粒子群优化粒子滤波算法及其在循环流化床床温辨识中的应用

Particle Swarm Optimized Particle Filter and Its Application in Identification of CFB Bed Temperature

在基于现场数据的神经网络训练中,粒子滤波算法对非线性、非高斯噪声系统的适应性及全局寻优能力都优于传统BP算法。针对粒子滤波算法的粒子贫乏以及初始状态未知时需要大量粒子才能进行鲁棒状态预估等问题,将粒子群优化思想引入粒子滤波过程,使粒子在权重更新前更加趋向于高似然区域;同时,优化过程使得远离真实状态的粒子趋向于真实状态出现概率较大的区域,提高了每个粒子的作用效果。将优化前后的粒子滤波算法训练2-1-1结构的神经网络并进行比较,结果表明优化算法提高了神经网络模型预测精度,降低了精确预估所需粒子数。最后,应用该算法对基于现场数据的循环流化床床温神经网络模型进行训练并预测,验证了算法的有效性。

In the training process of the neural network based on the field datasets,both the adaptability to nonlinear and non-gaussian noise system and global optimization ability of particle filter（ PF） are better than the traditional BP algorithm. To the problem of particle impoverishment and needing a large sample size for robust state estimation when initial state is unknown,particle swarm optimization is introduced into generic particle filter. Through particle swarm optimized particle filter（ PSO-PF）,particles are moved towards regions where they have larger values of likelihood function before weights update,at the same time,the optimization process drives particles far away from the true state to move to the area with larger true state probablility. The effect of each particle is improved. Using PF and PSO-PF to train 2- 1- 1 neural network respectively,the results suggest that the prediction accuracy is improved and the sample size necessary for accurate state estimation is reduced. Finally,the neural network based on the filed datasets of circulating fluidized bed. temperature is trained by PSO-PF,the bed temperature is predicted,the validity and usefulness of this algrithm are verified.