Photovoltaic power station group string level monitoring and rapid response method based on 5G communication

Abstract

The application discloses a photovoltaic power station group string level monitoring and rapid response method based on 5G communication, high-frequency acquisition of group string current, voltage and backboard temperature data is carried out, multidimensional evaluation such as transverse power deviation and longitudinal power attenuation is carried out, a state evaluation vector is generated, state classification judgment is carried out according to the vector on the edge side, a trigger flag is locally generated for the highest level alarm, decision message is differentially transmitted through 5G uRLLC and eMBB slice channels, when the trigger flag and a cloud transmission completion flag are simultaneously effective, a switch is driven to rapidly isolate a fault group string, and global analysis and closed-loop management are carried out in the cloud. The application realizes fine perception and millisecond-level rapid isolation response of group string level faults.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power plant monitoring technology, specifically a method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication. Background Technology

[0002] Photovoltaic power plant monitoring systems aim to achieve fault early warning and response through data analysis. Existing technologies mostly focus on inverter-level or power plant-level monitoring, and are insufficient in detecting subtle anomalies at the string level.

[0003] Patent CN209729025U discloses a photovoltaic power plant monitoring system based on Wi-Fi and 4G network communication. While it can remotely report data, its core functionality lies in simplifying on-site parameter configuration through Wi-Fi and utilizing 4G as a remote communication backhaul channel. This solution does not address real-time processing and multi-dimensional anomaly assessment of high-frequency data at the photovoltaic string level. Its communication architecture is essentially a single-channel data pass-through, failing to achieve quality of service isolation between emergency alarm data and regular monitoring data. Furthermore, it lacks the ability to quickly identify severe faults at the edge and trigger millisecond-level physical isolation responses. Therefore, when facing sudden string-level faults, unavoidable communication and processing delays exist from the occurrence of the anomaly to the execution of isolation. To address these issues, it is necessary to propose a 5G-based method for monitoring and rapidly responding to photovoltaic power plant string-level faults. Summary of the Invention

[0004] This invention provides a method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication, aiming to solve the problems of slow string-level fault perception, coarse location, and large response delay in existing technologies.

[0005] The objective of this invention can be achieved through the following technical solutions: A method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication includes the following steps: Step S1: The string acquisition terminal synchronously acquires the current value, voltage value and backsheet temperature value of each photovoltaic string at a preset acquisition period, and stores the acquired data in the local buffer to form the original data frame of each string.

[0006] Step S2: Read the original data frame and perform the following operations to construct the state evaluation vector for each string group: Preferably, the real-time power value of each string is obtained by multiplying the current value and the voltage value. Preferably, a preset formula is called to determine the power deviation rate of each string based on the real-time power value and the average real-time power of all strings under the same inverter.

[0007] Preferably, a preset algorithm is invoked to determine the power attenuation rate of each string based on the real-time power value and the historical reference power value of the string in the corresponding irradiance range.

[0008] Preferably, the temperature exceedance of each string is calculated based on the difference between the backplate temperature value and the preset temperature threshold.

[0009] Preferably, the power deviation rate, the power attenuation rate, and the temperature exceedance are written into the state evaluation vector of each string.

[0010] Step S3: Read the state evaluation vector, and according to the preset hierarchical decision rules, determine the current state of each string as one of multiple levels, including normal, warning and alarm, and generate a state decision message containing the decision result; wherein, when the decision result is the preset highest level alarm, a trigger flag is generated locally.

[0011] Step S4: Read the status decision message and send it to the cloud monitoring platform through a pre-established differentiated 5G transmission channel according to its priority marker; wherein, high-priority status decision messages are transmitted through the 5G uRLLC slice channel, and ordinary-priority status data are transmitted through the 5G eMBB slice channel; when the cloud monitoring platform receives the high-priority message, it returns an acknowledgment frame and sets the transmission completion flag locally.

[0012] Step S5: When the trigger flag and the transmission completion flag are both valid, an isolation command is sent to the switch of the branch where the faulty string is located to drive it to disconnect and physically isolate the faulty string; at the same time, an event record data frame containing information about this isolation event is generated.

[0013] Step S6: The cloud monitoring platform summarizes and analyzes the received status judgment messages and event record data frames, identifies clusters with a persistent low incidence trend, and generates an operation and maintenance task list; for clusters that have been isolated, it issues an intervention reminder to the operation and maintenance personnel, and receives a manually issued reset command after the maintenance is completed.

[0014] Preferably, before step S2, the method further includes the step of establishing and updating the historical reference power value: the edge computing node maintains a power sample queue for each string according to the irradiance intensity interval, performs validity screening before adding newly collected power samples to the queue, and after the number of valid samples in the queue reaches a preset threshold, updates the historical reference power value of the corresponding irradiance intensity interval according to the statistical processing calculation of the samples in the queue.

[0015] Preferably, the state evaluation vector constructed in step S2 includes weight coefficients corresponding to the power deviation rate, the power attenuation rate, and the temperature exceedance, and the weight coefficients are determined based on the reliability of the irradiance value, ambient temperature value, and historical reference power value in the current acquisition cycle.

