The optimization of hydrocarbon production is vital in the petroleum industry. Slug flow, however, can lead to production stoppages due to damage to surface equipment. As reservoir pressure declines during oil production, slug flow may occur in surface pipelines. Therefore, developing intelligent separators and implementing effective flow regime control methods are crucial for achieving this goal. This study constructs a smart laboratory pilot to collect experimental data, including liquid level, separator pressure, input mass flow rates, and control signals, under Model Predictive Control (MPC). We employ machine learning techniques, specifically Long Short-Term Memory (LSTM), to develop proxy models for a 3D reservoir simulation, significantly reducing computational time. The LSTM proxies are then integrated into a comprehensive production model that includes a horizontal gas-liquid separator equipped with an MPC controller. The controller efficiently regulates the separator's liquid level and operating pressure in real-time. Experimental results demonstrate that the proposed system effectively mitigates slug flow by adjusting separator pressure, maintaining stable operation across various flow regimes. In a 20-year field-scale simulation, the integrated LSTM-MPC system increased cumulative oil production by approximately 40% compared to a non-optimized system. This study presents a novel approach that combines data-driven reservoir modeling with advanced control strategies, offering a significant improvement in production optimization and flow assurance for the petroleum industry.