Abstract: For the sake of improving the reliability of measured data, the bi-sensors data fusion method is widely applied to obtain the information on product status in the process control field, and the linearly unbiased estimation theory used as the data fusion method is widely adopted to deal with measurement data. For any estimation process, both variance and bias are the significant indices evaluating the performance of the measurement data. Because the unbiased measurement data possesses unbiased merit, variance can be used to represent the reliability of the unbiased measurement data. However, we can not believe that the unbiased measurement data must be highly reliable. In order to enhance the measurement reliability of the bi-sensors, a novel data fusion method arising from biased estimation was proposed. First, the conclusion that biased measurement could increase the reliability of a single-sensor was proved. Second, the convex combination method was used to derive biased estimation data fusion given the bi-sensors. Finally, the conclusion that the proposed fusion process was always superior to single-sensor measurement in terms of mean square error was proved. Simulation and application in a real plant illustrated that the biased estimation fusion could improve the reliability of the measured data effectively.

Key words: measurement, bi-sensors, reliability, biased estimation, data fusion

摘要： 为了提高测量数据可靠性，过程控制领域广泛采用双冗余传感器测量生产状态信息，而目前处理双传感器测量数据常用的方法为线性无偏估计融合理论。通过估计理论得到的测量数据同时涉及方差和偏差两个统计特性。由于无偏测量数据具有无偏性的优良性质，所以测量方差能够全面地描述无偏测量数据的可靠性，然而不能由此认为无偏测量数据一定具有高可靠性。为了进一步提高双传感器测量数据的可靠性，提出有偏估计数据融合方法。首先，证明了有偏测量能够改善单传感器测量数据的可靠性；其次，采用凸组合方法推导了双传感器有偏估计融合表达式；最后，证明了有偏估计融合的均方误差小于任意单传感器的均方误差。仿真分析与实例应用均表明有偏估计数据融合可以有效地提高双传感器测量数据的可靠性。

关键词: 测量, 双传感器, 可靠性, 有偏估计, 数据融合