基于RBF神经网络的软测量算法的SoC实现

Implementation of SoC of soft measurement algorithm based on RBF neural network

提出了一种面向片上系统(SoC)的RBF神经网络的软测量算法,在OMAP—L137双核处理器SoC硬件平台上成功实现了整个训练与预测算法。针对SoC计算速度和存储空间等资源有限,对网络结构、权值更新模式和步长以及数据预处理方式等参数提出了具体的解决方案。经过相关数据集的测试结果表明:提出的算法移植方法完全满足工业应用的要求,且具有便携性、低成本、可扩展等多种优点。

An algorithm of RBF neural network soft measurement based on SoC is proposed. The entire training and prediction algorithms are implemented successfully on the hardware platform of dual-core SoC processor. In order to apply algorithm effectively on the platform whose compute speed and memory are limited, specific solutions about the structure of the network, weight update pattern and step and data preprocessing mode is proposed. The data set test results prove that the algorithm transplantation method satisfies the requirements of industrial application, which has a variety of advantages such as portable, low-cost and extensible.