A distributed monitoring and control method for photovoltaic power stations

Abstract

The present invention discloses a distributed monitoring and control method for photovoltaic power stations, which belongs to the technical field of photovoltaic power station monitoring and specifically includes the following steps: S1. When any local monitoring unit detects an abnormal fluctuation of a parameter that meets a preset trigger condition, it is marked as an event initiating unit; S2. The event initiating unit uses a preset topological relationship diagram to take other local monitoring units within a set surrounding range as associated monitoring units and sends a collaborative diagnosis activation instruction; S3. The associated monitoring units perform enhanced data collection and generate a lightweight pre-diagnosis report in combination with local historical operating mode data; S4. Each associated monitoring unit sends a report to the event initiating unit, which constructs a propagation characteristic model of the abnormal fluctuation; S5. A preliminary disposal instruction is generated according to the abnormal fluctuation event type output by the model, and sent to the affected local monitoring unit for execution; the present invention improves the flexibility and reliability of photovoltaic power stations in complex operating environments.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power station monitoring, and in particular to a distributed monitoring and control method for a photovoltaic power station. Background Art

[0002] As the global energy transition accelerates, the large-scale deployment of photovoltaic power plants, especially distributed photovoltaic systems, is becoming a trend. Distributed photovoltaic power plants offer the advantages of flexible consumption and localized power supply by distributing photovoltaic strings across diverse locations, such as rooftops, mountainous terrain, and water surfaces. However, they also face challenges such as wide equipment distribution, strong environmental interference, and complex electrical connections. Efficient monitoring and control are key to ensuring safe power plant operation and improving power generation efficiency. The key lies in real-time detection of local anomalies, accurate assessment of their impact, and dynamic adjustment of control strategies.

[0003] Currently, photovoltaic power station monitoring and control technologies are mainly based on centralized polling and fixed partitioning modes. In the centralized polling mode, the central management node collects data from all devices at a fixed period, determines abnormal conditions based on preset thresholds, and triggers an alarm and executes disposal instructions once a single parameter exceeds the set range. This mainly relies on the central node's periodic polling to collect data, resulting in delayed abnormal response and a lack of correlation analysis of the operating parameters of surrounding equipment, making it difficult to identify regional transmission risks caused by local abnormalities. The fixed partitioning mode divides the power station into multiple fixed areas based on physical location or electrical hierarchy. Data interaction and collaborative control are limited to each area, severing the natural electrical correlation and spatial proximity between cross-regional devices. It is difficult to identify the diffusion path of abnormalities between different areas, which in turn affects the efficiency of fault location and disposal. Summary of the Invention

[0004] The purpose of the present invention is to provide a distributed monitoring and control method for a photovoltaic power station to solve the following technical problems:

[0005] Current photovoltaic power station monitoring and control technologies are mainly based on centralized polling and fixed partitioning modes, relying on central nodes to periodically poll and collect data. This makes it difficult to identify the diffusion path of anomalies between different areas and cannot achieve distributed collaborative diagnosis.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] A distributed monitoring and control method for a photovoltaic power station, wherein the photovoltaic power station includes a plurality of dispersed photovoltaic strings and corresponding local monitoring units, comprises the following steps:

[0008] S1. Each local monitoring unit continuously monitors the real-time operating parameters of the corresponding PV string. When any local monitoring unit detects an abnormal fluctuation in the real-time operating parameters that meets a preset trigger condition, the local monitoring unit is marked as an event initiating unit.

[0009] S2. The event initiating unit selects other local monitoring units within a set range around the location as associated monitoring units according to a preset photovoltaic power station topology diagram, and sends a collaborative diagnosis activation instruction to all associated monitoring units;

[0010] S3. Each associated monitoring unit that receives the collaborative diagnosis activation instruction performs enhanced data collection and generates a lightweight pre-diagnosis report containing a device status label and anomaly confidence level based on the data collection results and the stored local historical operation mode data;

[0011] S4. Each associated monitoring unit sends the lightweight pre-diagnosis report to the event initiation unit. The event initiation unit constructs a propagation characteristic model of the abnormal fluctuation within a set range of the surrounding area based on the summarized lightweight pre-diagnosis report and its own abnormal fluctuation parameters;

[0012] S5. Generate a preliminary handling instruction according to the abnormal fluctuation event type output by the propagation characteristic model, and the event initiating unit sends the preliminary handling instruction to the affected local monitoring unit for execution.

[0013] As a further solution of the present invention: in S1, the abnormal fluctuation of the preset trigger condition is specifically:

[0014] The absolute value of the rate of change of the real-time operating parameter in n consecutive set sampling periods exceeds the first set threshold, and the absolute value of the rate of change continues to remain above the second set threshold in the subsequent n set sampling periods;

[0015] Wherein, n is a natural number, a fixed proportional relationship exists between the first set threshold and the second set threshold, and the first set threshold is greater than the second set threshold; the direction of the change rate remains consistent within the n consecutive set sampling periods, and the duration of the set sampling period is adaptively determined by the electrical time constant of the photovoltaic string.

