Application of experiment planning methods in expert procedures for machine learning of a resource distribution model in fire response

Abstract

Introduction. Intellectualization of control systems, including decision support systems, is an urgent need in the field of fire management. There are various types of support for decision makers. The paper focuses on mathematical and referential support, which makes it possible, on the basis of machine-learned optimization models, to prepare for the decision maker a set of suboptimal solutions consistent with his system of preferences. As a basis for the distribution of limited resources of fire and rescue units with simultaneous calls, a transport model is used, the parameters of the payoff function of which are trained on a sample of retrospective data formed with the participation of an experienced decision maker. To select solutions, expert procedures are used that are optimized by experiment planning methods, which can significantly reduce the time to prepare a decision without losing its quality, which makes the study relevant. Purpose of research. The purpose of the article is to construct and illustrate algorithms for applying the principles of optimal design of an experiment in the procedures for expert evaluation of options for distributing homogeneous resources of fire and rescue units over simultaneous calls. Methods. Methods of optimization, expert evaluation, machine learning of optimization models, simulation modeling, experiment planning were used. Results and discussion. The proposed technology for accelerating expert procedures for choosing the optimal distribution of homogeneous resources of fire and rescue units for simultaneous calls based on the principles of optimal planning of the experiment is illustrated by the example of two fire and rescue units and three calls. This example shows the effect of reducing the number of necessary assessments performed by decision makers by more than 70 %. In the case of distribution of different types of resources for each of them, a separate procedure should be built, similar to that described in this article. Conclusion. The results of a study of the application of the principles of optimal design of an experiment to speed up expert procedures for choosing by a decision maker the best option for distributing limited resources of fire and rescue units between simultaneous calls are presented. The proposed algorithm seems appropriate to be included in the expert system for supporting the management of fires. Keywords: fire and rescue unit; simultaneous calls; resources; transportation task; expert assessments; experiment planning; decision maker.