(ω1, ω2,..., ωm) representing the percentage of generation asset allocated among m supply contracts. The random vector y is the expected profit in each of these contracts. Let the density function of y be p(.). Given a decision ω , the probability of ƒ(ω,y) not exceeding a threshold α is represented as
Given a confidence level β and a fixed ω, the portfolio VaR is defined as
The is defined as the expected value of loss that exceeds
Rockafellar and Uryasev in [16], [17] defined a function Fβ(ω,α) to solve the above CvaR
where [t]+ = max{t,0}.
Then, we have
The discrete version to approximate Fβ(ω,α) with totally N samples could be chosen as
By introducing an auxiliary variable Zk for k=1, 2, ..., N, the above Fβ (ω,α) could be written as
The optimization problem is formed as maximizing profit with risk as the constraint. Recall the way that CVaR is defined, i.e.
, this optimization problem is a bi-level
optimization problem as
where V is the risk tolerance level specified by the GenCo. This in our case could be further written as
Where (α*, z*) is the solution of
As proved in [18], if the above constraint
is active and the corresponding
multiplier of the constraint is not equal to zero, the above bi-level optimization model is equivalent to a single layer linear optimization model as below:
đang được dịch, vui lòng đợi..
