A method and system for inverting photovoltaic power generation based on plant growth status

Abstract

This invention discloses a method and system for photovoltaic power generation inversion based on plant growth status. The method includes acquiring remote sensing image data and hyperspectral image data of a target area at a reference time point; obtaining enhanced plant state characteristics based on the hyperspectral image data; using the enhanced plant state characteristics and auxiliary environmental data, acquiring the plant's utilization of light energy through photosynthesis via a plant-photosynthetic response sub-model; mapping the light energy utilization to a total surface irradiance time series using a light inversion sub-model; inputting this time series into a photovoltaic power output model to obtain a photovoltaic system power reconstruction sequence; and obtaining a photovoltaic power reconstruction curve, error confidence interval, and missing data completion report based on the photovoltaic system power reconstruction sequence, thus completing the photovoltaic power generation inversion. This invention enables the reconstruction of historical photovoltaic power, completion of missing data, and prediction of power generation trends, improving the monitoring and evaluation capabilities of photovoltaic systems.

Description

Technical Field

[0001] This invention relates to the fields of photovoltaic power generation, environmental monitoring and artificial intelligence algorithm technology, specifically to a method and system for photovoltaic power generation inversion based on plant growth status. Background Technology

[0002] Currently, photovoltaic (PV) power prediction or historical power reconstruction mainly relies on meteorological measurements (such as solar irradiance, cloud cover, temperature, and wind speed) or direct historical PV output data. However, in many remote or data-scarce areas, incomplete meteorological station data, low-resolution satellite irradiance data, and missing PV output records lead to difficulties in accurately estimating historical power. Furthermore, in scenarios such as agricultural PV and forest-solar hybrid systems, vegetation conditions, changes in shading, or variations in surface reflectivity significantly affect the amount of irradiance received by the modules, while traditional models often neglect the coupling relationship between vegetation, irradiance, and power. Therefore, a new technical solution is urgently needed that can utilize vegetation growth status as input to invert historical light conditions and further reconstruct PV power output, thereby filling data gaps and improving estimation accuracy. Summary of the Invention

[0003] The technical problem to be solved by this invention is to address the issues of over-reliance on meteorological observation data, satellite remote sensing irradiance data, or direct photovoltaic power output history in existing technologies, which lead to incomplete meteorological station data, low satellite data resolution, and missing photovoltaic power output records in remote areas and data-scarce regions. This invention provides a photovoltaic power generation inversion method and system based on plant growth status to improve the accuracy and applicability of photovoltaic power estimation.

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

[0005] A method for inverting photovoltaic power generation based on plant growth status includes the following steps:

[0006] S1. Obtain remote sensing image data and hyperspectral image data of the target area at a reference time point through satellite remote sensing, UAV multispectral image and ground hyperspectral sensor, obtain plant growth status index data based on the remote sensing image data, extract the features of the hyperspectral image data, and obtain enhanced plant status features.

[0007] S2. Based on enhanced plant state characteristics and auxiliary environmental data, the plant's utilization of light energy through photosynthesis is obtained through a plant-photosynthetic response sub-model. The utilization of light energy is then mapped to a time series of total surface irradiance using an illumination inversion sub-model.

[0008] S3. Input the total surface irradiance time series into the photovoltaic power output model to obtain the power reconstruction sequence of the photovoltaic system;

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

[0010] Furthermore, in step S1, the remote sensing image data is sequentially processed by radiometric calibration and atmospheric correction, temporal interpolation, spatial registration, and noise filtering to obtain plant growth status index data.

[0011] Plant growth status indicators include normalized vegetation index, solar-induced chlorophyll fluorescence, and leaf area index;

[0012] Features extracted from hyperspectral image data include:

[0013] The hyperspectral image data is embedded and modeled using a self-supervised learning model based on Jaccard similarity. The vegetation canopy structure variation characteristics are calculated using a piecewise radial monotonicity analysis method, with the specific formula as follows:

[0014] ;

[0015] in, This indicates the variation characteristics of vegetation canopy structure in the radial direction. Indicates radial variation mode, This represents the total number of radial segments. Indicates the first The average vegetation index of radial segments. Indicates the first Average vegetation index of radial segments;

[0016] The hyperspectral image data is modeled using a self-supervised embedding loss function, with the specific formula as follows:

[0017] ;

[0018] ;

[0019] ;

[0020] in, This represents the ground truth value of Jaccard similarity. Indicates the first in hyperspectral image data One sample, Indicates the first in hyperspectral image data One sample, Indicates and Related sets, Indicates and Related sets, Indicates the prediction error. Both represent weight matrices. express transpose, Indicates by and The feature representation is a joint feature vector composed of combined features. This represents the final total loss value. This represents the total number of sample pairs. This represents the total number of samples in the hyperspectral image data;

[0021] The hyperspectral embedding feature vector is obtained by minimizing the self-supervised embedding loss function. ;

[0022] Using feature fusion function to Hyperspectral embedding feature vector Sun-induced chlorophyll fluorescence Leaf area index Multi-source feature fusion processing was performed to obtain enhanced plant state features. The specific formula is as follows:

[0023] ;

[0024] in, This represents the feature fusion function.

