Meteorological-physical mapping new energy multi-time scale output prediction method based on Transform
By constructing a deeply coupled meteorological-physical mapping model using Transformer, the problem of poor adaptability of new energy output forecasting under extreme weather conditions was solved, achieving high-precision multi-timescale forecasting and grid dispatch support.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN UNIV
- Filing Date
- 2026-01-20
- Publication Date
- 2026-05-15
AI Technical Summary
Existing methods for predicting renewable energy output have poor adaptability to extreme weather conditions, unclear mechanisms, and difficulty in meeting the multi-stage dispatching needs of the power grid. In particular, they fail to effectively constrain and utilize spatiotemporal uncertainties in complex environments.
We employ a Transformer-based data-driven dynamic meteorological forecasting model that deeply couples data with a rigorous photovoltaic physics mapping model. We construct a hybrid dynamic meteorological forecasting model combining CausalTransformer, CVAE-GAN, and GRU, and combine it with the photovoltaic physics model to predict power output at multiple time scales.
It achieves high-precision multi-timescale forecasting under extreme weather conditions, improves forecast accuracy and robustness, supports multi-stage grid dispatching needs, and provides reliable technical support for new energy power forecasting in complex environments.
Smart Images
Figure CN122051930A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of new energy output prediction technology, specifically a method for predicting new energy output across multiple time scales based on meteorological-physical mapping using Transformer. Background Technology
[0002] Currently, new energy output forecasting mainly relies on physical models driven by numerical weather prediction or machine learning methods based on historical data. While the mechanisms of physical models are well-defined, their accuracy is limited by the accuracy of the input meteorological data and the completeness of equipment parameters, and they are insufficiently adaptable to rapidly evolving extreme weather events such as sandstorms. Data-driven models perform well when there are sufficient samples, but their "black box" nature leads to a lack of interpretability in the forecasting process, and their generalization ability is weak in newly built bases with scarce data or under rare weather conditions. Existing hybrid-driven methods mostly employ simple serial processing or result fusion, failing to achieve deep coupling between the data-driven process and physical mechanisms. The spatiotemporal uncertainties of meteorological forecasts are not effectively constrained and utilized by physical models, and there is a lack of a unified multi-timescale forecasting architecture to support multi-stage grid dispatching. These problems are particularly prominent in complex environments such as deserts and Gobi, urgently requiring a new forecasting method that can deeply integrate spatiotemporal dynamic forecasting and physical mapping, and has multi-scale output capabilities. Summary of the Invention
[0003] To address the aforementioned problems, the present invention aims to provide a Transformer-based method for predicting renewable energy output across multiple time scales using meteorological-physical mapping. By deeply coupling data-driven dynamic meteorological forecasting with a rigorous photovoltaic physical mapping model using Transformer, this method solves the core problems of existing methods, such as poor adaptability under extreme weather conditions, unclear mechanisms, and difficulty in meeting the multi-stage dispatching requirements of the power grid. This provides reliable technical support for renewable energy power prediction in complex environments. The technical solution is as follows:
[0004] A method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping using Transformer includes the following steps:
[0005] Step S1: Data Acquisition and Preprocessing: Collect multi-source time-series meteorological datasets of the target new energy base and its surrounding areas, and preprocess the multi-source time-series meteorological datasets;
[0006] Step S2: Key meteorological element extraction: Perform feature analysis on the data in the preprocessed multi-source time-series meteorological dataset to screen out the meteorological elements that have a key impact on meteorological dynamics prediction and photovoltaic power output prediction;
[0007] Step S3: Build a dynamic meteorological prediction model based on a generative model and train it using a dataset; the dynamic meteorological prediction model includes a meteorological feature prediction module based on the Transformer architecture, a dynamic meteorological prediction module based on a conditional variational autoencoder combined with GAN denoising, and a meteorological feature prediction module based on GRU.
[0008] Step S4: Build a physical model for predicting new energy power output;
[0009] Step S5: When predicting the output of new energy sources, input the input data into the meteorological dynamic prediction model to obtain meteorological prediction results at different time scales, and then input the meteorological prediction results into the physical model to obtain the new energy output prediction results.
