An island prevention rapid judgment method and system based on edge computing and HPLC

Abstract

The application discloses an islanding prevention rapid judgment method and system based on edge computing and HPLC, and belongs to the technical field of power system protection. The system adopts a layered edge computing architecture, including a terminal layer, an edge layer and a cloud platform layer: the terminal layer collects voltage and current waveforms through an HPLC communication module and extracts parameters such as frequency deviation and harmonic impedance; the edge layer runs a lightweight AI model containing a sliding window FFT module and a fuzzy logic decision engine, and judges islanding in combination with frequency deviation, harmonic impedance and load mutation rate; and the cloud platform layer updates edge end model parameters through a federated learning algorithm. The method adopts a multi-criterion fusion decision algorithm and a lightweight model compression technology, greatly improves the response rate and accuracy of islanding detection, and significantly reduces the misjudgment rate.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power system protection, and particularly relates to a method and system for rapid islanding prevention based on edge computing and HPLC, which is suitable for islanding effect detection and protection of a distributed photovoltaic grid-connected system. BACKGROUND

[0002] In a distributed photovoltaic grid-connected system, islanding effect refers to the phenomenon that the distributed photovoltaic system still supplies power to local loads after the power grid is disconnected due to failure or power failure. The traditional passive islanding prevention detection method, such as overvoltage / underfrequency detection method, has a detection blind area in the new energy high penetration scenario, and the misjudgment rate is as high as 5% to 8%. The active detection method, such as frequency deviation method, relies on centralized master station decision, and the response time is greater than 2 seconds, the action time is long, and it cannot meet the demand of rapid and accurate detection of islanding. Therefore, there is an urgent need for an islanding prevention detection method and system that can improve the detection speed and accuracy. SUMMARY

[0003] The purpose of the present application is to provide a method and system for rapid islanding prevention based on edge computing and HPLC, which can realize rapid and accurate detection of islanding effect of a distributed photovoltaic system, reduce the misjudgment rate, and improve the response speed through a hierarchical edge computing architecture and a multi-criteria fusion decision algorithm.

[0004] To solve the above technical problems, the technical solution adopted by the present application is to provide a system for rapid islanding prevention based on edge computing and HPLC, which adopts a three-level hierarchical edge computing architecture of "terminal layer-edge layer-cloud platform layer" to realize the collaborative linkage of data acquisition, local decision and cloud optimization, wherein:

[0005] The terminal layer is installed at the photovoltaic inverter side or the grid-connected switch, and undertakes the key tasks of data acquisition and feature preprocessing. Specifically, the HPLC communication module equipped in the terminal layer adopts a high-performance communication chip conforming to the frequency band of 3-500 kHz, which can acquire voltage / current waveforms in real time, has strong anti-interference ability and stable data transmission, and ensures that high-frequency signal details are not lost. The feature extraction module built-in the terminal layer adopts an optimized lightweight feature extraction algorithm, which quickly and accurately calculates and extracts key feature parameters (including frequency deviation , harmonic impedance ) from the acquired voltage / current waveform data, providing a basis for subsequent islanding judgment. The calculation formula of the frequency deviation is ; the calculation formula of the harmonic impedance is: ; wherein is the real-time frequency of the voltage / current signal, The fundamental frequency of the national grid AC power (the fundamental frequency of the national grid AC power in China is 50 Hz), The harmonic voltage amplitude, The harmonic current amplitude.

[0006] Edge layer: an edge computing node is deployed as a carrier of the intelligent terminal of the transformer area, and the node runs a lightweight AI model (model size ≤ 50 KB) specially designed for edge devices, effectively reducing the demand for hardware resources. The lightweight AI model includes two core modules:

[0007] Sliding window FFT module: the sliding window technology is adopted, the window length is set to 100 ms, and the overlap rate is 50%. The feature parameters (including frequency deviation , harmonic impedance ) transmitted by the terminal layer are subjected to spectrum analysis, the load power calculated by the local load power sensor is used to calculate the load mutation rate (ΔP is the load active power at the current time, which is collected by the local load power sensor in real time; is the load active power at the previous time; is the sampling time interval, which is usually set to 0.1 s or less according to the system response requirement, so as to capture the rapid change of the load), further mining the feature information in the signal, improving the time resolution and stability of the frequency calculation, and improving the accuracy of the frequency calculation while ensuring the real-time performance;

[0008] Fuzzy logic decision engine: the engine takes frequency deviation , harmonic impedance and load mutation rate as input variables. By constructing a three-dimensional decision space, a fuzzy rule base is developed according to the actual power system operation experience and a large amount of experimental data, and intelligent judgment of island effect is realized.

