A photovoltaic power station performance monitoring and diagnosis method based on digital twinning

Abstract

The present application mainly relates to the technical field of photovoltaic power station operation and maintenance, in order to solve the problems of coarse granularity of PR index in photovoltaic power station operation and maintenance, incomplete loss decomposition, lack of multi-dimensional comparison ability, single data source of fault diagnosis, non-fusion of multi-source data, and non-reuse of fault knowledge, the present application provides a kind of photovoltaic power station performance monitoring and diagnosis method based on digital twinning, the core is, establish PR index deconstruction system of photovoltaic power station, based on PR index deconstruction system positioning specific level or equipment that influences current PR index, based on ID big data diagnosis, IV inverter IV scanning and CV machine vision inspection multi-source data, for the specific level or equipment that the PR value is low, identify and quantify the loss factor that influences PR index;Finally, based on the loss factor identification result generates hierarchical alarm and standardized operation work order, carries out fault repair, and automatically reviews the output of equipment and PR index recovery, realizes the hierarchical diagnosis of photovoltaic fault diagnosis.

Description

Technical Field

[0001] This invention mainly relates to the field of photovoltaic power plant operation and maintenance technology, and in particular to a method for monitoring and diagnosing the performance of photovoltaic power plants based on digital twins. Background Technology

[0002] The quality of operation and maintenance (O&M) of photovoltaic (PV) power plants directly affects their power generation efficiency and asset lifespan. With the continuous expansion of PV installations, the number of power plants has surged, equipment types have become more diverse, and operating environments have become more complex. Traditional O&M models relying on manual inspections and simple threshold alarms are no longer sufficient to meet the demands of refined and intelligent management. How to accurately monitor power plant performance, quickly locate energy efficiency losses, and accurately diagnose equipment faults has become a core technical problem that urgently needs to be solved in the current PV power plant O&M field.

[0003] In terms of photovoltaic power plant performance monitoring, the performance ratio (PR) is an internationally recognized indicator for evaluating power plant energy efficiency. Its standard definition is the ratio of the actual grid-connected electricity generated by the plant in the past 12 months to the theoretical power generation. However, in the existing technology, the PR indicator is only used as a comprehensive indicator of a single dimension, which has the following obvious shortcomings: (1) The granularity is coarse. It can only present the PR value at the power plant level or subarray level, and cannot penetrate to the collector line level or equipment level, so the root cause of the low PR value cannot be accurately located; (2) The loss decomposition is incomplete. The existing technology only divides the PR loss into three categories: component loss, inverter loss and environmental loss. It does not systematically cover the full-link loss such as curtailment loss, planned maintenance loss, various line losses, and equipment inefficiency loss, and cannot meet the refined requirements of operation assessment for the proportion of various losses; (3) It lacks multi-dimensional comparison capabilities. It cannot realize the comparison of PR values ​​between different components, different tilt angles, different orientations, different brackets, different inverters, and different transformer boxes. It is difficult to find defects in equipment selection, installation design and other aspects through horizontal comparison.

[0004] In terms of fault diagnosis of photovoltaic power plants, the existing technologies are mainly divided into four categories: physical detection method, energy loss method, IV curve method, and time-series voltage and current method. All of the above methods have obvious limitations and are mostly single data source diagnoses with low integration. Specifically: (1) Low accuracy of single data source diagnosis: Diagnosis based on electrical data is easily affected by noise and environmental interference and cannot identify visual defects such as component microcracks and dust accumulation; diagnosis based on machine vision is greatly affected by weather and lighting conditions and cannot reflect abnormal electrical performance of equipment; diagnosis based on IV scanning requires the inverter to be taken out of operation, which will cause human power loss and cannot achieve real-time diagnosis. (2) Multi-source data is not effectively integrated: In the existing technologies, IV scanning, machine vision inspection, big data alarm and other functions are mostly independent modules. The data cannot be interconnected and the diagnostic results are independent of each other, which cannot form a collaborative diagnostic effect and makes it difficult to achieve accurate fault location and root cause analysis. (3) Lack of standardization and accumulability of fault database: The existing technology has not built a unified fault pool, the fault alarm types of different diagnostic modules are not uniform, fault knowledge cannot be accumulated and reused, which leads to maintenance personnel having to spend a lot of time sorting out fault information, resulting in low maintenance efficiency.

