Design of Adaptive Control Systems for IoT Devices on the Tuya Development Platform Using Reinforcement Learning

Considering existing smart home management systems for both local and remote devices operating on the Tuya or Smart Home platform, several deficiencies have been observed during operation, impeding adaptive control and the overall maintenance of system functionality. The key shortcomings include the limited effectiveness of software tools for automating scenarios involving various actions on hardware devices, as well as an external dependency on the remote servers of the Tuya Development Platform. As consequence of these shortcomings, considerable electricity costs arise due to devices remaining idle while in an active state. The proposed solution encompasses a module for local device management which accounts for the analysis of various contextual factors such as user behavior patterns and external sensor data through a Proximal Policy Optimization model developed using Reinforcement Learning. The outcomes of the developed software include a flexible data collection system, an enhanced Proximal Policy Optimization algorithm for adaptive learning based on a Long Short-Term Memory model with attention layer, and a web interface for user interaction. The output of such a system enables the resolution of automation-related issues in device management due to the flexible reward mechanism of the training model, as well as addressing problems associated with device energy consumption.