[0009] The rules of the fuzzy rule base are:

[0010] If , it is determined as non-island effect;

[0011] If , it is determined as island effect;

[0012] If , there is island probability, and island probability determination should be combined with harmonic impedance and load mutation rate , specifically:

[0013] Only when , and When the three conditions are met at the same time, it is determined that the island probability is high;

[0014] wherein 、 is a frequency deviation threshold value, is an impedance threshold value, is a load mutation threshold value.

[0015] The cloud platform layer: as the intelligent hub of the whole system, it is responsible for summarizing and analyzing the data of the edge layer. Through the federated learning algorithm, it can establish a cross-regional island feature library without leaking the data privacy of each edge node, integrate the island event features in different regions and under different working conditions, provide more abundant reference basis for the island judgment of the edge layer, and update the parameters of the optimized micro model to the edge layer to continuously improve the detection accuracy and adaptability of the edge AI model.

[0016] The application also provides an island prevention rapid judgment method based on edge computing and HPLC, which comprises the following steps:

[0017] The terminal layer collects voltage / current waveforms and extracts frequency deviation and harmonic impedance characteristic parameters as basic characteristic parameters for the decision of the edge layer;

[0018] The edge layer performs spectral analysis and processing on the characteristic parameters output by the terminal layer through a sliding window FFT module, improves the real-time performance and accuracy of frequency characteristic calculation, and calculates the load mutation rate in real time through a local load power sensor to reflect the dynamic change of the load;

[0019] The fuzzy logic decision engine is used to fuse the frequency deviation , harmonic impedance and load mutation rate three criteria, and a multi-criteria fusion decision is made for island judgment, and the decision result is output and a control signal (warning or tripping) is triggered;

[0020] As a further description of the above technical solution, the knowledge distillation technology is also used for light-weight model compression, and the cloud large-scale LSTM model is compressed to a micro model (compressed by 90%) deployable on the edge, the parameter quantity is compressed from 10 6 to 10 4 , the precision loss is less than 3%, and it is suitable for the low computing power and low power consumption characteristics of the edge device.

[0021] As a further description of the above technical scheme, the island event data on the edge layer is uploaded to the cloud platform layer, a cross-region island feature library is constructed through a federated learning algorithm, a globally optimized LSTM model is trained, and updated model parameters are obtained; a knowledge distillation technology is used for lightweight model compression, the LSTM model is compressed into a micro model deployable on the edge, and the micro model is issued to the edge layer to update the parameters of the fuzzy logic decision engine on the edge.

[0022] As a further description of the above technical scheme, the multi-criterion fusion decision includes constructing a three-dimensional decision space, formulating a fuzzy rule base for island judgment according to actual power system operation experience and a large amount of experimental data, realizing multi-physical quantity joint decision, and solving the blind area problem of a single criterion.

[0023] If , it is determined that there is no island effect, and island early warning is not triggered.

[0024] If , it is determined that there is island effect, and island early warning is triggered.

[0025] If , there is island probability, and island probability determination must be combined with harmonic impedance and load mutation rate , and the specific island probability determination is as follows:

[0026] Only when , and three conditions are met at the same time, it is determined that the island probability is high, and island early warning is triggered.

[0027] If one of the above three conditions is not met, it is determined that the island probability is low, and island early warning is not triggered for the time being, but close attention must be paid.

