Photovoltaic power generation forecasting method based on spectrum forecasting

Abstract

The invention discloses a photovoltaic power generation forecasting method based on spectrum forecasting, and aims at solving the problem that errors are large due to the fact that only temperature and total irradiance are considered in an existing photovoltaic power generation forecasting method. The method comprises the following steps: downloading global forecast data and carrying out lattice point coding conversion; then carrying out dynamic downscaling and calculating future key meteorological variables; calculating the spectral optical thickness of each atmospheric component by using the spectral absorption and scattering coefficients of different gas components in the HITRAN database, and obtaining direct and scattered spectral irradiance according to the spectral optical thickness; carrying out calculation result verification and artificial intelligence correction by utilizing a solar spectrum radiometer; calculating generated power of different wavelengths; and calculating the final generation power through integral calculation in combination with the loss of other parts of the photovoltaic module. And forecasting a radiation spectrum at a future moment according to different forecasted meteorological variables, and converting radiation into power generation power according to the spectrum, thereby realizing refined conversion of optical power.

Description

Technical Field

[0001] This invention relates to the field of photovoltaic power generation prediction technology, and specifically to a photovoltaic power generation prediction method based on spectral forecasting. Background Technology

[0002] Solar power generation forecasting is primarily based on radiative transfer theory, combining ground-based meteorological observations, upper-air atmospheric sounding, satellite remote sensing, and numerical simulations. It establishes solar radiation forecasting models by acquiring relevant parameters such as water vapor content, cloud cover, temperature, and humidity. Solar radiation forecasting methods are mainly divided into three categories: first, statistical forecasting methods based on statistical regression and artificial intelligence, utilizing historical data and machine learning techniques for modeling and prediction; second, image-based forecasting methods based on satellite cloud images and ground-based sky imaging, predicting by analyzing cloud dynamics and radiation changes; and third, physical methods based on numerical weather prediction, simulating changes in solar radiation through atmospheric physical processes. Each of these three methods has its strengths and can provide effective forecast results at different time scales and application scenarios. Statistical methods perform better in short-term forecasts, image-based methods are more suitable for ultra-short-term forecasts, while physical methods are commonly used for medium- and long-term forecasts.

[0003] Currently, almost all solar radiation forecasting models predict total irradiance, and photovoltaic power forecasting models use relatively simple formulas to convert total irradiance to total power. This conversion process only considers temperature and total irradiance, neglecting some details and leading to process errors; for example, Chinese patent CN118868030A. In reality, photovoltaic panels made of different materials have different conversion efficiencies for solar radiation of different wavelengths, and different meteorological variables also have different reduction effects on solar radiation of different wavelengths. Therefore, predicting only total irradiance may overlook many details, leading to errors in variable conversion. Summary of the Invention

[0004] This invention addresses the problem of large errors caused by current photovoltaic power generation forecasting methods that only consider temperature and total irradiance. It proposes a photovoltaic power generation forecasting method based on spectral forecasting, which forecasts the radiation spectrum at future times based on different meteorological variables, and converts the radiation into power generation by dividing the spectrum, thus achieving refined conversion of light power.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a photovoltaic power generation forecasting method based on spectral forecasting, comprising the following steps: S1. Download global forecast data and perform gridded encoding on the data; use the WRF model to perform dynamic downscaling and calculate key future meteorological variables; S2. Using the spectral absorption and scattering coefficients of different gas components in the HITRAN database, calculate the spectral optical thickness of each atmospheric component, and obtain the direct and scattered spectral irradiance based on the spectral optical thickness. S3 utilizes a solar spectroradiometer to verify calculation results and perform artificial intelligence correction. S4 calculates the power generation at different wavelengths by combining the external quantum efficiency of the photovoltaic cell; the final power generation is calculated by integration and by combining the losses of other parts of the photovoltaic module.

[0006] This technical solution proposes a photovoltaic power generation forecasting method based on spectral forecasting. Instead of directly forecasting the total irradiance, it forecasts the radiation spectrum at future times based on forecasted meteorological variables (water content in the air, water cloud amount, and ice cloud amount), and converts the radiation into power generation power by dividing the spectrum, thus achieving refined conversion of light power.

[0007] The present invention is further configured such that step S1 includes: S11, download global forecast data including several meteorological variables, and then convert the data into a standard format supported by WPS using a grid-based interpolation method; S12, the data processed by WPS is input into the WRF model, the WRF model is used for dynamic downscaling, and the key meteorological variables for the future period are calculated based on the WRF model.

[0008] In this technical solution, steps S11 and S12 mainly involve data acquisition and preprocessing, dynamic downscaling, and variable calculation. The calculated variables can provide a basis for subsequent calculations of optical thickness and spectral irradiance.