[0025] Furthermore, in step S2, the total surface irradiance time series obtained includes the following:

[0026] Supporting environmental data include temperature, humidity, soil moisture, surface reflectance, and aerosol optical thickness;

[0027] Initial values ​​of irradiance based on climate statistics and plant physiological constraints were used to set the initial values ​​of the light inversion sub-model.

[0028] Enhance plant state characteristics The auxiliary environmental data is input into the illumination inversion sub-model, and a parameterized dedicated encoder network is used. Encoding is performed to obtain the initial posterior distribution estimate of the latent variables of surface irradiance components. This distribution estimate includes the mean estimate and the covariance matrix decomposition factor.

[0029] A hybrid optimization strategy combining alternating minimization and Markov chain Monte Carlo is employed, with the joint loss function encompassing the illumination inversion sub-model and the observation operator as the objective, to iteratively optimize the parameters of the initial posterior distribution estimate and the illumination inversion sub-model, thereby obtaining the optimal illumination inversion sub-model parameters. Dedicated encoder network parameters Optimal estimation of surface radiation components ;

[0030] The specific formula for the joint loss function is as follows:

[0031] ;

[0032] in, This represents the value of the joint loss function. Indicates the number of Monte Carlo samples. Represents remote sensing image data, This represents the illumination inversion sub-model. Indicates the first Potential variables of surface irradiance composition from Monte Carlo sampling This represents auxiliary environment data. The parameters represent the illumination inversion sub-model. Represents the covariance matrix of remote sensing image data. This represents a plant-photosynthetic response model. Indicates irradiance measured on the ground. Represents the covariance matrix of irradiation. This represents a coefficient indicating the control penalty. Represents the regularization term, Indicates potential variables of surface irradiation components;

[0033] The optimal estimate of surface irradiance components was calculated using an uncertainty quantification mechanism in the illumination inversion sub-model. Jacobian matrix at the location ,get posterior covariance matrix The specific formula is as follows:

[0034] ;

[0035] in, Denotes the covariance matrix of the prior distribution. express transpose;

[0036] Optimal estimation of surface radiation components The time series of total surface irradiance was obtained by comprehensively evaluating and processing the subsequent covariance matrix.

[0037] Furthermore, in step S3, the obtained power reconstruction sequence of the photovoltaic system includes the following:

[0038] The total surface irradiance time series is normalized and sliced ​​into time windows to obtain a fixed-length total surface irradiance time series. This time series is then input into the photovoltaic power output model, and spatiotemporal joint encoding and attention aggregation are performed using a spatial enhanced attention module to obtain enhanced feature representations.

[0039] The enhanced feature representation is processed using a memory retrieval feedforward network module to obtain the power reconstruction sequence of the photovoltaic system.

[0040] Furthermore, the spatially enhanced attention module includes a feature embedding unit, a two-dimensional rotational position encoding unit, and an attention weighting computation unit;

[0041] The feature embedding unit performs a linear mapping on the multimodal features in the fixed-length total surface irradiance time series. The two-dimensional rotational position encoding unit encodes the temporal and spatial location information of the mapped multimodal features in the total surface irradiance time series. The attention weighting calculation unit performs joint attention aggregation in the spatial and temporal dimensions to obtain the enhanced feature representation. The specific formula is as follows:

[0042] ;

[0043] in, Represents the query matrix. Represents the key matrix, Represents a value matrix, Both represent points in time. This represents the total number of points in time. Represents a mapping function. The dimension representing the key / query vector. Spatial representation enhances attention. Indicates the first Time-point query matrix Indicates the first Key matrix at time points, Indicates the first The value matrix at time points, Indicates the first Key matrix at time points.

