An Experiential Learning System for a Resources Planning Problem VASILE MAZILESCU, DAN CĂPRIŢĂ, CORNELIA UDUDEC, MIHAELA NECULIŢĂ Department of Economic Informatics Dunărea de Jos University 47, Domneasca Street, Galati ROMANIA e-mail: vasile.mazilescu@ugal.ro ; caprita_dan@yahoo.com , cornelia.novac@ugal.ro neculitam@yahoo.fr Abstract: - This paper refers the problem to capture by a human agent of a specific fuzzy knowledge model using a Learning Hybrid System (LHS) for improving the knowledge model based on the classical theory of diagnosis. For this, we have used the parsimonious covering theory proposed by Peng and Reggia taking into account a fuzzy planning problem. The goal of this research is represented by the design and implementation of the LHS, using the basic methods of Artificial Intelligence (AI). The system was implemented in Delphi, experimentally obtaining a prototype of the LHS as well as a series of results that confirm the conceptual elements on which LHS relies on. Keywords: - Artificial intelligence, Diagnosis, Fuzzy planning 1. Introduction The learning process consists in the behaviour modifications of an agent (human or artificial) with adaptive characteristics, that allow the agent to execute the same task or problem solving specific tasks, in a more efficiently way. At conceptual level, the learning process refers to:  knowledge improvement by own means;  adaptation to new situations or different ones from those known;  management and qualitative analysis of the knowledge structure, including the generalization or specialization of a conceptual model. The main concern in most cases of learning is the incremental acquirement, the modification, consolidation and adjustment of the knowledge models for a specific domain. In a large sense, the learning process could be divided in to two categories: i) supervised learning: for every training instance is assigned a valid classification by a “professor”; ii) unsupervised learning: there are no training instances, and then the learning algorithms must find significant classifications. In our LHS, the learning process is supervised and the goal of this problem is that human agent can assimilate in a gradual way the planning knowledge so that he becomes, as far as possible autonomous in a restricted time l T (meaning learning time). In the following, we will designate the learning process the synthesizing process of structured knowledge models M 0 ,…,M k . They are specific to a control strategy based on the observation of certain cases (problem instances), that means supervised learning. The existence of a number of fuzzy knowledge models M 0 ⊂ M 1 ⊂ …⊂ M k , means a gradual and incremental learning process [2,4]. The conceived learning system consists of:  controlled process (P) defined by a certain structure and dynamics, with a precisely goal, that will be “inherited” by the knowledge based control system goal. This represents the domain problem;  control expert system (CES) of the process, which includes more models of the process;  diagnosis system (DS) embedded in the LHS structure, starts from a set of manifestations and generates the hypotheses (explanations). Based on the generated explanations, the DS activates a certain intern knowledge model of the process that will be used by the control expert system. If these explanations are valid, than they represent the sum of knowledge which permit the advance of the learning problem. This paper is structured as follows:  In Section 2 we present the main characteristics of Peng and Reggia theory, starting from the formal definition of a diagnostic problem, the problem solution and the diagnostic problem-solving algorithm;  In Section 3 the structure of the LHS system is pointed out;  In Section 4 the main results obtained from the testing of the LHS system functionality for the learning problem of the proposed planning problem; Proceedings of the 11th WSEAS International Conference on Sustainability in Science Engineering ISSN: 1790-2769 126 ISBN: 978-960-474-080-2