The operation performance of a solar heating system is largely influenced by the coupling of outdoor climate conditions and building energy operation patterns, and is characterized by strong randomness and large heating fluctuations, which makes the optimal control rules for solar heating systems in dynamic state. Although the traditional RBC(rule-based control), whose rules is preset, has a fast response speed, the fixed-rules are hardly to make the solar heating system in optimal state. Furthermore, solar heating systems is generally characterized by a large time delay in the heat transfer process. The control point lag occurs in the control signal of the dynamic control strategy. To address these issues, this paper proposes a model predictive control-based (MPC) control strategy for solar heating systems. The method uses seq2seq-LSTM to predict major operating parameters including ambient temperature, solar radiation, the load inside the building for the next 24 h and combines the heating system operation model to simulate and obtain the control signal at the next moment. The results showed that, compared with traditional control strategies including RBC and PID control strategies, the efficiency of solar collector regulated by MPC increased by 9%. Taken the tank temperature at the simulation end as the energy calculation reference point, the solar heating system’s total energy consumption employing MPC is 34.2% lower than that employing traditional control strategy.