You are on page 1of 5

Proceedings of the 6th World Congress on Intelligent Control and Automation, June 21 - 23, 2006, Dalian, China

Case-Based Real-Time Controller and its Application in Combustion Control of Hot Blast Stoves
Sun Jinsheng, Yang Jinchuan and Meng Ru
Department of Electronic & Information Engineering Hebei Polytechnic University Tangshan, Hebei 063009, China sunjinsheng@heut.edu.cn
Abstract - For the last few years, AI offers with Cased-Based Reasoning (CBR) a powerful methodology, which has already proved its potentials in different industrial applications. However the related applications up to now are mainly focused on non real-time targets. In our work, a novel CBR based real time controller was proposed and named CBRTC (Case-Based RealTime Controller). CBRTC may be looked as a kind of direct expert controller, but it applies case-based rather than rule-based reasoning methodology to make control decisions. CBRTC has been put into operation for combustion control of hot blast stoves in some in-situ applications. It is shown with the applications that CBRTC may be a powerful tool for dealing with the control problems of complicated industrial processes. Index Terms CBRTC, Case-based reasoning, Combustion control, Hot blast stove

Yin Jingtao
Department of Mechanical & Electrical Engineering Hebei Vocational Institute of Energy Technology Tangshan, Hebei 063000, China llp728@sina.com stoves. Section IV describes the procedures of applying CBRTC to control hot blast stoves combustion process. Section V ends with the conclusions. II. CASE-BASED REAL-TIME CONTROLLER For the last few years, AI offers with Cased-Based Reasoning (CBR) a powerful methodology, which has already proved its potentials in different industrial applications. Casebased reasoning has been used to solve problems in diverse areas including decision support, help desk support, product cataloguing and maintenance support, etc. Case-based reasoning is a method that compares the present problem with previous ones and applies the problem solving of the past to the present problem. In other words, we can reuse the problem solving technique that was used in the past and apply it to the present problem [6, 7]. A typical CBR cycle is shown in Fig.1.
Problem New Case RETRIVE Retrieved Cases REUSE Previous Cases General Knowledge New case, Solution and Result Experts Knowledge Revised knowledge RETAIN REVISE Confiirmed Solution Suggested Solution

I. INTRODUCTION A blast furnace is used to produce molten pig iron from iron oxides, coke and flux. One of the major sources of energy for this process is the sensible heat coming from the preheated air, referred to as blast air, which is injected into the furnace. This air is preheated in tall, cylindrical, refractory-filled thermal regenerators called blast stoves. To ensure efficient furnace operation, the required flow-rate and temperature of the blast air must be maintained. Unfortunately, this goal is not easy to be touched, since the blast stoves have inherently time-delay, time-varying and non-linear characteristics. Traditional control strategies such as PID are no longer in force while confronting the combustion control problem of hot blast stoves, because there are no exact mathematical models for describing the stoves characteristics. Therefore some advanced control strategies have been researched and implemented for solving this combustion control problem, such as fuzzy control or expert control methods reported in literatures [3, 4]. But the natures of these intelligent control strategies are of rule-based, consequently the obstacle or so called bottle neck for gaining the expert knowledge, which is the basis for realizing the above mentioned strategies, is inevitable eventually. So a novel control strategy was proposed [5], which utilizes CBR rather than RBR as its reasoning machine for getting control decisions, and this method has been proved to be effective and easy to be implemented in several in-situ applications. The article is structured as follows. Section I is an introduction. Section II presents the concept of CBRTC and its architecture. Section III is a brief description of hot blast

Fig.1 CBR cycle

A CBRTC (Case-Based Real-Time Controller) is defined as a real-time controller which deploys case-based reasoning methodology for making its control decisions. The CBRTCs architecture is very similar to a direct expert controller. The main difference between them only lies in the correspondent reasoning method. A CBRTCs control decisions are made on the basis of case-based reasoning, yet an expert controllers decisions are reasoned out based on rule-based reasoning. The architecture of a control system using CBRTC is shown in Fig.2. The characteristic set obtained after the character identifying and information processing based on the measured process parameters is called a new query (or problem description). This new query then is submitted to the casebased reasoning algorithm, which is a basic part of the CBRTC. As is illustrated in the control systems schema (see Fig.2), the control decisions made by CBRTC have two possible results:

