Flow inferential measurement for proportional control valves by combining wavelet denoising and a dual-attention-based LSTM network

Abstract

Accurate flow inferential measurement enables accurate real-time acquisition of the flow rate for hydraulic systems, effectively replacing conventional expensive and space-consuming flowmeters. However, the high nonlinearity and complexity of valve flow pose significant challenges for achieving accurate flow inferential measurements. To address this issue, this paper proposes a novel method based on wavelet denoising and a dual-attention-based long short-term memory (DA-LSTM) network. The DA-LSTM network is innovatively proposed to learn the flow mapping relationship, and incorporates the intervariable attention mechanism and the variable self-attention mechanism to enhance learning performance. Additionally, considering the presence of noise contamination in the measurement datasets, the wavelet thresholding denoising method is employed to increase the data quality. Furthermore, the real-time performance of the proposed method is also considered. The trained model is validated against test datasets, and is also compared to three other neural network-based flow estimation methods. The experimental results demonstrate that the proposed method accurately realizes the flow inferential measurement of the proportional control valve, with a mean square error percentage of 0.6494 %. This establishes a robust foundation for accurate flow control of proportional control valves.