One of the principal challenges with Smart Grids is the very slow rate of development originating from the lack of investments from the governments and major companies. As a solution, the concept of the Internet of Energy (IoE) is capable of differing the need for massive investments and changing the business model of energy sharing so that end-users participate in the development process. The IoE envisions the next generation of smart grids as a fully interconnected network, including advanced metering infrastructures, distributed energy resources, and bidirectional communication systems. The first key question is how to convince end-users to participate and invest in upgrading the current power system. The feasibility of any possible solution is linked with the profitability of the whole process by reducing the electricity cost and maximizing the profit of energy trading. Consequently, optimizing operational scheduling and electricity routing are two fundamental problems that need to be addressed. However, the accuracy and originality of data must be guaranteed prior to utilizing it in solving scheduling and routing problems. The open architecture of the IoE-based smart grid results in manifold security concerns, especially the risk of False Data Injection attacks. The attack may target the technical aspects of a system since fabricating the network's data misleads power scheduling and routing strategies and interrupts the healthy operation of the power system. Also, the high penetration of smart devices in IoE-enabled smart grids, besides decentralization originating from employing renewable resources, faces the power system with intricate optimization problems, including operational scheduling and electricity routing problems. Accordingly, this thesis is on the application of the Internet of Energy in smart grids using Deep Reinforcement Learning, aiming to reduce costs and losses for both generator and consuming sides, considering the correctness of data in the system. The first objective of this research is to enhance the cyber defense of the Internet of Energy-enabled power systems against False Data Injection attacks. To this end, an intelligent intruder is first developed to generate innovative threats that the model has not previously seen. Moreover, well-known attack strategies are modeled to create passive attacks simultaneously. Next, the quality of the developed attack is examined using the proposed defense algorithm in the literature to demonstrate the necessity of a more powerful attack detection mechanism. Then, a Multi-Layer cyber defense mechanism is developed to detect both passive and active threats. After guaranteeing the originality and correctness of data, the second objective is optimizing the operational scheduling of all energy components in the system. Accordingly, a novel algorithm named Probabilistic Delayed Double Deep Q-Learning, which is a combination of the tuned version of Double Deep Q-Learning and Delayed Q-Learning has been proposed to optimize energy scheduling problems in IoE-based power systems. This algorithm makes a trade-off between overestimation and underestimation biases, guaranteeing sample complexity by applying a delay in updating the rule. Finally, to fulfill the last objective, which is optimizing electricity routing, a novel algorithm titled Approximate Reasoning Reward-based Adaptable Deep Double Q-Learning (A2R-ADDQL) is introduced specially to optimize electricity routing in residential units. As a result, both positive and negative biases are reduced compared to other deep Q-Learning-based algorithms. Moreover, the sample complexity of the model is decreased due to utilizing a fuzzy approximate reasoning reward function.