[0009] The present invention is further configured such that: the utilization of the spectral absorption and scattering coefficients of different gas components in the HITRAN database includes: The water vapor absorption coefficient was obtained by searching and calculating the HITRAN database; The absorption and scattering coefficients of water clouds or ice clouds were calculated using Mie scattering theory.

[0010] In this technical solution, since the absorption of radiation by water vapor molecules is much greater than the scattering, the effect of the scattering effect of water vapor molecules on solar shortwave radiation can be ignored within the scope of this solution. Therefore, only the absorption effect is considered here.

[0011] The present invention is further configured such that: the calculation of the spectral optical thickness of each atmospheric component includes calculating the optical thickness of the entire atmosphere and the scattering optical thickness. The optical thickness of the entire atmosphere is: the absorption coefficient and the scattering coefficient at each altitude from the ground to the top of the atmosphere are added together and then integrated. The scattering optical thickness is: the scattering coefficient at each altitude or pressure layer is considered and integrated in the altitude direction, or the scattering coefficient of each layer in the discrete layers is multiplied by its physical thickness and then summed.

[0012] In this technical solution, the optical thickness of the entire atmosphere can be obtained based on the calculated future absorption and scattering coefficients of each atmospheric component.

[0013] The present invention is further configured such that the spectral irradiance of the direct radiation is expressed as: The radiation intensity at the top of the atmosphere at that wavelength is multiplied by a first exponential attenuation factor, which is determined by the total optical thickness of the atmosphere at that wavelength.

[0014] In this technical solution, the first exponential decay factor is exp(-τ) total,λ The intensity of direct radiation is equal to the portion of the original radiation intensity that is reduced due to absorption and scattering.

[0015] The present invention is further configured such that the spectral irradiance of the scattered radiation is expressed as: The original radiation intensity at the top of the atmosphere is multiplied by a second exponential attenuation factor, and then multiplied by an empirical scattering factor; the second exponential attenuation factor is determined by the scattering optical thickness at that wavelength.

[0016] In this technical solution, the second exponential decay factor is 1 minus exp(-τ) s，λ ).

[0017] The present invention is further configured such that step S3 includes: The deviation between the actual measured spectral radiation and the calculated radiation value of the solar spectroradiometer is obtained. This deviation is corrected by constructing an artificial neural network model. The artificial neural network model is achieved by optimizing an objective function, which is to minimize the sum of errors between the observed values ​​and the model correction function across all wavelength ranges.

[0018] In this technical solution, the optimal mapping relationship is obtained by solving the above objective function, and then the calculation results after the above model correction are obtained.

[0019] The present invention is further configured such that: the calculation of power generation at different wavelengths based on the external quantum efficiency of the photovoltaic cell includes: Based on the external quantum efficiency of photovoltaic cells and the key performance parameters set at the time of manufacture of solar cell modules, the spectral cell efficiency under standard test conditions is calculated; the spectral power density of the cell is then calculated, which is the product of the direct radiation intensity at that wavelength and the first empirical coefficient, plus the product of the scattered radiation intensity at that wavelength and the second empirical coefficient, and then multiplied by the spectral cell efficiency at that wavelength.

[0020] In this technical solution, two empirical coefficients (the first empirical coefficient and the second empirical coefficient) are used to measure the influence of direct radiation and diffuse radiation on the output power of photovoltaic cells, respectively. These empirical constants can be determined through multiple experimental measurements.

[0021] The present invention is further configured such that: step S4 includes the process of generating total power energy of a single photovoltaic module within a certain time range and spectral range: after multiplying the spectral power density per unit wavelength and per unit time by the module power conversion coefficient, the total energy density is obtained by first performing a double integration over the wavelength range and time range; then multiplying it by the ratio of the actual total area of ​​the photovoltaic module to the standard test area; and then multiplying it by the efficiency loss coefficient caused by surface contamination of the module to obtain the total power generation energy of a single module.

[0022] In this technical solution, the component power conversion coefficient represents the overall energy utilization efficiency of the component; the area ratio can extend the results under standard test conditions to the actual area of ​​the component; and the pollution loss coefficient can correct for the attenuation of energy output caused by dust, dirt, etc.

[0023] The present invention is further configured such that step S4 also includes the process of calculating the total power generation energy of the entire photovoltaic system: the total power generation energy of the entire photovoltaic system is equal to the product of the total power generation of a single component and the number of components, and then multiplied by an overall system efficiency coefficient.

[0024] In this technical solution, the overall system efficiency coefficient comprehensively considers the energy losses in the inverter, lines, cables, and other power conversion and transmission processes.

