Photovoltaic power generation cabinet with leakage protection function and working method thereof

Abstract

The invention discloses a photovoltaic power generation cabinet with a leakage protection function and a working method thereof, and relates to the technical field of leakage protection. Comprising the following steps that a leakage current dynamic reference model is established based on photovoltaic system operation environment parameters and historical data, and the environment parameters at least comprise illumination intensity, environment temperature and component open-circuit voltage; time domain, frequency domain and time-frequency domain joint analysis is carried out based on direct current side and alternating current side leakage current information, and multi-dimensional feature vectors including a waveform distortion rate, a high-frequency harmonic energy ratio and a transient energy entropy value are extracted; according to the method, the leakage current dynamic reference model based on the environmental parameters is constructed, the electric leakage risk can be accurately evaluated in real time in the operation of the photovoltaic system by means of time-frequency domain conjoint analysis and multi-dimensional feature vector extraction, differentiated protection strategies are automatically executed according to risk scores, and in order to improve the detection accuracy, the method is simple and convenient to operate. The system adopts self-adaptive notch filtering and FIR low-pass filtering, so that high precision can be kept even in extreme environments such as lightning surge and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of leakage protection, and particularly to a photovoltaic power generation cabinet with a leakage protection function and its working method. Background Technique

[0002] When a photovoltaic power generation system is working, since the current input by photovoltaic power generation is usually uncertain and affected by various environmental factors, including light intensity, environmental temperature, and open-circuit voltage of components, etc., the leakage protection of the system is particularly important.

[0003] Traditional leakage protection devices usually use fixed thresholds for protection. However, in actual applications, due to changes in environmental parameters, this static method is very likely to cause leakage abnormalities that cannot be detected and processed in a timely manner, thus increasing the risk of system instability.

[0004] To sum up, it is obvious that the existing technology has inconveniences and defects in actual use, so it is necessary to improve. Summary of the Invention

[0005] The purpose of the present invention is to provide a photovoltaic power generation cabinet with a leakage protection function and its working method to solve the problems raised in the above background technique.

[0006] To achieve the above purpose, the present invention provides the following technical solution: A photovoltaic power generation cabinet with a leakage protection function and its working method, including the following steps:

[0007] S11: Based on the operating environmental parameters and historical data of the photovoltaic system, establish a dynamic reference model for leakage current, where the environmental parameters at least include light intensity, environmental temperature, and open-circuit voltage of components;

[0008] S12: Based on the leakage current information on the DC side and the AC side, conduct joint analysis in the time domain, frequency domain, and time-frequency domain, and extract a multi-dimensional feature vector including waveform distortion rate, high-frequency harmonic energy ratio, and transient energy entropy value;

[0009] S13: By fusing the multi-dimensional feature vector and the change amount of environmental parameters, calculate the real-time risk score, and the risk assessment calculation satisfies: where α, β, γ are dynamic weight coefficients, I meas is the measured current, I base is the reference current, E hf is the high-frequency energy, Δ Rins is the insulation resistance change rate;

[0010] S14: Implement a differential protection strategy based on the risk score level, including:

[0011] When S(t) ∈ [0.7, 0.9), start local insulation monitoring and issue a warning;

[0012] When S(t) ≥ 0.9 or dI / dt > 50 mA / ms, cut off the corresponding photovoltaic module string circuit;

[0013] When the high-frequency arc characteristics are detected, perform an emergency power-off for the entire system.

[0014] As a specific solution of the technical solution of the present application, the method for establishing the dynamic reference model includes the following steps:

[0015] Based on the system initialization operation stage, collect environmental parameters and leakage current data for at least 24 hours;

[0016] Obtain the functional relationship between the reference current and the environmental parameters by least squares fitting;

[0017] I base = Ki·G + K2·(T - 25) + K3·In(Voc / Vrated), where G is the light intensity, T is the environmental temperature, Voc is the open-circuit voltage of the component, and the fitting coefficients K1, K2, and K3 are automatically updated every 30 days.

