Generation Expansion Planning with high shares of Renewable Energy Sources: Single and Multi objective optimization based on Metamodel-assisted Evolutionary Algorithms

Abstract

The scope of this thesis is to develop and examine single and multi objective Metamodel-assisted Evolutionary Algorithms (MAEA) for Generation Expansion Planning (GEP) models in the presence of high shares of generation by Renewable Energy Sources (RES). A GEP model may facilitate decision making in mid towards long-term energy scheduling. Commonly, a GEP model is employed to provide a road-map towards an affordable, sustainable and secure operation of a power system. These road-maps are generated in the form of scenarios for the evolution of a power system and are determined by considering both possible investments in capacity additions and the short-term operation of the power system, e.g. to determine an economic and/or environmentally optimal investment plan that is adequate to meet the growing power demand and exhibits a reliable short-term operation. Such a problem can be formulated as an optimization problem. The pursue for emission-free power system and the increasing shares of RES, triggered by environmental considerations, have led to the introduction of various aspects to GEP models. Such an aspect is to capture economic and technical challenges related to the short-term operation of a power system with increased detail. This could be required to assess the synergy of the conventional generating fleet with the increasing installations of RES. In particular, the variability and uncertainty associated with the latter has been reported to increase the operational flexibility requirements of future generating fleets. It has also been reported that underestimating these requirements could have economic implications on the reliable and efficient short-term operation. Towards this aim, endeavours have been made to integrate, within a GEP model, a more detailed representation of the short-term operation of a power system in terms of spatial, temporal and technical detail. However, the integration of GEP model with such detail can lead to an increased computational cost. Therefore, simplifications are required. Evolutionary Algorithms (EA) are nature-inspired algorithms which employ stochastic operators to improve a set of candidate solutions. As derivative-free algorithms, EAs can be used as direct search methods; this feature has rendered them applicable for complex optimization problems. In addition, Multi-Objective EAs (MOEA) are well established approaches for Multi-Objective Optimization (MOO). On the other hand, one main limitation is the relatively large number of function evaluations required for the algorithm to converge. This can be binding for optimization problems involving computational costly simulations. For such applications, EAs coupled with Approximating Models (AM) have been developed which are commonly referred to as MAEAs or Surrogate-Assisted EAs. The AMs replace in part the original models and provide an estimate for the adequacy of a candidate solution to reduce the computational burden. This thesis focuses on MAEA applications for single and multi objective GEP optimization problems that include Simulation Models (SM) for the short-term operation of a power system. The most important contributions of this thesis are the following: A single objective multi-period GEP approach based on MAEAs is presented. The GEP model includes a Simulation Model (SM) to provide an indicator of the cost of the short-term operation. The adopted SM is an optimization model for the short-term operation of a power system including simplifications e.g. spatial detail is not examined. However, it exhibits an increased level of technical and temporal detail w.r.t. the context of long-term planning, and it is adopted to assess on-line the operating flexibility of a candidate installed capacity. The formulation exploits problem-specific characteristics. This is implemented by employing AMs to provide an estimate of the SM's output and reduce the number of simulations required to achieve a near-optimal solution. The AMs are Radial Basis Functions (RBF). These are built off-line and updated on-line to improve the accuracy of the achieved approximation. Both local and global AMs are built in different stages of the search. Problem specialized operators are developed to enhance the performance of the EA examined which is Differential Evolution (DE). The performance of the MAEA and the problem-specialized operators are assessed. The MAEA achieved satisfactory results based on the performed numerical experiments. Moreover, among the developed problem-specialized operators, a repair heuristic, addressing the constraint nature of the optimization problem, provided the largest improvement in the performance of the base DE algorithm. The impact of including the SM is also examined. The results indicate the importance of capturing operational flexibility requirements to adequately assess the flexibility providers considered as investment options. The metrics employed to examine the accuracy of the attained AMs indicated that a decent approximation had been achieved. Therefore, a visual analysis of the sensitivity of the operating cost towards the installed capacity of the derived near-optimal solution was carried out. A multi-objective static GEP approach based on MAEAs is presented that aims at capturing cost trade-offs emerging for a MOO GEP. Operational flexibility is assessed by an adopted SM that includes technical, spatial and temporal detail. The approach is developed based on MOEA and frameworks for surrogate-assisted derivative-free optimization. Approximation models are employed to address the computational restrictions. RBF and Polynomial Regression (PR) are used as the AMs. These are updated on-line by criteria that prioritize feasibility of the planning constraints, the spatial allocation of the attained training set w.r.t. the search space, and a possible Hypervolume improvement. A local phase is also included in which gradient-based local search is implemented employing local RBF, PR and an ensemble model. The performance of the approach is examined on a MOO benchmark test suite. Numerical experiments are carried out to assess the performance optimization approach on a MOO GEP formulation neglecting the short-term operation and on five MOO GEP formulations including a SM. The latter are repeated for two different levels of temporal detail. The results attained suggest an acceptable performance of the optimization approach w.r.t. the computational restriction. Moreover, the achieved accuracy of the AMs varied among the numerical experiments. The main factors influencing the performance of the AMs are identified. An analysis of the derived cost trade-offs for each of the five formulations examined can provide a detailed evaluation of the impact of a diverse set of alternatives. This could reveal incentives required for strategic energy policy decision making. For example, based on the extreme values of the non-dominated front attained for the considered operating and investment cost functions, a 96% reduction of the investment cost could result in a nearly 40% increase of operating cost. Decision support tools could facilitate the complex and evolving decision making process of GEP. Economic, environmental and social criteria must be considered along with aspects that are progressively identified as essential. Towards this aim, the developed EA-based approaches have been presented. Despite their heuristic nature, the results suggested that these could be promising tools to support well established state-of-the-art GEP models that could facilitate decision makers, such as investors and energy policy makers, when high shares of RES generation are considered.