Method and device for detecting direct current arc discharge of energy storage system

Abstract

The invention discloses a direct current arc discharge detection method and device of an energy storage system. The direct current arc discharge detection method of the energy storage system comprises the following steps: collecting a bus current signal and a bus voltage signal of the energy storage system; carrying out adaptive wavelet noise reduction and feature extraction on the collected bus current signal to obtain a current feature signal; performing harmonic resonance analysis and feature extraction on the acquired bus voltage signal to obtain a voltage feature signal; inputting the extracted current characteristic signal and the voltage characteristic signal into a WCNN model, and outputting an arc discharge probability value; and when the arc discharge probability value meets a set protection condition, it is confirmed that a direct current arc discharge fault occurs, and the direct current arc discharge fault is used for triggering opening protection. According to the invention, the accuracy and reliability of direct current arc discharge detection are significantly improved.

Description

A method and device for detecting DC arcing in an energy storage system Technical Field

[0001] This invention belongs to the field of high voltage DC protection technology, specifically relating to a DC arc detection method and device for energy storage systems. Background Technology

[0002] In the field of high voltage DC protection technology, the DC arc detection of high voltage energy storage system has the following defects: (1) Defects in the reliability of sensing scheme, including: 1) Magnetic saturation risk, traditional electromagnetic sensors have a significant risk of magnetic saturation in high voltage scenarios of ≥1500V, resulting in current sampling distortion; 2) Excessive response delay, existing isolation sampling schemes have a detection delay of >10ms due to the length of the signal chain, which cannot meet the millisecond-level arc response requirements; (2) Insufficient noise interference suppression, including: 1) High frequency noise pollution, during the converter switching process (especially SiC device 150k (1) Hz ripple) strongly interferes with the characteristics of high-frequency electric arc, causing the effective signal to be buried; 2) The false alarm rate is too high. The mainstream detection scheme has not established a dynamic noise filtering mechanism, and the actual false alarm rate is >1.5%, which significantly affects the availability of the system; (3) The disconnection safety mechanism is missing, including: 1) the risk of secondary arcing. The high-voltage arc disconnection process lacks energy gradient release design, and the probability of arc reignition is >34% (measured data); 2) the disconnection stress is out of control. The current change rate di / dt of the existing scheme during the disconnection process is >1000A / μs, which causes dangerous overvoltage and threatens the equipment insulation. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method and apparatus for detecting DC arcing in energy storage systems, which significantly improves the accuracy and reliability of DC arcing detection.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] Firstly, a method for detecting DC arcing in an energy storage system is provided, comprising: acquiring bus current signals and bus voltage signals of the energy storage system; performing adaptive wavelet denoising and feature extraction on the acquired bus current signals to obtain current feature signals; performing harmonic resonance analysis and feature extraction on the acquired bus voltage signals to obtain voltage feature signals; inputting the extracted current feature signals and voltage feature signals into a WCNN model to output an arcing probability value; and confirming the occurrence of a DC arcing fault when the arcing probability value meets the set protection conditions, thereby triggering trip protection.

[0006] Furthermore, the current characteristic signal includes: high-frequency subband energy mutation rate and time-domain waveform kurtosis coefficient.

[0007] Furthermore, the high-frequency subband energy mutation rate is calculated using the following formula:

[0008] ;

[0009] in, This represents the energy mutation rate of high-frequency subbands. This indicates the energy value of the high-frequency sub-band within a specified frequency band. This represents the total high-frequency energy within the specified frequency band. This indicates the maximum value of the change in current.

[0010] Furthermore, the voltage characteristic signal includes: calculating the harmonic distortion rate to the high-frequency noise energy ratio.

[0011] Furthermore, the WCNN model includes: a wavelet preprocessing layer, used to perform db4 wavelet decomposition on the input current feature signal and voltage feature signal to obtain a first preprocessed signal reflecting the signal trend and arc characteristics of the input signal; a dual-channel convolutional network, used to process the first preprocessed signal in parallel through time-domain and frequency-domain channels to obtain a convolutional processing signal reflecting time-domain waveform abrupt changes and time correlation, as well as reflecting high-frequency energy accumulation areas and harmonic resonance modes; and a feature fusion layer, used to perform feature fusion on the convolutional processing signal, dynamically weighting time-frequency features through a gated attention mechanism to obtain the arcing probability value.

