An internet-of-things cooperative distributed photovoltaic grid-connected abnormal early warning system and method

Abstract

The application discloses an Internet of Things cooperative distributed photovoltaic grid-connected abnormal early warning system and method, and belongs to the technical field of power early warning. The system adopts a three-layer Internet of Things architecture of terminal sensing, edge cooperation and cloud decision. The terminal side realizes synchronous data acquisition and adaptive preliminary screening. The edge side executes reference node iteration purification, virtual reference construction and multi-node cooperative verification. The cloud realizes implicit impedance accurate identification and propagation risk assessment. The method core includes six steps of time-space synchronous data acquisition, single-node adaptive threshold preliminary screening, iterative purification type virtual reference cooperative verification, reactive micro-perturbation increment decoupling implicit impedance identification, risk entropy propagation assessment and multi-dimensional hierarchical early warning. The application solves the reference node self-disturbance caused reference distortion problem, greatly reduces the false positive rate, improves the implicit abnormal identification accuracy, and improves the early warning advance by 2-3 times, and is suitable for large-scale distributed photovoltaic cluster low-voltage grid-connected safety monitoring and active operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of power early warning technology, and in particular to an IoT-coordinated distributed photovoltaic grid connection anomaly early warning system and method. Background Technology

[0002] Currently, residential and commercial distributed photovoltaic (PV) systems on the low-voltage distribution network are experiencing explosive growth, with a large number of grid-connected nodes, wide distribution, and complex operating environments. Grid connection anomalies (including poor terminal contact, line oxidation and aging, impedance mismatch, excessive harmonics, islanding effects, etc.) not only reduce PV power generation efficiency but may also cause safety accidents such as malfunctioning distribution network protection and electrical fires, making it one of the core safety hazards for the large-scale development of distributed PV.

[0003] Existing distributed photovoltaic grid connection early warning technologies mainly suffer from the following technical shortcomings: (1) The benchmark of multi-node collaborative scheme is prone to distortion: Some existing multi-node collaborative schemes directly select adjacent physical nodes as reference benchmarks without verifying the purity of the reference node's own operating status. When the reference node has component aging, partial occlusion or hidden contact anomalies, it will lead to the distortion of the benchmark data, a significant decrease in the accuracy of collaborative verification, and even missed judgments and misjudgments.

[0004] (2) Insufficient ability to identify latent anomalies and high false positive rate: Existing impedance anomaly identification methods are mostly based on harmonic data under steady-state operation to calculate equivalent impedance, which cannot distinguish between apparent impedance shifts caused by photovoltaic power output fluctuations and load changes and real physical impedance anomalies caused by terminal oxidation and line aging. A large number of impedance shifts caused by operating condition fluctuations will be misjudged as anomalies, resulting in a high false positive rate for latent anomaly identification and wasting operation and maintenance resources.

[0005] (3) Logic defects in risk assessment models and passive early warning mode: Existing risk assessments mostly adopt simple weighting methods, and some models that introduce information entropy have problems with physical meaning contradictions, which cannot accurately reflect the positive relationship between fault transmission probability and risk; at the same time, they are generally "post-event alarm" modes, which only trigger alarms after the abnormality worsens, and cannot assess the scope and impact of the abnormality, making it difficult to support proactive operation and maintenance and pre-emptive handling, and unable to adapt to the refined operation and maintenance needs of large-scale photovoltaic clusters. Summary of the Invention

[0006] The purpose of this invention is to provide an IoT-coordinated distributed photovoltaic grid-connected anomaly early warning system and method, which can achieve accurate identification and early warning of grid-connected anomalies, significantly reduce false alarm rate, improve the accuracy of hidden anomaly identification, and support proactive safety operation and maintenance of large-scale distributed photovoltaic systems.

[0007] To achieve the above objectives, the present invention provides an IoT-coordinated distributed photovoltaic grid-connected anomaly early warning system, comprising a terminal sensing layer, an edge collaboration layer, and a cloud platform decision layer that are sequentially connected from top to bottom, and the layers are connected to each other via a power wireless private network or a 5G / 4G public network. The terminal sensing layer includes several IoT sensing terminals deployed on the AC side of each distributed photovoltaic grid-connected point. The IoT sensing terminal integrates an electrical acquisition module, a harmonic analysis module, an environmental acquisition module, a time synchronization module, and an edge computing unit. It is used to synchronously collect electrical data and environmental data of the grid-connected point, perform single-node adaptive preliminary anomaly screening, and support interaction with the photovoltaic inverter to execute composite perturbation commands. The edge collaboration layer includes several regional edge gateways deployed according to low-voltage power supply areas. The regional edge gateway has a built-in reference group purification engine, virtual benchmark fitting module and collaborative verification unit, which are used to aggregate terminal data in the jurisdiction, perform candidate reference node iterative purification, virtual benchmark node construction, multi-node spatiotemporal collaborative verification and anomaly confidence calculation, and send collaborative sampling and composite perturbation control commands to the terminals. The cloud platform decision layer includes an impedance anomaly identification module, a propagation risk assessment module, and a tiered early warning module. It is used to receive abnormal data uploaded by the edge gateway, perform grid connection impedance authenticity verification, hidden anomaly identification, anomaly propagation risk assessment, and tiered early warning issuance, and link with the operation and maintenance management system to form a closed-loop handling mechanism.