[0016] As a further solution of the present invention: in S2, the photovoltaic power station topology diagram is specifically:

[0017] During the PV power station deployment phase, record the physical installation coordinates of each local monitoring unit and the electrical connection path of the DC bus branch to which it belongs;

[0018] The Euclidean distance between any two local monitoring units is calculated based on the physical installation coordinates; the hierarchical depth of each local monitoring unit in the DC bus network is analyzed based on the electrical connection path; the Euclidean distance and the hierarchical depth are combined to generate a topological weight coefficient; the topological weight coefficient represents the spatial and electrical correlation strength between any two local monitoring units; the topological relationship diagram is stored as a graph structure with local monitoring units as nodes and topological weight coefficients as edge attributes.

[0019] As a further solution of the present invention: in S3, the enhanced data collection and historical operation mode data are specifically:

[0020] The enhanced data acquisition method is to increase the sampling frequency to an integer multiple of the conventional monitoring frequency during the validity period of the collaborative diagnosis activation instruction; synchronously collect the output current and output voltage parameters of the photovoltaic strings, and perform time alignment processing on the collected current and voltage parameter sequences;

[0021] The local historical operation mode data is a cluster of current-voltage characteristic curves of the photovoltaic string under the same solar irradiance range as the current environment, and the cluster of current-voltage characteristic curves includes a standard operating curve and an allowable deviation boundary.

[0022] As a further solution of the present invention: in S3, the process of generating a lightweight pre-diagnosis report including the device status label and the abnormality confidence level is as follows:

[0023] Mapping the current-voltage parameter sequence obtained by enhanced data acquisition to the current-voltage characteristic curve cluster space in the local historical operation mode data;

[0024] Calculating the dynamic time warping distance between the current-voltage parameter sequence and the standard working curve; assigning a device status label based on whether the dynamic time warping distance exceeds the allowable deviation boundary of the characteristic curve cluster; the device status label includes an in-boundary state label, a boundary fluctuation state label, and an out-boundary state label;

[0025] Based on the ratio of the number of times and duration that the current parameter sequence crosses the allowable deviation boundary, an abnormality confidence value is calculated; the lightweight pre-diagnosis report includes a device status label, an abnormality confidence value and a local monitoring unit identifier.

[0026] As a further solution of the present invention: the process of constructing the propagation characteristic model of abnormal fluctuations within the surrounding set range in S4 is:

[0027] Extracting the device status labels and abnormal confidence values ​​from all lightweight pre-diagnosis reports; establishing a spatial coordinate system centered on the event initiating unit based on the preset photovoltaic power station topology diagram;

[0028] In the spatial coordinate system, the abnormal confidence value of each associated monitoring unit is converted into a spatial density field value; the spatial density field value decays exponentially as the Euclidean distance between the associated monitoring unit and the event initiating unit increases;

[0029] The spatial density field values ​​of all associated monitoring units are superimposed to generate a spatial density distribution cloud map of abnormal fluctuations. At the same time, the hierarchical depth of each associated monitoring unit in the DC convergence path is determined, and the gradient slope of the abnormal confidence value changing with the hierarchical depth is calculated;

[0030] Combining the morphological features of the spatial density distribution cloud map with the sign and absolute value of the gradient slope, a propagation characteristic model parameter set is output.

[0031] As a further solution of the present invention: in S5, the process of the propagation characteristic model determining the type of abnormal fluctuation event is as follows:

[0032] If the spatial density distribution cloud graph presents a single peak with the peak located at the coordinates of the event initiating unit, and the gradient slope is zero, it is determined to be a highly localized independent device event;

[0033] If the spatial density distribution cloud diagram presents a multi-peak morphology or a continuous diffusion morphology, and the gradient slope is a negative value, it is determined to be a regional event with positive electrical propagation characteristics;

[0034] If the spatial density distribution cloud diagram presents an irregular distribution pattern and the gradient slope is positive, it is determined to be a sign of a systemic event with the characteristics of reverse electrical propagation; the positive electrical propagation refers to the diffusion of abnormal fluctuations along the direction of normal current, and the reverse electrical propagation refers to the diffusion of abnormal fluctuations in the opposite direction of current.

[0035] As a further solution of the present invention: in S5, the process of generating the preliminary handling instruction is:

[0036] When it is determined to be a highly localized independent device event, the initial handling instructions only include the output limit operation of the PV strings associated with the event initiating unit;

[0037] When it is determined to be a regional event with forward electrical propagation, the instruction includes the global power reduction coefficient of the bus branch where the event initiating unit and all monitoring units associated with the state tags outside the boundary are located;

[0038] When it is determined to be a systemic event sign of reverse electrical propagation, the instruction includes an emergency disconnect command for all PV strings in the DC bus area to which the event initiating unit belongs and an alarm trigger signal for the central management node;

[0039] The output limiting operation, the global power reduction factor and the threshold value of the emergency disconnect command are all dynamically calculated through the propagation characteristic model parameter set.

[0040] As a further solution of the present invention: in S5, after the abnormal fluctuation event is handled, the set of local monitoring units actually affected is recorded; and the overlap between the set of local monitoring units and the set of associated monitoring units determined according to the topological relationship diagram is calculated, wherein the overlap is calculated using the Jaccard similarity coefficient formula;

[0041] According to the degree of deviation of the overlap from the expected threshold, the weight coefficients of the relevant edges in the topological relationship graph are adaptively adjusted, and the adjustment amplitude of the weight coefficients is positively correlated with the overlap deviation value. The expected threshold is determined by the statistical mean of historical events.

