A photovoltaic power generation power inversion method based on spectral composition

Abstract

The application discloses a photovoltaic power inversion method based on spectral components, which firstly quantifies the light energy conversion efficiency of wavelength-selective photovoltaic units through a spectral weighted fusion model; secondly, realizes accurate inversion of solar irradiance based on a double-signal self-compensation mechanism; then constructs a hierarchical Bayesian generation model to quantize the full-chain uncertainty of irradiance and the photovoltaic power derived therefrom; finally, generates a power inversion sequence and its confidence interval, and provides conditional prediction for missing periods, forming a complete power generation performance evaluation report. The method is particularly suitable for scenes where spectral dynamic changes are obvious, meteorological data is missing or historical records are incomplete, and can effectively improve the power supply reliability and operation stability of the photovoltaic power generation system, providing reliable data support for power scheduling, energy storage configuration and grid interaction of photovoltaic power stations.

Description

Technical Field

[0001] This invention relates to a method for inverting photovoltaic power generation based on spectral composition, belonging to the field of photovoltaic power generation technology. Background Technology

[0002] Currently, photovoltaic (PV) power prediction or historical power inversion mainly relies on meteorological station observation data or the PV system's own output records. However, in remote areas or scenarios with weak data infrastructure, problems such as missing meteorological data, insufficient spatiotemporal resolution of satellite irradiance products, or incomplete historical power records often arise, making it difficult to accurately reconstruct the actual power generation sequence of the PV system. Furthermore, in systems employing spectrally selective cells or affected by complex surface environments, dynamic changes in the incident spectrum significantly alter the spectral distribution and irradiance reaching the PV cells. Traditional power models typically treat solar radiation as uniform white light, failing to fully consider the coupling mechanism between spectral composition and cell photoelectric conversion efficiency, thus limiting estimation accuracy in environments with dynamically changing spectral distributions. Therefore, a new method is urgently needed that can use spectral composition as a key input, integrating environmental parameters and probabilistic modeling to achieve highly reliable inversion of illumination conditions and PV power generation. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a photovoltaic power generation inversion method based on spectral composition, which takes spectral composition as input and achieves high-reliability inversion of illumination conditions and photovoltaic power generation by integrating environmental parameters and probabilistic modeling.

[0004] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0005] A method for inverting photovoltaic power generation based on spectral composition includes the following steps:

[0006] S1. This method achieves quantitative evaluation and system optimization of light energy conversion efficiency by coupling spectral response characteristics with the spectral transmission characteristics of photovoltaic units. The specific process is as follows: First, the spectral response curves under blue, green, and red monochromatic light are obtained as reference photoelectric parameters. Second, the spectral photon flux distribution arriving at the photovoltaic unit is accurately calculated by combining the radiative transfer model with the spectral transmission curve of the wavelength-selective photovoltaic unit. Then, the reference photoelectric parameters are spectrally weighted and fused using the light intensity weights of the blue, green, and red light bands. The spectral conversion efficiency of the photovoltaic unit under the corresponding spectral environment is directly predicted through the spectral weighted fusion model. At the same time, a spectral composition efficiency index based on the photovoltaic spectral efficiency curve is introduced to evaluate the photoelectric effectiveness of the incident spectrum.

[0007] S2. Acquire the generated light energy data; input the light energy data signal into the pre-constructed irradiance inversion model, and calculate the predicted solar irradiance value by combining it with the pre-stored calibration parameters; wherein, the irradiance inversion model is configured to: dynamically compensate the irradiance-sensitive main sensing signal according to the coupling relationship between at least two internal signals with different response characteristics parsed from the original light energy data signal, so as to eliminate the measurement error caused by changes in ambient temperature; output the predicted solar irradiance value;

[0008] S3. Obtain historical solar irradiance data and corresponding auxiliary environmental data; train a hierarchical Bayesian generative model, which takes the predicted solar irradiance value and auxiliary environmental data as input and can output the posterior probability distribution of solar irradiance; propagate the posterior distribution of irradiance forward through a photovoltaic physical model that incorporates inverter power constraints to calculate the probability distribution of photovoltaic system output power and overload clipping loss, thereby realizing the quantification of uncertainty in the entire chain from irradiance to photovoltaic power;

[0009] S4. Based on the power inversion sequence of the photovoltaic system, obtain the photovoltaic power inversion curve, error confidence interval and missing measurement period supplement report, and complete the photovoltaic power generation inversion.

