A photovoltaic branch power generation loss decoupling and health diagnosis method, device, equipment, medium and product

Abstract

This invention discloses a method, apparatus, equipment, medium, and product for decoupling power generation losses and diagnosing health conditions in photovoltaic (PV) branches. The method includes: collecting current operating data and meteorological environmental data of the PV branch, and determining an ideal power generation baseline based on the current operating data and the meteorological environmental data; determining a loss time series based on the ideal power generation baseline, and performing loss quantification decomposition based on the loss time series to obtain loss decomposition component results; and determining fault type identification results and health status assessment conclusions based on the loss decomposition component results. Through the technical solution of this invention, existing operating data and meteorological environmental data of the power plant can be used to calculate the ideal power generation baseline, quantitatively decouple power generation losses, identify fault types, and assess the health status of each PV branch, enabling timely detection and accurate diagnosis of power generation anomalies in each PV branch.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of intelligent operation and maintenance and condition monitoring and diagnosis technology of photovoltaic power plants, and particularly to a method, device, equipment, medium and product for decoupling photovoltaic branch power generation loss and health diagnosis. Background Technology

[0002] With the continuous growth of global photovoltaic (PV) installed capacity, refined operation and maintenance management of PV power plants has become a core element in ensuring power generation revenue and reducing operating costs. A PV power plant typically consists of several inverters and multiple PV branches (strings) connected to each inverter. Branch-level anomalies are a significant factor causing power generation loss. Timely detection and accurate diagnosis of power generation anomalies in each branch are crucial for improving the overall economic efficiency of the power plant. Summary of the Invention

[0003] This invention provides a method, apparatus, equipment, medium, and product for decoupling and diagnosing power generation losses in photovoltaic branch circuits, enabling timely detection and accurate diagnosis of power generation anomalies in each photovoltaic branch circuit.

[0004] According to one aspect of the present invention, a method for decoupling photovoltaic branch power generation losses and diagnosing health is provided, comprising: Collect current operating data and meteorological environmental data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and the meteorological environmental data; The loss time series is determined based on the ideal power generation reference power, and the loss is quantified and decomposed based on the loss time series to obtain the loss decomposition component results; The fault type identification result and health status assessment conclusion are determined based on the loss decomposition component results.

[0005] According to another aspect of the present invention, a photovoltaic branch power generation loss decoupling and health diagnosis device is provided, the device comprising: The data acquisition and determination module is used to acquire the current operating data and meteorological environment data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and the meteorological environment data; The determination and decomposition module is used to determine the loss time series based on the ideal power generation reference power, and to perform loss quantification decomposition based on the loss time series to obtain the loss decomposition component results; The result determination module is used to determine the fault type identification result and the health status assessment conclusion based on the loss decomposition component result.

[0006] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the photovoltaic branch power generation loss decoupling and health diagnosis method according to any embodiment of the present invention.

[0007] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the photovoltaic branch power generation loss decoupling and health diagnosis method according to any embodiment of the present invention.

[0008] According to another aspect of the present invention, embodiments of the present invention also provide a computer program product, the computer program product including a computer program, which, when executed by a processor, implements the photovoltaic branch power generation loss decoupling and health diagnosis method described in any embodiment of the present invention.

[0009] This invention collects current operating data and meteorological environmental data of photovoltaic (PV) branches, and determines the ideal power generation baseline based on this data. It then determines the loss time series based on the ideal power generation baseline, and performs quantitative decomposition of the loss based on the time series to obtain the loss decomposition component results. Finally, it determines the fault type identification results and health status assessment conclusions based on the loss decomposition component results. Through this invention's technical solution, existing operating data and meteorological environmental data of the power plant can be used to calculate the ideal power generation baseline, quantitatively decouple power generation losses, identify fault types, and assess the health status of each PV branch, enabling timely detection and accurate diagnosis of power generation anomalies in each PV branch.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a photovoltaic branch power generation loss decoupling and health diagnosis method according to an embodiment of the present invention; Figure 2This is a schematic diagram of the structure of a photovoltaic branch power generation loss decoupling and health diagnosis device according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device for implementing the photovoltaic branch power generation loss decoupling and health diagnosis method according to embodiments of the present invention. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0014] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and their derivatives, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.

[0016] Example 1 Currently, the industry mainly uses the following technical solutions for diagnosing branch-level faults and health status in photovoltaic systems, but all of them have significant shortcomings: Option 1: Threshold alarm method based on total deviation ratio (PR value) Current mainstream systems calculate a performance ratio (PR value) by comparing the actual power generation of a branch with the theoretical power generation calculated using a simple conversion based on slope irradiance and component nominal parameters. An alarm is triggered when the PR value falls below a preset threshold. This method has the following drawbacks: (1) The theoretical power calculation model is too simplified and does not take into account systematic factors such as component factory deviation and cable transmission loss, resulting in low accuracy of the reference power itself and high false alarm and false alarm rates.

[0017] (2) Only one total deviation index is output, which cannot distinguish the specific source of power generation loss. Maintenance personnel cannot formulate targeted solutions and often need to go to the site to check one by one, resulting in low maintenance efficiency.

[0018] (3) It relies on a fixed threshold and is difficult to dynamically adapt to changes in environmental conditions and the natural degradation process of components.

[0019] Option 2: Monitoring and Alarm Method Based on Pure Data-Driven Black Box Model This method utilizes extensive historical operational data to train a machine learning model that directly predicts branch power based on meteorological and operational condition inputs, triggering an alarm when the actual power deviates significantly from the predicted value. However, this method has the following drawbacks: (1) Modeling requires a large amount of high-quality sample data under conditions of "good equipment performance, no dust, and no obstruction". In actual power plants, it is extremely difficult to obtain such samples, the data screening workload is large, and the sample quality varies significantly between different power plants and different seasons.

[0020] (2) The model is a black box structure, and the prediction process lacks physical interpretability. Under complex and transient weather conditions (such as fragmented cloud weather), the power prediction value may fluctuate violently, which is contrary to the physical characteristics of the components.

[0021] (3) The model has limited ability to generalize to unseen working conditions, and can only determine whether it is abnormal, but cannot reveal the specific cause of the abnormality or what maintenance measures should be taken.

[0022] Option 3: Add hardware-assisted diagnostic methods Some technical solutions propose installing branch-level irradiance sensors, dust meters, IV (Current-Voltage Characteristic Curve) scanning devices, or shadow imaging equipment at the power plant. While these solutions can provide more accurate local information, they have the following drawbacks: (1) The cost of upgrading existing power stations is extremely high, and it is difficult to deploy on a large scale across hundreds or even thousands of branch lines throughout the station.

[0023] (2) The additional sensor itself also needs to be calibrated and maintained regularly, and there is a risk of drift and failure.

[0024] (3) The complexity of data fusion increases, and the synchronization and alignment of multiple heterogeneous data is a difficult point in engineering applications.

[0025] How to obtain an accurate theoretical power benchmark for a branch circuit that reflects an ideal, clean, unobstructed, and well-functioning state, using only branch-level electrical data from SCADA (Supervisory Control And Data Acquisition) systems commonly found in photovoltaic power plants and environmental data from meteorological stations, while also automatically decomposing the total loss between the actual power generation and the ideal benchmark into physical components of different natures such as dust accumulation, shading, aging degradation, and electrical faults, and providing specific fault type identification results and health status assessment conclusions, has become a technical challenge that urgently needs to be overcome in this field.