[0042] Beneficial effects of the present invention:

[0043] The present invention achieves innovative breakthroughs based on the ripple-type collaborative diagnosis triggered by local abnormal events and the dynamic response path of power station topology adaptation. The present invention changes the periodic passive collection mode of centralized polling through the local abnormal triggering mechanism. When a single local monitoring unit detects an abnormal fluctuation, it immediately triggers the collaborative diagnosis of the surrounding associated monitoring units, significantly shortening the abnormal response time and achieving a leap from minute-level polling to second-level triggering; relying on the dynamic topological relationship diagram that integrates physical distance and electrical level, it breaks the static barriers of fixed partitions, upgrades abnormal diagnosis from regional isolated analysis to cross-regional spatial and electrical correlation analysis, effectively identifies the diffusion trend of abnormalities in the spatial dimension and the transmission path of the electrical level, and improves the accuracy of fault location; adopts lightweight pre-diagnosis report Mechanism, each associated unit only returns key information such as status label and abnormal confidence, reducing data transmission redundancy. At the same time, through hierarchical disposal strategy, power limiting operation is performed for local events, gradient adjustment is implemented for regional events, and global warning is issued for system risks, avoiding the extensiveness of traditional one-size-fits-all disposal, and realizing accurate classification of abnormal properties and dynamic matching of disposal strategies. The present invention constructs a complete closed loop of triggering, coordination, diagnosis, disposal and optimization, comprehensively improving the real-time perception, dissemination and judgment and accurate disposal capabilities of distributed photovoltaic power stations for abnormal events, and improving the flexibility and reliability of photovoltaic power stations in complex operating environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below with reference to the accompanying drawings.

[0045] Figure 1 It is a schematic flow diagram of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0047] See also Figure 1 As shown, the present invention is a distributed monitoring and control method for a photovoltaic power station, wherein the photovoltaic power station includes a plurality of dispersed photovoltaic strings and corresponding local monitoring units, comprising the following steps:

[0048] Step S1: Abnormal fluctuation detection and event initiation unit marking

[0049] Each local monitoring unit continuously collects real-time operating parameters of the PV string, including current, voltage, and module temperature. If the rate of change of any parameter significantly exceeds the primary threshold over multiple consecutive sampling cycles and remains above the secondary threshold in the same direction over subsequent cycles, it is identified as an abnormal fluctuation, triggering the unit to become an event initiator, thus avoiding misjudgments caused by environmental interference.

[0050] Step S2: Positioning of associated units based on topological relationship diagram

[0051] The event initiating unit determines the scope of collaboration based on a pre-set topological relationship diagram. This diagram integrates the physical installation coordinates of each unit with the electrical hierarchy of the DC bus path. It uses Euclidean distance to represent spatial proximity and hierarchical depth to reflect electrical connections, generating a topological weight coefficient. Based on the weight or distance threshold, it selects associated monitoring units within a specific surrounding range. For example, units with a physical distance of less than 50 meters and a similar electrical hierarchy are sent a collaborative diagnosis activation command.

[0052] Step S3: Enhanced data collection and pre-diagnosis report generation

[0053] Upon receiving the command, the associated monitoring unit increases the sampling frequency to several times the normal rate, synchronously collecting current and voltage parameters and performing time alignment. It then calls a cluster of historical current-voltage characteristic curves for the same irradiance range, compares the real-time parameter sequence with the historical data, and calculates dynamic time distances. Based on whether the parameters exceed the specified boundaries and the degree of deviation, a lightweight report is generated containing normal, suspicious, and abnormal status labels, along with the confidence level of the anomaly, retaining only critical diagnostic information.

[0054] Step S4: Construction of abnormal propagation feature model

[0055] The event-initiating unit aggregates reports and its own data, establishing a spatial coordinate system centered on itself. This unit converts the anomaly confidence level into a spatial density field value that decays with distance, generating a spatial density distribution cloud map and analyzing the anomaly's diffusion pattern. Furthermore, the gradient slope of the confidence level as it changes with level is calculated, taking into account the depth of each unit in the confluence path, to determine whether the anomaly is propagating in or against the direction of the current.

[0056] Step S5: Generation and execution of hierarchical disposal instructions

[0057] Based on the propagation characteristic model, if a localized, isolated event is identified, power limiting is applied only to the strings associated with the event-initiating unit. If the event is a regional event with forward propagation, power reduction is implemented on the busing branch containing the event-initiating unit and associated abnormal units. If the event is a systemic risk with reverse propagation, emergency disconnection of all strings in the region is instructed, and an alert is sent to the central node. After the event is resolved, the system adaptively adjusts the topology weight coefficient based on the overlap between the actual affected units and the topology-screened units, optimizing subsequent coordination efficiency.

[0058] In a preferred embodiment of the present invention, in S1, the abnormal fluctuation of the preset trigger condition is specifically:

[0059] Key PV string operating parameters, such as output current and voltage, are sampled at high frequency. The sampling period is not fixed but adaptively determined based on the PV string's electrical time constant. The electrical time constant is a key parameter reflecting the string's dynamic response speed. For example, strings with long cables or large capacitors have a larger electrical time constant, so the corresponding sampling period is automatically extended to match the device's actual dynamic characteristics, avoiding missed detections or misjudgments due to inappropriate sampling frequency.

