Resumen de Medium-term power planning in electricity markets with renewable generation sources
Laura Marí Tomàs
The problem addressed in this thesis is the medium-term generation planning over a yearly horizon of a generation company participating in a liberalized electricity market with pool auction of generation and consumption and with bilateral contracts between generation companies and distributions companies or big consumers. It is assumed that the generation portfolio of several generation companies includes a significant proportion of dispatchable renewables (hydro generation with storage reservoirs and pumping schemes) and non-dispatchable renewables as wind power and solar photovoltaic generation. It is also assumed than more than one generation company are able to influence market-price levels through their bidding in the auction so that the market could be oligopolistic. The results obtained are of interest to price-maker generation companies, but also to price-taker generators, and to the market operator to check whether the participants in the market behave as a cartel or seeking an equilibrium. The stochasticity of parameters in the medium-term planning is modeled in two ways. Regarding consumers load and generation unit outages, through the use of the probabilistic method of load matching: by representing the load through predicted load-duration curves of each period into which the yearly horizon is subdivided, by considering the capacity and an outage probability of each generation unit and by using the existing convolution techniques and the linear-inequality load-matching constraints. Regarding renewable energy sources, stochastic programming is used. Hydro-generation scenarios of inflows are developed for each period. As for non-dispatchable renewables (wind power and solar photo-voltaic generation), a novel model of representing them through two pseudo-units: one base unit with small outage probability and a crest unit with large outage probability is proposed, and scenarios are developed for the relevant parameters of the pseudo-units. The solar photo-voltaic generation model requires splitting each period into three subperiods with the dark hours, with the medium-light hours and with the bright hours. Quasi-Monte Carlo techniques have been employed to create a large scenario fan later reduced to a scenario tree with a reduced number of scenarios. Market prices are taken into account through an endogenous linear market-price function of load duration whose intercept depends on total hydro generation level and on wind power and solar photovoltaic level in each node of the scenario tree. With such market price function, the endogenous cartel solution and the equilibrium solutions to the medium-term planning can be obtained. To avoid having to consider the total exponential number of load-matching constraints, a load matching heuristic has been employed where small batches of new load matching constraints are generated after successive optimizations considering only the generated load matching constraints. For equilibrium solutions, the Nikaido-Isoda relaxation algorithm of successive solutions is employed using the successive optimizations of the load-matching heuristic. In mixed-market systems with auction and bilateral contracts, a time-share hypothesis is formulated and the profits function for generation companies with the generation left after honoring their bilateral contracts is formulated. The profit function obtained is non-convex, and a direct global optimization solver was tried, but proved not to be practical for the size of problem to be solved. A non-linear interior-point constrained optimization solver, also employed for problems in pure pool markets, was tried with several special techniques to circumvent the troubles caused by the non-convexity of the objective function and satisfactory results were obtained. A novel model of multi-period medium-term pumping was presented and employed. Results for several realistic test cases having different generation settings have been presented and analyzed.