[0026] The purpose of this invention is to overcome the shortcomings of the prior art and provide a fully data-driven method for decoupling and diagnosing photovoltaic branch power generation losses based on an ideal reference power, requiring no additional hardware deployment. The method is divided into two parts: an "offline modeling stage" and an "online monitoring and diagnosis stage," with a core architecture of an "ideal reference power engine + multi-dimensional loss feature decoupling and diagnostic tool." Specifically, Stage 1 (offline modeling): Constructing an ideal power generation reference model for the branch to obtain an accurate theoretical power reference. Stage 2 (online monitoring and diagnosis): Real-time calculation of the total power generation loss time series, extraction of multi-dimensional loss features, loss quantification decomposition and fault type identification, and finally outputting a health assessment and operation and maintenance recommendations.

[0027] Figure 1 This is a flowchart of a photovoltaic branch power generation loss decoupling and health diagnosis method according to an embodiment of the present invention. This embodiment is applicable to the case of photovoltaic branch power generation loss decoupling and health diagnosis based on an ideal reference power. The method can be executed by the photovoltaic branch power generation loss decoupling and health diagnosis device according to the embodiment of the present invention. The device can be implemented in software and / or hardware, such as... Figure 1 As shown, the method specifically includes the following steps: S101. Collect the current operating data and meteorological environment data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and meteorological environment data.

[0028] In this embodiment, the current operating data can be SCADA operating data, such as the current DC-side current I_dc and DC-side voltage V_dc of each photovoltaic branch.

[0029] For example, meteorological environmental data may include total horizontal irradiance (GHI) (unit: W / m²). 2 Horizontal diffuse irradiance DHI (if available, unit: W / m²) 2 Ambient temperature T_amb (unit: ℃), wind speed v_wind (unit: m / s), etc.

[0030] It should be noted that the ideal power generation reference power can be the power that should be generated when the branch is ideally clean, unobstructed, and the equipment is in good working order, under the assumption that the external irradiance, temperature and other conditions are exactly the same as the current conditions.

[0031] Specifically, the DC-side current I_dc and DC-side voltage V_dc of each photovoltaic branch are collected, and the total horizontal irradiance GHI (unit: W / m²) of the environmental meteorological station is also collected. 2 Horizontal diffuse irradiance DHI (if available, unit: W / m²) 2 Meteorological environmental data such as ambient temperature T_amb (unit: ℃) and wind speed v_wind (unit: m / s) are used to calculate the ideal power generation baseline for each photovoltaic branch (string) using the existing SCADA operation data and meteorological environmental data of the power station.

[0032] S102. Determine the loss time series based on the ideal power generation reference power, and perform loss quantification decomposition based on the loss time series to obtain the loss decomposition component results.

[0033] It should be noted that the loss time series can be the total power generation loss time series obtained by subtracting the ideal power generation reference power from the current real-time power according to the timestamp alignment.

[0034] In this embodiment, loss quantization decomposition can be achieved by using a physically constrained linear decomposer to decompose the loss into three virtual components with clear physical meaning.

[0035] The loss decomposition results may include: dust accumulation loss component, shadow occlusion loss component, and aging / abnormality loss component.

[0036] Specifically, a loss time series is constructed based on the ideal power generation benchmark, and a linear decomposer with physical constraints is used to decompose the loss into three virtual components with clear physical meaning.

[0037] S103. Determine the fault type identification results and health status assessment conclusions based on the loss decomposition component results.

[0038] Among them, the fault type identification result can identify the specific fault diagnosis type of the photovoltaic branch, and the health status assessment conclusion can be the assessment conclusion of the health status of the photovoltaic branch.

[0039] Specifically, based on the results of loss decomposition components, a well-trained lightweight classifier is used to output specific fault diagnosis labels, thereby obtaining fault type identification results; based on the cumulative diagnosis results over several days or weeks and the historical sequence of each loss component, multi-dimensional indicators that can quantify the health status of branches are calculated, thereby obtaining health status assessment conclusions.

[0040] This invention collects current operating data and meteorological environmental data of photovoltaic (PV) branches, and determines the ideal power generation baseline based on this data. It then determines the loss time series based on the ideal power generation baseline, and performs quantitative decomposition of the loss based on the time series to obtain the loss decomposition component results. Finally, it determines the fault type identification results and health status assessment conclusions based on the loss decomposition component results. Through this invention's technical solution, existing operating data and meteorological environmental data of the power plant can be used to calculate the ideal power generation baseline, quantitatively decouple power generation losses, identify fault types, and assess the health status of each PV branch, enabling timely detection and accurate diagnosis of power generation anomalies in each PV branch.

[0041] Optionally, current operating data and meteorological environmental data of the photovoltaic branch are collected, and the ideal power generation reference power is determined based on the current operating data and meteorological environmental data, including: Collect the current operating data of the photovoltaic branch and determine the current real-time power based on the current operating data.

[0042] In this embodiment, the current real-time power can be calculated based on the current DC-side current I_dc and DC-side voltage V_dc of the photovoltaic branch, P_actual = I_dc × V_dc.

[0043] Specifically, the DC-side current I_dc and DC-side voltage V_dc of each photovoltaic branch are collected, and the real-time power P_actual = I_dc × V_dc is calculated (or the power value is read directly). The data acquisition interval can be set to Δt = 1~5 minutes, preferably 1 minute.

[0044] Meteorological environmental data were collected, and preliminary theoretical power was determined based on the component installation parameters of the photovoltaic branch and the meteorological environmental data.

[0045] First, collect the total horizontal irradiance GHI (unit: W / m²). 2 Horizontal diffuse irradiance DHI (if available, unit: W / m²) 2 Meteorological environmental data such as ambient temperature T_amb (unit: °C) and wind speed v_wind (unit: m / s) are used. If the power station is equipped with a component backplane temperature sensor, its temperature data T_back is used directly; otherwise, it is estimated using ambient temperature and wind speed in subsequent steps.

[0046] Secondly, the meteorological environment data and the current real-time power data are strictly aligned according to the timestamp to form a unified time series dataset D={(t_i,X_i,P_actual,i)}, which serves as the basis for subsequent modeling. Here, X_i is a meteorological feature vector, which can be obtained by extracting specified data features from the meteorological environment data.

[0047] Finally, for any photovoltaic branch at any time t, based on its component installation design parameters (i.e., component installation parameters, such as the installation angle of the photovoltaic branch components) and real-time meteorological data, the preliminary theoretical power P_th0(t) is calculated using a physical mechanism method. Specific sub-steps include: 1) Irradiance conversion In the specific implementation process, an anisotropic sky irradiance conversion model (preferably HDKR (Hay-Davies-Klucher-Reindl Model, HDKR anisotropic scattering model or Hay-Davies-Klucher-Reindl tilted surface scattering model) or Perez (Perez Anisotropic Sky Model or Perez All-Weather Sky Model)) can be used, combined with the installation tilt angle β, azimuth angle γ_az and ground reflectivity ρ_g of the branch component, to convert the total horizontal irradiance GHI into the total irradiance G_poa on the inclined surface of the component: G_poa=GHI×R_b+DHI×R_d+GHI×ρ_g×R_r; Where R_b is the direct irradiance conversion factor, R_b=cos(θ_i) / cos(θ_z); θ_i is the incident angle (the angle between the sunlight and the normal to the component), which is calculated from the solar altitude angle, azimuth angle and the component installation tilt angle and azimuth angle; θ_z is the zenith angle; R_d is the diffuse irradiance conversion factor, which is calculated using an anisotropic model; R_r is the reflected irradiance conversion factor, R_r=(1-cosβ) / 2.

[0048] 2) Incident angle correction The effective irradiance G_eff on the component surface is calculated using the incident angle correction model: G_eff = G_poa × IAM(θ_i); Wherein, IAM(θ_i) is the incident angle correction function, and the ASHRAE model (ASHRAE Clear Sky Model) is preferred: IAM(θ_i)=1-b_0×(1 / cos(θ_i)-1); Where b_0 is a component-specific incident angle correction coefficient, typically ranging from 0.05 to 0.15, which can be obtained by fitting an IAM (Incidence Angle Modifier Curve) curve provided by the component manufacturer.

