Abstract: The MPTCP is one of the important strategies currently used for vertical switching in heterogeneous networks. A fluctuations of network throughput during vertical switching in heterogeneous networks can be reduced by a congestion control strategy. In this paper, a congestion control strategy generation mechanism for MPTCP based on reinforcement learning is proposed, and applies it to the heterogeneous network vertical switching of power Internet of Things. The main contributions of this paper are as follows: 1) Reinforcement learning environment based on MPTCP is constructed with taking heterogeneous network time-varying attributes as environment elements, and MPTCP congestion control policy as agent behavior policy. Interactive learning between the agent and heterogeneous network simulation environment is realized through the execution of vertical switching, and the optimal congestion control strategy is generated. 2) Vertical switching of heterogeneous networks based on the improved MPTCP congestion control strategy is realized, and the network switching process of heterogeneous networks is optimized. Simulation and actual test results show that the proposed method avoids data breakpoints in the heterogeneous network switching process of power IoT, and effectively reduces throughput fluctuation during the heterogeneous network switching process.