Search

In this article, a machine-learning-based model for the regression of cloud-to-ground lightning location and peak current from time-domain waveforms of lightning-induced voltage measurements on overhead transmission lines is presented. A principal component analysis (PCA) procedure is applied for extracting significant features and decreasing the dimension of the input vector. Then, a shallow neural network is trained with the results of the PCA. The obtained results show that the proposed approach can be the base for a tool able to regress lighting location with an accuracy comparable to or even better than traditional methods [i.e., lightning location system (LLS)] and provide a peak current estimate more accurate than LLS and more actual and widespread than direct tower measurements (which are limited to a reduced number of recorded events in some specific regions). Such a tool would also have significant advantages in terms of costs, since it would not require a dedicated instrumentation.

In this contribution the authors use a deep neural network (DNN) based approach for the optimization of an electric motor, taking into account both electromagnetic (EM) and mechanical constraints, i.e., approaching the problem from the multiphysics point of view. In the design process of high speed electric motors, the mechanical design of the rotor is of noteworthy importance, and in case of reluctance motors it cannot be separated from the electromagnetic design. The multiphysics model is created by using a commercial finite element method (FEM) software, and a multiobjective optimization procedure is run by using the before mentioned software. This is the selected tool for the generation of the training dataset used to train a DNN, which is used to refine the sub-optimal solutions previously obtained. The results show that the use of a two-step optimization leads to a better solution.

Purpose: This study aims to investigate the possible use of a deep neural network (DNN) as an inverse solver. Design/methodology/approach: Different models based on DNNs are designed and proposed for the resolution of inverse electromagnetic problems either as fast solvers for the direct problem or as straightforward inverse problem solvers, with reference to the TEAM 25 benchmark problem for the sake of exemplification. Findings: Using DNNs as straightforward inverse problem solvers has relevant advantages in terms of promptness but requires a careful treatment of the underlying problem ill-posedness. Originality/value: This work is one of the first attempts to exploit DNNs for inverse problem resolution in low-frequency electromagnetism. Results on the TEAM 25 test problem show the potential effectiveness of the approach but also highlight the need for a careful choice of the training data set. [ABSTRACT FROM AUTHOR]

In this paper, a decision-directed method is proposed for channel estimation and equalization in Powerline communication (PLC) based on orthogonal frequency-division multiplexing (OFDM). In the DDE (decision-directed estimation) method proposed, the received subcarrier-symbols are presented to a competitive neural network that adaptively finds the clusters of the QAM (Quadrature Amplitude Modulation) constellation. This method does not require a priori knowledge on the powerline then allowing a blind estimation of the channel with tracking capabilities limited by constraints on the phase changes. Simulations on a realistic indoor power-line system show that the proposed method achieves very good channel estimation and equalization performances and that it is robust to impulsive noise and non-linearities. (1) Index Terms--Powerlines, Neural Networks.

Planning and managing the operation of cogenerative plants (CHP) is increasingly becoming an industrial necessity because of the participation of CHP plants in the day-ahead energy market and the need to deal with heat demand fluctuations. Short-term operational planning usually considers power and heat demand forecast, which can widely fluctuate daily and seasonally, to maximise the net revenue and fulfil the heat requirements. In this framework, a Thermal Storage (TS) can balance day-night fluctuations due to outdoor temperature, as well as unexpected energy surplus and deficiencies caused by heat demand forecast errors, thus giving the CHP plant more operational flexibility. In this paper, the capability of a TS to balance thermal load fluctuations and forecast errors is investigated when a Machine Learning (ML) thermal load forecast for short-term predictions up to 48 h is used. The TS is modelled as a layered storage tank with perfectly mixed layers. Weather and consumption data from 2018 to 2020 related to a large greenhouse powered by a CHP plant located in Tuscany were used as a case study to perform the training and validation of the forecast model as well as to analyse the TS capability in balancing fluctuations with volumes ranging from 500 up to 10,000 m3. Several ML algorithms were used and compared against a naive prediction based on load persistency. Support Vector Machine (SVM) resulted as the best-performing algorithm. Using SVM leads to better exploitation of the TS capacity, compared to persistence, leading to a more regular State of Charge (SOC) trend and allowing the system to operate within expected conditions up to 80 % of the year. In contrast, a more naïve forecast approach brings to relevant volume size increase to achieve equal performance. Finally, a more accurate forecast reduced the TS size to 50 %, potentially cutting the investment and operational costs compared to the load persistency forecast strategy.

The pantograph-catenary subsystem is a fundamental component of a railway train since it provides the traction electrical power. A bad operating condition or, even worse, a failure can disrupt the railway traffic creating economic damages and, in some cases, serious accidents. Therefore, the correct operation of such subsystems should be ensured in order to have an economically efficient, reliable and safe transportation system. In this study, a new arc detection method was proposed and is based on features from the current and voltage signals collected by the pantograph. A tool named mathematical morphology is applied to voltage and current signals to emphasize the effect of the arc, before applying the fast Fourier transform to obtain the power spectrum. Afterwards, three support vector machine-based classifiers are trained separately to detect the arcs, and a fuzzy integral technique is used to synthesize the results obtained by the individual classifiers, therefore implementing a classifier fusion technique. The experimental results show that the proposed approach is effective for the detection of arcs, and the fusion of classifier has a higher detection accuracy than any individual classifier.

Assessment of the current collection properties of a pantograph–catenary system mounted on a train is of great importance. Excessive electric arcing can lead to wear of the system components, and, at the same time, it can be an index of wear status. In this paper we investigate the possibility of detecting arcing events in the pantograph–catenary collection system without the need of additional equipment installed on-board the train. Data that is currently measured and recorded for modern high-speed trains (i.e. voltage and current) are analysed in order to detect and quantify electric arcs and shed light on the current collection quality of the pantograph–catenary system. This work was performed in cooperation with Trenitalia s.p.a. who provided the data it collects on-board high-speed trains in regular passenger service.

The aim of this manuscript is to describe the first year of our experience using extracorporeal membrane oxygenation support.