[0060] A hierarchical threshold mechanism is used to identify parameter change rates. For example, if the absolute value of a parameter's rate of change suddenly exceeds a first threshold over five consecutive sampling cycles—for example, if the current rate of change exceeds 20% / s of the rated value—and then remains above a second threshold, such as 10% / s, for the next five cycles, while maintaining the same direction of change (either continuously increasing or decreasing), a valid abnormal fluctuation is identified. The first threshold is significantly higher than the second, typically about twice the former. This creates a dual-judgment logic: the first threshold identifies sudden abnormal changes, while the second threshold eliminates transient interference. For example, if a PV string's current continuously decreases due to a hidden crack in a module, its rate of change will first quickly exceed the first threshold, then maintain at the second threshold as the fault develops, triggering an event initiation unit flag. Parameter fluctuations caused by natural factors, such as irradiance changes caused by temporary cloud cover, may experience a brief increase in rate of change but are unlikely to remain above the second threshold, effectively filtering them out.

[0061] In another preferred embodiment of the present invention, in S2, the photovoltaic power station topology diagram is specifically:

[0062] During the power plant deployment phase, the precise physical installation coordinates of each local monitoring unit (such as latitude and longitude or X / Y coordinates in a plane coordinate system) are first recorded using a geographic information system. Electrical design drawings are then used to analyze the DC busbar connection paths for each unit, clarifying the hierarchical relationships between strings, combiner boxes, and inverters. Strings under the same combiner box belong to the same electrical hierarchy, with the combiner box closer to the inverter at a higher level. For example, a string at level 1 would have a combiner box at level 2 and an inverter at level 3.

[0063] Based on the physical installation coordinates, the system calculates the Euclidean distance between any two local monitoring units, which directly reflects the spatial proximity of the devices. Adjacent units may be affected by the same environmental factors, such as localized shadows and wind speed fluctuations, or there may be physical fault transmission potential, such as bracket vibration causing poor contact between adjacent strings. Furthermore, the system analyzes the hierarchical depth of each unit in the DC bus network based on the electrical connection path. The smaller the difference in hierarchical depth, the closer the units are electrically connected, and the greater the direct impact of fault propagation. For example, a string fault can directly affect the combiner box input parameters, thereby affecting inverter operation.

[0064] In order to generate a topological weight coefficient that can comprehensively characterize the strength of the association between units, the system quantitatively integrates the Euclidean distance and the hierarchical depth. The reciprocal of the Euclidean distance is used as the spatial correlation factor. The closer the distance, the higher the spatial correlation factor; the absolute value of the hierarchical depth difference is used as the electrical correlation factor. The closer the hierarchy, the higher the electrical correlation factor. Through weighted calculation, the topological weight coefficient is finally formed. For example, two units with a distance of 50 meters and a hierarchical difference of 1 have a higher weight coefficient than two units with a distance of 100 meters and a hierarchical difference of 2. The topological relationship diagram uses local monitoring units as nodes and stores topological weight coefficients as edge attributes to form graph structure data. Each node can quickly query its surrounding high-weight associated nodes, providing an accurate data basis for determining the scope of collaborative diagnosis in abnormal events.

[0065] This topology diagram supports a dynamic update mechanism. When the plant's physical configuration changes, such as adding new strings, replacing combiner boxes, or adjusting electrical connections, the central management node recollects the coordinate information of the relevant units, parses the new electrical connection paths, recalculates the hierarchical depth and Euclidean distance, generates updated weight coefficients, and distributes them to all local monitoring units, ensuring that the topology model remains consistent with the actual plant structure. This dual-attribute modeling approach overcomes the static limitations of traditional fixed partitioning, enabling the system to dynamically determine the coordinated scope of anomaly diagnosis based on the actual spatial layout and electrical connection strength of the equipment. For example, when an anomaly occurs in a string, the system not only triggers coordinated diagnosis of physically adjacent strings but also connects to downstream and upstream electrical devices on the same busbar. This comprehensive analysis captures the spatial spread of anomalies and the transmission paths at the electrical level. This allows for effective identification of anomaly propagation characteristics in complex scenarios, such as string faults propagating along the busbar to upstream inverters or mechanical vibration affecting adjacent strings due to spatial proximity.

[0066] In another preferred embodiment of the present invention, in S3, the enhanced data collection and historical operation mode data are specifically:

[0067] Enhanced data collection is a high-frequency data acquisition mechanism executed during the activation of collaborative diagnosis instructions. Its core goal is to accurately capture the dynamic details of abnormal fluctuations by increasing sampling density and synchronously collecting multiple parameters. Specifically, upon receiving the collaborative diagnosis instruction, the system increases the data collection frequency from the normal frequency to an integer multiple of the normal frequency, for example, from once per minute to 10 times per second. This allows it to capture the high-frequency changes characteristic of abnormal events, such as transient signals such as current dips or voltage jumps during the initial formation of module hot spots. Simultaneously, the system collects the output current and output voltage parameters of the PV strings. This is because current and voltage are parameters with strong physical correlations, and their synchronous changes can reveal the true operating status of the equipment. For example, when a string experiences an internal series resistance increase fault, the current drops while the voltage exhibits nonlinear changes. Synchronous analysis can effectively distinguish between the fault type and environmental interference. During the acquisition process, the system time-aligns the current and voltage parameter sequences. A unified timestamp calibration mechanism ensures that the multi-parameter data is fully synchronized in the temporal dimension, preventing analysis errors caused by sampling timing deviations.

