KR-20260067028-A - Artificial intelligence-based real-time serial arc fault detection method

Abstract

The artificial intelligence-based real-time serial arc fault detection method of the present invention is characterized by comprising the steps of: extracting normal current data and fault current data at a point where an arc fault may occur in a power converter connected to a power storage device, and performing data preprocessing using a pre-selected scaler; training the preprocessed data to a predetermined artificial intelligence algorithm; uploading the trained model to a digital signal processing device (DSP), and then detecting a DC serial arc fault at the said point in real time.

Inventors

• 박화평
• 전민서

Assignees

• 국립금오공과대학교 산학협력단

Dates

Publication Date

20260512

Application Date

20241105

Claims (5)

1. A step of extracting normal current data and fault current data at a point where an arc fault may occur in a power converter connected to a power storage device, and performing data preprocessing using a pre-selected scaler; A step of training preprocessed data into a predetermined artificial intelligence algorithm; and A step of uploading a trained model to a digital signal processing device (DSP) and then detecting a DC serial arc fault at the above point in real time; A real-time serial arc fault detection method including
2. In paragraph 1, The above scaler is, A real-time serial arc fault detection method characterized by being one of a standard scaler, a min-max scaler, and a mean scaler.
3. In paragraph 1, The above-mentioned predetermined artificial intelligence algorithm is A real-time serial arc fault detection method characterized by including a CNN (Convolutional Neural Network) layer, wherein the CNN (Convolutional Neural Network) layer classifies two states: a normal state and an arc state.
4. A step of extracting normal current data and fault current data at a point where an arc fault may occur in a power converter connected to a power storage device; A step of training a predetermined artificial intelligence algorithm with normal current data and fault current data; and The method includes the step of uploading a learned model to a digital signal processing device (DSP) and then detecting a DC serial arc fault at the above point in real time. A real-time serial arc fault detection method characterized by the above artificial intelligence algorithm being configured such that the preprocessed values of the mean, standard deviation, minimum value, and maximum value scaler of the input data are connected to the layers and pipelines of the artificial intelligence model and included in the learning weights of the artificial intelligence.
5. In paragraph 4, The above-mentioned predetermined artificial intelligence algorithm is A real-time serial arc fault detection method characterized by including a CNN (Convolutional Neural Network) layer, wherein the CNN (Convolutional Neural Network) layer classifies two states: a normal state and an arc state.

Description

Artificial intelligence-based real-time serial arc fault detection method The present invention relates to a serial arc fault detection method, and more specifically, to an artificial intelligence-based real-time serial arc fault detection method. As the use of renewable energy, particularly photovoltaic (PV) energy, increases, the safety of photovoltaic systems is emerging as an important issue. Photovoltaic (PV) systems and energy storage systems (ESS) use DC power, and arc failures can occur due to loose connectors, damaged cables, etc. Parallel arc faults can be easily detected due to rapid changes in current, but series arc faults are difficult to detect because the changes in current are minimal, so they are emerging as a major problem in the use of photovoltaic (PV) systems and energy storage (ESS) systems. To detect DC series arc faults, various techniques such as FFT, wavelets, and statistics are used to determine fault and normal states. Figure 1 is a diagram showing an FFT size comparison algorithm for fault detection. In existing DSPs and FPGAs, DC serial arc fault detection devices detect faults by extracting AC information of the DC current after high-speed sampling at an average of 100 KS/s as shown in Fig. 1, and using techniques such as FFT and wavelet. However, since these conventional technologies require complex signal processing, high-performance digital signal processing (DSP) units are essential. Additionally, they have the disadvantage of being vulnerable to external noise and switching noise. Figure 1 is a diagram showing an FFT size comparison algorithm for fault detection. Figure 2 is an example of 1KS/s normal data preprocessed with min/max. FIG. 3 is a diagram showing the process of detecting an arc fault in the arc fault detection device of the present invention. FIG. 4 is a diagram showing a real-time serial arc fault detection method according to a first embodiment of the present invention. FIG. 5 is a diagram showing a real-time serial arc fault detection method according to a second embodiment of the present invention. Hereinafter, in order to explain in detail enough for a person skilled in the art to easily implement the technical concept of the present invention, embodiments of the present invention will be described with reference to the attached drawings. FIG. 2 is an example of 1KS/s normal data preprocessed with min/max, FIG. 3 is a diagram showing the process of detecting an arc fault in the arc fault detection device of the present invention, FIG. 4 is a diagram showing a real-time serial arc fault detection method according to the first embodiment of the present invention, and FIG. 5 is a diagram showing a real-time serial arc fault detection method according to the second embodiment of the present invention. Referring to FIGS. 2 to 5, the real-time serial arc fault detection method of the arc fault detection device according to the present embodiment is performed as follows. First of all, unlike conventional methods, the present invention uses an artificial intelligence model (ML/DL Model), so it only requires enough data for the model to accurately classify normal and faulty conditions, and thus can sufficiently detect normal and faulty conditions with 1 KS/s. That is, the present invention detects DC series arc faults by extracting normal and fault current data at a point where a DC series arc fault may occur, selecting scalers such as Standard, Min-Max, and Mean to preprocess the data, training an ML/DL model, and then uploading the trained model to a digital signal processing device (DSP). The arc fault detection device of the present invention can detect faults and normal conditions by extracting normal and fault data, training an ML/DL model using Standard and Min-Max Scaler, and uploading it to a DSP. By extracting normal and fault data containing external noise and training the artificial intelligence model (Machine/Deep Learning Model), the weakness of the digital signal processing unit (DSP) being vulnerable to noise can be compensated for. In addition, while existing digital signal processing devices (DSPs) and FPGAs use methods such as high-speed sampling, FFT, and Wavelet, which result in high complexity, using an AI model as in the arc fault detection device of the present invention eliminates the need for high-speed sampling and allows for the use of preprocessing steps such as Standard and Min-Max, which are general preprocessing steps for artificial intelligence (AI), thereby maintaining simplicity compared to existing FFT or Wavelet statistical techniques. FIG. 4 is a diagram showing a real-time serial arc fault detection method according to a first embodiment of the present invention. Referring to FIG. 4, first, normal current data and fault current data at a point where an arc fault may occur in a power converter connected to a power storage device are extracted, and data preprocessing is performed using a pre-selected scaler (S12, S14). Next,