Energy system planning with risk consideration based on the chance-constrained programming model

Abstract

Greenhouse gas (GHG) emissions must be limited in order to mitigate global warming. Energy system planning is an analytical tool for a detailed implementation to meet GHG emission targets, and it involves determining the energy technologies to be deployed in the future. Since the 1990s, uncertainty factors have been considered in the planning of energy systems. In addition, various risks (e.g., policy risk and technological change) have begun to be considered in energy system planning as climate change issues have intensified. With the advent of distributed energy resources (DERs), which are small, modular, energy generation, and storage technologies, studies that consider various stakeholders have also started to increase since the 2010s. Consequently, we demonstrated the process of energy system planning with risk using chance constraints. First, we defined the uncertainty factors for energy system planning and systemically constructed a scenario tree. Second, based on the constructed data, we constructed a multi-stage chance-constrained model to analyze the trade-off between risk and cost. Finally, we expanded the energy system planning considering the risk of multi-agent modeling for DERs. In the first study, we constructed a stochastic process model of a reasonable solar PV module price using the learning curve and the AR (1) model. In addition, we proposed stopping criteria for constructing a stable scenario tree. Upon comparing the projection intervals for validation, it was confirmed that the scenario tree was well constructed by covering 70%-90%. We devised a systematic process to construct a scenario tree of uncertainty factors for use in stochastic programming-based energy system planning by summarizing the existing methodology. In the second study, we defined risk as the probability of not meeting carbon emissions target. On this basis, we construct a multi-stage chance-constrained energy system planning model. In addition, we reduced the computational time through a heuristic algorithm that removes scenarios through measures h_1 (i.e., the cost difference between scenarios) and h_2 (i.e., the cost difference between stages). Through analysis of the relationship between cost and risk, it was found that the current national plan would meet the carbon emissions target with a probability of approximately 50%. We proposed a heuristic solution approach that combines the sample average approximation and sampling and discarding methods. We also provided a new analysis of carbon emission policies. Finally, we considered a chance-constrained game for the optimal sizing of multi-microgrids. To solve the chance-constrained game, we applied a sampling approach and replace the simulation model with a convolutional neural network model to reduce the computational burden. We analyzed the planning capacity change of a microgrid according to the number of agents, reliability, and correlation among scenarios. This study provided Nash equilibria as behavioral changes of a microgrid in multiagent aspects. In conclusion, we demonstrated the process of energy system planning with risk through systematic steps. The limitation of these studies is the lack of theoretical foundations, such as rigorous proof. Nevertheless, the proposed methods are essentially modifications of existing methods, and their practicality was demonstrated by showing that they are well applied through numerical analysis. Follow-up studies using realistic data or robust optimization will further advance our research.