Industrial Load Modeling and Optimization for Market-Based Interaction with Power Systems

Industrial loads account for more than 60% of electricity consumption in China and offer substantial flexibility for balancing variable power systems. Their market participation remains limited by complex production constraints, incomplete information, and the computational burden of coordinating large portfolios. This dissertation develops modeling and optimization methods for market-based interaction between industrial loads and power systems. First, unified formulations based on the Linearized State Task Network and continuous Resource Task Network represent discrete and continuous industrial processes for power-system optimization. In a representative steelmaking case, they reduce solution time from more than 24 hours to less than 30 minutes while preserving modeling accuracy. Second, a privacy-preserving identification method combines process knowledge with hourly smart-meter data to infer internal production parameters. Using 21 days of observations, it achieves errors of 5.2%-8.5% for cement and steel-powder production, more than halving the errors of conventional machine-learning baselines. Third, a data-driven method converts high-dimensional, nonconvex flexibility regions into compact linear representations. For a steelmaking process with more than 10,000 binary variables, the resulting models require only 24-48 continuous variables and incur errors of 3.6%-10.3%. Finally, a co-optimization framework combines dimension-reduced bidding with exact disaggregation, allocating power among tens of thousands of resources within milliseconds while maintaining device-level feasibility. In a representative comparison, it reduces interaction costs by 40% relative to a simplified strategy. Together, these methods provide a tractable pipeline from industrial process modeling and parameter identification to flexibility aggregation and power system interation.