The Design And Research Of Soft-sensor Model Of The Carbon Content In Coal Ash

In modern industrial production control process, there are a lot of important process variables, which are often closely related to production efficiency, product quality and business efficiency, need to be strictly controlled. However, for technical, technological or economic reasons, sometimes it is difficult even can not be measured directly on these process variables, even if use some physical or chemical methods to measure, there will still exist hysteresis nature, strong interference, strong coupling and other factors that led to the measured value inaccurate, large delay, poor robustness results. Thus, soft-sensor technique for its strong real-time, strong representation and modeling methods diversification to become one of the best choice to solve this problem, and has been developed for industrial control and process testing field favorite. The core of soft-sensor technology is soft sensor model, its modeling approach are mechanism modeling, data modeling and hybrid modeling.This paper first analyzes the main factors affecting the soft-sensor model. Among them, the auxiliary variable selection and data pre-processing is the basis of the soft-sensor modeling, has a significant impact on the performance and accuracy of the modeling. On the auxiliary variable selection of this paper, it take the application of mechanism-based analysis and gray theory based methods. The data pre-processing has fully considered the data acquisition, data processing and error analysis, then use based on the deviation of the reference value with a test method for data preprocessing.The researchers then studied the modeling method of soft-sensor techniques, such as the mechanism modeling method, the regression modeling method and neural network modeling method. Take application of regression method and BP neural network method to build soft-sensor model of ash carbon content, which variable is important but not easily measured, and compared the model output value and the actual laboratory values, and analyze with scroll optimal correction.The paper also takes these types of modeling methods on horizontal comparison in approach of the advantages and disadvantages of each modeling. Through algorithm analysis comparison, precision analysis and application of environmental summed up the pros and cons between the modeling methods in different circumstances similarities and differences. Add ash carbon content of the soft-sensor model in the campus energy information management system under WinCC operating platform, real-time monitoring of the trend of the ash carbon content, and incorporated into the offline database query function.Finally, use soft-sensor techniques to take the expansion research of the boiler power consumption and coal consumption of a single day, through the establishment of the soft-sensor model, the power consumption and coal consumption on the day is real-time inference estimated that play a guiding role of the operation of the boiler, and also to make some contribution to low-carbon production, energy conservation.