[0028] Compared with the prior art, the present application has the following beneficial effects:

[0029] The system of the present application adopts a layered architecture, the terminal layer focuses on data acquisition and feature extraction, the edge layer realizes localized rapid decision, and the cloud platform layer is responsible for global model optimization, forming an "end-edge-cloud" collaborative closed loop, realizing local processing and rapid decision of data, reducing the dependence on the cloud, and improving the real-time performance and reliability of the system; the method of the present application adopts a multi-criterion fusion decision algorithm combined with frequency deviation , harmonic impedance , and load mutation rate Three types of feature parameters are used to construct a three-dimensional judgment space, develop a fuzzy rule base, cover the blind area of traditional passive / active detection, and improve the accuracy of detection; lightweight model compression enables complex AI models to run on edge devices without relying on high-bandwidth communication and cloud real-time computing, reducing hardware costs and computing resource requirements. The island detection system and method of the present application greatly improves the response rate and reduces the misjudgment rate. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 is a system structure schematic diagram of example one.

[0031] Figure 2 is a method flowchart of example two. DETAILED DESCRIPTION

[0032] The technical solutions of the present application will be described clearly and completely in combination with the drawings and examples. Obviously, the described examples are only a part of the examples of the present application, not all examples. Based on the examples in the present application, all other examples obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0033] Example one

[0034] The present embodiment provides a fast island prevention judgment system based on edge computing and HPLC, as shown in Figure 1 , a three-level hierarchical edge computing architecture of "terminal layer-edge layer-cloud platform layer" is adopted to realize the collaborative linkage of data acquisition, local decision and cloud optimization, specifically:

[0035] A new high-speed HPLC communication module is installed on the photovoltaic inverter side of a certain distributed photovoltaic system. The module uses a communication chip that meets the 3-500 kHz frequency band, with an anti-interference capability improved by 30%. Real-time acquisition of voltage and current signals output by the photovoltaic inverter is performed with a sampling rate set to 10 kHz. After preprocessing, the collected signals are input to a feature extraction module integrated with a lightweight feature extraction algorithm to quickly and accurately calculate the frequency deviation and harmonic impedance , and transmit these feature parameters to the edge layer through the HPLC communication module.

[0036] An edge computing node is deployed on the intelligent terminal of the transformer area. The node uses a low-power, high-performance processor. The edge computing node runs a sliding window FFT module to perform frequency spectrum analysis on the voltage and current signals transmitted from the terminal layer, with a window length set to 100 ms and an overlap rate of 50% to improve the accuracy and real-time performance of frequency calculation. At the same time, the edge computing node receives the , and the load power data collected by the local sensor, to calculate the load mutation rate The three variables are input into the fuzzy logic decision engine to make decisions according to the established fuzzy rule base to determine whether the island effect occurs.

[0037] The cloud platform collects island event data and model training data uploaded by each edge layer, adopts a federated learning algorithm, trains and updates a large LSTM model in the cloud without leaking user privacy. The updated model parameters are transmitted to each edge layer through a secure channel to update the AI model at the edge, establish a cross-regional island feature library, and improve the system's detection capability for island effects in different regions and different scenarios.

[0038] Embodiment Two

[0039] This embodiment provides a fast island prevention judgment method based on edge computing and HPLC, as shown in the flowchart Figure 2 Through hierarchical architecture design and multi-technology collaboration, fast and accurate detection of island effects is achieved, including three core links of data collection and feature extraction, edge multi-criteria fusion decision, and cloud model optimization. The specific steps are as follows:

[0040] S1, terminal layer data collection and feature preprocessing:

[0041] The HPLC communication module of the terminal layer continuously collects voltage and current waveform data on the photovoltaic grid side to form time series data; the feature extraction module uses a lightweight feature extraction algorithm to calculate the real-time frequency from the voltage and current waveform (where is the fundamental frequency of the national grid alternating current) ; using a specific harmonic injection and detection method, the harmonic impedance is calculated.

[0042] S2, edge layer multi-criteria fusion decision:

[0043] The sliding window FFT module of the edge layer processes the collected voltage and current waveform data, combines the frequency deviation and harmonic impedance data, and the load power collected by the local load power sensor, to calculate the load mutation rate . The fuzzy logic decision engine combines the rules of the fuzzy rule base to determine the island, where the fuzzy rule base rules are:

[0044] If , it is determined that there is no island effect, and no island warning is triggered;

[0045] If If the frequency deviation is greater than 0.1 Hz, the island effect is determined, and the island warning is triggered.