[0008] Preferably, the regional edge gateway supports an edge-to-edge collaborative wake-up mechanism. When a single node reports a suspected anomaly, it automatically wakes up all nodes in the candidate reference group to enter the high-frequency sampling mode simultaneously. After the collaborative verification is completed, it automatically restores the normal sampling frequency.

[0009] This invention also provides an IoT-coordinated method for early warning of anomalies in distributed photovoltaic grid connection, comprising the following steps: S1. Deploy IoT sensing terminals at each distributed photovoltaic grid connection point, and achieve microsecond-level spatiotemporal synchronization of all nodes through a time synchronization module to routinely collect electrical and environmental data at the grid connection point; S2. The IoT sensing terminal extracts electrical characteristic quantities locally, completes the initial anomaly screening of a single node based on an adaptive dynamic threshold corrected for ambient temperature, and reports suspected abnormal data to the regional edge gateway. S3. After receiving a suspected abnormal signal, the regional edge gateway performs iterative purity screening on the candidate reference node group to obtain a purified reference group. Then, it constructs a virtual benchmark node based on purity weighting and calculates the anomaly confidence level through the spatiotemporal correlation between the node to be inspected and the virtual benchmark to determine whether it is a confirmed anomaly. S4. For nodes that are confirmed to be abnormal, first calculate the rough comprehensive impedance offset rate based on the natural operating harmonic data, then trigger the composite perturbation sequence of the fundamental reactive power superimposed with the harmonic current of the photovoltaic inverter output, and calculate the real physical impedance by decoupling the electrical increments before and after the perturbation to obtain the effective impedance offset rate and identify the real hidden impedance anomaly. S5 and cloud platforms combine the low-voltage distribution network topology and calculate the risk of abnormal nodes spreading to surrounding nodes and the upper-level distribution network based on an improved risk entropy model. S6. Integrate the abnormal confidence level, effective impedance offset rate and propagation risk level to calculate the comprehensive early warning value and classify the early warning level, issue corresponding early warning instructions and trigger collaborative handling.

[0010] Preferably, in S2, the formula for calculating the adaptive dynamic threshold is: ; In the formula, For the first i The node k Dynamic alarm thresholds for electrical features; This is the average of the data from the past 30 days of normal operation for this feature; This represents the standard deviation of the data from the past 30 days of normal operation for this feature. This is the ambient temperature correction factor, with a value range of [0.8, 1.2], which increases linearly with increasing ambient temperature.

[0011] Preferably, step S3 specifically includes the following steps: S31. From the normal nodes in this area that have not triggered suspected abnormalities, select the 8 nodes that are closest to the node to be inspected as the initial candidate reference group. S32. Calculate the feature purity of each candidate node, iteratively remove contaminated nodes with purity below the preset threshold, until all remaining nodes meet the purity requirements and the number of nodes is not less than 3, forming a purification reference group. S33. Using the purity of each node in the purification reference group as the weight, perform weighted fitting on the feature sequences of each node to generate a standard feature sequence of the virtual benchmark node without local perturbation. S34. For the corresponding electrical characteristics that trigger the anomaly, calculate the spatiotemporal correlation coefficient between the feature sequence of the node under test and the virtual reference feature sequence, and derive the anomaly confidence of the node under test. When the anomaly confidence is greater than or equal to the preset confidence threshold, it is determined to be a confirmed anomaly.

[0012] Preferably, in S32, the formula for calculating the purity of node features is: ; In the formula, For the first jThe feature purity of each candidate reference node, with a value range of [0, 1]; This represents the total number of electrical feature dimensions involved in the verification. Candidate nodes j No. k The current time-period mean of the feature; For all nodes in the current candidate reference group k The mean of the features; For all nodes in the current candidate reference group k The standard deviation of the feature; if the number of nodes in the purified reference group is less than 3 after iteration, the range of candidate nodes is expanded and re-screened; In S33, the formula for calculating the feature sequence of the virtual reference node is: ; ; In the formula, for t The eigenvalues ​​of the virtual reference node at each time step; The total number of nodes in the purification reference group; For the first j The weighting coefficients of each purification reference node; For the first j Characteristic purity of each purification reference node; for t Time of the first j The characteristic instantaneous values ​​of each purified reference node; In S34, the formula for calculating the spatiotemporal correlation coefficient is: ; The formula for calculating the anomaly confidence level is: ; In the formula, Node to be inspected i The spatiotemporal correlation coefficient with the virtual reference node ranges from [0, 1]. Node to be inspected i exist t The corresponding abnormal feature value at any given time; The mean of the feature sequence of the node to be inspected; The mean of the feature sequence of the virtual baseline node; This represents the number of time sampling points for the feature sequence. Node to be inspected i The abnormal confidence level ranges from [0, 1].

