Decision Support System for Porous Ceramic Matrix-based Burner by Hybrid Genetic Algorithm-Supervised Kohonen Map: A Comparative Assessment of Performance of Neural Network Under Different Minor Attributes

Hybrid Genetic Algorithm (GA)-supervised Kohonen (SK) map is explored for the decision support system in the operation of porous ceramic matrix (PCM)-based burners. Four features of PCM are selected for defining the regime of operation of PCM-based burner. Based on the values of these four features, 16 distinct regime of operations are identified. Hybrid GA-SK map is fed with the numerical generated temperature profiles of the PCM. The GA is able to give the architectural details of the best SK map. The SK map is then supplied with new samples from PCM, and good prediction of their regime of operation is obtained. Minor attributes of hybrid GA-SK map are altered and analyzed for the higher accuracy in prediction of regimes. Hexagonal grid under eigen values initialization of weights was able to give highest average class prediction (acp) of 57.14%. Under initialization of weights by eigen values, a network of 10 x 10 size and 300 epochs gives high optimization criterion of 0.79, while maintaining high frequency of 0.6. Present work intends to strengthen the hybrid GA-SK map approach for decision support system for PCM-based burner.

Automated summary

This study explores the application of a hybrid genetic algorithm (GA)-supervised Kohonen (SK) map in the decision support system for porous ceramic matrix (PCM)-based burners. Four selected features of the PCM are used to define the burner's operating regime, resulting in 16 distinct regime classifications. The hybrid GA-SK map is trained using the generated temperature profiles of the PCM, and is able to accurately predict the regime of operation for new PCM samples. Minor attributes of the hybrid GA-SK map are adjusted to improve the accuracy of regime prediction. The use of a hexagonal grid with eigen values initialization of weights achieves the highest average class prediction (acp) of 57.14%. A 10 x 10 network size with 300 epochs and eigen values weight initialization achieves a high optimization criterion of 0.79, while maintaining a high frequency of 0.6. This research aims to strengthen the hybrid GA-SK map approach for decision support in PCM-based burners.