基于BP神经网络的霍普菲尔德模型改进研究

为了分析对流层延迟的时空变化规律、提高对流层延迟的改正精度,利用BP神经网络处理非线性问题的优势,改进传统的霍普菲尔德模型得到一种新的融合模型(Hop+BP模型)。分别对比Hop+BP模型与传统的霍普菲尔德模型、多元线性回归模型、BP神经网络等模型的计算结果,得到如下结论:霍普菲尔德模型存在一个明显的系统误差,精度较低;多元线性回归的预测精度有所提高,但是其本质是将数据强制拟合,缺少物理解释,难以推广使用;传统的BP神经网络的计算精度较之霍普菲尔德模型有80%的提高,但存在明显的不稳定性;Hop+BP模型具有预测精度高、稳定性好等优点,预测中误差为1.1cm,明显优于传统方法。     

径向基神经网络在沉降预测中的应用

针对大型建筑物的沉降监测存在较多的不等时间间隔沉降监测数据,现有模型需要对此类数据进行等间隔处理后建立沉降预测模型的现状,该文采用无需对监测数据进行等间隔预处理的径向基神经网络对沉降监测数据建立沉降预测。通过对西安某大厦基坑开挖对地表和周围建筑物影响的沉降监测数据进行实例分析,并与非等间隔灰色GM(1,1)预测模型进行对比,利用模型评价指标评价预测模型精度。结果表明:采用径向基神经网络建立预测模型处理过程简便,其预测精度优于非等间隔灰色GM(1,1)预测模型。     

离散Hopfield神经网络的稳定性

研究离散Ｈｏｐｆｉｅｌｄ神经网络的串行稳定性和并行稳定性，分别提出了条件较弱的串行稳定性和并行稳定性的新判据．     

Parallel Neural Network-Based Motion Controller for Autonomous Underwater Vehicles

A parallel neural network-based controller （PNNC） is presented for the motion control of underwater vehicles in this paper. It consists of a real-time part, a self-learning part and a desired-state programmer, and it is different from normal adaptive neural network controller in structure. Owing to the introduction of the self-learning part, on-line learning can be performed without sample data in several sample periods, resulting in high learning speed of the controller and good control performance. The desired-state programmer is utilized to obtain better learning samples of the neural network to keep the stability of the controller. The developed controller is applied to the 4-degree of freedom control of the AUV “IUV- IV” and is successful on the simulation platform. The control performance is also compared with that of neural network controller with different structures such as normal adaptive neural network and different learning methods. Current effects and surge velocity control are also included to demonstrate the controller＇ s performance. It is shown that the PNNC has a great possibility to solve the problems in the control system design of underwater vehicles.     

Dynamic analysis of vessel/riser/equipment system for deep-sea mining with RBF neural network approximations

Abstract

Deep-sea mining (DSM) is an advanced technology. This article is focused on the dynamic analysis of a coupled vessel/riser/equipment system of a DSM based on radial basis function (RBF) neural network approximations while considering vessel dynamic positioning (DP) and active heave compensation (AHC). A coupled model including the production support vessel (PSV), lifting riser, and slurry pump is established containing simulated DP and AHC models. Furthermore, dynamic simulations are implemented to obtain the results of the vessel motions, thruster forces, pump motions and riser tensions. Using optimal Latin hypercube sampling, an RBF neural network approximation model is established, the input includes environmental factors and the output includes the dynamic responses of the pump motion and riser tension. Calculations are performed using RBF network approximations instead of a coupled model. The obtained results show that the PSV wave frequency (WF) motions have significant influence on the dynamic responses of the subsea system. Moreover, the current load affects the compensation effect. The RBF network approximation model can be used to reduce the required calculation time.     

褐藻胶结构的统计学描述及其应用

从褐藻胶线性链块结构的本质出发,运用结构单元的特征性及简单的统计原理,得到了各种结构单元频率之间的相互关系,并阐述了这些结构参数的化学意义,及其初步的应用,从而发现了一些实验现象的理论依据。为探讨褐藻胶组成、结构、功能之间的关系,建立一个对组成、结构定量描述的参数系数。     

基于不变性矩特征的人工神经网络2-D形状识别

在2—D不变性矩快速计算的基础上，利用人工神经网络，使用不变性矩完成2—D形状的识别。实验结果与现有文献的结果比较，本文采用的方法具有特征数目少、意义统一、识别快速、准确的特点。