1-4244-0332-4/06/$20.00 2006 IEEE

7792

Fig.2 Control systems architecture with CBRTC

(i) If the case-based reasoning algorithm finds a stored case similar to this new query from the sample case-base, then the correspondent solution description of the similar case, i.e. the control decision in other words, is reused referring to the effect description and output to the actuator. (ii) If the case-based reasoning algorithm finds no similar case from the case-base, the learning and revising functions will be executed. They generally give out a trial solution for this new problem, which is a smallest adjustment unit of the actuator. After this trial control decision has been output to the actuator, its adjusting result will be measured and evaluated, therefore a new sample case with all the three basic elements, <problem, solution, effect>, will be formed and be added in the sample case-base. As for the similarity determination between a new query and each sample case, it is calculated in terms of their weighted Euclidean distances described as follows:

d ( X ,Y ) =

a (X Y )
i i i i =1

(1)

In our work, a threshold value dmin is pre-determined before the on-line retrieving procedure is doing. If one or more cases with their respective distances less than dmin are found, then the sample case which has the smallest distance is taken as the retrieved and reused sample. A CBRTCs implementation is typically divided into two phases: Data Preparation and On-Line Operation. The tasks need to do in the data preparation phase include the followings: (i) Define the sample cases representation, in other words, determine which input, disturb and output variables should be included in a sample case. These variables are related with the control decisions made by operators, and should be chosen by referring to operators experiences and/or analysing the historical data. (ii) Determine both the case-bases structure and retrieving method, and draw out some adjusting samples from the recorded historical data. These sample cases will be taken as the initial seed of the case-base while a CBRTC beginning its on-line operation. The tasks of a CBRTC need to do in its on-line operation phase are just like a typical CBR cycle, which can be divided into the following steps: (i) Getting data and submitting problem. A CBRTC gets the real-time data reflecting the controlled objects operation status and technique parameters. If some problems are found from the on-line data, such as a bias between a required temperature and the measured one is greater than a predetermined threshold, they will be submitted to the CBRTCs

reasoning machine in the form of a problem description vector to look for a suitable solution. (ii) Retrieving and matching. A CBRTCs reasoning machine compares the submitted problem with each sample case stored in the case-base to find a similar case. The similar case is defined as a sample case which has the smallest space distance between it and the submitted problem vector. And the space distance is calculated using Equation (1). (iii) Adopting and reusing. If the similar cases distance calculated with Equation (1) is less than a pre-determined threshold, this cases solution description will be adopted and reused, and taken directly as the reasoning result or the control decision of the CBRTC. (iv) Learning and revising. If no sample cases distance in the case-base is less than a pre-determined threshold, a CBRTC will take a trial adjustment as its control decision. This trial adjustment is made based on some known control rules. And this new problem and new solution will be evaluated and added to the sample case-base to further extend the CBRTCs problem solving ability. Although some industrial processes could not be directly regulated using mathematical model based control methods because of their uncertainty and complexity, experienced operators can manipulate such processes contentedly simply by remembering and referring to their accumulated practices. Such successful manipulating practices are the solid basis for constructing the experience-based or knowledge-based CBRTCs. And the more the experiences are accumulated in the controllers sample case-base, just like more knowledge stored in a persons brain, the more effectively the CBRTC will function while dealing with different and various technique conditions and variations. III. PROCESS DESCRIPTION OF HOT BLAST STOVES Iron manufacturing in blast furnace requires a large flow of air preheated at high temperature (above 1000). A battery of 3~4 regenerative heat exchangers (stoves), operated cyclically, is located in the vicinity of the furnace. A stove is a tall, cylindrical vessel filled with refractory (checkers) used to accumulate and later release energy. The brick geometry and its properties usually vary in sections, as the upper section operates at much higher temperature than the lower section. During the first part of a working cycle (on-gas), the checkers are heated by combustion of blast furnace gas (BFG), possibly enriched with coke oven gas (COG), as shown in Fig.3(a). When the temperature is high enough, compressed air (blast) circulates through the stove in the opposite way (see Fig. 3(b)). During this on-blast operation, a by-pass valve controls the preheat temperature. The hot blast is fed to the furnace usually until a determined duration is passed. At that moment, the next stove is put into blast operation and a new cycle is repeated (approximately every 180 minutes). Typically, a small or medium sized stove is on-gas during 115 minutes, and on-blast during 55 minutes. Transition between operation phases requires several minutes, needed to switch valves, pressurize or blow-down the stove.