[0013] Preferably, step S4 specifically includes the following steps: S41. Based on the voltage and current data of each harmonic under natural operating conditions, calculate the grid-connected equivalent impedance under each harmonic, and then obtain a rough comprehensive impedance deviation rate. S42. When the rough composite impedance offset rate reaches the trigger threshold, the regional edge gateway sends a composite perturbation command to the corresponding photovoltaic inverter, controlling the inverter to output the fundamental reactive perturbation within ±5% of the rated reactive capacity, while superimposing the 3rd, 5th and 7th small harmonic current perturbations. S43. Collect the amplitude and phase changes of each harmonic voltage and current before and after the perturbation, calculate the incremental impedance of the frequency band, and eliminate common-mode interference caused by operating condition fluctuations. S44. Calculate the impedance accuracy coefficient, correct the rough composite impedance deviation rate to obtain the effective composite impedance deviation rate, and determine whether there is a true hidden impedance anomaly.

[0014] Preferably, in S41, the calculation formula for the equivalent impedance of each harmonic connected to the grid is as follows: ; The formula for calculating the overall impedance offset ratio is as follows: ; In the formula, For the first h Equivalent impedance of grid connection under subharmonics; , The first h The amplitude of the second harmonic voltage and current; , The first h Phase angles of subharmonic voltages and currents; This is a rough estimate of the overall impedance offset rate. For the first h Reference impedance under subharmonics; For the first h The weighting coefficients for subharmonics are as follows: higher order harmonics have a higher weighting than lower order harmonics. The amplitude of the fundamental reference impedance; In S43, the formula for calculating the incremental impedance of the frequency band is: ; In the formula, For the first h The incremental impedance under subharmonics characterizes the true physical impedance characteristics of the grid-connected circuit. , These are the numbers before and after the perturbation. h The magnitude change of the second harmonic voltage and current; , These are the numbers before and after the perturbation. h Phase changes of subharmonic voltage and current; In S44, the formula for calculating the impedance accuracy coefficient is: ; The formula for calculating the effective composite impedance offset is: ; In the formula, The impedance accuracy coefficient has a value range of [0, 1]. The closer the value is to 1, the higher the proportion of impedance deviation caused by real physical changes. To achieve an effective overall impedance offset rate; This is a rough estimate of the overall impedance offset rate.

[0015] Preferably, in S6, the formula for calculating the comprehensive early warning value is: ; In the formula, The comprehensive early warning value ranges from [0, 1]. , , These are the weighting coefficients for the anomaly confidence level, effective impedance offset rate, and propagation risk level, respectively, and the sum of the three is 1; , These are the normalized baseline values ​​for impedance offset rate and propagation risk, respectively.

[0016] Preferably, in S5, the formula for calculating the risk of abnormal propagation is: ; In the formula, The abnormal propagation risk level has a value range of [0, 1]. The average failure propagation probability across all propagation paths; For the first k The probability of fault propagation along a propagation path is determined by both line impedance and node capacity matching. Total number of transmission paths; , These are the conduction strength weight and uncertainty weight, respectively, with default values ​​of 0.6 and 0.4. The capacity impact coefficient is the smaller value between the node capacity percentage and 0.1. This is the topology coupling coefficient; the closer the node is to the transformer in the distribution area, the larger the value.

[0017] Therefore, the beneficial effects of the above-mentioned IoT-coordinated distributed photovoltaic grid-connected anomaly early warning system and method are as follows: (1) This invention effectively eliminates contaminated nodes affected by local occlusion and component aging interference through candidate reference node iterative purification and virtual benchmark construction technology, and eliminates the distortion problem of the reference benchmark itself. Compared with the traditional multi-node average verification scheme, it significantly improves the accuracy of collaborative verification and greatly reduces the overall false alarm rate. (2) This invention can accurately distinguish between apparent impedance deviation caused by operating condition fluctuations and real physical contact anomalies by utilizing only the inverter's own control capabilities, thereby improving the accuracy of hidden anomaly identification and significantly reducing invalid maintenance work orders.

[0018] (3) The improved risk entropy model of this invention integrates fault conduction intensity and propagation uncertainty, with clear physical meaning and reasonable numerical range, which solves the logical defect of the traditional entropy model where risk and probability are reversed. Combined with the monitoring of the gradual change of multi-harmonic impedance, it can detect gradual hidden anomalies such as increased contact resistance and line aging 2-3 times in advance, realizing the upgrade from passive alarm to active prediction.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the structure of an embodiment of the distributed photovoltaic grid-connected anomaly early warning system with Internet of Things collaboration according to the present invention; Figure 2 This is a schematic diagram illustrating the steps of an embodiment of the distributed photovoltaic grid-connected anomaly early warning method with Internet of Things collaboration according to the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0023] Example 1: like Figure 1As shown, this embodiment provides an IoT-coordinated distributed photovoltaic grid-connected anomaly early warning system, which includes a terminal sensing layer, an edge collaboration layer, and a cloud platform decision layer that are connected in a top-to-bottom communication manner, and the layers are connected to each other through a power wireless private network.

