Optimal generation scheduling in large-scale electric power systems.

摘要

A knowledge-based dynamic programming algorithm for thermal pumped-hydro generation scheduling, and a neural network design for thermal generation scheduling in power systems are presented in this dissertation.; In the knowledge-based dynamic programming algorithm, an expert system is embedded into the dynamic programming algorithm routine to reduce its solution search space. A set of expert rules is developed, and the corresponding knowledge base is established using facts and information from a previously available optimal schedule, its original load profile, and the new load profile for the systems. The algorithm utilizes these expert rules and the facts in the knowledge base to modify the schedule to satisfy the current system requirements. The algorithm is designed to automatically select the number of steps required to approach the new load profile.; A modified variable truncation dynamic programming technique and a neighboring state selection technique are also proposed and added into the set of expert rules. In the modified variable truncation technique, a step look back and forward condition and a more efficient state-saving scheme are included, thus making the technique amenable to systems with energy storage units. The neighboring state selection technique is a heuristic method which limits the number of decisions to be considered by the dynamic programming algorithm.; A neural network consisting of two sub-networks which correspond to different types of variables in the thermal generation scheduling problem is derived and simulated for solving the thermal problem. The first sub-network is a neural net which solves the economic dispatch problem, and whose outputs indicate the power generation of on-line units. The second level is a Boltzmann machine, a stochastic neural network which determines the off/on status of units.; Computational results by a medium-size power system show that fast and high-quality solutions can be obtained using the proposed knowledge-based dynamic programming algorithm. Simulation results using the neural network prove its feasibility for solving the generation scheduling problem fast and near-optimally.