[0016] Preferably, the step of identifying the persistent low-frequency tendency in step S6 includes: the cloud monitoring platform calculates the proportion of low-frequency strings across the entire site; if the proportion exceeds a first global threshold, the persistent low-frequency judgment of all strings is paused; if the proportion is lower than a second global threshold, the persistent low-frequency judgment is performed normally; if the proportion is between the first global threshold and the second global threshold, the inverter is divided into sub-regions for partition verification, and the persistent low-frequency judgment in the sub-regions that meet the partition conditions is paused only.

[0017] Preferably, the step of generating the operation and maintenance task list in step S6 includes: the cloud monitoring platform matching the historical sequence of the state evaluation vector of the target string with a preset fault feature rule library, inferring the fault type of the target string based on the matching result, and writing the inferred fault type into the operation and maintenance task list.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention achieves physical isolation transmission of alarm data and regular data by combining string-level high-frequency acquisition and multi-dimensional real-time evaluation at the edge, along with 5G network slicing. It also innovatively uses a dual-setting of "local trigger flag" and "cloud transmission completion flag" as isolation action conditions. Under the premise of ensuring system information consistency, it compresses the entire link latency of perception-decision-physical isolation of serious faults to the millisecond level, effectively preventing the fault from escalating and significantly reducing power generation loss and operation and maintenance costs. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings: Figure 1 This is a flowchart of a method for string-level monitoring and rapid response of a photovoltaic power station based on 5G communication, as proposed in an embodiment of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] First embodiment: This invention provides a string-level monitoring and rapid response method for photovoltaic power plants based on 5G communication. It leverages the low latency and high bandwidth of the 5G communication network to achieve high-frequency real-time data acquisition, precise fault location, and millisecond-level rapid isolation response for each string in the photovoltaic power plant, thereby significantly reducing downtime and power generation loss. Detailed explanations follow.

[0022] Please see Figure 1 , Figure 1 This is a flowchart illustrating a method for string-level monitoring and rapid response in photovoltaic power plants based on 5G communication, as disclosed in an embodiment of the present invention. Figure 1 As shown, the method may include the following steps: Step S1: Under the power supply of the storage battery, the string acquisition terminal continuously collects the current value, voltage value and backplane temperature value of each string at a preset acquisition cycle, and stores the above three types of data in the local buffer to form the original data frame of the current cycle corresponding to each string.

[0023] In this embodiment of the invention, the string acquisition terminal can be deployed at the combiner box or installed near the combiner node of each string. Each acquisition terminal is responsible for acquiring data from one or more strings. The acquisition terminal has a built-in current sensor, voltage sampling circuit, and thermal element. The three types of sensors trigger acquisition at the same time to ensure that the parameters in the same data frame are strictly aligned in time, avoiding abnormal misjudgments caused by misalignment of acquisition time.

[0024] In this embodiment of the invention, the preset acquisition period can be flexibly configured according to the power plant scale and communication resources. In one embodiment of the invention, the acquisition period is set to 100 milliseconds, that is, 10 data frames are generated per second, to meet the need for rapid perception of abnormal trends in subsequent steps. After completing one cycle of acquisition, the acquisition terminal packages and compresses the current value, voltage value, and backplane temperature value according to a predefined data frame format, writes the group string identifier and acquisition timestamp into the frame header, and temporarily stores the data frame in the local buffer, waiting for further processing in step S2.

[0025] Step S2: Read the original data frame of the current cycle output in step S1 from the local buffer, calculate the real-time power value of each string, and compare the real-time power value of each string with the average power value of all parallel strings under the same inverter in the current cycle to obtain the power deviation rate of each string; at the same time, compare the real-time power value of each string with the historical reference power value of the string itself in the vertical direction to obtain the power attenuation rate of each string; in addition, calculate the difference between the backplane temperature data and the temperature threshold to obtain the temperature exceedance of each string; write the power deviation rate, power attenuation rate and temperature exceedance together into the state evaluation vector of the string in the current cycle as the input of step S3.

[0026] In this embodiment of the invention, the real-time power value is obtained by directly multiplying the current value and voltage value collected in step S1. The calculation is completed locally on the acquisition terminal, without the need to upload to the cloud, thus saving communication bandwidth.

[0027] The power mean used for horizontal comparison is the real-time arithmetic mean of the power of all strings connected to the same inverter within the current acquisition period. This mean is updated in real time and does not rely on historical data, thus exhibiting good adaptability to current light intensity and temperature conditions—when overall light intensity suddenly decreases, the power of all strings decreases synchronously, and the mean also decreases, preventing false alarms from being triggered by the horizontal deviation rate. The historical reference power value is the average power of the string under the same irradiance range on several recent sunny days, and is updated periodically by the edge computing nodes. In one embodiment of the present invention, the power deviation rate is calculated as follows: The difference between the current string's real-time power and the average power of all strings in the same inverter is divided by the average power to obtain a signed ratio value. A positive value indicates that the string's power is higher than the average, and a negative value indicates that it is lower than the average.

