Abstract
Navigation
- Introduction
- Relevance of the topic
- Goals and objectives
- The scientific novelty
- Expected practical results
- A review of studies
- Own research on the topic of master's work
- Conclusion
- References
Introduction
The process of coking is one of the oldest manufacturing processes. Coking technology, operated manufacturing equipment, instrumentation and automation mainly put in place in the early second half of last century and have remained to this time of significant changes. At the same time the process has significant energy reserves using of which in Ukraine is extremely important because coal remains actually the only available energy source. One of the possibilities for reducing energy consumption from the coking process is the introduction tools of monitoring and control based on microprocessor technology and modern management techniques.
Relevance of the topic
Trends of development of modern management systems of chemical-technological processes indicate that they must be adaptive, intelligent. Adaptability of these systems provides the possibility of its configuration: the different types of raw materials, the type of products, hardware and technology design. Experience of managing complex, inertia, potentially dangerous objects of class, which the coking process refers to, shows that the management and staff training necessary to consider not only the requirements for the process run in production mode, but also best practices and knowledge of highly skilled operators, chemists, engineers, scientists who are experts in this subject area. Thus, the development of decision support system to control coking process that includes intelligent features to enable a semantic solution to the formalized problem, digital mathematical model and subsystem of predictions of temperature of coking is actual nowadays.
Goals and objectives
The main goal of master's work is the development of modifications, methods and algorithms for decision support system to improve the management of the process of coking.
To achieve the goal it is necessary to solve the main problems:
- Analyze existing decision support systems to manage the process of coking.
- Develop a mathematical model of the process of coking and algorithms to control of its parameters.
- Test and implement the developed models and algorithms to evaluate the effectiveness of their use.
The object of research – the process of coking.
Subject of research – Methods, models, decision support algorythms in the systems to control the process of coking.
Research methods. There are systematic approach and analysis, methods for representation of knowlendge, methods of prediction are used to achieve the objectives. Methods of mathematical modeling and mathematical statistics, such as factor analysis are using to simulate the process using.
The scientific novelty
First proposed the structure of decision support systems to control the process of coking, which allows to predict the temperature of coking with given parameters. Gained further development of planning model of the process of coking with using the methods of mathematical statistics.
Expected practical results
The developed software system based on a mathematical model will reduce the number of abnormal or emergency situations in the workplace, will assist in making decisions, will justify the choice of solutions in a given situation.
A review of studies
In Ukraine the issue of operational prediction of the temperature conditions and the management of the process of coking was interested by V.N. Tkachenko. He describes the tasks that he had to decide, and gives the results of experimental and calculated values for coking temperatures in conditions of Avdiivka Coke Plant in his paper [1]. Among other things, there is an algorithm of his system to contol the process of coking [1, p.222].
Own research on the topic of master's work
Development of a mathematical model
It was developed a mathematical model describing the temperature dependence of the parameters of the process of coking (1) [2]:
The vector of input parameters
where 
and 
are characteristics of the charge and heating gas accordingly.
The vector of model's parameters and coefficients,
where 
and 
are characteristics of the furnace and charge accordingly.
Consumption of heating gas 
is a model's vector of control action 
.
where:
λ is a thermal conductivity;
К is a heat transfer coefficient of the heating wall;
L is a width of cell;
is an area of the heating wall;
is a gas' coefficient of thermal heating;
is a thermal conductivity of the wall;
is a wall thickness;
is a thermal diffusivity of the charge;
is a heat heating gas;
is a flow of charge;
is the temperature of the charge at the wall;
is an initial temperature of the charge;
is an initial temperature of the heating ga;
is a final temperature of the heating gas;
is a thermal diffusivity of the charge at the wall;
is a specific heat of the charge at the wall.
Development of decision support system
It was decided to develop an active [3] decision support system (DSS) of the entire enterprise [4, 5], which could make a proposal, which solution to choose, and will be connected to large stores of information and serve to many managers of enterprise.
Figure 1 – Example of coking temperature after adjusting the parameters of the coking by decision support system in emergency situations on the basis of the prediction of coking temperature; the red line - the minimum allowable temperature of coking
(animation: 7 shots, 3 cycles of repetition, 51 KB)
There was decided to use the logical model of knowledge representation in the DSS [6]. The main idea of the approach in the construction of such models is that all information needed for applications is regarded as a collection of facts and statements which are presented as formulas in some logic. The knowledge appear as a set of that formulas, and generation of new knowledge is reduced to implement the procedures of logical inference. The basis of logical models of knowledge representation is the notion of formal theory given by a tuple:
where:
- B is a countable set of basic symbols (alphabet);
- F is a set called formulas;
- A is a priori selected subset of true formulas (axioms);
- R is a finite set of relations between the formulas, called rules of inference.
