Development of strategy of management of the difficult projects on the basis of tutoring algorithms with the reinforcement

In this work the main approaches to formation of strategy of management of the difficult projects on the basis of methods of machine learning with a reinforcement are considered. The scale and a variety of tasks of the difficult projects, the number of performers and the used resources can create very the compound combinatory circuits therefore effective expeditious interpretation of such structures and adoption of the justified decisions guaranteeing keeping of terms represent the considerable complexity. To overcome the designated difficulties, the intellectual systems capable to analyze and predict the temporary and resource characteristics relating both to separate tasks, and to their set are applied. Such process is integrated into Markov model of a decision making. Development of the system of indexes as the quantitative, and the qualitative, intended for job evaluation projects is its part. All these aspects play an essential role in determination of those technical and economic risks which can arise during implementation of projects a path of definition of strategy. The research objective consists in formation of intellectual model and algorithm of a decision making when choosing strategy of implementation of the difficult projects on the basis of models of machine learning with a reinforcement. Research techniques. For the solution of a problem of the choice of strategy of management of the difficult projects the Markov model of acceptance of decisions used for assessment of value of states and actions of the agent when choosing strategy on the basis of the temporal difference method was used. As a result, assessment of value of actions was made on the basis of the algorithm SARSA allowing to receive optimal variants of actions on each problem of implementation of the project depending on factors of internal and external indeterminacy. Results. The methods presented in article provide the effective tool for operational permission of a wide range of the tasks which are inevitably arising at the embodiment of the difficult projects, considering both internal, and external factors of indeterminacy. Thanks to the model of machine learning with a reinforcement based on the Markovian process the basis for the system of support of a decision making is created. This system is capable to estimate dynamically the course of implementation of the project and to form adaptive and exact strategy with low level of an error. Various sub models, such as regressions, qualifiers, clustering and deep neural networks can be integrated into its structure. Conclusion. The received results in to the complete measure are applicable for formation of effective strategy of management of the difficult projects. It is proved that use of Markov models of a decision making fully allows to level indeterminacy when determining nature of the distribution law of a random value of a universe of the data necessary for tutoring of the project. Besides the discretization in MPPR corresponds to the nature of formation and management of the difficult project which is carried out by variation of parameters ε and γ and in case the project lasts the considerable time perhaps of use of parameter β which effectiveness of application is subject to further researches.