A method for constructing an energy efficiency prediction model of a photovoltaic power station considering coupling effects of dust accumulation and temperature

Abstract

The application discloses a kind of photovoltaic power station energy efficiency prediction model construction methods considering dust accumulation and temperature coupling effect, it is related to photovoltaic power generation technical field, the method includes: coupling factor of dust accumulation and temperature is constructed;Calculate dust accumulation comprehensive attenuation coefficient, determine dust accumulation spatial non-uniform shading correction factor, utilize dust accumulation comprehensive attenuation coefficient and dust accumulation spatial non-uniform shading correction factor to double correction is carried out to the original irradiance intensity obtained in advance, obtain irradiance intensity;Coupling factor, irradiance intensity and wind speed are regarded as feature, generate initial working condition boundary, and in combination with system energy efficiency ratio mutation interval and coupling factor threshold value, initial working condition boundary is calibrated in coordination, and based on calibration result, multiple operating conditions are divided out;Respectively construct multiple nonlinear regression model with system energy efficiency ratio as prediction target.The application can more accurately simulate the actual operation of photovoltaic power station, to provide more accurate energy efficiency prediction result.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation technology, and more specifically, to a method for constructing a photovoltaic power plant energy efficiency prediction model that considers the coupling effect of dust accumulation and temperature. Background Technology

[0002] As the installed capacity of large-scale photovoltaic power plants continues to climb, energy efficiency directly determines the return on investment and the stability of energy supply. Energy efficiency prediction has become one of the core technologies for intelligent operation and maintenance of photovoltaic power plants. From a physical perspective, the photoelectric conversion efficiency of photovoltaic modules is significantly affected by environmental factors. Among them, the bidirectional coupling effect of dust accumulation and temperature is a key bottleneck restricting the accuracy of energy efficiency prediction: dust accumulation not only directly reduces light absorption efficiency by blocking solar radiation, but the loose covering layer it forms also changes the thermal conductivity of the module surface, leading to an increase in module operating temperature. The increase in module temperature further reduces photoelectric conversion efficiency (the typical temperature coefficient of crystalline silicon modules is about -0.4% / ℃ to -0.5% / ℃), and exacerbates the adsorption and solidification of dust particles, forming a vicious cycle of dust deposition - increased thermal resistance - increased temperature - decreased energy efficiency.

[0003] Current research on energy efficiency prediction in the industry largely focuses on quantifying the impact of single environmental factors, with insufficient exploration of the coupling mechanism between dust accumulation and temperature. On the one hand, existing models generally fail to distinguish the differentiated impact of dust accumulation types (such as dry dust, wet dust, and sticky dust) on heat transfer efficiency, nor do they consider the dynamic effect of temperature changes on dust adhesion characteristics, resulting in significant deviations in the quantification of coupling effects. On the other hand, the energy efficiency degradation of photovoltaic modules also interacts with parameters such as solar irradiance and wind speed. Traditional models often use linear superposition to handle the influence of multiple factors, which cannot adapt to complex and variable actual operating conditions. Against this backdrop, existing energy efficiency prediction technologies have created a contradiction between prediction accuracy and ease of engineering use: traditional simplified models, due to neglecting the coupling mechanism and the interaction of multiple factors, generally have prediction errors exceeding 12%, making it difficult to meet the needs of precise operation and maintenance; while accurate prediction models relying on complex machine learning algorithms such as deep learning and ensemble learning require high levels of algorithm expertise and computing resources, making it difficult for non-AI professionals to learn, reproduce, and debug them, resulting in high barriers to engineering implementation and hindering the large-scale promotion and application of the technology.

[0004] There are currently no effective solutions to the problems in the relevant technologies. Summary of the Invention

[0005] To address the problems in related technologies, this invention proposes a method for constructing a photovoltaic power plant energy efficiency prediction model that considers the coupling effect of dust accumulation and temperature, in order to overcome the aforementioned technical problems existing in the current related technologies.

[0006] Therefore, the specific technical solution adopted by the present invention is as follows:

[0007] A method for constructing a photovoltaic power plant energy efficiency prediction model that considers the coupling effect of dust accumulation and temperature, the method comprising:

[0008] S1. Obtain the percentage of dry dust accumulation area, wet dust accumulation area, and sticky dust accumulation area on the surface of the photovoltaic module, as well as the real-time temperature of the photovoltaic module surface. Combine this with the normalized thermal resistance coefficient corresponding to different dust accumulation types to construct a coupling factor between dust accumulation and temperature.

[0009] S2. Calculate the comprehensive attenuation coefficient of dust accumulation based on the coupling factor, and divide the light-receiving surface of the photovoltaic module into several detection zones. Determine the non-uniform shading correction factor of the dust accumulation space based on the proportion of the total dust accumulation area of ​​each detection zone. Use the comprehensive attenuation coefficient of dust accumulation and the non-uniform shading correction factor of the dust accumulation space to double correct the pre-obtained original irradiance intensity to obtain the irradiance intensity.

[0010] S3. Using coupling factor, irradiance and wind speed as features, generate initial operating condition boundaries, and combine the system energy efficiency ratio mutation range and coupling factor threshold to perform collaborative calibration of the initial operating condition boundaries. Based on the calibration results, classify multiple operating conditions.

[0011] S4. Based on each type of operating condition, construct a multivariate nonlinear regression model with the system energy efficiency ratio as the prediction objective to achieve energy efficiency prediction of the photovoltaic power plant's operating status.

[0012] Furthermore, the expression for the coupling factor is:

[0013] ;

[0014] In the formula, This is the coupling factor between dust accumulation and temperature; Correction factor for component type; This represents the percentage of dry ash accumulation area. The normalized thermal resistance coefficient of dry ash accumulation; The percentage of wet dust accumulation area; The normalized thermal resistance coefficient of wet ash accumulation; This represents the percentage of area with sticky ash accumulation. The normalized thermal resistance coefficient of viscous ash; This refers to the temperature sensitivity coefficient. Real-time surface temperature of photovoltaic modules; This is the standard reference temperature.

[0015] Furthermore, based on the coupling factor, the comprehensive attenuation coefficient of dust accumulation is calculated, and the light-receiving surface of the photovoltaic module is divided into several detection zones. The non-uniform shading correction factor for the dust accumulation space is determined based on the proportion of the total dust accumulation area in each detection zone. The pre-obtained original irradiance intensity is then double-corrected using both the comprehensive attenuation coefficient of dust accumulation and the non-uniform shading correction factor for the dust accumulation space, resulting in the irradiance intensity including:

[0016] S21. Based on the coupling factor, the light transmittance attenuation coefficient of each type of ash accumulation and the preset irradiance attenuation sensitivity coefficient are combined to calculate the comprehensive attenuation coefficient of ash accumulation.

[0017] S22. Divide the light-receiving surface of the photovoltaic module into several detection zones, calculate the proportion of total dust accumulation area in each detection zone, and determine the non-uniform shading correction factor of the dust accumulation space based on the difference between the maximum and minimum values ​​of the proportion of total dust accumulation area in each detection zone.

[0018] S23. Obtain the original solar irradiance intensity and use the comprehensive attenuation coefficient of ash accumulation and the non-uniform shading correction factor of ash accumulation space to double correct the original irradiance intensity to obtain the irradiance intensity.

[0019] Furthermore, the expression for the overall ash accumulation attenuation coefficient is as follows:

[0020] ;

[0021] In the formula, The overall attenuation coefficient of ash accumulation; This is the irradiation attenuation sensitivity coefficient; This represents the percentage of dry ash accumulation area. The light transmittance attenuation coefficient for dry dust accumulation; The percentage of wet dust accumulation area; The light transmittance attenuation coefficient for wet dust accumulation; This represents the percentage of area with sticky ash accumulation. The light transmittance attenuation coefficient for sticky dust accumulation; This is the coupling factor between dust accumulation and temperature; This is the upper limit of the coupling factor.

[0022] Furthermore, coupling factor, irradiance, and wind speed are used as features to generate initial operating condition boundaries. These boundaries are then collaboratively calibrated using the system energy efficiency ratio abrupt change range and coupling factor threshold. Based on the calibration results, multiple operating conditions are identified, including:

[0023] S31. Obtain real-time wind speed, combine coupling factor and irradiance intensity to construct a feature set, and standardize the feature set to generate a standardized feature set.

[0024] S32. Based on the standardized feature set, unsupervised clustering is performed using the K-means clustering algorithm, and the optimal number of clusters is determined by the principle of maximizing the silhouette coefficient, so as to obtain the initial working condition center and the initial working condition boundary in the standardized space.

[0025] S33. Inversely map the initial operating condition center and the initial operating condition boundary to the original physical dimensions, and combine the acquired historical operating data to perform binning statistics according to the coupling factor dimension and the irradiance intensity dimension, and calculate the average change rate of the system energy efficiency ratio between adjacent bins.

[0026] S34. Compare the average change rate of the system energy efficiency ratio with a preset threshold. Based on the comparison results, determine the bin boundaries where the average change rate of the system energy efficiency ratio exceeds the preset threshold as the sudden change range of the system energy efficiency ratio.

