Raquel García-Bertrand

This paper proposes a new acceleration technique and a representative day aggregation procedure for the risk-averse two-stage stochastic generation and transmission network expansion planning problem, in which the conditional value-at-risk is used. We use a finite set of scenarios to model uncertainty in the peak demand level of loads, along with the capacity and marginal production cost of generating units. Moreover, we use representative days to model the operational variability of the electrical demand and renewable generation. The combination of scenarios and representative days involves many variables and constraints, which may lead to computationally intractable problems. Therefore, we propose a new relaxed version of the constraint generation-based algorithm that reduces the computational time of the problem. We additionally present a two-stage aggregation procedure that combines the modified maximum dissimilarity algorithm and the priority chronological time-period clustering in order to reduce the resolution of the representative days and to pay attention to extreme conditions. The numerical results of modified versions of the IEEE 24-bus Reliability Test System and the IEEE 118-bus Test System show reductions in the computational time of more than 89% for the relaxed constraint generation-based algorithm, and of more than 94% for the two-stage aggregation procedure.

Information about the paper titled "LEARNING POWER SYSTEM SIMULATION IN AN INTERNATIONAL FRAMEWORK" at IATED Digital Library

This paper addresses the two-stage transmission network expansion planning problem considering a dynamic robust approach, in which investment decisions can be made in different years of the planning horizon. In contrast to previous multi-year approaches, we focus on modeling the non-convex and inter-temporal operation of technologies such as storage facilities, whose performance promotes the incorporation of renewable generating units into power systems. An adaptive robust optimization approach is considered to model the uncertainty in, for instance, the maximum power consumption of loads. Furthermore, representative days are used to model the operational variability of electrical demand and the production of climate-based renewable generating units. We propose combining acceleration techniques with the nested column-and-constraint generation algorithm to tackle the resulting problem. Numerical results show that the use of the proposed model leads to reductions in the total worst-case cost of up to 5% in comparison with a static approach.

The increasing incorporation of renewable generating units into power systems has made it imperative to model variabilities during the resolution of generation and transmission expansion planning problems. This is commonly carried out by considering representative time periods, which are attained using clustering techniques, since the use of all historical data may involve computationally intractable problems. The chronological information concerning the input data should be maintained when representative time periods are used to model inter-temporal constraints. This is the case of long-term storage, which may be critical as regards mitigating the fluctuations related to renewable production. Recent clustering techniques have resolved this drawback at the cost of not ensuring the feasibility of the problem or underestimating extreme values of input data, which may influence investment decisions. This paper proposes a new clustering method that is based on previous techniques, maintains the chronological information regarding the input data throughout the entire time horizon and assigns more priority to the representation of extreme values. Numerical results show that the use of the proposed clustering method involves a reduction in the computational burden, since investment plans closer to the optimal solution are attained while considering less representative time periods than occurs with other methods.

The use of historical data in transmission network expansion planning problems is key to represent the short-term uncertainties in demand and stochastic renewable production conditions. Nevertheless, the use of all available historical data leads to intractable problems. For this reason, input data should be reduced while keeping important information about the system under study. Several clustering methods have been used in the technical literature for this purpose, but these generally do not represent extreme conditions such as peak demand levels, which may be critical to avoid load shedding. This letter proposes a novel approach to obtain representative time periods based on the maximum dissimilarity algorithm, which properly represents these extreme conditions. Numerical results show that the load is completely supplied using the proposed technique in all cases and that the number of required representative time periods is significantly reduced in comparison with other techniques, which translates into a reduction of the complexity of the transmission expansion planning problem.