[0049] Alternatively, a simplified cosine correction model can be used to obtain IAM(θ_i)=cos(θ_i).

[0050] 3) Battery temperature estimation If the measured value of the module backsheet temperature is not provided on-site, the cell operating temperature T_cell will be calculated using a temperature estimation model: T_cell=T_amb+(G_eff / G_ref)×[a+b×v_wind^c]×(NOCT-20) / 800×G_poa×(1-η_stc); Wherein, G_ref=800W / m 2 η_i represents the reference irradiance; a, b, and c are wind speed correction factors, preferably a=3, b=-1.5, and c=0.5 (based on the Faiman model correction); v_wind is the ambient wind speed (m / s); NOCT (Nominal Operating Cell Temperature) is the nominal operating temperature of the module, provided by the module datasheet, with a typical value of 45±3℃; η_stc is the photoelectric conversion efficiency of the module under standard test conditions (STC).

[0051] If a backplane temperature sensor is available on-site, the measured backplane temperature T_back will be used directly, and the battery temperature will be estimated using an empirical thermal resistance model. T_cell=T_back+G_eff / G_ref×ΔT_offset; Wherein, ΔT_offset is the typical temperature difference between the backplane and the battery junction temperature, with a value range of 2~5℃.

[0052] 4) Preliminary theoretical power calculation Based on the component's nominal parameters and the calculated G_eff and T_cell, calculate the preliminary theoretical power P_th0: P_th0=N_s×N_p×P_stc×(G_eff / G_stc)×[1+γ×(T_cell-T_stc)]×η_inv×η_cable; Where N_s is the number of series-connected components in the branch; N_p is the number of parallel-connected components (usually 1); P_stc is the nominal power of the component (under STC conditions, unit: W); G_stc = 1000 W / m 2The standard test condition irradiance is T_stc = 25℃, which is the standard test condition temperature; γ is the peak power temperature coefficient of the module (unit: / ℃), with a typical value of -0.35% / ℃ to -0.45% / ℃; η_inv is the inverter weighted efficiency, ranging from 0.96 to 0.99, which can be obtained by interpolation based on the inverter efficiency curve at the current power point; η_cable is the collector line loss coefficient, η_cable = 1 - L_cable, where L_cable is the line loss rate, with a typical value of 0.01 to 0.03.

[0053] The mechanistic model established by the above steps ensures that the power calculation results can correctly and stably follow the physical trends of changes in external conditions such as solar irradiance and component temperature.

[0054] Meteorological and environmental data are input into the deviation compensation model to obtain the prediction deviation value.

[0055] In this embodiment, the deviation compensation model can be a pre-trained model used to predict deviations based on current meteorological characteristics. The predicted deviation value can be the predicted deviation value output by the deviation compensation model based on the current meteorological characteristics.

[0056] Specifically, the current meteorological characteristics corresponding to the meteorological environment data are input into the deviation compensation model to obtain the prediction deviation value.

[0057] The ideal power generation reference power is determined based on the preliminary theoretical power and the predicted deviation value.

[0058] Specifically, in online operation, for any target branch, after each data collection time window, the following calculations are performed: P_ideal(t)=P_th0(t)+f_delta(X(t)); Where P_th0(t) is the preliminary theoretical power calculated in real time; f_delta(X(t)) is the predicted deviation value output by the trained deviation compensation model based on the current meteorological characteristics X(t). P_ideal(t) is the ideal power generation reference power of this branch at time t. It represents the power that the branch should generate when it is ideally clean, unobstructed, and in good working order, under the assumption that external irradiance, temperature, and other conditions are exactly the same as they are now.

[0059] For calculating ideal reference power, existing technical solutions are either pure mechanistic models or pure data-driven black-box models. However, the technical solution of this invention is a two-layer architecture of "mechanistic model feedforward calculation + data-driven deviation compensation". Compared with the existing technology, the reference power calculated by the technical solution of this invention is significantly improved, and it has both physical interpretability and data adaptability.

[0060] Optionally, the loss time series is determined based on the ideal power generation baseline, and loss quantification decomposition is performed based on the loss time series to obtain the loss decomposition component results, including: The loss time series is determined based on the current real-time power and the ideal power generation baseline.

[0061] In the actual implementation process, the total power loss of each photovoltaic branch at time t is calculated in real time: ΔP_loss(t)=P_ideal(t)-P_actual(t); To eliminate the scaling effect of different irradiance intensities on the absolute value of the loss and to facilitate the generality of subsequent feature extraction, the loss rate is calculated simultaneously: LR(t)=ΔP_loss(t) / P_ideal(t)×100%; The above indicators are stored continuously according to timestamps to form a minute-level loss time series {LR(t_i)}.

[0062] A sliding window feature extraction method is used to extract features from the loss time series, resulting in a set of feature vectors.

[0063] The feature vector set includes: trend and persistence features, fluctuation and mutation features, irradiance correlation features, time and symmetry features, string group comparison features, and natural recovery features.

[0064] In practice, a configurable sliding analysis window W_analysis can be set. The window length can be adaptively selected according to the diagnostic target: short-term window (for detecting sudden failures): T_short = 1~4 hours, preferably 2 hours; medium-term window (for assessing dust accumulation trends): T_medium = 1~7 days, preferably 3 days; long-term window (for assessing aging trends): T_long = 30~90 days, preferably 60 days.

[0065] Next, multi-dimensional feature engineering is performed on the loss time series {LR(t_i)} and branch-related parameters within the window to extract feature vectors F that reflect different physical causes. For example, feature categories and typical feature terms include: (1) Trend and persistence characteristics 1) Mean rate of change in loss rate (reflecting the rate of dust accumulation): f_trend=mean(dLR / dt)=(1 / N)×Σ_{i=1}^{N-1}[LR(t_{i+1})-LR(t_i)] / Δt; Where N is the number of valid sampling points within the window. Dust accumulation typically exhibits a slow, monotonous upward trend over several days to several weeks (f_trend > 0 and persists).

[0066] 2) Monotonic persistence index (distinguishing between gradual and volatile losses): f_mono = N_mono / N; Where N_mono is the maximum number of consecutive sampling points in which the loss rate sequence changes continuously in the same direction (monotonically increasing or monotonically decreasing) within the window. Dust accumulation usually exhibits high monotonic persistence (f_mono→1), while cloud shadow occlusion exhibits low monotonic persistence (f_mono is smaller).

[0067] 3) Long-term linear fit slope (reflecting aging trend): A linear fit LR(t) = k × t + b is performed on the loss rate sequence within the window, and the slope k is extracted as the aging trend feature f_age = k. The component aging degradation exhibits a very slow linear trend at the monthly or even annual level (|k| is small but has a consistent sign).

[0068] (2) Fluctuation and mutation characteristics 1) Standard deviation and coefficient of variation of loss rate: f_std=σ_LR=sqrt((1 / N)×Σ_{i=1}^{N}[LR(t_i)-μ_LR] 2 ); f_cv=σ_LR / μ_LR (when μ_LR≠0); Localized cloud shadows or temporary occlusions often cause sharp, jagged fluctuations in the loss rate (with large values ​​for f_std and f_cv).

[0069] 2) Loss rate kurtosis (reflects the severity of outliers): f_kurt=(1 / N)×Σ_{i=1}^{N}[(LR(t_i)-μ_LR) / σ_LR]^4-3; Hard faults such as short circuits in bypass diodes may cause a sudden and constant increase in the loss rate within a short period of time (f_kurt increases significantly).

[0070] 3) Step mutation detection: Calculate the change in loss rate between adjacent sampling points ΔLR_i = LR(t_{i+1}) - LR(t_i). If there are N_jump consecutive sampling points satisfying |ΔLR_i| > ΔLR_th, then a step jump is determined to have occurred. N_jump ranges from 2 to 5, preferably 3; ΔLR_th is the step threshold, ranging from 5% to 15%, preferably 10%.