7793

Dome

Combustion chamber Blast air Combustion chamber Combustion air Gas Cold air

(a) on-gas phase (b) on-blast phase Fig. 3 Two working phases of a hot blast stove

Our study comes from the No.2 blast furnace at Xingtai Steel Company in Heibei, China. The No.2 blast furnace runs a three-stove system operation. Two stoves will be in on-gas cycle operation and one stove will be in on-blast cycle operation at any given time. Because a hot blast stoves heat level could not be directly and simply measured, the automatic control system usually takes two indirect parameters as its controlled variables, which are the stoves dome temperature and its waste air temperature. The adjusted variables are the air flow rate and the fuel (BFG) flow rate, and a suitable ratio of air/fuel should be maintained to ensure an optimal and energysaving combustion status. Ideal curves of the dome and waste air temperatures are shown in Fig.4. A total on-gas phase could be divided into 2 stages: the fast heating stage and the heat storing stage. Consequently, the control missions are distinguished in terms of the different stages:
Dome temperature

increasing rate of the waste air temperature should be guaranteed at the same time. Therefore three CBRTCs are constructed to carry out the above mentioned three different control tasks respectively: a CBRTC for the dome temperatures increasing rate control during the fast heating stage, a CBRTC for the stable value control of the dome temperature and a CBRTC for the waste air temperatures increasing rate control during the heat storing stage. The control variables determined by the CBRTCs are the gas flow rate and the air flow rate. Since the main difference among these three CBRTCs lies merely in their case representations, the CBRTC for the stable value control of the dome temperature is taken as the explanation example for this control method. A. Case Representation of CBRTC In a CBRTCs case-base, each sample case is described as a vector form which includes three elementary components, that is: a problem description, a solution description and an effect description. Before a CBRTC is put into on-line operation, its case representation should be carefully researched and constructed. For controlling a stoves dome temperature during its heat storing stage, in addition to the directly related adjusting variables (like its gas flow rate and air flow rate), other relevant parameters should also be taken into consideration. In our work, the case-bases case representation is constructed based on the following considerations: (i) The suggestions of experienced operators and the analysis of historical data should be taken into account first of all. (ii) Secondly, the selected variables or attributes should come out of easy to be measured or calculated process parameters. Finally, the case representation of the CBRTC for stable dome temperature control is determinedand described as in Table 1.
Table 1 Case representation
No. 1 2 3 4 5 6 7 8 9 10 11 Name On-gas duration Dome temperature Change of dome temperature Waste gas temperature Change of waste air temperature Pressure of BFG Flow rate of gas Flow rate of air Ration of gas/air Position of gas valve Position of air valve Symbol Description On-gas Duration from the beginning to the current time Current dome temperature Difference between the current dome temperature and the precedent one Current waste air temperature Difference between the current waste air temperature and the precedent one Current pressure of BFG (gas) Current flow rate of BFG Current flow rate combustion air Current ratio of BFG/air Precedent position of gas valve Precedent position of air valve of

