航天发射场制冷系统优化建模研究

Study on the Optimization Modeling of Refrigeration System in Spacecraft Launching Site

针对现有航天发射场空调制冷系统的控制精度较低,不利于给航天器提供良好的发射环境,建立基于神经网络的预测模型,采用优化建模方法,能很好地提高控制精度,主要问题在于神经网络结构和参数寻优算法的确定。首先分析并确定了制冷系统的神经网络结构及相关参数,然后针对传统神经网络算法(梯度法)寻优速度慢和"过拟合"的缺点,提出了一种新型的神经网络训练算法——MPGA-BPNN,利用多种群遗传算法MPGA搜索网络的阈值和权值等参数,具有全局优化能力且寻优速度快的优点。所建模型具有很好的预测精度,说明了目标优化控制方法对于发射场制冷系统具有很好的适用性。

At present, the refrigeration system in spacecraft launching site is of low control precision, which is harmful to keeping good launching environment for spacecraft. For building neural network predictive model and adopting optimization model contribute to the improvement of control precision, its main difficulty is the determination of network configuration and optimization algorithm. Firstly the network structure and related parameters of neural network of refrigeration system were well analyzed and determined, then with an eye to the slow convergence speed and over - fitting of traditional network training algorithm, a novel training method MPGA - BPNN was proposed, which makes use of multiple population genetic algorithm （MPGA） to search the optimum weight value and threshold value of BPNN, and has the merits of global optimum ability and fast convergence speed. The network is of high prediction precision ,which proves that the optimization modeling method is suitable for refrigeration system in spacecraft launching site.