[0044] Furthermore, the memory retrieval feedforward network module includes a memory storage unit, a retrieval matching unit, and a dynamic expert combination unit;

[0045] The enhanced feature representation is stored as key-value pairs in the memory storage unit of the memory retrieval feedforward network module. The retrieval matching unit in the memory retrieval feedforward network module calculates the similarity between the stored enhanced feature representation and the key vectors of all historical patterns in the memory bank. The highest matching score is then selected using the Softmax function. The historical experience representation vector is obtained by weighted summation of each memory unit, using the following formula:

[0046] ;

[0047] ;

[0048] in, Indicates the first Attention weights for each memory unit The feature vector representing the time series of total surface irradiance. express transpose, Indicates the first The key vector of each memory unit, Indicates the first The key vector of each memory unit, Represents a vector of historical experience. Indicates the first A vector of values ​​for each memory unit;

[0049] After concatenating the enhanced feature representation with the historical experience representation vector, the gating weights of each expert are dynamically calculated using a dynamic expert combination unit. All gating weights are then weighted and fused to obtain the power reconstruction sequence of the photovoltaic system. The specific formula is as follows:

[0050] ;

[0051] in, For the first The gating weight of each expert, Indicates the first A network of experts This represents the power reconstruction sequence of a photovoltaic system.

[0052] Furthermore, in step S4, a photovoltaic power reconstruction curve is plotted based on the power reconstruction sequence of the photovoltaic system; calculations are performed. The probability range of power estimates in the posterior covariance matrix is ​​used to obtain the error confidence interval; the photovoltaic power output model is used to predict the irradiance and power of the missing period, and the credibility of the supplementary values ​​is evaluated in combination with the error confidence interval, so as to obtain a missing period supplementary report containing power estimates and accuracy indicators.

[0053] Furthermore, this invention also proposes a photovoltaic power generation inversion system based on plant growth status, comprising:

[0054] The feature acquisition module is used to acquire remote sensing image data and hyperspectral image data of the target area at a reference time point through satellite remote sensing, UAV multispectral image and ground hyperspectral sensor, acquire plant growth status index data based on the remote sensing image data, extract the features of the hyperspectral image data, and obtain enhanced plant status features.

[0055] The time series acquisition module is used to acquire the light energy utilization of plants through photosynthesis based on enhanced plant state characteristics and auxiliary environmental data, through a plant-photosynthetic response sub-model, and to map the light energy utilization into a time series of total surface irradiance using an illumination inversion sub-model.

[0056] The inversion module is used to input the total surface irradiance time series into the photovoltaic power output model to obtain the power reconstruction sequence of the photovoltaic system; based on the power reconstruction sequence of the photovoltaic system, the photovoltaic power reconstruction curve, error confidence interval and missing measurement period supplement report are obtained to complete the inversion of photovoltaic power generation.

[0057] Furthermore, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the photovoltaic power generation inversion method based on plant growth status.

[0058] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the aforementioned photovoltaic power generation inversion method based on plant growth status.

[0059] Compared with the prior art, the present invention, employing the above technical solution, has the following technical effects:

[0060] 1. This invention does not rely on complete meteorological observation data or photovoltaic power output history. Instead, it uses vegetation growth status as an indirect record carrier of the environment and power generation history, making it suitable for scenarios where meteorological data is missing or the historical power output record of the photovoltaic system is incomplete.

[0061] 2. This invention fully considers the coupling relationship between vegetation, irradiance, and power. By accurately inverting light conditions through vegetation status indicators, it can effectively adapt to scenarios with significant dynamic changes in vegetation shading, such as agricultural photovoltaics and forest-photovoltaic complementarity, thereby improving the pertinence and accuracy of power estimation.

[0062] 3. This invention integrates plant photosynthetic response mechanisms with machine learning algorithms, combined with a multi-module collaborative intelligent prediction framework and hybrid optimization strategy, to achieve high-precision inversion of light conditions and photovoltaic power, while providing error confidence intervals to ensure the reliability of the results.

[0063] 4. This invention can reconstruct the historical power of photovoltaic power plants, supplement missing data, and predict power generation trends, significantly improving the monitoring and evaluation capabilities of photovoltaic systems and providing scientific support for the optimized operation and planning design of photovoltaic power plants. Attached Figure Description

[0064] Figure 1 This is a flowchart illustrating the overall implementation of the present invention.

[0065] Figure 2 This is a graph showing the changes in plant growth status index data over time in an embodiment of the present invention.

[0066] Figure 3 This is a graph showing the results of the total surface irradiance time series obtained in an embodiment of the present invention.

