用于神经网络直接逆控制的连续空间遗传算法

The Design of NN Direct Inverse Controller Based on Genetic Algorithms

为解决神经网络直接逆控制中训练样本的获取问题 ,提出了一种神经控制器的设计方法。通过对连续空间遗传算法的编码和遗传算子进行适当改进 ,采用保留精英的线性排序选择 ,避免成熟前收敛 ,并给出算术交叉算子和乘法变异算子 ,使算法同时具有好的搜索精度和搜索效率 ;然后采用这种改进的遗传算法对非线性动态系统的控制进行优化 ,获得了基于一定性能指标的期望的状态轨迹及相应的最优控制序列 ,并以此训练神经网络控制器。最后给出了以同步机为控制对象的仿真结果。

This paper presents a generalized design metheod of neuro-controllers to solve the knowledge acquisition of training data in the neural network direct inverse control. We improve genetic algorithms in continuous space in coding and genetic operators to overcome premature convergence to posess good precision and search effciency simultaneously by utilizing linear ranking selection, arithmetic crossover and multiplier mutation. Based on the modified genetic algorithms in continuous space, we optimize control inputs of dynamic systems, and then trains neuro-controller with the obtained desirable response trajectory and control signals that produce it as training data. The synchronous machine is employed as a test-bed to demonstrate the effectiveness of the proposed design method and the simulation results are given at the end of the paper.