[0018] As a specific solution of the technical solution of the present application, the time-frequency domain joint analysis includes the following steps:

[0019] Perform complex Morlet wavelet transform on the DC-side current signal to extract the energy ratio in the 20 - 50 kHz frequency band;

[0020] Construct a 1D convolutional neural network for the AC-side zero-sequence current, and the input layer receives the original waveform data in a 200 ms time window;

[0021] When the wavelet energy ratio > 15% and the CNN output confidence > 90%, it is determined that there is a high-frequency leakage fault.

[0022] As a specific solution of the technical solution of the present application, the hierarchical protection strategy further includes:

[0023] In the first-level warning stage, start the virtual grounding detection mode and inject a 1 kHz detection signal into the DC bus;

[0024] When the detected grounding impedance Z < 500 Ω, directly jump to the third-level protection action;

[0025] Record the fault characteristics and upload them to the cloud platform for pattern learning.

[0026] As a specific solution of the technical solution of the present application, the noise suppression step further included before step S12 specifically includes:

[0027] Establish an inverter switching frequency fingerprint library, including characteristic frequency points of 50 kHz and 100 kHz;

[0028] An adaptive notch filter is used to dynamically filter the characteristic frequency points, and the filtering depth is ≥ 40 dB;

[0029] When a lightning surge is detected, it automatically switches to the FIR low-pass filtering mode.

[0030] As a specific solution of the technical solution of this application, the method specifically uses a heterogeneous computing architecture as follows:

[0031] The FPGA chip executes wavelet transform and transient feature extraction, and the processing delay < 100 μs;

[0032] The AI coprocessor runs the neural network model, and the inference time per cycle ≤ 2 ms;

[0033] The dual processors exchange feature data through shared memory, and the synchronization error < 10 μs.

[0034] Compared with the prior art, the beneficial effects of the present invention are:

[0035] The photovoltaic power generation control cabinet with leakage protection function and its working method can, by constructing a leakage current dynamic reference model based on environmental parameters and with the aid of time-frequency domain joint analysis and multi-dimensional feature vector extraction, accurately evaluate the leakage risk in real time during the operation of the photovoltaic system, and automatically execute differentiated protection strategies according to the risk score. To improve the detection accuracy, the system uses adaptive notch filtering and FIR low-pass filtering, and can maintain high precision even in extreme environments such as lightning surges. The data processing uses a heterogeneous computing architecture. The FRGA chip processes transient feature extraction and wavelet transform, the AI coprocessor runs the neural network model, and the dual processors efficiently synchronize data through shared memory, greatly improving the response speed and reliability of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the present invention;

[0037] Figure 2 is a flow chart of the time-frequency domain joint analysis of the present invention;

[0038] Figure 3 is a flow chart of the noise suppression of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0040] It should be noted that in the description of the present invention, the orientation or positional relationship indicated by the terms "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention.

[0041] In addition, it should be understood that for the convenience of description, the dimensions of the various components shown in the drawings are not drawn in actual proportional relationship. For example, the thickness or width of some layers may be exaggerated relative to other layers.

[0042] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined or described in one drawing, it will not be necessary to further discuss and describe it specifically in the description of the subsequent drawings.

[0043] As Figures 1 - 3 shown, the present invention provides a technical solution: a working method of a photovoltaic power generation control cabinet with a leakage protection function, including the following steps:

[0044] S11: Based on the operating environment parameters and historical data of the photovoltaic system, establish a dynamic reference model for leakage current. The environment parameters at least include light intensity, ambient temperature, and component open-circuit voltage. It should be clearly understood that in the embodiments of the present application, taking standard test conditions as an example, when the light intensity is 1000W / m 2 ², the ambient temperature is 25 °C, the open-circuit voltage of the photovoltaic module is 45.5V. By using the dynamic discrete equivalent model and the least squares method to identify the parameters of the input and output data, an accurate dynamic reference model for leakage current can be established.

[0045] S12: Based on the leakage current information on the DC side and the AC side, perform joint analysis in the time domain, frequency domain, and time-frequency domain, and extract a multi-dimensional feature vector including waveform distortion rate, high-frequency harmonic energy ratio, and transient energy entropy value. It should be clearly understood that in the present application, by analyzing the leakage current waveform characteristics through the Hausdorff distance algorithm, waveform distortion can be effectively identified. The high-frequency harmonic energy ratio is an important feature for leakage detection. In the present application, the high-order harmonic components of the leakage current are relatively large and need to be extracted through frequency domain analysis. The transient energy entropy value reflects the transient characteristics of the leakage current and is extracted through joint time-frequency domain analysis.