[0067] Figure 4 This is a diagram showing the result of obtaining the power reconstruction sequence of the photovoltaic system in an embodiment of the present invention. Detailed Implementation

[0068] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0069] To achieve the above objectives, this invention proposes a method for inverting photovoltaic power generation based on plant growth status, such as... Figure 1 As shown, the specific steps are as follows:

[0070] S1. Acquire remote sensing and hyperspectral image data of the target area at a reference time point through satellite remote sensing (such as Landsat, Sentinel-2, etc.), UAV multispectral imagery, and ground-based hyperspectral sensors. Based on this remote sensing image data, obtain plant growth status index data, extract features from the hyperspectral image data, and obtain enhanced plant status features; specifically:

[0071] The remote sensing image data is processed as follows: 1. Radiometric calibration and atmospheric correction: The original digital quantization values ​​are converted into surface reflectance to eliminate the effects of atmospheric scattering and absorption; 2. Temporal interpolation: For data gaps, distributed temporal interpolation algorithms (such as linear interpolation or spline interpolation) are used to fill the data gaps to ensure the continuity of the time series; 3. Spatial registration: The multi-source remote sensing image data are unified to the same geographic coordinate system and spatial resolution; 4. Noise filtering: Sliding window filtering or wavelet denoising methods are used to remove noise from the images. Through the above processing, continuous and consistent plant growth status index data are obtained.

[0072] Plant growth status indicators include NDVI (Normalized Difference Vegetation Index), SIF (Solar Induced Chlorophyll Fluorescence), and LAI (Leaf Area Index).

[0073] Features extracted from hyperspectral image data include:

[0074] The hyperspectral image data is embedded and modeled using a self-supervised learning model based on Jaccard similarity. The vegetation canopy structure variation characteristics are calculated using a piecewise radial monotonicity analysis method, with the specific formula as follows:

[0075] ;

[0076] in, This indicates the variation characteristics of vegetation canopy structure in the radial direction. Indicates radial variation mode, This represents the total number of radial segments. Indicates the first The average vegetation index of radial segments. Indicates the first Average vegetation index of radial segments;

[0077] The core objective of the segmented radial monotonicity analysis method is to extract quantitative features reflecting changes in the three-dimensional structure of vegetation canopy from macroscopic remote sensing images. Healthy vegetation canopies in their vigorous growth phase often exhibit specific gradient distributions in leaf area density and chlorophyll content in both vertical and radial directions. For example, for many tree species, the outer edges of the canopy receive more sunlight, resulting in higher vegetation indices (such as NDVI). By calculating this radial monotonicity pattern, it is possible to effectively distinguish vegetation at different growth stages, among different species, or under environmental stress—structural information that traditional vegetation indices (such as NDVI) cannot fully capture.

[0078] The hyperspectral image data is modeled using a self-supervised embedding loss function, with the specific formula as follows:

[0079] ;

[0080] ;

[0081] ;

[0082] in, This represents the true value of Jaccard similarity; Indicates the first in hyperspectral image data One sample; Indicates the first in hyperspectral image data One sample; Indicates and Related sets, Indicates and Related sets are used to calculate similarity; Indicates prediction error; Both represent weight matrices. Used to transform intermediate features Used for transformations in the output layer; express transpose; Indicates by and The feature representation is a joint feature vector composed of combined features; This represents the final total loss value; This represents the total number of sample pairs; This represents the total number of samples in the hyperspectral image data;

[0083] The hyperspectral embedding feature vector is obtained by minimizing the self-supervised embedding loss function. It is used to quantify vegetation chlorophyll concentration, photosynthetic intensity, and canopy structure information;

[0084] Using feature fusion function to Hyperspectral embedding feature vector Sun-induced chlorophyll fluorescence Leaf area index Multi-source feature fusion processing was performed to obtain enhanced plant state features. The specific formula is as follows:

[0085] ;

[0086] in, Represents the feature fusion function;

[0087] It is used as input for subsequent illumination inversion sub-models to achieve high-precision inversion and reconstruction of irradiation conditions and photovoltaic power.

[0088] S2. Based on enhanced plant state characteristics and auxiliary environmental data, the plant's utilization of light energy through photosynthesis is obtained through a plant-photosynthetic response sub-model. Then, using a light inversion sub-model, the light energy utilization is mapped to a time series of total surface irradiance. Specifically:

[0089] A plant-photosynthetic response sub-model was constructed to estimate the time series of light energy available to plants or photosynthetically active radiation. This model takes into account plant physiological constraints, such as the light compensation point and the light saturation point.

[0090] Ancillary environmental data, including temperature, humidity, soil moisture, surface reflectance, and aerosol optical thickness, are crucial for constructing accurate plant-photosynthetic response models. For example, temperature directly affects enzyme activity, thus influencing photosynthetic efficiency; soil moisture stress causes stomatal closure, reduces carbon assimilation, and alters the plant's efficiency in utilizing light energy. Ancillary environmental data can be obtained from reanalysis data (such as ERA5), ground weather stations, or specialized satellite products (such as MODIS surface temperature and soil moisture products). All ancillary environmental data must undergo rigorous temporal and spatial matching with plant growth status indicators to ensure data spatiotemporal consistency, which is fundamental to successful inversion.