[0071] (3) Irradiance correlation characteristics The data within the analysis window is divided into K intervals (bins) based on the effective irradiance G_eff: Low irradiance chamber: G_eff∈[0,G_low), G_low ranges from 200 to 400 W / m 2 300W / m is preferred 2 ; Medium irradiance chamber: G_eff∈[G_low,G_high), G_high ranges from 600 to 800 W / m 2 700W / m is preferred 2 ; High irradiance chamber: G_eff∈[G_high,1200]W / m 2 .

[0072] Calculate the average loss rate μ_LR,k (k=1,2,3) within each irradiance box: f_rad,k=μ_LR,k; And calculate the difference ratio of loss rate between different irradiance chambers: f_rad_diff=max(|μ_LR,k-μ_LR,j|) / mean(μ_LR,k);k,j∈{1,2,3},k≠j; The loss rate caused by uniform dust accumulation is relatively similar across different irradiance ranges (f_rad_diff is small); while fixed shading at specific locations (such as the shadow of a utility pole) only produces significant losses at specific solar locations (corresponding to specific irradiance ranges) (f_rad_diff is large).

[0073] (4) Time and symmetry characteristics 1) Asymmetric index of loss rate during morning and evening hours: Divide the day into a morning period (ω∈[-ω_max,0)) and an afternoon period (ω∈(0,ω_max]) based on the solar hour angle, where ω_max corresponds to 2-3 hours after sunrise / before sunset. Calculate: f_asym=|μ_LR,morning-μ_LR,afternoon| / max(μ_LR,morning,μ_LR,afternoon); East- or west-facing building obstruction usually results in a significant asymmetry in the loss rate before noon or after noon (f_asym→1).

[0074] 2) Shift in the peak time of the loss rate: Calculate the offset of the peak loss rate time t_peak within the window relative to the local solar noon time t_noon: f_offset=|t_peak-t_noon|; Fixed occlusion usually causes the peak loss rate to deviate from noon (larger f_offset), while the peak loss rate of uniform dust accumulation is usually close to noon (smaller f_offset).

[0075] (5) Characteristics of string group comparison Calculate the coefficients of variation of the loss rate of all parallel branches under the same inverter within the same analysis window: f_dispersion=σ_LR,group / μ_LR,group; Where σ_LR,group and μ_LR,group are the standard deviation and mean of the loss rate of all branches under this inverter, respectively.

[0076] Calculate the deviation of the target branch loss rate from the median of the group of branches: f_deviation=|LR_target-LR_median| / LR_median (when LR_median≠0); When dust accumulates uniformly across the entire area, the loss rate of all branches increases synchronously, and the dispersion remains at a low level (f_dispersion<0.1); however, local occlusion or faults in a single branch cause the deviation of that branch to increase significantly (f_deviation>0.3).

[0077] (6) Natural recovery characteristics If a rainfall event (precipitation P_rain > P_th, where P_th ranges from 1 to 5 mm, preferably 2 mm) or an automatic cleaning event is detected within the analysis window, calculate the step decrease in the average loss rate of the target branch before and after the event: f_recovery=μ_LR,before-μ_LR,after; Where μ_LR,before represents the average loss rate of the N_pre sampling points before the event occurs, and μ_LR,after represents the average loss rate of the N_post sampling points after the event occurs. The values ​​of N_pre and N_post correspond to 1 to 3 hours.

[0078] The recoverable loss portion (f_recovery>0) represents washable dust; the unrecoverable loss portion remaining after cleaning tends to be component aging or permanent contamination.

[0079] The feature vector is input into a physically constrained linear decomposer to perform loss quantization decomposition, and the loss decomposition component results are obtained.

[0080] The constraints of the linear decomposer with physical constraints include: the ash accumulation component satisfies the upper limit of daily change rate constraint, the shading component satisfies the lower limit of fluctuation and irradiance correlation constraint, and the aging / abnormal component satisfies the upper limit of short window change rate constraint.

[0081] Specifically, a physically constrained linear decomposer is used to decompose the average total loss rate LR_mean of the current analysis window into three virtual components with explicit physical meaning: LR_mean=LR_dust+LR_shade+LR_degrad; Wherein, LR_dust is the dust accumulation loss component; LR_shade is the shadow occlusion loss component; and LR_degrad is the aging / abnormality loss component.

[0082] In actual operation, the constraint conditions are decomposed as follows: Constraint 1 (Dust Component Constraint): LR_dust≥0, and satisfies the trend constraint: |dLR_dust / dt|≤α_dust×LR_dust; Among them, α_dust is the upper limit coefficient of the ash accumulation change rate, with a value range of 0.01~0.05 / day, reflecting the characteristic of ash accumulation being slow and gradual over several days to several weeks.

[0083] Constraint 2 (Occlusion Component Constraint): LR_shade≥0, and satisfies the fluctuation constraint: σ_LR_shade≥β_shade×LR_shade; Wherein, β_shade is the lower limit coefficient of shading fluctuation, with a value range of 0.2~0.5, reflecting the high volatility of shading loss; at the same time, LR_shade is required to be positively correlated with the irradiance correlation feature f_rad_diff and the time asymmetry feature f_asym.

[0084] Constraint 3 (Aging / Abnormal Component Constraint): LR_degrad≥0, and the residual constraint is satisfied: |dLR_degrad / dt|≤γ_degrad×LR_degrad; Among them, γ_degrad is the upper limit coefficient of aging change rate, with a value range of 0.001~0.01 / day, reflecting the characteristic that aging is approximately constant or has step changes within a short window.

[0085] The optimization objective is to minimize the decomposition residuals: min||LR_mean-(LR_dust+LR_shade+LR_degrad)|| 2 +λ_1×Reg(LR_dust)+λ_2×Reg(LR_shade)+λ_3×Reg(LR_degrad); Where Reg(·) is the regularization term, and λ_1, λ_2, and λ_3 are regularization coefficients, with values ​​ranging from 0.01 to 0.1.

[0086] This decomposition process achieves quantitative decoupling of the loss sources through feature constraints, and outputs the contribution value of each component to the total loss.

[0087] Regarding the multidimensional decoupling mechanism for loss, existing technical solutions only calculate the total PR value or total loss. However, the technical solution of this invention is a six-dimensional feature extraction (trend, fluctuation, irradiance correlation, time symmetry, string discretization, and restorativity) + linear decomposition with physical constraints. Compared with existing technologies, the technical solution of this invention can achieve quantitative decoupling of loss sources, moving from "knowing the total amount" to "knowing the composition".

[0088] Optionally, the fault type identification results and health status assessment conclusions are determined based on the loss decomposition component results, including: Based on the target classifier, fault diagnosis labels are determined according to the loss decomposition component results and feature vector set, and fault type identification results are determined according to the fault diagnosis labels.

[0089] In this embodiment, the target classifier can be a gradient boosting tree (the parameter configuration can be: the number of trees n_estimators ranges from 50 to 200, the maximum depth max_depth ranges from 3 to 8, and the learning rate learning_rate ranges from 0.01 to 0.1) or a fuzzy decision tree (the membership function adopts Gaussian or triangular shape).

[0090] Specifically, based on the loss decomposition component results and the original multidimensional feature vector F, a well-trained lightweight classifier is used to output specific fault diagnosis labels, and the fault type identification result can be determined based on the fault diagnosis labels.

[0091] For example, a typical set of output labels Y includes: y_1: normal; y_2: slight uniform dust accumulation; y_3: severe uniform dust accumulation; y_4: temporary local shading (cloud shadows / vegetation); y_5: fixed building / structure shading; y_6: component performance degradation (aging and decay); y_7: branch electrical anomaly (bypass diode short circuit / hot spot risk, etc.); y_8: inverter branch efficiency decline.