[0028] For example, if a string currently has a power output of 2.1 kW, and the average power output of all strings under the same inverter is 2.5 kW, then the power deviation rate of this string is -16%, indicating that this string is experiencing significant underpowerment. Temperature exceeding the limit is calculated by subtracting a preset temperature threshold from the current backplane temperature measurement. This threshold can be configured according to the component model and ambient temperature. In one embodiment of this invention, the temperature threshold is set to 75 degrees Celsius.

[0029] It should be noted that step S2 introduces both horizontal and vertical comparison dimensions to address two different types of anomaly scenarios. Horizontal comparison can quickly detect anomalies that are significantly lower than neighboring strings at the current moment, making it suitable for identifying sudden faults such as occlusion, short circuits, and poor contact. Vertical comparison, on the other hand, can detect slow degradation relative to the string's historical levels, making it suitable for identifying gradual efficiency losses such as component aging and long-term dust accumulation. The comparison results of both dimensions are uniformly passed to the next step in vector form, enabling subsequent fault classification decisions to simultaneously consider both scenarios and avoid missed or false positives when using a single-dimensional comparison.

[0030] Step S3: Read the status evaluation vectors of each string output in Step S2. Based on the preset hierarchical decision rules, determine the current status of each string as one of four levels: normal, level 1 warning, level 2 warning, or level 3 alarm. Encapsulate the decision result, along with the corresponding string's identifier and the current timestamp, into a status decision message. Strings determined to be at level 1 warning or higher have their status decision messages marked as high priority and must immediately enter the transmission process in Step S4 after this step. Strings determined to be normal have their status data enter the regular reporting process in Step S4 according to normal priority.

[0031] In this embodiment of the invention, the graded judgment rule can be executed according to the following logic: if the absolute value of the power deviation rate is less than 10%, the power attenuation rate is less than 5%, and the temperature exceedance is not greater than zero, it is judged as normal; if the absolute value of the power deviation rate is between 10% and 20%, or the power attenuation rate is between 5% and 15%, or the temperature exceedance is between 0 and 5 degrees Celsius, it is judged as a first-level warning; if the absolute value of the power deviation rate exceeds 20% and is accompanied by a temperature exceedance greater than 0 degrees Celsius, or the power attenuation rate exceeds 15%, it is judged as a second-level warning. If the absolute value of the power deviation rate exceeds 35%, the temperature exceeds the limit by more than 10 degrees Celsius, or the current value suddenly drops below 20% of the rated value, it will be directly judged as a Level 3 alarm. All of the above thresholds are configurable parameters and can be adjusted based on the climate characteristics of the power plant's location, the component brand, and operation and maintenance experience during actual deployment.

[0032] In this embodiment of the invention, under the three-level alarm state, a trigger flag is immediately generated locally. This flag will serve as one of the prerequisites for the rapid isolation action in step S5, and will participate in the execution judgment of step S5 together with the transmission result of step S4.

[0033] It should be noted that the four-level state classification and the design of directly generating trigger flags for level three alarms locally are key arrangements in this invention for improving fault response speed. If all state decisions rely on the cloud for completion before issuing instructions, there will inevitably be a network round-trip delay between the occurrence of an anomaly and the isolation action. However, by decentralizing the identification of level three alarms and the generation of trigger flags to the edge side, the most urgent isolation action no longer depends on cloud confirmation, thus laying the foundation for millisecond-level rapid isolation in step S5.

[0034] Step S4: Read the status decision message generated in step S3. Based on the priority marker of the message, send the message to the cloud monitoring platform through a pre-established differentiated 5G transmission channel. Among them, the status decision message carrying the high priority marker is transmitted through the 5G uRLLC slice channel, which has an end-to-end latency guarantee of no more than 10 milliseconds and is used to carry emergency messages of level 1 warning and above. The status data carrying the ordinary priority marker is transmitted through the 5G eMBB slice channel, which has a large transmission bandwidth and is used to carry the periodic statistical data reporting of each group under normal conditions. After receiving the high priority message, the cloud monitoring platform returns a reception confirmation frame to the sender, records the reception time of the confirmation frame as the completion time of this high priority transmission, and writes the status flag "high priority message transmission completed" into the local transmission completion flag for use in step S5.

[0035] In this embodiment of the invention, the establishment of the uRLLC slice channel and the eMBB slice channel relies on the network slicing technology of the 5G network. They are isolated from each other at the levels of radio resource scheduling and core network forwarding priority, and do not interfere with each other. In actual deployment scenarios, when the photovoltaic power plant operator accesses the 5G network, it negotiates and configures the corresponding slice parameters with the operator. After the slice configuration is completed, when sending a message, it only needs to select the corresponding logical channel according to the priority flag in the message header, without needing to perform additional routing judgments during transmission. For example, when a string is determined to be a level 3 alarm, a status decision message carrying the alarm information is sent through the uRLLC slice channel, completing transmission within 10 milliseconds and receiving a confirmation frame from the cloud. The transmission completion flag is then set. The entire process does not require waiting for the transmission results of ordinary reported data, and the messages of the two slice channels do not block each other.

