Furnace thermal efficiency modeling using an improved convolution neural network based on parameter-adaptive mnemonic enhancement optimization

Abstract Thermal efficiency is a very important factor in furnaces. Accurate thermal efficiency modeling plays a key role in optimizing and improving thermal efficiency. Industrial process data own complex characteristics of high nonlinearity and time series. Convolutional neural network (CNN) has been widely used to build models due to the strong ability in processing high-nonlinear and time-series data. However, in the traditional CNN method, the repeatability of calculations is not considered when updating weights; only the latest updated weights are reserved. As a result, some important computation experience is abandoned, which weakens the performance of CNN. In order to solve this problem, an improved CNN method integrating with parameter-adaptive mnemonic enhancement optimization (AMEO-CNN) is proposed in this paper. In the proposed AMEO-CNN, previous computational experiences are adaptively reserved to minimize output errors using optimized weights. The continuity of the algorithm and the anisotropy of the initial values are proved through theoretical analyses. In order to validate the performance of the proposed AMEO-CNN model, thermal efficiency modeling simulations of an industrial heating furnace are carried out. Compared with other three models, simulation results show that the proposed AMEO-CNN model can achieve higher modeling accuracy, which verifies the effectiveness of the proposed AMEO-CNN method.