[0091] The initial values ​​of the parameters of the light inversion sub-model were set using initial values ​​of irradiance based on climate statistics and plant physiological constraints (such as light compensation point and light saturation point) to ensure that the optimization process converged and stabilized.

[0092] Enhance plant state characteristics The auxiliary environmental data is input into the illumination inversion sub-model, and a parameterized dedicated encoder network is used. Encoding is performed to obtain the initial posterior distribution estimate of the latent variables of surface irradiance components. This distribution estimate includes the mean estimate and the covariance matrix decomposition factor.

[0093] Among them, parameterized dedicated encoder networks This includes a dimensionality reduction preprocessing layer, a mean prediction branch, a covariance prediction branch, and residual connections;

[0094] After the dimensionality reduction preprocessing layer projects enhanced plant state characteristics and auxiliary environmental data onto the physical observation space, the mean prediction branch is used to obtain mean estimates of the latent variables of surface irradiance components. The covariance matrix decomposition factor of the latent variables of surface irradiance components is obtained by using the covariance prediction branch. ,satisfy The residual connection is used to connect the output of the dimensionality reduction preprocessing layer to the mean prediction branch, thereby improving network stability.

[0095] A hybrid optimization strategy combining alternating minimization and Markov chain Monte Carlo is employed, with the joint loss function encompassing the illumination inversion sub-model and the observation operator as the objective, to iteratively optimize the parameters of the initial posterior distribution estimate and the illumination inversion sub-model, thereby obtaining the optimal illumination inversion sub-model parameters. Dedicated encoder network parameters Optimal estimation of surface radiation components ;

[0096] The specific formula for the joint loss function is as follows:

[0097] ;

[0098] in, This represents the value of the joint loss function. Indicates the number of Monte Carlo samples. Represents remote sensing image data, This represents the illumination inversion sub-model. Indicates the first Potential variables of surface irradiance composition from Monte Carlo sampling This represents auxiliary environment data. The parameters represent the illumination inversion sub-model. Represents the covariance matrix of remote sensing image data. This represents a plant-photosynthetic response model. Indicates irradiance measured on the ground. Represents the covariance matrix of irradiation. This represents a coefficient indicating the control penalty. Represents the regularization term, Indicates potential variables of surface irradiation components;

[0099] The optimal estimate of surface irradiance components was calculated using an uncertainty quantification mechanism in the illumination inversion sub-model. Jacobian matrix at the location ,get posterior covariance matrix The specific formula is as follows:

[0100] ;

[0101] in, Denotes the covariance matrix of the prior distribution. express transpose;

[0102] Optimal estimation of surface radiation components The time series of total surface irradiance (including direct, diffuse, and cloud-shading components) is obtained by comprehensively evaluating and processing the subsequent covariance matrix.

[0103] S3. Input the total surface irradiance time series into the photovoltaic power output model to obtain the power reconstruction sequence of the photovoltaic system; specifically:

[0104] The total surface irradiance time series is normalized and sliced ​​into time windows to obtain a fixed-length total surface irradiance time series. This time series is then input into the photovoltaic power output model, and spatiotemporal joint encoding and attention aggregation are performed using a spatial enhanced attention module to obtain an enhanced feature representation that incorporates complex spatiotemporal dependencies.

[0105] In agricultural photovoltaic (PV) applications, uneven spatial distribution of vegetation beneath or around PV arrays leads to variations in the irradiance received by PV modules at different locations, after reflection or shading by the vegetation. The spatial enhancement attention module aims to capture this complex spatial dependency. Two-dimensional rotational position encoding enables the model to understand the specific location (spatial dimension) of each module within the array and its time point (temporal dimension) within a specific time series. The attention mechanism dynamically learns and weights spatially related and temporally synchronized module node information. For example, when dense vegetation in a certain area causes shading, this module automatically enhances the weight of downstream or adjacent affected module node features in the prediction, thereby achieving more accurate power estimation.

[0106] The spatial enhanced attention module includes a feature embedding unit, a two-dimensional rotational position encoding unit, and an attention weighting computation unit;

[0107] The feature embedding unit performs a linear mapping on the multimodal features in the fixed-length total surface irradiance time series. The two-dimensional rotational position encoding unit encodes the temporal and spatial location information of the mapped multimodal features in the total surface irradiance time series. The attention weighting calculation unit performs joint attention aggregation in the spatial and temporal dimensions to obtain an enhanced feature representation that incorporates complex spatiotemporal dependencies. The specific formula is as follows:

[0108] ;

[0109] in, Represents the query matrix. Represents the key matrix, Represents a value matrix, Both represent points in time. This represents the total number of points in time. Represents a mapping function. The dimension representing the key / query vector. Spatial representation enhances attention. Indicates the first Time-point query matrix Indicates the first Key matrix at time points, Indicates the first The value matrix at time points, Indicates the first Key matrix at time points;

[0110] The enhanced feature representation, which incorporates complex spatiotemporal dependencies, is processed using a memory retrieval feedforward network module to obtain the power reconstruction sequence of the photovoltaic system.