[0027] S35. The real-time wind speed is processed using a dynamic wind speed compensation mechanism to obtain the coupling factor threshold. Combined with the sudden change range of the system energy efficiency ratio, the initial operating condition boundary under the original dimensions is calibrated collaboratively to classify multiple operating conditions.

[0028] Furthermore, a dynamic wind speed compensation mechanism is used to process real-time wind speed to obtain a coupling factor threshold. Combined with the system energy efficiency ratio abrupt change range, the initial operating condition boundary under the original dimensions is collaboratively calibrated, resulting in the classification of multiple operating conditions, including:

[0029] S351. Using real-time wind speed as input, calculate the basic coupling factor threshold according to the preset wind speed interval segmentation rules.

[0030] S352. Obtain the continuous wind speed sequence within the preset time window, calculate the cumulative duration of the first wind speed interval and the second wind speed interval respectively, perform time-series correction on the basic coupling factor threshold based on the cumulative duration, and output the coupling factor threshold.

[0031] S353. The initial operating condition boundary that is inversely mapped to the original dimension is used as the basic anchor point, and the sudden change interval of the system energy efficiency ratio is used as the physical correction anchor point. Based on the coupling factor threshold, the initial operating condition boundaries of the coupling factor dimension and the irradiance intensity dimension are calibrated in a coordinated manner to obtain the coupling factor division threshold and the irradiance intensity division threshold.

[0032] S354. Based on the coupling factor division threshold and the irradiance division threshold, cross-division is performed according to the preset binary combination logic to obtain multiple operating condition intervals.

[0033] Furthermore, using the initial operating condition boundary, which is inversely mapped to the original dimensions, as the basic anchor point, and the abrupt change range of the system energy efficiency ratio as the physical correction anchor point, the initial operating condition boundaries of the coupling factor dimension and the irradiance intensity dimension are co-calibrated according to the coupling factor threshold, respectively, to obtain the coupling factor partitioning threshold and the irradiance intensity partitioning threshold, including:

[0034] S3531. Decouple the initial working condition boundary from the inverse mapping to the original dimension by dimension, extract the initial boundary values ​​of the coupling factor dimension and the initial boundary values ​​of the irradiance intensity dimension respectively, and construct the basic anchor point.

[0035] S3532. Based on the system energy efficiency ratio mutation interval, extract the mutation critical value set in the coupling factor dimension and the mutation critical value set in the irradiance dimension respectively, and construct physical correction anchor points.

[0036] S3533. Using the coupling factor threshold as a dynamic reference benchmark, and combining the basic anchor point and the physical correction anchor point, the initial boundary values ​​of the coupling factor dimension and the irradiance dimension are calibrated according to the dual-dimensional collaborative calibration rule, so as to generate the initial coupling factor division threshold and the initial irradiance division threshold.

[0037] S3534. Using preset physical boundary constraints, constrain the initial coupling factor division threshold and the initial irradiance division threshold to obtain the coupling factor division threshold and the irradiance division threshold.

[0038] Furthermore, using the coupling factor threshold as a dynamic reference benchmark, and combining the basic anchor point and the physical correction anchor point, the initial boundary values ​​of the coupling factor dimension and the irradiance dimension are calibrated according to the dual-dimensional collaborative calibration rule, respectively, to generate the initial coupling factor division threshold and the initial irradiance division threshold, including:

[0039] The mean of the set of abrupt change critical values ​​in the coupling factor dimension is extracted from the physical correction anchor points as the coupling change feature value. If a coupling change feature value exists in the physical correction anchor points, the absolute difference between the coupling change feature value and the coupling factor threshold is calculated, and the absolute difference is compared with half of the historical bin width. When the absolute difference exceeds half of the historical bin width, the coupling factor threshold is used as the initial coupling factor partitioning threshold. When the absolute difference does not exceed half of the historical bin width, the weighted average between the coupling change feature value and the coupling factor threshold is calculated as the initial coupling factor partitioning threshold. If no coupling change feature value exists in the physical correction anchor points, the coupling factor threshold is used as the initial coupling factor partitioning threshold.

[0040] The mean of the set of critical values ​​for abrupt changes in irradiance intensity dimension is extracted from the physical correction anchor points as the irradiance abrupt change feature value. If there is an irradiance abrupt change feature value in the physical correction anchor points, the relative deviation between the irradiance abrupt change feature value and the initial irradiance intensity threshold in the base anchor points is calculated, and the relative deviation is compared with a preset ratio. When the relative deviation exceeds the preset ratio, the initial irradiance intensity threshold is used as the initial irradiance intensity division threshold. When the relative deviation does not exceed the preset ratio, the weighted average between the irradiance abrupt change feature value and the initial irradiance intensity threshold is calculated as the initial irradiance intensity division threshold. If there is no irradiance abrupt change feature value in the physical correction anchor points, the initial irradiance intensity threshold is used as the initial irradiance intensity division threshold.

[0041] Furthermore, based on each type of operating condition, a multivariate nonlinear regression model is constructed with the system energy efficiency ratio as the prediction objective to achieve energy efficiency prediction of the photovoltaic power plant's operating status, including:

[0042] S41. Obtain historical samples that have undergone outlier removal and missing value imputation, and construct training datasets according to the operating condition categories respectively;

[0043] S42. Based on each type of operating condition, with the system energy efficiency ratio as the prediction target, construct an initial multivariate nonlinear regression model, and train the initial multivariate nonlinear regression model using the training dataset of the corresponding operating condition to obtain the multivariate nonlinear regression model for each type of operating condition.

[0044] S43. Calculate the coupling factor and irradiance of the sample to be predicted, and combine the coupling factor division threshold and the irradiance division threshold to determine the boundary buffer where the sample to be predicted is located. Based on the judgment result, use the corresponding multivariate nonlinear regression model to make a prediction and output the predicted value of the system energy efficiency ratio to realize the energy efficiency prediction of the photovoltaic power station's operating status.

[0045] Furthermore, the coupling factor and irradiance of the sample to be predicted are calculated, and the boundary buffer in which the sample to be predicted is located is determined by combining the coupling factor partitioning threshold and the irradiance partitioning threshold. Based on the determination result, the corresponding multivariate nonlinear regression model is used for prediction, and the predicted value of the system energy efficiency ratio is output to achieve the energy efficiency prediction of the photovoltaic power plant's operating state, including:

[0046] By calculating the coupling factor and irradiance of the sample to be predicted, the two-dimensional difference of the sample to be predicted is obtained; wherein, the two-dimensional difference includes the absolute value of the difference between the coupling factor and the coupling factor division threshold, and the absolute value of the difference between the irradiance and the irradiance division threshold.

[0047] When the difference between the two dimensions is less than the corresponding buffer threshold, the sample to be predicted is determined to be within the boundary buffer; otherwise, the sample to be predicted is determined to be outside the boundary buffer.

[0048] If the sample to be predicted is outside the boundary buffer, the multivariate nonlinear regression model corresponding to the operating condition to which the sample to be predicted belongs is used for prediction, and the predicted value of the system energy efficiency ratio is output.

[0049] If the sample to be predicted is within the boundary buffer, then the two types of operating conditions adjacent to the sample to be predicted are identified, and the multivariate nonlinear regression models corresponding to the two types of operating conditions are called simultaneously for prediction, outputting two sets of initial system energy efficiency ratio prediction values; the Euclidean distance from the sample to be predicted to the center of the two types of operating conditions in the standardized feature space is calculated, and the weights corresponding to the two sets of initial system energy efficiency ratio prediction values ​​are calculated using the inverse distance method, and the two sets of initial system energy efficiency ratio prediction values ​​are weighted and fused through the weights to obtain the system energy efficiency ratio prediction value.

[0050] The beneficial effects of this invention are as follows:

[0051] 1. By introducing a coupling factor between ash accumulation and temperature, and combining it with a multivariate nonlinear regression model, this invention can more accurately simulate the actual operation of photovoltaic power plants, thereby providing more accurate energy efficiency prediction results and reducing prediction errors.

[0052] 2. This invention lowers the technical threshold for model building and application by designing intuitive and calculable coupling factors and classifying different working conditions, making it easy for non-AI professionals to master and use.

[0053] 3. This invention not only considers the coupling effect between ash accumulation and temperature, but also incorporates the influence of parameters such as irradiance and wind speed, as well as their interaction, thereby better adapting to the complex and ever-changing operating environment in reality and improving the model's versatility and applicability.

[0054] 4. This invention ensures the stability and reliability of the model, enabling it to maintain high prediction accuracy under different weather conditions.

[0055] 5. This invention reduces predictive abrupt changes caused by switching between adjacent operating conditions through weighted fusion of operating condition division and boundary buffer. This provides an executable technical basis for the intelligent operation and maintenance of photovoltaic power plants.

[0056] 6. This invention provides strong technical support for the intelligent operation and maintenance of photovoltaic power plants, which helps to improve the return on investment of photovoltaic projects and the stability of energy supply. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 This is a flowchart of a method for constructing a photovoltaic power plant energy efficiency prediction model that considers the coupling effect of dust accumulation and temperature according to an embodiment of the present invention;

[0059] Figure 2 This is a flowchart of the operating condition division of a photovoltaic power plant energy efficiency prediction model construction method considering the coupling effect of dust accumulation and temperature according to an embodiment of the present invention.

