The Internet of Things (IoT) links a wide range of physical devices to the Internet, facilitating cutting-edge applications in fields like the military, healthcare, agriculture, and transportation. These IoT applications have grown in popularity due to their ability to tackle real-time challenges effectively. However, despite the benefits they provide IoT systems are notably susceptible to security vulnerabilities, making them targets for a range of cyberattacks. These threats include DDoS, MiTM, sinkhole attacks, eavesdropping, and DoS attacks. In this work, a novel Real-Time Intrusion Detection in IoT with a Deep Learning-based Multi-Head Attention BiGRU (RIGRU) approach has been proposed to accurately classify IoT attacks. Data is collected from IoT sensors on network devices. It goes through data cleaning and standardization. Dingo Optimization is used to select relevant features iteratively for classification tasks. These features are then fed into a Multi-Head Attention BiGRU Network to detect MiTM, DDoS attack and normal. The proposed RIGRU approach was calculated utilizing various metrics, namely precision, f1score, recall, and accuracy. The RIGRU model advances the overall accuracy by 0.87%, 0.95%, and 0.75%, over the PCC-CNN, DIDS and GNN, respectively.