Real-Time Monitoring and Analysis of Flame Combustion Optimized by Fusion Convolutional Autoencoder

Intelligent algorithms have been widely used in many fields because of their unsupervised and self-learning characteristics. At present, the application of intelligent algorithm combined with combustion monitoring in the direction of combustion characteristics collection has some problems, such as sample collection missing and feature recognition not obvious. Therefore, an attention-mechanism optimized convolutional self-coding combined with feedforward neural network is proposed to construct a combustion monitoring and diagnosis model for combustion feature extraction and combustion state diagnosis. The results show that the Attention-Convolutional Autoencoder and Radial basis function neural network (ACAE-RBF) model maintains a diagnostic accuracy of more than 95%, the area under the PR curve is 0.83, the balance point value is 0.79, and the prediction diagnosis error rate after 40 iterations is as low as 4%, which is significantly better than the comparison model. The above research results show that the optimization model has excellent performance in the accuracy and precision of combustion detection and diagnosis, and provides a new idea for artificial intelligence in the field of flame combustion monitoring.