[0024] The terminal sensing layer consists of several IoT sensing terminals deployed on the AC side of each distributed photovoltaic grid-connected point. Each terminal integrates a voltage transformer, a current transformer, a harmonic analysis module, an ambient temperature sensor, a component temperature sensor, a BeiDou / GPS dual-mode time synchronization module, and a local edge computing unit. The core functions of the terminal include: electrical and environmental data acquisition under microsecond-level time synchronization, local feature extraction and adaptive initial screening, dynamic switching of sampling frequency, and command interaction with the photovoltaic inverter. Under normal conditions, the terminal collects data at a frequency of 100Hz; when a suspected anomaly is detected, it automatically increases the sampling frequency to 2kHz and receives commands from the regional edge gateway to perform composite perturbation coordination.

[0025] The edge collaboration layer consists of regional edge gateways deployed according to low-voltage power supply areas. A single gateway can manage 5-50 photovoltaic terminal nodes within its jurisdiction. The gateway integrates a reference group purification engine, a virtual benchmark fitting module, a collaborative verification unit, and an inverter control interface. Core functions include: data aggregation and timing calibration of nodes within the jurisdiction, iterative purification of candidate reference nodes, construction of virtual benchmark nodes, anomaly confidence calculation, edge-end collaborative sampling scheduling, and issuance of composite perturbation commands. Over 90% of routine data and preliminary anomaly verification are completed at the edge, with only confirmed anomalies uploaded to the cloud, significantly reducing communication bandwidth consumption.

[0026] The cloud platform's decision-making layer is deployed at the regional photovoltaic operation and maintenance center, and includes modules for impedance anomaly identification, propagation risk assessment, tiered early warning, and operation and maintenance linkage interfaces. Core functions include: global distribution network topology management, in-depth identification of latent impedance anomalies, anomaly propagation risk entropy assessment, multi-dimensional tiered early warning issuance, automatic generation of operation and maintenance work orders, and closed-loop management of handling. The cloud platform also features historical data storage, model iteration optimization, and global situation visualization capabilities.

[0027] like Figure 2 As shown, the early warning method based on the above-mentioned photovoltaic grid connection anomaly early warning system specifically includes the following steps: S1. IoT node deployment and spatiotemporal synchronized data acquisition: IoT sensing terminals are deployed on the AC incoming line side of each distributed photovoltaic grid-connected point. A BeiDou / GPS dual-mode time synchronization module achieves microsecond-level time synchronization across all nodes in the distribution area, with a synchronization accuracy better than 1μs. In normal operation, the terminals collect data at a sampling frequency of 100Hz, including instantaneous three-phase voltage, instantaneous three-phase current, active power, reactive power, amplitude and phase of harmonics from the 2nd to the 25th orders, ambient temperature, and photovoltaic module surface temperature. When a suspected anomaly is detected by the terminal's local initial screening, an abnormal status is automatically triggered, increasing its sampling frequency to 2kHz and reporting the suspected anomaly signal to the regional edge gateway.

[0028] S2. Terminal-side single-node feature extraction and adaptive threshold initial screening: The IoT sensing terminal preprocesses the collected data locally, extracting core electrical characteristics, including voltage deviation rate, three-phase current imbalance, total harmonic distortion rate of voltage, total harmonic distortion rate of current, and power factor deviation. Based on the node's nearly 30 days of normal operation history data, an adaptive dynamic threshold that is corrected for ambient temperature is constructed to complete the initial anomaly screening.

[0029] The formula for calculating the adaptive dynamic threshold is: In the formula, For the first i The node k Dynamic alarm thresholds for electrical features; This is the average of the data from the past 30 days of normal operation for this feature; This represents the standard deviation of the data from the past 30 days of normal operation for this feature. This is the ambient temperature correction factor, with a value range of [0.8, 1.2]. It increases linearly with increasing ambient temperature and is used to offset the effects of temperature changes on conductor resistance and electrical equipment parameters.

[0030] When any feature of a node exceeds the corresponding dynamic threshold When this happens, the terminal marks the node as "suspected abnormal" and uploads the high-frequency sampling data for the corresponding time period to the regional edge gateway.

[0031] S3. Collaborative verification of regional edge nodes and calculation of anomaly confidence: An iterative purification-based virtual benchmark collaborative verification mechanism is adopted to solve the benchmark distortion problem caused by the self-disturbance of traditional physical reference nodes. The specific steps include: S31, Preliminary selection of candidate reference group: Upon receiving a suspected abnormal signal, the regional edge gateway, based on the distribution network topology database, selects eight nodes from the normal nodes within its area that have not triggered any suspected anomalies and have no abnormal records in the past 24 hours. These nodes, with the closest electrical distance to the node to be inspected, form an initial candidate reference group. The electrical distance is calculated based on the distribution network line length, wire diameter, and topology connection relationship.

[0032] S32, Iterative purity screening: Calculate the feature purity of each node within the initial candidate group. Purity reflects the degree to which the node's operating state deviates from the normal level of the population. Set a purity threshold. =0.6, remove contaminated nodes with purity below the threshold; the remaining nodes are regrouped into a candidate group, purity is recalculated and iteratively removed until the purity of all nodes in the candidate group is greater than or equal to the threshold and the number of nodes is not less than 3. The final purified reference group is obtained, and the number of nodes is denoted as . If fewer than three nodes remain after the iteration, the candidate node range is expanded to the four nodes with the second closest electrical distance, and the iterative screening is performed again.

