Particle filter-based parameter estimation in a model of the human circadian rhythm

Recent insights into the effects of light on human health call for a more human-centric approach in automatic lighting control systems. We contribute to the provisioning of lighting settings tailored to the needs of individuals by addressing the challenge of predicting the response of an individual’s circadian rhythm to light exposure. Existing models of the human circadian rhythm are not tailored to individual physiological characteristics such as intrinsic circadian period, light sensitivity and age. We propose to improve model accuracy by using Bayesian statistical inference to estimate the values of model parameters that reflect these physiological characteristics. We illustrate our generic method by applying to a combination of two popular models of the circadian rhythm. By processing individual light exposure- and actigraphy data recoded during a field trial with 20 human subjects with a Particle Filter, we estimate each subject’s intrinsic circadian period. When correlating these to the subjects’ Munich Chronotype Questionnaire Midsleep on Free Days time, a significant relationship was found: r > 0.6 and p < 0.01. This shows the proposed method has good potential for improving model accuracy.