Research on the Application of Deep Reinforcement Learning Algorithm for Sensor Networks in Intelligent Environment Control

Abstract

Real-time search of objects, data, and services has grown to be a crucial and practical problem in the IoT world in recent years due to the fast deployment of numerous Internet of Things (IoT) sensors. In addition to offering solutions for real-time sensor data collecting in the IoT, IoT data prediction models also provide more useful applications than conventional IoT event detection models. Using complicated time series created by several types of sensors, we have developed an enhanced neural network model in this research for multi-dimensional feature selection and outlier identification. Comparing our approach to conventional data prediction methods, we can see an improvement in the stability and accuracy of long-term forecasts for IoT sensor data. Finally, we compare the training capabilities of traditional neural network models with random forest models to assess the efficacy of our prediction model using sensor data from the Intel Berkeley Research Laboratory. Our model achieves an accuracy rate that is respectively more than 12% and 68% higher than the two comparison models, while the corresponding loss values are less than 0.325 and 0.595, respectively.