Approximate LASSO Model Predictive Control for Resource Constrained Systems
Approximate LASSO Model Predictive Control for Resource Constrained Systems
复制标题
资源受限系统的近似LASSO模型预测控制
DOI：
--
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
A. Wallace
中科院分区：
文献类型：
--
作者：
Yun Wu;J. Mota;A. Wallace
套索预测控制是解决滚动域内最优控制问题的一种流行方法。然而，随着视界长度的延长，由于密集的内存使用和计算成本，在资源受限的系统(如嵌入式平台)上部署套索MPC是具有挑战性的。通过利用一种应用于最近梯度下降(PGD)的降低精度的近似技术，我们展示了一种在资源受限的可重构器件上的实现，例如现场可编程门阵列(FPGA)。我们的实验显示了与高精度优化求解器相同的性能，但在逻辑成本和内存带宽方面都有显著改善，分别减少了60%和80%，功耗节省高达70%。
LASSO MPC is a popular method for solving optimal control problems within a receding horizon. It is, however, challenging to deploy LASSO MPC on resource constrained systems, such as embedded platforms, due to the intensive memory usage and computational cost as the horizon length is extended. By exploiting a reduced precision, approximation technique applied to Proximal Gradient Descent (PGD), we demonstrate an implementation on a resource constrained, reconfigurable device, such as a Field Programmable Gate Array (FPGA). Our experiments show equivalent performance to a high-precision optimisation solver, but with significant improvements to both logic cost and memory bandwidth, up to 60% and 80% reduction respectively, with up to 70% power savings.