A short-circuit current adaptive detection method for a distributed green electricity direct connection system

Abstract

The application discloses a short-circuit current adaptive detection method of a distributed green electricity direct connection system, belongs to the field of power system automation, and aims to solve the problems of misoperation, refusal to operate and response delay caused by fixed threshold, single feature, insufficient transient identification and centralized architecture in the prior art. The method comprises the following steps: collecting multi-dimensional electrical quantities in real time to construct a space-time correlation data set; fusing the current change rate, voltage drop, harmonic distortion and inverter state to generate a dynamic fault feature vector; dynamically adjusting the threshold based on illumination, wind speed and load working condition; identifying the fault through multi-node weighted consistency collaboration; and outputting high-precision alarm and positioning results. The system is deployed on an edge computing node, contains an FPGA and an ARM coprocessor, and is integrated with a transient modeling and online learning module. Through the above scheme, millisecond-level adaptive detection, high-reliability collaborative identification and continuous performance optimization are realized.

Description

Technical Field

[0001] This invention belongs to the field of power system automation, specifically, it relates to an adaptive detection method for short-circuit current in a distributed green electricity direct-connection system. Background Technology

[0002] With the deepening of the "dual carbon" goals, the proportion of green electricity in the energy structure continues to increase. Distributed green electricity direct connection systems are gradually becoming an important part of the new power system due to their advantages such as high efficiency, low carbon emissions, and local consumption. These systems typically supply power directly to specific loads from distributed renewable energy sources such as photovoltaics and wind power through dedicated lines, eliminating the complex step-up, transmission, and step-down processes in traditional power grids, and significantly improving energy utilization efficiency and power supply economy.

[0003] Short-circuit current detection is a core component in ensuring the safe and stable operation of distributed green energy direct-connection systems. Because the output characteristics of green energy sources (such as inverter-type photovoltaics and wind turbines) differ fundamentally from those of traditional synchronous generators, their fault current amplitude is small, rises rapidly, has high harmonic content, and is significantly affected by control strategies. This makes it difficult for traditional short-circuit protection methods based on power frequency steady-state models to accurately identify and quickly respond to fault conditions.

[0004] Existing technologies for short-circuit current detection in distributed green energy direct-connection systems exhibit multiple shortcomings: First, most detection methods rely on fixed thresholds or preset time windows, failing to adapt to the dynamic changes in short-circuit current characteristics under different operating conditions, easily leading to false tripping or failure to trip. Second, existing solutions generally do not fully integrate multi-dimensional electrical quantities such as voltage, current, power, and topology information, resulting in insufficient fault feature extraction and weak anti-interference capabilities. Third, for high-penetration renewable energy access scenarios, there is a lack of effective identification mechanisms for transient processes caused by inverter current limiting control, low-voltage ride-through, and other operating mode switching. Finally, in complex topologies with multi-source collaborative power supply, the output of each power source fluctuates significantly, making it difficult for traditional centralized detection architectures to achieve millisecond-level fault location and isolation. These problems severely restrict the large-scale application of distributed green energy direct-connection systems in high-reliability power consumption scenarios, necessitating a short-circuit current detection method with adaptive sensing, multi-source collaborative criteria, and rapid response capabilities. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive detection method for short-circuit current in a distributed green electricity direct-connection system, which mainly solves the problems of false activation, failure to activate, and response delay caused by fixed thresholds, single features, insufficient transient identification, and centralized architecture in existing technologies.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] An adaptive short-circuit current detection method for a distributed green electricity direct-connection system includes the following steps:

[0008] S1 collects multi-dimensional electrical quantity data in real time and constructs a multi-source spatiotemporal correlation dataset by combining it with network topology information;

[0009] S2, based on the multidimensional electrical quantity data, and simultaneously introducing the inverter operating status flag bit, a dynamic fault feature vector containing feature parameters of more than a predetermined dimension is formed.

[0010] S3 performs adaptive threshold determination, dynamically adjusts the threshold boundary of the short-circuit fault criterion according to the current system operating conditions, and continuously monitors the feature vector through a sliding time window mechanism;

[0011] S4, initiate multi-source collaborative fault identification, collect feature vectors from multiple or more adjacent nodes, and use a weighted consensus algorithm to verify the fault direction and location;

[0012] S5 outputs short-circuit fault alarm and location results. If the multi-source collaborative judgment confirms the existence of the fault, an alarm message is generated and sent to the relevant circuit breaker for isolation operation through the high-speed communication channel within a predetermined delay time.

