<p>Excessive workloads have a negative impact on healthcare quality. However, hospitalizations exhibit a varying distribution throughout the week, with a peak on Mondays and a decrease on weekends. This imbalance creates disparities for patients to have the same medical services level during peak times compared to non-peak times. So, we propose a dynamic programming model to minimize variance in daily service amounts by optimizing the number of admissions to smooth hospitalization demand and prevent overloads. Furthermore, we limit the maximum waiting size and service capacity to ensure system efficiency. However, the exponential growth of the state space with varying lengths of patient stays makes it challenging for traditional dynamic programming solutions. Therefore, we introduce a multi-agent constrained QMIX (C-QMIX) reinforcement learning algorithm to deal with complex states and get a stable solution. Finally, 81 weeks of data from a tertiary hospital in western China are used to test the algorithm. The results indicate a maximum reduction of 14% in variance while maintaining reasonable levels of average waiting time and overall service quantity and reducing the excessive occupancy rate to mitigate medical risks and enhance healthcare quality.</p>