[0012] Furthermore, the dynamically weighted time-frequency features include: calculating the attention weights for the time-domain feature map and the frequency-domain feature map.

[0013] ;

[0014] in, This represents the Sigmoid function, ⊕ indicates concatenation, and W represents trainable weights. Represents the time-domain feature map; Represents the frequency domain feature map; This indicates the "importance ratio" of frequency domain features relative to time domain features;

[0015] Perform feature-weighted fusion:

[0016] ;

[0017] in, This represents the weighted fusion of features from the time-domain feature map and the frequency-domain feature map.

[0018] Furthermore, the protection condition is set as follows: the arcing probability value exceeds the dynamic threshold three times consecutively.

[0019] Furthermore, the trip protection includes: Level 1 protection, pre-charge resistor switching start, used to limit the peak fault current and suppress the initial arc energy; Level 2 protection, IGBT soft turn-off start, used to control the current change rate and prevent voltage spikes; Level 3 protection, vacuum contactor trip start, used to physically isolate the fault circuit.

[0020] Secondly, a DC arcing detection device for an energy storage system is provided, comprising: a data acquisition module for acquiring bus current signals and bus voltage signals of the energy storage system; a current feature extraction module for performing adaptive wavelet noise reduction and feature extraction on the acquired bus current signals to obtain current feature signals; a voltage feature extraction module for performing harmonic resonance analysis and feature extraction on the acquired bus voltage signals to obtain voltage feature signals; a probability calculation module for inputting the extracted current feature signals and voltage feature signals into a WCNN model and outputting an arcing probability value; and an arcing judgment module for confirming the occurrence of a DC arcing fault when the arcing probability value meets the set protection conditions, and for triggering trip protection.

[0021] Furthermore, the data acquisition module acquires the bus voltage signal through a magneto-optical current sensor and through a high-voltage differential probe.

[0022] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention obtains current characteristic signals by performing adaptive wavelet denoising and feature extraction on the collected bus current signals; it obtains voltage characteristic signals by performing harmonic resonance analysis and feature extraction on the collected bus voltage signals; the extracted current and voltage characteristic signals are input into a WCNN model to output arcing probability values; when the arcing probability values ​​meet the set protection conditions, the trip protection is triggered, which significantly improves the accuracy and reliability of DC arcing detection, realizes rapid and reliable disconnection of DC arcing faults, and provides core technical support for the safe operation of new power systems. Attached Figure Description

[0023] Figure 1 is a schematic diagram of the system hardware architecture of a DC arc detection device for an energy storage system provided in an embodiment of the present invention;

[0024] Figure 2 is a structural diagram of the WCNN model in an embodiment of the present invention;

[0025] Figure 3 is a timing diagram of the three-stage tripping operation in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0027] Example 1

[0028] A method for detecting DC arcing in an energy storage system includes: acquiring bus current and bus voltage signals of the energy storage system; performing adaptive wavelet denoising and feature extraction on the acquired bus current signals to obtain current characteristic signals; performing harmonic resonance analysis and feature extraction on the acquired bus voltage signals to obtain voltage characteristic signals; inputting the extracted current and voltage characteristic signals into a WCNN model to output an arcing probability value; and confirming a DC arcing fault when the arcing probability value meets the set protection conditions, thereby triggering trip protection.

[0029] As shown in Figures 1 to 3, this invention takes the application of the DC arc detection method in a 1500V / 2MWh energy storage system as an example for further explanation.

[0030] Hardware configuration: (1) Magneto-optical current sensor: HIOKI CT7060 (bandwidth DC-15MHz, accuracy ±0.5%) adopts dual optical path differential compensation design, temperature drift coefficient ≤5ppm / ℃; (2) Edge computing unit: deploys quantized WCNN model (INT8 accuracy, model size <5MB).

[0031] 1. Collect the bus current signal and bus voltage signal of the energy storage system.

[0032] S1. Acquire positive and negative bus current signals using a magneto-optical current sensor (withstand voltage ≥1500V, bandwidth ≥10MHz).

[0033] A magneto-optical current sensor with a withstand voltage ≥1500V was selected to meet the insulation requirements of high-voltage environments; a bandwidth ≥10MHz was chosen to capture the high-frequency transient current generated by the electric arc, with a typical arc spectrum in the range of 100kHz to 10MHz. The magneto-optical current sensor utilizes the Faraday magneto-optical effect to measure the bus current non-contactly, avoiding high-voltage isolation issues.