[0010] Preferably, S1 is implemented through the following steps:

[0011] (1) Dynamic weighted fusion of photoelectric parameters based on spectral distribution: First, the baseline spectral response parameters under pure blue light (B: 400-500 nm), pure green light (G: 500-600 nm), and pure red light (R: 600-700 nm) conditions are obtained by measurement or modeling, including: maximum spectral conversion efficiency. Initial spectral response coefficient and base loss coefficient ,in, .

[0012] Secondly, regarding the actual spectral photon flux density distribution at the photovoltaic unit in a wavelength-selective photovoltaic (WSPV) system. Calculate the intensity weights of the blue, green, and red light bands. The calculation formula is:

[0013] ,

[0014] in, Indicates band The light intensity weighting coefficient, and They represent the bands respectively. The minimum and maximum wavelength boundaries, The integral in the denominator represents the spectral photon flux density distribution at the photovoltaic unit, and the total photon flux of the entire spectrum utilizing the effective radiation band represents the total photon flux. For band The spectral conversion efficiency function is obtained through photovoltaic unit calibration; This is a non-linear adjustment parameter, with a value range of [value range missing]. It is used to adjust the weight allocation strategy for each band.

[0015] Subsequently, the core innovative formula—spectrally weighted photoelectric parameter synthesis—is executed, dynamically fusing the monochromatic light photoelectric parameters with real-time spectral weights to generate equivalent photoelectric parameters corresponding to the current composite spectrum:

[0016] ,

[0017] ,

[0018] ,

[0019] in, and These represent the equivalent maximum spectral conversion efficiency, equivalent initial spectral response coefficient, and equivalent basic loss coefficient synthesized under the current composite spectrum, respectively. and These represent the photoelectric parameters of the photovoltaic unit measured under pure monochromatic light (blue, green, and red), respectively.

[0020] (2) Prediction of spectral conversion efficiency and quantitative evaluation of spectral composition efficiency: Using the equivalent photoelectric parameters synthesized in (1), the spectral conversion efficiency of the photovoltaic unit under WSPV spectral conditions is calculated through the photoelectric response model. .

[0021] Meanwhile, in order to directly evaluate the impact of spectral composition on light energy conversion efficiency, and going beyond the traditional spectral efficiency index, the following innovative formula for quantifying spectral composition efficiency is applied for calculation:

[0022] ,

[0023] ,

[0024] in, and These represent effective spectral flux and spectral utilization effective radiation, respectively. This represents the spectral photon flux density distribution; The curve representing the relative spectral response coefficients of the spectral conversion; The spectral curve representing the relative effect of spectral conversion.

[0025] Preferably, S2 is implemented through the following steps:

[0026] Process the raw light energy data signal to obtain the first signal value. With the second signal value , It has an approximately linear relationship with the incident irradiance. It is strongly correlated with ambient temperature; it calls pre-stored calibration parameters, including reference irradiance. First reference signal value Second reference signal value and compensation coefficient Based on the first signal value Second signal value And the calibration parameters, and calculate the intermediate irradiance estimate using the first relational formula. :

[0027] ,

[0028] in, This represents the intermediate irradiance estimate (W / m²) based on a linear proportional relationship. This represents the known reference irradiance value (W / m²) under standard test conditions. This indicates the first signal value acquired. Indicates that at an irradiance of The first signal reference value is obtained under the calibration conditions; subsequently, based on the second signal value... Second reference signal value and compensation coefficient The dynamic compensation factor is calculated using the second relational formula. :

