Modelo Cliente-Servidor sin Control de Estado para Aprendizaje Profundo de Datos en Dispositivos IoT Aplicados a Parámetros Ambientales

Abstract

The exponential growth of IoT devices requires the research and development of new architectures for the management of sensor access protocols, client-server operations, and the analysis of large volumes of data with multiple related parameters. The objective of this work is to investigate and propose a client-server model, stateless, for access to IoT sensors, with MQTT brokers and REST architecture. Through in-depth analysis, the model groups the CO2 concentration values (objective variable) of a given place, to then correlate the results with the possible effects on people's health, over time. The sensors are accessible in real time through GNSS gateways (with access to LTE-M1, WiFi mesh or Lorawan cellular networks), monitored and managed with SNMP/Netconf protocols. The normalization of the variable is done with external environmental data obtained by geolocation. We compared the results of K-NN. K-Means and GMM for machine learning (supervised and unsupervised) and location risk group assignment for the CO2 concentration variable, in time ranges. With the information obtained, correction (or alarm) actions can be carried out on other devices controlled by IoT to regulate the ventilation of the place and its operational capacity.