[0005] Regarding the current state of related research, the article "A Review of Photovoltaic Array Fault Detection Methods" published in the journal *Electrical Engineering* details the advantages and disadvantages of various fault detection methods, pointing out the limitations of single diagnostic methods, but does not propose a multi-source fusion diagnostic architecture. Some patents have proposed the idea of ​​multi-source data fusion, but these only integrate electrical and meteorological data, excluding IV scanning and machine vision data, and do not construct a unified fault pool. Therefore, the diagnostic precision and accuracy still need improvement.

[0006] In summary, existing technologies have significant shortcomings in the refined management of photovoltaic (PV) power plant performance indicators (PR indicators) and multi-source integrated fault diagnosis. There is an urgent need for a PV diagnostic technology solution that can achieve multi-level penetrating monitoring of PR indicators, full-link loss decomposition, and multi-dimensional comparative analysis to improve the refinement and intelligence of PV power plant operation and maintenance, reduce power generation losses, and ensure the stable and efficient operation of the power plant throughout its entire life cycle. Summary of the Invention

[0007] This invention provides a photovoltaic power plant performance monitoring and diagnosis method based on digital twins, which solves the problems of coarse granularity of PR indicators, incomplete loss decomposition, lack of multi-dimensional comparison capabilities, single data source for fault diagnosis, lack of integration of multi-source data, and inability to reuse fault knowledge in the operation and maintenance of existing photovoltaic power plants.

[0008] The technical solution adopted by the present invention to solve the above-mentioned technical problems

[0009] A method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins includes the following steps:

[0010] Step S1: Establish a photovoltaic power plant PR index deconstruction system, and locate the specific level or equipment affecting the current PR index based on the PR index deconstruction system; the PR index structure system includes power plant PR index, collector line PR index, subarray PR index, and equipment-level PR index;

[0011] Step S2: Based on multi-source data from ID big data diagnosis, IV inverter IV scanning and CV machine vision inspection, identify and quantify the loss factors affecting the PR index for the specific level or equipment with low PR value.

[0012] Step S3: Generate graded alarms and standardized operation and maintenance work orders based on the loss factor identification results, carry out fault repair, and automatically verify the recovery status of equipment output and PR index.

[0013] Furthermore, in step S1, the LSTM time series prediction model and the random forest feature screening algorithm are used to locate the specific level or device that affects the current PR index.

[0014] The LSTM time series prediction model performs outlier removal, missing value filling, and time series alignment on the collected multi-source time series data; predicts theoretical power generation based on historical irradiance and temperature time series data; identifies the time series deviation between actual power generation and theoretical prediction values, and outputs the deviation magnitude and duration.

[0015] The random forest feature selection algorithm takes the time-series bias results output by LSTM as input, selects key loss factors from multiple potential loss factors, quantifies and sorts the contribution of each loss factor, and outputs the key loss types that affect the PR value and their proportions.

[0016] Furthermore, the loss factors mentioned in step S2 include non-technical loss factors and technical loss factors. The non-technical loss factors include curtailment loss, planned maintenance loss, shading loss, dust accumulation loss, and snow and ice cover loss. The technical loss factors include component and DC line loss, inverter conversion loss, transformer and AC line loss, main transformer loss, outgoing line loss, equipment failure loss, and equipment inefficiency loss.

[0017] Furthermore, the formulas for calculating various losses are as follows: ;in This represents the theoretical power generation capacity when there are no corresponding losses. This represents the actual power generation when corresponding losses exist.

[0018] Furthermore, when ,and At that time, it was determined that the photovoltaic power station was experiencing losses, among which, For the set deviation threshold, For the set deviation time, The set deviation time threshold.

[0019] Furthermore, in step S2:

[0020] The ID big data diagnosis is based on the equipment-level theoretical power generation model. By comparing the deviation between theoretical output and actual power generation data in real time, it identifies equipment abnormalities and initially delineates the scope of the fault impact.

[0021] The IV inverter IV scan is triggered for abnormal devices identified by ID diagnosis. By comparing the IV characteristic curve with the standard model, electrical performance defects are identified.

[0022] The CV machine vision inspection targets fault locations identified through ID diagnosis and IV scanning. It leverages UAV infrared or visible light inspection data and uses image recognition algorithms to identify visual defects, thereby achieving visual verification and meter-level precise location of the fault.

[0023] Furthermore, the method also includes: comparing and analyzing the PR indicators between different component types, different installation tilt angles, different installation orientations, different bracket types, different inverter models, and different transformer specifications, and selecting the combination scheme with the optimal power generation efficiency.

