Fuel Classification based on Flame Characteristics using a Time Series Analysis with Fuzzy Support Vector Machine Algorithm

Flame monitoring and categorization become important in efficient power generation and energy conservation sectors. Flicker Frequency Range (FFR) is considered as the key factor to monitor flames and fuel identification. The fluctuation of combustion arises due to the varying quality of fuel, which causes the overlapping of FFR for different fuels leading to misclassification. In this work, we attempt to classify fuels based on flame characteristics output from a Digital Flame Scanner. A new model is proposed by combining Time Series Analysis with Fuzzy Support Vector Machine (TSA‐FSVM) for training, classification, and prediction of fuel types. With the help of 12,000 real‐time data collected from the Bharat Heavy Electricals Limited (BHEL), Tuticorin branch (a public sector company, which manufactures power plant equipment), a comparative analysis is performed using different Machine Learning algorithms with the proposed technique. From the results, it is found that TSA‐FSVM outperforms existing as well as other Machine Learning methods by increasing the accuracy of predicting the right fuel. Thus, it helps to avoid the boiler system explosion/OFF state, which leads to conserving the energy, increases electricity production, and cost‐efficiency.