Reservoir computing (RC) is a powerful computational framework that addresses the need for efficient, low-power, and high-speed processing of time-dependent data. While RC has demonstrated strong signal processing and pattern recognition capabilities, its practical deployment in physical hardware is hindered by a critical challenge: the lack of efficient, scalable parameter optimization methods for real-world implementations. Traditionally, RC optimization has relied on software-based modeling, which limits the adaptability and efficiency of hardware-based systems, particularly in high-speed and energy-efficient computing applications. Herein, an in situ optimization approach was employed to demonstrate an optoelectronic delay-based RC system with digital delayed feedback, enabling direct, real-time tuning of system parameters without reliance on external computational resources. By simultaneously optimizing five parameters, normalized mean squared error (NMSE) values of 0.028, 0.561, and 0.271 are achieved in three benchmark tasks: waveform classification, time series prediction, and speech recognition, outperforming simulation-based optimization with NMSEs 0.054, 0.543, and 0.329, respectively, in two of the three tasks. This method enhances the feasibility of physical reservoir computing by bridging the gap between theoretical models and practical hardware implementation.