This paper introduces a new integrated framework for short-term time series forecasting used as decision logic support in Energy Management Systems (EMS). The proposed framework defines data preprocessing, forecast model training, and adaptation techniques. For data preprocessing, Gaussian and Kalman filtering were used in offline learning on historical data and online (real-time) forecasting. Optimal parametrisation for both filters was selected based on the forecast accuracy. Next, a multiple-input, single-output forecast model based on the Long Short-Term Memory (LSTM) neural network was tuned for each device. The forecast horizon is single-step, and chaining of predictions was used for longer horizons. Finally, the best model was adapted to the changes in the time series data. A significant drop was observed in the forecast accuracy on the past data after adapting the model to the new data using a naive approach. Due to this drop, a novel sample rehearsal selection strategy for adaptation of the model is proposed and compared to other different strategies. All the techniques are demonstrated based on the examples of forecasting multivariate time series for different household devices: a water heater, a refrigerator and an HVAC device. The use of data preprocessing improved the forecast accuracy by up to 56% compared to no preprocessing. Moreover, by introducing a sample rehearsal, knowledge retention was improved by up to 76.6% compared to the naive approach, and by up to 62% compared to the random selection of samples.

This paper proposes measures to improve the protection of MV distribution networks operating with feeders in a closed-loop arrangement. Bi-directional overcurrent relays (OCRs) are discussed, the selectivity of which is achieved through the timing coordination of their operation. The classic approach is formulated as a minimization of the operating times of all the OCRs. The proposed approach enhances the selectivity by considering the maximum operating time of substation OCRs and the unwanted trips of inloop OCRs. Moreover, the sensitivity is also increased by introducing an objective function that minimizes the pickup-current settings of all the OCRs together with their operating times. Furthermore, to fulfill the demanding requirements for operating times, variable penalties are introduced. Thus, the optimization procedure is forced towards the region with viable solutions for the optimization problem. Two variants of self-adaptive differential evolution have been used that both show better convergence when compared to the classic differential evolution. Moreover, ten mutation strategies were tested, where “rand/1/bin” showed the best results. A comparison with other methods for timing coordination shows that the proposed optimization results in a comparable value for the OCRs’ operating times. In order to further reduce the operating times, GOOSE communications between the OCRs are adopted. The proposed measures for improved protection operation are fully confirmed through dynamic simulations of the faults in the discussed 20-kV network. Moreover, the proposed protection design is already implemented and permanently operates in a 20-kV network with more than 5000 customers, whereas the field results show selective and reliable protection operation.

This paper deals with rotary and linear synchronous reluctance machines and synchronous permanent magnet machines. It proposes a general method appropriate for determining the two-axis dynamic models of these machines, where the effects of slotting, mutual interaction between the slots and permanent magnets, saturation, cross-saturation, and—in the case of linear machines—the end effects, are considered. The iron core is considered to be conservative, without any losses. The proposed method contains two steps. In the first step, the dynamic model state variables are selected. They are required to determine the model structure in an arbitrarily chosen reference frame. In the second step, the model parameters, described as state variable dependent functions, are determined. In this way, the magnetically nonlinear behavior of the machine is accounted for. The relations among the Fourier coefficients of flux linkages and electromagnetic torque/thrust are presented for the models written in dq reference frame. The paper presents some of the experimental methods appropriate for determining parameters of the discussed dynamic models, which is supported by experimental results.

This paper deals with the detection of saturation in the magnetic core of a welding transformer which is a part of a middle-frequency direct current (MFDC) resistance spot welding system (RSWS). It consists of an input rectifier, which produces dc bus voltage, an inverter, a welding transformer, and a full-wave rectifier that is mounted on the output of a transformer. During normal RSWS operation welding transformer's magnetic core can become saturated due to the unbalanced resistances of both transformer secondary windings and different characteristics of output rectifier diodes, which causes current spikes and over-current protection switch-off of the entire system. In order to prevent saturation of the transformer magnetic core, the RSWS control must detect that the magnetic core is approaching the saturated region. The aim of this paper is to present a reliable method for detection of magnetic core saturation that does not require an additional sensor. It is based on the artificial neural network (ANN). Its input is the measured primary current of the welding transformer. The applied ANN is trained to recognize the waveform of the current spikes in the primary current caused by the magnetic core saturation, which is used for magnetization level detection.

Clouds moving at a high speed in front of the Sun can cause step changes in the output power of photovoltaic (PV) power plants, which can lead to voltage fluctuations and stability problems in the connected electricity networks. These effects can be reduced effectively by proper short-term cloud passing forecasting and suitable PV power plant output power control. This paper proposes a low-cost Internet of Things (IoT)-based solution for intra-minute cloud passing forecasting. The hardware consists of a Raspberry PI Model B 3 with a WiFi connection and an OmniVision OV5647 sensor with a mounted wide-angle lens, a circular polarizing (CPL) filter and a natural density (ND) filter. The completely new algorithm for cloud passing forecasting uses the green and blue colors in the photo to determine the position of the Sun, to recognize the clouds, and to predict their movement. The image processing is performed in several stages, considering selectively only a small part of the photo relevant to the movement of the clouds in the vicinity of the Sun in the next minute. The proposed algorithm is compact, fast and suitable for implementation on low cost processors with low computation power. The speed of the cloud parts closest to the Sun is used to predict when the clouds will cover the Sun. WiFi communication is used to transmit this data to the PV power plant control system in order to decrease the output power slowly and smoothly.