Optimal Short-Term Thermal Unit Commitment Using Neural Network

Abstract

Unit commitment and economic dispatch, when combined together is a useful tool to find the most economic generation schedule with which demand and all generating unit constraints are satisfied. This developed unit commitment and economic dispatch program provide dispatchers a robust tool for planning both operation and market strategies with consideration of cost minimization, risk management. Fuel cost savings can be obtained by proper commitment of the available generating units. This thesis describes a Dynamic programming method for the commitment of thermal units over a period of up to 24 hours. Back propagation neural network has been applied for solving the unit commitment schedule of thermal generating units of 3 thermal power plants with very promising results. The total cost includes both the fuel cost and cost associated with the start up and shut down of units. A variety of spinning reserve requirements is observed and equality constraints of power balance, inequality plant generation capacity constraints. The inputs to the neural network contain the total load supplied. The electric power generation of 3 thermal power plants is taken as the output of the neural network. A MATLAB code has been developed to generate training and test pattern for the developed network. To conclude, the performance and time taken for execution of the neural network is compared with conventional dynamic programming method.