Self-Adaptive Improvement Of Genetic Algorithm And Application In Reactive Power Optimization

Genetic algorithms(GA) is a kind of evolutionary algorithms presented by John Holland in 1975. The GA is efficient in solving some complex problems, such as combinational and nonlinear optimizations, which are difficult to be solved by traditional optimization methods. In recent years, increasing wide attention and applications are drawn in GA for its robustness, parallelity and global optimization characteristics.Summarizing former research, the simple GA (SGA) is analysised and the parameter self-adapting improved GA(adaptive GA, AGA) is presented in this paper. Compared with SGA, the AGA is improved in the reproduction and the mutation operations. The main parameters are adjusted according to the process of optimization. In the prior period, the selection pressure should be reduced to reserve the diversity of gene in solution group. Therefore the early-converging problem of SGA is overcomed, and the optimization results can be improved. In the posterior period, the selection pressure should be increased to improve the converge speed. The optimization on traditional functions shows that, the proposed AGA overcomes the early-converging problem of SGA, and gets better solution in less time.Based on the research in AGA, the reactive power optimization and planning problems are modeled and solved. The reactive power optimization is to minimize active losses by switching shunt capacitors and adjusting tap-changers of transformers, which is modeled as a nonlinear combinational optimization problem. The control variables are coded in binary code. The expert knowledge and experiences are introduced in mutation operation, which improves the solutions. To accelerate the process, the hash table is utilized to store fitness functions.The reactive power planning is to minimize energy losses by allocating shunt capacitors. The model of reactive power planning, similar to that for reactive power optimization, is also a nonlinear combinational optimization problem. The allocation and capacity of capacitors are coded in binary code. Different load levels in a long period are considered in calculation. Simulation results show that, the AGA solves the reactive power optimization and planning models effectively. The solutions reduce active power and energy losses significantly.