[0092] This classifier only needs to be trained using a small number of typical abnormal event samples that have been manually verified and labeled in history (no less than N_label samples for each class label, with N_label ranging from 20 to 50, preferably 30) to achieve high-accuracy fault identification.

[0093] The health status index of the photovoltaic branch is determined based on the loss decomposition component results, and the health status assessment conclusion is determined based on the health status index.

[0094] Specifically, based on the cumulative diagnostic results over several days or weeks and the historical sequence of each loss component, a multi-dimensional health status index that can quantify the health status of the branch is calculated, and then a health status assessment conclusion is drawn based on the health status index.

[0095] The health status indicators include: cleanliness and health index, long-term health index, and risk level of sudden failure.

[0096] 1) Cleanliness and Health Index (CHI) CHI=max(0,min(100,(1-LR_dust,7d)×100)); Wherein, LR_dust,7d is the moving average of the ash accumulation loss component over the past 7 days. This index reflects the severity of power generation loss caused by ash accumulation in the current branch and the urgency of cleaning. The lower the index, the more urgently cleaning needs to be arranged.

[0097] For example, the index classification is as follows: CHI≥95: Good cleanliness, no cleaning required; 85≤CHI<95: Slight dust accumulation, it is recommended to pay attention during routine inspections; 70≤CHI<85: Moderate dust accumulation, it is recommended to arrange cleaning within 2 weeks; CHI<70: Severe dust accumulation, it is recommended to arrange cleaning within 7 days.

[0098] 2) Long-term Health Index (LHI) LHI=max(0,min(100,(1-LR_degrad,90d)×100)); Wherein, LR_degrad,90d is the moving average of the aging loss component over the most recent 90 days. This index tracks the irreversible performance degradation process of branch components, providing a quantitative basis for medium- to long-term technical upgrades or component replacement decisions.

[0099] For example, the index classification is as follows: LHI≥95: excellent module performance; 90≤LHI<95: slight aging, it is recommended to pay attention during the annual maintenance; 80≤LHI<90: moderate aging, it is recommended to arrange IV curve testing to confirm; LHI<80: severe aging, it is recommended to assess the necessity of module replacement.

[0100] 3) Risk level of sudden failure If the fault type inference layer outputs a high-risk label such as "branch electrical anomaly" (y_7), a high-risk warning is immediately triggered. The risk level R is determined based on the loss rate: R=1 (low risk): LR_mean < 10%; R=2 (medium risk): 10% ≤ LR_mean < 20%; R=3 (high risk): 20% ≤ LR_mean < 30%; R=4 (emergency): LR_mean ≥ 30%.

[0101] For fault identification architecture, existing technical solutions are single threshold alarms or black box classification, while the technical solution of this invention is a two-level serial hybrid expert model of "loss quantification decomposition + fault type reasoning". Compared with the existing technology, the technical solution of this invention can improve the accuracy of fault type identification and output interpretable fault labels.

[0102] For health assessment systems, existing technical solutions rely on a single PR threshold alarm, while the technical solution of this invention uses Clean Health Index (CHI) + Long-Term Health Index (LHI) + Sudden Risk Level. Compared with existing technologies, the technical solution of this invention provides quantitative health assessment across multiple time scales, directly driving differentiated operation and maintenance decisions.

[0103] Optionally, the training process for the bias compensation model is as follows: Historical operating data of photovoltaic branches are collected, and historical real-time power is determined based on the historical operating data.

[0104] Specifically, historical SCADA operation data for each photovoltaic branch is collected, including DC-side current and DC-side voltage, and historical real-time power is calculated (or the power value is read directly). The data collection interval is set to Δt = 1~5 minutes, preferably 1 minute.

[0105] Historical meteorological data was collected, and the historical theoretical power was determined based on the component installation parameters of the photovoltaic branch and the historical meteorological data.

[0106] In actual operation, this step is implemented on the same principle as the step of "collecting meteorological environmental data and determining the preliminary theoretical power based on the component installation parameters of the photovoltaic branch and meteorological environmental data", so it will not be described in detail here.

[0107] A regression model is trained based on historical meteorological data, historical real-time power, and historical theoretical power to obtain a bias compensation model.

[0108] It should be noted that before training the regression model, an automated screening process for high-quality, dust-free, and unobstructed working conditions is required. This process aims to automatically obtain a small number of high-quality samples needed to build a high-precision benchmark model with minimal manual intervention. Specifically, this process includes the following steps: 1) Hard filtration The following unqualified data are excluded: Nighttime data: GHI < G_night, where G_night is the nighttime irradiance threshold, with a value range of 10~50W / m 2 , preferably 20W / m 2 . Data during the branch fault alarm period: Determined according to the SCADA system fault flag bit or current abnormality (I_dc < 0.1A and GHI > 100W / m 2 ). Transition data at the moment of inverter start and stop: The first N_start sampling points after the inverter is started and the N_stop sampling points before shutdown, N_start and N_stop have a value range of 3~10, preferably 5 sampling points. Invalid data with sensor dead values or obvious out-of-limit: Including GHI > 1500W / m 2 (exceeding the physical limit), T_amb < -40°C or T_amb > 60°C, v_wind < 0 or v_wind > 50m / s, etc.

[0109] 2) Irradiance stability detection A sliding time window W_stable is set, with a window length T_w = 10~15 minutes, that is, containing N_w = T_w / Δt sampling points. Calculate the standard deviation σ_G and coefficient of variation CV of GHI in the window: CV=σ_G / μ_G; where μ_G is the mean value of GHI in the window. Only the period data with CV < CV_th are retained, where CV_th is the stability threshold, with a value range of 0.03~0.08, preferably 0.05 (i.e., 5%). This batch of data represents stable working conditions without fast-moving cloud occlusion.

[0110] 3) Module cleaning state benchmark locking Prioritize calling the power station's historical cleaning records (cleaning dates and cleaning branch ranges). The complete daytime period on the N_dry-th day after each cleaning (to ensure the module surface is completely dry and stable, N_dry has a value range of 2~5, preferably 3) that meets the irradiance stability condition and is in the medium-high irradiance interval (GHI > G_mid, G_mid has a value range of 300~500W / m 2 , preferably 400W / m 2 ) is marked as the "benchmark cleaning window" W_clean.

[0111] If there are no historical cleaning records, or the cleaning records are insufficient to cover the representative working conditions of each season, an alternative scheme is enabled: select several time periods in the historical data where the relative deviation between the power of each branch and the median P_median of the power of all branches under the affiliated inverter in the same period is less than δ_clean as candidate cleaning windows: |P_actual-P_median| / P_median<δ_clean; Wherein, δ_clean is the cleanliness consistency threshold, with a value ranging from 0.01 to 0.03, preferably 0.02 (i.e., 2%). At the same time, it is required that the seasonal performance index (monthly PR value) of this branch within the candidate window is in the upper quartile range of the historical operating data.

[0112] 4) Shadow Occlusion Elimination Shadow identification is performed using the current dispersion rate among multiple branches with the same installation tilt angle under the same inverter. For the branch set S_inv under the same inverter, the relative dispersion rate of each branch current I_k and the median current I_median of the set is calculated under steady-state conditions: RD_k=|I_k-I_median| / I_median; If RD_k > RD_th, then the k-th branch is considered to have a local temporary shadow, and the data samples corresponding to this period are not included in the high-quality sample set. RD_th is the current discrete threshold, with a value range of 0.03 to 0.08, preferably 0.05 (i.e., 5%).