[0029] ,

[0030] in, This represents the dynamic compensation factor (dimensionless) used to correct for the effects of temperature. The compensation coefficient represents the degree to which the change in the second signal affects the photoelectric conversion efficiency. This indicates the acquired second signal value. This represents the second signal reference value obtained under calibration conditions; finally, it is obtained by combining intermediate irradiance estimates. With the dynamic compensation factor The predicted solar irradiance is calculated and output using the third relation. :

[0031] ,

[0032] The first, second, and third relational expressions together constitute an integrated irradiance inversion model. This model achieves self-compensation by using two signals with different physical responses obtained from the photosensitive measurement unit, thus enabling accurate inversion of solar irradiance without relying on an independent external temperature sensor.

[0033] Preferably, S3 is implemented through the following steps:

[0034] First, collect historical solar irradiance observations. and corresponding auxiliary environment data (Including temperature, humidity, soil moisture, surface reflectance, and aerosol optical thickness). Construct a hierarchical Bayesian generative model whose optimization objective is to maximize the following joint variational lower bound:

[0035] ,

[0036] in, As a potential variable characterizing the spatiotemporal features of irradiance; For encoder, For likelihood models; and The parameters to be optimized are: the first term is the inversion loss, the second term is the KL regularization term, and the third term... For adaptive regularization, its form is:

[0037] ,

[0038] in, For the model at time The prediction variance The weight function is monotonically decreasing. and These are the first-order gradient operator and the second-order difference operator in time, respectively. To control sparsity, use hyperparameters. Train the model using historical data (e.g., the previous 90 days) and optimize the parameters until convergence.

[0039] After the model optimization converges, the posterior uncertainty of the latent variables is refined using the following covariance update formula:

[0040] ,

[0041] in, For the likelihood model in the optimal estimation of latent variables Jacobian matrix at the location, To observe the noise covariance matrix, Let be the prior covariance matrix.

[0042] Finally, the refined posterior distribution is obtained. By distribution Mid-sampling and using photovoltaic physical models Forward calculations were performed, and the photovoltaic AC power output was statistically obtained using the Monte Carlo method. and overload clipping loss The expected value, variance, and confidence interval.

[0043] Preferably, S4 is implemented through the following steps:

[0044] Based on the posterior distribution obtained from S3 The sampled power sequence photovoltaic units generate a probabilistic inversion curve of the historical power of the photovoltaic system and its dynamic error confidence interval. For the missing periods in the historical data, a pre-trained hierarchical Bayesian generative model is invoked, using environmental data from the previous and subsequent valid periods as conditions, to generate conditional prediction distributions of irradiance and power output for the missing periods. Finally, a comprehensive report is output, including the power inversion sequence, dynamic confidence interval, power compensation values ​​for the missing periods, as well as their uncertainty measures and credibility ratings, for use in photovoltaic system power generation accounting and performance reliability analysis.

[0045] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0046] 1. The power inversion method proposed in this invention comprehensively utilizes spectral analysis, photoelectric response modeling, irradiance inversion and probabilistic uncertainty propagation to achieve high-precision inversion of photovoltaic power generation and supplementation of missing data in complex spectral environments.

