A Neural Network Approach For Evaluating Levelized Cost Of Electricity For Generating Electricity Across Various Generation Technologies

This paper proposes a novel method to determine the LCOE of energy sources based on uncertain parameters. Generally, to determine the global sensitivity analysis of energy technology, Monte Carlo simulation (MCS) is used. The drawback of MCS is its computational expense to generate a large number of random variables. So an alternate method of using the NARX network is proposed. Instead of generating many random variables, the NARX network is trained, and the network directly estimates the Levelized cost of Electricity (LCOE). Thus the repetitive approach of calculating the LCOE based on the uncertain variables is no longer needed. Thereby dramatically reducing the computational burden of the process. For the study, coal-based thermal energy technology and solar PV are considered. This particular technology is chosen due to the large number of uncertain variables associated with LCOE. The global sensitivity analysis results for both MCS and NARX are compared. The results showed that NARX requires less time than MCS and generates the same result as MCS with low error. Also, the computation time has is reduced with almost replicating results of MCS