[0113] 5) Confirmation of excellent performance For the candidate data ranges after the above screening, it is further required that the ratio of the actual power of the branch to the calculated preliminary theoretical power (i.e., the PR value) falls within the high value range of the branch's historical operating data: PR = P_actual / P_th0 > PR_min; Among them, PR_min is the high-performance threshold, with a value range of 0.85~0.92, preferably 0.88. At the same time, it is required that the PR value does not have obvious step decay phenomenon within this period (the rate of change of PR value between adjacent sampling points |ΔPR / Δt|<0.05 / h).

[0114] After the above automatic filtering process, a batch of "high-quality operating condition sample segments" are extracted for each branch. Each sample point contains a meteorological feature vector X and the corresponding SCADA measured power P_actual. X=[G_eff,T_cell,v_wind,DHI_ratio,t_hour,sin(δ_sun),cos(δ_sun)]; Wherein, G_eff is the effective irradiance; T_cell is the cell temperature; v_wind is the wind speed; DHI_ratio=DHI / GHI is the ratio of diffuse irradiance (if DHI data is available, otherwise set to a missing value or default value of 0.2); t_hour is the number of hours in a day (0~24), used to capture intraday nonlinear effects; δ_sun is the solar declination angle, sin(δ_sun) and cos(δ_sun) are used to capture seasonal variation patterns.

[0115] For each high-quality sample point, calculate the power deviation: ΔP = P_actual - P_th0; Construct a lightweight regression model f_delta, using the meteorological feature vector X as the model input and the corresponding time-time ΔP as the model output target, and perform supervised learning training. The model type can be one of the following: Support Vector Regression, with the preferred kernel function being RBF (Radial Basis Function Kernel), a penalty parameter C ranging from 10 to 1000, and a kernel parameter γ ranging from 0.001 to 0.1; Gradient Boosting Tree (LightGBM or XGBoost), with the number of trees n_estimators ranging from 50 to 200, the maximum depth max_depth ranging from 3 to 8, and the learning rate learning_rate ranging from 0.01 to 0.1; or a shallow fully connected neural network with 12 hidden layers, 1664 neurons per layer, ReLU activation function, Adam optimizer, and a learning rate of 0.001 to 0.01.

[0116] Since the variation in deviation ΔP is much smaller than the variation in absolute power, the model converges more easily. Furthermore, the learned patterns primarily reflect systematic factors such as inverter efficiency nonlinearity, low-irradiance weak light effects, and the slight dispersion of component factory parameters, without including noise introduced by dust or shading. After training, a deviation compensation model f_delta is obtained, bound to each branch (or cluster of similar branches).

[0117] For the automatic screening of high-quality working conditions, existing technical solutions involve manual screening or simple threshold filtering. However, the technical solution of this invention is a multi-level automatic screening (hard filtering → stable detection → cleaning lock → shadow elimination → performance confirmation). Compared with the existing technology, the technical solution of this invention does not require manual annotation, automatically obtains high-quality samples, and improves the sample screening efficiency.

[0118] Optionally, after determining the fault type identification results and health status assessment conclusions based on the loss decomposition component results, the following may also be included: Based on the fault type identification results and health status assessment conclusions, maintenance work order suggestions are generated.

[0119] Specifically, based on the fault type identification results and health status assessment conclusions obtained from the above process, maintenance work order suggestions with priority ranking are automatically generated.

[0120] For example, the suggested format for an operation and maintenance work order can be: "[Branch Number]: [Fault Type Label], Cleanliness and Health Index [CHI Value] Score, Long-Term Health Index [LHI Value] Score, [Risk Level] Level Risk, Suggested [Specific Measures], Suggested Completion Deadline [Time Range]". Typical examples: "#12 branch: Severe and uniform dust accumulation, cleanliness and health index 75 points, long-term health index 96 points, level 1 risk, it is recommended to arrange branch cleaning, recommended completion time limit within 7 days"; "#05 branch: suspected tree shadow obstruction in the morning, cleanliness and health index 98 points, long-term health index 95 points, level 1 risk, it is recommended to trim or check the environment around the array, recommended completion time limit within 30 days"; "#02 branch: component performance degradation (aging and attenuation), cleanliness and health index 92 points, long-term health index 88 points, level 2 risk, it is recommended to arrange IV curve testing to confirm the degree of attenuation, recommended completion time limit within 60 days"; "#08 branch: branch electrical abnormality (bypass diode short circuit risk), cleanliness and health index 97 points, long-term health index 94 points, level 3 risk, it is recommended to immediately stop the machine to check the branch electrical connection and bypass diode status, recommended completion time limit within 24 hours."

[0121] As an exemplary description of an embodiment of the present invention, a large-scale ground-mounted photovoltaic power station with a rated capacity of 10MW is used as an example. This power station is equipped with 50 string inverters, each connected to approximately 20 photovoltaic branches. Each branch consists of 28 monocrystalline silicon modules with a rated power of 450W connected in series, and the nominal power of each branch is 12.6kW. The power station is equipped with a standard weather station, providing 1-minute data on total horizontal irradiance (GHI), ambient temperature, and wind speed. The SCADA system collects and stores the DC current and DC voltage of each branch at 1-minute intervals.

[0122] Phase 1 (Offline Modeling): Construct an ideal power generation benchmark model for the branch circuit to obtain an accurate theoretical power benchmark.

[0123] 1) Data Acquisition and Preprocessing The system collects SCADA branch-level operational data and weather station data for the power station for 12 months, from January 1, 2023 to December 31, 2023. After time alignment and format standardization, a unified historical database is formed. The total amount of data is approximately: 50 inverters × 20 branches / unit × 525,600 minutes / year ≈ 525.6 million records.

[0124] 2) Preliminary theoretical power calculation For each branch, based on its component installation tilt angle β=23°, azimuth angle γ_az=0° (due south direction), and ground reflectivity ρ_g=0.2, the slope irradiance G_poa is calculated using the HDKR model, and G_eff is obtained after incident angle correction (ASHRAE model, b_0=0.1).

[0125] T_cell was calculated using a temperature estimation model with the following parameters: NOCT = 42℃; η_stc = 0.204 (20.4%); wind speed correction factors a = 3, b = -1.5, c = 0.5.

[0126] Preliminary theoretical power calculation parameters: N_s=28; N_p=1; P_stc=450W; γ=-0.35% / ℃=-0.0035 / ℃; η_inv=0.985 (dynamic interpolation based on the inverter efficiency curve at the current power point); η_cable=0.98 (line loss rate 2%).

[0127] 3) Automatic screening of high-quality operating conditions Execute the high-quality operating condition automatic screening program, parameter configuration: nighttime threshold G_night=20W / m 2 Inverter transition data removal: N_start=N_stop=5 sampling points; Stability detection window T_w=15 minutes (15 sampling points); Stability threshold CV_th=0.05; Drying waiting days after cleaning N_dry=3 days; Medium-high irradiance threshold G_mid=400W / m 2 Cleanliness consistency threshold δ_clean=0.02; Current dispersion threshold RD_th=0.05; Excellent performance threshold PR_min=0.88.

[0128] The power station undergoes a comprehensive cleaning every quarter, resulting in four cleaning records per year. By accessing these records, the high-irradiance period from 10:00 AM to 2:00 PM on the third day after each cleaning can be identified as the baseline cleaning window. Branch periods with current dispersion exceeding 5% under the same inverter are excluded. Ultimately, approximately 400 to 600 high-quality sample points are selected for each branch, totaling approximately 20,000 to 30,000 sample points per branch throughout the year.

[0129] 4) Training of the deviation compensation model X = [G_eff, T_cell, v_wind, DHI_ratio, t_hour, sin(δ_sun), cos(δ_sun)] is the input feature vector, and ΔP = P_actual - P_th0 is the target variable. A LightGBM regression model is trained independently for each branch as f_delta. Model parameter configuration: number of trees n_estimators = 100; maximum depth max_depth = 6; learning rate learning_rate = 0.05; number of leaf nodes num_leaves = 31; feature sampling ratio (column sampling) feature_fraction = 0.8; data sampling ratio (row sampling) bagging_fraction = 0.8; bagging frequency bagging_freq = 5.