[0033] Formula for calculating node feature purity: In the formula, For the first j The feature purity of each candidate reference node ranges from [0, 1]. A higher value indicates that the node's operating state is closer to the normal level of the population. This represents the total number of electrical feature dimensions involved in the verification. Candidate nodes j No. k The current time-period mean of the feature; For all nodes in the current candidate reference group k The mean of the features; For all nodes in the current candidate reference group k The standard deviation of the feature.

[0034] S33. Construction of Virtual Reference Nodes: Based on the purity of each node in the purified reference group, a virtual benchmark node without local perturbation is constructed through weighted fitting. Its feature sequence serves as a unified benchmark for collaborative verification, replacing the average verification of traditional discrete physical nodes, thus eliminating the interference of local perturbation of single nodes from the root.

[0035] The formula for calculating the feature sequence of a virtual reference node is as follows: In the formula, for tThe eigenvalues ​​of the virtual reference node at each time step; The total number of nodes in the purification reference group; For the first j The weighting coefficients of each purification reference node; For the first j Characteristic purity of each purification reference node; for t Time of the first j The characteristic instantaneous value of a purified reference node.

[0036] S34. Anomaly Confidence Calculation: For the corresponding electrical characteristics that trigger suspected anomalies, the spatiotemporal correlation coefficient between the feature sequence of the node under investigation and the feature sequence of the virtual reference node is calculated, and then further fused to obtain the anomaly confidence level. The higher the spatiotemporal correlation coefficient, the stronger the consistency between the operating patterns of the node under investigation and the normal group, and the lower the probability of an anomaly.

[0037] Formula for calculating spatiotemporal correlation coefficient: In the formula, Node to be inspected i The spatiotemporal correlation coefficient with the virtual reference node ranges from [0, 1]. Node to be inspected i exist t The corresponding abnormal feature value at any given time; The mean of the feature sequence of the node to be inspected; The mean of the feature sequence of the virtual baseline node; This represents the number of time sampling points for the feature sequence. The formula for calculating the anomaly confidence level is: Node to be inspected i The anomaly confidence level ranges from [0, 1], with a higher value indicating a higher probability of an anomaly.

[0038] Set confidence threshold =0.7, when When the value is ≥0.7, it is judged as "confirmed anomaly" and proceeds to the next step of deep identification of latent anomalies; when When the value is less than 0.7, it is determined to be a false trigger caused by environmental disturbance. The abnormal state is then lifted, and the terminal resumes normal sampling.

[0039] S4. Dynamic estimation of grid-connected equivalent impedance and identification of latent anomalies: A composite perturbation incremental decoupling method combining fundamental reactive power and harmonic current is adopted to solve the problems of natural harmonic impedance estimation being easily affected by power output fluctuations and being unable to distinguish between apparent deviations and actual physical anomalies. The specific steps include: S41. Rough impedance estimation based on natural operating harmonics: For nodes that are confirmed to be abnormal, the equivalent impedance of the grid connection under each harmonic is first calculated using the voltage, current amplitude and phase data of each harmonic under natural operating conditions. Then, it is compared with the reference impedance to obtain a rough comprehensive impedance deviation rate.

[0040] Formulas for calculating the equivalent impedance of the grid under each harmonic: Rough formula for calculating overall impedance offset: In the formula, For the first h Equivalent impedance of grid connection under subharmonics; , The first h The amplitude of the second harmonic voltage and current; , The first h Phase angles of subharmonic voltages and currents; This is a rough estimate of the overall impedance offset rate. For the first h The reference impedance under subharmonics is the historical average value under normal operating conditions during the initial commissioning of the node. For the first h The weighting coefficient of higher harmonics is higher than that of lower harmonics because poor contact and aging of the line are more sensitive to changes in impedance of higher harmonics. The amplitude of the fundamental reference impedance is used for normalization.

[0041] Set the perturbation trigger threshold to 10%, when When the value is ≥10%, the composite perturbation precision verification process is triggered.

[0042] S42, Inverter composite perturbation triggering: The regional edge gateway sends multi-band composite perturbation commands to the photovoltaic inverters of the nodes under inspection, controlling the inverters to output fundamental reactive power perturbations within ±5% of their rated reactive power capacity. Simultaneously, it superimposes 3rd, 5th, and 7th order small harmonic current perturbations, keeping the total harmonic distortion rate within the national standard's allowable range. The perturbation amplitude is strictly controlled, ensuring no impact on normal grid power supply or photovoltaic active power output throughout the process. The perturbation lasts for 5 power frequency cycles, with the terminal maintaining 2kHz high-frequency sampling, simultaneously recording electrical data for 10 cycles before and after the perturbation.

[0043] S43. Incremental impedance decoupling calculation: The amplitude and phase changes of each harmonic voltage and current before and after the perturbation are collected. The common-mode interference caused by photovoltaic power output fluctuations and load changes is eliminated by incremental calculation, and the incremental impedance that only reflects the true physical characteristics of the grid-connected circuit is obtained.