[0036] It's important to note that this step combines 5G network slicing technology with a message priority classification mechanism. This allows emergency alarm data and regular monitoring data to physically share the same 5G infrastructure, while simultaneously achieving strict service quality isolation at the logical level—a capability that traditional 4G or RS485 communication solutions cannot provide. In 4G networks, a large number of concurrent data reporting requests compete for channel resources, making it difficult to guarantee the actual latency of emergency alarm messages. The 5G network slicing mechanism fundamentally solves this problem, ensuring the real-time transmission of emergency alarms even in scenarios where photovoltaic power plants simultaneously report data from large-scale parallel strings.

[0037] Step S5: Read the trigger flag generated in step S3 and the transmission completion flag generated in step S4. When both are set simultaneously, send an isolation command to the DC circuit breaker or smart combiner switch of the corresponding string branch, driving it to complete the disconnection within the preset action time and physically isolate the faulty string from the combiner branch. At the same time, package the string identifier, trigger reason category, command issuance time and switch disconnection completion time of this isolation action together as an event record data frame and upload it to the cloud monitoring platform for operation and maintenance personnel to view and analyze.

[0038] In this embodiment of the invention, the issued isolation command is transmitted to the drive circuit of the DC circuit breaker or intelligent combiner switch via a wired control bus. The command format includes the address code of the target switch, the action type (opening or closing), and a check code. The entire command frame length does not exceed 32 bytes to ensure the response speed of the control link. In one embodiment of the invention, the total time from the simultaneous setting of the trigger flag and the transmission completion flag to the completion of the switch opening does not exceed 150 milliseconds, of which the time for the command to be issued to the switch drive circuit does not exceed 5 milliseconds, the mechanical action time of the switch body does not exceed 50 milliseconds, and the remaining time is for verification and confirmation. In this embodiment of the invention, requiring the transmission completion flag as one of the necessary prerequisites for the isolation action is to ensure that the cloud has been aware of the abnormal state of the string, thereby avoiding the situation where the isolation action is performed locally in isolation and the cloud monitoring platform cannot detect it—this design ensures both response speed and information consistency of the entire monitoring system.

[0039] It should be noted that this step uses the dual reset bits of the "local trigger flag" and the "cloud transmission completion flag" as the trigger condition for the isolation action, which is a design trade-off that balances response speed and system reliability. On the one hand, the trigger flag is directly generated by the local edge side in step S3, without relying on cloud instructions, ensuring the lower limit of latency for the entire response chain; on the other hand, the introduction of the transmission completion flag avoids the problem of inconsistent states caused by unilateral execution of isolation actions locally in extreme scenarios such as network anomalies. Both flags are maintained in the form of local registers, and the latency of the read operation is negligible and will not have a substantial impact on the overall response speed.

[0040] In step S6, after receiving the status judgment messages of each string uploaded in step S4 and the event record data frames uploaded in step S5, the cloud monitoring platform summarizes and analyzes the status evaluation vector data of all strings in the entire station within the current collection period, identifies the set of strings with a persistent low incidence trend, generates an operation and maintenance task list containing string identifier, abnormal duration, suggested maintenance priority, and inferred fault type, and pushes it to the operation and maintenance personnel through the operation and maintenance management interface of the cloud platform. For strings that have already undergone isolation operations in step S5, the cloud monitoring platform will also issue a manual intervention reminder to the operation and maintenance personnel, prompting them to manually issue a reset command through the platform interface to release the isolation status after confirming the cause of the fault on-site and completing the repair.

[0041] In this embodiment of the invention, the cloud monitoring platform performs rolling analysis of the status data of each string in units of collection cycles. When identifying a persistent low power consumption trend, the absolute value of the power deviation rate exceeding a preset persistence threshold is typically used as the criterion. For example, if the power deviation rate of a string is below -15% for 30 consecutive collection cycles (i.e., 3 consecutive seconds), and the overall irradiance of the entire station decreases during the current period, the cloud platform marks the string as having a persistent low power consumption trend, includes it in the maintenance task list, and infers its fault type based on historical records of similar anomalies, such as shading, component attenuation, or abnormal DC cable loss. The platform presents the inferred fault type as a "suggested investigation" rather than a mandatory conclusion to avoid misleading maintenance personnel into ignoring other possible causes of the fault.

[0042] In this embodiment of the invention, Figure 1 Steps S1 to S6 of the method shown constitute a complete technical link of acquisition, processing, transmission, response, and closed-loop management. Step S1 completes data acquisition, and the obtained raw data frame flows to step S2; Step S2 completes multi-dimensional comparison calculation, and the obtained state evaluation vector flows to step S3; Step S3 completes hierarchical decision-making, and the obtained state decision message and trigger flag bit flow to steps S4 and S5 respectively; Step S4 completes differentiated transmission, and the obtained transmission completion flag bit flows to step S5; Step S5 completes isolation action execution and sends the event record data frame to step S6; Step S6 completes global summary analysis and maintenance task generation in the cloud, and issues a reset command to the site at an appropriate time, completing the closed-loop management of the isolation operation in step S5. The data flow and signal linkage between each step constitute an overall collaborative technical solution, enabling the present invention to achieve rapid perception and millisecond-level response to string-level anomalies in photovoltaic power plants while ensuring acquisition accuracy and system reliability.