[0130] Using 5-fold cross-validation, after training convergence, the mean absolute error (MAE) of the ideal baseline power model on the validation set is approximately 30W. Relative to the nominal branch power of 12.6kW, the relative error is approximately 0.24%. Compared to the pure mechanistic model (MAE approximately 120W, relative error 0.95%), the accuracy is improved by approximately 75%.

[0131] Phase Two (Online Monitoring and Diagnosis): Real-time calculation of total power generation loss sequence, extraction of multi-dimensional loss features, quantitative decomposition of losses and identification of fault types, and finally output of health assessment and operation and maintenance recommendations.

[0132] Taking the online diagnostic process of a branch circuit anomaly that occurred on a certain morning at the power station as an example: 1) Real-time loss calculation At 10:15 AM, the system calculated the following indicators for branch #08 in real time: Current meteorological conditions: GHI = 680 W / m 2 Ambient temperature T_amb=22℃, wind speed v_wind=3.5m / s; Real-time output of the ideal reference power engine: P_ideal(08)=8920W; SCADA displays the actual power of this branch: P_actual(08)=6580W; Calculate the total loss: ΔP_loss=8920-6580=2340W; Loss rate: LR=2340 / 8920×100%=26.2%.

[0133] 2) Multidimensional feature extraction Multidimensional feature extraction was performed using the cumulative window data (255 one-minute sampling points) from 06:00 on the current day to 10:15 at the current time. Trend and persistence characteristics: The loss rate within the window remained relatively stable in the range of 3% to 5% from 06:00 to 10:00, and then suddenly changed after 10:00; the first-order difference mean f_trend was close to 0 before 10:00, and then suddenly increased to a positive value after 10:00; the monotonic persistence index f_mono decreased after the sudden change.

[0134] Fluctuation and abrupt change characteristics: Before 10:00, the standard deviation of the loss rate f_std≈0.8% and the coefficient of variation f_cv≈0.2; after 10:00, f_std suddenly increased to 4.5% and f_cv suddenly increased to 0.35; the kurt value f_kurt suddenly increased from -0.5 to 5.2 around 10:00, indicating a significant step change.

[0135] Irradiance correlation characteristics: Data after 10:00 were binned according to G_eff (low: [0,300), medium: [300,700), high: [700,1200] W / m 2 The study found that the average loss rate was μ_LR,high = 28.5% for the high-irradiance box, μ_LR,mid = 15.2% for the medium-irradiance box, and μ_LR,low = 4.1% for the low-irradiance box. The irradiance difference ratio f_rad_diff = 0.86, indicating that the loss is strongly correlated with irradiance.

[0136] Temporal and Symmetry Characteristics: The loss rate μ_LR,morning = 12.3% during the morning session (06:00-12:00). Assuming symmetry in the afternoon, the expected loss rate should be lower. However, since the current time is morning, the asymmetry index f_asym needs to be calculated using historical afternoon data. The historical afternoon loss rate for the same period is approximately 4.5%, and f_asym = |12.3-4.5| / 12.3 = 0.63, indicating a significant asymmetric loss during the morning session.

[0137] String group comparison characteristics: Under the same inverter, the mean loss rate of the other 19 branches within the same window is μ_LR,group=4.2%, the standard deviation is σ_LR,group=0.8%, and the coefficient of variation is f_dispersion=0.19. The loss rate of branch #08 is LR_target=26.2%, and the deviation from the median LR_median=4.1% is f_deviation=|26.2-4.1| / 4.1=5.39, which is significantly outside the normal range.

[0138] Natural recovery characteristic: No rainfall or cleaning event records are found within the window, f_recovery=0.

[0139] 3) Loss decomposition and fault identification The above features are then decomposed into loss quantification and fault type identification.

[0140] Decomposition results: Dust loss component LR_dust=1.5% (based on trend constraints and string group synchronization judgment; since the loss rate of other branches is only about 4%, this component reflects the overall basic dust accumulation level of the station); Shading loss component LR_shade=22.0% (based on high fluctuation constraints, strong irradiance correlation, and time asymmetry judgment); Aging / abnormality loss component LR_degrad=2.7% (residual component, including basic aging and measurement noise).

[0141] Verification: LR_dust + LR_shade + LR_degrad = 1.5% + 22.0% + 2.7% = 26.2% = LR_mean, decomposition closure.

[0142] Fault type inference: Based on the loss decomposition results and original features, the LightGBM classifier outputs the label: "#08 branch: temporary local occlusion (suspected shadow of building or equipment appearing after 10 o'clock)".

[0143] 4) Maintenance work order generation suggestion The system generates an alarm work order: "#08 Branch: Temporary partial shading (suspected shadow of building or equipment after 10:00), cleanliness and health index 98 points, long-term health index 95 points, level 3 risk. It is recommended to immediately check for new shading sources around the #08 branch array after 10:00 (such as temporary equipment, vegetation growth, etc.). It is recommended to complete the task within 24 hours." Maintenance personnel went to the site for verification and found that a recently added outdoor air conditioning unit cast a strip-shaped shadow on the lower half of the components of branch #08 after 10:00 AM, confirming the diagnosis. After removing the obstruction source, the branch power immediately returned to normal. This diagnostic process was completed without the aid of any additional sensors, relying entirely on SCADA and existing weather station data.

[0144] It should be noted that the technical methods of this invention are not only applicable to large-scale ground-mounted photovoltaic power plants, but also, after parameter adaptation, applicable to: distributed rooftop photovoltaic systems (adjusting installation tilt angle and azimuth angle parameters); floating photovoltaic power plants (adjusting the wind speed correction coefficient in the temperature estimation model); bifacial photovoltaic module power plants (adding a back irradiance term in irradiance conversion); and integrated photovoltaic and energy storage power plants equipped with energy storage systems (considering the impact of energy storage charging and discharging in power calculation).

[0145] Regarding hardware dependence, existing technical solutions require the addition of irradiance sensors, dust meters, IV scanning equipment, etc., while the technical solution of this invention only utilizes existing SCADA and weather station data, requiring zero additional hardware. Compared with existing technologies, the technical solution of this invention reduces the transformation cost and is suitable for rapid full-coverage deployment in existing power plants.

[0146] The technical solution of this invention includes: a dual-driven ideal benchmark power engine based on "mechanism + data" to overcome the dual bottlenecks of insufficient accuracy of pure mechanism models and large sample requirements of pure data models; a multi-level automatic high-quality operating condition screening mechanism to solve the problem of obtaining high-quality samples and achieve automatic sample screening without human intervention; a six-dimensional loss feature extraction and physical constraint decomposition mechanism to achieve quantitative decoupling of branch-level power generation losses from total amount to components; a two-level hybrid expert fault identification architecture to achieve accurate mapping from loss decomposition to interpretable fault labels; and a dual-index health assessment system to upgrade the operation and maintenance mode from passive alarm to proactive prevention.

[0147] Example 2 Figure 2 This is a schematic diagram of a photovoltaic branch power generation loss decoupling and health diagnosis device according to an embodiment of the present invention. This embodiment is applicable to situations involving photovoltaic branch power generation loss decoupling and health diagnosis based on an ideal reference power. The device can be implemented using software and / or hardware, and can be integrated into any device that provides photovoltaic branch power generation loss decoupling and health diagnosis functions, such as… Figure 2 As shown, the photovoltaic branch power generation loss decoupling and health diagnosis device specifically includes: a data acquisition and determination module 201, a determination and decomposition module 202, and a result determination module 203.