[0025] The present invention can bring the following beneficial effects: This application uses a method to predict the spectral distribution of solar radiation by variables such as clouds and water vapor, and then uses the conversion efficiency of photovoltaic panels for solar radiation of different wavelengths to make more precise predictions of photovoltaic output. Numerical models are used to dynamically downscale global meteorological forecast data to predict future atmospheric conditions, including temperature, pressure, clouds, and water vapor in the vertical layers, providing a forecast basis for subsequent calculations.

[0026] Using HITRAN (High Resolution Transmission Molecular Absorption Database) and radiation formulas, the absorption coefficients and scattering coefficients of different atmospheric components are calculated, thereby calculating the future optical thickness of the atmosphere and obtaining predictions of direct and scattered spectral radiation. Based on the predicted solar radiation spectrum distribution, the power generation of a single photovoltaic module and the power generation of the system are calculated. The algorithm has a sound scientific basis, clear implementation steps, and the results are more accurate than traditional algorithms. Attached Figure Description

[0027] Figure 1 This is a flowchart of steps S1 and S2 of a photovoltaic power generation forecasting method based on spectral forecasting according to this application.

[0028] Figure 2 This is a flowchart of steps S3 and S4 of a photovoltaic power generation forecasting method based on spectral forecasting according to this application. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0030] While solar power generation is growing rapidly, current grid connection rates remain limited. Due to the difficulty in accurately predicting power generation and its mismatch with electricity demand, some regions are experiencing difficulties in absorbing solar power, or even curtailment. The volatility, randomness, and intermittency of solar power generation are major obstacles to its large-scale grid integration. These characteristics not only affect the stability and security of the power grid but also increase reliance on backup energy storage. Therefore, improving the accuracy of solar power generation forecasting is fundamental to grid dispatching and planning, and is crucial for the stable operation of the grid under high penetration rates. By improving forecasting accuracy, grid dispatching can be effectively optimized, ensuring the safe and stable operation of the system and laying a solid technical foundation for future large-scale photovoltaic grid connection.

[0031] Solar power generation forecasting is primarily based on radiative transfer theory, combined with ground-based meteorological observations, upper-air atmospheric sounding, satellite remote sensing, and numerical simulations. It establishes solar radiation forecasting models by acquiring relevant parameters such as water vapor content, cloud cover, temperature, and humidity. In recent years, with the booming development of the solar energy industry and the widespread adoption of smart grids, the market demand for solar energy forecasting has been continuously increasing. Developed countries have established numerous photovoltaic forecasting laboratories, and solar energy forecasting technology has developed rapidly. In recent years, my country has also gradually established specialized solar energy forecasting research institutions and companies, and its technological level has steadily improved. Depending on the application scenario, solar radiation forecasting can be divided into various time scales. Ultra-short-term forecasts are mainly used for real-time dispatching and power generation management within the next 4 hours; short-term forecasts cover the next 72 hours, providing support for economic dispatching and system reserve arrangements; medium- and long-term forecasts are applicable to the next 1-2 weeks or even longer, and are of great significance for power plant planning, grid layout, and emergency event analysis.

[0032] Currently, solar radiation forecasting methods are mainly divided into three categories: first, statistical forecasting methods based on statistical regression and artificial intelligence, which utilize historical data and machine learning techniques for modeling and prediction; second, image-based forecasting methods based on satellite cloud images and ground-based sky imaging, which predict solar radiation by analyzing cloud dynamics and radiation variations; and third, physical methods based on numerical weather prediction, which simulate changes in solar radiation through atmospheric physical processes. Each of these three methods has its strengths and can provide effective forecast results at different time scales and application scenarios. Statistical methods perform better in short-term forecasts, image-based methods are more suitable for ultra-short-term forecasts, while physical methods are commonly used for medium- and long-term forecasts.

[0033] Photovoltaic power forecasts are divided into two categories: direct forecasts and indirect forecasts. Indirect forecasts first forecast solar irradiance and then obtain photovoltaic power through a photovoltaic-to-electricity model, while direct forecasts forecast photovoltaic power directly from changes in meteorological elements.

[0034] Currently, almost all solar radiation forecasting models predict total irradiance, and photovoltaic power forecasting models use relatively simple formulas to convert total irradiance to total power. However, these conversion processes only consider temperature and total irradiance, neglecting some details and leading to errors. In reality, photovoltaic panels made of different materials have varying conversion efficiencies for different wavelengths of solar radiation. Furthermore, different meteorological variables have varying effects on reducing solar radiation at different wavelengths. Therefore, forecasting only total irradiance may overlook many details, resulting in errors during variable conversion.

