The Research Of Multivariable PID Neural Network In BBD Ball Mill Control System

The BBD ball mill is one of the key equipments of pulverizing system in power plants, which is widely used in thermal power generating units, its security and economy running state effects the main technologic and economic targets of the process system directly, such as production ability and energy consumption. Thus, the research of the automatic control and the optimal operation is so significant to decrease of the per-consuming and cutting energy cost.The BBD ball mill is such a multivariable coupling system that it has the characters of nonlinearity, slow time-varying and huge delay. There's a strict coupling between the controlled and controlling variables, a complex character of the dynamic system, it is so difficult for us to build an accurate mathematic model, also, it's hard to obtain a satisfied results through the ordinary controlling methods. Thus, in this paper, we make a deep research of the controlling of the BBD ball mill from the points of the intelligence control. The main content of this paper contains the following:(1) Make an introduction of the gross structure and working principle of BBD ball mill and make detailed analysis of the dynamic properties and influencing factors. After that, make a deeply research of the mill's mathematical models, control objectives and requirements.(2) By doing a research of the structure and algorithm based on multivariable PID neural network, I designed a PID neural network control system of BBD ball mill for the property of the milling system.(3) Make a research of particle swarm optimization (PSO) and obtain a deep understanding of its improving method. Then I raised an improved PSO based on the queue theory, not only gave out the improving idea and implement procedures but also the formula experiment analysis. The result showed that the improved PSO had a quicker convergence speed and a better training accuracy than the standard PSO, so it could effectively keep the search away from falling into the local minimum value. (4) Utilize the improved PSO to optimize the initial parameters of the PID neural network controller, and then insert the optimal solution to the network to adjust the network weight online combined with using error backward BP algorithm. The simulation results show that, the multivariable PID neural network controller which is improved by the PSO has a better steady and dynamic capabilities, also with a strong self-learning and self-adaptive abilities, so it can solve the problem of the coupling and time-varying for the milling system commendably.