基于RBF神经网络的温度场重建算法研究

Temperature field reconstruction algorithm based on RBF neural network

在声学法锅炉炉膛温度场测量中,重建算法是实现炉膛温度场重建的关键。本文提出一种基于径向基函数神经网络的复杂温度场重建算法。该算法首先对被测温度场用离散余弦变换,建立离散余弦变换低阶次项DCT系数向量与声波路径平均温度向量的映射关系,然后利用RBF神经网络良好的函数逼近能力实现该映射关系,并通过正交最小二乘法进行学习和训练,实现被测温度场的重建。本文对3种原型温度场进行了重建,并在40 dB、30 dB和20 dB等3种不同噪声水平下进行了重建实验。仿真及初步实验结果表明,该算法具有温度场重建精度高、速度快、抗干扰能力强的特点。

Reconstruction algorithm is essential to reconstruct temperature field image in boiler temperature field survey. This article puts forward a new algorithm based on RBF Neural Network. The algorithm first uses Discrete Cosine Transform （DCT） on temperature field, and establishes a mapping relation between low order term coefficient vector and sound wave path average temperature vector, then implements the mapping relation using RBF Neural Network that has strong function fitting ability. Through training with Orthogonal Least Square Method, the temperature field can be reconstructed. Three primary model temperature fields were reconstructed in a simulation experiment. Further experiments were carried out with noisy data under noise levels of 40 dB, 30 dB, 20 dB SNR respectively. Simulation results show that the algorithm features high precision, fast speed and good noise-rejection ability.