[0010] The beneficial effects of this invention are:
[0011] This invention addresses the challenges of rapid meteorological changes and strong spatiotemporal heterogeneity in photovoltaic (PV) output forecasting in desert and Gobi regions under extreme weather conditions such as sandstorms. It proposes a multi-timescale PV output forecasting method that deeply couples a Transformer with data-driven dynamic meteorological forecasting and combines it with a rigorous meteorological-PV output physical mapping model. By constructing a hybrid dynamic meteorological forecasting model integrating CausalTransformer, CVAE-GAN, and GRU, it solves the core problems of existing methods, such as poor adaptability under extreme weather conditions, unclear mechanisms, and difficulty in meeting the multi-stage dispatching requirements of the power grid. This achieves high-precision, multi-timescale forecasting of regional irradiance spatiotemporal evolution. Deeply coupling the PV physical mechanism model reliably maps predicted meteorological elements to output, effectively integrating the adaptability of data-driven approaches with the interpretability of physical models. This significantly improves forecasting accuracy and model robustness, supports multi-timescale collaborative output from short-term to medium-to-long-term, meets the multi-stage dispatching requirements of the power grid, and provides reliable technical support for PV power forecasting in complex environments. Attached Figure Description
[0012] Figure 1 This is a schematic diagram of the prediction model framework of the present invention.
[0013] Figure 2 The predicted and actual GHI (Global Horizontal Irradiance) values for the target area obtained by this method.
[0014] Figure 3 The predicted and actual values of DNI (Direct Normal Irradiance) for the target area obtained by this method.
[0015] Figure 4These are the predicted and actual values of DHI (Diffuse Horizontal Irradiance) for the target area obtained using this method.
[0016] Figure 5 This is a graph showing the differential calculation results of the three irradiance predictions for the target area obtained by this method.
[0017] Figure 6(a) Comparison of the predicted photovoltaic power output of the target area on April 5, 2019 with the actual power output using this method.
[0018] Figure 6(b) shows a comparison between the predicted photovoltaic power output of the target area on April 19, 2019, and the actual power output, obtained using this method.
[0019] Figure 6(c) Comparison of the predicted photovoltaic power output of the target area on May 11, 2019 with the actual power output using this method.
[0020] Figure 6(d) Comparison of the predicted photovoltaic power output of the target area on May 18, 2019 with the actual power output using this method.
[0021] Figure 7(a) Comparison of the predicted photovoltaic power output of the target area on May 24, 2019 with the actual power output using this method.
[0022] Figure 7(b) shows a comparison between the predicted photovoltaic power output of the target area on May 25, 2019, and the actual power output, obtained using this method.
[0023] Figure 7(c) Comparison of the predicted photovoltaic power output of the target area on June 8, 2019 with the actual power output using this method.
[0024] Figure 7(d) Comparison of the predicted photovoltaic power output of the target area on June 10, 2019 with the actual power output using this method.
[0025] Figure 8(a) Comparison of the predicted photovoltaic power output of the target area on April 2, 2019 with the actual power output using this method.
[0026] Figure 8(b) Comparison of the predicted photovoltaic power output of the target area on May 6, 2019 with the actual power output using this method.
[0027] Figure 8(c) Comparison of the predicted photovoltaic power output of the target area on June 9, 2019 with the actual power output using this method.
[0028] Figure 8(d) Comparison of the predicted photovoltaic power output of the target area on June 12, 2019 with the actual power output using this method.
[0029] Figure 9(a) Comparison of the predicted photovoltaic power output of the target area on April 12, 2019 with the actual power output using this method.
[0030] Figure 9(b) shows a comparison between the predicted photovoltaic power output of the target area on April 21, 2019, and the actual power output, obtained using this method.
[0031] Figure 9(c) Comparison of the predicted photovoltaic power output of the target area on April 24, 2019 with the actual power output using this method.
[0032] Figure 9(d) Comparison of the predicted photovoltaic power output of the target area on April 25, 2019 with the actual power output using this method. Detailed Implementation
[0033] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0034] A method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping using Transformer includes the following steps:
[0035] Step S1: Data Acquisition and Preprocessing: Collect multi-source time-series meteorological datasets of the target new energy base and its surrounding areas, and preprocess the data.
[0036] The multi-source time-series meteorological dataset for the target new energy base and its surrounding area is defined as follows: ,in, Represents satellite remote sensing cloud images, Images representing incident solar radiation on the Earth's surface. Represents temperature, Represents wind speed. Represents wind direction. Represents relative humidity. This represents air pressure. A set of data is measured every 15 minutes to form a sample point.
[0037] The method for handling outlier data in the dataset is as follows:
[0038] For missing data: take the average of the data at the two adjacent times of the missing time point as a replacement. If missing data also appears at adjacent times, discard the sample point at that time.
[0039] For remote sensing cloud images and solar irradiance images, if the image is missing, damaged, or otherwise abnormal, the sample point is discarded directly.