[0046] If the frequency deviation is greater than 0.1 Hz, the island effect is determined, and the island warning is triggered. If the island probability exists, the island probability determination must be combined with the harmonic impedance and the load mutation rate , specifically:

[0047] Only when the three conditions of , and are met at the same time, it is determined that the island probability is high, and the island warning is triggered.

[0048] S3, cloud platform layer model optimization and update:

[0049] The edge layer periodically uploads the collected data and the decision results of the fuzzy logic decision engine to the cloud platform layer. The cloud platform layer uses federated learning technology to integrate the data uploaded by multiple edge layers, and trains and optimizes the large LSTM model in the cloud. For example, after collecting a large amount of data, the cloud platform layer finds that there is a certain correlation between the frequency deviation and the load mutation rate when an island event occurs in a certain load characteristic in a certain area, and integrates this rule into the LSTM model. Then, through the knowledge distillation technology, the optimized large LSTM model is compressed into a micro model that can be deployed on the edge, and the updated model parameters are distributed to the edge layer, and the edge layer updates the local lightweight AI model to improve the accuracy and adaptability of subsequent island detection. In the subsequent running process, when similar situations occur again, the edge layer can more accurately judge whether it is an island event, effectively avoiding misjudgment and significantly reducing the misjudgment rate.

[0050] Application example:

[0051] A kind of island prevention fast judgment method based on edge computing and HPLC, the steps are as follows:

[0052] S1, terminal layer data acquisition and feature preprocessing:

[0053] The HPLC communication module of the terminal layer continuously acquires voltage and current waveform data on the photovoltaic grid side at a sampling rate of 10 kHz, forming time series data; the feature extraction module uses a lightweight feature extraction algorithm to calculate the real-time frequency =50.25 Hz, the frequency deviation =0.25 Hz ( =50 Hz) is obtained; by using a specific harmonic injection and detection method, the harmonic impedance =25 V / 19.2 A=1.3 Ω is calculated.

[0054] S2, edge layer multi-criterion fusion decision:

[0055] The sliding window FFT module of the edge layer processes the collected voltage and current waveform data, combines the frequency deviation and harmonic impedance data, and the load power collected by the local load power sensor to calculate the load mutation rate . Assuming that the load power changes from 30 kW to 32.7 kW in this period, the time interval is Δt = 1 s, and the calculation result is = 9% / s. The fuzzy logic decision engine combines the rules in the fuzzy rule base to determine the island: since = 0.25 Hz, = 1.3 Ω, = 9% / s, the three conditions of “ , and ” are not met simultaneously (where , are empirically set to 0.1 Hz and 0.4 Hz, respectively; is empirically set to 1.5 Ω; is set to 10% / s), so the island probability is low, and the island warning is not triggered.

[0056] S3, cloud platform layer model optimization and update:

[0057] The edge layer regularly uploads the collected data and the decision results of the fuzzy logic decision engine to the cloud platform layer. The cloud platform layer uses federated learning technology to integrate the data uploaded by multiple edge layers and trains and optimizes the large LSTM model on the cloud. For example, after collecting a large amount of data, the cloud platform layer finds that in a certain region, there is a certain correlation between the frequency deviation and the load mutation rate when an island event occurs under certain load characteristics, and integrates this rule into the LSTM model. Then, through the knowledge distillation technology, the optimized large LSTM model is compressed into a micro model that can be deployed on the edge, and the updated model parameters are distributed to the edge layer, and the edge layer updates the local lightweight AI model to improve the accuracy and adaptability of subsequent island detection. In the subsequent running process, when similar situations occur again, the edge layer can more accurately determine whether it is an island event, and the misjudgment rate is reduced by 4%.

[0058] The present application realizes the "fast detection-accurate decision-continuous optimization" closed loop of the distributed photovoltaic system island effect through the island prevention fast judgment system and method based on edge computing and HPLC, and provides an innovative solution for the safety of the power grid under the high proportion of new energy grid-connected scenario.