[0068] Local historical operating pattern data is constructed based on the normal operating characteristics of PV strings under the same environmental conditions. Considering that solar irradiance is a key environmental factor affecting PV string output characteristics, the system categorizes and stores historical operating data by irradiance range, such as [0-200W / m²], [200-400W / m²], and [400-600W / m²]. Each range corresponds to a cluster of current-voltage characteristic curves. Each curve cluster includes a standard operating curve and tolerance limits. The standard operating curve is generated by statistically averaging historical normal operating data (i.e., data from fault-free, unshaded periods) and represents the ideal operating state of the string within that irradiance range. The tolerance limits are determined based on the standard deviation or percentile of the historical data. For example, a 5% current fluctuation range or a 3% voltage fluctuation range is extended above and below the standard curve to form a boundary range that encompasses normal operating fluctuations. This categorized modeling approach enables the system to quickly retrieve matching historical pattern data under the current ambient irradiance for real-time comparison, improving diagnostic accuracy and timeliness.

[0069] In a preferred embodiment of the present invention, in S3, the process of generating a lightweight pre-diagnosis report including a device status label and anomaly confidence level is as follows:

[0070] First, the current-voltage parameter sequence obtained by enhanced data acquisition is mapped to the corresponding historical current-voltage characteristic curve cluster space, establishing a spatial mapping relationship between real-time data and historical patterns. Subsequently, the distance between the real-time parameter sequence and the standard operating curve is calculated using the dynamic time warping (DTW) algorithm. This algorithm can effectively match time series with different speeds or phases by flexibly adjusting the time axis alignment, making it suitable for analyzing the shape differences of current-voltage curves. If the calculated dynamic time warping distance is less than the threshold of the allowable deviation boundary, the current parameter sequence is determined to be within the boundary and assigned an "in-boundary" status label, indicating that the string is operating normally. If the distance exceeds the boundary threshold but does not completely deviate from the curve cluster range, it is determined to be a "boundary fluctuation" status label, indicating that the operating status is suspicious and may be an early sign of a fault. If the distance significantly exceeds the boundary and continuously deviates from the standard curve, it is determined to be an "out-of-boundary" status label, indicating that a clear abnormality has occurred in the string.

[0071] The calculation of the anomaly confidence value is based on the frequency and duration ratio of the parameter sequence crossing the allowable deviation boundary. Specifically, the system counts the number of times the parameter sequence crosses the boundary and the proportion of time outside the boundary during the enhanced data acquisition cycle, and generates a quantitative value between 0 and 1 through weighted calculation. For example, if the parameter sequence crosses the boundary three times within 10 seconds, and the cumulative proportion of time outside the boundary is 40%, the anomaly confidence can be determined to be 0.6, indicating that there is a medium-confidence anomaly possibility. The lightweight pre-diagnosis report only contains key information such as the device status label, anomaly confidence value, local monitoring unit identifier and time stamp, and eliminates redundant data to reduce the communication transmission load. For example, the data volume of a single report can be controlled within hundreds of bytes to ensure fast transmission and processing in the distributed network.

[0072] In another preferred embodiment of the present invention, the process of constructing the propagation characteristic model of abnormal fluctuations within the surrounding set range in S4 is:

[0073] First, the event initiation unit extracts key information from the lightweight pre-diagnosis reports of all associated monitoring units, including device status tags and anomaly confidence values. This data reflects each unit's response to the abnormal event. Then, based on a preset PV plant topology diagram, the system establishes a two-dimensional or three-dimensional spatial coordinate system with the physical installation coordinates of the event initiation unit as the origin. The position of each associated monitoring unit is mapped into this coordinate system, forming a local spatial network centered on the event initiation point.

[0074] In the spatial coordinate system, the anomaly confidence value is converted into a spatial density field value. This conversion process follows the principle of distance attenuation: the closer the monitoring unit is to the event-initiating unit, the higher the contribution of its anomaly confidence to the spatial density; conversely, the contribution decreases exponentially with increasing distance. Specifically, the spatial density field value is calculated by combining the confidence value with the distance attenuation function. For example, for every 10-meter increase in distance, the density contribution decreases by 50%, simulating the physical law that the impact of anomalies decreases with increasing spatial distance. By superimposing the spatial density field values ​​of all associated monitoring units, the system generates a spatial density distribution cloud map of anomaly fluctuations. This cloud map intuitively presents the aggregation and diffusion characteristics of anomalies in physical space in the form of color depth or contour lines. If the cloud map shows a single peak with the peak located near the coordinates of the event-initiating unit, it indicates that the anomaly is highly localized; if there are multiple peaks or a continuous diffusion pattern, it indicates that the anomaly may spread to surrounding areas.