[0111] The memory retrieval feedforward network module includes a memory storage unit, a retrieval matching unit, and a dynamic expert combination unit.

[0112] The enhanced feature representation is stored as key-value pairs in the memory storage unit of the memory retrieval feedforward network module. The retrieval matching unit in the memory retrieval feedforward network module calculates the similarity between the stored enhanced feature representation and the key vectors of all historical patterns in the memory bank. The highest matching score is then selected using the Softmax function. The scattered historical experiences are aggregated into a comprehensive historical experience representation vector by weighted summation of each memory unit, as shown in the specific formula:

[0113] ;

[0114] ;

[0115] in, Indicates the first Attention weights for each memory unit The feature vector representing the time series of total surface irradiance. express transpose, Indicates the first The key vector of each memory unit, Indicates the first The key vector of each memory unit, Represents a vector of historical experience. Indicates the first A vector of values ​​for each memory unit;

[0116] After concatenating the enhanced feature representation with the historical experience representation vector, the gating weights of each expert are dynamically calculated using a dynamic expert combination unit. All gating weights are then weighted and fused to obtain the power reconstruction sequence of the photovoltaic system. The specific formula is as follows:

[0117] ;

[0118] in, For the first The gating weight of each expert, Indicates the first A network of experts This represents the power reconstruction sequence of a photovoltaic system;

[0119] The power output of photovoltaic systems is closely related to historical meteorological conditions, and there are many reusable patterns. The memory retrieval feedforward network module essentially constructs a queryable experience base, storing typical historical patterns of light-temperature-power changes in key-value pairs. When new inputs arrive, the module retrieves the "historical memories" most similar to the current meteorological conditions and dynamically combines the corresponding "experiences" to generate the final prediction. This allows the memory retrieval feedforward network module to quickly call upon historical experience and make more stable and accurate predictions when facing similar weather events that have occurred before (such as sudden clearing and overcasting, or persistent cloudy weather). The value of this module is particularly prominent when measured power data is scarce.

[0120] In the calibration and optimization phase of the memory retrieval feedforward network module, the model is adaptively calibrated using a limited number of measured power samples to correct the module's residuals and systematic errors. A sliding window-based dynamic update strategy is adopted to enable the module to adapt online under different seasonal and meteorological conditions. During the optimization process, the Adam optimizer is used for parameter updates, and an early stopping criterion is used to prevent overfitting.

[0121] S4. Based on the power reconstruction sequence of the photovoltaic system, obtain the photovoltaic power reconstruction curve, error confidence interval, and missing measurement period supplementary report to complete the inversion of photovoltaic power generation; specifically:

[0122] Plot the photovoltaic power reconstruction curve based on the power reconstruction sequence of the photovoltaic system; calculate... The probability range of power estimates in the posterior covariance matrix is ​​used to obtain the error confidence interval. The photovoltaic power output model is then used to predict the irradiance and power for the missing periods. The reliability of the supplementary values ​​is evaluated based on the error confidence interval, resulting in a missing period supplementary report containing power estimates and accuracy indicators. These outputs can be used for historical power analysis, system optimization, and power generation trend prediction for photovoltaic power plants.

[0123] Example 1:

[0124] To verify the effectiveness of the method of the present invention, an agricultural photovoltaic power station was selected as the target area. This area has high vegetation coverage and serious lack of meteorological data.

[0125] Data Acquisition: Collect Sentinel-2 multispectral image data of the target area over several consecutive years, extract indicators such as NDVI, SIF, and LAI, and acquire auxiliary environmental data such as temperature and humidity.

[0126] Illumination inversion: The total surface irradiance time series was obtained by inverting the illumination inversion sub-model. The inversion results were compared with the measured irradiance data, and the root mean square error was reduced by about 15%.

[0127] Power reconstruction: The total surface irradiance time series is input into the photovoltaic power output model to obtain the power reconstruction sequence of the photovoltaic system. The reconstruction error of this sequence is within 5% when compared with the measured power, and the accuracy of filling in the missing measurement period is over 90%.

[0128] The above data shows that the method of the present invention can still reconstruct photovoltaic power with high accuracy even in scenarios with missing data, and has strong practical value.