[0035] Example 1 To address the shortcomings of existing technologies, this embodiment proposes a photovoltaic power generation forecasting method based on spectral forecasting, referencing... Figure 1 and Figure 2 It mainly includes the following steps.

[0036] Step S1: First, download the global forecast data, then perform gridded encoding on the data; subsequently, use the WRF model for dynamic downscaling, and finally calculate the key meteorological variables for the future.

[0037] Step S1 specifically includes data acquisition and preprocessing, as well as dynamic downscaling and variable calculation.

[0038] Step S11: Download global forecast data, which includes several meteorological variables, and then convert the data into a standard format supported by WPS using a grid-based interpolation method.

[0039] After completing step S11, proceed to step S12, input the data processed by WPS into the WRF model, perform dynamic downscaling based on the WRF model, and calculate the key meteorological variables for the future period based on the WRF model.

[0040] In this technical solution, steps S11 and S12 mainly involve data acquisition and preprocessing, dynamic downscaling, and variable calculation. The calculated variables can provide a basis for subsequent calculations of optical thickness and spectral irradiance.

[0041] In this embodiment, the global forecast data is specifically downloaded from ECMWF (European Centre for Medium-Range Weather Forecasts). The global forecast data covers a variety of meteorological variables in the surface and pressure layers, such as temperature, humidity, air pressure, and wind speed.

[0042] In order for various meteorological variable data to be read by WPS (Weather Processing System), ECMWF data needs to be converted into gridded codes.

[0043] Specifically, the data from the surface layer and the pressure layer are converted into standard formats supported by WPS (such as GRIB or NetCDF format) through grid-based interpolation methods, ensuring that they meet the input requirements of the WRF model.

[0044] More specifically regarding step S12 above, the WRF model is an advanced numerical weather model used for weather forecasting and climate research. Using WRF for dynamic downscaling can refine large-scale global forecast information to a higher resolution.

[0045] In this embodiment, key meteorological variables for the next 240 hours can be calculated based on WRF simulations.

[0046] These key meteorological variables mainly include: Downward shortwave radiation (SWDOWN): The amount of shortwave radiation transmitted from the atmosphere to the ground. Water vapor mixing ratio (QVAPOR): The ratio of the mass of water vapor in the atmosphere to the mass of dry air; Cloud-water mixing ratio (QCLOUD): The ratio of the mass of liquid water cloud in the atmosphere to the mass of dry air; Cloud-ice mixing ratio (QICE): The ratio of the mass of solid cloud ice in the atmosphere to the mass of dry air; Cloud water particle concentration (QNCLOUD); Ice crystal particle concentration (QNICE); Disturbed atmospheric pressure (P): Pressure disturbance term for each layer; Reference pressure (PB): The actual air pressure of each layer can be calculated by summing the perturbed atmospheric pressure and the reference pressure.

[0047] Temperature perturbation term (T): The temperature perturbation term for each layer in the WRF model. The actual temperature of each layer can be obtained by summing the temperature perturbation term with the normal background temperature, which is represented by T. 00 .

[0048] After completing step S1, proceed to step S2, using the spectral absorption and scattering coefficients of different gas components in the HITRAN database to calculate the spectral optical thickness of each atmospheric component, and obtain the direct and scattered spectral irradiance based on the spectral optical thickness.

[0049] More specifically, the spectral optical thickness of each atmospheric component is calculated by combining the spectral absorption and scattering coefficients of the gas components and using the key meteorological variables predicted in step S1.

[0050] The process of obtaining the spectral absorption and scattering coefficients of the above-mentioned different gas components specifically includes the calculation of the absorption and scattering coefficients of water vapor and water clouds or ice clouds.

[0051] The absorption coefficient of water vapor was obtained by searching and calculating using the HITRAN database. Since the absorption of radiation by water vapor molecules is much greater than that of scattering, the effect of scattering by water vapor molecules on solar shortwave radiation can be ignored within the scope of this scheme. Therefore, only the absorption effect is considered here.

[0052] The absorption and scattering coefficients of water clouds or ice clouds were calculated based on Mie scattering theory.

[0053] The HITRAN database is a high-resolution transport molecule absorption database.

[0054] Specifically, the water vapor absorption coefficient can be calculated using HITRAN. The HITRAN database can be obtained by installing the HAPI package in Python. The HAPI package is an interactive tool for HITRAN applications, and the `hapi.absorptionCoefficient_Voigt` function can calculate the absorption coefficient. In this function, the Diluent variable can adjust the amount of water vapor input: Diluent is represented as {'air':f air ,'H2O':f H2O}, where the water vapor volume fraction f H2O and air volume fraction f air It can be calculated using the variables QVAPOR and the molecular weight of water vapor and the average molecular weight of dry air.