[0013] Furthermore, in this invention, the multidimensional electrical quantity data includes three-phase voltage, three-phase current, active power, reactive power, and system frequency, which are acquired by deploying high-sampling-rate synchronous phasor measurement units at each power source outlet, load access point, and key node of the line in the distributed green electricity direct connection system.

[0014] Furthermore, in this invention, in step S2, multidimensional electrical quantity data is used to calculate the rate of change of current, voltage sag depth, harmonic distortion rate, power surge, and zero-sequence component; wherein, the rate of change of current is calculated using the five-point difference method, and its expression is: , where Δt is the sampling interval; the harmonic distortion rate is calculated by extracting harmonic components within a predetermined number range through fast Fourier transform and then using the total harmonic distortion rate formula; the inverter operating status flag includes four states: normal power generation, current-limited operation, low voltage ride-through, and shutdown.

[0015] Furthermore, in this invention, the adjustment of the adaptive threshold is based on a preset working condition mapping table, which is obtained through offline simulation and field measurement data training and covers a predetermined number of typical operating scenarios. When the system is running online, the threshold parameter closest to the current working condition is called in real time through the nearest neighbor matching algorithm.

[0016] Furthermore, in this invention, the weight coefficients of the weighted consensus algorithm are dynamically allocated based on the electrical distance from each node to the suspected fault point, and the weight calculation expression is as follows: ,in Let be the normalized electrical distance of the i-th node, and ε be a small constant greater than 0.

[0017] Furthermore, in this invention, in S4, when the dynamic fault feature vector of any node continuously exceeds the adaptive threshold for a predetermined time period, a regional collaborative discrimination mechanism is triggered to achieve feature vector aggregation; wherein, the regional collaborative discrimination mechanism requires at least two non-adjacent nodes to simultaneously detect a sudden increase in current in the same direction and a synchronous drop in voltage, then it is confirmed as a real short-circuit fault.

[0018] Furthermore, in this invention, the alarm information includes the fault type, the time of occurrence, the location segment, and the confidence level. The fault location segment is determined by comparing the phase difference of the current at both ends of each line segment. When the absolute value of the phase difference is greater than a preset angle threshold and the duration exceeds a predetermined time period, it is determined that a short circuit has occurred inside the segment.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] (1) This invention breaks through the limitations of traditional fixed threshold detection. By establishing a threshold mapping table covering multiple working conditions such as light, wind speed, and load level, and combining the nearest neighbor matching algorithm to dynamically adjust the fault judgment threshold, it can adapt to the changes in the form of short-circuit current under the full-scene operation of the distributed green electricity direct connection system. Compared with the traditional fixed threshold scheme, the fault identification accuracy is improved, and the problem of difficulty in balancing detection sensitivity and reliability under different output working conditions is effectively solved.

[0021] (2) This invention breaks through the traditional detection limitations of single current characteristics. It integrates eight parameters such as current change rate, voltage drop depth, harmonic distortion rate, power mutation amount, and inverter operating status to construct a dynamic fault feature vector, which fully covers the transient and steady-state characteristics of inverter-type green power supply faults. It can effectively distinguish fault transients from non-fault conditions such as load fluctuations and low voltage ride-through. The anti-interference capability is effectively improved compared with traditional solutions.

[0022] (3) The present invention adopts an edge-side distributed collaborative detection architecture and achieves cross-verification of faults through a multi-node weighted consensus algorithm, avoiding the communication delay bottleneck of traditional centralized detection. The total time for fault detection and location can be controlled within 10ms, which is an order of magnitude faster than the traditional centralized solution. It can support the large-scale application of distributed green electricity direct connection system in high-reliability power consumption scenarios such as data centers and precision manufacturing.

[0023] (4) This invention is specifically designed with feature extraction and discrimination logic for the operating characteristics of inverter-type green power sources, such as current limiting control and low voltage ride-through. The inverter operating status is directly incorporated into the fault feature vector, which can effectively identify special transient processes such as power output fluctuation and control mode switching of green power sources. This solves the problem that the traditional detection method based on synchronous generator model is not adaptable to high-penetration green power scenarios. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the method of the present invention.

[0025] Figure 2 This is a schematic diagram of the topology of the distributed green electricity direct connection system in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0027] like Figure 2 As shown, this embodiment is applied to a distributed green electricity direct-connection system in a typical industrial park. The system consists of a rooftop photovoltaic array (total installed capacity of 2.5 MW), a small wind turbine generator (0.8 MW), and several critical loads (including data centers, precision manufacturing equipment, etc.), interconnected by three dedicated 10 kV feeders to form a multi-power, multi-load, closed-loop switchable topology. Edge intelligent detection nodes are deployed at each power outlet, load access point, and feeder sectionalizing switch in the system to perform short-circuit current adaptive detection tasks.