[0129] This invention reconstructs photovoltaic power output by inverting historical light conditions through vegetation growth status, overcoming the reliance of traditional methods on meteorological data and historical power records. By integrating plant physiological mechanisms with machine learning algorithms, it achieves high-precision and robust power reconstruction, applicable to various complex scenarios. Those skilled in the art can adjust parameters and model structure according to the above embodiments to adapt to specific application needs.

[0130] Example 2:

[0131] Using a 500 kW agricultural photovoltaic power station as the target area, relevant data on maize canopy growth indicators were collected from April to August in the target area. The data information is shown in Table 1.

[0132] Table 1 Data for the target area

[0133]

[0134] Based on the data provided in Table 1, in this typical agricultural photovoltaic scenario in real life, the crop growth status can indirectly record historical light intensity, and therefore is very suitable for the plant state-driven light inversion of this invention.

[0135] To reconstruct missing light levels, maize canopy growth indicators were extracted from Sentinel-2 remote sensing imagery data from April to July. These indicators included NDVI (reflecting chlorophyll content and canopy density), SIF (reflecting photosynthetic efficiency), and LAI (reflecting leaf area and vegetation growth stage). The changes in these growth indicators over time are shown below. Figure 2 As shown. From Figure 2 The typical growth characteristics of corn from greening to tasseling to peak fruiting can be clearly seen. These growth indicators truly reflect the integrated response of plants to environmental factors such as light, temperature, and soil moisture, and serve as an indirect recorder of historical light conditions.

[0136] The results of the time series of total surface irradiance obtained by inverting using the illumination inversion sub-model are as follows: Figure 3 As shown, from Figure 3 The data shows that the radiation intensity increases slowly in April and May, and increases significantly in June and July, exhibiting a peak in strong light, which is consistent with the characteristics of the Jianghuai region.

[0137] The results of obtaining the power reconstruction sequence of the photovoltaic system using the photovoltaic power output model are as follows: Figure 4 As shown, from Figure 4 The output photovoltaic power can be seen as... Figure 3 The irradiance variation trend is highly consistent, and it can simulate the output fluctuations caused by typical weather conditions such as rainy → alternating cloudy and sunny → sunny peak. The peak power is about 100-120 kW, which is consistent with the industry experience of 20-30% capacity reduction caused by crop shading in 500 kW agricultural photovoltaic power stations.

[0138] The power reconstruction sequence of the photovoltaic system obtained by this invention is compared with the measured power sequence. The specific data are shown in Tables 2 and 3.

[0139] Table 2 Comparison of power sequences under sunny conditions

[0140]

[0141] As can be seen from Table 2, under sunny conditions, the power reconstruction sequence of the photovoltaic system obtained by this invention is highly consistent with the measured power sequence, and the relative error is less than 5%.

[0142] Table 3 Comparison of power sequences under cloudy and fluctuating conditions

[0143]

[0144] As can be seen from Table 3, under the condition of large power fluctuations and cloudy fluctuations, the power reconstruction sequence of the photovoltaic system obtained by this invention still maintains high accuracy, with a relative error of about 8%, contributing the main part of the overall MAPE.

[0145] Tables 2 and 3 show comparisons between the power reconstruction sequence of the photovoltaic system obtained by this invention and the measured values ​​under two typical operating conditions: sunny and cloudy / fluctuating conditions. It can be seen that the power reconstruction sequence of the photovoltaic system obtained by this invention performs well under different operating conditions: the relative error is small under sunny conditions (as shown in Table 2), and the error remains stable under fluctuating conditions (as shown in Table 3). The overall performance index, Mean Absolute Percentage Error (MAPE), is 7.7%, which is calculated based on all test data including the above operating conditions.

[0146] This invention also proposes a photovoltaic power generation inversion system based on plant growth status, including a feature acquisition module, a time series acquisition module, an inversion module, and a computer program that can run on a processor. It should be noted that each module in the above system corresponds to a specific step of the method provided in this invention embodiment, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention embodiment.

[0147] This invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. It should be noted that when the processor executes the computer program, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0148] This invention also proposes a computer-readable storage medium storing a computer program. It should be noted that when the computer program is executed by a processor, it corresponds to the specific steps of the method provided in this invention, possessing the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in this invention.