[0055] Liquid water droplets and ice crystals in water clouds or ice clouds have different optical properties than water vapor, and their absorption and scattering coefficients can be calculated using Mie scattering theory.

[0056] Absorption coefficient β a This can be expressed as the absorption efficiency factor Q. a Multiply by three-quarters, then multiply by the air density ρ air The product of r and Q is ultimately divided by r. eff Where Q refers to the mixing ratio output in WRF mode (QCLOUD for cloud and water, QICE for ice and cloud), in kg / kg; r eff Let r be the effective radius. eff Expressed as 3Q divided by 4πρ w N w Then take the cube root of this result. Where ρ w It is the density of liquid water or ice, N. w This refers to particle concentration, specifically QNCLOUD (cloud droplet particle concentration) and QNICE (ice crystal particle concentration) in the WRF output variables, with units of kg. -1 .

[0057] Similarly, the scattering coefficient β s This can be expressed as the scattering efficiency factor Q. s Multiply by three-quarters, then multiply by the air density ρ air The product of r and Q is ultimately divided by r. eff In addition, the air density ρ in each layer air The gas constant of dry air can be obtained by dividing the air pressure P′ of the layer by the first coefficient, which is the gas constant of dry air multiplied by the air temperature of the layer, and then multiplied by the sum of the variable QVAPOR (0.61 times) and 1. The value of the gas constant of dry air is approximately 287 J / (kg·K).

[0058] The absorption and scattering efficiency factors Q in the above calculations a and Q s The values ​​depend on the particle size and wavelength and can be obtained from tables or models in Mie scattering theory. In this scheme, the values ​​are obtained from the PyMieScatt package in Python, and the PyMieScatt.MieQ function is used for calculation.

[0059] For other atmospheric gases, since their content does not change much, they are considered constant in the short term in this embodiment. The calculation is performed using mid-latitude continental standard vertical distribution data, and the calculation method is similar to that for water vapor, using the HITRAN dataset.

[0060] The calculation process for the spectral optical thickness of each atmospheric component mainly includes the calculation process for the optical thickness of the entire atmosphere and the calculation process for the scattering optical thickness.

[0061] The calculation process for the optical thickness of the entire atmosphere is as follows: from the ground to the top of the atmosphere, the absorption coefficient and scattering coefficient at each altitude are added together and then integrated.

[0062] The calculation process for the scattering optical thickness is as follows: consider the scattering coefficient at each altitude or pressure layer and integrate it in the altitude direction, or multiply the scattering coefficient of each layer by its physical thickness and sum them in discrete layers.

[0063] In this technical solution, the optical thickness of the entire atmosphere can be obtained based on the calculated future absorption and scattering coefficients of each atmospheric component.

[0064] Wherein, the total optical thickness τ of the entire atmosphere at a certain wavelength total，λ The optical thickness of the atmosphere is the sum of the absorption and scattering coefficients of each layer along a vertical path from the ground to the top of the atmosphere, integrated along the height direction. This integral can be discretized into a summation of multiple pressure layers. That is, at each pressure level, the sum of the absorption and scattering coefficients of that layer is multiplied by the physical thickness of that layer, and then the results of all layers are added together to obtain the total optical thickness of the entire atmosphere.

[0065] It is worth noting that although the total optical thickness takes into account both absorption and scattering processes, solar radiation does not completely disappear in the atmosphere after scattering. It may continue to propagate and reach the ground through multiple scattering and refractions. This portion of solar radiation that reaches the ground is called "diffuse horizontal irradiance" (DHI), and its effects can be described by the diffuse optical thickness.

[0066] The spectral irradiance of direct radiation can be calculated by multiplying the radiation intensity of the top of the atmosphere at that wavelength by a first exponential attenuation factor, which is determined by the total optical thickness of the atmosphere at that wavelength.

[0067] The calculation of the spectral irradiance of scattered radiation can be expressed as the original radiative intensity at the top of the atmosphere multiplied by a second exponential attenuation factor, and then multiplied by an empirical scattering factor; the second exponential attenuation factor is determined by the scattering optical thickness at that wavelength.

[0068] In this technical solution, the first exponential decay factor is exp(-τ) total，λ The intensity of direct radiation is equal to the portion of the original radiation intensity that is reduced due to absorption and scattering.

[0069] In this technical solution, the second exponential decay factor is 1 minus exp(-τ) s,λ ).

[0070] The basis of radiative transfer is Beer's Law, which states that at a given wavelength, radiation incident from the top of the atmosphere will have a radiant intensity I after passing through the atmosphere. λ It will gradually attenuate due to absorption and scattering. After passing through an atmosphere of thickness z, the radiation intensity will decrease to the original intensity I. 0,λ Multiplied by a factor exp(-τ) that decreases with increasing optical thickness. λ，z ). I λ，z τ is the radiation intensity of wavelength λ after passing through an atmospheric layer of thickness z. λ，z It refers to the optical thickness at wavelength λ, after passing through the physical thickness z of the atmosphere, or the degree of attenuation.

