Applied Mathematics and Control Sciences

The purpose of this work is to provide decision support for the management of software development information systems. The novelty of the results is determined by the fact that the developed models and algorithms for managing the process of developing information systems software provide an increase in the quality of management decisions by ensuring the accounting of uncertainty in the estimates of experts at the initial stage of the project, automated calculation of the volume of changes as a proportion of requirements implemented in the previous stages/iterations and affected by the implementation of new requirements, based on the calculated relationships between the elements of the project and the automated calculation of estimates of the complexity and timing of software development based on fuzzy inference systems, dynamically adjusted as retrospective data are accumulated.

The goal of implementing autopilot system into vehicle is the automation of its movement on a dynamic landscape. Vehicle movement task can be considered as a particular case of a classification task, because the problem of every autopilot system is an optimal set of agent actions selection depending on the vehicle state. Over the recent years a great success has been achieved in self-driving vehicles learning. This article is about current approaches to implementation of autopilot system with the focus on portable vehicles. Particularly, it covers neural networks’ algorithms, appropriate for such project, and analyses them between each other according to available state-of-the-art reference sources. Work quality, learning speed, convergence guarantee and AI tasks stack coverage extent were chosen as the comparative characteristics. As a result, DQN (Deep Q-Learning) algorithm was chosen as the best option for implementation in terms of both virtual and real portable self-driving vehicles. This research paper also contains basic details about useful software solutions: virtual platforms and frameworks.

Statement of the rating and control of complex objects problem under uncertainty is formulated. Seven methods are presented for assessing the state of individual properties of a complex object, designed for various conditions of uncertainty. The source of uncertainty can serve as objective tools of control, that is, estimates of particular properties of an object can be either exact values on a set of real numbers or interval estimates, in addition to this, the state of individual properties can have a probability distribution about the possible state of an object. The qualitatively described properties of an object are subject to uncertainty, the source of which is the subjectivity of the judgments of experts involved in their evaluation. It is important that experts can also evaluate quantitative indicators, for example, in cases where the tools of objective control are not available, that is, subjective input methods are applicable not only to qualitative indicators, but also quantitative ones. In order to aggregate information about private properties, it is proposed to use the matrix mechanisms of complex assessment that are known and obtained by the author. The defined matrix mechanisms of complex evaluation are as follows: the tree fuzzy rating mechanism with an additive-multiplicative approach and two max-min approaches to set theoretic operations, as well as two continuous complex evaluation mechanisms. Six model examples demonstrate procedures of aggregating two criteria with accurate real values, as well as with different uncertainty such as interval values, fuzzy values and F-Fuzzy values.