[0043] Second embodiment: This embodiment is a further refinement of the "historical reference power value" establishment and update mechanism in the first embodiment. It details how to dynamically build and maintain a historical performance reference database for each string based on irradiance intensity intervals and using a sliding window method with validity screening, aiming to ensure the accuracy and adaptability of longitudinal power attenuation rate calculation.

[0044] The second embodiment mainly describes in detail the specific implementation process of the historical reference power value establishment and dynamic update mechanism, the weighted composition method of the state evaluation vector in step S2, and the global irradiance exclusion logic and fault type inference mechanism in step S6.

[0045] The second embodiment can be implemented in conjunction with the first embodiment, or a portion thereof can be selected and deployed separately based on the first embodiment.

[0046] Before performing the longitudinal comparison in step S2, a historical reference power database needs to be pre-established for each string. This database uses irradiance intensity range as the index dimension and stores the historical steady-state power reference values ​​of the string under each irradiance intensity range. The specific implementation process is as follows: Edge computing nodes continuously receive raw data frames uploaded by each string acquisition terminal and extract the real-time power value and current irradiance value for each acquisition cycle. The irradiance value is collected by an irradiometer deployed at the power plant site and reported to the edge computing node via the same 5G communication channel, aligning with the string power data in time. In one embodiment of the invention, the irradiance intensity is divided into several intervals based on power per square meter, with the boundary values ​​of adjacent intervals increasing in steps of 50 watts per square meter, for example, 0 to 50, 50 to 100, 100 to 150, and so on, up to 1000 to 1050 watts per square meter. The edge computing node maintains an independent power sample queue for each string and each irradiance intensity interval, with the queue length upper limit set to 200 samples in one embodiment of the invention. Whenever a new power sample falls into a certain irradiance range, the edge computing node determines whether the sample queue corresponding to that range is full. If it is not full, the new sample is directly added to the end of the queue. If it is full, the earliest sample stored in the queue is removed, and then the new sample is added. This adopts a first-in-first-out sliding window mechanism to ensure that the benchmark value can reflect the recent true performance level of the string, and will not be continuously lowered or raised by premature historical data.

[0047] Before adding samples to the queue, edge computing nodes need to perform validity screening on new samples to remove abnormal acquisition values ​​from contaminating the baseline. The screening rules include the following: First, if the power value of a new sample is lower than 60% of the mean of existing samples in the current irradiance range, it is considered that the string may be in an abnormal state during the acquisition period corresponding to the sample, and it will not be included in the baseline sample queue; Second, if the real-time power of more than half of the strings under the inverter is low during the time period corresponding to the acquisition timestamp, it is considered that the current time period is a station-wide low irradiance period, and all samples generated during this period will not be included in the baseline queue to avoid data from overall low irradiance conditions such as cloudy days or cloud cover from being mixed into the clear day baseline, causing the baseline value to be systematically low; Third, if the standard deviation of the real-time power value of a string exceeds 15% of its current mean within 5 consecutive acquisition periods, it is considered that the irradiance fluctuation is large during the current period, and samples during this period will also not be included in the baseline queue.

[0048] When at least 20 valid samples have accumulated in the sample queue for a certain irradiance intensity range, the edge computing node sorts the power values ​​of all samples in the queue from smallest to largest, removes the lowest 10% and highest 10% of samples (i.e., removes approximately 2 to 20 samples from the tail and head of the queue), calculates the arithmetic mean of the remaining samples, and uses this mean as the historical baseline power value of the corresponding string in the current irradiance intensity range, writing it into the baseline power database. This baseline value is recalculated every 5 new valid samples to achieve dynamic rolling updates. In step S2, when performing longitudinal comparison, the historical baseline power value of the corresponding range is read from the baseline power database using the irradiance intensity corresponding to the current acquisition cycle as the index, and the difference is calculated with the current real-time power value to obtain the power attenuation rate. If the current irradiance intensity range has not accumulated enough valid samples (i.e., the number of samples is less than 20), the longitudinal comparison is skipped in this cycle, and only the lateral comparison is performed. The power attenuation rate is recorded as an invalid value in the state evaluation vector. In step S3, when an invalid value is read, the corresponding longitudinal comparison condition does not participate in the classification decision of the current cycle.