[0060] Figure 3 This is a schematic diagram of a photovoltaic power plant energy efficiency prediction model construction system that considers the coupling effect of dust accumulation and temperature according to an embodiment of the present invention.

[0061] Figure 4 This is a schematic diagram of the overall process of constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature according to an embodiment of the present invention.

[0062] Figure 5 This is a schematic diagram of the temperature coupling mechanism in a photovoltaic power plant energy efficiency prediction model construction method that considers the coupling effect of dust accumulation and temperature according to an embodiment of the present invention. Detailed Implementation

[0063] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0064] According to an embodiment of the present invention, a method for constructing a photovoltaic power plant energy efficiency prediction model that considers the coupling effect of dust accumulation and temperature is provided.

[0065] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 and Figure 4 As shown, according to an embodiment of the present invention, a method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature is provided. The method includes:

[0066] S1. Obtain the percentage of dry dust accumulation area, wet dust accumulation area, and sticky dust accumulation area on the surface of the photovoltaic module, as well as the real-time temperature of the photovoltaic module surface. Combine this with the normalized thermal resistance coefficient corresponding to different dust accumulation types to construct a coupling factor between dust accumulation and temperature.

[0067] In this optional embodiment, the expression for the coupling factor is:

[0068] ;

[0069] In the formula, This is the coupling factor between dust accumulation and temperature; Correction factor for component type; This represents the percentage of dry ash accumulation area. The normalized thermal resistance coefficient of dry ash accumulation; The percentage of wet dust accumulation area; The normalized thermal resistance coefficient of wet ash accumulation; This represents the percentage of area with sticky ash accumulation. The normalized thermal resistance coefficient of viscous ash; This refers to the temperature sensitivity coefficient. Real-time surface temperature of photovoltaic modules; This is the standard reference temperature.

[0070] It should be added that, based on the two-way coupling mechanism between the thermal resistance effect of dust accumulation on component temperature and the effect of temperature on dust adhesion strength, and considering the differences in the thermal characteristics of different dust accumulation types, a coupling factor is designed to quantify this coupling relationship; the coupling factor... The expression is: In the formula: Ash accumulation-temperature coupling factor (dimensionless, typical range) ); For component type correction factor (for crystalline silicon components) Thin film module ); , , The area percentages of dry, wet, and sticky ash accumulation, respectively (meeting the requirements). The uncovered area is the area without dust accumulation, which can be obtained through component surface image recognition, inspection photo partition recognition, dust accumulation sensor detection, or manual sampling and calibration. , , These are the normalized thermal resistance coefficients corresponding to the ash accumulation type ( ,in For reference thermal resistance, , , ); Temperature sensitivity coefficient ( The specific value is determined by the type of photovoltaic module, the type of dust accumulation, and the climate environment of the power station: scenarios with a high proportion of dry dust accumulation, strong interaction between the module surface material and dust molecules, and large temperature fluctuations. Take the higher value ( In scenarios with a high proportion of wet, sticky dust and minimal temperature fluctuations, Take the lower value ( ); Real-time surface temperature of photovoltaic modules ( ); Standard reference temperature (25) ).

[0071] S2. Calculate the comprehensive attenuation coefficient of dust accumulation based on the coupling factor, and divide the light-receiving surface of the photovoltaic module into several detection zones. Determine the non-uniform shading correction factor of the dust accumulation space based on the proportion of the total dust accumulation area of ​​each detection zone. Use the comprehensive attenuation coefficient of dust accumulation and the non-uniform shading correction factor of the dust accumulation space to double correct the pre-obtained original irradiance intensity to obtain the irradiance intensity.

[0072] In this optional embodiment, based on the coupling factor, the comprehensive attenuation coefficient of dust accumulation is calculated, and the light-receiving surface of the photovoltaic module is divided into several detection zones. The non-uniform shading correction factor for the dust accumulation space is determined based on the proportion of the total dust accumulation area in each detection zone. The pre-obtained original irradiance intensity is then double-corrected using both the comprehensive attenuation coefficient of dust accumulation and the non-uniform shading correction factor for the dust accumulation space, resulting in an irradiance intensity including:

[0073] S21. Based on the coupling factor, the light transmittance attenuation coefficient of each type of ash accumulation and the preset irradiance attenuation sensitivity coefficient are combined to calculate the comprehensive attenuation coefficient of ash accumulation.

[0074] In this optional embodiment, the expression for the overall ash accumulation attenuation coefficient is:

[0075] ;

[0076] In the formula, The overall attenuation coefficient of ash accumulation; This is the irradiation attenuation sensitivity coefficient; This represents the percentage of dry ash accumulation area. The light transmittance attenuation coefficient for dry dust accumulation; The percentage of wet dust accumulation area; The light transmittance attenuation coefficient for wet dust accumulation; This represents the percentage of area with sticky ash accumulation. The light transmittance attenuation coefficient for sticky dust accumulation; This is the coupling factor between dust accumulation and temperature; This is the upper limit of the coupling factor.

[0077] S22. Divide the light-receiving surface of the photovoltaic module into several detection zones, calculate the proportion of total dust accumulation area in each detection zone, and determine the non-uniform shading correction factor of the dust accumulation space based on the difference between the maximum and minimum values ​​of the proportion of total dust accumulation area in each detection zone.

[0078] S23. Obtain the original solar irradiance intensity and use the comprehensive attenuation coefficient of ash accumulation and the non-uniform shading correction factor of ash accumulation space to double correct the original irradiance intensity to obtain the irradiance intensity.

[0079] It should be added that, such as Figure 5 As shown, a comprehensive attenuation coefficient for dust accumulation is introduced. Correction for solar irradiance intensity: , In the formula: The irradiation attenuation sensitivity coefficient (take) (Determined by fitting historical data). , , Light transmittance attenuation coefficients for different dust accumulation types (dry type taken as...) Moisture Viscosity ); This is an upper limit for the coupling factor to prevent... Negative values ​​may occur. For the measured irradiance ( ), This is the corrected irradiance intensity. When... At that time, set To ensure physical plausibility. The value is a correction factor for non-uniform occlusion in the dust accumulation space. When the difference in the proportion of dust accumulation area in different detection zones of the component is greater, The smaller the value.

[0080] For scenarios where dust is evenly distributed, Take close For scenarios with obvious localized mud streaks, point-like shading, or edge accumulation, the light-receiving surface of the component is divided into... Each detection zone is calculated to determine the percentage of total dust accumulation area in each zone. and according to each The difference between the maximum and minimum values ​​is determined. The more uneven the dust distribution, The smaller the value, the better it reflects the string mismatch loss caused by local occlusion. When hour, Pick ;when hour, Pick ;when hour, Pick .

[0081] S3. Using coupling factor, irradiance and wind speed as features, generate initial operating condition boundaries, and combine the system energy efficiency ratio mutation range and coupling factor threshold to perform collaborative calibration of the initial operating condition boundaries. Based on the calibration results, classify multiple operating conditions.

[0082] In this optional embodiment, the coupling factor, irradiance, and wind speed are used as features to generate the initial operating condition boundary. The initial operating condition boundary is then collaboratively calibrated by combining the system energy efficiency ratio abrupt change range with the coupling factor threshold. Based on the calibration results, multiple operating conditions are classified, including:

[0083] S31. Obtain real-time wind speed, combine coupling factor and irradiance intensity to construct a feature set, and standardize the feature set to generate a standardized feature set.

[0084] S32. Based on the standardized feature set, unsupervised clustering is performed using the K-means clustering algorithm, and the optimal number of clusters is determined by the principle of maximizing the silhouette coefficient, so as to obtain the initial working condition center and the initial working condition boundary in the standardized space.

[0085] S33. Inversely map the initial operating condition center and the initial operating condition boundary to the original physical dimensions, and combine the acquired historical operating data to perform binning statistics according to the coupling factor dimension and the irradiance intensity dimension, and calculate the average change rate of the system energy efficiency ratio between adjacent bins.

[0086] S34. Compare the average change rate of the system energy efficiency ratio with a preset threshold. Based on the comparison results, determine the bin boundaries where the average change rate of the system energy efficiency ratio exceeds the preset threshold as the sudden change range of the system energy efficiency ratio.

[0087] S35. The real-time wind speed is processed using a dynamic wind speed compensation mechanism to obtain the coupling factor threshold. Combined with the sudden change range of the system energy efficiency ratio, the initial operating condition boundary under the original dimensions is calibrated collaboratively to classify multiple operating conditions.

[0088] In this optional embodiment, a dynamic wind speed compensation mechanism is used to process the real-time wind speed to obtain a coupling factor threshold. Combined with the system energy efficiency ratio abrupt change range, the initial operating condition boundary under the original dimensions is collaboratively calibrated, resulting in the classification of multiple operating conditions, including:

[0089] S351. Using real-time wind speed as input, calculate the basic coupling factor threshold according to the preset wind speed interval segmentation rules.

[0090] S352. Obtain the continuous wind speed sequence within the preset time window, calculate the cumulative duration of the first wind speed interval and the second wind speed interval respectively, perform time-series correction on the basic coupling factor threshold based on the cumulative duration, and output the coupling factor threshold.