[0047] 2. The power inversion method proposed in this invention is particularly suitable for scenarios with significant dynamic changes in the spectrum, missing meteorological data, or incomplete historical records, and can effectively improve the reliability of photovoltaic system monitoring, evaluation, and optimized operation. Attached Figure Description

[0048] Figure 1 This is a flowchart of the photovoltaic power generation inversion method based on spectral composition according to the present invention;

[0049] Figure 2 This is a comparison chart of the inversion power and measured power of the method proposed in this invention and the traditional method on a typical working day. Detailed Implementation

[0050] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0051] This embodiment uses a rooftop distributed photovoltaic (PV) system on a public building in a city as the implementation object. The total installed capacity of the system is 120kW, and it uses polycrystalline silicon PV units. Because the rooftop PV array is arranged at multiple azimuth angles, and surrounding buildings intermittently shade some units during the morning and evening, this scenario represents a typical complex urban spectral and shading environment. This embodiment aims to fully demonstrate the high-precision inversion process of historical PV power generation under conditions of missing meteorological data, dynamic spectral changes, and local shading. The following will strictly adhere to... Figure 1 The core steps shown are explained in detail one by one in the implementation process of this embodiment, and are accompanied by corresponding formulas and exemplary parameter values.

[0052] Step 1: Spectral analysis and quantitative evaluation of light energy conversion efficiency

[0053] The core of this step is to dynamically evaluate the photoelectric conversion efficiency of the photovoltaic unit under the current spectral conditions based on real-time spectral composition. First, acquire spectral observation data arriving at the photovoltaic unit surface within the target time period (e.g., 15 consecutive days), with a time resolution of 1 hour. Data sources can include ground-based spectrometers or atmospherically corrected satellite spectral products. Simultaneously, acquire the baseline photoelectric parameters of the polycrystalline silicon photovoltaic unit under monochromatic light irradiation in pure blue (400-500nm), pure green (500-600nm), and pure red (600-700nm) wavelengths. Assume the following reference values ​​are obtained through laboratory measurements: the maximum spectral conversion efficiency in the blue light band. Initial spectral response coefficient Basic loss coefficient The parameters corresponding to the green light band are as follows: , , The red light band is , , .

[0054] For a specific moment (e.g., 10:00 on a certain day), based on the measured or inverted spectral photon flux density distribution... First, calculate the intensity weights for the blue, green, and red bands. Assume that the calculation yields the following result: , , .

[0055] Subsequently, the monochromatic light reference parameters are dynamically fused with real-time weights to generate equivalent photoelectric parameters corresponding to the current composite spectrum, and spectral weighted photoelectric parameter synthesis is performed:

[0056] ,

[0057] ,

[0058] ,

[0059] To further quantify spectral effectiveness, a spectral composition effectiveness assessment is introduced. The effective spectral flux (ESF) and spectral utilized effective radiation (SUER) are calculated. The calculated ESF and SUER values ​​can be compared with historical data to evaluate the "quality" of the current spectrum. This step then outputs the equivalent photoelectric parameters considering the actual spectral components at the current moment. And spectral performance indicators (ESF, SUER) provide a physical basis for subsequent accurate inversion of irradiance and power.

[0060] Step 2: Dual-signal self-compensated irradiance inversion

[0061] This step aims to achieve high-precision and highly stable solar irradiance inversion using the dual-channel signals within the photovoltaic sensing unit without the need for an external temperature sensor. The system utilizes a dual-channel photodetector integrated within the monitoring unit to simultaneously acquire two channels of raw solar energy data signals: the first channel signal... Sensitive to the 400-700nm visible light band, its intensity is approximately linearly related to the incident irradiance; second signal It is sensitive to thermal radiation, and its intensity is strongly correlated with ambient temperature and the temperature rise of the photovoltaic unit. Assuming a reference value is obtained under standard test conditions (STC: irradiance 1000 W / m², temperature 25°C): Reference Irradiance First reference signal value (Counting), second reference signal value (Counting), and compensation coefficient .

[0062] At a certain real-time moment, the real-time signal value is acquired: , First, calculate the intermediate irradiance estimate based on linear scaling. :

[0063] ,

[0064] Next, the dynamic compensation factor is calculated. Used to correct for increases in temperature (manifested as) Signal drift caused by (increased)

[0065] ,

[0066] Finally, the corrected predicted solar irradiance value was obtained through compensation calculations. :

[0067] ,

[0068] This model effectively eliminates measurement errors caused by temperature variations through an internal dual-signal self-compensation mechanism. This step outputs an accurate solar irradiance time series that eliminates major temperature interference. This serves as the core input for subsequent power modeling.

