Causality-aware many-objective intra-day dispatch for wind-PV-thermal-storage complementary power systems

Abstract

Intraday dispatch of wind-PV-thermal-storage complementary power systems is challenged by volatile renewable generation, strong inter-temporal state coupling, narrow and evolving feasible regions, and conflicting economic, low-carbon, renewable-accommodation, and operational-security objectives. In such systems, variable renewable generation, controllable thermal units, energy storage, flexible demand, and external power exchange provide complementary regulation capabilities over the intraday horizon. Existing scheduling and constrained evolutionary optimization methods usually improve local rolling correction or feasible search, but still inadequately represent causal temporal dependencies, recursive state propagation, and the rolling evolution of feasible regions. This paper reconstructs the problem as a many-objective dynamic optimization task under causal time-sequence constraints and develops a Causal Time-Sequence Constraint-Aware Many-Objective Intra-day Dynamic Dispatch Optimization model, termed CTS-MIDDO. Structural causal priors are introduced to encode physically admissible dependencies among states, controls, disturbances, and objective responses, while causal time-sequence constraints capture non-anticipative execution, recursive feasibility, and future reachability. To solve the resulting narrow-feasible-region problem, a Temporal-Reachability and Feasible-Boundary Guided Many-Objective Evolutionary Algorithm, termed TRFB-MaOEA, is proposed by combining temporal encoding, forward state decoding, causal repair, feasible-boundary guidance, and rolling warm starts. Experiments based on real operating data, covering 15 CTS-MIDDO test instances and 10 typical-day closed-loop scenarios, show that TRFB-MaOEA achieves an average final feasible-solution ratio of 98.71% and obtains lower IGD values than each baseline under common repair support on at least 10 of the 15 instances. The 96-step closed-loop experiments further demonstrate stable dispatch performance under multiple operating stresses, confirming the value of integrating causal time-sequence structure into both model construction and many-objective evolutionary search.