[0075] At the same time, the system analyzes the hierarchical depth of each associated monitoring unit in the DC convergence path. Hierarchical depth is defined as the number of nodes a unit is from the string end in the string-combiner box-inverter electrical link. For example, if the string level is 1, the corresponding combiner box level is 2, and the connected inverter level is 3. By calculating the gradient slope of the anomaly confidence value as it changes with hierarchical depth, the propagation direction of the anomaly within the electrical hierarchy can be determined. A negative slope means that the confidence decreases with increasing hierarchy, indicating that the anomaly may be propagating in a forward direction from lower-level strings to higher-level devices. A positive slope indicates that the confidence increases with increasing hierarchy, suggesting that the anomaly is propagating in the reverse direction, from the inverter to the string. The absolute value of the gradient slope reflects the propagation strength; a larger absolute value indicates a more significant transmission effect of the anomaly between electrical hierarchies.

[0076] Ultimately, the propagation characteristic model combines the morphological characteristics of the spatial density distribution cloud map, such as single peak, multi-peak, and diffusion, with the positive and negative signs and absolute values ​​of the gradient slope, and outputs a parameter set that includes the spatial diffusion pattern, electrical propagation direction, and the degree of abnormal concentration. For example, if the cloud map shows single peak concentration and a gradient slope of zero, it means that the anomaly is limited to the vicinity of the event initiating unit and has not propagated to the electrical upstream and downstream; if the cloud map shows multi-peak diffusion and a negative gradient slope, it indicates that the anomaly is both spatially diffused to the periphery and transmitted to high-level equipment along the current direction. These parameters provide a key basis for the subsequent determination of abnormal event types and the generation of disposal strategies, enabling the system to distinguish between local independent events, regional propagation events, or signs of systemic risks, avoiding the diagnostic limitations caused by traditional methods that rely only on single-dimensional data.

[0077] In another preferred embodiment of the present invention, in S5, the process of the propagation characteristic model determining the type of abnormal fluctuation event is as follows:

[0078] When the spatial density distribution cloud graph presents a single peak, and the peak is strictly located at the coordinate position of the event initiating unit, and the gradient slope of the abnormality confidence changing with the depth of the DC bus path layer is zero, it indicates that the abnormality is limited to the photovoltaic string associated with the event initiating unit and has not spread to the surrounding space or electrical upstream and downstream. This situation is usually caused by a single device failure, such as hidden cracks in the internal components of the string, local poor contact, etc., and its impact range is strictly limited to the physical location and the lowest level of electrical connection. For example, a string has abnormal current due to loose terminal blocks, and the monitoring data of the surrounding adjacent strings has not changed significantly, and the parameters of high-level equipment such as the junction box and inverter remain normal. At this time, the gradient slope is zero, and the spatial density is concentrated at the location of the faulty string.

[0079] If the spatial density distribution cloud shows a multi-peaked or continuously diffuse pattern, it indicates that the anomaly has spread beyond the confines of a single device and to other physically adjacent strings. Furthermore, a negative gradient indicates that the anomaly confidence decreases with increasing DC bus path levels, that is, along the current flow direction, from the string to the combiner box and inverter. This propagation pattern typically corresponds to regional faults, such as a failure in a combiner box component that simultaneously affects multiple strings connected to it, or localized shadowing that causes power imbalances in adjacent strings and triggers a chain reaction through electrical connections. For example, if a combiner box has poor contact at the input, current anomalies are detected in all three strings connected to it, and the anomaly confidence gradually decreases from the string level to the combiner box level, the spatial density cloud shows a multi-peaked distribution centered on the combiner box. In this case, it is determined to be a regional event with positive electrical propagation.

[0080] When the spatial density distribution cloud map shows an irregular distribution pattern and the gradient slope is positive, the anomaly confidence increases with the level, indicating that the anomaly may originate from high-level equipment, such as the inverter and the grid side, and propagate in the opposite direction of the current to the lower-level strings. This pattern has systemic risk characteristics and may be caused by inverter failure, grid voltage fluctuations, or energy storage system anomalies. Its impact is not limited to a specific area, but spreads to the upstream strings through electrical connections. For example, a short circuit failure on the DC side of the inverter may cause the voltage of all connected combiner boxes to drop sharply, thereby affecting the normal operation of hundreds of strings. In this case, the spatial density cloud map shows that the anomaly distribution has no obvious aggregation center and the gradient slope is positive, indicating that the anomaly propagates from the inverter to the combiner box and the strings in the opposite direction.

[0081] In a preferred embodiment of the present invention, the process of generating the preliminary handling instruction in S5 is as follows:

[0082] For highly localized anomalies, the initial response directive applies only to the PV strings associated with the event-initiating unit, executing output limiting operations. For example, if a string's current remains consistently below the specified value due to a hot spot effect, the system limits its maximum power output, for example, to 70% of the rated value, to prevent further failure while maintaining normal operation of other strings. The limiting threshold is dynamically calculated using the propagation signature model parameter set, combined with anomaly confidence and historical operating data, to ensure that limiting suppresses the fault without excessively impacting overall power generation.

[0083] For regional anomalies with forward electrical propagation, the system implements global power reduction for the busbar containing the event-initiating unit and all associated monitoring units with an out-of-bounds status. The power reduction factor is determined based on the spatial distribution density of the anomaly confidence level and the electrical level propagation intensity. For example, if 30% of the strings in a busbar are marked as out-of-bounds and the spatial density cloud map shows a spreading trend, the system may reduce the overall power of that busbar by 20% to reduce the risk of fault propagation. Simultaneously, an early warning is sent to the next higher-level electrical node, prompting it to monitor changes in input parameters and prepare for further control.