[0091] S353. The initial operating condition boundary that is inversely mapped to the original dimension is used as the basic anchor point, and the sudden change interval of the system energy efficiency ratio is used as the physical correction anchor point. Based on the coupling factor threshold, the initial operating condition boundaries of the coupling factor dimension and the irradiance intensity dimension are calibrated in a coordinated manner to obtain the coupling factor division threshold and the irradiance intensity division threshold.

[0092] In this optional embodiment, the initial operating condition boundary, inversely mapped to the original dimensions, is used as the basic anchor point, and the abrupt change range of the system energy efficiency ratio is used as the physical correction anchor point. Based on the coupling factor threshold, the initial operating condition boundaries of the coupling factor dimension and the irradiance intensity dimension are co-calibrated to obtain the coupling factor partitioning threshold and the irradiance intensity partitioning threshold, including:

[0093] S3531. Decouple the initial working condition boundary from the inverse mapping to the original dimension by dimension, extract the initial boundary values ​​of the coupling factor dimension and the initial boundary values ​​of the irradiance intensity dimension respectively, and construct the basic anchor point.

[0094] S3532. Based on the system energy efficiency ratio mutation interval, extract the mutation critical value set in the coupling factor dimension and the mutation critical value set in the irradiance dimension respectively, and construct physical correction anchor points.

[0095] S3533. Using the coupling factor threshold as a dynamic reference benchmark, and combining the basic anchor point and the physical correction anchor point, the initial boundary values ​​of the coupling factor dimension and the irradiance dimension are calibrated according to the dual-dimensional collaborative calibration rule, so as to generate the initial coupling factor division threshold and the initial irradiance division threshold.

[0096] In this optional embodiment, the coupling factor threshold is used as a dynamic reference benchmark. Combined with the base anchor point and the physical correction anchor point, and following a two-dimensional collaborative calibration rule, the initial boundary values ​​for the coupling factor dimension and the irradiance dimension are calibrated respectively, generating the initial coupling factor division threshold and the initial irradiance division threshold, including:

[0097] The mean of the set of abrupt change critical values ​​in the coupling factor dimension is extracted from the physical correction anchor points as the coupling change feature value. If a coupling change feature value exists in the physical correction anchor points, the absolute difference between the coupling change feature value and the coupling factor threshold is calculated, and the absolute difference is compared with half of the historical bin width. When the absolute difference exceeds half of the historical bin width, the coupling factor threshold is used as the initial coupling factor partitioning threshold. When the absolute difference does not exceed half of the historical bin width, the weighted average between the coupling change feature value and the coupling factor threshold is calculated as the initial coupling factor partitioning threshold. If no coupling change feature value exists in the physical correction anchor points, the coupling factor threshold is used as the initial coupling factor partitioning threshold.

[0098] The mean of the set of critical values ​​for abrupt changes in irradiance intensity dimension is extracted from the physical correction anchor points as the irradiance abrupt change feature value. If there is an irradiance abrupt change feature value in the physical correction anchor points, the relative deviation between the irradiance abrupt change feature value and the initial irradiance intensity threshold in the base anchor points is calculated, and the relative deviation is compared with a preset ratio. When the relative deviation exceeds the preset ratio, the initial irradiance intensity threshold is used as the initial irradiance intensity division threshold. When the relative deviation does not exceed the preset ratio, the weighted average between the irradiance abrupt change feature value and the initial irradiance intensity threshold is calculated as the initial irradiance intensity division threshold. If there is no irradiance abrupt change feature value in the physical correction anchor points, the initial irradiance intensity threshold is used as the initial irradiance intensity division threshold.

[0099] S3534. Using preset physical boundary constraints, constrain the initial coupling factor division threshold and the initial irradiance division threshold to obtain the coupling factor division threshold and the irradiance division threshold.

[0100] S354. Based on the coupling factor division threshold and the irradiance division threshold, cross-division is performed according to the preset binary combination logic to obtain multiple operating condition intervals.

[0101] It should be added that, such as Figure 2 As shown, using -means clustering algorithm, with coupling factor Corrected irradiance Wind speed Feature parameters: For the original data... -Score standardization; determine the optimal number of clusters based on the principle of maximizing the silhouette coefficient. ; -means clustering is used to determine the historical samples in , , The initial operating condition center and initial boundary in the three-dimensional feature space; combined with the system energy efficiency ratio in historical operating data. Follow , and The changing abrupt intervals map the cluster boundaries back to their original dimensions and calibrate them for engineering interpretability. and Thresholds are used to form four types of operating conditions. The mutation interval can be determined by sorting historical samples according to... and Divide into boxes and calculate the contents of each box. Mean and compare adjacent bins The rate of change of the mean is determined.

[0102] The four operating conditions are as follows: Weak coupling - Low radiation: and Weak coupling - high radiation: and Strong coupling - low radiation: and Strong coupling - high radiation: and Wind speed compensation mechanism: When the wind speed... At that time, the adhesion of accumulated dust weakens, and the coupling factor threshold increases to The upper limit is ;when hour, ;when At that time, ash deposition intensifies. Meanwhile, in The duration of high wind speeds is recorded within a preset time window. The cumulative duration is not less than When a wind speed cleanliness maintenance state is determined, the coupling factor threshold is appropriately increased; when The cumulative duration is not less than When this occurs, it is determined to be a prolonged low wind speed state, and the coupling factor threshold is appropriately lowered.

[0103] For example, taking a certain moment as an example, the measured dry dust accumulation area ratio is 0.25, wet dust accumulation is 0.15, and viscous dust accumulation is 0.05. The real-time surface temperature of the module is 45℃, and the corresponding normalized thermal resistance coefficients are 0.40, 0.67, and 1.13, respectively. Substituting these values ​​into the formula, the coupling factor is calculated to be 1.62. Using this coupling factor and the light transmittance attenuation coefficient of each type of dust accumulation, the comprehensive dust accumulation attenuation coefficient is calculated to be 0.72. At the same time, the module is divided into 9 detection zones. The maximum value of the total dust accumulation area ratio of each zone is 0.52, the minimum value is 0.31, and the difference is 0.21. Based on this, the non-uniform shading correction factor of the dust accumulation space is determined to be 0.96, and the original irradiance intensity of 850W / m² is finally corrected to 587.5W / m². A real-time wind speed of 3.8 m / s was acquired. A three-dimensional feature set was constructed using a coupling factor of 1.62 and a corrected irradiance of 587.5 W / m². After Z-score standardization, the standardized feature vectors of this sample were approximately 0.85, 0.32, and 0.67. K-means clustering was performed based on the standardized feature set. The silhouette coefficient reached its maximum value of 0.71 when the number of clusters was 4. Four initial operating conditions were determined, and the initial operating condition centers and boundaries in the standardized space were obtained. After inverse mapping back to the original dimensions, the initial coupling factor boundary values ​​were 0.82 and 1.31, and the initial irradiance boundary value was 403 W / m². Historical 180-day operating data was acquired, and the system was divided into bins according to a coupling factor of 0.2 and an irradiance of 100 W / m². The mean change rate of the energy efficiency ratio of adjacent bins was calculated. The mean of all change rates was 3.8%, and the standard deviation was 2.1%. Based on this, a preset threshold of 5.9% was set. Upon comparison, the average change rate of the energy efficiency ratio at the boundary of the sub-compartment with a coupling factor in the range of 1.25 to 1.45 and an irradiance intensity in the range of 380 to 420 W / m² reached 7.3%, exceeding the threshold. Therefore, this sub-compartment boundary was determined as the abrupt change range of the system's energy efficiency ratio. Using a real-time wind speed of 3.8 m / s as input, the high wind speed segmentation rule was triggered, and the basic coupling factor threshold was calculated to be 1.04. The continuous wind speed sequence of the past 24 hours was obtained, and the cumulative duration of wind speeds not lower than 3 m / s was 3.2 hours, triggering the upward adjustment condition. The basic threshold was increased by 0.08, and the output coupling factor threshold was 1.12. Using the initial coupling factor cutoff values ​​of 0.82 and 1.31 as basic anchor points, and the coupling factor mutation critical value sets of 1.25, 1.38, and 1.45 and the irradiance mutation critical value sets of 385, 405, and 418 in the energy efficiency ratio mutation interval as physical correction anchor points, the mean of the coupling mutation critical value set of 1.36 and the mean of the irradiance mutation critical value set of 403 were extracted.A two-dimensional collaborative calibration is performed: In the coupling factor dimension, the coupling mutation feature value of 1.36 exists, and its absolute difference from the coupling factor threshold of 1.12 is 0.24. The historical binning half-width is 0.1, and the difference exceeds the half-width, so 1.12 is directly used as the initial coupling factor partitioning threshold. In the irradiance dimension, the relative deviation between the irradiance mutation feature value 403 and the initial irradiance boundary value 403 is 0%, not exceeding the preset proportion of 15%, so the weighted average value 403 is calculated as the initial irradiance partitioning threshold. After physical boundary constraints, the final output coupling factor partitioning threshold is 1.12, and the irradiance partitioning threshold is 403 W / m². The system is divided into four operating conditions based on binary combinatorial logic: weak coupling - low irradiance corresponds to a coupling factor less than 1.12 and an irradiance intensity less than 403 W / m²; weak coupling - high irradiance corresponds to a coupling factor less than 1.12 and an irradiance intensity not less than 403 W / m²; strong coupling - low irradiance corresponds to a coupling factor not less than 1.12 and an irradiance intensity less than 403 W / m²; and strong coupling - high irradiance corresponds to a coupling factor not less than 1.12 and an irradiance intensity not less than 403 W / m². This invention uses dynamic wind speed compensation to adaptively adjust the coupling factor threshold according to environmental conditions, effectively avoiding the misjudgment of mild dust accumulation as a strong coupling condition. Simultaneously, the irradiance boundary value is verified using the system energy efficiency ratio abrupt change range, ensuring precise alignment between the irradiance dimension boundary and the power plant's actual efficiency inflection point.