[0071] Considering the different ways radiation reaches the ground and the varying conversion efficiencies of photovoltaic panels, we divide the total radiation into direct radiation and scattered radiation. Direct radiation is weakened by absorption and scattering, therefore the optical thickness is the total optical thickness τ. total，λ Scattered radiation is mainly affected by scattering, therefore only τ was used. total，λ f here diff It is the empirical scattering coefficient, which is usually between 0.1 and 0.3.

[0072] The spectral irradiance of total radiation can be written as the sum of the two parts mentioned above (i.e., the direct part and the scattered part).

[0073] After completing all the calculations in step S2, proceed to step S3, which uses a solar spectroradiometer to verify the calculation results and perform artificial intelligence correction. Specifically, based on the deviation between the actual spectral radiation measured by the solar spectroradiometer and the calculated radiation value, an artificial neural network model is constructed to correct this deviation. The artificial neural network model is implemented by optimizing an objective function, which is specifically expressed as: minimizing the sum of errors between the observed values ​​and the model correction function across all wavelength ranges.

[0074] In this technical solution, the optimal mapping relationship is obtained by solving the above objective function, and then the calculation results after the above model correction are obtained.

[0075] More specifically, a solar spectroradiometer measures the intensity of solar spectral radiation from the ultraviolet to the near-infrared band, typically covering the spectral range of 280–4000 nm. Since common photovoltaic panels generate photoelectric effects within the range of 300–1200 nm, this range can be fully covered. The 300–1200 nm spectrum measured can be used to verify calculation results and for artificial intelligence correction. For a given wavelength, the difference (i.e., deviation) between the actual measured spectral radiation and the calculated radiation value can be defined as the difference between the two. This deviation can serve as the basis for constructing an artificial neural network model for correction.

[0076] To achieve the correction objective, it is done by optimizing the aforementioned objective function, which is typically solved within the wavelength range of 300 nm to 1200 nm. Here, the observed values... This refers to the spectral radiation intensity measured by a spectroradiometer, while the calculated value... This is the result obtained from the model calculations in the aforementioned steps. The purpose of the correction function is to optimize and adjust the original calculation results to make them closer to the measured values.

[0077] After completing step S3, proceed to step S4 to calculate the power generation at different wavelengths based on the external quantum efficiency of the photovoltaic cell; then calculate the final power generation by integration and combining it with the losses of other parts of the photovoltaic module.

[0078] The calculation of power generation at different wavelengths includes the following process: The spectral cell efficiency η is calculated under standard test conditions using the external quantum efficiency of photovoltaic cells and key performance parameters set at the time of manufacture of solar cell modules. λ，STC The external quantum efficiency EQE is equal to the charge q and the open-circuit voltage V. ocThe product of the photon energy and the fill factor FF is then divided by the energy of the photon at that wavelength. The photon energy is determined by Planck's constant h and the corresponding optical frequency ν. ph The product determines the outcome.

[0079] The variables mentioned above are explained in detail below: η λ，STC EQE represents the spectral cell efficiency at a specific wavelength λ under Standard Test Conditions (STC); EQE is an abbreviation for External Quantum Efficiency, which describes the number of electron-hole pairs generated per incident photon per unit time at a specific wavelength. q is the elementary charge of an electron, approximately 1.602 × 10⁻⁶. -19 Kulun; V oc It is the open-circuit voltage, an important parameter of the photovoltaic cell itself; FF stands for Fill Factor, which is the ratio of the maximum output power of a solar cell when it is working under illumination to the product of the short-circuit current and open-circuit voltage under ideal conditions. It reflects the non-ideal nature of the cell. h is Planck's constant, approximately 6.626 × 10⁻⁶. -34 Joule second; ν ph It is the frequency of the incident photon, which is related to the wavelength of the incident photon, ν. ph = c / λ, where c is the speed of light 3.00 × 10⁻⁶ 8 rice.

[0080] After determining the spectral cell efficiency, the cell spectral power density is calculated. Specifically, the cell spectral power density is expressed as the product of the direct radiation intensity at that wavelength and the first empirical coefficient, plus the product of the scattered radiation intensity at that wavelength and the second empirical coefficient, and then multiplied by the spectral cell efficiency at that wavelength.

[0081] In this technical solution, two empirical coefficients (the first empirical coefficient and the second empirical coefficient) are used to measure the influence of direct radiation and diffuse radiation on the output power of photovoltaic cells, respectively. These empirical constants can be determined through multiple experimental measurements.

