Showing papers by "Lauri Kutt published in 2020"

Residential Load Forecasting Using Recurrent Neural Networks

Abstract: In Electrical systems, load forecasting is very important as it has implications on flexibility, smooth operation, and economical aspects as well. The residential load depends on household size, weather season, numbers of load, number of occupants and their behavior, types of devices, etc. Thus, making its accurate forecasting a very difficult job. In this research, machine learning and deep learning-based Recurrent Neural Networks (RNN) algorithms are used for the day-ahead load forecasting of an Estonian household. A data set based on measured load values of an Estonian household is used in the development of this forecasting model. The simulation results indicate that the RNN based algorithm gives better forecasting based on lower Root Mean Square Error (RMSE) value.

Time Dependency of Current Harmonics for Switch-Mode Power Supplies

Abstract: This paper presents the time-dependent variance in the current harmonics emission by power supplies during power quality measurements. Power quality problems are becoming more significant with the adoption of power electronic-based circuits such as power supplies. The switch-mode power supplies are widespread as industrial, commercial, and domestic electrical loads. They draw non-sinusoidal current from the utility and inject current harmonics. Therefore, they are the reason for poor power quality and reduction in the power factor. The current harmonics emission from these power supplies depends on the circuit topology, operating conditions, and filter inside them. The harmonic emission estimations are critical for network operators; however, various uncertainties have made it a complicated task. The time-dependent stability affects the magnitude and phase angle of the harmonic current measurements and estimation of power quality indices. This paper investigates the variation in current harmonics emitted by the power supply during the initial unstable period under constant load and operating conditions.

Performance Comparison of PD Data Acquisition Techniques for Condition Monitoring of Medium Voltage Cables

Abstract: Already installed cables are aging and the cable network is growing rapidly. Improved condition monitoring methods are required for greater visibility of insulation defects in the cable networks. One of the critical challenges for continuous monitoring is the large amount of partial discharge (PD) data that poses constraints on the diagnostic capabilities. This paper presents the performance comparison of two data acquisition techniques based on phase resolved partial discharge (PRPD) and pulse acquisition (PA). The major contribution of this work is to provide an in-depth understanding of these techniques considering the perspective of randomness of the PD mechanism and improvements in the reliability of diagnostics. Experimental study is performed on the medium voltage (MV) cables in the laboratory environment. It has been observed that PRPD based acquisition not only requires a significantly larger amount of data but is also susceptible to losing the important information especially when multiple PD sources are being investigated. On the other hand, the PA technique presents improved performance for PD diagnosis. Furthermore, the use of the PA technique enables the efficient practical implementation of the continuous PD monitoring by reducing the amount of data that is acquired by extracting useful signals and discarding the silent data intervals.

Forecasting of Energy Consumption and Production Using Recurrent Neural Networks

Abstract: Energy forecasting for both consumption and production is a challenging task as it involves many variable factors. It is necessary to calculate the actual production of energy and its consumption as it is very beneficial in maintaining demand and supply. The reliability and smooth functioning of any electrical system are dependent on this management. In this article, the Recurrent Neural Network (RNN) based algorithm is used for energy forecasting. The algorithm is used for making three days ahead prediction of energy for both generation and consumption in Estonia. A comparison is also made between our proposed algorithm and the forecasting algorithm used by Estonian energy regulatory authority. The results of both algorithms indicate that our proposed algorithm has lower Root Mean Square Error (RMSE) and is giving better forecasting.

Case study of non-linear PV inverter devices attached to the LV distribution network

Abstract: Every year, more and more solar power plants are connected to the grid, producing electricity in an environmentally sustainable manner. The increasing number of photovoltaic (PV) installations and their integration into the low voltage (LV) distribution network (DN) is having an impact in terms of power quality (PQ). For example, the voltage in the DN can sustain high distortion values. The impact of a PV installation on the LV network is analysed in this research. The field measurements were carried out over a 3-week period at a solar power plant with a total output power of 160 kW in an Estonian rural municipality. The measurement results provide the opportunity to look more closely at the effect of the solar power plant on the supply voltage of the LV DN. Parameters such as voltage variation within a one-minute period, the asymmetry of the voltages and the total harmonic distortion of the voltages are discussed here.

