1410 IEEE TRANSACTIONS ON MAGNETICS, VOL. 47, NO. 5, MAY 2011 Layer Recurrent Neural Network Solution for an Electromagnetic Interference Problem Dan D. Micu , Levente Czumbil , Georgios Christoforidis , and Andrei Ceclan Electrical Engineering Department, Technical University of Cluj Napoca, Romania Electrical Engineering Department, Technological Institute of West Macedonia, Greece The paper presents an original contribution related to the implementation of a neural network artificial intelligence (AI) technique through Matlab environment, on the study of induced AC voltage in the underground metallic pipeline, due to nearby high voltage grids. The advantage yields in a simplified computation model compared to FEM, and implicitly a lower computational time. In comparison with other neural network solutions identified in the literature, where the induced AC potential is directly evaluated, the authors of this paper propose a new neural network solution to evaluate MVP on the studied domain, using a larger training database for a large panel of different geometries. Index Terms—Electromagnetic compatibility, electromagnetic fields, finite element method, neural networks, pipelines, transmission lines. I. INTRODUCTION C URRENTLY, to reduce construction costs of underground metallic pipelines (designed to transport liquid or gaseous substances like oil, water, gas, etc.) they are placed in the same distribution corridor as power system transmission lines. Due to electromagnetic interference between high voltage (HV) power lines (PL) and these metallic pipelines (MP), induced AC po- tentials will appear. This AC voltage may be dangerous on both the operating personal and on the structural integrity of MP, last due to corrosive effects of the induced current [1] Usually the electromagnetic interference problems are studied with the finite element method (FEM). In order for this method to be applied on electromagnetic interference prob- lems, it is known that it requires expensive computation time, because a new mesh is required for each construction geometry considered. Therefore, it may be of interest a scaling method of the results from one configuration case to another, so as to provide a lower computational time. [2], [3] Thus, an artificial intelligence based method is presented with emphasize on the input variables and on the architecture of the neural network. II. ELECTROMAGNETIC INTERFERENCE PROBLEM We proposed the evaluation of the magnetic vector potential (MVP) on the surface of MP, where the geometry of the electro- magnetic interference problem (HVPL and sky wires) is the one presented in [1]. The calculated MVP will be used to determine the induced AC potential in the metallic pipeline. Fig. 1 shows the cross section of the HVPL-MP common dis- tribution corridor. The problem refers to a buried metallic gas pipeline which shares for 25 [km] the same distribution corridor with a 145 kV/50 [Hz] frequency HVPL. The transmission line consists in HAWK ACSR conductors and two 4 mm radius sky wires. End effects are neglected for the inductive interference calcu- lations, therefore leading to a two dimensional problem. Manuscript received May 29, 2010; accepted October 26, 2010. Date of cur- rent version April 22, 2011. Corresponding author: D. D. Micu (e-mail: Dan. Micu@et.utcluj.ro). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TMAG.2010.2091494 Fig. 1. Cross section of the system under investigation. Fig. 2. Implementation and training of the proposed NN. The HVPL-MP system is represented by a two dimensional problem, which depends on the separation distance , soil resis- tivity and coordinates of the point where the MVP has to be determined. The previous assumption leads to a linear two- dimensional electromagnetic diffusion problem, for the -direc- tion components of MVP and the total current density vector. Thus, taking into account the cross section of the studied problem, the -direction component of the magnetic vector po- tential and of the total current density are described by the following equation system: (1) 0018-9464/$26.00 © 2011 IEEE