一种改进的定性趋势识别方法

摘要

Signal trend comprises important process state information, and trend identification can be used for fault diagnoses. But under the influence of random noise, external disturbance and signal fluctuation, it’s hard to identify linearity primitives. According to statistics, this paper inducts confidence interval of derivative value. It converts the enactment of linearity primitive derivative threshold into confidence degree enactment, improving the adaptability and robustness of trend identification arithmetic. Examples indicate that this way can identify linearity primitives exactly.%　　信号趋势包含重要的过程状态信息，识别趋势可以进行故障诊断。但由于随机噪声、外部干扰及信号波动等因素的影响，很难准确识别线性基元。从统计学的角度出发，引入导数值置信区间。把线性基元导数值阈值的设定转换成置信度的设定，从而提高了趋势识别算法的适应性和鲁棒性。实例验证表明该方法能准确识别线性基元。