[0046] S13: By fusing the multi-dimensional feature vector and the change amount of the environment parameters, calculate the real-time risk score. The risk assessment calculation satisfies: where α, β, and γ are dynamic weight coefficients, I meas is the measured current, I base is the reference current, E hf is the high-frequency energy, Δ Rins is the insulation resistance change rate; in this application, when the leakage current exceeds the standard, the quotient h of the Hausdorff distance mismatch exceeds the threshold of 2, indicating that the leakage current is abnormal.

[0047] S14: Execute a differential protection strategy based on the risk scoring level, including:

[0048] When S(t) ∈ [0.7, 0.9), start local insulation monitoring and issue a warning;

[0049] When S(t) ≥ 0.9 or dI / dt > 50 mA / ms, cut off the corresponding photovoltaic module string circuit;

[0050] When high-frequency arc characteristics are detected, perform an emergency power-off for the entire system.

[0051] The method for establishing a dynamic reference model includes the following steps:

[0052] Based on the system initialization operation stage, collect environmental parameters and leakage current data for at least 24 hours;

[0053] Obtain the functional relationship between the reference current and environmental parameters through least squares fitting;

[0054] I base = Ki·G + K2·(T - 25) + K3·In(Voc / Vrated), where G is the light intensity, T is the environmental temperature, Voc is the open-circuit voltage of the module, and the fitting coefficients K1, K2, and K3 are automatically updated every 30 days. It should be clear that in the embodiments of this application, when collecting data points, G, T, Voc, and I need to be recorded once per hour, and there are 24 data points in 24 hours, 270 data points in 30 days, and the collected data is organized to obtain a data set. The data set is {G i , T i , Voc i , I mcas , i}, where i = 1 to N. In this application, N is the number of data points, and then a design matrix is constructed, and matrices A and vectors b are defined: Then every 30 days, extract the data of the past 30 days from the system, recalculate K, and update these coefficients in the dynamic reference model. To establish the leakage current dynamic reference model, 24-hour data needs to be collected, and I is fitted once every 30 days using the least squares method basc = K1G + K2T + K3Voc, and the coefficients are updated regularly.

[0055] The joint time-frequency domain analysis includes the following steps:

[0056] Perform complex Morlet wavelet transform on the DC-side current signal to extract the energy ratio in the 20 - 50 kHz frequency band;

[0057] Construct a 1D convolutional neural network for the AC-side zero-sequence current, and the input layer receives the original waveform data in a 200 ms time window;

[0058] When the wavelet energy ratio > 15% and the CNN output confidence > 90%, it is determined that there is a high-frequency leakage fault. It should be clear that in the embodiments of this application, first, the DC-side current signal needs to be collected, and the sampling frequency f s is 1 MHz (i.e., 1,000,000 samples / second), the sampling time window length T is 10 ms, and the number of sampling points N = f s ·T = 10000 points, and then calculate the energy ratio in the 20 - 50 kHz frequency band: In this application, assume that the energy E 20-50kHZ in the 20 - 50 kHz frequency band = 1500, and the total energy E total = 10000, Then use the labeled dataset to train the network. The labeled dataset includes the zero-sequence current waveforms of normal and high-frequency leakage faults. After training, input the zero-sequence current waveform data in a new 200 ms time window, and the network outputs a confidence level P, indicating the probability of a high-frequency leakage fault. Assume that the network output confidence level P = 0.95 (i.e., 95%). When the wavelet energy ratio > 15% and the CNN output confidence > 90%, it is determined that there is a high-frequency leakage fault. Assume that the wavelet energy ratio is 15% and the CNN output confidence is 95%, and the determination result is that there is a high-frequency leakage fault.

