A PSO Based Artificial Neural Network Approach for Short Term Unit Commitment Problem

UC (Unit Commitment) is a non-linear, large scale, complex, mixed-integer combinatorial constrained optimization problem. This paper proposes, a new hybrid approach for generating UC schedules using SI (Swarm Intelligence) learning rule based NN (Neural Network). The training data has been generated using DP (Dynamic Programming) for machines without valve point effects and using genetic algorithm for machines with valve point effects. A set of load patterns as inputs and the corresponding unit generation schedules as outputs are used to train the network. The NN fine tunes the best results to the desired targets. The proposed approach has been validated for three thermal machines with valve point effects and without valve point effects. The results are compared with the approaches available in the literature. The PSO (Particle Swarm Optimization) -ANN (Artificial Neural Network) trained model gives better results which show the promise of the proposed methodology.