[0028] At the system architecture level, each edge intelligent detection node consists of the following core hardware modules: a high-sampling-rate synchronous phasor measurement unit (PMU), a dual-core ARM Cortex-A72 processor, a Xilinx Artix-7 series FPGA coprocessor, a gigabit Ethernet communication interface, a BeiDou / GNSS dual-mode timing module, and a local storage and power management unit. The synchronous phasor measurement unit uses the Analog Devices AD7606C-18 18-bit synchronous sampling ADC chip, which has 8 channels of analog input capability and supports the synchronous acquisition of three-phase voltage (accessed via PT ratio 10 kV / 100 V), three-phase current (accessed via CT ratio 400 A / 1 A), and DC-side auxiliary signals. The sampling frequency is fixed at 12.8 kHz, meeting the sampling requirement of no less than 10 kHz. The analog front-end of this ADC chip integrates an anti-aliasing low-pass filter with a cutoff frequency set at 5.12 kHz, effectively suppressing high-frequency noise interference.

[0029] The synchronization phasor measurement unit is connected to the FPGA coprocessor via an SPI bus. The FPGA internally includes a high-speed data buffer FIFO, a digital filter bank, and initial feature screening logic. The FPGA simultaneously receives a 1 PPS (pulses per second) signal and a serial time code from the BeiDou / GNSS dual-mode timing module, achieving a time synchronization accuracy better than 0.8μs, ensuring strict alignment of data acquisition across all nodes in the system. The acquired raw electrical quantity data is encapsulated within the FPGA using the IEEE 1588v2 Precision Time Protocol (PTP) and then transmitted via Gigabit Ethernet interface using UDP multicast to the local ARM processor and adjacent nodes, with communication latency controlled within 200μs.

[0030] The dual-core ARM processor runs a customized Linux real-time operating system (RT-Linux). The main core is responsible for executing multi-source collaborative fault identification algorithms, adaptive threshold calling, fault location logic, and alarm information generation; the secondary core is dedicated to maintaining the historical fault database, performing online learning tasks, and communicating with the upper-level monitoring system. The ARM processor interconnects with the FPGA at high speed via a PCIe Gen2 x1 interface, with a bandwidth of up to 5 Gbps, ensuring low-latency transmission of intermediate data such as feature vectors. The local storage unit uses an industrial-grade eMMC chip (32 GB capacity) to cache the raw sampling data and fault event records of the most recent 72 hours.

[0031] The entire distributed system adopts a star-ring hybrid network topology. Each edge node forms a redundant communication ring network through an industrial-grade gigabit switch, supporting RSTP (Rapid Spanning Tree Protocol) for link fault self-healing. System topology information (including node number, line impedance parameters, switch status, etc.) is periodically distributed by the central configuration server and cached in the non-volatile memory of each node for use in fault location.

[0032] Under the support of the above system architecture, such as Figure 1 As shown, the specific execution process of the workflow in this embodiment is as follows:

[0033] First, the synchronous phasor measurement units at each edge node continuously acquire the three-phase voltage u at a sampling rate of 12.8 kHz. a u b u c Three-phase current i a i b i cThe system calculates active power P, reactive power Q, and system frequency f in real time based on instantaneous power theory. After all sampled data is timestamped within the FPGA, it is combined with locally stored network topology information (such as the feeder number to which this node belongs, adjacent node IDs, line length, etc.) to construct a multi-source associated dataset containing spatiotemporal coordinates. This dataset is packaged and uploaded to the ARM main core at 10ms intervals as the basis for subsequent feature extraction.

[0034] The ARM main core calls the preprocessed current sequence from the FPGA and calculates the rate of change of current di / dt using the five-point central difference method. Specifically, for the current sampling point k, the formula for calculating the rate of change of current is:

[0035]

[0036] Where Δt = 1 / 12800 ≈ 78.125 μs. This method effectively suppresses the influence of high-frequency noise on derivative estimation while ensuring computational efficiency. Simultaneously, the system extracts the 2nd to 50th harmonic components using a sliding window FFT (window length 20ms, overlap rate 50%) and calculates the total harmonic distortion (THD) according to IEC 61000-4-7 standard. Voltage sag depth is defined as the percentage deviation between the current effective voltage value and the rated value; power surge is obtained by comparing the active / reactive power difference between the current 10ms window and the previous window; zero-sequence components are calculated by synthesizing three-phase currents. Furthermore, the inverter controller reports its operating status flags (0=normal generation, 1=current-limited operation, 2=low voltage ride-through, 3=shutdown) in real time via the CAN bus; these flags are directly embedded in the feature vector. Finally, the system integrates the above 8-dimensional parameters (i.e., the predetermined dimensions are 8, including the maximum value of three-phase di / dt, voltage sag depth, THD, active power surge, reactive power surge, zero-sequence current amplitude, system frequency deviation, and inverter status flag bits) to form a dynamic fault feature vector. .