[0082] In this technical solution, the external quantum efficiency can be obtained from literature. The key new energy parameters set by the solar cell module at the time of manufacture include open circuit and fill factor.

[0083] After obtaining the battery spectral power density, the total power generation of a single photovoltaic module within a certain time range and spectral range is calculated based on the battery spectral power density. The specific calculation process is as follows: After multiplying the spectral power density per unit wavelength and per unit time by the module power conversion factor, the total energy density is obtained by double integration over the wavelength range and time range. Then, it is multiplied by the ratio of the actual total area of ​​the photovoltaic module to the standard test area. Finally, it is multiplied by the efficiency loss coefficient caused by surface contamination of the module to obtain the total power generation energy of a single module.

[0084] In this technical solution, the component power conversion coefficient represents the overall energy utilization efficiency of the component; the area ratio can extend the results under standard test conditions to the actual area of ​​the component; and the pollution loss coefficient can correct for the attenuation of energy output caused by dust, dirt, etc.

[0085] After obtaining the total power generation energy of a single photovoltaic module within a certain time range and spectral range, the total power generation energy of the entire photovoltaic system is calculated using this information. The calculation process is as follows: the total power generation energy of the entire photovoltaic system is equal to the product of the total power generation of a single module and the number of modules, and then multiplied by an overall system efficiency coefficient.

[0086] In this technical solution, the overall system efficiency coefficient comprehensively considers the energy losses in the inverter, lines, cables, and other power conversion and transmission processes.

[0087] This embodiment proposes a photovoltaic power generation forecasting method based on spectral forecasting. Instead of directly forecasting the total irradiance, it forecasts the radiation spectrum at future times based on forecasted meteorological variables (water content in the air, water cloud amount, and ice cloud amount), and converts the radiation into power generation by spectrum, thus achieving refined conversion of light power.

[0088] This embodiment innovatively proposes a method for forecasting the spectral distribution of solar radiation using variables such as clouds and water vapor, and then refines the forecast of photovoltaic power output based on the conversion efficiency of photovoltaic panels for solar radiation of different wavelengths. In the first stage of the algorithm, a numerical model is used to dynamically downscale global meteorological forecast data to predict future atmospheric conditions, including temperature, air pressure, clouds, water vapor, etc., in the vertical layers, providing a forecasting basis for subsequent calculations. In the second stage, the HITRAN (High Resolution Transmission Molecular Absorption Database) and radiation formulas are used to calculate the absorption and scattering coefficients of different atmospheric components, thereby calculating the future atmospheric optical thickness and obtaining forecasts of direct and scattered spectral radiation. In the third stage, based on the predicted solar radiation spectral distribution, the photovoltaic power generation of a single module and the system power generation are calculated. This algorithm has a sound scientific basis, clear implementation steps, and higher accuracy than traditional algorithms.

[0089] The technical terms used in this embodiment are explained below.

[0090] Optical Depth / Optical Thickness: Optical thickness describes the degree of attenuation of light as it passes through a medium due to absorption and scattering. It reflects the amount by which radiation intensity decreases after traversing a certain path. A greater optical thickness indicates more severe absorption and scattering of light as it passes through the atmosphere. In photovoltaic models, it is used to predict the intensity of light reaching the ground under different weather conditions.

[0091] Absorption coefficient: The absorption coefficient describes a material's ability to absorb light of a specific wavelength; that is, how much light energy is absorbed when light passes through a unit length of medium. In the atmosphere, different components (such as water vapor and carbon dioxide) have different absorption coefficients for different wavelengths of solar radiation, resulting in the weakening of some radiant energy. These atmospheric components have different absorption capacities for different wavelengths of radiation, which is reflected in the differences in their spectral absorption coefficients.

[0092] Scattering Coefficient: The scattering coefficient describes the ability of particles in a medium to scatter incident light in different directions. The size and shape of different particles affect the scattering intensity. Scattering affects the direction of light propagation but does not consume light energy. The presence of scattered light is an important factor affecting indirect radiation (DHI) in photovoltaic power generation.