[0069] Step 3: Hierarchical Bayesian Generative Model and Uncertainty Quantification

[0070] This step transforms the determined irradiance values ​​into a probability distribution and propagates it to the final power output through a physical model, achieving full-chain uncertainty quantification from irradiance to photovoltaic power. First, historical solar irradiance observations are collected. and corresponding auxiliary environment data (Including temperature, humidity, aerosol optical thickness, etc.).

[0071] Model Building and Training: Construct a hierarchical Bayesian generative model whose optimization objective is to maximize the following joint variational lower bound:

[0072] ,

[0073] Among the regularization terms The form is:

[0074] ,

[0075] Posterior Uncertainty Refinement: After model optimization convergence, the posterior covariance of the latent variables is refined using the covariance update formula.

[0076] ,

[0077] in, .

[0078] Power uncertainty propagation:

[0079] Next, uncertainty propagation and power calculation are performed, starting from the distribution. Medium sampling Each time, one latent variable is obtained from the sampling. Then, a possible irradiance sequence is obtained through a likelihood model. . Irradiance sequence Input photovoltaic physical model incorporating inverter constraints The model, combined with the equivalent photoelectric parameters obtained in the first step, calculates the DC power. Then consider the inverter efficiency curve and rated maximum AC power To obtain the AC output power And calculate the overload clipping loss. The photovoltaic AC output power was obtained by statistically analyzing all the sampling results. and overload clipping loss The probability distribution is calculated, including its expected value, variance, and 95% confidence interval. This step outputs a probabilistic inversion sequence of the photovoltaic system's output power, with a dynamic confidence interval for each time point, quantifying the uncertainty of the inversion results.

[0080] Step 4: Power sequence inversion and integrated report generation

[0081] This step integrates all intermediate results to generate a comprehensive report that can be directly used for operation and maintenance analysis. First, power sequence inversion is performed: based on the power probability distribution output in the third step, its expected value is taken as the best estimated power sequence, and at the same time, dynamic error confidence intervals for each time point are generated (e.g., 95% confidence interval bands formed by the upper and lower 2.5% quantiles).

[0082] For missing time periods in historical data (e.g., no data available from 2-4 PM on a certain afternoon due to equipment failure), missing data is filled in by calling a pre-trained Bayesian generative model and using environmental data from valid time periods before and after the missing time periods. Given the conditions, the model generates conditional prediction distributions of irradiance and power output for that period, and extracts the best estimate from the distribution to achieve intelligent data imputation.

[0083] Finally, the system automatically generates a comprehensive report containing the following: 1) a power inversion curve, showing the comparison between the best estimated power sequence and the measured values ​​(during the validation period); 2) a dynamic confidence interval plot, showing the power sequence and its uncertainty range; 3) missing data completion results, listing the completed time period, power values, and their uncertainty measures (such as standard deviation) and confidence rating (based on prediction variance); 4) a key statistical indicator table, summarizing inversion accuracy indicators, such as daily-scale correlation coefficient and hourly-scale mean absolute error. The report includes, for example... Figure 2 The power comparison curves shown clearly demonstrate the inversion effect. To verify the effectiveness of this invention, a photovoltaic power physical model based on total irradiance-temperature is used as a traditional comparison method. This model takes total horizontal irradiance and photovoltaic cell temperature as inputs and calculates power using the standard photovoltaic conversion efficiency formula. The dynamic changes in spectral composition are not considered. The specific formula is as follows:

[0084] ,

[0085] in, Indicates total irradiance. Indicates the effective area of ​​the photovoltaic unit. This indicates the photoelectric conversion efficiency of the photovoltaic unit under standard test conditions. Indicates the power temperature coefficient. This indicates the actual operating temperature of the photovoltaic unit. This indicates the reference temperature under standard test conditions. This indicates the inverter's conversion efficiency. This represents the overall loss coefficient of the system excluding the inverter.

