Search

Current transformer saturation is a key issue for power systems because it negatively affects the operation of relays, resulting in the malfunction of protection devices. Recently, deep learning has been used in many academic fields with promising results. Here, we present a technique to compensate for saturated waveforms using deep learning to reconstruct an undistorted waveform. The optimal structure was obtained using pre- and fine-tuning mechanisms, which yielded good performance and initialized the optimum weight during the pre-training stage. The deep learning parameters were determined using particle swarm optimization prior to training. Finally, deep learning performance was evaluated using newly introduced conditions that were not observed during the training stage.

The amount of electrical consumption rises gradually every year which puts an enormous strain on electrical power systems and electrical protective systems. Dealing with this issue, automatic protective systems have been developed relentlessly; however, DC-offset in fault current is still the main issue to solve since it makes many protective devices operate inaccurately which lead to both unreliable and unsecured power systems. Modifying Discrete Fourier Transform (DFT) in phasor domain is quite well known for solving DC-offset, however the drawback of this method is that more than one cycle waveform is essential to complete the task. This paper proposed Deep Neural Network method to estimate fault current phase angle using only 90.625% of a cycle waveform during DC Offset phenomenon. With enough intensive DNN training, the DNN algorithm manage to show an accurate result while using less input signal which could cut down the operating time. Not only DC-offset signal, but the results show that DNN is also robust against DC-offset with noise and harmonics.

Current Transformer (CT) saturation is regarded as one of the major problems in power system field due to the reason that it negatively impacts the operation of relays, resulting in malfunction protective devices. Recently, deep learning methods have been commonly implemented in most academic fields as the reason of significant generated results. This paper presents a compensation method for saturated waveform by applying deep learning to the aforementioned problem. To achieve a good network structure, pre-training and fine-tuning mechanism have been implemented because it shows a great performance as it well initializes the optimal weight in the pre-training stage. Finally, a training model is evaluated by the newly-introduced conditions, in which has never been experienced during the training stage.