Firstly , studied feed - forward neural network and put forward a new algorithm on bp network , called bp algorithm based on robust error function ( bparef ) , and the algorithm is proved to be effective for approaching nonlinear system 首先研究了前馈神经网络，提出了基于鲁棒误差函数的bp神经网络的算法，并且验证了其对非线性系统逼近的有效性。
Back analysis of thermal and mechanics parameters are carried out by using dfp method and improving minimum two - multiplied error function based on the practical datas of temperature and strain of the hole 结合引水洞的温度和应变实测资料，利用考虑时间效应、无量纲化的最小二乘误差函数，采用梯度优化方法中变尺度方法对其热、力学参数进行了确定性反演。
At present , during the course of bp neural network ' s learning and training , we often adopt the algorithm of back - error propagation , which is based on global error function ' s gradient descent 特别是bp网络近年来广泛应用于模式识别、预测评估等领域，并取得良好的效果。目前bp网络采用误差逆传播算法学习训练神经网络，该算法是基于网络误差函数梯度下降的。
At present , during the course of bp neural network ' s learning and training , we often adopt the algorithm of back - error propagation , which is based on global error function ' s gradient descent , whose essence is a point - to - point search algorithm 目前bp神经网络采用误差逆传播算法学习训练神经网络，该算法是基于网络误差函数梯度下降的，其本质是点到点的搜索方法。
The performance of the differential mode loop , the characteristic of mrc combining and the high - pass characteristic of the error transfer function of the uniform sampling second - order dpll are simulated in the matlab , the simulation results are given 在matlab环境下对分集合并中的差模环特性、最大比合并特性以及均匀采样二阶数字锁相环误差函数高通特性进行了仿真,给出了仿真结果。
Because the weight space formed by the bp neural network ' s global error function , which include extreme point , is a hyper - surface , considering the initial parameters of bp neural network ' s structure , so the bp neural network has a inherent deficiency of easily falling into local minimum 由于bp神经网络的全局误差函数构成的权值空间是包含多个极值点的超曲面，加之即网络训练开始网络结构参数是随机给定的，因而bp神经网络在学习训练中容易陷入局部极小。
And then the stability and the stable error response of the dpll are analyzed . there is an obvious doppler shift frequency offset in the received signal with flight at a high speed , in the dissertation the uniform sampling second - order dpll is used to remove the frequency bias for its excellent high - pass characteristic of the error transfer function 由于高速飞行的遥测目标在接收信号中会出现较大的多普勒频率分量,而典型均匀采样二阶数字锁相环dpll ( digitalphase - lockedloop )的误差函数具有高通特性,所以本文提出一种利用其高通特性来去除多普勒频率分量的新方法。
6 . since the monitor precision of quasi - dynamic calibration system is very important , the model of the four channel electric pressure - measuring system in frequency domain is built up . the cause of variance is analyzed . the error function of the system is defined and the formula of correcting the system error is derived 6由于落锤液压动标装置校准系统的校准压力监测精度极为重要，本文对四路电测压力系统进行频域建模，分析了产生不一致的原因，定义了系统的误差函数及推导了统一误差模型的修正公式。
Automatically scan , decision feedback loop and costas loop combined together are adopted . simulation proved the performance accord with system need . further more , a synchronization project using error function linearization as the core is brought forward 在此基础上提出了大多普勒频移下载波捕获的设计方案，采用自动扫描、判决反馈环和costas环相结合，通过了仿真证明了其性能符合系统要求，进一步提出了一个以误差函数线性化为核心的同步方案以改善低信噪比下载波同步性能。
Using the first order and the second order derivative information of the network error function to the learning rate factor / / and the momentum factor ? , dynamic optimization of the learning rate factor / / and the momentum factor ? is obtained during the network training process , which efficiently speeds up the network learning rate 并利用网络误差函数对训练速率系数和冲量系数的一阶和二阶导数信息，在网络训练过程中动态优化调整训练速率系数和冲量系数，有效地加快了网络的训练速度。
