Neural Network Predictive Control Of Power Plant Water Treatment System

To solve the industry control problem of power plant water process, an improvedidentification and predictive control based on neural network model is presented in thispaper. Power plant water treatment system include condensing water and feed waterammonia treatment, feed water hydrazine treatment, boil water phosphate treatment,and power plant industrial waste water treatment. These systems have some obviouscharacteristics for industry control. First, it has large time delay, because it has delayprocess like mixing, reaction, measurement. Secondly, it has strong nonlinear, especiallystrong acid and alkali neutralization process. Thirdly, it is a time-variables anduncertainties process. In addition, change of load, error of measurement, switching ofequipments, which are difficult to control. For these plants, PID control method isdifficult, modern control theory ask too much for model accuracy. So, predictive controlmight be a good choice, because it demands little for model and can achieve real-timecontrol. This paper reviewed the study achievements of predecessors for nonlinearpredictive control and PH neutralization control. According to mechanism, we foundedthe model of PH neutralization process. On these bases, the method of optimization forweights is improved. And the control method based on neural network model andcontroller was applied for control of PH neutralization process, the simulation resultproved the validity of the method. this paper make following contributions.Identification methods based on mixture neural network and Bayesian-Gaussianneural network (BGNN) are proposed. It enhanced the identification precision andcapable of fast tracking via improving the construction of model and optimizationalgorithm. The model simulation experiment is conducted on PH neutralization process.The simulation result proved the effectiveness and accuracy of the new method, whichalso meet the demand of online identification.To solute the problem of PH neutralization process control, an RBFNNidentification method based on SA-DPSO algorithm is proposed applied on model ofstrong acid equivalent of PH neutralization process. It raised the identificationefficiency and model prediction accuracy through identification of delay time. Thecombination of RBFNN identifier and predictive controller realized predictive controlto PH neutralization process of wastewater. The simulation results show that the methodis efficient for PH value control, and it obtained the effect of less error and save medicament.Combined with the power plant feed water ammonia treatment process, the methodwas applied for multiple input multiple output system. This paper reviewed theadvantage of neural network predictive control, and also pointed the disadvantage suchas real-time of algorithm, generalization ability and option of construction, proof ofrobust stability, et al. At last, the paper summed up whole contents.