Coordinated optimization of reserve fund scale and emergency resource allocation for power utilities under multi-type...

Large-scale blackouts increasingly arise from heterogeneous causes such as natural disasters, equipment failures and cyber attacks, which force power utilities to decide, before any event is known, both how large a financial reserve to hold and how to convert it into emergency response capability. This paper develops a coordinated optimization of reserve fund scale and emergency resource allocation under multi-type large-scale blackout risk. A unified set of multi-type blackout scenarios is constructed by Monte Carlo sampling and condensed by a tail-aware scenario reduction into a tractable representative set. The restoration technology distinguishes a repair channel, in which crews and materials enter as complementary inputs, from a re-energization channel supplied by mobile equipment, and both channels carry cause-dependent productivity so that the response mechanism differs across disasters, failures and cyber attacks. A two-stage risk-averse stochastic program co-optimizes the first-stage reserve fund and resource pre-positioning together with the second-stage post-event response, embeds a conditional-value-at-risk measure to limit extreme losses, and is solved by a sample-average-approximation scheme with an L-shaped decomposition. Case studies on a stylized 118-node planning instance show that the coordinated risk-averse plan attains an out-of-sample risk-adjusted cost of 352.1 $M against 558.2 $M for a decoupled scheme and 604.7 $M for a risk-neutral scheme, while holding CVaR0.95 to 154.3 $M and expected unserved energy to 1478 MWh. The tail-aware reduction keeps the CVaR0.95 error at 1.08% of its raw-sample value with 200 retained scenarios, against 11.91% for random subsampling. The cause mix governs the resource portfolio, and the value of lost load is the dominant economic driver of the appropriate reserve scale, with a standardized regression coefficient of 0.66 under joint parameter uncertainty. The study is presented as a methodological proof of concept: the instance is synthetic and network operability is not modelled, so the results characterize the structure of the preparedness trade-off rather than a validated prescription for a specific utility.