[0086] Based on the same inventive concept, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the aforementioned photovoltaic power generation inversion method based on spectral composition.

[0087] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned steps of the photovoltaic power generation inversion method and system based on spectral composition.

[0088] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0089] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0090] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0091] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0092] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A method for photovoltaic power generation inversion based on spectral composition, characterized in that, Includes the following steps: Step 1: In the target environment where photovoltaic power generation is to be inverted, obtain the reference photoelectric parameters of the wavelength-selective photovoltaic unit in the photovoltaic system under the blue, green and red monochromatic light bands respectively; calculate the spectral weighting coefficients of the blue, green and red light bands based on the actual spectral photon flux density distribution at the wavelength-selective photovoltaic unit; use the spectral weighting coefficients to perform spectral weighted fusion of the reference photoelectric parameters to obtain the equivalent photoelectric parameters under the blue-green-red composite spectrum. The spectral conversion efficiency under the blue-green-red composite spectrum is calculated based on the equivalent photoelectric parameters under the blue-green-red composite spectrum, and the effective spectral flux and effective radiation for spectral utilization are calculated using the actual spectral photon flux density distribution. Step 2: Obtain the first signal and the second signal from the spectral conversion efficiency, effective spectral flux, and effective spectral radiation under the blue-green-red composite spectrum. The first signal is sensitive to the visible light band, and its intensity is linearly related to the incident irradiance. The second signal is sensitive to thermal radiation, and its intensity is strongly correlated with the ambient temperature. Based on the first and second signals, predict the solar irradiance using a pre-constructed irradiance inversion model. Specifically, based on the first and second signals, a pre-constructed irradiance inversion model is used to predict solar irradiance, as detailed below: Calculate the estimated intermediate irradiance: ， in, This represents the intermediate irradiance estimate based on a linear proportional relationship. This represents a known reference irradiance value. This indicates the first signal value acquired. Indicates that at an irradiance of The first signal reference value obtained under the calibration conditions; Calculate the dynamic compensation factor: ， in, This represents the dynamic compensation factor used to correct for the effects of temperature. A compensation coefficient representing the degree to which the change in the second signal affects the photoelectric conversion efficiency. This indicates the acquired second signal value. This represents the second signal reference value obtained under calibration conditions; The predicted solar irradiance is calculated using the intermediate irradiance estimate and the dynamic compensation factor. : ； Step 3: Based on Step 1 and Step 2, obtain the predicted values ​​of solar irradiance, auxiliary environmental data, and photovoltaic power generation of the photovoltaic system under historical time to form a dataset for training the hierarchical Bayesian generative model. The model takes the predicted values ​​of solar irradiance and auxiliary environmental data as input and the photovoltaic power generation inversion value as output. Step 4: Use the trained hierarchical Bayesian generative model to invert the photovoltaic power generation, obtain the photovoltaic power generation inversion sequence, and then obtain the photovoltaic power generation inversion curve and its dynamic error confidence interval, thus realizing the inversion of photovoltaic power generation.

