二进离散小波神经网络在传感器逆向建模中的应用

Research on Wavelet Neural Networks and Its Application to Sensor Reverse Modeling

给出了快速收敛的离散二进小渡神经网络的初始化．构造和权值确定的详细方法。并将这类小波神经网络应用于传感器的非线性校正，并给出了仿真实验结果。相对使用随机贪心算法训练的神经网络，快速收敛小波神经网络利用离散二进小波变换的便利，采用启发式的构造算法；具有构造过程复杂度低，构造完成后高度接近目标模型，训练次数少，并可有效避免陷入局部极小点的优点。有效解决了小波神经网络尺度和平移系数在训练时需对小波函数进行求导而影响网络收敛速度的问题。

details of initial, construction and parameters determine for a fast-converging discrete binary wavelet neural networks （WNN） has been introduced, This kind of WNN is used in sensors non-liner error correction and a simulation result is given, Compare to other neural networks trained by stochastic gradient method, theory of discrete binary wavelet transform speeds up the training procedure of fast-converging WNN and makes this WNN avoid converging to a local maximum point. This fast version of wavelet neural network is also not complicate in building, It solved the problem of slow convergence of dilation and translation parameters which caused by differentiating wavelets.