[0024] Furthermore, the method also includes constructing a fault pool, specifically including:

[0025] The fault alarm types, coding rules and classification standards of unified ID big data diagnosis, IV inverter IV scanning and CV machine vision inspection are divided into three levels: general fault, relatively serious fault and severe fault according to the scope of fault impact, degree of loss and urgency.

[0026] A fault knowledge graph is constructed based on the diagnostic characteristics, handling procedures, repair effects, and PR index recovery status of historical faults, enabling automatic matching of similar faults and reuse of diagnostic experience.

[0027] Furthermore, the method also includes: synchronously marking the event information of planned maintenance or fault maintenance onto the time axis of the time dimension analysis system, associating the power generation loss during maintenance with maintenance events, and realizing the accurate collection and control of maintenance losses.

[0028] Furthermore, the measured value of the PR index is compared with the preset benchmark value. Based on the comparison results, the health status of the equipment is divided into three levels: normal status, warning status, and abnormal status, and the results are displayed visually. The normal status indicates that the index value is equal to or better than the benchmark value, the warning status indicates that the index value is worse than the benchmark value and the excess is within the preset critical range, and the abnormal status indicates that the index value is significantly worse than the benchmark value and the excess exceeds the preset critical range.

[0029] The beneficial effects of this invention are:

[0030] This invention constructs a multi-level PR index system covering the power plant level, collector line level, subarray level, and equipment level, achieving layer-by-layer penetrating monitoring of PR indicators from the power plant level to the equipment level. It can accurately locate specific levels or equipment with low PR values, solving the shortcomings of traditional technologies that have coarse PR granularity and cannot accurately locate specific issues. By decomposing the power plant level PR value into different loss factors, it achieves accurate quantification and clear proportion of various losses, making the lost power completely transparent and traceable. By constructing a multi-dimensional PR comparison and evaluation system, it realizes the comparison and deviation analysis of PR values ​​between different component types, installation tilt angles, installation orientations, bracket types, inverter models, and transformer specifications, enabling the selection of equipment combination schemes with optimal power generation efficiency and providing quantitative basis for subsequent power plant design and selection. Attached Figure Description

[0031] Figure 1 This is a flowchart of a photovoltaic power plant performance monitoring and diagnosis method based on digital twins, as described in this invention. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are only for explaining the invention and do not constitute a limitation on the scope of protection of this invention.

[0033] I. Establishing a PR Index Decomposition System for Photovoltaic Power Plants

[0034] like Figure 1 As shown, firstly, based on the physical topology of the photovoltaic power station, a digital twin model covering the entire hierarchy of the power station, collector line, subarray, and equipment is built to restore the spatial layout, equipment parameters, electrical topology relationships, installation conditions, and geographical environment information of the power station in a 1:1 ratio.

[0035] Secondly, a fully automated data acquisition and preprocessing system is constructed to collect three core categories of data in real time:

[0036] Environmental operating data: including irradiance, component temperature, ambient temperature and humidity, and time-series irradiance angle;

[0037] Electrical operating data includes inverter AC / DC side voltage and current, string electrical parameters, transformer / main transformer operating parameters, power supply to the substation, and switch signals.

[0038] Specialized testing data includes inverter IV scanning data, drone CV machine vision inspection data, equipment ledgers, historical operation and maintenance records, and repair ledgers.

[0039] The above data is cleaned, time-series aligned, outlier removed, and format standardized to eliminate data noise interference with subsequent calculations and diagnostics.

[0040] First, determine whether there is any power loss in the entire photovoltaic power station. The specific determination method is as follows:

[0041] when ,and At that time, it was determined that the photovoltaic power station was experiencing losses, among which, This represents the theoretical power generation without corresponding losses (obtained from the power simulation system through control variables). This represents the actual power generation with corresponding losses. For the set deviation threshold, For the set deviation time, The set deviation time threshold.

[0042] Then, based on the collected data, the PR value of each level is calculated. In this embodiment, the power station level PR = the actual grid-connected power generation of the power station in the past 12 months / theoretical power generation = 92%; the collector line level PR = the actual grid-connected power generation within the coverage area of ​​each collector line / theoretical power generation, where the PR of the 3rd collector line is 85%; the subarray level PR = the actual power generation of each subarray / theoretical power generation, where the PR of the 5th subarray under the 3rd collector line is 80%; and the equipment level PR = the actual output of each equipment / theoretical output, where the PR of the 3rd inverter under the 5th subarray is 75%. Through layer-by-layer penetration, the specific equipment with a low PR value is accurately located as the 3rd inverter under the 5th subarray.