[0049] It is important to note that the core intention of the aforementioned benchmark establishment mechanism is to provide a "reasonable expected power under the same illumination conditions" for longitudinal comparison, rather than a simple average of the string under all conditions throughout the day. This distinction is crucial. If benchmarks are not established by dividing the irradiance into intervals, power data from the morning, noon, and evening will overlap, causing the benchmark value to lose its reference value. Consequently, longitudinal comparisons will be unable to effectively distinguish between normal solar cycle variations and actual performance degradation. The interval sliding window design allows the benchmark value to gradually decrease along with the natural aging process of the components, avoiding the continuous false alarm problems that arise with increasing service life when using a fixed factory rated power as the benchmark.

[0050] Third embodiment: This embodiment is an optimization and refinement of the "state evaluation vector" in the first embodiment. It introduces a dynamic weighting coefficient mechanism, assigning different weights to the three evaluation indicators—power deviation rate, power attenuation rate, and temperature exceedance—based on real-time environmental conditions (such as irradiance and temperature) and data reliability. This allows the subsequent state classification decision to maintain more reasonable sensitivity under different operating conditions.

[0051] In step S2, when writing the power deviation rate, power attenuation rate, and temperature exceedance into the state evaluation vector, the three parameters are not stored in a parallel manner with equal weights. Instead, each parameter carries a weight coefficient, which reflects the reliability of the corresponding parameter under the current illumination and temperature conditions. The specific implementation process is as follows: When constructing the state evaluation vector, the edge computing node first reads the irradiance and ambient temperature values ​​of the current acquisition cycle, and assigns weight coefficients to the three parameters according to the following rules: For the power deviation rate, if the current irradiance is higher than 200 watts per square meter, a higher weight is assigned, which is set to 0.6 in one embodiment of the present invention; if the current irradiance is lower than 200 watts per square meter, it is considered that the power difference between strings is small under low irradiance conditions, and the distinguishability of horizontal comparison decreases, so a lower weight is assigned, which is set to 0.3 in one embodiment of the present invention. For the power attenuation rate, if the number of historical benchmark samples within the current irradiance intensity range is sufficient (i.e., not less than 50), the benchmark value is considered reliable, and a weight of 0.3 is assigned; if the number of samples is between 20 and 49, the benchmark value is considered to be in the initial establishment stage, and the reliability is reduced, so a weight of 0.15 is assigned; when the number of samples is less than 20, the power attenuation rate has been recorded as an invalid value, and the weight is reset to zero. Regarding temperature exceeding limits, when the ambient temperature exceeds 35 degrees Celsius, the backplane temperature will naturally be higher, and the reliability of the absolute value of the temperature exceeding the limit will decrease, thus assigning a lower weight of 0.1; under other ambient temperature conditions, a weight of 0.2 is assigned. The actual values ​​of the above three weights are all configurable parameters and can be adjusted according to the geographical location and climate characteristics of the power station during system deployment. The sum of the three weights is not fixed at 1, and the threshold in the hierarchical decision rule of step S3 has been calibrated accordingly according to the above weight range.

[0052] In one embodiment of the present invention, the data structure of the state evaluation vector consists of the following fields in order: string identifier (4-byte integer), acquisition timestamp (8-byte Unix time, precision in milliseconds), real-time power value (4-byte floating-point number, unit in watts), power deviation rate (4-byte signed floating-point number), power deviation rate weight (1 byte, integer value from 0 to 255, corresponding to floating-point weights from 0 to 1.0 magnified by 255 times and then rounded), power decay rate (4-byte signed floating-point number, filled with a specific flag value of 0xFF when invalid), power decay rate weight (1 byte, encoding method as above), temperature exceedance (4-byte signed floating-point number, unit in degrees Celsius), temperature exceedance weight (1 byte, encoding method as above), and vector validity flag (1 byte, used to identify which parameters participated in the valid calculation in this period). The length of the entire vector is fixed at 31 bytes so that step S3 does not need to parse the variable-length structure when reading.

[0053] It should be noted that the purpose of introducing the weighting coefficient is to ensure that the classification decision in step S3 maintains reasonable sensitivity under different environmental conditions, rather than simply weighting and summing the three parameters to obtain a single score and then comparing it with the threshold. Step S3 still compares each parameter with its corresponding threshold separately. The role of the weighting coefficient is reflected in the judgment of the weighted deviation by the classification decision rule, so that the contribution of the power deviation rate under low irradiance conditions is appropriately reduced, and the large number of false alarms in the early morning and evening hours can be avoided to distract the attention of maintenance personnel.

[0054] Fourth embodiment: This embodiment focuses on the false positive elimination logic of the cloud platform when "identifying a persistent low-frequency trend" in the first embodiment. It provides a hierarchical global and regional analysis mechanism to distinguish between a general power reduction caused by a decrease in total irradiance (such as cloud cover) and a genuine individual string fault, thereby improving the accuracy of fault identification.