[0044] Formula for calculating frequency band incremental impedance: In the formula, For the first h The incremental impedance under subharmonics characterizes the true physical impedance characteristics of the grid-connected circuit. , These are the numbers before and after the perturbation. h The magnitude change of the second harmonic voltage and current; , These are the numbers before and after the perturbation. h The phase change of subharmonic voltage and current.

[0045] S44. Calculation of Effective Impedance Offset Rate and Anomaly Detection: Define an impedance accuracy coefficient to quantify the proportion of the actual physical impedance change in the coarse impedance deviation, correct the coarse deviation rate, and obtain an effective comprehensive impedance deviation rate to eliminate misjudgments caused by operating condition fluctuations.

[0046] Formula for calculating impedance accuracy coefficient: In the formula, This is the impedance accuracy coefficient, with a value range of [0, 1]. The closer the value is to 1, the higher the proportion of impedance deviation caused by changes in actual physical contact; the lower the value, the more the impedance deviation is caused by fluctuations in operating conditions. When the calculated result is greater than 1, it is taken as 1 to ensure the reasonableness of the physical meaning.

[0047] The formula for calculating the effective composite impedance offset is: In the formula, To effectively synthesize impedance offset rate, This is a rough estimate of the overall impedance offset rate.

[0048] The threshold for identifying latent anomalies is set at 15%. When the value is ≥15%, the node is determined to have a real hidden abnormality in grid connection impedance. Abnormality types include poor contact of wiring terminals, oxidation and aging of lines, loose connections, etc.

[0049] S5. Calculation of anomaly propagation risk degree based on improved risk entropy: The cloud platform, combined with the low-voltage distribution network topology, introduces an improved risk entropy model that integrates fault propagation intensity and propagation uncertainty. It calculates the risk level of fault propagation from abnormal nodes to surrounding nodes and the upstream distribution network, supporting tiered early warning and proactive maintenance priority ranking. The propagation path refers to the electrical path through which an abnormal node propagates a fault to adjacent nodes and transformer substations via distribution network lines; the number of paths is determined by the distribution network topology.

[0050] Formula for calculating the risk of abnormal propagation: In the formula, The abnormal propagation risk level has a value range of [0, 1]. The average failure propagation probability across all propagation paths; For the first k The probability of fault propagation along a propagation path is determined by both line impedance and node capacity matching. Total number of transmission paths; , These are the conduction strength weight and uncertainty weight, respectively, with default values ​​of 0.6 and 0.4. This is the capacity influence coefficient. , The rated capacity of abnormal nodes is the proportion of the total photovoltaic capacity of the area. When the proportion reaches 10%, the capacity impact is saturated to avoid the risk value of small capacity nodes being too low. is the topology coupling coefficient, with a value range of [0.3, 1]. The closer the node is to the transformer in the distribution area, the larger the value, and the stronger the impact on the upstream distribution network.

[0051] Based on the risk level, the risk is divided into three levels: low risk (… <0.3), medium risk (0.3≤ <0.7), high risk ( ≥0.7).

[0052] S6. Multi-dimensional integrated hierarchical early warning and collaborative response closed loop: By integrating three core dimensions—abnormal confidence level, effective impedance offset rate, and propagation risk—a comprehensive early warning value is calculated. Based on the comprehensive early warning value, early warning levels are classified, and corresponding response strategies are triggered.

[0053] Comprehensive warning value calculation formula: In the formula, The comprehensive early warning value ranges from [0, 1]. , , These are the weighting coefficients for the anomaly confidence level, effective impedance offset rate, and propagation risk level, respectively, and the sum of the three is 1; , These are the normalized baseline values ​​for impedance offset rate and propagation risk, respectively.

[0054] The status and warning levels are classified as follows: Normal state: If the value is less than 0.4, no warning will be triggered; continue routine monitoring. Level III (General Warning): 0.4≤ If the value is less than 0.6, a patrol reminder will be sent to the maintenance personnel, and on-site verification is recommended within 72 hours. Level II (Important Warning): 0.6≤ When the value is less than 0.8, the regional edge gateway will link with the corresponding photovoltaic inverter to perform current limiting operation (limiting the output to 70% of the rated value), and at the same time generate a priority maintenance work order to arrange on-site handling within 24 hours; Level I (Emergency Warning): If the value is ≥0.8, the regional edge gateway will directly trigger an emergency shutdown command for the photovoltaic inverter, and at the same time, it will link the distribution network protection device to push an emergency alarm to the operation and maintenance duty center and immediately arrange on-site repairs.

[0055] After the handling is completed, the operation and maintenance personnel will backfill the handling results to the cloud platform, and the system will automatically update the node baseline data and model parameters, forming a complete closed loop of early warning-handling-optimization.