[0104] S4. Based on each type of operating condition, construct a multivariate nonlinear regression model with the system energy efficiency ratio as the prediction objective to achieve energy efficiency prediction of the photovoltaic power plant's operating status.

[0105] In this optional embodiment, based on each type of operating condition, a multivariate nonlinear regression model is constructed with the system energy efficiency ratio as the prediction objective to achieve energy efficiency prediction of the photovoltaic power plant's operating status, including:

[0106] S41. Obtain historical samples that have undergone outlier removal and missing value imputation, and construct training datasets according to the operating condition categories respectively;

[0107] S42. Based on each type of operating condition, with the system energy efficiency ratio as the prediction target, construct an initial multivariate nonlinear regression model, and use the training dataset of the corresponding operating condition to train the initial multivariate nonlinear regression model to obtain the multivariate nonlinear regression model for each type of operating condition.

[0108] S43. Calculate the coupling factor and irradiance of the sample to be predicted, and combine the coupling factor division threshold and the irradiance division threshold to determine the boundary buffer where the sample to be predicted is located. Based on the judgment result, use the corresponding multivariate nonlinear regression model to make a prediction and output the predicted value of the system energy efficiency ratio to realize the energy efficiency prediction of the photovoltaic power station's operating status.

[0109] In this optional embodiment, the coupling factor and irradiance of the sample to be predicted are calculated, and the boundary buffer where the sample to be predicted is located is determined by combining the coupling factor partition threshold and the irradiance partition threshold. Based on the determination result, the corresponding multivariate nonlinear regression model is used for prediction, and the predicted value of the system energy efficiency ratio is output to achieve the energy efficiency prediction of the photovoltaic power station's operating status.

[0110] By calculating the coupling factor and irradiance of the sample to be predicted, the two-dimensional difference of the sample to be predicted is obtained; wherein, the two-dimensional difference includes the absolute value of the difference between the coupling factor and the coupling factor division threshold, and the absolute value of the difference between the irradiance and the irradiance division threshold.

[0111] When both differences between the two dimensions are less than the corresponding buffer threshold, the sample to be predicted is determined to be within the boundary buffer; otherwise, the sample to be predicted is determined to be outside the boundary buffer.

[0112] If the sample to be predicted is outside the boundary buffer, the multivariate nonlinear regression model corresponding to the operating condition to which the sample to be predicted belongs is used for prediction, and the predicted value of the system energy efficiency ratio is output.

[0113] If the sample to be predicted is within the boundary buffer, then the two types of operating conditions adjacent to the sample to be predicted are identified, and the multivariate nonlinear regression models corresponding to the two types of operating conditions are called simultaneously for prediction, outputting two sets of initial system energy efficiency ratio prediction values; the Euclidean distance from the sample to be predicted to the center of the two types of operating conditions in the standardized feature space is calculated, and the weights corresponding to the two sets of initial system energy efficiency ratio prediction values ​​are calculated using the inverse distance method, and the two sets of initial system energy efficiency ratio prediction values ​​are weighted and fused through the weights to obtain the system energy efficiency ratio prediction value.

[0114] It should be added that a multivariate nonlinear regression model is independently constructed for each type of operating condition, using the system energy efficiency ratio as the basis. For the prediction target: In the formula: the boundary buffer weighted fusion is used to avoid abrupt changes in the predicted value when switching between adjacent working conditions. The weight can be determined according to the distance between the sample to be predicted and the center of the adjacent working condition. The closer the working condition is, the greater the weight of the model. For model parameters, through the Regularized least squares fitting, regularization coefficient pass Cross-validation confirms this. This is the random error term; and These are the key interaction terms, representing the irradiation-ash accumulation coupling effect and the temperature-heat dissipation synergistic effect, respectively.

[0115] The model building process includes: using The criteria include removing outliers to ensure data quality; and using linear interpolation to handle missing values ​​in time series data, where the missing value rate exceeds a certain threshold. The data segments were removed; the historical samples were first classified into the corresponding working conditions according to the working conditions, and then used as input variables. and output tags ;according to The training and validation sets are divided proportionally to maintain temporal continuity; [the following is introduced] Regularization prevents overfitting; the regularization strength is optimized through cross-validation. During real-time prediction, it is first calculated... and It also determines the operating condition and calls the corresponding operating condition model to output. When the sample is in or boundary When within the boundary buffer, simultaneously call adjacent working condition models and fuse them by distance weighting; and For accuracy acceptance standards, feature engineering is performed again for operating conditions that do not meet the standards.

[0116] For example, 4320 operational data entries were obtained from 180 historical days. After removing 17 outliers and performing linear interpolation to complete the data for periods with a missing rate below 15%, 4260 valid samples were retained. These samples were then categorized into four groups based on their operating conditions: weakly coupled-low irradiance, weakly coupled-high irradiance, strongly coupled-low irradiance, and strongly coupled-high irradiance, containing 1210, 1085, 895, and 1070 samples respectively. Each dataset was further divided based on the system energy efficiency ratio. For the label, the input features include the corrected irradiance. Component surface temperature Coupling factor Wind speed and interactive items and Build a model for each type of working condition. ; Using belt Regularized least squares fitting parameters, regularization coefficients Five-fold cross-validation was used to determine (under strong coupling-high irradiation conditions) The validation set mean square error is minimized when the value is 0.15. After training, the validation set determination coefficients for the four types of working conditions are... The mean absolute percentage errors were 0.92, 0.94, 0.91, and 0.93, respectively. The percentages were 4.2%, 3.8%, 4.5%, and 4.1%, respectively, all meeting the accuracy acceptance criteria. In actual prediction, at a given time, the measured percentages of dry dust accumulation were 0.20, wet dust accumulation 0.12, and viscous dust accumulation 0.03, with a module temperature of 42℃, an initial irradiance of 820W / m², and a range of dust accumulation area percentages in each zone of 0.18. The calculated coupling factor was 1.14, the comprehensive dust accumulation attenuation coefficient was 0.76, and the non-uniform shading factor was 0.96. The corrected irradiance intensity... Combining the coupling factor threshold of 1.12 and the irradiance threshold of 403 W / m², the absolute value of the coupling factor difference is obtained. Absolute value of the difference in irradiance intensity The latter is much greater than 20W / m², which does not meet the condition that the difference between the two dimensions is simultaneously less than the buffer threshold. Therefore, the sample is determined to be outside the buffer, and the strongly coupled-high-irradiation working condition model is directly called and substituted. , , , , , Output The predicted value is 0.82. If the irradiance of another sample is 410 W / m² and other conditions remain unchanged, then... Difference and Difference Simultaneously satisfying the buffer condition, it is determined that it is located within the boundary buffer of strong coupling-low irradiance and strong coupling-high irradiance. The two types of working condition models are called simultaneously to obtain the initial predicted values ​​respectively. , The Euclidean distance from the sample to the center of the strong-low load condition in the standardized space is calculated to be 0.35, and the distance to the center of the strong-high load condition is calculated to be 0.65. Normalization using the inverse distance method yields weights of 0.65 and 0.35, respectively. The final weighted fusion is then performed. .

[0117] Example 1:

[0118] 1. Basic parameters of the power station: This example selects a large-scale ground-mounted crystalline silicon photovoltaic power station in Northwest China, with an installed capacity of 50MW. The photovoltaic modules are monocrystalline silicon modules, and the module type correction factor is... The region has a dry climate, with predominantly dry ash, a small amount of moist ash, and no sticky ash. The temperature fluctuates greatly between day and night, which is consistent with... Therefore, we choose the scenario where a higher value is desired. .

[0119] 2. Data Acquisition: Continuously collect operational data of the power station from April to mid-July of a certain year, with a time resolution of 15 minutes. Acquisition parameters include: percentage of ash accumulation area, dry ash accumulation... Moist ash Sticky dust Real-time surface temperature of components The temperature range is 15~65℃; measured solar irradiance. ,scope Ambient wind speed ,scope Historical energy efficiency ratio , used as labels for model training.

[0120] 3. Coupling factor calculation:

[0121] Set parameters, temperature sensitivity coefficient Reference temperature Normalized thermal resistance coefficient , , ;

[0122] Substitute into the formula to calculate the coupling factor : Calculated The range is .

