基于LSTM的单级倒立摆NMPC优化控制算法研究

Research on LSTM-Based NMPC Optimization Control Algorithm for Single Inverted Pendulum

摘要: 本研究针对单级倒立摆系统的非线性控制难题，创新性地将长短期记忆网络(LSTM)与非线性模型预测控制(NMPC)相结合，提出了一种新型优化控制算法。针对传统NMPC方法存在的模型依赖性强、计算复杂度高等问题，本研究采用LSTM构建数据驱动的预测模型，有效提升了系统的控制性能。仿真对比结果表明本文所提出的LSTM-NMPC算法较传统NMPC方法具有显著优势，系统稳定时间缩短50%从7秒降至3.5秒，角度超调量由0.691弧度降低至0.527弧度改善幅度达23.7%，角速度超调量减少33.3%从12.517弧度每秒降至8.344弧度每秒。这些改进充分验证了该算法在响应速度与稳定性方面的优越性，不仅为欠驱动系统智能控制提供了新思路，其方法还可推广应用于复杂非线性工业系统的控制领域。

Abstract: This study addresses the nonlinear control challenges of single inverted pendulum systems by innovatively integrating Long Short-Term Memory (LSTM) networks with Nonlinear Model Predictive Control (NMPC), proposing a novel optimized control algorithm. To overcome the limitations of conventional NMPC methods, including strong model dependency and high computational complexity, this research employs LSTM to construct a data-driven predictive model, significantly enhancing the system’s control performance. Comparative simulation results demonstrate that the proposed LSTM-NMPC algorithm exhibits remarkable advantages over traditional NMPC methods: the stabilization time is reduced by 50% from 7 seconds to 3.5 seconds, the angle overshoot is decreased from 0.691 radians to 0.527 radians (representing a 23.7% improvement), and the angular velocity overshoot is reduced by 33.3% from 12.517 rad/s to 8.344 rad/s. These enhancements fully validate the algorithm’s superior performance in both response speed and stability, providing new insights for intelligent control of underactuated systems while demonstrating potential for extension to complex nonlinear industrial control applications.