[ad_1]
Chinese scientists have developed a new system that can detect DC arcs in solar power installations through a backpropagation neural network. This new technology reportedly guarantees detection times of less than 200 ms.
Researchers from China’s Tianjin University and inverter manufacturer Ginlong Solis have developed a new AI-based method for DC arc detection in photovoltaic systems.
When a discharge occurs between conductive parts on this side of the system, a DC arc occurs in the PV array. The high energy and temperature involved in this process ionizes the surrounding gas, creating a plasma path for the current to continue flowing. “PV DC arc faults can cause fires, damage property and endanger lives,” the group said. “Many countries have established DC arc detection standards and defined DC arc fault protection in view of the dire consequences of DC arcs.”
This new technique is based on fast Fourier transform (FFT) algorithms and backpropagation neural network (BPNN) machine learning estimation methods. FFT is a commonly used algorithm for signal analysis and was used in this study to transform the PV current into the frequency domain. BPNN is a machine learning training method that modifies itself by changing the weights given to different layers of the understanding mechanism.
In the first step of the proposed method, a digital signal processor (DSP) unit samples the AC component of the PV side current up to several hundred points. Then, use the FFT algorithm to obtain the frequency domain results and remove the components in the low frequency range (below 41 kHz) and the high frequency range (above 102.5 kHz).
In the next step, the remaining samples are divided into eight groups and fed into the BPNN analysis. Using these eight groups, the AI model processes the data and determines whether an arc exists. If present, sends a command to the PV end DC/DC to interrupt the arc.
“The arc interruption process can be performed locally, eliminating the need for decisions made by the inverter level layer or the cloud layer,” they added. “As a result, the reliability of the system is relatively high. The disadvantage is that the algorithm is mainly implemented at the PV end layer. Moreover, the training data is stored locally, so the local unit has no data storage. requires more physical memory.”
AI models are trained in two ways: online and offline. When offline, the training data collected is input into algorithms within the computer to upgrade the capabilities of the DSP control software. If offline-based recognition fails, online training is initiated. If it fails, the recorded data from the detection system is sent to the cloud, where the AI is further trained. It then sends the results back to the physical unit.
“Using FFT and AI analysis, arcs can be identified instantly, with an arc detection time of less than 200 milliseconds when an arc occurs,” the scientists said.
To test the detection method, they built a test platform. They connected an arc generator, a device that can generate an arc for the AI to identify, between the platform’s PV and inverter.
They found that “the training results closely match the actual values.” “Furthermore, 40 sets of test data are used to validate the trained model. There is a good agreement between the test results and the created model. The success rate is 97.5%, with only 40 falsely identified. There is only one set in the set.The arc detection time when an arc occurs is less than 200ms.
This new approach is presented in “Methods for DC Arc Detection in Photovoltaic (PV) Systems.” engineering achievements.
This content is copyrighted and may not be reused. If you would like to collaborate with us and reuse some of our content, please contact us at [email protected].
[ad_2]
Source link