[0056] Therefore, this invention adopts the above-mentioned IoT-coordinated distributed photovoltaic grid-connected anomaly early warning system and method. It solves the reference distortion problem caused by the perturbation of the reference node itself through the reference group purification mechanism, distinguishes between apparent impedance offset and real physical anomaly through reactive power micro-perturbation incremental decoupling technology, and realizes quantitative assessment of anomaly propagation risk through an improved risk entropy model. It can significantly reduce the false alarm rate, improve the accuracy of hidden anomaly identification, and increase the early warning lead time by 2-3 times. It is suitable for low-voltage grid-connected safety monitoring and proactive operation and maintenance of large-scale distributed photovoltaic clusters.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An IoT-coordinated distributed photovoltaic grid-connected anomaly early warning system, characterized in that: It includes the terminal perception layer, the edge collaboration layer, and the cloud platform decision layer, which are connected in a top-down manner. The terminal sensing layer includes several IoT sensing terminals deployed on the AC side of each distributed photovoltaic grid-connected point. The IoT sensing terminal integrates an electrical acquisition module, a harmonic analysis module, an environmental acquisition module, a time synchronization module, and an edge computing unit. It is used to synchronously collect electrical data and environmental data of the grid-connected point, perform single-node adaptive preliminary anomaly screening, and support interaction with the photovoltaic inverter to execute composite perturbation commands. The edge collaboration layer includes several regional edge gateways deployed according to low-voltage power supply areas. The regional edge gateway has a built-in reference group purification engine, virtual benchmark fitting module and collaborative verification unit, which are used to aggregate terminal data in the jurisdiction, perform candidate reference node iterative purification, virtual benchmark node construction, multi-node spatiotemporal collaborative verification and anomaly confidence calculation, and send collaborative sampling and composite perturbation control commands to the terminals. The cloud platform decision layer includes an impedance anomaly identification module, a propagation risk assessment module, and a tiered early warning module. It is used to receive abnormal data uploaded by the edge gateway, perform grid connection impedance authenticity verification, hidden anomaly identification, anomaly propagation risk assessment, and tiered early warning issuance, and link with the operation and maintenance management system to form a closed-loop handling mechanism.

2. The distributed photovoltaic grid-connected anomaly early warning system with IoT collaboration according to claim 1, characterized in that: The regional edge gateway supports an edge-to-end collaborative wake-up mechanism. When a single node reports a suspected anomaly, it automatically wakes up all nodes in the candidate reference group to enter the high-frequency sampling mode simultaneously. After the collaborative verification is completed, it automatically restores the normal sampling frequency.

3. A distributed photovoltaic grid-connected anomaly early warning method based on Internet of Things (IoT) collaboration, implemented using the early warning system described in any one of claims 1-2, characterized in that, Includes the following steps: S1. Deploy IoT sensing terminals at each distributed photovoltaic grid connection point, and achieve microsecond-level spatiotemporal synchronization of all nodes through a time synchronization module to routinely collect electrical and environmental data at the grid connection point; S2. The IoT sensing terminal extracts electrical characteristic quantities locally, completes the initial anomaly screening of a single node based on an adaptive dynamic threshold corrected for ambient temperature, and reports suspected abnormal data to the regional edge gateway. S3. After receiving a suspected abnormal signal, the regional edge gateway performs iterative purity screening on the candidate reference node group to obtain a purified reference group. Then, it constructs a virtual benchmark node based on purity weighting and calculates the anomaly confidence level through the spatiotemporal correlation between the node to be inspected and the virtual benchmark to determine whether it is a confirmed anomaly. S4. For nodes that are confirmed to be abnormal, first calculate the rough comprehensive impedance offset rate based on the natural operating harmonic data, then trigger the composite perturbation sequence of the fundamental reactive power superimposed with the harmonic current of the photovoltaic inverter output, and calculate the real physical impedance by decoupling the electrical increments before and after the perturbation to obtain the effective impedance offset rate and identify the real hidden impedance anomaly. S5 and cloud platforms combine the low-voltage distribution network topology and calculate the risk of abnormal nodes spreading to surrounding nodes and the upper-level distribution network based on an improved risk entropy model. S6. Integrate the abnormal confidence level, effective impedance offset rate and propagation risk level to calculate the comprehensive early warning value and classify the early warning level, issue corresponding early warning instructions and trigger collaborative handling.

4. The method for early warning of distributed photovoltaic grid-connected anomalies in IoT collaboration according to claim 3, characterized in that: In S2, the formula for calculating the adaptive dynamic threshold is: ； In the formula, For the first i The node k Dynamic alarm thresholds for electrical features; This is the average of the data from the past 30 days of normal operation for this feature; This represents the standard deviation of the data from the past 30 days of normal operation for this feature. This is the ambient temperature correction factor, with a value range of [0.8, 1.2], which increases linearly with increasing ambient temperature.

5. The method for early warning of distributed photovoltaic grid-connected anomalies in IoT collaboration according to claim 3, characterized in that: S3 specifically includes the following steps: S31. From the normal nodes in this area that have not triggered suspected abnormalities, select the 8 nodes that are closest to the node to be inspected as the initial candidate reference group. S32. Calculate the feature purity of each candidate node, iteratively remove contaminated nodes with purity below the preset threshold, until all remaining nodes meet the purity requirements and the number of nodes is not less than 3, forming a purification reference group. S33. Using the purity of each node in the purification reference group as the weight, perform weighted fitting on the feature sequences of each node to generate a standard feature sequence of the virtual benchmark node without local perturbation. S34. For the corresponding electrical characteristics that trigger the anomaly, calculate the spatiotemporal correlation coefficient between the feature sequence of the node under test and the virtual reference feature sequence, and derive the anomaly confidence of the node under test. When the anomaly confidence is greater than or equal to the preset confidence threshold, it is determined to be a confirmed anomaly.

