| Superheated steam temperature is a significant monitoring variable of the coal-fired boiler,excessive fluctuation of superheated steam temperature poses a serious problem to the security and economy of unit running.Because the superheated steam temperature system is time-varying and affected by many variables,when coal fired units participate in the fast and deep load-changing running of power grid frequency adjustment and peak adjustment,the wind,coal and water systems which are closely related to the superheated steam temperature are adjusted rapidly,and the regular cascade PID control often causes the superheated steam temperature to deviate from the target value and fluctuate too much.To ensure that the superheated steam temperature does not overstep the limit,the operator often needs to manually adjust the controllers’ steam temperature set value(bias)or cut the water injection valve to manual control,which increases the work intensity.Therefore,it is necessary to design a more effective and practical spray desuperheating control system.Based on the discussion of the theory of superheated steam temperature system,the theory of DMC algorithm and the principles of neural network,the thesis combines neural network and Dynamic Matrix Control applied to superheated steam temperature system and develops the design and simulation research of superheated steam temperature control scheme.(1)A step-forced test is conducted with the support of a600 MW simulator to derive data on the variation of superheater inlet and outlet temperatures at every level with the opening of water injection valves,and a transfer function model is identified by the genetic algorithm,which results in the devise of a DMC-PID cascade control scheme.(2)Given the differences in the characteristics of steam temperature objects under different working conditions,multiple DMC control models are established for various loads,and real-time control quantity calculation is carried out for diverse working conditions.(3)Given the relatively complex model obtaining and the difficulty of adjusting many parameters for multi-model DMC control,two different control strategies are designed by adding neural networks to the DMC-PID cascade control scheme.Matlab real-time online control programs are written for each of the mentioned control schemes,and full-time simulation tests are carried out on a600 MW supercritical unit simulator to verify the control effect of each scheme.The results of the simulation tests show that the four control schemes are better than the original control scheme of the unit,and the fluctuation range of superheated steam temperature is reduced.Compared to the DMC-PID control strategy,superheated steam temperature fluctuations are further reduced and valve opening conditions are better under the multi-model DMC control scheme.Of the two schemes incorporating neural networks,both outperformed the DMC-PID simulation,and the schemes further improve the stability of the superheated steam temperature system.Compared to the combined neural network inverse and DMC-PID cascade control schemes.The DMC-PID cascade control with neural network feedforward compensation results in smaller deviations from the setpoint or better control effect. |
