Journal of Shanghai Jiao Tong University ›› 2023, Vol. 57 ›› Issue (1): 36-44.doi: 10.16183/j.cnki.jsjtu.2021.300

Special Issue: 《上海交通大学学报》2023年“船舶海洋与建筑工程”专题

• Naval Architecture, Ocean and Civil Engineering • Previous Articles     Next Articles

Sequential Prediction Method of Single Parameter for Thermal System Based on MWSA

XIAO Pengfeia, NI Hea(), JIN Jiashanb   

  1. a. College of Power Engineering, Naval University of Engineering, Wuhan 430033, China
    b. College of Naval Architecture and Marine Engineering, Naval University of Engineering, Wuhan 430033, China
  • Received:2021-08-05 Revised:2021-10-06 Online:2023-01-28 Published:2023-01-13
  • Contact: NI He


The changes in the status parameters of the thermal system reflect the operating status of the system in real time. The forecast results of the trend extraction and time series prediction of the current equipment status parameters can be used as a reference for the next operation management strategy and equipment maintenance, which can be used for the long-term system safe and stable operation. In this paper, a method which is described as MWSA, based on the midpoint and regression based empirical mode decomposition (MREMD), the wavelet threshold denoising (WTD) and midpoint and techniques, the singular value decomposition (SVD) and optimized parameter permutation entropy (PE), and an auto regressive integrated moving average model (ARIMA), is applied to the single-parameter time series prediction of thermal systems. First, the MREMD is used to decompose the monitored operating state parameters into a number of intrinsic mode functions (IMF) and residual components. Next, the components that do not meet the screening conditions are subjected to wavelet thresholding. The denoised components and the components that originally meet the filtering conditions are recomposed into new IMF components. Finally, the K-means clustering algorithm based on SVD and PE is used to classify the recomposed IMF components, the component with a lower entropy value is selected and reconstructed into a trend item, and ARIMA is used to predict. An actual case verifies that this method can effectively overcome the interference of high-frequency noise in the original parameter timing, and the prediction accuracy is higher than that of similar methods without noise reduction treatment.

Key words: sequential prediction, mode decomposition, threshold denoising, clustering algorithm, auto regressive model

CLC Number: