Neural Network Inverse System Online Decoupling Control Of Air Dense Medium Fluidized Bed

The dry cleaning of coal with ADMFB (Air Dense Medium Fluidized Bed) is an efficient dry separation technology which apply fluidization technology to coal separation. Compared with the traditional wet coal separation method, it has many advantages, has opened a new method for coal separation and has the significant economic value and broad application prospects. The key to dry cleaning of coal with ADMFB is control fluidized bed high and density to make them steady within the scope of the process design. However, due to the strong coupling between the high and density of the fluidized bed in ADMFB, the traditional control method is difficult to achieve good control. Based on the linearize and coupling theory of inverse system and the superiority of neural network applied to nonlinear system control, this paper bring in neural network inverse system method witch combine the advantages of the two methods in order to achieve decoupling control of the system. Moreover, a strategy of adjusting the weight online is proposed based on training the neural network inverse system offline.This paper first introduce the background of the neural network inverse system method and the advantage of problem-solving, analyses neural network inverse system theoretically, introduce the construction method of neural network inverse system in detail. After that, this paper gives the proof of the reversibility of the controlled object based on the mathematical model of ADMFB and the inverse system theory. A static neural network and integrators are used to construct the neural network inverse system of the controlled object, it get network initial parameters through off-line training, and is placed in series with the original system as an original inverse controller. The other neural network with the same structure is used to identify the inverse system of original system on-line, and the parameters obtained from the on-line identification is assigned to the neural network as the inverse controller in order to dynamic adjust network parameters and achieve the system on-line decoupling. Placed neural network inverse system in series with controlled object, the composite system is decoupled into two independent pseudo-linear subsystems:a first-order bed high subsystem and a first-order density subsystem. On this basis, this paper adopts the mature reliable PID controller as additional controllers to achieve high performance control of the system.Finally, simulations are eventually performed for the neural network inverse control system of ADMFB by using Matlab software for the platform. The simulation results indicate that, the neural network inverse method realizes high precision decoupling control of ADMFB, and neural networks inverse controller based on online identification has stronger robustness to variation of parameters than neural networks inverse controller based on off-line identification.