[0043] II. Identification and Quantification of Loss Factors by Integrating ID+Ⅳ+CV Multi-Source Data

[0044] For the located inverter No. 3, initiate multi-source integrated diagnostics:

[0045] ID Big Data Diagnosis: Based on the device-level theoretical power generation model, the deviation between theoretical output and actual power generation data is compared in real time (in this invention, ID big data diagnosis refers to big data diagnosis based on the comparison of historical data and real-time data of device identifier (ID)). In this embodiment, the theoretical output power of the inverter is predicted by the LSTM time series prediction model, and it is identified that the actual output is about 8% lower than the theoretical value, thus initially identifying an anomaly in the inverter.

[0046] IV Inverter IV Scan: An IV scan is triggered for abnormal devices identified by ID diagnostics to obtain the IV characteristic curves of the strings connected to that inverter. Comparison with the standard model reveals significant curve distortion, characterized by abnormal series resistance, suggesting potential component microcracks or poor connector contact.

[0047] CV (Machine Vision) Inspection: For fault locations identified through ID (Indicator ID) diagnosis and IV (Number IV) scanning, infrared / visible light inspection data from a drone was used. Image recognition algorithms revealed significant dust accumulation on the three components connected to the inverter, and the infrared image showed abnormally high temperatures at the connectors.

[0048] Based on the combined data from the three sources, the confirmed cause of the fault is: dust accumulation on the components leads to poor heat dissipation, which in turn causes overheating of the connectors and a decrease in conversion efficiency. This fault results in a loss of approximately 5% in the inverter's PR value.

[0049] At the same time, for the specific levels identified as causing the low PR value, the specific loss values ​​of various loss factors were calculated respectively. :

[0050] .

[0051] III. Generate tiered alarms and standardized maintenance work orders, and implement closed-loop management.

[0052] Based on the diagnostic results, a tiered alarm is automatically triggered. Simultaneously, the operation and maintenance management module is linked to automatically generate a standardized operation and maintenance work order, containing the following information: fault location, fault type, handling priority, handling process, and rectification suggestions. The work order is automatically dispatched to the corresponding operation and maintenance personnel. The operation and maintenance personnel complete the maintenance according to the work order requirements. After the handling is completed, the system automatically verifies the recovery status of equipment output and performance indicators.

[0053] IV. Multi-dimensional PR Comparison Analysis

[0054] Preferably, this invention includes a multi-dimensional PR comparison analysis, specifically including:

[0055] Horizontal comparison at the equipment level: PR values ​​were compared for different component types, mounting bracket types, and inverter models within the power station. For example, the results showed that the average PR value of subarrays using tracking brackets was 93%, while the average PR value of subarrays using fixed brackets was 88%, with tracking brackets outperforming fixed brackets by approximately 5%. The average PR value of subarrays using component A was 92%, while the average PR value of subarrays using component B was 87%, with component A outperforming component B by approximately 5%. Based on this, a combination of "tracking brackets + component A" is recommended for future new power stations.

[0056] Longitudinal trend analysis over time: Examining the power plant's PR (Polarization Rate) trends over the past two years at a quarterly granular level reveals that PR values ​​are generally about 3% lower each summer. Marking events on the timeline shows that summer coincides with the peak season for sandstorms, significantly increasing ash accumulation and losses. Therefore, it is recommended to increase the frequency of cleaning during the summer.

[0057] Geographic spatial distribution analysis: Automatic PR ranking was performed on a subarray basis. It was found that subarrays with lower PR values ​​were mostly located on the west side of the power station, and this area has a certain slope, with afternoon shadows from the mountains. Therefore, it is recommended to prioritize the use of modules with stronger anti-shadowing capabilities when selecting modules for the western area.

[0058] V. Fault Pool Construction and Knowledge Accumulation

[0059] As a further optimization, this invention also constructs a unified fault pool, standardizing the fault alarm types, coding rules, and grading standards for the three diagnostic modules: ID, IV, and CV. Fault information is stored in the fault pool to construct a fault knowledge graph. When similar faults occur subsequently, the system can automatically match historical cases and recommend handling solutions, enabling the reuse of diagnostic experience.

