SOLVING PROBLEMS OF CLASSIFICATION, FORECASTING AND MANAGING DESIGN OF CONSTRUCTION PROJECTS

SOLVING PROBLEMS OF CLASSIFICATION, FORECASTING AND MANAGING DESIGN OF CONSTRUCTION PROJECTS Akhtam Sharafutdinovich Daliyev,Candidate of economic sciences, Associate Professor,Republic of Uzbekistan The process of designing construction objects, both when considering its stages separately and as a whole, belongs to the class of complex multiparametric objects from the mathematical modeling perspective.When solving such problems, it is almost impossible to build one single model that sufficiently describes the process under consideration. Therefore, several models are built, each of which reflects some aspect of the object under study and is included as an element in the general system of models. From a mathematical point of view, any complex object or complex system is a set in which a predetermined relation R with fixed properties P is realized. Such a relation is usually represented by requirements of a certain order, connections between elements of the system (object): events occurring in one of the elements of the system affect events in other elements. When optimizing residential development in microdistricts, it is necessary to optimize the demographic ratio of apartments. However, this condition is not always possible to fulfill, since in standard projects the necessary ratio of apartments differs from the required one. Subsequently, this problem of apartment layout closely affects the general optimization problem. Complex systems are characterized by three groups of variables: input variables generated by systems external to the studied system: X = X₁, X₂, ..., Xₙ, output variables generated by the studied system, determining the impact of the system on the environment: Y = Y₁, Y₂, ..., Yₘ, state coordinates characterizing the dynamic behavior of the studied system. Control is optimal if the transition of the system from one state to another, corresponding to the achievement of the goal, is accompanied by minimal costs. Design processes, as an object of study, are dynamic complex systems. They are characterized bythe following features: • diversity of structure; • difference in the nature of elements; • multi-connectedness of elements; • dynamism; • multi-criteriality; • probabilistic nature of the processes taking place. In the context of solving the main problem of any optimal control system for a complex object, several subproblems are solved, without solving which it is impossible to achieve the set control goal. Such problems include control problems, operational forecasting of ongoing processes and selection of control actions on them, depending on the current and forecast state, as a result of which the set control goal is achieved. For normal operation of the design control system, as with any complex object, it must have: • a control goal; • an ideal model of the object's functioning, calculated so that the goal is achieved in an optimal way; • a model of the actual state; • decision procedures that eliminate deviations between the ideal model and the actual one and implement the control goals. Pattern recognition methods have long been successfully used in solving problems of classification, forecasting and control of complex objects, in particular complex design processes. A set M of objects is given, relative to which classification is performed. It is known that M is represented as a sum of subsets called classes, M = ∪K. The task is to specify an algorithm A such that A(Z) = A(I, Sq) = ||βᵥ||qₓₗ, Information I is standard training information: I = {(S₁,...,Sₘ), α(S₁), α(S₂), ..., α(Sₘ)}, α(Sᵢ) = (αᵢ₁, αᵢ₂, ... αᵢₗ). Here αᵢⱼ = P(Sᵢ) = "Sᵢ ∈ Kⱼ", {0,1,Δ}; Sᵢ = (aᵢ₁, aᵢ₂, ..., aᵢₙ), i = 1,...,m, where 0, 1, Δ respectively mean that Sᵢ ∈ Kⱼ, Sᵢ ∉Kⱼ and it is not known whether S belongs to class Kⱼor not. The widespread use of recognition methods is mainly due to the fact that their use does not require a high level of formalization of training information. A finite list is required with applicants and an indication of the class of belonging. In recognition itself, there are two approaches to solving the recognition problem. One of the approaches is called the discriminant approach, the other is structural. Within the framework of the discriminant approach, the problem of classifying an admissible object is solved (supervised learning) or dividing an admissible set of objects into classes (taxonomy). Structural recognition is more general in the range of problems solved - in addition to the classification problem, it solves the problems of describing the structure of a complex object and identifying the elements of its structure that provide information about the behavioral features of a complex object in various description classes. In other words, the use of structural methods allows you to build a model of the object under study and use it when solving the problem of object control [1,2,3]. In this case, the following problems are solved: • setting a system of features and describing learning objects in it; • reducing the feature space; • analyzing learning objects and identifying structural elements of description; • describing learning objects in a structural form; • recognition. Of these tasks, two tasks are traditionally distinguished due to their importance: • evaluation and selection of the most important parameters of description and reduction of the initial feature space; • forecast of the course of the current technological process by promptly assigning it to different classes by the quality of the output product. This exhausts the traditional application of pattern recognition methods. The use of advisory control systems includes a set of solutions to the above problems, i.e. first, evaluation and selection of features, then, based on them, in real time, forecasting the course of the process, and, in addition, directly for the control task - identifying modes in the class of optimal ones that are closest to the current mode and setting recommendations on the values of important parameters that need to be changed to bring the mode into the optimal class [4]. There are a number of works in which, along with the use of pattern recognition methods, other methods are also used, for example, regression analysis. The process of constructing a model itself consists of two stages: • dividing a set of production situations into homogeneous groups of taxa; • constructing a regression model of each taxon. As a result, the resulting combined model is more adequate to the production process than the general regression model. The basis of the consulting control systems are structural methods of pattern recognition, which, among other things, by identifying the characteristic features of classes, allow us to analyze and adequately describe complex objects. This gives rise to several problems: • firstly, the same parameter responses can be achieved by different control actions; • secondly, some control actions cannot be feasible and must be replaced by adequate ones; • thirdly, different control actions can differ in price or speed of action, etc. Thus, recognition methods are successfully used in problems of forecasting, classification and management of complex processes, which include the design process. REFERENCES 1. Dzhumabaev Kh.R., Nefedov L.I., Daliev A.Sh., Mardiev N. Mathematical and software support for CAD of residential development. Tashkent: Fan. 1992. 256 p. 2. Boltyansky V.T. Optimal control. Moscow: Knowledge. 1974. 307 p. 3. Bondorenko B.I. Calculation of a set of houses in a development that meets demographic conditions. In the collection: Mathematical methods in urban development. Issue 2. - Kyiv: Budivelnik. 1969., pp. 23-32. 4. Daliev A.Sh. Algorithms and models of structural recognition for solving problems of classification, forecasting and management of complex processes. "Yangi O‘zbekiston Iqtisodiyoti" - "Economy of the new Uzbekistan" (ISBN 978-9943-5256-3-4), National University of Uzbekistan named after Mirzo Ulugbek, T. 2024. pp. 24-26.