[0034] This invention uses 10MHz current + voltage high-frequency analysis to break through the traditional detection bandwidth limitation (conventional methods ≤100kHz), and can accurately capture microsecond-level electric arcs.

[0035] S2. A high-voltage differential probe (common-mode rejection ratio ≥140dB@1MHz) is used to synchronously acquire the bus voltage signal, thereby obtaining the bus voltage fluctuation signal. The voltage probe has a built-in RC integration network to suppress switching noise above 150kHz.

[0036] A high-voltage differential probe with a common-mode rejection ratio (CMRR) ≥140dB@1MHz is selected to obtain strong anti-interference capability and accurately extract minute voltage fluctuations. The high-voltage differential probe synchronously captures sudden drops or oscillations in bus voltage (voltage anomalies caused by impedance changes due to electric arcing).

[0037] The edge computing unit of the quantized WCNN model is deployed and mainly performs the following operations: adaptive wavelet denoising and feature extraction on the collected bus current signal to obtain the current feature signal; harmonic resonance analysis and feature extraction on the collected bus voltage signal to obtain the voltage feature signal; input the extracted current feature signal and voltage feature signal into the WCNN model and output the arcing probability value.

[0038] S3. Perform adaptive wavelet denoising on the current signal to extract the energy mutation rate and time-domain waveform kurtosis coefficient of the high-frequency subband (1-5MHz).

[0039] Adaptive wavelet denoising can filter out environmental noise (switching noise) while preserving the characteristics of high-frequency electric arcs. The instantaneous energy release of an electric arc surges in the 1-5MHz frequency band, and the energy mutation rate of the high-frequency subband (1-5MHz) is used to reflect this characteristic of the electric arc.

[0040] The formula for calculating the energy mutation rate of high-frequency subbands is as follows:

[0041] ;

[0042] in, This represents the energy mutation rate of high-frequency subbands. This indicates the energy value of the high-frequency sub-band within a specified frequency band. In this invention, [the energy value is...]. Focusing on the energy value of a specific high-frequency sub-band around 5MHz, arcing will produce characteristic high-frequency electromagnetic radiation. This frequency band is highly sensitive to the arcing phenomenon and is the key frequency band for detection. It represents the total high-frequency energy within a set frequency band, serving as a "normalization factor" to eliminate the interference of overall high-frequency energy fluctuations in the system on the detection, making the indicators more targeted; This represents the maximum value of the current change. During arcing, the electrical circuit parameters change abruptly, and the "rapid change" of current is a typical electrical characteristic of arcing. Taking the "maximum value" can highlight the most severe degree of the change and enhance the distinguishability of the characteristics.

[0043] The high-frequency subband energy mutation rate reflects the relative mutation level of high-frequency energy by measuring the ratio of a specific high-frequency subband (around 5MHz) to the total energy. This, combined with the maximum rate of change of current, strengthens the correlation between "electrical parameter mutations"—during arcing, the proportion of high-frequency radiated energy in the characteristic frequency band and the current mutation rate both differ significantly from normal operating conditions. This formula, through the fusion calculation of multi-dimensional features, accurately characterizes the high-frequency energy mutation characteristics caused by arcing, providing a quantifiable indicator for subsequent identification of "whether arcing exists."

[0044] The arc current is pulsed, resulting in a significantly increased kurtosis value. Therefore, the kurtosis coefficient of the time-domain waveform is used to quantify the sharpness of the current waveform. The calculation of the kurtosis coefficient of the time-domain waveform is prior art and will not be described in detail in this application.

[0045] S4. Perform harmonic resonance analysis on the voltage signal and calculate the third harmonic distortion rate to the high-frequency noise energy ratio.

[0046] The third harmonic is sensitive to the electric arc, so the harmonic distortion caused by the nonlinear characteristics of the electric arc can be reflected by the third harmonic distortion rate.

[0047] Based on the characteristic that the wideband electromagnetic radiation generated by the ionization of gas by the electric arc is superimposed on the voltage signal, the high-frequency noise energy ratio is calculated.

[0048] The calculation of the third harmonic distortion rate and the high-frequency noise energy ratio is prior art and will not be elaborated here.