[0040] During data preprocessing, the divided data is normalized using the Min-Max normalization method to linearly map the data to the range [0,1], thereby removing the influence of different dimensions of the data.
[0041] ;
[0042] In the formula, For the normalized data, The minimum value in the batch set. This is the maximum value.
[0043] Step S2: Extraction of key meteorological elements: Perform feature analysis on the data in the multi-source time-series meteorological dataset to screen out the meteorological elements that have a key impact on meteorological dynamic forecasting and photovoltaic power output forecasting.
[0044] The feature analysis and screening method is constructed based on the Pearson correlation coefficient and the Spearman correlation coefficient. The Pearson correlation coefficient is calculated as follows:
[0045] ;
[0046] In the formula, and Here, n is the sample mean, and n is the sample size. and This is the sample used for correlation analysis. Meteorological factors with an absolute correlation coefficient greater than 0.3 were selected as key influencing factors.
[0047] The Spearman correlation coefficient is calculated as follows:
[0048] ;
[0049] In the formula, n is the sample size. and Sample pairs In the sequence respectively and rank in and It is obtained by rearranging the data from the two sample sets whose correlation is to be analyzed from largest to smallest. Meteorological factors with an absolute correlation coefficient greater than 0.3 are selected as key influencing factors.
[0050] Finally, by taking the union of the key factors obtained in the two previous steps, the resulting multi-source time-series meteorological dataset supporting new energy processing and prediction is constituted as follows: .
[0051] Step S3: Build a dynamic meteorological prediction model based on a generative model and train it using the dataset.
[0052] The constructed meteorological dynamic prediction model consists of the following components: a meteorological feature prediction module based on the Transformer architecture (Causal Transformer), a meteorological dynamic prediction module based on conditional variational autoencoder combined with GAN (Generative Adversarial Network) for noise reduction (CVAE-GAN), and a meteorological feature prediction module based on GRU (GRU). Its model structure is as follows: Figure 1 As shown.
[0053] For the Transformer part, the Causal Transformer model was chosen as the main structure, which receives solar irradiance images. . The shape is W and H are the width and height of the solar irradiance image. Before inputting the model, it is cut into m*n shapes. The sub-blocks are processed by a multi-layer CNN structure for feature extraction and dimensionality reduction. After dimensionality reduction, the feature vectors are flattened to obtain m*n 32-dimensional vectors. These vectors are then used as word vectors and input into the CausalTransformer model. The objective function of the Causal Transformer is defined as follows:
[0054] ;
[0055] Where B represents the amount of data used for training; T represents the sequence length of the irradiated images used for prediction; and t represents time. Represents a sub-block obtained from the segmentation, index Representative by The resulting m*n sub-block sequence; In Teacher Forcing training mode, The corresponding predicted value.
[0056] For the meteorological dynamic prediction model part of the conditional variational autoencoder combined with GAN denoising, its received irradiation image data The image at time t is used as a condition, and the image at time t+1 is used as the target image for generation. Simultaneously, a GAN model is introduced to discriminate the generated images, suppressing noise in the images generated by the conditional variational autoencoder. The objective function for the generation stage is defined as follows:
[0057] ;
[0058] in, The irradiation image of the b-th batch at the t-th time step. To pass The calculated prior distribution; the neural network outputs a set of Gaussian distribution parameters, Z is a random variable determined by these distributions, and the sampled values of Z are denoted as... ; To pass through sampled values and irradiation images The calculated posterior distribution; For the decoder to use sampled values The decoded image; for The probability that the discriminator determines it to be true; This represents the KL divergence operator.
[0059] The objective function for the identification stage is as follows:
[0060] ;
[0061] The GRU part receives wind speed, wind direction, and temperature data, and predicts the wind speed, wind direction, and temperature at the next moment. Its objective function is as follows:
[0062] ;
[0063] in, and These are the temperature at time t and the predicted temperature at the next time point, respectively. and These represent the wind speed at time t and the predicted wind speed for the next time step, respectively. and These represent the wind direction at time t and the predicted wind direction at the next time point, respectively.
[0064] for Figure 1 The hybrid network of the meteorological dynamic prediction model receives the predicted values of the radiation image at time t from the Transformer part and the conditional variational autoencoder part, and outputs the final predicted radiation image at time t. Its objective function is:
[0065] ;
[0066] Step S4: Build a physical model for predicting the output of new energy sources.