[0093] External quantum efficiency (EQE) describes the efficiency with which a photovoltaic cell converts each photon absorbed at a given wavelength into an electron-hole pair. It is used to evaluate the response capability of a photovoltaic cell at different wavelengths. A higher EQE means that a photovoltaic cell can generate more charge carriers under illumination at a specific wavelength. The EQE of photovoltaic cells made of different materials varies at different wavelengths; for example, materials such as amorphous silicon (a-Si) and cadmium telluride (CdTe) have different response bands.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A photovoltaic power generation forecasting method based on spectral forecasting, characterized in that, Includes the following steps: S1. Download global forecast data and perform gridded encoding on the data; use the WRF model to perform dynamic downscaling and calculate key future meteorological variables; S2. Using the spectral absorption and scattering coefficients of different gas components in the HITRAN database, calculate the spectral optical thickness of each atmospheric component, and obtain the direct and scattered spectral irradiance based on the spectral optical thickness. S3 utilizes a solar spectroradiometer to verify calculation results and perform artificial intelligence correction. S4 calculates the power generation at different wavelengths by combining the external quantum efficiency of the photovoltaic cell; the final power generation is calculated by integration and by combining the losses of other parts of the photovoltaic module.

2. The photovoltaic power generation forecasting method based on spectral forecasting according to claim 1, characterized in that, Step S1 includes: S11, download global forecast data including several meteorological variables, and then convert the data into a standard format supported by WPS using a grid-based interpolation method; S12, the data processed by WPS is input into the WRF model, the WRF model is used for dynamic downscaling, and the key meteorological variables for the future period are calculated based on the WRF model.

3. A photovoltaic power generation forecasting method based on spectral forecasting according to claim 1 or 2, characterized in that, The utilization of spectral absorption and scattering coefficients of different gas components from the HITRAN database includes: The water vapor absorption coefficient was obtained by searching and calculating the HITRAN database; The absorption and scattering coefficients of water clouds or ice clouds were calculated using Mie scattering theory.

4. The photovoltaic power generation forecasting method based on spectral forecasting according to claim 3, characterized in that, The calculation of the spectral optical thickness of each atmospheric component includes calculating the optical thickness of the entire atmosphere and the scattering optical thickness. The optical thickness of the entire atmosphere is calculated by integrating the absorption coefficient and scattering coefficient at each altitude from the ground to the top of the atmosphere. The scattering optical thickness is calculated by considering the scattering coefficient at each altitude or pressure layer and integrating it in the altitude direction, or by multiplying the scattering coefficient of each layer by its physical thickness and summing them in discrete layers.

5. The photovoltaic power generation forecasting method based on spectral forecasting according to claim 4, characterized in that, The spectral irradiance of the direct radiation is expressed as: The radiation intensity at the top of the atmosphere at that wavelength is multiplied by a first exponential attenuation factor, which is determined by the total optical thickness of the atmosphere at that wavelength.

6. The photovoltaic power generation forecasting method based on spectral forecasting according to claim 4, characterized in that, The spectral irradiance of the scattered radiation is expressed as: The original radiation intensity at the top of the atmosphere is multiplied by a second exponential attenuation factor, and then multiplied by an empirical scattering factor; the second exponential attenuation factor is determined by the scattering optical thickness at that wavelength.

7. A photovoltaic power generation forecasting method based on spectral forecasting according to claim 1 or 2, characterized in that, Step S3 includes: The deviation between the actual measured spectral radiation and the calculated radiation value of the solar spectroradiometer is obtained. This deviation is corrected by constructing an artificial neural network model. The artificial neural network model is achieved by optimizing an objective function, which is to minimize the sum of errors between the observed values ​​and the model correction function across all wavelength ranges.

8. The photovoltaic power generation forecasting method based on spectral forecasting according to claim 1, characterized in that, The calculation of power generation at different wavelengths based on the external quantum efficiency of photovoltaic cells includes: Based on the external quantum efficiency of photovoltaic cells and the key performance parameters set at the time of manufacture of solar cell modules, the spectral cell efficiency under standard test conditions is calculated; the spectral power density of the cell is then calculated, which is the product of the direct radiation intensity at that wavelength and the first empirical coefficient, plus the product of the scattered radiation intensity at that wavelength and the second empirical coefficient, and then multiplied by the spectral cell efficiency at that wavelength.

9. A photovoltaic power generation forecasting method based on spectral forecasting according to claim 1 or 8, characterized in that, Step S4 includes the process of total power generation energy of a single photovoltaic module within a certain time and spectral range: after multiplying the spectral power density per unit wavelength and per unit time by the module power conversion coefficient, the total energy density is obtained by double integration over the wavelength range and the time range. Then multiply it by the ratio of the actual total area of ​​the photovoltaic module to the standard test area; Multiply by the efficiency loss coefficient caused by surface contamination of the component to obtain the total power generation of a single component.

10. A photovoltaic power generation forecasting method based on spectral forecasting according to claim 9, characterized in that, Step S4 also includes the process of calculating the total power generation energy of the entire photovoltaic system: the total power generation energy of the entire photovoltaic system is equal to the product of the total power generation of a single component and the number of components, and then multiplied by an overall system efficiency coefficient.

Images