The rapid growth of IoT-based smart environments has created a multitude of considerable security vulnerabilities due to heterogeneous devices and constantly changing network conditions. Traditional intrusion detection systems (IDS) often lack the sophistication necessary to effectively detect new attacks, requiring innovative, dynamic methods. In this paper we propose a unique intrusion detection framework that utilizes a Hierarchical Bi-directional Long Short Term Memory (Bi-LSTM) model optimized with a Multi-Objective Bat Algorithm (MOBA). The Bi-LSTM model learns from future, and past data sequences to improve intrusion detection performance by enhancing accuracy when learning temporal patterns. The MOBA model optimizes the Bi-LSTM model by learning to select the best features for a supervised classifier, optimizing the model concurrently with optimizing the parameter tuning. Experimental verification of the proposed framework utilizing benchmark datasets confirms the proposed IDS framework is generally superior to the other methodologies for detection accuracy, total cost of false alarms, and total time to execute the IDS framework. The proposed IDS also shows a distinctive balance of balance feature selection and optimizing parameters, and great opportunity of performance without sacrificing execution times, potentially enabling the deployment of the IDS in resource-limited IOT environments and the ability to create an intelligent and scalable response to changing cyber-threats.