[0059] The hierarchical protection strategy also includes:

[0060] In the first-level warning stage, start the virtual grounding detection mode and inject a 1 kHz detection signal into the DC bus;

[0061] When the detected grounding impedance Z < 500 Ω, directly jump to the third-level protection action;

[0062] Record the fault characteristics and upload them to the cloud platform for pattern learning. It should be clear that in this application, the injected monitoring signal is a sine wave, and the amplitude is determined according to the system design, usually between a few milliamperes and dozens of milliamperes. Then, by measuring the voltage V and current I of the injected signal on the DC bus, the grounding impedance Z can be calculated and calculated through the grounding impedance formula: Among them, V is the measured voltage and I is the injected current. As can be seen from the previous text, when the grounding impedance Z < 500 Ω, the system immediately executes a three-level protection action to cut off the corresponding photovoltaic module string circuit to prevent the expansion of the fault. Assuming the grounding impedance Z = 400 Ω, the system immediately cuts off the circuit. At the same time, the system will record the detailed characteristics when the fault occurs, including the grounding impedance Z, the injected current I, the measured voltage V, and the timestamp, etc. Assuming that when the fault occurs, the grounding impedance Z = 400 Ω, the injected current I = 5 mA, the measured voltage V = 2.0 V, and the timestamp is 2024-10-10 10:30:00, the fault characteristic data is uploaded to the cloud platform through the network for pattern learning and analysis. The cloud platform uses machine learning algorithms to analyze the uploaded data and extract the fault characteristic patterns to optimize future fault detection and protection strategies.

[0063] The noise suppression steps specifically include:

[0064] Establish an inverter switching frequency fingerprint library, including characteristic frequency points of 50 kHz and 100 kHz;

[0065] Adopt an adaptive notch filter to dynamically filter the characteristic frequency points, and the filtering depth ≥ 40 dB;

[0066] When a lightning surge is detected, it automatically switches to the FIR low-pass filter mode. It should be clear that in the embodiments of this application, 50 kHz and 100 kHz are the characteristic frequencies of the inverter switching frequency. Usually, significant noise will be generated during the switching operation of the inverter. The selection of these frequencies is based on the operating frequency of the inverter and common noise spectrum analysis. The output signal of the inverter under different loads and operating conditions is collected by a spectrum analyzer, and the fast Fourier transform is used to perform spectrum analysis on the collected signal. The amplitude and phase information of the 50 kHz and 100 kHz frequencies are extracted and stored in the fingerprint database for subsequent noise suppression processing. It should also be clear that in this application, a practical adaptive filter, such as the least mean square error algorithm, is used to dynamically adjust the system of the filter. By selecting filter parameters, sufficient filtering effect is ensured. In this application, the filter parameter is the best at 64 orders, and the step factor U is between 0.001 and 0.01 to balance the convergence speed and stability. The initial weight is set to a zero vector or a small random value. Then, the collected inverter output signal is input into the adaptive notch filter, the error between the filter output and the desired signal is calculated, and according to the error signal and the input signal, the LMS algorithm is used to update the weight of the filter to determine that the attenuation of the filter at the characteristic frequency reaches more than 40 dB. By monitoring the amplitude and change rate of the input signal, when it is detected that the signal amplitude exceeds the preset threshold or the change rate is abnormal, it is determined as a lightning surge. In this application, an inverter switching frequency fingerprint database is established, including the 50 kHz and 100 kHz characteristic frequencies. An adaptive notch filter is used to dynamically filter the characteristic frequencies, and the filtering depth is ≥ 40 dB. When a lightning surge is detected, it automatically switches to the FIR low-pass filter mode.

[0067] The specific time of the heterogeneous computing architecture is as follows:

[0068] The FPGA chip performs wavelet transform and transient feature extraction, and the processing delay < 100 μs;

[0069] The AI coprocessor runs the neural network model, and the inference time per cycle ≤ 2 ms;

[0070] The dual processors exchange feature data through shared memory, and the synchronization error is < 10 μs. It should be clear that in the embodiments of the present application, the FPGA chip is responsible for performing wavelet transform and transient feature extraction, and the processing delay is less than 100 us. The FPGA chip accelerates the wavelet transform through parallel computing to ensure that the calculation is completed within 100 us. The FPGA chip extracts transient features through high-speed parallel processing, and the processing delay is also less than 100 us. In the present application, assuming that the computational complexity of the wavelet transform is O(NlogN), where N is the number of data points, the parallel processing ability of the FPGA can significantly reduce the calculation time. Assuming that the computational complexity of transient feature extraction is O(N), the parallel processing ability of the FPGA can also reduce the calculation time. At the same time, it should also be clear that the AI coprocessor runs the neural network model, and the inference time per cycle does not exceed 2 ms. Assuming that the computational complexity of the neural network model is O(M), where M is the number of model parameters, the AI coprocessor ensures that the inference time is within 2 ms through parallel computing and optimized hardware architecture.