[0148] The acquisition and determination module 201 is used to acquire the current operating data and meteorological environment data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and the meteorological environment data. The determination and decomposition module 202 is used to determine the loss time series based on the ideal power generation reference power, and to perform loss quantification decomposition based on the loss time series to obtain the loss decomposition component results; The result determination module 203 is used to determine the fault type identification result and the health status assessment conclusion based on the loss decomposition component result.

[0149] Optionally, the acquisition and determination module 201 is specifically used for: Collect the current operating data of the photovoltaic branch and determine the current real-time power based on the current operating data; Meteorological environmental data is collected, and preliminary theoretical power is determined based on the component installation parameters of the photovoltaic branch and the meteorological environmental data; The meteorological and environmental data are input into the deviation compensation model to obtain the predicted deviation value; The ideal power generation reference power is determined based on the preliminary theoretical power and the predicted deviation value.

[0150] Optionally, the determination and decomposition module 202 is specifically used for: The loss time series is determined based on the current real-time power and the ideal power generation reference power. Sliding window feature extraction is performed on the loss time series to obtain a set of feature vectors; The feature vector is input into a physically constrained linear decomposer for loss quantization decomposition to obtain the loss decomposition component results.

[0151] Optionally, the result determination module 203 is specifically used for: Based on the target classifier, fault diagnosis labels are determined according to the loss decomposition component results and the feature vector set, and fault type identification results are determined according to the fault diagnosis labels. The health status index of the photovoltaic branch is determined based on the loss decomposition component results, and the health status assessment conclusion is determined based on the health status index.

[0152] Optionally, the training process of the deviation compensation model is as follows: Collect historical operating data of the photovoltaic branch and determine the historical real-time power based on the historical operating data; Historical meteorological data is collected, and the historical theoretical power is determined based on the component installation parameters of the photovoltaic branch and the historical meteorological data; Based on the historical meteorological data, the historical real-time power, and the historical theoretical power, a regression model is trained to obtain a deviation compensation model.

[0153] Optionally, the device further includes: The suggestion generation module is used to generate maintenance work order suggestions based on the fault type identification results and the health status assessment conclusions.

[0154] The above-mentioned products can perform the photovoltaic branch power generation loss decoupling and health diagnosis method provided in any embodiment of the present invention, and have the corresponding functional modules and beneficial effects of the method.

[0155] Example 3 Figure 3 A schematic diagram of an electronic device 30 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0156] like Figure 3 As shown, the electronic device 30 includes at least one processor 31 and a memory, such as a read-only memory (ROM) 32 or a random access memory (RAM) 33, communicatively connected to the at least one processor 31. The memory stores computer programs executable by the at least one processor. The processor 31 can perform various appropriate actions and processes based on the computer program stored in the ROM 32 or loaded from storage unit 38 into the RAM 33. The RAM 33 can also store various programs and data required for the operation of the electronic device 30. The processor 31, ROM 32, and RAM 33 are interconnected via a bus 34. An input / output (I / O) interface 35 is also connected to the bus 34.

[0157] Multiple components in electronic device 30 are connected to I / O interface 35, including: input unit 36, such as keyboard, mouse, etc.; output unit 37, such as various types of monitors, speakers, etc.; storage unit 38, such as disk, optical disk, etc.; and communication unit 39, such as network card, modem, wireless transceiver, etc. Communication unit 39 allows electronic device 30 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0158] Processor 31 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 31 include, but are not limited to, central processing unit (CPU), graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 31 performs the various methods and processes described above, such as the photovoltaic branch power generation loss decoupling and health diagnosis method: Collect current operating data and meteorological environmental data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and the meteorological environmental data; The loss time series is determined based on the ideal power generation reference power, and the loss is quantified and decomposed based on the loss time series to obtain the loss decomposition component results; The fault type identification result and health status assessment conclusion are determined based on the loss decomposition component results.

[0159] In some embodiments, the photovoltaic branch power generation loss decoupling and health diagnosis method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 38. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 30 via ROM 32 and / or communication unit 39. When the computer program is loaded into RAM 33 and executed by processor 31, one or more steps of the photovoltaic branch power generation loss decoupling and health diagnosis method described above can be performed. Alternatively, in other embodiments, processor 31 can be configured to perform the photovoltaic branch power generation loss decoupling and health diagnosis method by any other suitable means (e.g., by means of firmware).

[0160] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0161] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0162] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0163] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0164] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0165] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0166] In one embodiment, the present invention further includes a computer program product, which includes a computer program that, when executed by a processor, implements the photovoltaic branch power generation loss decoupling and health diagnosis method of any embodiment of the present invention.

[0167] In implementing the computer program product, computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0168] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0169] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for decoupling photovoltaic branch power generation losses and diagnosing health issues, characterized in that, include: Collect current operating data and meteorological environmental data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and the meteorological environmental data; The loss time series is determined based on the ideal power generation reference power, and the loss is quantified and decomposed based on the loss time series to obtain the loss decomposition component results; The fault type identification result and health status assessment conclusion are determined based on the loss decomposition component results.

2. The method according to claim 1, characterized in that, Collect current operating data and meteorological environmental data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and the meteorological environmental data, including: Collect the current operating data of the photovoltaic branch and determine the current real-time power based on the current operating data; Meteorological environmental data is collected, and preliminary theoretical power is determined based on the component installation parameters of the photovoltaic branch and the meteorological environmental data; The meteorological and environmental data are input into the deviation compensation model to obtain the predicted deviation value; The ideal power generation reference power is determined based on the preliminary theoretical power and the predicted deviation value.

3. The method according to claim 2, characterized in that, The loss time series is determined based on the ideal power generation baseline, and loss quantification decomposition is performed based on the loss time series to obtain the loss decomposition component results, including: The loss time series is determined based on the current real-time power and the ideal power generation reference power. Sliding window feature extraction is performed on the loss time series to obtain a set of feature vectors; The feature vector is input into a physically constrained linear decomposer for loss quantization decomposition to obtain the loss decomposition component results.

4. The method according to claim 1, characterized in that, Based on the results of the loss decomposition components, the fault type identification results and health status assessment conclusions are determined, including: Based on the target classifier, fault diagnosis labels are determined according to the loss decomposition component results and the feature vector set, and fault type identification results are determined according to the fault diagnosis labels. The health status index of the photovoltaic branch is determined based on the loss decomposition component results, and the health status assessment conclusion is determined based on the health status index.

5. The method according to claim 2, characterized in that, The training process of the deviation compensation model is as follows: Collect historical operating data of the photovoltaic branch and determine the historical real-time power based on the historical operating data; Historical meteorological data is collected, and the historical theoretical power is determined based on the component installation parameters of the photovoltaic branch and the historical meteorological data; Based on the historical meteorological data, the historical real-time power, and the historical theoretical power, a regression model is trained to obtain a deviation compensation model.

6. The method according to claim 1, characterized in that, After determining the fault type identification result and health status assessment conclusion based on the loss decomposition component results, the process also includes: Based on the fault type identification results and the health status assessment conclusions, maintenance work order suggestions are generated.

7. A device for decoupling and diagnosing power generation losses in a photovoltaic branch circuit, characterized in that, include: The data acquisition and determination module is used to acquire the current operating data and meteorological environment data of the photovoltaic branch, and determine the ideal power generation reference power based on the current operating data and the meteorological environment data; The determination and decomposition module is used to determine the loss time series based on the ideal power generation reference power, and to perform loss quantification decomposition based on the loss time series to obtain the loss decomposition component results; The result determination module is used to determine the fault type identification result and the health status assessment conclusion based on the loss decomposition component result.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the photovoltaic branch power generation loss decoupling and health diagnosis method according to any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a processor, implement the photovoltaic branch power generation loss decoupling and health diagnosis method as described in any one of claims 1-6.

10. A computer program product, characterized in that, The method includes a computer program that, when executed by a processor, implements the photovoltaic branch power generation loss decoupling and health diagnosis method according to any one of claims 1-6.

Images