[0084] If a systemic risk of reverse electrical propagation is identified, the response is escalated to an emergency disconnection and global alarm. First, all PV strings in the DC bus area of ​​the unit initiating the incident are disconnected from the grid to prevent the anomaly from spreading through the inverter to the grid. Second, an emergency event report containing complete summary information and preliminary analysis conclusions is sent to the central management node, triggering a site-wide fault response process, such as activating backup power supplies and dispatching operations and maintenance personnel to the site for investigation. The triggering conditions and scope of the emergency disconnection command are determined by the absolute value of the gradient slope and the degree of spatial irregularity in the propagation characteristic model, ensuring rapid isolation of potential fault sources before the risk escalates.

[0085] This handling mechanism avoids the drawbacks of the traditional one-size-fits-all approach by deeply integrating anomaly propagation characteristics with handling strategies. This prevents both overhandling due to localized failures and underhandling due to systemic risks. Furthermore, a dynamically calculated threshold mechanism enables handling instructions to adapt to the severity of different failure scenarios. For example, for the same localized event, a string with an anomaly confidence level of 0.9 will have a larger limiting amplitude than a string with a confidence level of 0.6, demonstrating the technical principle of precise control.

[0086] In another preferred embodiment of the present invention, in S5, after the abnormal event is handled, the set of local monitoring units actually affected is recorded. The set is dynamically determined based on whether each unit performs power limiting, decoupling and other operations during the handling process, and truly reflects the scope of equipment involved in the abnormal propagation.

[0087] Subsequently, the overlap between the actual set of affected units and the set of associated monitoring units preset in the topological relationship diagram is calculated, and the degree of match between the two is quantified using the Jaccard similarity coefficient. The overlap is compared with a pre-set expected threshold. If the overlap is significantly lower than expected, it indicates that the weight coefficients of the relevant units in the topological relationship diagram do not accurately reflect the actual strength of the association and need to be adjusted. The adjustment logic is to dynamically adjust the weight coefficients of the relevant edges based on the overlap deviation value. The greater the deviation, the greater the adjustment. For example, the weight coefficients of units that are actually affected but not included in the associated set will be increased with the event-initiating unit to enhance their relevance during subsequent screening.

[0088] The adjusted topology weight coefficient is directly applied to subsequent abnormal events, making the selection of associated monitoring units more in line with the actual propagation range. For example, if the actual affected units in a regional fault exceed the initial associated set, the system will increase the electrical hierarchy weight or spatial proximity weight between the corresponding regional units to avoid screening omissions in similar events. This mechanism dynamically adapts to changes in power plant equipment characteristics, layout adjustments, or new fault modes through a closed loop of disposal, feedback, and optimization, solving the lag problem of traditional static topology models. It upgrades the selection of associated units from a fixed logic that relies on initial deployment to dynamic learning combined with historical experience, continuously improving the system's prediction accuracy for abnormal propagation and the efficiency of collaborative diagnosis.

[0089] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A distributed monitoring and control method for a photovoltaic power station, wherein the photovoltaic power station comprises a plurality of dispersed photovoltaic strings and corresponding local monitoring units, characterized in that: The following steps are involved: S1. Each local monitoring unit continuously monitors the real-time operating parameters of the corresponding PV string. When any local monitoring unit detects an abnormal fluctuation in the real-time operating parameters that meets a preset trigger condition, the local monitoring unit is marked as an event initiating unit. S2. The event initiating unit selects other local monitoring units within a set range around the location as associated monitoring units according to a preset photovoltaic power station topology diagram, and sends a collaborative diagnosis activation instruction to all associated monitoring units; S3. Each associated monitoring unit that receives the collaborative diagnosis activation instruction performs enhanced data collection and generates a lightweight pre-diagnosis report containing a device status label and anomaly confidence level based on the data collection results and the stored local historical operation mode data; S4. Each associated monitoring unit sends the lightweight pre-diagnosis report to the event initiation unit. The event initiation unit constructs a propagation characteristic model of the abnormal fluctuation within a set range of the surrounding area based on the summarized lightweight pre-diagnosis report and its own abnormal fluctuation parameters; S5. Generate a preliminary handling instruction based on the abnormal fluctuation event type output by the propagation feature model, and the event initiating unit sends the preliminary handling instruction to the affected local monitoring unit for execution; In S4, the process of constructing the propagation characteristic model of abnormal fluctuations within the surrounding set range is as follows: Extracting the device status labels and abnormal confidence values ​​from all lightweight pre-diagnosis reports; establishing a spatial coordinate system centered on the event initiating unit based on the preset photovoltaic power station topology diagram; In the spatial coordinate system, converting the abnormal confidence value of each associated monitoring unit into a spatial density field value; The spatial density field value decays exponentially as the Euclidean distance between the associated monitoring unit and the event initiating unit increases; The spatial density field values ​​of all associated monitoring units are superimposed to generate a spatial density distribution cloud map of abnormal fluctuations. At the same time, the hierarchical depth of each associated monitoring unit in the DC convergence path is determined, and the gradient slope of the abnormal confidence value changing with the hierarchical depth is calculated; Combining the morphological features of the spatial density distribution cloud map with the sign and absolute value of the gradient slope, a propagation characteristic model parameter set is output.

