Jelena Milojković

Comparison of some ANN based forecasting methods implemented on short time series

Several ANN architectures were implemented for forecasting based on reduced time series encountered in the subject of waste management of electrical and electronic products. Along with the existing ones new original training set organization and ANNpsilas structures are proposed that perform favorably comparing with the existing ones.

Short-term forecasting in electronics

Short time series are characterised by a lack of the following: trend information, randomness and periodicity. This makes prediction based on them very difficult or even impossible. This unfortunately is frequently the case in modern electronic developments. In this study, we propose the implementation of some architectures of artificial neural networks as a potential solution to that problem. Examples will be given related to the prediction of quantities of obsolete computers and verification of Moores law in modern electronic production.

Forecasting based on short time series using ANNs and grey theory - some basic comparisons

Two modern forecasting methods based on short time series are compared. Results obtained by use of artificial neural nets (ANNs), are contrasted to the ones produced by use of the so called grey theory or Grey Model (GM). Specifically, the Feed-Forward Accommodated for Prediction (FFAP) and the Time Controlled Recurrent (TCR) ANNs are used along with the GM(1,1) algorithm for one- and two-steps-ahead forecasting of various quantities (electricity loads, number of fixed telephones lines, obsolete computers, etc). Advantages of the ANN concept are observed.

Dynamic one step ahead prediction of electricity loads at suburban level

One step ahead prediction based on short time series is presented. It will be shown here first that for the subject of short term prediction of electricity load, even though a large a-mount of data may be available, only the most recent of it may be of importance. That gives rise to prediction based on limited amount of data. We here propose implementation of some instances of architectures of artificial neural networks as potential systematic solution of that problem as opposed to heuristics that are in use. To further rise the dependability of the predicted data averaging of two independent predictions is proposed. Examples will be given related to short-term (hourly) forecasting of the electricity load at suburban level. Prediction is carried out on real data taken for one suburban transformer station. Implementation of an on-line real time prediction system is presented.

ICT and power: Synergy and hostility

An attempt is made in this paper to summarize the state of the art in the interaction between the ICT (information communication technologies) that becomes ubiquitous and the electrical power production and distribution. Both are considered to have a difficult task to fulfill enormous rise of demand that is becoming above all expectations. In the same time they technologically interfere in the sense that they mutually help in the fulfillment of their main task while in the same time loading each other with problems inherent to the respective technology. In the paper we will try to merge our (LEDA laboratory of the University of Niš) own results with ones available in the literature in order to give as complete a picture of the subject as possible.

On prediction in electronics

We here propose implementation of some architectures of artificial neural networks as a potential solution of that problem prediction in electronics based on short time series. Examples are given related to predictions for verification of Moores law in modern electronic developments.

Low-Voltage Circuit Breakers Based on WBG Solid-State Devices

Conventional circuit breakers suffer from two main deficiencies: they are slow to operate and develop an electrical arc. These may be overcome by using solid-state switches which in turn introduce other problems, most significantly power dissipated while in the on-state. Nevertheless, a number of solid-state devices are candidates for implementation as low-voltage circuit breakers and there are several options based on the semiconductor material that may function as high-power switches. This paper presents a unique, extensive and systematic evaluation of these options. Voltage-controlled devices are selected due to the simplicity of the controlling circuit and their resilience to dv/dt-induced switching. Properties of fully solid-state circuit breakers are established and systematic comparisons are made among switches built of silicon and other wide bandgap (WBG) devices such as SiC MOS and GaN HEMT transistors. Using SPICE simulation it is shown that solid-state circuit breakers (SSCBs) based on WBG devices exhibit superior characteristics compared with silicon devices, with faster switching and higher voltage and current ratings. Hybrid circuit breakers, combining both conventional and solid-state switches, are discussed too and a new design circuit is simulated and compared to both conventional and fully solid-state designs.

ANN application for the next day peak electricity load prediction

One step ahead prediction of peak electricity loads based on artificial neural networks (ANN) is presented. Two architectures of ANNs were implemented to produce predictions that were used to generate the final value as an average. The time instants when daily peak loads occur are produced simultaneously. Examples will be given confirming both the feasibility of the method and the need for further elaboration of the procedure.

ANN versus Grey theory based forecasting methods implemented on short time series

Two modern concepts implemented for forecasting based on reduced time series are contrasted. Results obtained by use of artificial neural nets (ANNs), already discussed at this conference, are compared with the ones obtained by implementation of the so called Grey theory or Grey Model (GM). Particularly, feed-forward accommodated for prediction (FFAP) and time controlled recurrent (TCR) ANNs are used along with the GM(1,1) algorithm for one- and two-steps-ahead forecasting of various quantities (obsolete computers, electricity loads, number of fixed telephones etc). Advantages of the ANN concept are observed. The GM(1,1) was studied in the appendix and compared with no advantages against the least-mean-squares approximation by an exponential.