Boosting Algorithm Research And Its Application In Spectrum Analysis

Ensemble learning algorithms can significantly improve the generalization ability of learning systems through training a finite number of weak learners and combine their results. Boosting is a representative algorithm of ensemble learning and has received a great deal of research and application, but it is mainly focused on the problem of classification.This paper takes a research on the Boosting regression algorithm. Firstly, it introduces the Gradient Boosting theory and the no weight regression algorithm based on this theory, then it presents the experimental results of a practical problem. Secondly, the paper shows the results from a great number of experiments of the weight algorithm that chooses samples by probability. The experiments point out that weight algorithm has a better generalization ability than the no weight algorithm. But a single weight algorithm is unstable and it needs tremendous time of calculation to combine many single algorithms into a stable one. So it can not be universally applied. Thirdly, in order to improve the deficiency of these two algorithms, the paper presents an improved algorithm. Experimental results show that, this new algorithm is far more excellent. Its calculating time is based on the precision of the algorithm. So, one who uses this algorithm in practical problems can balance the performance and the calculating time as needed. Lastly, a combination method of no weight algorithm and this improved algorithm is presented to solve a little deficiency. This combination method can't improve the performance of original algorithm, but it can faster the convergence rate in the first few steps.This paper also explains the importance of environment detection and thenecessity of the soft-sensor water analysis instrument. It introduces a multi-parameters water analysis instrument based on the theory of UV spectrum analysis. It also explains the measurement theory, hardware structure, software function and intelligent computing methods in detail. And applies the improved algorithm mentioned before into the intelligent analysis model. Through the experiment on the COD data and compared with some commonly used regression methods, it proves the efficiency and applicability of the ensemble learning Boosting regression algorithm.