The advantages of logical models:
- As a "foundation" is used the classical apparatus of mathematical logic, methods which are well studied and formally founded.
- There are quite efficient inference procedures using the mechanisms of automatic theorem proving to search and logically meaningful output.
- In the knowledge bases can be stored only the set of axioms, and all other knowledge obtained from them by the rules of inference, as well as data, facts and other information about people, objects, events and processes.
The problem of prediction of temperature of coking
In general, the problem of finding the prediction of the process of coking reduces to finding the temperature at certain points in time after the last measurement. To solve this problem, you can choose one of the methods of time series prediction.
-
Regression methods.
Multiple regression model is generally described by the expression: [7]
(10)
In a simpler version of the linear regression model dependence of the dependent variable takes the form of expression:
(11)
Here
are gleaned regression coefficients, 
is a component of the error. It is assumed that all errors are independent and normally distributed.To construct the regression models we should have a database of observations kind of this :
variables independentdependent # X1 X2 ... XN Y 1 x_11 x_12 ... x_1N Y_1 2 x_21 x_22 ... x_2N Y_2 ... ... ... ... ... ... m x_M1 x_M2 ... x_MN Y_m With a table of values of past observations, we can choose (for example, the method of least squares) regression coefficients, thus setting the model.
-
Learning genetic algorithm LGAP.
LGAP (Learning Genetic Algorithm for Prognosis) based on an idea previously used in the algorithms, and ZET WANGA, which is usually used to fill gaps in the empirical data tables. This algorithm requires a large amount of system resources for its work. It can be easily parallelized and shows good results when working on multiprocessor systems [8], but this solution is very costly.
-
Neural network prediction models. Multilayer perceptrons.
Currently, using of neural networks is the most promising quantitative method of prediction. We can mention a lot of advantages of neural networks over other algorithms. One of them is that the use of neural networks easy to study the dependence of the predicted value of the independent variables [7].
The disadvantage of of neural networks is their indeterminacy. This refers to the fact that after training the neural network there is a "black box", which somehow works, but the logic of the decision-making of the neural network completely hidden from an expert.
The simplest version of the application of artificial neural networks is to use of conventional perceptron with one, two, or (at most) three hidden layers. In this case usually a set of parameters is the inputs of the neural network based on which (according to the expert) network can successfully predict.
Conclusion
In conclusion I would like to note that the process of coking is characterized by its complexity and accuracy in constructing a mathematical model. Any mistakes and large errors can lead to the fact that at the end of the process will be received poor quality coke, which will lead to a loss on the sale. Therefore, to solve the problem of reducing the dimension of feature space was chosen as the most accurate method, factor analysis, to minimize the loss of information [2].
To select a method for solving the problem of prediction of temperature of coking, it is necessary to conduct additional studies of algorithms regression models, LGAP and neural network models. The selection criteria will perform the speed and accuracy of the algorithms, as well as their dependence on the hardware.
- Attention
- Master's work is not complete yet at the time of writing this essay. Estimated date of completion: December 2012. Full text of the paper, as well as materials on the subject may be obtained from the author or his supervisor after that date.
References
-
[Электронный ресурс] – Режим доступа: http://www.iamm.ac.donetsk.ua/upload/iblock/46e/tkachenkomatmodelirovanie.pdf
-
Логическая модель представления знаний.
[Электронный ресурс] – Режим доступа: http://ru.wikipedia.org/wiki/Логическая_модель_представления_знаний.
-
[Электронный ресурс] – Режим доступа: http://www.neuroproject.ru/forecasting_tutorial.php.
-
Большаков А.А. Методы обработки многомерных данных и временных рядов / А.А. Большаков, Р.Н. Каримов. Учебное пособие для вузов – М.: Горячая линия – Телеком, 2007. – 522 с.
-
Кауфман А.А, Харлампович Г.Д. Технология коксохимического производства. К 30/ А.А. Кауфман, Г.Д. Харламвович. Учебное пособие – Екатеринбург: ВУХИН-НКА, 2005. 288 стр.