Specificity of risk assessment in project management

Abstract

The work is devoted to the issues of risk forecasting in project management systems. The paper proposes to use the analytical capabilities of neural networks of direct propagation. This is an option that allows you to take into account the influence of subjective factors when determining risk in project management. The model of such a network makes it possible to use bifurcation dependencies as a formula for activating synapses. This approach has its difficulties in training a neural network. It allows you to formalize its work in conditions of uncertainty of input signals for individual neurons. A formula for taking into account bifurcation dependencies when using them in neural networks of direct propagation is proposed. The scope of application of such a dependence in the training models of the neural network of direct propagation is shown. This is shown on the example of a project management system, where both the contractor and the customer have a subjective factor. The possibilities of such models for analyzing possible risks both in the creation and in the implementation of projects are shown. At the same time, the risk of losing any factors from among those that lead to the emergence of risk-forming scenarios and events is excluded. This shows how easy it is to take into account the uncertainty in signal activation systems when training a neural network. Such an approach can make it possible to find solutions in a selected area for a wide range of similar problems. However, the requirement for network architecture can be problematic because of its individuality for each type of task. For the selected system, key scenarios have been identified, including subjective ones, which depend on the actual output functions allocated through the work of the trained perceptron. They reflect the risk system in such projects. And we do this in a generalized version, without specification for design systems. The proposed model of a direct propagation network can be extended and specified for a variety of tasks of this type, differing only in the meaning of input signals as logical statements. In this aspect, the material can be of a methodological nature and be applied not only in project management, but also in training systems