[0059] The above merely describes the best mode of the present application. It should be noted that for those skilled in the art, the technical solutions of the present application can be modified or replaced equivalently without departing from the principles of the present application, and the technical effects of the present application can also be achieved, which should be considered as belonging to the protection scope of the present application.

Claims

1. An islanding rapid judgment system based on edge computing and HPLC, adopting a hierarchical edge computing architecture, comprising: a terminal layer: arranged at a photovoltaic inverter side or a grid-connected switch, for real-time acquisition of voltage / current waveforms and extraction of output characteristic parameters; an edge layer: deployed at a district intelligent terminal, for running a lightweight AI model for islanding judgment; a cloud platform layer: establishing a cross-regional islanding feature library through a federal learning algorithm, and updating AI model parameters of the edge end; characterized in that: the lightweight AI model comprises: Sliding window FFT module: using sliding window technology, the terminal layer of the characteristic parameters transmitted by the spectrum analysis, through the local load power sensor to collect the load power calculation load mutation rate ; Fuzzy logic decision engine: takes frequency deviation , harmonic impedance and load rate of change as input variables, and realizes intelligent judgment of islanding effect through multi-criteria fusion decision; the multi-criterion fusion decision-making comprises constructing a three-dimensional judgment space, and formulating a fuzzy rule base for islanding judgment according to actual power system operation experience and a large amount of experimental data; the fuzzy rule base contains the following rules: If then determine as non-islanding; If then islanding is determined; If , there is an island probability, and the harmonic impedance and load mutation rate must be combined to determine the island probability, specifically: Only when , and three conditions are met at the same time, it is determined that the island probability is high, triggering island warning; wherein , is a frequency deviation threshold value, is an impedance threshold value, is a load jump threshold value.

2. The system of claim 1, wherein, the terminal layer comprises: an HPLC communication module: for real-time acquisition of voltage / current waveforms; Feature extraction module: integrate lightweight feature extraction algorithm, used to extract and output frequency deviation And harmonic impedance .

3. The system of claim 2, wherein: the working frequency band of the HPLC communication module is 3-500 kHz, and the sampling rate is 10 kHz.

4. The system of claim 2, wherein: The frequency deviation The calculation formula is: ; harmonic impedance The calculation formula is: ; wherein is the real-time frequency of the voltage / current signal, is the fundamental frequency of the national grid alternating current, is the harmonic voltage amplitude, is the harmonic current amplitude.

5. A method for rapid judgment of islanding prevention based on edge computing and HPLC, characterized in that, comprising the following steps: The terminal layer collects voltage / current waveforms, extracts frequency deviation And harmonic impedance Characteristic parameters, as the basis of edge layer decision characteristic parameters; The edge layer performs spectrum analysis on the characteristic parameters output by the terminal layer through a sliding window FFT module, and calculates the load mutation rate in real time through a local load power sensor ; Fuzzy logic decision engine fuses frequency deviation , harmonic impedance , and load rate of change Three criteria, multi-criteria fusion decision is used to judge island, output decision results and trigger control signal; the multi-criterion fusion decision-making comprises constructing a three-dimensional judgment space, and formulating a fuzzy rule base for islanding judgment according to actual power system operation experience and a large amount of experimental data; the fuzzy rule base contains the following rules: If then determine as non-islanding; If then islanding is determined; If , there is an island probability, and the harmonic impedance and load mutation rate must be combined to determine the island probability, specifically: Only when , and three conditions are met at the same time, it is determined that the island probability is high, triggering island warning; wherein , is a frequency deviation threshold value, is an impedance threshold value, is a load jump threshold value.

6. The method of claim 5, wherein, Also include: edge layer on the island event data to the cloud platform layer, through the federal learning algorithm constructs the cross-region island feature library, trains the global optimization's LSTM model, obtains the update model parameter, the model parameter includes the frequency deviation threshold 、 , impedance threshold And load mutation threshold .

7. The method of claim 6, wherein, further comprising: adopting a knowledge distillation technology for lightweight model compression, compressing an LSTM model into a micro model deployable at the edge end, and issuing it to the edge layer to update the fuzzy logic decision engine parameters of the edge end.

Images