[0049] S5. Input the current and voltage features into the WCNN model (Wavelet Convolutional Neural Network, a deep learning model that combines wavelet transform and traditional convolutional neural network CNN), and output the arc probability value.

[0050] The WCNN model is a CNN that incorporates wavelet transform, giving it a stronger ability to process high-frequency features. It includes:

[0051] The wavelet preprocessing layer is used to perform DB4 wavelet decomposition on the input current and voltage characteristic signals to obtain the first preprocessed signal that reflects the signal trend and arc characteristics of the input signal. The wavelet preprocessing layer uses the Mallat algorithm (fast discrete wavelet transform) to implement DB4 wavelet (Daubechies 4th order wavelet) decomposition and extracts high-frequency detail coefficients of 1.25-5MHz.

[0052] The wavelet preprocessing layer performs multi-scale decomposition on the input current / voltage signal, extracting low-frequency profiles (signal trends) and high-frequency details (arc features), providing high-quality time-frequency input for subsequent dual-channel processing. It exhibits high regularity and is suitable for capturing the non-stationary signal characteristics (current pulses, voltage spikes) of transient arcs.

[0053] As shown in Table 1, a dual-channel convolutional network is used to process the first preprocessed signal in parallel through a time-domain channel (1D-CNN) and a frequency-domain channel (2D-CNN) to obtain a convolutional signal that reflects the temporal waveform abrupt changes and time correlation, as well as the high-frequency energy accumulation region and harmonic resonance mode.

[0054] Table 1: Dual-channel convolutional networks

[0055]

[0056] The dual-channel parallel processing of time-domain and frequency-domain features avoids information loss in traditional single-channel models and improves sensitivity to multi-scale arc features.

[0057] The feature fusion layer is used to fuse features of the convolutional signal. It dynamically weights time-frequency features through a gating attention mechanism to obtain the arcing probability value.

[0058] The core mechanism of the feature fusion layer is the gated attention mechanism, and its operation process includes:

[0059] Dynamic weight allocation, calculating the attention weights for the time-domain feature map and the frequency-domain feature map:

[0060] ;

[0061] in, This represents the Sigmoid function, ⊕ indicates concatenation, and W represents trainable weights. Represents the time-domain feature map; Represents the frequency domain feature map; This indicates the "importance ratio" of frequency domain features relative to time domain features;

[0062] Perform feature-weighted fusion:

[0063] ;

[0064] in, This represents the weighted fusion of features from the time-domain feature map and the frequency-domain feature map.

[0065] The feature fusion layer can adaptively enhance strong features that are sensitive to electric arcs (emphasizing the frequency domain channel during high-frequency bursts and the time domain channel during steady-state distortions) and suppress weak feature responses caused by noise interference.

[0066] Conventional CNNs cannot directly process high-frequency transient signals. This invention uses WCNN to achieve dynamic response to microsecond-level electric arcs through wavelet pre-decomposition and a dual-channel structure, breaking through the bottleneck of traditional technology. The gated attention mechanism enables the model to maintain a decision accuracy of >99.2% (actual data) even under complex working conditions (such as motor starting interference).

[0067] This invention supports deployment in embedded systems (such as energy storage converter controllers) and meets real-time protection requirements of ≤20ms. The architecture of this invention provides an end-to-end solution for 1500V high-voltage energy storage systems, from feature extraction to intelligent decision-making, significantly improving the reliability of DC arcing detection.

[0068] The input features of the WCNN model are the four-dimensional feature vectors extracted by S3 and S4 (two current terms + two voltage terms). The output of the WCNN model is the arcing probability value (0~1), which is used to quantify the probability of failure.

[0069] This invention uses a dual-channel verification method of high-frequency current surge and voltage harmonic distortion to avoid misjudgment of a single signal (such as switching surge).

[0070] When the arcing probability value meets the set protection conditions, a DC arcing fault is confirmed, and the trip protection is triggered.

[0071] S6. Dynamic threshold protection mechanism: When the arcing probability value exceeds the dynamic threshold three times in a row, non-instantaneous interference is confirmed, and three-level trip protection is triggered.

[0072] Level 1 protection, with pre-charge resistor switching start, is used to limit the peak fault current and suppress the initial arc energy; the pre-charge resistor limits the inrush current to below 20% of the rated value.

