Artificial Intelligence, Machine Learning and Neural Networks for Tomography in Smart Grid – Performance Comparison between Topology Identification Methodology and Neural Network Identification Methodology for the Distribution Line and Branch Line Length Approximation of Overhead Low-Voltage Broadband over Power Lines Network Topologies

Abstract

Until now, the neural network identification methodology for the branch number identification (NNIM-BNI) has identified the number of branches for a given overhead low-voltage broadband over powerlines (OV LV BPL) topology channel attenuation behavior [1]. In this extension paper, NNIM-BNI is extended so that the lengths of the distribution lines and branch lines for a given OV LV BPL topology channel attenuation behavior can be approximated; say, the tomography of the OV LV BPL topology. NNIM exploits the Deterministic Hybrid Model (DHM) and the OV LV BPL topology database of Topology Identification Methodology (TIM). By following the same methodology of the original paper, the results of the neural network identification methodology for the distribution line and branch line length approximation (NNIM-LLA) are compared against the ones of the newly proposed TIM-based methodology, denoted as TIM-LLA.

Big Data and Neural Networks in Smart Grid - Part 2: The Impact of Piecewise Monotonic Data Approximation Methods on the Performance of Neural Network Identification Methodology for the Distribution Line and Branch Line Length Approximation of Overhead Low-Voltage Broadband over Powerlines Networks

“…Except for the previous default operation settings, the following assumptions of [1] are also taken into account in this companion paper, namely: (i) The number of branches of the examined indicative OV LV BPL topologies (say, suburban and rural cases of Table 1 of [1]) is assumed to be known; (ii) the database representativeness, which is analyzed in [23] for the operation of NNIM-LLA, is assumed during the application of the default operation settings B' and C'; (iii) the exclusion of the symmetrical OV LV BPL topologies from the OV LV BPL topology database so as not to disrupt the approximations due to the symmetry of BPL topologies described in [41], [42]; (iv) the inclusion of the examined suburban and rural cases into the TIM OV LV BPL topology database; and (v) the mechanism described in [1] for preventing the unacceptable NNIM-LLA approximations of [23] (i.e., at least one of the approximated distribution and branch line lengths is below zero given the fixed length of 1000m between the transmitting and receiving ends for all the applied OV LV BPL topologies).…”

“…Deterministic Hybrid Model (DHM), which describes the BPL signal propagation and transmission across the topologies of the overhead low voltage (OV LV) BPL networks [12]- [20], has acted as the channel model basis while artificial intelligence (AI), machine learning (ML) and neural network (NN) features have been concatenated after it in [1]. Indeed, exploiting the available big data of the Topology Identification Methodology (TIM) BPL topology database for the OV LV BPL topologies of [21], [22] and AI -ML -NN functionalities, the neural network identification methodology for the distribution line and branch line length approximation (NNIM-LLA) has been proposed for the OV LV BPL topologies in [23] while its performance has been assessed in [1] when measurement differences of various intensities may occur. In fact, measurement differences between experimental and theoretical OV LV BPL topology channel attenuation values may be observed due to several practical reasons and "real" life conditions, as shown in [22], [24].…”

“…In accordance with [1], [22], [24]- [27], a typical scenario to take into account the measurement differences during the BPL topology channel attenuation analysis is their handling as error distributions, such as the Continuous Uniform Distributions (CUDs) of various amplitudes that are superimposed to the coupling scheme transfer function theoretical numerical results from DHM. In [1], NNIM-LLA has been deployed and benchmarked by exploiting the already knowledge and experience of [3], [23], namely: (i) the list of the indicative OV LV BPL topologies; (ii) default operation settings B of [23]; (iii) default operation settings C of [1]; (iv) the assumption of a priori knowledge of the number of branches of the examined indicative OV LV BPL topologies in each case (i.e. not-blind NNIM-LLA approximations); (v) the database representativeness, which is analyzed in [23] for the operation of NNIM-LLA; and (vi) the mechanism proposed in [1] against the unacceptable NNIM-LLA approximations of [23].…”

“…In [1], NNIM-LLA has been deployed and benchmarked by exploiting the already knowledge and experience of [3], [23], namely: (i) the list of the indicative OV LV BPL topologies; (ii) default operation settings B of [23]; (iii) default operation settings C of [1]; (iv) the assumption of a priori knowledge of the number of branches of the examined indicative OV LV BPL topologies in each case (i.e. not-blind NNIM-LLA approximations); (v) the database representativeness, which is analyzed in [23] for the operation of NNIM-LLA; and (vi) the mechanism proposed in [1] against the unacceptable NNIM-LLA approximations of [23]. In accordance with [3], [23], Root-Mean-Square Deviation (RMSD) has been chosen as the performance metric so that the impact of the measurement differences on the NNIM-LLA approximation performance can be evaluated as well as the countermeasure techniques against them.…”

“…not-blind NNIM-LLA approximations); (v) the database representativeness, which is analyzed in [23] for the operation of NNIM-LLA; and (vi) the mechanism proposed in [1] against the unacceptable NNIM-LLA approximations of [23]. In accordance with [3], [23], Root-Mean-Square Deviation (RMSD) has been chosen as the performance metric so that the impact of the measurement differences on the NNIM-LLA approximation performance can be evaluated as well as the countermeasure techniques against them. In [1], it has been revealed that NNIM-LLA presents an inherent mitigation efficiency against CUD measurement differences of low a CUD values (i.e., a CUD values that remain lower than 5dB).…”