Buzdolaplari için hassas sicaklik kontrolü

This thesis explores the development of an advanced adaptive PID controller for optimal temperature control in refrigerators while maintaining low power consumption. The proposed controller combines PID and reinforcement learning with a Radial Basis Function network, designed under an Actor-Critic structure. The framework adapts PID parameters based on reinforcement learning techniques, enabling real-time adjustments in response to system conditions. The paper's focus on leveraging reinforcement learning ensures that the controller can learn from experience and fine-tune its behavior, resulting in improved efficiency and reduced power consumption. The Proportional, Integral, and Derivative components are crucial for achieving precise control for PID. The reinforcement learning approach introduces a novel adaptive weight updating rule adjustment over RBF network, allowing for on-line tuning of the PID parameters based on predictive outputs and reinforcement signals. By sharing the hidden layer between the Actor and Critic, the storage space requirement is minimized, and computational costs are reduced. Moreover, customer profiling through door actuation and ambient temperature data from IoT refrigerators allows for user-adaptive cooling algorithms. These algorithms improve cooling performance, reduce energy consumption, and minimize temperature fluctuations. This integration of reinforcement learning into adaptive PID controllers reflects a shift toward more intelligent and flexible control systems in modern refrigeration technology. The data collection phase involves obtaining anonymized information on refrigerator door openings, ambient temperature, and humidity levels from IoT refrigerators. The deployment stage consists of both Cloud Deployment and Control Board Deployment, with clustering algorithms like K-means used to categorize data and generate user profiles. By deploying these profiles to the cloud and control board, the learning rate and PID parameters are optimized, resulting in a more efficient cooling system with reduced energy consumption.