Design And Implementation Of Neural Network Soft Measurement Platform Based On SOC

With more and more application of soft measurement algorithm facing complex industry process, in view of the high cost and low portability of high performance computer or digital devices realizing, this paper studied a general soft measurement platform based on SOC whose core is dual-core processor OMAP-L137. This platform combines the real-time task scheduling and the floating-point operation precision. The all steps of RBF neural network algorithm have been transplanted into the embedded system after analysing and researching several key factors within the design and implementation process detailedly. Based on the comprehensive index of the prediction accuracy and calculate speed, the problem of poor real-time ability about algorithm embedded application has been solved successfully, and has good computing accuracy.Firstly, the hardware is built on the unique idea of modular design, which consists of core board, motherboard, I/O interface board and user interface board. Then, a deep research on the embedded application of neural network soft measurement algorithm is carried on. On the basis of C language level realizing about algorithm, some key elements have been studied such as network structure, weights update mode, weights update step, data preprocessing method, algorithm stop condition and so on. The solutions have been put forward and incorporated into a whole optimization strategy.The experiment results on the relative data sets show that the whole training and prediction process are realized in the SOC successfully, which also satisfy the calculation speed and precision requirements. At the same time, some necessary system application software have been wrote with the core algorithm, so that the platform can accomplish the soft measurement tasks, and has the advantages of portability, versatility, low cost, small volume and extensibility. It can not only be applied to various industrial process object, but also be extended to other intelligent algorithm transplantation or other application fields.