[0071] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A working method of a photovoltaic power generation control cabinet with a leakage protection function, characterized in that, It includes the following steps: S11: Based on the operating environment parameters and historical data of the photovoltaic system, establish a dynamic leakage current reference model, where the environment parameters at least include light intensity, ambient temperature, and component open-circuit voltage; S12: Conduct joint time-domain, frequency-domain, and time-frequency-domain analysis based on the leakage current information on the DC side and AC side, and extract a multi-dimensional feature vector including waveform distortion rate, high-frequency harmonic energy ratio, and transient energy entropy value; S13: Calculate the real-time risk score by fusing the multi-dimensional feature vector and the change in environmental parameters, and the risk assessment calculation satisfies: where α, β, and γ are dynamic weight coefficients, I meas is the measured current, I base is the reference current, E hf is the high-frequency energy, Δ Rins is the insulation resistance change rate; S14: Implement a differential protection strategy based on the risk score level, including: When S(t) ∈ [0.7, 0.9), start local insulation monitoring and issue a warning; When S(t) ≥ 0.9 or dI / dt > 50 mA / ms, cut off the corresponding photovoltaic module string circuit; When high-frequency arc characteristics are detected, perform an emergency power-off for the entire system.

2. The working method of a photovoltaic power generation control cabinet with a leakage protection function according to claim 1, characterized in that: The method for establishing the dynamic reference model includes the following steps: Based on the initial operation stage of the system, collect the environment parameters and leakage current data for at least 24 hours; Obtain the functional relationship between the reference current and the environment parameters through least squares fitting; I base = K1·G + K2·(T - 25) + K3·ln(Voc / Vrated), where G is the light intensity, T is the ambient temperature, and Voc is the open-circuit voltage of the module. The fitting coefficients K1, K2, and K3 are automatically updated every 30 days.

3. The working method of a photovoltaic power generation control cabinet with a leakage protection function according to claim 1, characterized in that: The joint time-frequency-domain analysis includes the following steps: Perform complex Morlet wavelet transform on the DC side current signal and extract the energy ratio in the 20 - 50 kHz frequency band; Construct a 1D convolutional neural network for the AC side zero-sequence current, and the input layer receives the original waveform data within a 200 ms time window; When the wavelet energy ratio > 15% and the CNN output confidence > 90%, it is determined that there is a high-frequency leakage fault.

4. The working method of a photovoltaic power generation control cabinet with a leakage protection function according to claim 1, characterized in that: The hierarchical protection strategy further includes: In the first-level warning stage, start the virtual ground detection mode and inject a 1 kHz detection signal into the DC bus; When the detected grounding impedance Z < 500 Ω, directly jump to the third-level protection action; Record the fault characteristics and upload them to the cloud platform for pattern learning.

5. The working method of a photovoltaic power generation control cabinet with a leakage protection function according to claim 1, characterized in that: Before step S12, there is also a noise suppression step, which specifically includes: Establish an inverter switching frequency fingerprint library, including characteristic frequency points of 50 kHz and 100 kHz; Use an adaptive notch filter to dynamically filter the characteristic frequency points, and the filtering depth ≥ 40 dB; When lightning surges are detected, automatically switch to the FIR low-pass filter mode.

6. The working method of a photovoltaic power generation control cabinet with a leakage protection function according to claim 1, characterized in that: The method is implemented through a heterogeneous computing architecture, specifically: The FPGA chip executes wavelet transform and transient feature extraction, and the processing delay < 100 μs; The AI coprocessor runs the neural network model, and the inference time per cycle ≤ 2 ms; The two processors exchange feature data through shared memory, and the synchronization error < 10 μs.

7. A photovoltaic power generation control cabinet with a leakage protection function, characterized in that, The working method of a photovoltaic power generation control cabinet with a leakage protection function as described in any one of claims 1 to 6 is used.

Images