A novel heliostat aiming optimization framework via differentiable Monte Carlo ray tracing for solar power tower systems

Optimization of heliostat aiming points is crucial for ensuring the safe and efficient operation of Solar Power Tower (SPT) systems. However, traditional optimization algorithms typically encounter challenges, such as simplified simulation, high computational costs, and low optimization efficiency. To address these issues, an efficient and accurate gradient-based optimization framework, DiffMCRT, is proposed. This framework adopts a novel differentiable full-path Monte Carlo ray tracing method that can accurately simulate the Radiative Flux Density Distribution (RFDD) and calculate the gradients of the simulation process. The gradients are then backpropagated to optimize heliostat aiming points in a continuous solution space. The framework is highly parallelized and implemented in Taichi with automatic differentiation on GPU, enabling accurate and efficient optimization for large heliostat fields. Validation against measured data demonstrates a root mean square error of RFDD of approximately 0.1 W/m2, confirming the accuracy of DiffMCRT. Furthermore, experiments on the commercial Gemasolar field show that the framework can optimize heliostat aiming points about 5 s, with a loss of less than 1 % compared to desired flux distributions, outperforming the heuristic algorithm by two orders of magnitude in efficiency. Moreover, DiffMCRT successfully handles both steady-state and transient conditions, reproducing the desired flux distribution while protecting the receiver. This framework not only advances real-time optimization in SPT systems but also holds promise for addressing more complex inverse problems in the future.