[0037] The system dynamically adjusts the judgment thresholds for each feature dimension based on the current operating conditions. Operating condition information includes: real-time light intensity (unit: W / m²) and wind speed (m / s) provided by the weather station, total load level (percentage of rated capacity) fed back by the load monitoring module, and a summary of the control modes of each inverter. These operating condition parameters are quantified into feature vectors. The system then uses a nearest neighbor matching algorithm to retrieve the most similar historical scenario from a pre-stored operating condition mapping table. This mapping table contains 128 typical operating combinations (i.e., a predetermined number of operating scenarios, such as "sunny day + full power generation + light load", "cloudy day + wind turbine full power generation + heavy load", etc.), and each combination corresponds to a set of optimal thresholds, such as the current delay change rate threshold T. di / dt∈ [50, 500] A / ms, voltage drop depth threshold T vd ∈ [10%, 90%]. The system uses a sliding time window (window length 5ms, step size 1ms) to process the feature vector V. f Continuous monitoring is performed, and if any dimension exceeds its adaptive threshold for a duration of ≥2ms, the multi-source collaborative fault identification mechanism is triggered.

[0038] When a node (denoted as NodeA) detects a suspected fault, it immediately requests a synchronization feature vector from its three or more electrically nearest neighbors (e.g., NodeB, NodeC, NodeD) via multicast. Each responding node sends back its current feature vector and local timestamp within 1ms. The master node (usually the node that triggered the fault first) collects this data, first verifies time synchronization (deviation < 5μs), and then executes a weighted consensus algorithm. Weighting coefficients... Based on the normalized electrical distance from each node to the suspected fault area Dynamic calculation:

[0039]

[0040] in The results are pre-calculated from the line impedance matrix and topology information. The weighted consensus algorithm requires at least two non-adjacent nodes to simultaneously meet the following conditions: (a) the current rate of change increases suddenly in the same direction (i.e., di / dt signs are consistent); (b) the voltage drop depth synchronously exceeds the threshold; and (c) the weighted eigenvector consensus score is higher than a preset threshold (e.g., 0.85). If the conditions are met, it is confirmed as a real short-circuit fault.

[0041] When a real short-circuit fault exists, the system first performs precise location based on the current phase difference between the two ends of each feeder segment. For any feeder segment L j Its first and last nodes are denoted as N. start and N end The system calculates the difference Δφ between the fundamental phase angles of the currents at the two nodes. j If |Δφ j If the angle is greater than 30° and the duration is ≥3ms, then the short circuit is determined to have occurred at L. jInternally, the core theoretical basis of this criterion is the phase change law of the fault current in the distributed green electricity direct-connection system: when a short circuit occurs within the line section, the fault current at the beginning and end of the section is provided by different power sources (photovoltaic, wind power, backup power), and the impedance change at the fault point will cause a significant shift in the fundamental phase of the current at the beginning and end. Verified by PSCAD / EMTDC simulation and field measurements, the absolute value of this phase difference is greater than 30° under all operating conditions. However, when a short circuit occurs outside the line section or the system is operating normally (including inverter power fluctuations and low-voltage ride-through), the current at the beginning and end of the section is provided by the same power source or affected by the same fault point, and the absolute value of the phase difference is less than 30°. Furthermore, the 30° threshold is selected as the optimal distinction point between fault and non-fault states, verified by simulations of 128 typical operating conditions. The criteria are applicable to fault location of 10kV dedicated feeder sections in distributed green electricity direct-connection systems. Specifically, they are suitable for: topologies where inverter-type green electricity sources such as photovoltaic and wind power are directly connected to the grid; inverters are in four operating states: normal power generation, current-limiting operation, low-voltage ride-through, and shutdown; system loads are constant power / constant impedance loads such as data centers and precision manufacturing equipment; and edge detection node deployment with a sampling frequency ≥10kHz and time synchronization accuracy better than 0.8μs.