[0055] Step S6, when identifying a persistent low power trend, needs to exclude station-wide power drops caused by a decrease in overall station irradiance (e.g., cloud cover, dust storms), to avoid misjudging station-wide low power as a persistent low power fault in an individual string. The specific implementation process is as follows: Within each analysis cycle, the cloud-based monitoring platform first counts the number of strings across the entire site with a negative power deviation rate exceeding 15% in absolute terms. The ratio of this number to the total number of strings across the entire site is recorded as the overall low-emission ratio. If the overall low-emission ratio exceeds 70%, the cloud platform considers the current period to be a site-wide low-irradiance or shading event, pausing the counting of continuous low-emission periods for all strings within the current analysis cycle. The accumulated number of consecutive low-emission cycles remains unchanged but does not increase, waiting for data from subsequent collection cycles to return to normal before resuming the count. If the overall low-emission ratio is below 30%, the current low-emission is considered an anomaly in a localized, individual string, and the continuous low-emission counting logic is executed normally. If the overall low-frequency transmission rate is between 30% and 70%, the process will proceed to regional verification: The cloud platform will divide all strings in the station into several sub-regions according to their respective inverters, and calculate the low-frequency transmission rate of each sub-region. If the low-frequency transmission rate of a certain sub-region exceeds 80% while the low-frequency transmission rate of the adjacent sub-region is less than 20%, it is considered that the sub-region may be under local cloud cover, and the continuous low-frequency transmission count of the strings in the sub-region will be suspended; while strings in other sub-regions where the deviation persists will continue to be counted normally for continuous low-frequency transmission.

[0056] It should be noted that the three-stage exclusion logic described above—global pause, normal execution, and regional verification—is more effective than the simple "one-size-fits-all global threshold" approach in handling complex scenarios like localized cloud cover, which are extremely common in large ground-mounted power plants. Large power plants often cover hundreds of acres or even more, and a moving cloud may only obscure one or a few inverter sub-regions. If a global exclusion logic is used, individual string faults that actually exist within the obscured sub-region will be masked, missing the opportunity for early detection. However, regional verification retains the ability to identify persistently low-frequency strings in non-cloud-covered areas even under background noise in the cloud-covered region.

[0057] Fifth embodiment: This embodiment is a detailed description of the "fault type inference mechanism" of the cloud platform in the first embodiment. It uses a predefined rule base to perform pattern matching based on the feature combination of multiple parameters in the historical state data of the string, thereby automatically inferring possible fault types (such as occlusion, component aging, open circuit, etc.) and providing intelligent preliminary diagnostic suggestions for operation and maintenance personnel.

[0058] Step S6 involves adding a string to the maintenance task list. Based on the combination characteristics of each parameter in the historical sequence of the string's state evaluation vector, rule matching is used to infer its most likely fault type. The specific implementation process is as follows: The cloud monitoring platform maintains a fault feature rule base, which contains several pre-defined fault type rules. Each rule consists of a feature description of three parameters and a corresponding fault type label. In one embodiment of the present invention, the rule base includes the following typical rules: Rule 1: If the absolute value of the power deviation rate of a certain string is consistently large (more than 20% in one embodiment of the present invention) in multiple consecutive sampling cycles, and the deviation amplitude is significantly greater during high irradiation periods than during low irradiation periods, while the power attenuation rate is within the normal range (i.e. no significant decrease is observed relative to the historical benchmark), and the backplane temperature does not exceed the limit, then the fault type is presumed to be obstruction type. Possible causes include obstruction by foreign objects, obstruction by projection of adjacent components, or bird droppings contamination.

[0059] Rule 2: If the power deviation rate of a certain string increases slowly over time (i.e., the daily average deviation rate shows a monotonically increasing trend for several consecutive days) and the power attenuation rate increases synchronously, and the backplane temperature is slightly higher than normal, then the fault type is presumed to be component efficiency degradation. Possible causes include component aging, long-term accumulation of surface dirt, or the development of microcracks.

[0060] Rule 3: If the power deviation rate of a certain string suddenly jumps to a large negative value (i.e. the deviation rate changes by more than 25 percentage points between two adjacent acquisition cycles), the current value drops sharply and remains at a low level in subsequent cycles, and the backplane temperature does not show obvious abnormalities, then the fault type is suspected to be an open circuit on the DC side. Possible causes include poor contact in the junction box, open circuit in the DC cable, or blown fuse.

[0061] Rule 4: If the backplane temperature of a certain string continues to exceed the limit (more than 8 degrees Celsius in one embodiment of the present invention) and the power deviation rate is slightly negative or close to zero (i.e. the power has not yet dropped significantly), the fault type is presumed to be early-stage hot spot, that is, the component has local high temperature points but has not yet caused significant power loss, and manual thermal imaging verification is recommended first.

[0062] Rule 5: If the voltage value of a certain string is lower than the historical reference (more than 5% lower in one embodiment of the present invention), but the current value is relatively normal, the power deviation rate is in the range of Level 1 warning to Level 2 warning, and the backplane temperature is normal, then the fault type is presumed to be a partial short circuit in the string series circuit. Possible causes include bypass diode failure or partial component short circuit.

