Modeling And Control Of Direct-fired System With Duplex Inlet&Outlet Ball Pulverizer

A double-input and double-out coal ball mill (DIDO-CBM) is widely used in the direct-fired system. It requires that the output coal is equal to boiler consumption. Thus the reference of the DIDO-CBM depends on subsequent unit load that varies with production demand. However, the DIDO-CBM is a serious time delay process, resulting in poor stability and adaptability of unit and load, respectively. It is very significant to study the control of mill output. Although the control of material level is discussed, the control of mill output is few.So, this dissertation is attempted to control mill output by using model predictive control method. The main works and achievements are shown as follows:(1) A uniform incidence degree method is proposed to obtain the gray sequence of influence factors by using analytic hierarchy process. Because the mill output affected by many factors with different levels, if these are all regarded as input parameters of model, it will cause computing time too long, and is not conducive to implement on-line. If the gray sequence of influence factors was obtained, the important factors can be selected, and the model complexity can be reduced.(2) A dynamic density-weighted LS-SVM method is proposed to establish the on-line prediction model, according to the problem that the mill output is difficult to measure on-line. Firstly, a certain neighborhood radius was determined by the sample size. Secondly, each sample within the neighborhood radius was calculated to get their density index, and then to get the possible contribution index to the model. The data which has the largest contribution degree to model from each selection has formed a sample set. This set is support vector sample set. Finally, based on the selection of the auxiliary variables, the support vector sample set was increased and cut samples dynamically, and the dynamic prediction model of mill output was established. The experimental results that contrasted with dynamic LS-SVM and dynamic weighted LS-SVM model respectively in steady-state condition, load-up condition and abnormal condition show that the dynamic density-weighted LS-SVM model has little error.(3) A model predictive control strategy is presented to control mill output based on dynamic density-weighted least squares support vector machines (LS-SVM) and particle swarm optimization (PSO). Firstly, using the dynamic density-weighted LS-SVM model forecast the output at the current time of mill output on-line. The difference between current output and actual value of mill output was superimposed on the future forecast output value of the model after feedback, and then got the predicted output of next cycle after correction. After comparing the future output and the reference trajectory of mill output, then by the particle swarm optimization algorithm, the signal output of control was realized.(4) A preview reference trajectory is proposed to predict load changing based on unit load model, according to the lag of mill output in the process of load changing. Firstly, analysis and use the historical data to establish the load model. In order to prediction the target value ahead multi-step, and based on the preview control principle, the feed-forward coefficient matrix was got, and the reference trajectory of mill output has been obtained. Tests results show that the adjusting time of this predictive control is short, error is smaller, and system dynamic stability is better.(5) A feed-forward compensation of steam pressure is proposed to control mill output under load mutation. By the input and output energy balance principle of boiler, the function between steam pressure and fuel quantity was deduced. Then, by the function, the difference between pressure actual value and set value of vapor was changed to the order of demand value of mill output, and as the feed-forward of mill output control system. Adopting steam pressure feed-forward will enable the control system to perceive the steam pressure changes, and timely adjust the mill output to sustain load. The simulation results show that this method has faster regulating speed. So it is more conducive to eliminate disturbance.(6) A real-time monitoring and control system is designed for the simulated experiment platform. In this system, the control algorithm designed in this paper was combined with hardware and software, the double CPU structure and multi-thread technology was introduced. The operation efficiency and reliability was improved, and has realized the mill output measurement on-line and control functions.Research in this dissertation have broken the before situation of separate control about pulverizing system. The pulverizing system, unit load and the main steam pressure were combined as a whole. The methods and conclusions in this dissertation can be referenced by the other direct-fired systems and unit load control.