[0042] The location results, combined with the fault type (judged as single-phase grounding, phase-to-phase short circuit, etc. based on zero-sequence component and harmonic characteristics), occurrence time (accurate to the μs level), and confidence level, generate a structured alarm message. This message is sent to the intelligent circuit breaker of the relevant feeder within 5ms via a high-speed communication channel (using the IEC 61850-9-2LE protocol), triggering tripping and isolation. The confidence level is derived from the consistency score, which is obtained by normalizing the consistency score of the weighted feature vectors of each node. The normalized result is the fault confidence level, ranging from 0 to 1. The consistency score is obtained by weighted summation of the matching degree and weight coefficient of the feature vectors of each node (the matching degree is 0-1, positively correlated with the similarity of electrical quantity characteristics). For example, if the weights of the three nodes are 0.6, 0.3, and 0.1, and the feature vector matching degrees are 1, 0.9, and 0.8, then the consistency score is 0.6×1+0.3×0.9+0.1×0.8=0.95, and the normalized fault confidence is 95%.

[0043] The above embodiments are merely one of the preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A short-circuit current adaptive detection method for a distributed green electricity direct-connection system, characterized in that, Includes the following steps: S1 collects multi-dimensional electrical quantity data in real time and constructs a multi-source spatiotemporal correlation dataset by combining it with network topology information; S2, based on the multi-dimensional electrical quantity data, and simultaneously introducing the inverter operating status flag bit, a dynamic fault feature vector containing feature parameters of a predetermined dimension or higher is formed; wherein, the inverter operating status flag bit includes four states: normal power generation, current limiting operation, low voltage ride-through, and shutdown. S3 performs adaptive threshold determination, dynamically adjusts the threshold boundary of the short-circuit fault criterion according to the current system operating conditions, and continuously monitors the feature vector through a sliding time window mechanism; S4, initiate multi-source collaborative fault identification, collect feature vectors from multiple adjacent nodes, and use a weighted consensus algorithm to verify the fault direction and location; wherein, the weight coefficients of the weighted consensus algorithm are dynamically allocated according to the electrical distance from each node to the suspected fault point, and the weight calculation expression is as follows: ,in Let be the normalized electrical distance of the i-th node, and ε be a small constant greater than 0; When the dynamic fault feature vector of any node continuously exceeds the adaptive threshold for a predetermined time period, the regional collaborative discrimination mechanism is triggered to realize feature vector aggregation. The regional collaborative discrimination mechanism requires at least two non-adjacent nodes to simultaneously detect a sudden increase in current in the same direction and a synchronous drop in voltage, which is then confirmed as a real short-circuit fault. S5 outputs short-circuit fault alarm and location results. If the multi-source collaborative judgment confirms the existence of the fault, an alarm message is generated and sent to the relevant circuit breaker for isolation operation through the high-speed communication channel within a predetermined delay time.

2. The short-circuit current adaptive detection method for a distributed green electricity direct-connection system according to claim 1, characterized in that, The multidimensional electrical quantity data includes three-phase voltage, three-phase current, active power, reactive power, and system frequency, which are acquired by deploying high-sampling-rate synchronous phasor measurement units at each power source outlet, load access point, and key node of the line in the distributed green electricity direct connection system.

3. The short-circuit current adaptive detection method for a distributed green electricity direct-connection system according to claim 2, characterized in that, In step S2, multidimensional electrical quantity data are used to calculate the rate of change of current, voltage sag depth, harmonic distortion rate, power surge, and zero-sequence component; wherein, the rate of change of current is calculated using the five-point difference method, and the expression is: , where Δt is the sampling interval; the harmonic distortion rate is calculated by extracting harmonic components within a predetermined number range through fast Fourier transform and then using the total harmonic distortion rate formula.

4. The short-circuit current adaptive detection method for a distributed green electricity direct-connection system according to claim 3, characterized in that, The adjustment of the adaptive threshold is based on a preset working condition mapping table, which is obtained through offline simulation and field measurement data training and covers a predetermined number of typical operating scenarios. When the system is running online, the threshold parameter closest to the current working condition is called in real time through the nearest neighbor matching algorithm.

5. The short-circuit current adaptive detection method for a distributed green electricity direct-connection system according to claim 4, characterized in that, The alarm information includes the fault type, occurrence time, location segment, and confidence level. The fault location segment is determined by comparing the phase difference of the current at both ends of each line segment. When the absolute value of the phase difference is greater than a preset angle threshold and the duration exceeds a predetermined time period, it is determined that a short circuit has occurred in the segment.

Images