Neural Networks: A Comprehensive Foundation
Neural Networks: A Comprehensive Foundation
A tutorial on support vector regression
Statistics and Computing
A fast grid search method in support vector regression forecasting time series
IDEAL'06 Proceedings of the 7th international conference on Intelligent Data Engineering and Automated Learning
Hi-index | 0.00 |
In numerous practical applications the interesting measurands are not explicitly available by existing sensors or unaffordable due to high cost of explicit sensor principle. Virtual sensors, as one particular means to design intelligent integrated sensory systems (I2S2), offer an solution to this problem, by merging various sources of information to generate the desired measurand for the given environmental stimuli. In this paper, radial-basis-function-networks (RBFN) and supportvector-regression (SVR) are compared for knock-detection in combustion engines with regard to ease of learning, generalization, and resourceefficiency. Additionally, the notion of a hybrid virtual sensor (HVS) is introduced here for invariance and complexity reasons. In our experiments, real-world engine data has been applied for method comparison and recommendations for parameter settings. SVR shows better generalization results than RBFN for the criteria correlation coefficient and absolute mean error are applied. In future work, we will integrate HVS concept in our emerging tool for automated I2S2 design.