Intelligent hydrogen flame recognition method: Comparisons among SVM, GA-SVM and PSO-SVM

In this paper, the Genetic Algorithm and Particle Swarm Optimization are used to obtain the kernel parameters of the radial basis function of the Support Vector Machine (SVM), and the optimized SVM algorithm is used to realize intelligent identification of hydrogen flame. The feasibility of this method is a crucial aspect that assures its practicality. Dual-band $(1.9 \mu \mathrm{~m}$ and $2.7 \mu \mathrm{~m}$) infrared sensors are used to collect hydrogen flame data in the special infrared band. The hydrogen flame data is used to extract its time-domain and frequent-domain characteristics, including the average voltage value of the hydrogen flame time-domain data and the frequency corresponding to the maximum frequency amplitude and maximum amplitude of the frequency domain data. The above feature data is used as the algorithm input. The classification accuracy of the training set, test set, optimization time, and error fitness curve are taken as the evaluation indexes of the algorithm, and the complex hydrogen flame identification problem is transformed into a binary classification problem with faster speed, higher efficiency, and higher accuracy by support vector machine algorithm. The experimental results show that the PSO-SVM has the highest accuracy on hydrogen flame, and SVM has the shortest optimization time. GA-SVM has the longest optimization time, and its accuracy is lower than that of PSOSVM. Moreover, the recognition accuracy of the PSO-SVM training set and the test set are ${9 9 \%}$ and ${1 0 0 \%}$, respectively, and its optimization time does not exceed 2 s. The recognition effect is better than that of the traditional support vector machine, and a genetic algorithm optimizes the support vector machine, which means that the hydrogen flame identification and detection method based on the SVM algorithm is feasible, and the particle swarm optimization can obtain better recognition accuracy.