This study presents an integrated framework for thermodynamic–economic modeling and dynamic operational optimization of a combined cooling, heating, and power system based on a Kalina cycle and an absorption chiller. The operational problem was formulated over a 24-hour horizon with hourly time steps, and the deep reinforcement learning Soft Actor–Critic (SAC) algorithm was employed for continuous decision-making. Three control policies, namely profit-oriented, balanced, and efficiency-oriented policies, were developed, and their performance, based on five independent runs, was compared with a linear programming-based reference policy. The results showed that the reference policy achieved a daily operating profit of €419.03 and an energy performance index of 0.07724. The mean daily profits of the profit-oriented, balanced, and efficiency-oriented policies were €538.318, €547.701, and €589.236, respectively, while their corresponding energy performance indices were 0.08036, 0.08014, and 0.08099. These results correspond to profit improvements of 28.5%, 30.7%, and 40.6% and energy performance index improvements of 4.0%, 3.8%, and 4.8%, respectively, compared with the reference policy. The efficiency-oriented policy achieved the highest mean values for both performance indicators, although the dispersion across independent runs demonstrated the sensitivity of performance to the stochastic nature of the training process. Moreover, an inference time of less than one millisecond indicated the computational potential of the trained policy for online implementation. Sensitivity analysis further demonstrated the direct effect of electricity price variations on economic performance and the limited sensitivity of the evaluated reinforcement learning policy outputs to demand variations within the investigated range.