[0073] Secondary protection, IGBT soft turn-off start-up, IGBT soft turn-off stage injects reverse current to accelerate arc suppression, used to control the current change rate and prevent voltage spikes;

[0074] Level 3 protection, vacuum contactor tripping start, used to physically isolate faulty circuits.

[0075] The dynamic threshold automatically adjusts its sensitivity based on system operating conditions (load current) to reduce false alarm rate. The actuator unit performs three-level trip protection actions, disconnecting fault circuits in stages to ensure safety redundancy.

[0076] Dynamic threshold calculation:

[0077] ;

[0078] in, The threshold is dynamic, and SOC is the state of charge of the battery.

[0079] The dynamic threshold protection mechanism can avoid sensitivity imbalance of fixed thresholds under light / heavy load conditions, thus improving reliability.

[0080] This invention uses a three-step judgment mechanism to prevent false triggering, and the three-level tripping action ensures complete fault isolation.

[0081] This invention achieved a 100% success rate in 30 artificial arc tests; the total energy of the circuit breaker tripping is <500J, meeting the IEC60947-2 high-voltage breaking standard; it can withstand continuous operation at 1500V DC, and the insulation impedance of the sampling circuit is >1GΩ; under 150kHz PWM interference, the feature extraction signal-to-noise ratio is improved to 35dB; the overall power consumption is <10W, and it supports wide temperature range operation from -40℃ to +85℃.

[0082] This invention is applicable to 1500V high-voltage DC systems such as photovoltaic energy storage power stations, electric vehicle charging piles, and hydrogen energy storage and transportation, solving the problem of DC arc protection in scenarios such as parallel battery clusters and DC switchgear. Through a closed-loop technology system of high-bandwidth sensing → multi-dimensional features → AI fusion decision-making → dynamic protection, this invention achieves a breakthrough in the accuracy and reliability of arc detection in high-voltage DC systems, providing crucial protection for the safety of high-voltage energy storage.

[0083] Example 2

[0084] Based on the DC arcing detection method for an energy storage system described in Embodiment 1, this embodiment provides a DC arcing detection device for an energy storage system, comprising: a data acquisition module for acquiring bus current signals and bus voltage signals of the energy storage system; a current feature extraction module for performing adaptive wavelet noise reduction and feature extraction on the acquired bus current signals to obtain current feature signals; a voltage feature extraction module for performing harmonic resonance analysis and feature extraction on the acquired bus voltage signals to obtain voltage feature signals; a probability calculation module for inputting the extracted current feature signals and voltage feature signals into a WCNN model and outputting an arcing probability value; and an arcing judgment module for confirming a DC arcing fault when the arcing probability value meets the set protection conditions, and triggering trip protection.

[0085] As shown in Figure 1, the magneto-optical sensor unit includes a Faraday rotator and a closed-loop temperature control system. Its linearity error is ≤±0.2%, eliminating the interference of ambient temperature on the magneto-optical effect. It can be configured with a data acquisition module to acquire bus voltage signals via a magneto-optical current sensor and a high-voltage differential probe, and transmit these signals to the signal conditioning unit. The magneto-optical sensor unit utilizes the Faraday magneto-optical effect to measure DC current non-contactly, avoiding the risks of direct high-voltage connection, and is suitable for 1500V systems.

[0086] The signal conditioning unit integrates a Xilinx Zynq UltraScale+ MPSoC (FPGA+ARM architecture), and can be configured with current feature extraction, voltage feature extraction, probability calculation, and arcing detection modules to perform edge computing on the input signal. It analyzes current / voltage characteristics in real time and outputs the arcing probability determination result. The edge computing latency is <5ms, meeting millisecond-level protection requirements.

[0087] After triggering the trip protection, the actuator unit performs a three-level trip protection action, as shown in Figure 2.

[0088] Table 2: Three-level tripping protection

[0089]

[0090] This invention employs a three-level tripping mechanism, forming multiple safeguards of "electrical limitation → soft shutdown → physical isolation" to mitigate the risk of single-action failure. The total fault clearing time is ≤10ms (from arc occurrence to complete disconnection). IGBT soft shutdown reduces di / dt to 1 / 10 of traditional solutions, significantly lowering shutdown overvoltage. It is suitable for 1500V DC systems such as photovoltaic energy storage power stations, electric vehicle supercharging piles, and high-voltage DC power distribution in data centers, addressing arc protection needs in high-risk scenarios such as parallel battery clusters and DC bus cabinets. Through a hardware-software collaborative design of high-precision sensing, intelligent decision-making, and hierarchical execution, this invention achieves rapid and reliable disconnection of DC arcing faults, providing core technical support for the safe operation of new power systems.