Impact of Cable Impedance on the Harmonic Emission of LED Lamps

Abstract: This paper presents the effect of installation cables on the current harmonics emission due to nonlinear loads. Climate change and volatile fuel prices are persuading policy-making institutions to move in the direction of energy efficiency practices. The energy-efficient models of household appliances contain power electronic circuits and, thus, a source of current harmonics. In order to estimate the influence of current harmonics on the distribution network, an accurate assessment of power quality is required when the non-linear sources are connected to the distribution grid. However, accurate assessment is a challenging task because of the involvement of various uncertainties. This paper focuses on the influence of installation cables on the current harmonics connected with the nonlinear devices.

Probabilistic bivariate modeling of harmonic current

Abstract: Modeling modern nonlinear loads with varying current poses a significant challenge. While the traditional numeric models provide an adequate representation of stable loads, if a device or a group of devices operate at different modes with varying harmonic currents, a statistical approach is necessary to represent the extent and the shape of the variation. This paper provides an overview of probabilistic modeling methods for harmonic currents and presents a novel method for empirical nonparametric harmonic current modeling.

Comparison of Current Harmonic Emission by Different Lighting Technologies

Abstract: This paper presents an analysis of current harmonic emission form different lighting technologies in the residential network. Lighting consumes a significant share of total electricity consumption. The governments are making policies to promote the usage of energy-efficient electrical appliances. The energy-efficient lamps are now replacing the inefficient incandescent lamps to reduce power consumption. These lamps encompass non-linear circuits and reduce the power quality of the distribution network. The large scale penetration of energy-efficient can have an enormous effect on power quality issues in the grid. This study provides an estimation of harmonic current emission by the gradual replacement of incandescent lamps with compact fluorescent and light-emitting diode lamps.

Characterization of Corona and Internal Partial Discharge under Increasing Electrical Stress using Time Domain Analysis

Abstract: Aging and abnormal stresses accelerate insulation degradation and reduce the lifetime of power equipment. Partial discharge (PD) measurement is an effective tool to study the condition of the insulation. Reliability of PD diagnosis depends on the accurate interpretation of the measured PD signals. PD itself is a complex phenomenon and the presence of different types of discharge sources makes interpretation of the PD data quite challenging. This paper investigates internal and corona PDs in order to distinguish them when both are active simultaneously. The presented work identifies the PD sources based on time domain analysis that provides a simplified solution as compared to identification techniques based on different statistical features leading to complex data processing. While superimposed phase-resolved partial discharge (PRPD) patterns provided incomplete information, time domain PD characteristics e.g. pulse repetition rate and pulse amplitude combined with PRPD mapping are analyzed to differentiate PD activity. Furthermore, the electrical stress (voltage level) is increased gradually in the experiments made and PD behavior is studied. The presented technique contributes to enhancing the accuracy of PD diagnosis that is necessary for appropriate decision making concerning the repair of affected components.

Requirements to Data Acquisition and Signal Analysis for Electrical Grid Condition Monitoring

Abstract: Electrical outages cause economical loss but in many cases lead also to broken devices. Similarly, repeated voltage fluctuation may result in overstress of the components of the grid which leads to electrical failure. Before any electrical fault, usually, some smaller-scale anomalies occur in load current or voltage. These anomalies can usually be detected from measured quantities by using the Fourier transform and by analyzing voltage changes indicated as additional frequency components. However, the noise present in the real signal makes this analysis less accurate. Therefore, a wavelet transform based denoising method is proposed here along with Fourier transform to overcome this problem. In this paper, a discussion on the options for real-time diagnostics of an electrical grid is presented. The methods are compared and analyzed to assess the reliability of the proposed method. The anomalies observed are linked to unexpected voltage RMS values which correspond to variations in frequency domain components.