[0067] The physical model architecture used for predicting new energy power output is as follows:
[0068] (1) Physical calculation model of photovoltaic inverter:
[0069] ;
[0070] Its model chain consists of five parts in sequence: tilted surface irradiance calculation, incident angle loss calculation, photovoltaic panel DC characteristic modeling, photovoltaic array power aggregation, and inverter AC conversion.
[0071] (2) Calculation of irradiance on inclined surface:
[0072] The direct irradiance of the inclined surface is calculated using geometric projection relationships, as follows:
[0073] ;
[0074] in, The angle of incidence of the sun; The solar altitude angle; The angle of inclination of the plate; This is the azimuth angle.
[0075] Scattered irradiance was calculated using the Perez model, which divides scattered light into three components: isotropic scattering, circumsolar scattering, and horizon scattering, as follows:
[0076] ;
[0077] in, and This is the brightness coefficient, which is related to the clarity and brightness of the sky. It is a geometric factor.
[0078] Ground reflected irradiance is calculated using total horizontal radiation and surface reflectivity, as follows:
[0079] ;
[0080] in, This represents the ground reflectance coefficient.
[0081] The total irradiance on the inclined surface is the sum of the three components of the inclined surface irradiance:
[0082] ;
[0083] (3) Calculation of incident angle loss:
[0084] When sunlight does not strike the glass cover of a photovoltaic module perpendicularly, some of the light is reflected and cannot reach the cells to be absorbed and generate electricity. This reflection loss follows Fresnel's equations in optics:
[0085] ;
[0086] in, IAM represents the irradiance effectively absorbed by the battery; IAM is the loss factor. The angle of refraction; The angle of incidence is denoted as .
[0087] (4) Modeling of DC characteristics of photovoltaic panels:
[0088] The DC output power of the photovoltaic module under standard test conditions (STC) needs to be calculated based on the total irradiance of the tilted surface, ambient temperature, and wind speed.
[0089] ;
[0090] in, This refers to the temperature of the solar cells. Ambient temperature; The temperature rise of the solar cells relative to the environment; and This is an empirical coefficient; This is the real-time wind speed.
[0091] The photovoltaic cell model parameters are solved iteratively using STC parameters and the temperature coefficient:
[0092] ;
[0093] in, Photocurrent; This is the reverse saturation current; It is a series resistor; d represents the parallel resistance; d is the diode factor. q is the thermal voltage; k is the Boltzmann constant; q is the elementary charge.
[0094] (5) Photovoltaic array power aggregation:
[0095] With multiple photovoltaic modules connected in series, the voltage is the sum of the module voltages, and the current is limited by the module with the worst power generation in the string.
[0096] ;
[0097] In series and parallel connections, the current is the sum of the currents in each series:
[0098] ;
[0099] (6) Inverter AC conversion:
[0100] AC power is calculated using the Sandia inverter model:
[0101] ;
[0102] in, This represents the efficiency curve coefficient.
[0103] Step S5: When predicting the output of new energy sources, input the input data into the meteorological dynamic prediction model to obtain the prediction results at different time scales. Input the meteorological prediction results into the physical model to obtain the new energy output prediction results.
[0104] To further verify the effectiveness of the Transformer-coupled joint embedding prediction method for predicting renewable energy output across multiple time scales using meteorological-physical mapping, a simulation experiment was conducted. Specifically, photovoltaic power generation data and measured meteorological data from a certain region from 2019 to 2023 were used for training and prediction.
[0105] This experiment was conducted using Python and the Tensorflow 2.0 framework. The computer used was configured with an Intel Core i5-9300H CPU @ 2.40 GHz and one NVIDIA GTX 1660Ti GPU.
[0106] Figure 2 , Figure 3 and Figure 4 (Collection date: April 9, 2023) The prediction results of the meteorological prediction model designed in this invention are shown respectively. It can be seen that the prediction error is relatively small and has sufficient accuracy. Figure 5 This is a graph showing the differential calculation results of the three irradiance predictions for the target area obtained by this method.
[0107] Figures 6(a)-6(d), 7(a)-7(d), 8(a)-8(d), and 9(a)-9(d) show the comparison between the calculated photovoltaic power output and the actual output under four dust storm events in the area where the photovoltaic power station is located in 2019, selected from meteorological yearbook reports. The calculated output was obtained by inputting the measured total horizontal irradiance (GHI), direct irradiance (DNI), diffuse irradiance (DHI), temperature, and wind speed into the physical model. The results demonstrate the prediction results of the multi-timescale renewable energy output prediction model designed in this invention, showing that the prediction error is relatively small and has sufficient accuracy.
