Using LSTM Deep Learning Technology to Adjust PID Control in the Application of Inverted Pendulum

羅宇彤

Abstract

This study aims to design a PID controller that can tune nonlinear systems effectively. The PID controller uses LSTM, a RNN algorithm to enhance its performance, using time series data to predict PID parameters. In conclusion, the LSTM deep learning model can be applied to nonlinear systems to increase its stability and to find suitable PID parameter values for the system promptly.

---

Research Purpose

1. Establish a LSTM deep learning model to obtain PID parameters.
2. Discuss the effect of adjusting the PID within different ranges of K_P values.
3. Discuss the effect of LSTM and Ziegler Nichols on adjusting the inverted pendulum PID under different m and l values.

---

Methodology

---

conclusions and future work

The LSTM PID controller developed in this study can balance nonlinear systems effectively and efficiently. The experiment findings reveal the following:

1. The K_P value must be larger than 1 to converge.
2. When the mass of the pendulum exceeds that of the pendulum cart, the LSTM deep learning model exhibits superior and more pronounced effects.
3. For values of l less than 1, the LSTM model outperforms Ziegler Nichols, with more significant and effective outcomes.

PID controllers in inverted pendulum systems can be applied in robotic arms, simulating the balance of the human during standing, and the optimization of stabilizers in aerospace.

Back