Hu Shaolin, Chen Xianxi, Sun Guo Xi
Guangdong Provincial Key Laboratory of Petrochemical Equipment Fault Diagnosis, Guangdong University of Petrochemical Technology, Maoming, People's Republic of China.
R Soc Open Sci. 2025 Aug 13;12(8):241957. doi: 10.1098/rsos.241957. eCollection 2025 Aug.
This paper presents a sliding window constrained fault-tolerant filtering method for sampling data in petrochemical instrumentation. The method requires the design of an appropriate sliding window width based on the time series, as well as the expansion of both ends of the series. By utilizing a sliding window constraint function, the method produces a smoothed estimate for the current moment within the window. As the window advances, a series of smoothed estimates of the original sampled data is generated. Subsequently, the original series is subtracted from this smoothed estimate to create a new series that represents the differences between the two. This difference series is then subjected to an additional smoothing estimation process, and the resulting smoothed estimates are employed to compensate for the smoothed estimates of the original sampled series. The experimental results indicate that, compared with sliding mean filtering, sliding median filtering and Savitzky-Golay filtering, the method proposed in this paper can more effectively filter out random errors and reduce the impact of outliers when dealing with sampling data contaminated by noise and outliers. It possesses strong fault tolerance and the ability to extract the true variations of the sampling data.
本文提出了一种用于石化仪表采样数据的滑动窗口约束容错滤波方法。该方法需要基于时间序列设计合适的滑动窗口宽度，以及对序列两端进行扩展。通过利用滑动窗口约束函数，该方法对窗口内的当前时刻产生一个平滑估计。随着窗口的推进，生成一系列原始采样数据的平滑估计。随后，从这个平滑估计中减去原始序列，以创建一个表示两者差异的新序列。然后对这个差异序列进行额外的平滑估计过程，并将得到的平滑估计用于补偿原始采样序列的平滑估计。实验结果表明，与滑动平均滤波、滑动中值滤波和Savitzky-Golay滤波相比，本文提出的方法在处理受噪声和异常值污染的采样数据时，能够更有效地滤除随机误差并减少异常值的影响。它具有很强的容错能力和提取采样数据真实变化的能力。
