An HMM Based Kiln Coal Feeding Trend Prediction Method

Rotary kiln belongs to large thermal equipment, which is important to the production process of basic industries. Such as cement, metallurgy and steel, etc.lt involves a lot of operating parameters and has complicated structure. In rotary control system, the kiln is a multivariate, nonlinear and strong coupling controlled object.Coal instead of fuel now is widely used in rotary kiln sintering department in China. The kiln’s conditions, skin structure, pulp ingredients and the coal’s quality have great influence to the coal feeding operation during the production process. Therefor, in some complex conditions, it is difficult to determine whether increase or decrease the amount of feed coal even for the experienced workers. Nowadays, remote monitoring system, which is controlled by computer, has been widely used in kiln’s control.Massive of time series, which contains a lot of key knowledge and relevant artificially control experience,has been obtained.But it has not been sufficiently used. HMM-based (Hidden Markov Model-based) time series analysis method has been widely researched and developed in speech recognition, computer intrusion detection, etc.But it is less involved in industrial field. In this paper HMM (Hidden Markov Model) is used to analyze the time series of industry field.Extract the trend features of the thermal data, and statistic the relationship between the time series’s change and the feed coal direction. A kind of kiln feed coal quantity trend forecast method based on hybrid hidden markov model is proposed in this paper, the concrete content as follows.(1) In view of the characteristic of multi-variable,the principal component analysis method was used to do attributes redcution and obtain the main factors. Then curve fitting and slopes distribution methods were used to extract the trend feature. According to the feed coal quantity, divide the existing database into rising database and dropping database.(2) According to the differences between CHMM (Continuous Hidden Markov Model) and DHMM (Discrete Hidden Markov Model), the CHMM model and the DHMM model were modeled in this paper to do comparison. Train them respectively, then the dropping HMM evaluation model and the rising HMM evaluation model can be obtained.(3) The traditional training algorithm of HMM is easily fall into the local optimum and it depends on the initial parameters greatly. Simulated annealing algorithm was used to do improvement in the paper.(4) As to the prediction (recognition). By calculating the likelihood possibility of the evaluate sequence to recognize the trendy of the coal feeding, then used it as a knowledge of expert system.The experimental results show that the HMM-based coal feeding trend prediction is feasiblely judging the operation of coal feed operation and it can provide reliable guidace to the artificial control and expert control of the kiln under complex conditions.