2. A distributed monitoring and control method for a photovoltaic power station according to claim 1, characterized in that: In S1, the abnormal fluctuation of the preset trigger condition is specifically: The absolute value of the rate of change of the real-time operating parameter in n consecutive set sampling periods exceeds the first set threshold, and the absolute value of the rate of change continues to remain above the second set threshold in the subsequent n set sampling periods; Wherein, n is a natural number, a fixed proportional relationship exists between the first set threshold and the second set threshold, and the first set threshold is greater than the second set threshold; the direction of the change rate remains consistent within the n consecutive set sampling periods, and the duration of the set sampling period is adaptively determined by the electrical time constant of the photovoltaic string.

3. A distributed monitoring and control method for a photovoltaic power station according to claim 1, characterized in that: In S2, the photovoltaic power station topology diagram is specifically: During the PV power station deployment phase, record the physical installation coordinates of each local monitoring unit and the electrical connection path of the DC bus branch to which it belongs; The Euclidean distance between any two local monitoring units is calculated based on the physical installation coordinates; the hierarchical depth of each local monitoring unit in the DC bus network is analyzed based on the electrical connection path; the Euclidean distance and the hierarchical depth are combined to generate a topological weight coefficient; the topological weight coefficient represents the spatial and electrical correlation strength between any two local monitoring units; the topological relationship diagram is stored as a graph structure with local monitoring units as nodes and topological weight coefficients as edge attributes.

4. A distributed monitoring and control method for a photovoltaic power station according to claim 1, characterized in that: In S3, the enhanced data collection and historical operation mode data are specifically: The enhanced data acquisition method is to increase the sampling frequency to an integer multiple of the conventional monitoring frequency during the validity period of the collaborative diagnosis activation instruction; synchronously collect the output current and output voltage parameters of the photovoltaic strings, and perform time alignment processing on the collected current and voltage parameter sequences; The local historical operation mode data is a cluster of current-voltage characteristic curves of the photovoltaic string under the same solar irradiance range as the current environment, and the cluster of current-voltage characteristic curves includes a standard operating curve and an allowable deviation boundary.

5. A distributed monitoring and control method for a photovoltaic power station according to claim 4, characterized in that: In S3, the process of generating a lightweight pre-diagnosis report including the device status label and the abnormality confidence level is as follows: Mapping the current-voltage parameter sequence obtained by enhanced data acquisition to the current-voltage characteristic curve cluster space in the local historical operation mode data; Calculating the dynamic time warping distance between the current-voltage parameter sequence and the standard working curve; assigning a device status label based on whether the dynamic time warping distance exceeds the allowable deviation boundary of the characteristic curve cluster; the device status label includes an in-boundary state label, a boundary fluctuation state label, and an out-boundary state label; Based on the ratio of the number of times and duration that the current parameter sequence crosses the allowable deviation boundary, an abnormality confidence value is calculated; the lightweight pre-diagnosis report includes a device status label, an abnormality confidence value and a local monitoring unit identifier.

6. A distributed monitoring and control method for a photovoltaic power station according to claim 1, characterized in that: In S5, the process of the propagation characteristic model determining the type of abnormal fluctuation event is as follows: If the spatial density distribution cloud graph presents a single peak with the peak located at the coordinates of the event initiating unit, and the gradient slope is zero, it is determined to be a highly localized independent device event; If the spatial density distribution cloud diagram presents a multi-peak morphology or a continuous diffusion morphology, and the gradient slope is a negative value, it is determined to be a regional event with positive electrical propagation characteristics; If the spatial density distribution cloud diagram presents an irregular distribution pattern and the gradient slope is positive, it is determined to be a sign of a systemic event with the characteristics of reverse electrical propagation; the positive electrical propagation refers to the diffusion of abnormal fluctuations along the direction of normal current, and the reverse electrical propagation refers to the diffusion of abnormal fluctuations in the opposite direction of current.

7. A distributed monitoring and control method for a photovoltaic power station according to claim 6, characterized in that: In S5, the process of generating the preliminary handling instruction is as follows: When it is determined to be a highly localized independent device event, the initial handling instructions only include the output limit operation of the PV strings associated with the event initiating unit; When it is determined to be a regional event with forward electrical propagation, the instruction includes the global power reduction coefficient of the bus branch where the event initiating unit and all monitoring units associated with the state tags outside the boundary are located; When it is determined to be a systemic event sign of reverse electrical propagation, the instruction includes an emergency disconnect command for all PV strings in the DC bus area to which the event initiating unit belongs and an alarm trigger signal for the central management node; The output limiting operation, the global power reduction factor and the threshold value of the emergency disconnect command are all dynamically calculated through the propagation characteristic model parameter set.

8. The distributed monitoring and control method for a photovoltaic power station according to claim 1, characterized in that: In S5, after the abnormal fluctuation event is handled, the set of local monitoring units actually affected is recorded; and the degree of overlap between the set of local monitoring units and the set of associated monitoring units determined according to the topological relationship diagram is calculated, wherein the degree of overlap is calculated using the Jaccard similarity coefficient formula. According to the degree of deviation of the overlap from the expected threshold, the weight coefficients of the relevant edges in the topological relationship graph are adaptively adjusted, and the adjustment amplitude of the weight coefficients is positively correlated with the overlap deviation value. The expected threshold is determined by the statistical mean of historical events.

Images