Neural Network Determined Thermal Regulation of Systems with Remote Inputs

Abstract

Human thermoregulation models, where metabolic rate is used to maintain thermal homeostasis, have been of interest to anticipate human physiological changes in extreme temperatures as well as to predict comfort. Existing thermal models, however, have limitations in health diagnostics due to the reliance on averages that are used to model the response of a canonical person. Under identical thermal environments, any two people will respond differently because of differences in body composition, vasculature, acclimatization, and so forth. Replacing the heuristics with a neural network control system trained on measured data could be advantageous for these healthcare applications. The challenge of using a neural net for control of the human body thermal model is that we can not measure those parameters that we want the neural network to sense as input nor that will be controlled as output. The brain has access to temperature sensors throughout the body and adjusts the thermal generation mechanisms to maintain homeostasis. Yet, we are unable to train a neural network to behave the same way because we do not have access to the internal sensors that are available to the brain. The initial steps for this work were successful. A preliminary study using a one-dimensional forward conduction model instead of the human body thermal model was performed first. The neural network was able to maintain a steady internal, core temperature in various varying environmental temperature patterns. Beginning work on the human body thermal model and neural network system is shared briefly.