The Research Of Two-wheeled Self-balanced Vehicle Based On PID Neural Network

The two-wheel self-balanced vehicle is a new kind of scooter. The novel driving experience attracts the young people, so does the electric drive mode and cabinet shape. It is a good choice for mitigating the increasingly serious air pollution, energy crisis and traffic congestion, which make the two-wheel self-balanced vehicle valuable. However, the structure of the two-wheel self-balanced vehicle is of low stability, challenging the control theory. In other words, the two-wheel self-balanced vehicle is a good platform for testing control algorithms. So far, PID control algorithm with simple control principle and using none accurate system model, becomes the most popular control algorithm for the two-wheel self-balanced vehicle, which shows a good performance. However, the controller parameters are usually needed to adjust artificially. Hence, it is not that much easy to get the best performance in current condition. With the running of the scooter, the practical model gradually diverges from theory model which is not so accurate essentially. What’s more, the running environment of the two-wheel self-balanced vehicle is uncertain. All that make precise control in a long time very difficult. This essay attempts to make full use of neural network’s self-learning ability to optimize the parameters of PID controller in real time, improving the performance of the controller and optimizing the vehicle’s balance performance.Firstly, this essay establishes the model of a two-wheel scooter using Newton method, and analyzes its state variables. Based on this model, a gyroscope accelerometer MPU6050is selected to detect the gesture information of the vehicle body, and the wheel encoders are used to detect the condition of the wheels. Based on the state variables, a high-performance AVR micro-controller ATmegal6executes the control algorithm, which is programmed to process the gesture information. Based on gesture signals, the micro-controller output PWM which driver the motors. So far, a bottom system based on the conventional PID control algorithm is established for a two-wheel self-balanced vehicle.Secondly, neural network identification codes NNI and neural network control codes NNC are programmed in C language on chip STM32F103ZET6. At first, NNI is used to identify the model of the two-wheel self-balanced vehicle system. Then, the trained NNC weights, that is, the optimized parameters are sent to the bottom controller to improve the balance performance of the two-wheel vehicle. Finally, a series of tests examine the balance performance and anti-interference capability of the two-wheel self-balanced vehicle. The results show that the balance performance and anti-interference capability have been significantly improved by neural network optimization algorithm.The essay prove that the balance performance and anti-interference capability of the two-wheel self-balanced vehicle system can be improved by optimizing the PID controller parameters using neural network.