遗传算法和神经网络结合的PSD非线性校正

针对光电位置传感器(PSD)检测系统在大坝变形观测中所呈现的非线性问题,建立改进的遗传算法和LM-BP神经网络结合的模型,对PSD的非线性进行补偿?该方法先用遗传算法对LM-BP网络的权阈值进行优化后再用LM-BP网络逼近任意非线性函数的特点对实际位置数据与理想值进行拟合后并进行测试,经过多次任意产生的种群优化后选择较为优秀个体作为神经网络的和阈值,并对任意位置进行校正,仿真结果表明,该方法克服了LM-BP网络对初始权阈值的依赖和泛化能力弱的特点,多次实验平均误差都小于1%,其泛化能力优于标准的遗传算法和神经网络结合的模型。

Nonlinear correction of PSD with genetic algorithm based on neural network

The position sensitive detector (PSD) detection system in the observation of dam deformation can present nonlinear problem, so an improved genetic algorithm based on LM BP neural network model is established to compensate the nonlinearity of PSD. Firstly, this method optimized the weight and threshold of LM BP network by genetic algorithm, and let the actual position data fitted with the ideal value with the characteristics of LM BP network to approximate any nonlinear function. The more excellent individual was chosen as the neural network weight value and threshold value after a population of optimizations repeatly and randomly. Finally, arbitrary position was corrected. The simulation results show that this method overcomes the dependent of LM BP network on initial weight threshold and no strongly generalization ability. The average error is less than 1% through several experiment observations. Also, the generalization ability is better than the standard genetic algorithm based on neural network.