A Hybrid Neural Network and Simulated Annealing Approach to the Unit Commitment Problem

Machine Learning Research Centre, School of Computer Science, QUT, Brisbane, Qld 4001, Australia
Source: OAI


In this paper, the authors present an approach combining the feedforward neural network and the simulated annealing method to solve unit commitment, a mixed integer combinatorial optimisation problem in power system. The artificial neural network (ANN) is used to determine the discrete variables corresponding to the state of each unit at each time interval. The simulated annealing method is used to generate the continuous variables corresponding to the power output of each unit and the production cost. The type of neural network used in this method is a multi-layer perceptron trained by the back-propagation algorithm. A set of load profiles as inputs and the corresponding unit-commitment schedules as outputs (satisfying the minimum up-down, spinning reserve and crew constraints) are utilized to train the network. A method to generate the training patterns is also presented. The experimental result demonstrates that the proposed approach can solve unit commitment in a reduced computational time with an optimum generation schedule.

Download full-text


Available from: Jaydev Sharma
    • "Later hybridization between two heuristic algorithms has been used to solve the GS problem [21] [22] [23]. Recently, Particle Swarm Optimization approach inspired by quantum computing has also been introduced in [24]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a Modified Non-dominated Sorting Genetic Algorithm-II (MNSGA-II) solution to Multi-objective Generation Scheduling (MOGS) problem. The MOGS problem involves the decisions with regards to the unit start-up, shut down times and the assignment of the load demands to the committed generating units, considering conflicting objectives such as minimization of system operational cost and minimization of emission release. Through an intelligent encoding scheme, hard constraints such as minimum up/down time constraints are automatically satisfied. For maintaining good diversity in the performance of NSGA-II, the concepts of Dynamic Crowding Distance (DCD) is implemented in NSGA-II algorithm and given the name as MNSGA-II. In order to prove the capability of the proposed approach 10 units, 24-hour test system is considered. The performance of the MNSGA-II are compared with NSGA-II and validated with reference Pareto front generated by conventional weighted sum method using Real Coded Genetic Algorithm (RGA). Numerical results demonstrate the ability of the proposed approach, to generate well distributed pareto front solutions for MOGS problem.
    No preview · Article · Jan 2015
    • "These hybrid methods are claimed to accommodate the constraints that are more complicated and claimed to have better quality solutions even though the system under consideration is very large181920. Later hybridization between two heuristic algorithms has been used to solve the GS problem212223. Recently, Particle Swarm Optimization approach inspired by quantum computing has also been introduced in [24]. "

    No preview · Conference Paper · Dec 2014
  • Source
    • "However, a great deal of operator interaction is required in this approach, making it inconvenient and time-consuming. Neural networks (most often multilayer perceptrons) are trained to recognize the most economical UC schedule associated with the pattern of the current load curve (Nayak & Sharma, 2000; Wong, Chung, & Wong, 2000). The training set contains typical load curves and corresponding UC schedules. "
    [Show abstract] [Hide abstract]
    ABSTRACT: An approach for solving the unit commitment problem based on genetic algorithm with binary representation of the unit start-up and shut-down times is presented. The proposed definition of the decision variables and their binary representation reduce the solution space and computational time in comparison to the classical genetic algorithm approach to unit commitment. The method incorporates time-dependent start-up costs, demand and reserve constraints, minimum up and down time constraints and units power generation limits. Penalty functions are applied to the infeasible solutions. Test results showed an improvement in effectiveness and computational time compared to results obtained from genetic algorithm with standard binary representation of the unit states and other methods.
    Full-text · Article · Nov 2013 · Expert Systems with Applications
Show more