With the advancement of human society and information and communication technology, the Internet of Things has penetrated into various aspects of the daily lives of people and industries. Emerging blockchain technology has become a viable solution to IoT security due to its inherent characteristics such as distribution, security, immutability, and traceability. However, the integration of IoT and blockchain has challenges such as latency, throughput, scalability, and device power limitation. Recent research has focused on the role of artificial intelligence methods in improving IoT performance in blockchain. According to the studies, there are few effects on improving the performance of IoT devices with limited power, so in this study, a conceptual model for improving blockchain performance in IoT devices with limited power by deep reinforcement learning is proposed. In this model, Internet devices with limited power can delegate their extraction task to the mobile edge computing layer. The presented model has six layers of perception, data, network, consensus, mobile edge computing and application which are explained in detail. In this model, to improve the throughput and select the mining method, a recommender located in the mobile edge computation layer is used. Recommender systems are adjusted by adjusting the size and time of building blocks to improve the throughput and also tries to minimize the delay and energy consumption of the mining operation by selecting suitable method. To achieve good performance in reinforcement learning, the use of Q learning and long-short term memory is suggested. The use of deep reinforcement learning is to set the block size by considering the transmission delay in order to increase throughput as well as mining with respect to the minimum delays, and energy consumption in the proposed conceptual model can improve the performance of blockchain in the IoT.