RBF神经网络校正长距离α测量的非线性研究

摘要

In this paper, the RBF neural network predicting models are established to predict the LRAD meas-urement results of ionization voltage with five measurement factors:test pipe diameter, pipe length, source-to-detector distance, air speed, air flow being inputs, the ionization voltage being output. 239 Puαsource whose radioactivity is 523. 3 Bq being radiation sources, 220 groups of data are measured. Compared with BP neural network, RBF neural network, the mean absolute error is 4. 35%, 0. 769 s of the training time;5. 65% and 1. 453 s of BP neural network. The experiment results show that: the capability of RBF neural network about the nonlinear correction of LRAD measurement is better than BP neural network.%论文以被测管管径、管长、测量距离、风速、空气流量五个测量因素为输入，测量结果电离电压值为输出，利用RBF神经网络对LRAD测量结果进行预测。在活度为523.3 Bq的239 Puα放射源环境下，测量220组数据。将BP神经网络与RBF网络对比，两者的平均绝对误差分别为5.65%、4.35%，训练时间分别为1.453、0.769 s。实验结果表明：RBF神经网络对LRAD测量结果的非线性校正效果优于BP神经网络。