalance of the continued ENS labels, which can reduce 30% of mean absolute error. More importantly, in addition to the uncertainties in IHESs, the uncertainty induced by the GRU network is further considered. The upper and lower bounds of ENS are calculated to tolerate the uncertainty of the GRU network, thereby enhancing the credibility of the proposed PRA method. The proposed method is validated in a modified Barry Island energy system, showing a similar accuracy with Monte-Carlo Simulation but saving 99.9% of time consumption.