[0123] 4. Irradiation intensity correction:

[0124] Setting parameters: Irradiation attenuation sensitivity coefficient Light transmittance attenuation coefficient , , Measured solar irradiance :scope .

[0125] Calculate the attenuation coefficient : , can be obtained The range of values ​​is ;

[0126] Corrected irradiance After correction The range of values ​​for is In this embodiment, the difference in dust accumulation between the detection zones is relatively small. Let's use 1.00 for example calculations.

[0127] 5. Operating condition classification:

[0128] (1) Data preprocessing (Z-score standardization calculation): Selecting coupling factors Corrected irradiance Ambient wind speed Three sets of feature parameters were used to calculate the mean based on 10,080 sampled data points. and standard deviation The results are shown in Table 1.

[0129] Table 1. Example Table of Mean and Standard Deviation

[0130] Z-score standardization is performed to eliminate dimensional differences. The standardization formula is: In the formula These are the original parameter values. The mean of the parameters, The standard deviation of the parameter is given.

[0131] All 10080 sampled data points were standardized to eliminate dimensional differences. Calculation example: [The following is a sample data set] , , After standardization , , .

[0132] (2) Determine the optimal number of clusters: Use the K-means clustering algorithm to test the number of clusters. Calculate each The average profile coefficient corresponding to the value is calculated using the following formula: In the formula, For the sample The average distance to other samples in the same cluster, For the sample The average distance to all samples in the nearest heterogeneous cluster; the average silhouette coefficient results for each cluster number are shown in Table 2.

[0133] Table 2. Example table of average silhouette coefficient results for each cluster number.

[0134] Based on the principle of maximizing the silhouette coefficient, the optimal number of clusters was finally determined to be 4.

[0135] (3) Wind speed compensation mechanism corrects coupling factor threshold:

[0136] The wind speed compensation rule designed in this invention is as follows: When hour, , upper limit 1.5; when hour, ;when hour, Within 24 hours When the cumulative duration is not less than 2 hours, the adjustment is increased. 0.05~0.10; within 24 hours When the cumulative duration is not less than 6 hours, the adjustment will be lowered. 0.05~0.10.

[0137] Calculation example: Take (Maximum value in this embodiment): (<1.5, take the calculated value); take : ;Pick (Minimum value in this embodiment): .

[0138] (4) Working condition classification:

[0139] Based on the compensation , combined and The interval was divided into four operating conditions, and 10,080 standardized data points were traversed, based on real-time... corresponding The operating conditions were determined, and the number and percentage of data for each operating condition were counted, as shown in Table 3.

[0140] Table 3. Example table of the number and percentage of data for each operating condition.

[0141] 6. Construction and validation of prediction models for different working conditions:

[0142] (1) with The guidelines handle outliers as follows:

[0143] Taking the weak coupling-high irradiation condition as an example, this condition has a total of 3830 lines. The data, calculated as follows: mean ; Standard deviation Outlier Detection Range The calculated interval is [0.715, 0.925]. Under this operating condition... There are a total of 42 data entries. There were 0 data entries in total. After removing 42 outliers, 3788 valid data entries remained. Similarly, outlier removal was performed for the other three types of operating conditions, resulting in the removal of 212 outliers from the entire dataset, representing 2.1% of the total data.

[0144] (2) Missing value handling (linear interpolation calculation):

[0145] In this embodiment, the data missing rate is 3.5%, with a total of 353 missing data points. Linear interpolation is performed based on the time series data, using the following formula: ; Calculation example: a certain time period , ; , ; Data is missing. Interpolation calculation: .

[0146] (3) Dataset partitioning (calculated based on temporal continuity): A total of 9868 valid data points were partitioned into training and validation sets in a 7:3 ratio. The data was arranged chronologically, with the first 70% of the data used as the training set and the last 30% as the validation set to ensure temporal continuity. Partition results: Training set: Data from the first 70% of the time period, totaling... Items; Validation set: Data from the last 30% of the time period, totaling [number missing] items. strip.

[0147] (4) Model parameter fitting (calculated using L2 regularized least squares method):

[0148] Formula for multivariate nonlinear regression model under different working conditions: .

[0149] The objective function for L2 regularization is: In the formula, These are the model's predicted values. The regularization coefficient is used; taking the weakly coupled high-irradiation condition as an example, the optimal coefficient is determined through 5-fold cross-validation. The 3788 data points for this working condition were divided into 5 parts, each with 758 data points; one part was selected as the validation set, and the other four parts as the training set. A model was fitted, and the validation set MAPE was calculated; then testing was conducted. Record each The corresponding average MAPE is shown in Table 4.

[0150] Table 4. Example of Average MAPE

[0151] Select the one with the smallest average MAPE The optimal regularization coefficients are used to fit the model parameters for this working condition: , , , , , , Similarly, parameter fitting is performed for the other three types of working conditions.

[0152] (5) Model validation results (MAPE and calculate):

[0153] MAPE calculation formula: ; Calculation formula: Taking the weakly coupled, high-irradiation condition as an example, substituting the 758 data points from the validation set, we calculated the following: , The final verification results for the four types of working conditions are shown in Table 5.

[0154] Table 5. Example of final verification results for four types of working conditions (one of the tables)

[0155] 7. Comparative Analysis of Results:

[0156] The model of this invention is compared with the traditional linear prediction model, and the results are as follows:

[0157] High prediction accuracy: Based on the backcalculation results of historical operation and maintenance data of the same period using traditional models and the publicly available error benchmark range in the photovoltaic industry, the overall MAPE of traditional linear models is 10-15%, while the overall MAPE of the model of this invention is reduced to 3.4%, improving prediction accuracy by 70%. Strong engineering usability: The model of this invention does not require professional algorithm computing power support. Operation and maintenance personnel can complete the calculation through Excel or the built-in formulas of the power plant monitoring system, shortening the model debugging time from 48 hours for traditional complex models to 6 hours. Robustness: In extreme sandstorm weather (… (approaching 1.2) and high temperatures ( Under these conditions, the prediction error of the model of this invention is stable within 5%, while the error of traditional models may be as high as 20% or more.

[0158] Example 2:

[0159] 1. Basic parameters of the power station: This example selects a rooftop thin-film photovoltaic power station in southern China with an installed capacity of 10MW. The photovoltaic modules are cadmium telluride thin-film modules. Module type correction factor. (Compliant with thin-film components) (The range of values); This region is rainy and humid, with wet ash being the main type, sticky ash accounting for a significant proportion, and dry ash being less common. Temperature fluctuations are also relatively mild, which is consistent with... Therefore, we choose the scenario where the lower value is selected. .

[0160] 2. Data Acquisition: Operational data of the power station was continuously collected from June to early September of a certain year during the rainy season, and 8960 sampling data points were generated after filtering. The time resolution was 15 minutes. Collected parameters included: percentage of ash accumulation area, dry ash accumulation... Moist ash Sticky dust (satisfy ); Real-time surface temperature of components :scope Measured solar irradiance :scope Ambient wind speed :scope Historical energy efficiency ratio : Used as labels for model training.

[0161] 3. Coupling factor calculation:

[0162] Setting parameters: Temperature sensitivity coefficient Reference temperature Normalized thermal resistance coefficient: , ; , ; , ;

[0163] Substitute into the formula to calculate the coupling factor : ; calculated The range is .

[0164] 4. Irradiation intensity correction:

[0165] Setting parameters: Irradiation attenuation sensitivity coefficient Light transmittance attenuation coefficient: , , (In humid southern regions, wet dust has a more compact structure due to its adsorption of moisture, strong adhesion, and uniform coverage, resulting in a significantly stronger shading effect against solar radiation.) Measured solar irradiance :scope ;

[0166] Calculate the attenuation coefficient : achievable The range of values ​​is Corrected irradiance After correction The range of values ​​for is In this embodiment Use 1.00 for example calculations; when there is significant local accumulation of sticky ash, a correction of 0.90~1.00 can be made based on the difference between the detection zones.

[0167] 5. Operating condition classification:

[0168] (1) Data preprocessing (Z-score standardization calculation): Selecting coupling factors Corrected irradiance Ambient wind speed Three sets of feature parameters were used to calculate the mean based on 8960 sampled data points. and standard deviation The results are shown in Table 6.

[0169] Table 6. Example Table of Mean and Standard Deviation (Part Two)

[0170] Z-score standardization is performed to eliminate dimensional differences. The standardization formula is: In the formula These are the original parameter values. The mean of the parameters, The standard deviation of the parameter is given.

[0171] All 8960 sampled data points were standardized to eliminate dimensional differences. Calculation example: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] , , After standardization , , .

[0172] (2) Determine the optimal number of clusters: Use the K-means clustering algorithm to test the number of clusters. Calculate each The average profile coefficient corresponding to the value is calculated using the following formula: In the formula, For the sample The average distance to other samples in the same cluster, For the sample The average distance to all samples in the nearest heterogeneous cluster; the average silhouette coefficient results for each cluster number are shown in Table 7.

[0173] Table 7. Example Table 2 of the average silhouette coefficient results for each cluster number.

[0174] Based on the principle of maximizing the silhouette coefficient, the optimal number of clusters was finally determined to be 4.