[0108] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements 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 predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping using Transformer, characterized in that, Includes the following steps: Step S1: Data Acquisition and Preprocessing: Collect multi-source time-series meteorological datasets of the target new energy base and its surrounding areas, and preprocess the multi-source time-series meteorological datasets; Step S2: Key meteorological element extraction: Perform feature analysis on the data in the preprocessed multi-source time-series meteorological dataset to screen out the meteorological elements that have a key impact on meteorological dynamics prediction and photovoltaic power output prediction; Step S3: Build a dynamic meteorological prediction model based on a generative model and train it using a dataset; the dynamic meteorological prediction model includes a meteorological feature prediction module based on the Transformer architecture, a dynamic meteorological prediction module based on a conditional variational autoencoder combined with GAN denoising, and a meteorological feature prediction module based on GRU. Step S4: Build a physical model for predicting new energy power output; Step S5: When predicting the output of new energy sources, input the input data into the meteorological dynamic prediction model to obtain meteorological prediction results at different time scales, and then input the meteorological prediction results into the physical model to obtain the new energy output prediction results.
2. The method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping according to claim 1, characterized in that, In step S1, the multi-source time-series meteorological dataset of the target new energy base and its surrounding area is defined as: ； in, Represents satellite remote sensing cloud images, Images representing solar radiation incident on the Earth's surface. Represents temperature, Represents wind speed. Represents wind direction. Represents relative humidity. Represents air pressure.
3. The method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping according to claim 2, characterized in that, Step S1 also includes processing the abnormal data in the multi-source time-series meteorological dataset, specifically including: For missing data: take the average of the data at the two adjacent times of the missing time point as a replacement; if missing data also appears at adjacent times, discard the sample point at that time. For satellite remote sensing cloud images and solar radiation images If the image is missing, damaged, or otherwise abnormal, the sample point is discarded directly.
4. The method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping according to claim 1, characterized in that, In step S1, when preprocessing the multi-source time-series meteorological dataset, the divided data is normalized by using the Min-Max normalization method to linearly map the data to the range [0,1], in order to remove the influence of different dimensions of the data. ； In the formula, For the normalized data, The minimum value in the batch set. This is the maximum value.
5. The method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping according to claim 2, characterized in that, Step S2 specifically includes: Step S21: Calculate the Pearson correlation coefficient among multi-source time-series meteorological data: ； In the formula, and Here, n is the sample mean, and n is the sample size. and These are samples used for correlation analysis; Meteorological factors with an absolute value of correlation coefficient greater than the set Pearson correlation coefficient threshold were selected as key influencing factors; Step S22: Calculate the Spearman correlation coefficient among multi-source time-series meteorological data: ； In the formula, n is the sample size. and Sample pairs In the sequence respectively and rank in and It is obtained by rearranging the data of the two sample sets whose correlation is to be analyzed from largest to smallest; Meteorological factors with an absolute value of correlation coefficient greater than the set Spearman correlation coefficient threshold were selected as key influencing factors; Step S23: Based on the obtained key influencing factors, construct a multi-source time-series meteorological dataset to support new energy processing and prediction. .
6. The method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping according to claim 2, characterized in that, In step S3, the meteorological dynamic prediction model specifically includes: 1) Meteorological feature prediction module based on Transformer architecture: The meteorological feature prediction module based on the Transformer architecture selects the Causal Transformer model as the main structure and receives solar irradiance images. ; The shape is W and H represent the width and height of the solar irradiance image. Before inputting it into the model, the image is divided into m*n sub-blocks. Each sub-block undergoes feature extraction and dimensionality reduction through a multi-layer CNN structure. After dimensionality reduction, the feature vectors are flattened to obtain m*n 32-dimensional vectors. These vectors are then used as word vectors and input into the Causal Transformer model. The objective function of the Causal Transformer is defined as follows: ； Where B represents the amount of data used for training; T represents the sequence length of the irradiated images used for prediction; and t is the time step. Represents a sub-block obtained from the segmentation, index Representative by The resulting m*n sub-block sequence; In Teacher Forcing training mode, The corresponding predicted value; 2) A dynamic meteorological forecasting module based on conditional variational autoencoder combined with GAN denoising: A meteorological dynamic prediction module based on conditional variational autoencoder combined with GAN