Modeling and Analysis of an Electrically Heated Molten-Salt Thermal Storage System for Thermal Power Flexibility

Electrically heated molten-salt storage converts surplus electricity into heat for later use. We combine a fixed-property heater model, a proper orthogonal decomposition–neural network (POD–NN) surrogate, and cyclic plant dispatch. The 210 cases computed with COMSOL Multiphysics 6.4.0.293 (COMSOL AB, Stockholm, Sweden) comprise 168 training, 21 validation, and 21 independent test cases. Two POD modes retain 99.998368% of the training fluctuation energy. Across all independent tests, mean case root mean square error (RMSE) is 0.716 K and pooled RMSE is 2.089 K. The worst case has an RMSE value of 9.424 K and maximum pointwise error of 20.098 K, mainly from coefficient prediction at a withheld grid corner. Five computed fields exceed 873.15 K, a numerical screening reference that is not an enforced temperature limit. These fields are retained to assess the numerical model and do not establish feasible operating conditions. Analytical comparisons verify model components; independent validation of the complete heater remains necessary. At the fixed nominal heater condition, cost minimization selects no charging, whereas emissions minimization avoids 91.930 t CO2/day at an additional cost of 2135 CNY/day. The results quantify the cost–emissions tradeoff for the stated model and daily scenario.