波动方程反演中基于加权的混合目标函数研究

基于 L_2范数的目标函数要求原始数据噪声遵从 Gauss 分布,因此对大的数据异常点过分敏感。为了解决这一问题,利用反正切函数作为加权因子构造了反正切混合目标函数,并利用模型试验进行了验证。简要阐述了最优化目标函数的构造方法；给出了利用反正切函数构造混合目标函数的基本原理,并对目标函数的微分特征和概率特征进行了分析；利用模型论述了反正切混合目标函数中控制参数μ和σ的意义,μ控制反正切函数变化的速度,σ控制数据误差项被处理或加权的方式。设计了一维层状模型,对比了反正切混合目标函数和其它混合目标函数的反演效果。反正切混合目标函数的模型试验表明,对于小数据误差项用近似 L_2范数的方式处理,对于较大的数据误差项用近似 L_1范数的方式处理,可以抑制虚假异常点对反演的影响。对反正切混合目标函数的梯度特征分析表明,反正切加权因子加大了有效模型参数误差梯度分量的权重,从而加速了迭代收敛过程。分析还表明反正切目标函数能够适应 Gauss 和 Cauchy 等不同概率密度分布的数据误差,因此,提高了目标函数的适应能力,扩大了波动方程反演的处理范围。

Elevator Group-Control Policy Based on Neural Network Optimized by Genetic Algorithm

Aiming at the diversity and nonlinearity of the elevator system control target, an effective group method based on a hybrid algorithm of genetic algorithm and neural network is presented in this paper. The genetic algorithm is used to search the weight of the neural network. At the same time, the multi-objective-based evaluation function is adopted, in which there are three main indicators including the passenger waiting time, car passengers number and the number of stops. Different weights are given to meet the actual needs. The optimal values of the evaluation function are obtained, and the optimal dispatch control of the elevator group control system based on neural network is realized. By analyzing the running of the elevator group control system, all the processes and steps are presented. The validity of the hybrid algorithm is verified by the dynamic imitation performance.