模糊神经网络竞争学习算法及在加速度表动态测试中的应用

A Competitive Learning Algorithm for Fuzzy Neural Network and Its Application to Dynamic Measurement of Accelerometer

加速度表是飞行器的重要传感器之一 ,使用前需进行动态测试、误差分离和补偿。而这些有赖于加速度表的模型。由于漂移、耦合等不确定因素 ,致使难以建立加速度表的准确模型 ,提出用模糊神经网络作为加速度表的建模手段 ,并通过把模糊神经网络的学习过程转化为竞争聚类和最小二乘优化 ,提出了一种基于竞争聚类的模糊神经网络学习算法 ,在某导弹加速度表上的实物实验对这一方法进行了较好的验证。

The accelerometer is one of the important device of the flight vehicle. We must make dynamic measurement so as to isolate and compensate the error before we use it. All of these depend on accelerometer′s model, but it is difficult to build the model due to the random factors such as drift and coupling. In section 1, we get a neural network model (eq.4) based on fuzzy logic, σ j k, X j k and Y k are to be determined. In section 2, we put forward a competitive learning algorithm for determining σ j k, X j k and Y k in eq(4). We divide the learning process into two steps, one is competitive cluster, the other is least square error optimization. In section 3, we use the algorithm to model the accelerometer in a certain flight vehicle. Fig.3 gives the error between the model and the real accelerometer, it shows that the algorithm is effective.