noise reduction receives solar irradiance images. The image at time t is used as a condition, and the image at time t+1 is used as the target for generation. Simultaneously, a GAN model is introduced to discriminate the generated images, suppressing noise in the images generated by the conditional variational autoencoder. The objective function of the generation stage is... The definition is as follows: ； in, This is the irradiation image of the b-th batch at the t-th time step. To pass The calculated prior distribution; the neural network outputs a set of Gaussian distribution parameters, Z is a random variable determined by these distributions, and the sampled values of Z are denoted as... ; To pass through sampled values and irradiation images The calculated posterior distribution; For the decoder to use sampled values The decoded image; for The probability that the discriminator determines it to be true; Represents the KL divergence operator; Identification stage objective function as follows: ； 3) GRU-based meteorological feature prediction module: The meteorological feature prediction module based on GRU receives wind speed, wind direction, and temperature data, and predicts the wind speed, wind direction, and temperature at the next moment; its objective function is... as follows: ； in, and These are the temperature at time t and the predicted temperature at the next time point, respectively. and These represent the wind speed at time t and the predicted wind speed for the next time step, respectively. and These represent the wind direction at time t and the predicted wind direction for the next time step, respectively. 4) For the network of the meteorological dynamic prediction model, it receives the predicted values of the solar irradiance image at time t from the meteorological feature prediction module based on the Transformer architecture and the meteorological dynamic prediction module based on conditional variational autoencoder combined with GAN denoising, and outputs the final predicted solar irradiance image at time t; its objective function is... for: 。 7. The method for predicting the output of new energy sources across multiple time scales based on meteorological-physical mapping according to claim 2, characterized in that, In step S4, the physical model architecture used for predicting new energy power output is as follows: (1) Physical calculation model of photovoltaic inverter: ； in, For AC power; the model chain is as follows: Calculation of irradiance on inclined surface Calculation of incident angle loss Modeling of DC characteristics of photovoltaic panels Photovoltaic array power aggregation AC conversion with inverter ; The three elements of horizontal irradiance are: direct normal irradiance (DNI), diffuse horizontal irradiance (DHI), and total horizontal irradiance (GHI). This refers to the temperature of the solar cells. (2) The irradiance of the inclined surface is calculated as follows: Direct irradiance of inclined surface The calculation, based on geometric projection relationships, is as follows: ； in, The angle of incidence of the sun; The solar altitude angle; The angle of inclination of the plate; It is the azimuth angle; Scattered irradiance of inclined surface Using the Perez model, the scattered light is divided into three components: isotropic scattering, heliotropic scattering, and horizon scattering, as follows: ； in, and This is the brightness coefficient, which is related to the clarity and brightness of the sky. Geometric factor; Ground reflected irradiance on sloping surfaces The calculations are performed using total horizontal radiation and surface reflectivity, as follows: ； in, Ground reflectance coefficient; Total irradiation of inclined surface The sum of the three components of the irradiance on the inclined surface: ； (3) The incident angle loss is calculated as follows: According to Fresnel's equations, when sunlight does not strike the glass cover of a photovoltaic module perpendicularly, the effective irradiance absorbed by the cells is... The calculation is as follows: ； in, IAM represents the irradiance effectively absorbed by the battery; IAM is the loss factor. The angle of refraction; Angle of incidence; (4) Modeling of DC characteristics of photovoltaic panels: Calculate the DC output power of the photovoltaic module under standard test conditions based on the total irradiance of the tilted surface, ambient temperature, and wind speed: ； in, Ambient temperature; The temperature rise of the solar cells relative to the environment; and This is an empirical coefficient; Real-time wind speed; This is the reverse saturation current; The photovoltaic cell model parameters are solved iteratively using standard test condition parameters and temperature coefficient: ； in, This is the equivalent current of the photovoltaic cell; Photocurrent; It is a series resistor; d represents the parallel resistance; d is the diode factor. q is the thermal voltage; k is the Boltzmann constant; q is the elementary charge; This refers to the number of solar cells connected in series in the module; The equivalent voltage of the photovoltaic cell; (5) Photovoltaic array power aggregation: The total voltage of a photovoltaic panel module connected in series For component voltage The sum of currents Limited by the photovoltaic panel module with the worst power generation in the string, the calculation is as follows: ； in, Let N be the equivalent current of the Nth series component; Total current of series photovoltaic modules connected in parallel For each string current sum: ； in, The output power of the photovoltaic panel modules connected in series; for The total voltage when series components are connected in parallel; (6) Inverter AC conversion: Calculating AC power using a Sandia inverter model : ； in, For efficiency curve coefficients; This refers to the rated AC output power of the inverter. This refers to the rated DC input power of the inverter. This represents normalized DC power.