Neural Network Control of Environmental Parameters in Neonatal Incubators

The neonatal incubator is a multi-connected, non-linear, non-stationary, stochastic object. Classical methods of control such an object not provide the desired quality of regulation. There is clinical demand for high-precision maintenance of microclimate parameters in the incubator. The author's method of gradient neural network control was used to control the multi-connected output parameters of the incubator. The structure of the dynamic system includes a neural network block and a block of a physical object (neonatal incubator). The dynamic system was described by a discrete model. The set of control coordinates at the step of the discrete process are the weight coefficients of the neural network. The objective of the control is set as a non-negative function which characterizes the deviations from the specified temperature and humidity parameters. The output of the neural network considered equal a control vector which was formed that affects the change in microclimate parameters and based on the values of the set of weight coefficients. Neural networks of direct propagation were used to build a neural network control system. The neural network control algorithm was based on the correction of weight coefficients using the gradient control method and error backpropagation. The regularization method was used to solve the problem of stability of the control process. The method of gradient neural network control in neonatal incubators is at the stage of practical implementation using a Raspberry Pi 4B 4GB RAM mini-computer and the Python programming language.