[0175] (3) Wind speed compensation mechanism corrects coupling factor threshold:

[0176] The wind speed compensation rule designed in this invention is as follows: When hour, , upper limit 1.5; when hour, ;when hour, Within 24 hours When the cumulative duration is not less than 2 hours, the adjustment is increased. 0.05~0.10; within 24 hours When the cumulative duration is not less than 6 hours, the adjustment will be lowered. 0.05~0.10.

[0177] Calculation example: Take (Maximum value in this embodiment): (<1.5, take the calculated value); take : ;Pick (Minimum value in this embodiment): .

[0178] (4) Working condition classification:

[0179] Based on the compensation , combined and The interval was divided into four operating conditions, and 8960 standardized data points were traversed, based on real-time... corresponding The operating conditions were determined, and the number and percentage of data for each operating condition were counted, as shown in Table 8.

[0180] Table 8. Example Table 2: Number and Percentage of Data Items for Each Operating Condition

[0181] (Note: Due to coupling factor) The minimum value is 1.214, and no data falls into the weak coupling range, which is consistent with the actual scenario of strong coupling between dust accumulation and temperature in humid southern regions.

[0182] 6. Construction and validation of prediction models for different working conditions:

[0183] (1) with The guidelines handle outliers as follows:

[0184] Taking the strongly coupled-high-irradiation operating condition as an example, this condition has a total of 4301 lines. The data, calculated as follows: mean ; Standard deviation Outlier Detection Range The calculated interval is [0.654, 0.906]. Under this operating condition... There are a total of 38 data entries. A total of 12 data entries were collected. After removing 50 outliers, 4251 valid data entries remained. Similarly, outlier removal was performed for the other three operating conditions, resulting in the removal of 268 outliers from the entire dataset, representing 3.0% of the total data.

[0185] (2) Missing value handling (linear interpolation calculation):

[0186] In this embodiment, the data missing rate is 4.0%, with a total of 358 missing data points. Linear interpolation is performed based on the time series data, using the following formula: ; Calculation example: a certain time period , ; , ; Data is missing. Interpolation calculation: .

[0187] (3) Data set partitioning (calculated based on time continuity): There are a total of 8692 valid data sets. They are divided according to the 7:3 ratio and the principle of time continuity: 6084 data sets for the training set (first 70% of the time period) and 2608 data sets for the validation set (last 30% of the time period).

[0188] (4) Model parameter fitting (calculated using L2 regularized least squares method):

[0189] Formula for multivariate nonlinear regression model under different working conditions: .

[0190] The objective function for L2 regularization is: In the formula, These are the model's predicted values. The regularization coefficient is used; taking the strongly coupled-high-irradiation condition as an example, the optimal value is determined through 5-fold cross-validation. The working condition data is divided into 5 parts. One part is selected as the validation set, and four parts as the training set. A model is fitted, and the validation set MAPE is calculated. Testing is then conducted. Record each The corresponding average MAPE is shown in Table 9.

[0191] Table 9. Example Table of Average MAPE (Part 2)

[0192] Select the one with the smallest average MAPE The optimal regularization coefficients are used to fit the model parameters for this working condition: , , , , , , Similarly, parameter fitting can be performed for other strongly coupled low-irradiation conditions.

[0193] (5) Model validation results (MAPE and (Calculation): The final verification results are shown in Table 10.

[0194] Table 10. Example Table of Final Verification Results for Four Types of Working Conditions (Part Two)

[0195] 7. Comparative Analysis of Results:

[0196] The model of this invention is compared with the traditional linear prediction model, and the results are as follows:

[0197] High prediction accuracy: Based on the backcalculation results of historical operation and maintenance data of the same period using traditional models and the publicly available error benchmark range in the photovoltaic industry, the overall MAPE of traditional linear models is 10-15%, while the overall MAPE of the model of this invention is reduced to 3.9%, improving prediction accuracy by approximately 70%. Strong ease of use in engineering: The model of this invention does not require professional algorithm computing power support; operation and maintenance personnel can complete the calculations using Excel or the built-in formulas of the power plant monitoring system. Robustness: The model is effective even after continuous rainfall and the accumulation of wet and sticky ash (… ) and hot and humid weather ( Under these conditions, the prediction error of the model of this invention is stable within 5%, while the error of traditional models may be as high as 20% or more.

[0198] like Figure 3 As shown, according to another embodiment of the present invention, a photovoltaic power plant energy efficiency prediction model construction system considering the coupling effect of dust accumulation and temperature is also provided. The system includes:

[0199] The coupling factor construction module is used to obtain the proportion of dry dust accumulation area, wet dust accumulation area, and sticky dust accumulation area on the surface of photovoltaic modules, as well as the real-time temperature of the photovoltaic module surface. It also combines the normalized thermal resistance coefficients corresponding to different dust accumulation types to construct a coupling factor between dust accumulation and temperature.

[0200] The irradiance intensity correction module is used to calculate the comprehensive attenuation coefficient of dust accumulation based on the coupling factor, divide the light-receiving surface of the photovoltaic module into several detection zones, determine the non-uniform shading correction factor of the dust accumulation space based on the proportion of the total dust accumulation area of ​​each detection zone, and use the comprehensive attenuation coefficient of dust accumulation and the non-uniform shading correction factor of the dust accumulation space to double correct the pre-acquired original irradiance intensity to obtain the irradiance intensity.

[0201] The operating condition classification module is used to generate initial operating condition boundaries by taking coupling factor, irradiance and wind speed as features, and to perform collaborative calibration of the initial operating condition boundaries by combining the system energy efficiency ratio mutation range and coupling factor threshold. Based on the calibration results, multiple operating conditions are classified.

[0202] The model building module is used to construct a multivariate nonlinear regression model with the system energy efficiency ratio as the prediction objective based on each type of operating condition, so as to realize the energy efficiency prediction of the operating status of photovoltaic power plants.

[0203] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature, characterized in that, The method includes: S1. Obtain the percentage of dry dust accumulation area, wet dust accumulation area, and sticky dust accumulation area on the surface of the photovoltaic module, as well as the real-time temperature of the photovoltaic module surface. Combine this with the normalized thermal resistance coefficient corresponding to different dust accumulation types to construct a coupling factor between dust accumulation and temperature. S2. Calculate the comprehensive attenuation coefficient of dust accumulation based on the coupling factor, and divide the light-receiving surface of the photovoltaic module into several detection zones. Determine the non-uniform shading correction factor of the dust accumulation space based on the proportion of the total dust accumulation area of ​​each detection zone. Use the comprehensive attenuation coefficient of dust accumulation and the non-uniform shading correction factor of the dust accumulation space to double correct the pre-obtained original irradiance intensity to obtain the irradiance intensity. S3. Using coupling factor, irradiance and wind speed as features, generate initial operating condition boundaries, and combine the system energy efficiency ratio mutation range and coupling factor threshold to perform collaborative calibration of the initial operating condition boundaries. Based on the calibration results, classify multiple operating conditions. S4. Based on each type of operating condition, construct a multivariate nonlinear regression model with the system energy efficiency ratio as the prediction objective to achieve energy efficiency prediction of the photovoltaic power plant's operating status.

2. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature as described in claim 1, characterized in that, The expression for the coupling factor is: ； In the formula, This is the coupling factor between dust accumulation and temperature; Adjustment factor for component type; This represents the percentage of dry ash accumulation area. The normalized thermal resistance coefficient of dry ash accumulation; The percentage of wet dust accumulation area; The normalized thermal resistance coefficient of wet ash accumulation; This represents the percentage of area with sticky ash accumulation. The normalized thermal resistance coefficient of viscous ash; Temperature sensitivity coefficient; Real-time surface temperature of photovoltaic modules; This is the standard reference temperature.

3. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature as described in claim 1, characterized in that, The process involves calculating the comprehensive attenuation coefficient of dust accumulation based on the coupling factor, dividing the light-receiving surface of the photovoltaic module into several detection zones, determining the non-uniform shading correction factor of the dust accumulation space based on the proportion of the total dust accumulation area in each detection zone, and using both the comprehensive attenuation coefficient of dust accumulation and the non-uniform shading correction factor of the dust accumulation space to perform a double correction on the pre-acquired original irradiance intensity, resulting in the irradiance intensity including: S21. Based on the coupling factor, the light transmittance attenuation coefficient of each type of ash accumulation and the preset irradiance attenuation sensitivity coefficient are combined to calculate the comprehensive attenuation coefficient of ash accumulation. S22. Divide the light-receiving surface of the photovoltaic module into several detection zones, calculate the proportion of total dust accumulation area in each detection zone, and determine the non-uniform shading correction factor of the dust accumulation space based on the difference between the maximum and minimum values ​​of the proportion of total dust accumulation area in each detection zone. S23. Obtain the original solar irradiance intensity and use the comprehensive attenuation coefficient of ash accumulation and the non-uniform shading correction factor of ash accumulation space to double correct the original irradiance intensity to obtain the irradiance intensity.

4. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature as described in claim 3, characterized in that, The expression for the comprehensive attenuation coefficient of ash accumulation is: ； In the formula, The overall attenuation coefficient of ash accumulation; This is the irradiation attenuation sensitivity coefficient; This represents the percentage of dry ash accumulation area. The light transmittance attenuation coefficient for dry dust accumulation; The percentage of wet dust accumulation area; The light transmittance attenuation coefficient for wet dust accumulation; This represents the percentage of area with sticky ash accumulation. The light transmittance attenuation coefficient for sticky dust accumulation; This is the coupling factor between dust accumulation and temperature; This is the upper limit of the coupling factor.

5. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature as described in claim 1, characterized in that, The initial operating condition boundary is generated by using coupling factor, irradiance, and wind speed as features. Combined with the system energy efficiency ratio abrupt change range and coupling factor threshold, the initial operating condition boundary is collaboratively calibrated. Based on the calibration results, multiple operating conditions are classified, including: S31. Obtain real-time wind speed, combine coupling factor and irradiance intensity to construct a feature set, and standardize the feature set to generate a standardized feature set. S32. Based on the standardized feature set, unsupervised clustering is performed using the K-means clustering algorithm, and the optimal number of clusters is determined by the principle of maximizing the silhouette coefficient, so as to obtain the initial working condition center and the initial working condition boundary in the standardized space. S33. Inversely map the initial operating condition center and the initial operating condition boundary to the original physical dimensions, and combine the acquired historical operating data to perform binning statistics according to the coupling factor dimension and the irradiance intensity dimension, and calculate the average change rate of the system energy efficiency ratio between adjacent bins. S34. Compare the average change rate of the system energy efficiency ratio with a preset threshold. Based on the comparison results, determine the bin boundaries where the average change rate of the system energy efficiency ratio exceeds the preset threshold as the sudden change range of the system energy efficiency ratio. S35. The real-time wind speed is processed using a dynamic wind speed compensation mechanism to obtain the coupling factor threshold. Combined with the sudden change range of the system energy efficiency ratio, the initial operating condition boundary under the original dimensions is calibrated collaboratively to classify multiple operating conditions.

6. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature as described in claim 5, characterized in that, The method employs a dynamic wind speed compensation mechanism to process real-time wind speed, obtains a coupling factor threshold, and combines this with the system energy efficiency ratio abrupt change range to collaboratively calibrate the initial operating condition boundary under the original dimensions, thus classifying multiple operating conditions, including: S351. Using real-time wind speed as input, calculate the basic coupling factor threshold according to the preset wind speed interval segmentation rules. S352. Obtain the continuous wind speed sequence within the preset time window, calculate the cumulative duration of the first wind speed interval and the second wind speed interval respectively, perform time-series correction on the basic coupling factor threshold based on the cumulative duration, and output the coupling factor threshold. S353. The initial operating condition boundary that is inversely mapped to the original dimension is used as the basic anchor point, and the sudden change interval of the system energy efficiency ratio is used as the physical correction anchor point. Based on the coupling factor threshold, the initial operating condition boundaries of the coupling factor dimension and the irradiance intensity dimension are calibrated in a coordinated manner to obtain the coupling factor division threshold and the irradiance intensity division threshold. S354. Based on the coupling factor division threshold and the irradiance division threshold, cross-division is performed according to the preset binary combination logic to obtain multiple operating condition intervals.

7. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature as described in claim 6, characterized in that, The initial operating condition boundary, which is inversely mapped to the original dimensions, is used as the basic anchor point, and the abrupt change range of the system energy efficiency ratio is used as the physical correction anchor point. Based on the coupling factor threshold, the initial operating condition boundaries of the coupling factor dimension and the irradiance intensity dimension are jointly calibrated to obtain the coupling factor division threshold and the irradiance intensity division threshold, including: S3531. Decouple the initial working condition boundary from the inverse mapping to the original dimension by dimension, extract the initial boundary values ​​of the coupling factor dimension and the initial boundary values ​​of the irradiance intensity dimension respectively, and construct the basic anchor point. S3532. Based on the system energy efficiency ratio mutation interval, extract the mutation critical value set in the coupling factor dimension and the mutation critical value set in the irradiance dimension respectively, and construct physical correction anchor points. S3533. Using the coupling factor threshold as a dynamic reference benchmark, and combining the basic anchor point and the physical correction anchor point, the initial boundary values ​​of the coupling factor dimension and the irradiance dimension are calibrated according to the dual-dimensional collaborative calibration rule, so as to generate the initial coupling factor division threshold and the initial irradiance division threshold. S3534. Using preset physical boundary constraints, constrain the initial coupling factor division threshold and the initial irradiance division threshold to obtain the coupling factor division threshold and the irradiance division threshold.

8. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature according to claim 7, characterized in that, The process of using the coupling factor threshold as a dynamic reference benchmark, combined with the basic anchor point and the physical correction anchor point, and calibrating the initial boundary values ​​of the coupling factor dimension and the irradiance dimension according to the dual-dimensional collaborative calibration rule, to generate the initial coupling factor division threshold and the initial irradiance division threshold includes: The mean of the set of abrupt change critical values ​​along the coupling factor dimension is extracted from the physical correction anchor points and used as the coupling change feature value. If a coupling change feature value exists in the physical correction anchor points, the absolute difference between the coupling change feature value and the coupling factor threshold is calculated, and the absolute difference is compared with half of the historical bin width. When the absolute difference exceeds half of the historical bin width, the coupling factor threshold is used as the initial coupling factor partitioning threshold; when the absolute difference does not exceed half of the historical bin width, the weighted average between the coupling mutation feature value and the coupling factor threshold is calculated as the initial coupling factor partitioning threshold; if there is no coupling mutation feature value in the physical correction anchor point, the coupling factor threshold is used as the initial coupling factor partitioning threshold. The mean of the set of abrupt change critical values ​​in the irradiance intensity dimension is extracted from the physical correction anchor points as the irradiance change feature value. If an irradiance change feature value exists in the physical correction anchor points, the relative deviation between the irradiance change feature value and the initial irradiance intensity threshold in the base anchor points is calculated, and the relative deviation is compared with a preset ratio. When the relative deviation exceeds the preset ratio, the initial irradiance threshold is used as the initial irradiance division threshold; when the relative deviation does not exceed the preset ratio, the weighted average between the irradiance mutation feature value and the initial irradiance threshold is calculated as the initial irradiance division threshold; if there is no irradiance mutation feature value in the physical correction anchor point, the initial irradiance threshold is used as the initial irradiance division threshold.

9. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature according to claim 1, characterized in that, The process of constructing a multivariate nonlinear regression model with the system energy efficiency ratio as the prediction objective for each type of operating condition, in order to predict the energy efficiency of the photovoltaic power plant's operating status, includes: S41. Obtain historical samples that have undergone outlier removal and missing value imputation, and construct training datasets according to the operating condition categories respectively; S42. Based on each type of operating condition, with the system energy efficiency ratio as the prediction target, construct an initial multivariate nonlinear regression model, and use the training dataset of the corresponding operating condition to train the initial multivariate nonlinear regression model to obtain the multivariate nonlinear regression model for each type of operating condition. S43. Calculate the coupling factor and irradiance of the sample to be predicted, and combine the coupling factor division threshold and the irradiance division threshold to determine the boundary buffer where the sample to be predicted is located. Based on the judgment result, use the corresponding multivariate nonlinear regression model to make a prediction and output the predicted value of the system energy efficiency ratio to realize the energy efficiency prediction of the photovoltaic power station's operating status.

10. The method for constructing a photovoltaic power plant energy efficiency prediction model considering the coupling effect of dust accumulation and temperature according to claim 9, characterized in that, The process involves calculating the coupling factor and irradiance of the sample to be predicted, and combining the coupling factor threshold and irradiance threshold to determine the boundary buffer where the sample is located. Based on the determination result, a corresponding multivariate nonlinear regression model is used for prediction, and the predicted system energy efficiency ratio is output to achieve energy efficiency prediction of the photovoltaic power plant's operating status. By calculating the coupling factor and irradiance of the sample to be predicted, a two-dimensional difference of the sample to be predicted is obtained; wherein, the two-dimensional difference includes the absolute value of the difference between the coupling factor and the coupling factor division threshold, and the absolute value of the difference between the irradiance and the irradiance division threshold. When both differences between the two dimensions are less than the corresponding buffer threshold, the sample to be predicted is determined to be within the boundary buffer; otherwise, the sample to be predicted is determined to be outside the boundary buffer. If the sample to be predicted is outside the boundary buffer, the multivariate nonlinear regression model corresponding to the operating condition to which the sample to be predicted belongs is used for prediction, and the predicted value of the system energy efficiency ratio is output. If the sample to be predicted is within the boundary buffer, then the two types of operating conditions adjacent to the sample to be predicted are identified, and the multivariate nonlinear regression models corresponding to the two types of operating conditions are called simultaneously for prediction, outputting two sets of initial system energy efficiency ratio prediction values; the Euclidean distance from the sample to be predicted to the center of the two types of operating conditions in the standardized feature space is calculated, and the weights corresponding to the two sets of initial system energy efficiency ratio prediction values ​​are calculated using the inverse distance method, and the two sets of initial system energy efficiency ratio prediction values ​​are weighted and fused through the weights to obtain the system energy efficiency ratio prediction value.

Images