【作者】 陈勇；

【导师】 孔峰；

【作者基本信息】 广西大学 ， 控制理论与控制工程， 2004， 硕士

【摘要】 遗传算法(GA)和误差反向传播算法的多层前馈网络(BPNN)都具有很强的寻优能力。遗传算法具有全局寻优能力。神经网络在局部寻优效果较好，但容易陷入局部极点。于是本文提出两者相结合来对PID调节器参数寻优。使得PID调节器稳定性、响应速度以及精度都得到提高。 由于GA和BPNN的寻优时收敛速度慢，使得基于GA-BPNN的PID调节器在实际控制系统中的运用受到了限制。为了克服这个局限性，根据无刷直流电机在调速系统中的特点，本文提出一种新型的控制策略：“离线粗调”和“在线细调”相结合。让大部分工作放到离线状态来做，使得已训练好BPNN-PID调节器来在线对无刷直流电动机调速。由于BPNN具有自适应能力和记忆能力，使得该控制系统具有自适应能力和响应速度快的特点。 对于智能算法：GA和BPNN分别寻优PID参数，采用GA优化BPNN的权值，并进行了仿真试验。对于新型控制策略也进行了仿真验证。并在北京闻亭科技发展有限责任公司(WINTECH)所开发的TMS320F240评估板上进行了部分硬件仿真，都得到了很好的控制效果。表明了这种智能算法、新型控制策略是可行的、可靠的。更多还原

【Abstract】 Genetic algorithm (GA) and Back Propagation neural network (BPNN) have a strong capacity of optimizing parameters. GA is able to optimize parameters in wide range, BPNN have good converge effect in part, but it is prone to run into apices in part, so, In this paper, combining them to optimize the parameters of PID Adjustor. In order to make PID Adjustor become better stability, more quick rate and higher precision.Because the convergent rate of GA and BPNN is very slow, PID Adjustor based on GA-BPNN can’t be widely used in real-time control system. In order to conquer the limitation, in the paper, the late-model control strategy is presented, which combines Off-line Roughly Adjust and On-line Precisely Adjust. In real-time adjust speed system, The Brush -less Direct Current Motor (BLDCM) is relatively fastness, so the mostly work is done in Off-line state, and PID Adjustor Based on trained BPNN is used to real-time control BLDCM. Because BPNN have capabilities of adaptive and memory, the control system have characters of adaptive and fast respond.For the intelligent algorithm: GA and BPNN optimize parameters of PID Adjustor respectively, which is simulated and validated. And in the paper, it use GA to optimize the initial values of the power of BPNN. The late-model control strategy is also simulated and validated. Based on TMS320F240 EVM board tapped by WINTECH Co. The intelligent algorithm and the late-model control strategy is partly hardware simulation. These simulations get very good control effects, which shows that the intelligent algorithm and the late-model control strategy are feasible and reliable.更多还原