Waste air temperature

Changeover Fast t1 heating Heat storing t

t0

t2 On-gas

t3 On-blast

t4

Fig.4 Ideal curves of the dome and waste air temperatures

(i) During the fast heating stage, increasing the dome temperature to its stable value as quick as possible, but without regard to the waste air temperatures change at the same time. (ii) During the heat storing stage, maintaining the dome temperature around its stable value, and keeping a suitable air/fuel ratio to save energy consumption. After the precedent stove ending its on-gas phase, adjusting the current on-gas stoves fuel feed to make it have an adequate raising slope of the waste air temperature for reaching the waste temperatures upper limit at the combustion terminal. IV. APPLYING CBRTC TO COMBUSTION CONTROL As explained in the above section, combustion control of hot blast stoves can be transferred to two temperature control problems: dome temperature control and waste air temperature control. During the fast heating stage, a definite increasing rate of the dome temperature is the control target. During the heat storing stage, a stable dome temperature value and a definite

tON-GAS
TDOME TDOME TEX-AIR TEX-AIR PBFG FBFG FAIR RGAS-AIR VGAS VAIR

7794

12 13 14

Increment of gas valve Increment of air valve Adjust effect

VGAS VAIR Eadj

Increment of gas valve of this adjustment Increment of air valve of this adjustment : normal, :better

B.

Drawing out Initial Seed

After a CBRTCs case representation is determined and checked, it should be filled up originally with sufficient sample cases before being put into on-line operation. These sample cases are called the initial seed [8] of the CBRTCs case-base. The initial seed can be drawn out from the historical data accumulated in the monitoring system of the blast stoves. After analysing several experienced operators manual manipulating habits, we catalogue their operating actions into three different types, they are: time-delay adjustment, progressive adjustment and go-back adjustment. These three adjustments are depicted with Fig. 5.
Dom e t em per at ur e

Ti m e- del ay adj ust m ent Pr ogr essi ve adj ust m ent Go- back adj ust m ent

t1 t2

t3

t4

Fig.5 Three types of manual adjustments

Let the process variables at the sampling time ti (items 1 to 11 in Table 1) be a set Pi, and the adjusting variables (items 12 and 13 in Table 1) be a set Si, then a pair set composed of both Pi and Si plus its effect become a sample case in the CBRTCs case-base, Ci. As regards the above mentioned three adjustment types, the calculation methods for drawing out the initial seed are described with equations (2~4) respectively, where Equation (2) is for time-delay adjustment, Equation (3) for progressive adjustment and Equation (4) for go-back adjustment. (2) Ci = S i (t4) P i (t1-T) Where is a reducing coefficient to the adjustment, 0<<1. (3) Ci = S i (t3) P i (t1-T) (4) Ci = S i (t2) P i (t1-T) It should be noted that a reducing coefficient is attached in Equation (2). The reason is that while an operator is aware of his adjustment being too late, he often subconsciously strengthens the adjusting action. Therefore if an operator is replaced with a computer, such kind of time-delay adjustment will not occur once more, so a weaker adjusting action is proper while drawing out the initial seed from such kind of manual operations. In our work, about 200 initial sample cases were drawn out from the historical operation data of 10 days. These initial samples cover nearly all the possible operation conditions, including various disturbances like pressure fluctuations of BFG, changeover of other stoves, etc. C. On-Line Operations

Differed from the non real-time applications, a CBRTC should do its retrieving, reusing and revising as fast as possible because of its real-time requirement. Thus some complicated similarity calculation methods are unsuitable to such an application. Therefore the simple k-nearest neighbour algorithm is used to realize searching and matching of the similar cases (see Equation 1). The essential idea of k-nearest neighbour is defining a certain space distance between a query case to be matched and each example case stored in the casebase, and taking the distance as a definite similarity degree measure. On-line operations of a CBRTC can be explained as follows: (i) At each sampling time, measure and calculate every process variables listed in Table 1, put them into a vector form: (tON-GAS, TDOME, TDOME, TEX-AIR, TEX-AIR, PBFG, FBFG, FAIR, RGAS-AIR, VGAS, VAIR). (ii) Compare the current dome temperature with its required stable value, if the difference is bigger than a threshold , then the CBRTC will be activated. (iii) Take the current process variables as a new query, and submit it to the reasoning algorithm to seek an answer. (iv) The CBRTC will check its memories (its case-base) to see whether a similar case is exist. (v) If such a similar case is there, then take this cases solution description as its answer to the submitted query: (VGAS, VAIR). (vi) If no similar case is found, the CBRTC will give a trial but definite response, which is a minimum adjusting unit (2%) of the air valve. This trial adjustments effect will be evaluated later.
Table 2 Adjustment Example of CBRTC
No. 1 2 3 4 5 6 7 8 9 10 TDOME 1284.5 1284.2 1283.8 1283.0 1283.0 1283.7 1282.8 1283.2 1283.9 1284.8 TEX-AIR 252.1 252.7 253.2 253.1 252.9 253.3 253.9 255.3 255.5 255.7 FBFG 20904 21100 21372 21538 20276 20647 21130 20964 20939 20198 FAIR 10191 9971 9858 9940 10345 9913 9920 9851 9910 9991 PBFG 5.06 5.13 5.20 5.21 5.01 5.05 5.08 5.09 5.08 5.09 VGAS 51 51 51 51 50 49 48 48 48 48 VAIR 60 60 60 60 60 60 60 60 60 60

