Design of an AI sunroom control system based on neural network speech recognition and multisensor intelligent perception fusion
Abstract
To address the shortcomings of traditional sunrooms, such as tedious manual control, numerous safety hazards, and sluggish environmental response, this paper designs and implements an intelligent sunroom control system centered around an STM32 microcontroller. The system adopts a hierarchical modular architecture to construct a multi-sensor collaborative perception network composed of raindrop, photosensitive, and MQ-2 smoke sensors. Furthermore, a decision-level weighted fusion strategy is introduced, which comprehensively evaluates the environmental status after normalizing the data from each sensor, thereby effectively suppressing false alarms from any single sensor. Through multichannel parallel data acquisition and a correlation control decision matrix, the environmental state is accurately identified. Concurrently, a positional PI closed-loop control algorithm is linked to dynamically regulate the rotational speed of the DC geared motor, which significantly enhances the anti-interference capability of the system under complex weather conditions while ensuring the smooth opening and closing of the roof. Regarding interaction and security protection, the system integrates an ASRPRO offline speech recognition module based on a neural network processor, a Bluetooth APP, and physical button controls. Specifically, the speech recognition module executes neural network forward inference on the device side, achieving millisecond-level responsiveness without requiring network connectivity. The system is equipped with a comprehensive safety mechanism encompassing smoke alarms, limit protection, and motor stall protection. Physical testing demonstrates that the system operates stably and exhibits excellent adaptive perception and precise control capabilities, thereby providing a feasible reference solution for the intelligent retrofitting of small and medium-sized residential buildings.