6. The method for early warning of distributed photovoltaic grid-connected anomalies in IoT collaboration according to claim 5, characterized in that: In S32, the formula for calculating the purity of node features is: ； In the formula, For the first j The feature purity of each candidate reference node, with a value range of [0, 1]; This represents the total number of electrical feature dimensions involved in the verification. Candidate nodes j No. k The current time-period mean of the feature; For all nodes in the current candidate reference group k The mean of the features; For all nodes in the current candidate reference group k The standard deviation of the feature; If the number of nodes in the purified reference group is less than 3 after iteration, the range of candidate nodes is expanded and the screening is repeated. In S33, the formula for calculating the feature sequence of the virtual reference node is: ； ； In the formula, for t The eigenvalues ​​of the virtual reference node at each time step; The total number of nodes in the purification reference group; For the first j The weighting coefficients of each purification reference node; For the first j Characteristic purity of each purification reference node; for t Time of the first j The characteristic instantaneous values ​​of each purified reference node; In S34, the formula for calculating the spatiotemporal correlation coefficient is: ； The formula for calculating the anomaly confidence level is: ； In the formula, Node to be inspected i The spatiotemporal correlation coefficient with the virtual reference node ranges from [0, 1]. Node to be inspected i exist t The corresponding abnormal feature value at any given time; The mean of the feature sequence of the node to be inspected; The mean of the feature sequence of the virtual baseline node; This represents the number of time sampling points for the feature sequence. Node to be inspected i The abnormal confidence level ranges from [0, 1].

7. The method for early warning of distributed photovoltaic grid-connected anomalies in IoT collaboration according to claim 3, characterized in that: S4 specifically includes the following steps: S41. Based on the voltage and current data of each harmonic under natural operating conditions, calculate the grid-connected equivalent impedance under each harmonic, and then obtain a rough comprehensive impedance deviation rate. S42. When the rough composite impedance offset rate reaches the trigger threshold, the regional edge gateway sends a composite perturbation command to the corresponding photovoltaic inverter, controlling the inverter to output the fundamental reactive perturbation within ±5% of the rated reactive capacity, while superimposing the 3rd, 5th and 7th small harmonic current perturbations. S43. Collect the amplitude and phase changes of each harmonic voltage and current before and after the perturbation, calculate the incremental impedance of the frequency band, and eliminate common-mode interference caused by operating condition fluctuations. S44. Calculate the impedance accuracy coefficient, correct the rough composite impedance deviation rate to obtain the effective composite impedance deviation rate, and determine whether there is a true hidden impedance anomaly.

8. The method for early warning of distributed photovoltaic grid-connected anomalies in IoT collaboration according to claim 7, characterized in that: In S41, the formula for calculating the equivalent impedance of each harmonic in the grid is as follows: ； The rough formula for calculating the overall impedance offset is: ； In the formula, For the first h Equivalent impedance of grid connection under subharmonics; , The first h The amplitude of the second harmonic voltage and current; , The first h Phase angles of subharmonic voltages and currents; This is a rough estimate of the overall impedance offset rate. For the first h Reference impedance under subharmonics; For the first h The weighting coefficients for subharmonics are as follows: higher order harmonics have a higher weighting than lower order harmonics. The amplitude of the fundamental reference impedance; In S43, the formula for calculating the incremental impedance of the frequency band is: ； In the formula, For the first h The incremental impedance under subharmonics characterizes the true physical impedance characteristics of the grid-connected circuit. , These are the numbers before and after the perturbation. h The magnitude change of the second harmonic voltage and current; , These are the numbers before and after the perturbation. h Phase changes of subharmonic voltage and current; In S44, the formula for calculating the impedance accuracy coefficient is: ； The formula for calculating the effective composite impedance offset is: ； In the formula, is the impedance accuracy coefficient, with a value range of [0, 1]; To achieve an effective overall impedance offset rate; This is a rough estimate of the overall impedance offset rate.

9. The method for early warning of distributed photovoltaic grid-connected anomalies in IoT collaboration according to claim 8, characterized in that: In S6, the formula for calculating the comprehensive early warning value is: ； In the formula, The comprehensive early warning value ranges from [0, 1]. , , These are the weighting coefficients for the anomaly confidence level, effective impedance offset rate, and propagation risk level, respectively, and the sum of the three is 1; , These are the normalized baseline values ​​for impedance offset rate and propagation risk, respectively.

10. The method for early warning of distributed photovoltaic grid-connected anomalies in IoT collaboration according to claim 3, characterized in that: In S5, the formula for calculating the risk of abnormal propagation is: ； In the formula, The abnormal propagation risk level has a value range of [0, 1]. The average failure propagation probability across all propagation paths; For the first k The probability of fault propagation along a propagation path is determined by both line impedance and node capacity matching. Total number of transmission paths; , These are the conduction strength weight and uncertainty weight, respectively, with default values ​​of 0.6 and 0.

4. This is the capacity influence coefficient; This is the topology coupling coefficient; the closer the node is to the transformer in the distribution area, the larger the value.

Images