基于改进粒子群优化的神经网络及应用

为了克服粒子群算法的早熟收敛,提出了一种改进的粒子群算法用于神经网络训练。该算法对种群进行均匀初始化,用多个粒子的信息引导个体的更新,以保证全局搜索的有效性,同时引入随机算子对陷入局部最优的粒子进行变异,提高了算法的寻优性能。将改进粒子群算法训练的神经网络应用于IRIS模式分类问题和短期电力负荷预测,与BP算法、遗传算法及粒子群算法比较,该算法在提高误差精度的同时可加快训练收敛的速度。

Neural network based on modified particle swarm optimization and its application

A modified particle swarm optimization algorithm applied to train the artificial neural network is proposed in order to overcome the problem of premature convergence observed in many applications of PSO.In the new algorithm,each particle is initialized equally and attracted towards the better previous positions visited by many particles to improve global convergence.At the same time, the algorithm uses random operator to change positions of the particles which plunged in the local optimization.The algorithm is successfully applied to IRIS pattern classification problems and short term power load forecasting.Compared with error back propagation algorithm,genetic algorithm and basic particle swarm optimization,the algorithm can improve the computing accuracy while speeding up the convergence process.