[0063] When the cloud monitoring platform performs fault type inference on a set of strings, it sequentially matches the historical data of the state evaluation vector of the set of strings over the most recent analysis periods with the aforementioned rules one by one. The fault type label corresponding to the successfully matched rule is written into the operation and maintenance task list as the inference result. If multiple rules match simultaneously, all matched fault type labels are listed together and sorted in descending order according to the matching confidence of each rule (i.e., the comprehensive score of the degree of conformity of each feature), for operation and maintenance personnel to make a comprehensive judgment. If no rule matches, the operation and maintenance task list is marked "Pending on-site investigation, no matching fault features" to avoid the platform providing incorrect guidance.

[0064] It should be understood that the terms “comprising” and “including” used in this disclosure and claims indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0065] It should also be understood that the terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims means any combination and all possible combinations of one or more of the associated listed items, and includes such combinations; The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication, characterized in that, Includes the following steps: The electrical and temperature parameters of each photovoltaic string are collected to form the raw data frame; Based on the processing operations of the original data frame, a state evaluation vector containing multiple anomaly evaluation indicators is constructed. The state evaluation vector is classified and judged according to preset rules, a state judgment message is generated, and a local trigger flag is generated when the judgment is the highest level alarm. Based on the priority of the status decision message, it is transmitted to the cloud through a differentiated 5G network slicing channel, and a transmission completion flag is generated when the transmission of high-priority messages is completed. When the local trigger flag and the transmission completion flag are both valid, an isolation command is issued to physically isolate the corresponding string. The received data is aggregated and analyzed in the cloud to generate an operation and maintenance task list and implement closed-loop management.

2. The method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 1, characterized in that, The steps for constructing the state evaluation vector specifically include: Calculate the real-time power value of each string group based on the electrical parameters; The power deviation rate is determined by calculating the real-time power value and the average real-time power of all strings under the same inverter. The power attenuation rate is determined based on the calculation of the real-time power value and the historical reference power value of the corresponding string; The temperature exceedance is calculated based on the difference between the temperature parameter and the preset temperature threshold. The power deviation rate, power attenuation rate, and temperature exceedance are written into the state evaluation vector.

3. The method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 2, characterized in that, Before constructing the state assessment vector, the method further includes the step of establishing and updating the historical reference power value, specifically: maintaining a power sample queue for each string according to the irradiance intensity interval, adding newly collected power samples to the corresponding queue after validity screening, and dynamically updating the historical reference power value of the corresponding irradiance intensity interval based on the statistical processing of valid samples in the queue.

4. The method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 2, characterized in that, The power deviation rate, power attenuation rate, and temperature exceedance in the state assessment vector are each associated with a dynamic weighting coefficient; the dynamic weighting coefficient is determined based on at least one of the irradiance value, ambient temperature value, and sample reliability of the historical reference power value in the current acquisition cycle.

5. The method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 1, characterized in that, The step of transmitting via differentiated 5G network slicing channels includes: transmitting high-priority status decision messages via 5G uRLLC slicing channels, and transmitting ordinary-priority status data via 5G eMBB slicing channels.

6. The method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 1, characterized in that, The method for identifying strings with a persistently low frequency of occurrence in the cloud-based aggregation and analysis step includes: Calculate the proportion of low-frequency serializations across the entire site; If the ratio exceeds the first global threshold, the judgment of continuous low occurrence of all strings across the entire site is suspended. If the ratio is lower than the second global threshold, the continuous low incidence judgment will be executed normally; If the ratio is between the first global threshold and the second global threshold, then the verification is performed by dividing the inverter into sub-regions, and the judgment of continuous low power generation is only suspended in the sub-regions that meet the regional conditions.

7. A method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 1 or 6, characterized in that, The steps for generating the operation and maintenance task list include: matching the historical sequence of the state evaluation vector of the target string with a preset fault feature rule base, inferring the fault type of the target string based on the matching result, and writing the inferred fault type into the operation and maintenance task list.

8. A method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 2, characterized in that, The rule for the graded judgment is as follows: the string status is judged into multiple levels including normal, warning and alarm; among which, the judgment conditions for the highest level alarm include at least one of the following: the absolute value of the power deviation rate exceeds the first deviation threshold, the temperature exceeds the first temperature threshold, or the current value in the electrical parameters is lower than the rated threshold.

9. A method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 3, characterized in that, The rules for valid screening of newly acquired power samples include: if the power value of a new sample is lower than a preset proportion of the average value of existing samples in the current irradiance range, or if more than half of the string power of the same inverter is low in the current time period, or if the power value fluctuates more than a preset fluctuation threshold in multiple consecutive acquisition cycles, then the sample is determined to be an invalid sample and is not included in the power sample queue.

10. A method for string-level monitoring and rapid response of photovoltaic power plants based on 5G communication according to claim 4, characterized in that, The rules for determining the dynamic weighting coefficients include: reducing the weighting coefficient of the power deviation rate when the current irradiance is lower than a preset irradiance threshold; reducing or setting the weighting coefficient of the power attenuation rate to zero when the number of valid samples of the historical reference power value is insufficient; and reducing the weighting coefficient of the temperature exceedance when the current ambient temperature is higher than a preset temperature threshold.

Images