Assessment of electrochemical CO$_2$-to-CO reduction technology using real options based on reinforcement learning under multiple time-varying uncertainties

Abstract

As an effort to reduce anthropogenic greenhouse gas (GHG) emissions which is the main cause of global warming has been emphasized, importance of carbon dioxide utilization (CDU) technology becomes much larger. Among many CDU technologies, electrochemical CO$_2$ reduction (ECO$_2$R) is promising because it can use electricity made from renewable energy sources, which have less CO$_2$ emissions. Since ECO$_2$R can accelerate the reduction of CO$_2$ emissions, commercialization of ECO$_2$R is necessary even though the process has low technology readiness level (TRL). Conventionally, net present value (NPV) method, which uses NPV computed by discounted cash flow analysis has been used to evaluate the commercialized project. However, this method cannot respond to dynamic uncertainties such as government policy and energy price, because it makes all decisions at the current timestep. To deal with dynamic uncertainties, real options such as expansion and delay have been introduced to make decisions. Currently, lattice tree-based model has been used, but the model has to set the levels of nodes and probabilities arbitrarily and computation becomes larger with the big problem. So far, there have been no cases that evaluate the project whose technology is at an early-stage, but dealing with it as commercialized in multi-period. In this work, optimized capacity of electrochemical CO$_2$-to-CO reduction technology is determined by real options based on reinforcement learning (RL) and the value of the technology is computed.