2014 | OriginalPaper | Buchkapitel
Real-Time Adaptive Residual Calculation for Detecting Trend Deviations in Systems with Natural Variability
verfasst von : Steven P. D. Woudenberg, Linda C. van der Gaag, Ad Feelders, Armin R. W. Elbers
Erschienen in: Advances in Intelligent Data Analysis XIII
Verlag: Springer International Publishing
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
Real-time detection of potential problems from animal production data is challenging, since these data do not just include chance fluctuations but reflect natural variability as well. This variability makes future observations from a specific instance of the production process hard to predict, even though a general trend may be known. Given the importance of well-established residuals for reliable detection of trend deviations, we present a new method for real-time residual calculation which aims at reducing the effects of natural variability and hence results in residuals reflecting chance fluctuations mostly. The basic idea is to exploit prior knowledge about the general expected data trend and to adapt this trend to the instance of the production process at hand as real data becomes available. We study the behavioural performance of our method by means of artificially generated and real-world data, and compare it against Bayesian linear regression.