Claims

1. A method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins, characterized in that, Includes the following steps: Step S1: Establish a photovoltaic power plant PR index deconstruction system, and locate the specific level or equipment affecting the current PR index based on the PR index deconstruction system; the PR index structure system includes power plant level PR index, collector line level PR index, subarray level PR index, and equipment level PR index. Step S2: Based on multi-source data from ID big data diagnosis, IV inverter IV scanning and CV machine vision inspection, identify and quantify the loss factors affecting the PR index for the specific level or equipment with low PR value. Step S3: Generate graded alarms and standardized operation and maintenance work orders based on the loss factor identification results, carry out fault repair, and automatically verify the recovery status of equipment output and PR index.

2. The method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins according to claim 1, characterized in that, In step S1, the LSTM time series prediction model and the random forest feature selection algorithm are used to locate the specific level or device affecting the current PR index; The LSTM time series prediction model performs outlier removal, missing value filling, and time series alignment on the collected multi-source time series data; predicts theoretical power generation based on historical irradiance and temperature time series data; identifies the time series deviation between actual power generation and theoretical prediction values, and outputs the deviation magnitude and duration. The random forest feature selection algorithm takes the time-series bias results output by LSTM as input, selects key loss factors from multiple potential loss factors, quantifies and sorts the contribution of each loss factor, and outputs the key loss types that affect the PR value and their proportions.

3. The method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins according to claim 1, characterized in that, The loss factors mentioned in step S2 include non-technical loss factors and technical loss factors. The non-technical loss factors include curtailment loss, planned maintenance loss, shading loss, dust accumulation loss, and snow and ice cover loss. The technical loss factors include component and DC line loss, inverter conversion loss, transformer and AC line loss, main transformer loss, outgoing line loss, equipment failure loss, and equipment inefficiency loss.

4. The method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins according to claim 3, characterized in that, The formulas for calculating various losses are as follows: ;in, This represents the theoretical power generation capacity when there are no corresponding losses. This represents the actual power generation when corresponding losses exist.

5. The method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins according to claim 4, characterized in that, when ,and At that time, it was determined that the photovoltaic power station was experiencing losses, among which, For the set deviation threshold, For the set deviation time, The set deviation time threshold.

6. The method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins according to claim 2, characterized in that, In step S2: The ID big data diagnosis is based on the equipment-level theoretical power generation model. By comparing the deviation between theoretical output and actual power generation data in real time, it identifies equipment abnormalities and initially delineates the scope of the fault impact. The IV inverter IV scan is triggered for abnormal devices identified by ID diagnosis. By comparing the IV characteristic curve with the standard model, electrical performance defects are identified. The CV machine vision inspection targets fault locations identified through ID diagnosis and IV scanning. It leverages UAV infrared or visible light inspection data and uses image recognition algorithms to identify visual defects, thereby achieving visual verification and meter-level precise location of the fault.

7. The method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins according to claim 1, characterized in that, The method also includes: comparing and analyzing the PR index between different component types, different installation tilt angles, different installation orientations, different bracket types, different inverter models, and different transformer specifications, and selecting the combination scheme with the best power generation efficiency.

8. A method for performance monitoring and diagnosis of photovoltaic power plants based on digital twins according to any one of claims 1-7, characterized in that, The method also includes constructing a fault pool, specifically including: The fault alarm types, coding rules and classification standards of unified ID big data diagnosis, IV inverter IV scanning and CV machine vision inspection are divided into three levels: general fault, relatively serious fault and severe fault according to the scope of fault impact, degree of loss and urgency. A fault knowledge graph is constructed based on the diagnostic characteristics, handling procedures, repair effects, and PR index recovery status of historical faults, enabling automatic matching of similar faults and reuse of diagnostic experience.

9. A method for performance monitoring and diagnosis of a photovoltaic power station based on digital twins according to any one of claims 1-7, characterized in that, The method also includes: synchronously marking the event information of planned maintenance or fault maintenance onto the time axis of the time dimension analysis system, and associating the power generation loss during maintenance with maintenance events to achieve accurate collection and control of maintenance losses.

10. A method for performance monitoring and diagnosis of a photovoltaic power station based on digital twins according to any one of claims 1-7, characterized in that, The method further includes: comparing the measured value of the PR index with a preset benchmark value, dividing the health status of the equipment into three levels: normal state, warning state, and abnormal state according to the comparison result, and displaying the results visually; wherein, the normal state indicates that the index value is equal to or better than the benchmark value, the warning state indicates that the index value is worse than the benchmark value and the excess is within a preset critical range, and the abnormal state indicates that the index value is significantly worse than the benchmark value and the excess exceeds the preset critical range.

Images