2. The photovoltaic power generation inversion method based on spectral composition according to claim 1, characterized in that, The specific process of step 1 is as follows: In the target environment where photovoltaic power generation is to be inverted, the reference photoelectric parameters of wavelength-selective photovoltaic units in the blue, green, and red light bands are obtained by measurement or modeling, including: maximum spectral conversion efficiency. Initial spectral response coefficient and base loss coefficient , , This indicates blue light, with a wavelength range of 400-500nm; It represents green light with a wavelength range of 500-600 nm; This indicates red light, with a wavelength range of 600-700 nm; Based on the actual spectral photon flux density distribution at the wavelength-selective photovoltaic unit, calculate the spectral weighting coefficients for the blue, green, and red light bands. and The calculation formula is: ， in, and They represent The minimum and maximum wavelength boundaries of the band. Indicates the wavelength of the wavelength-selective photovoltaic unit. The spectral photon flux density distribution at a given point, where the integral in the denominator represents the total photon flux of the entire spectrum utilizing the effective radiation band. For band The spectral conversion efficiency function; This is a non-linear adjustment parameter, with a value range of [value range missing]. ; By weighting and fusing the reference photoelectric parameters of all monochromatic lights with the corresponding spectral weighting coefficients, the equivalent photoelectric parameters under the blue-green-red composite spectrum are obtained: ， ， ， in, and These represent the equivalent maximum spectral conversion efficiency, equivalent initial spectral response coefficient, and equivalent basic loss coefficient synthesized under the blue-green-red composite spectrum, respectively. based on and The spectral conversion efficiency under the blue-green-red composite spectrum was calculated using a photoelectric response model. ; The impact of spectral composition on light energy conversion efficiency is evaluated using the following spectral composition performance indicators: ， ， in, and These represent effective spectral flux and spectral utilization effective radiation, respectively. The curve representing the relative spectral response coefficients of the spectral conversion. The spectral curve representing the relative effect of spectral conversion.

3. The photovoltaic power generation inversion method based on spectral composition according to claim 1, characterized in that, In step 3, the auxiliary environmental data includes temperature, humidity, soil moisture, surface reflectance, and aerosol optical thickness. The optimization objective of the hierarchical Bayesian generative model is to maximize the following joint variational lower bound: ， in, The objective function for model optimization, To support environmental data, This is the predicted value of solar irradiance. As a potential variable characterizing the spatiotemporal features of irradiance; For mathematical expectation operators, For latent variables for conditions The coding distribution, For and For the conditional irradiance likelihood model For Kullback–Leibler divergence, The weight coefficients for the regularization term. For adaptive regularization terms Weighting coefficients; and These are the parameters to be optimized for the encoder and the irradiance likelihood model, respectively. For inversion loss, Based on the standard prior distribution The regularization term is the basis. For adaptive regularization; It has the following forms: ， in, For the hierarchical Bayesian generative model at time... The prediction variance The weight function is monotonically decreasing. and Let represent the first-order gradient operator and the second-order difference operator with respect to time, respectively. Hyperparameters for controlling sparsity; After the hierarchical Bayesian generative model converges, the posterior uncertainty of the latent variables is refined using the following covariance update formula: ， in, The refined posterior covariance matrix of the latent variables. For the irradiance likelihood model, the optimal estimation of latent variables Jacobian matrix at the location, Here is the covariance matrix of the irradiance observation noise. The prior covariance matrix given by the encoder; By using the posterior probability distribution of solar irradiance Mid-sampling and using photovoltaic physical models Forward calculations were performed, and the photovoltaic AC power output was statistically obtained using the Monte Carlo method. and overload clipping loss Expectation, variance, and confidence interval. DC power This is the maximum AC output power of the inverter.

4. The photovoltaic power generation inversion method based on spectral composition according to claim 3, characterized in that, In step 4, the posterior probability distribution obtained in step 3 is used as a basis. Generate the inversion curve of historical photovoltaic power generation of the photovoltaic system and its dynamic error confidence interval; for the missing periods in the historical data, call the trained hierarchical Bayesian generative model, and use the auxiliary environmental data of the effective periods before and after the missing periods as conditions to generate the conditional prediction distribution of irradiance and photovoltaic power output for the missing periods; finally, output the photovoltaic power inversion sequence, dynamic confidence interval and photovoltaic power compensation value for the missing periods.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the photovoltaic power generation inversion method based on spectral composition as described in any one of claims 1 to 4.

6. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the photovoltaic power generation inversion method based on spectral composition as described in any one of claims 1 to 4.

Images