District Heating Networks (DHNs) are central to sustainable energy transition, but their efficiency depends on effective supply temperature control. Rule-based strategies, such as heating curves, remain common for their simplicity, but tuning them often requires long periods of suboptimal operation. In this paper, we propose an alternative approach that directly optimizes control parameters via gradient descent, based on a differentiable dynamic simulation model. We apply this approach to optimize the parameters of various heating curve formulations and evaluate its performance on a real meshed network using monitoring data. Results show a reduction of heat losses compared to both manual regulation and gradient-free optimization, demonstrating the potential of gradient-based methods for DHN control.