Dynamic energy system modeling using hybrid physics-based and machine learning encoder–decoder models

Three model configurations are presented for multi-step time series predictions of the heat absorbed by water and steam in a thermal power plant. The models predict over horizons 2, 4, 6 steps into future, where each step is 5-minute increment. evaluated pure machine learning model, novel hybrid physics-based with an incomplete dataset. deconstructs individual boiler absorption units: economizer, wall, superheater, reheater. Each configuration uses gated recurrent unit (GRU) or GRU-based encoder–decoder as deep architecture. Mean squared error used to evaluate compared target values. architecture 11% more accurate than GRU only models. dataset highlights importance manipulated variables system. 10% on average 20 iterations model. Automatic differentiation applied perform local sensitivity analysis identify most impactful 72 boiler. analyses discussion about optimizing