[0091] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting DC arcing in an energy storage system, characterized in that, include: The system collects bus current and bus voltage signals from the energy storage system; it then performs adaptive wavelet denoising and feature extraction on the collected bus current signals to obtain current feature signals. Harmonic resonance analysis and feature extraction are performed on the collected bus voltage signal to obtain voltage feature signals; the extracted current feature signals and voltage feature signals are input into the WCNN model to output the arcing probability value; when the arcing probability value meets the set protection conditions, a DC arcing fault is confirmed to have occurred, which is used to trigger the trip protection.

2. The DC arcing detection method for an energy storage system according to claim 1, characterized in that, The current characteristic signal includes: high-frequency subband energy mutation rate and time-domain waveform kurtosis coefficient.

3. The DC arcing detection method for an energy storage system according to claim 2, characterized in that, The high-frequency subband energy mutation rate is calculated using the following formula: ;in, This represents the energy mutation rate of high-frequency subbands. This indicates the energy value of the high-frequency sub-band within a specified frequency band. This represents the total high-frequency energy within the specified frequency band. This indicates the maximum value of the change in current.

4. The DC arcing detection method for an energy storage system according to claim 1, characterized in that, The voltage characteristic signal includes: calculating the harmonic distortion rate to the high-frequency noise energy ratio.

5. The DC arcing detection method for an energy storage system according to claim 1, characterized in that, The WCNN model includes: a wavelet preprocessing layer, used to perform db4 wavelet decomposition on the input current and voltage feature signals to obtain a first preprocessed signal reflecting the signal trend and arc characteristics of the input signal; a dual-channel convolutional network, used to process the first preprocessed signal in parallel through time and frequency channels to obtain a convolutional processing signal reflecting abrupt changes and time correlations in the time-domain waveform, as well as reflecting the energy accumulation area and harmonic resonance mode in the high-frequency band; and a feature fusion layer, used to perform feature fusion on the convolutional processing signal, dynamically weighting the time-frequency features through a gated attention mechanism to obtain the arcing probability value.

6. The DC arcing detection method for an energy storage system according to claim 5, characterized in that, Dynamically weighted time-frequency features include: calculating the attention weights for the time-domain feature map and the frequency-domain feature map. ;in, This represents the Sigmoid function, ⊕ indicates concatenation, and W represents trainable weights. Represents the time-domain feature map; Represents the frequency domain feature map; This represents the "importance ratio" of frequency domain features relative to time domain features; feature weighting and fusion are then performed. ;in, This represents the weighted fusion of features from the time-domain feature map and the frequency-domain feature map.

7. The DC arcing detection method for an energy storage system according to claim 1, characterized in that, The protection condition is set as follows: the arcing probability value exceeds the dynamic threshold three times consecutively.

8. The DC arcing detection method for an energy storage system according to claim 1, characterized in that, The tripping protection includes: Level 1 protection, pre-charge resistor switching start, used to limit the peak fault current and suppress the initial arc energy; Level 2 protection, IGBT soft turn-off start, used to control the current change rate and prevent voltage spikes; Level 3 protection, vacuum contactor tripping start, used to physically isolate the fault circuit.

9. A DC arc detection device for an energy storage system, characterized in that, include: The data acquisition module is used to acquire the bus current signal and bus voltage signal of the energy storage system; The current feature extraction module is used to perform adaptive wavelet noise reduction and feature extraction on the acquired bus current signal to obtain the current feature signal; the voltage feature extraction module is used to perform harmonic resonance analysis and feature extraction on the acquired bus voltage signal to obtain the voltage feature signal. The probability calculation module is used to input the extracted current and voltage feature signals into the WCNN model and output the arcing probability value. The arcing detection module is used to confirm the occurrence of a DC arcing fault when the arcing probability value meets the set protection conditions, and to trigger the trip protection.

10. The DC arcing detection device for an energy storage system according to claim 9, characterized in that, The data acquisition module acquires the bus voltage signal through a magneto-optical current sensor and a high-voltage differential probe.

Images