An adjustment example of the CBRTC is given out in Table 2. At the 4th sampling time, the dome temperature is 2 lower than the required value (1285) and the difference is bigger than (1.5), so the CBRTC is activated. This new problem is described as a vector: (22, 1283.0, -2.0, 253.1, 1.0, 21538, 9940, 2.17, 5.21, 51, 60). After the space distance calculations with each sample case in the case-base, a similar case is found, which is: C39 = (22, 1283.1, -1.9, 265.9, 1.6, 21683, 9824, 2.21, 5.27, 52, 58, -3, 0). Therefore this cases solution description is reused and put into the actuators as the control decision, which is the gas valves increment (-3, 0). V. CONCLUSIONS Compared with other intelligent control strategies, like fuzzy control or expert control, the presented case-based one has an obvious advantage: it is very easy to be implemented.

7795

The obstacle or bottle neck for obtaining the control rules is entirely overcome. And more importantly, since the control decisions in manual operations were made by the operators with taking into account several factors affecting the control results, and the decisions were fuzzy in some degree, thus the CBRTCs control decisions based on the sample cases drawn out from experienced practices would inherit naturally the characteristics of both fuzzy and multi-factors. It could be said that the CBRTC may be a powerful tool for dealing with the control problems of complicated industrial processes. This view point has been proved with several insitu applications like combustion control of hot blast stoves. REFERENCES
[1] Kenneth R. Muske, James W. Howse, Glen A. Hansen et al., Advanced Control of Operations in the Blast Furnace Project, LANL Technical Report: LA-UR-99-5051, 1999. [2] Kenneth R. Muske, James W. Howse, Glen A. Hanson et al., Temperature Profile Estimation for a Thermal Regenerator, Proceedings of the 38th Conference on Decision and Control, Phoenix, USA, 1998, pp.3944-3949. [3] Huang Zhaojun, Bai Fengshuang, Zhuang Bin et al., Flow Set and Control Expert System of Hot Stoves for Blast Furnace, Metallurgical Automation, no. 5, 2002, pp. 38-40. [4] Ma Zhuwu, Lou Shengqiang, Li Gang et al., The Intelligence Burning Control of the Hot Stove of Lianyuan Iron & Steel Group Co., Metallurgical Automation, no. 4, 2002, pp. 11-15. [5] Sun Jinsheng, Research on a Case-Based Reasoning AI Control Strategy for Blast Furnace Stove Combustion, Doctor Thesis, China University of Mining and Technology, 2005. [6] R. Bergmann, S. Breen, M. Goker et al., Developing Industrial Case-Based Reasoning Applications, The INRECA-Methodology, LNAI 1612, Springer, 1999. [7] Aamodt A., Plaza E., Cased-Based Reasoning: Foundational Issues, Methodological Variations and System Approaches, AI communications, 7(1), 39-59. [8] Joaquim Comas Matas, Development, Implementation and Evaluation of an Activated Sludge Supervisory System for the Granollers WWTP, Doctor Thesis, University of Girona, 2000.

7796

You might also like