[0149] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A photovoltaic power generation power inversion method based on the growth state of plants, characterized by, include: S1. Acquire remote sensing and hyperspectral image data of the target area at a reference time point through satellite remote sensing, UAV multispectral imagery, and ground-based hyperspectral sensors. Based on this remote sensing image data, obtain plant growth status index data, extract features from the hyperspectral image data, and obtain enhanced plant status features; specifically: The remote sensing image data is processed sequentially by radiometric calibration and atmospheric correction, temporal interpolation, spatial registration, and noise filtering to obtain plant growth status index data. Plant growth status indicators include normalized vegetation index, solar-induced chlorophyll fluorescence, and leaf area index; Features extracted from hyperspectral image data include: The hyperspectral image data is embedded and modeled using a self-supervised learning model based on Jaccard similarity. The vegetation canopy structure variation characteristics are calculated using a piecewise radial monotonicity analysis method, with the specific formula as follows: ； in, This indicates the variation characteristics of vegetation canopy structure in the radial direction. Indicates radial variation mode, This represents the total number of radial segments. Indicates the first The average vegetation index of radial segments. Indicates the first Average vegetation index of radial segments; The hyperspectral image data is modeled using a self-supervised embedding loss function, with the specific formula as follows: ； in, This represents the ground truth value of Jaccard similarity. Indicates the first in hyperspectral image data One sample, Indicates the first in hyperspectral image data One sample, Indicates and Related sets, Indicates and Related sets, Indicates the prediction error. Both represent weight matrices. express transpose, Indicates by and The feature representation is a joint feature vector composed of combined features. This represents the final total loss value. This represents the total number of sample pairs. This represents the total number of samples in the hyperspectral image data; The hyperspectral embedding feature vector is obtained by minimizing the self-supervised embedding loss function. ; Using feature fusion function to Hyperspectral embedding feature vector Sun-induced chlorophyll fluorescence and leaf area index Multi-source feature fusion processing was performed to obtain enhanced plant state features. The specific formula is as follows: ； in, Represents the feature fusion function; S2. Based on enhanced plant state characteristics and auxiliary environmental data, the plant's utilization of light energy through photosynthesis is obtained through a plant-photosynthetic response sub-model. The utilization of light energy is then mapped to a time series of total surface irradiance using an illumination inversion sub-model. S3. Input the total surface irradiance time series into the photovoltaic power output model to obtain the power reconstruction sequence of the photovoltaic system; S4. Based on the power reconstruction sequence of the photovoltaic system, obtain the photovoltaic power reconstruction curve, error confidence interval and missing measurement period supplement report, and complete the inversion of photovoltaic power generation.

2. The photovoltaic power generation inversion method based on plant growth status according to claim 1, characterized in that, In step S2, the total surface irradiance time series obtained includes the following: Supporting environmental data include temperature, humidity, soil moisture, surface reflectance, and aerosol optical thickness; Initial values ​​of irradiance based on climate statistics and plant physiological constraints were used to set the initial values ​​of the light inversion sub-model. Enhance plant state characteristics The auxiliary environmental data is input into the illumination inversion sub-model, and a parameterized dedicated encoder network is used. Encoding is performed to obtain the initial posterior distribution estimate of the latent variables of surface irradiance components. This distribution estimate includes the mean estimate and the covariance matrix decomposition factor. A hybrid optimization strategy combining alternating minimization and Markov chain Monte Carlo is employed, with the joint loss function encompassing the illumination inversion sub-model and the observation operator as the objective. This strategy iteratively optimizes the parameters of the initial posterior distribution estimate and the illumination inversion sub-model to obtain the optimal illumination inversion sub-model parameters. Dedicated encoder network parameters Optimal estimation of surface radiation components ; The specific formula for the joint loss function is as follows: ； in, This represents the value of the joint loss function. Indicates the number of Monte Carlo samples. Represents remote sensing image data, This represents the illumination inversion sub-model. Indicates the first Potential variables of surface irradiance composition from Monte Carlo sampling This represents auxiliary environment data. The parameters represent the illumination inversion sub-model. Represents the covariance matrix of remote sensing image data. This represents a plant-photosynthetic response model. Indicates irradiance measured on the ground. Represents the covariance matrix of irradiation. This represents a coefficient indicating the control penalty. Represents the regularization term, Indicates potential variables of surface irradiation components; The optimal estimate of surface irradiance components was calculated using the uncertainty quantification mechanism of the illumination inversion sub-model. Jacobian matrix at the location ,get posterior covariance matrix The specific formula is as follows: ； in, Denotes the covariance matrix of the prior distribution. express Transpose of; Optimal estimation of surface radiation components The time series of total surface irradiance was obtained by comprehensively evaluating and processing the subsequent covariance matrix.

3. The photovoltaic power generation inversion method based on plant growth status according to claim 1, characterized in that, In step S3, the power reconstruction sequence of the photovoltaic system includes the following: The total surface irradiance time series is normalized and sliced ​​into time windows to obtain a fixed-length total surface irradiance time series. This time series is then input into the photovoltaic power output model, and spatiotemporal joint encoding and attention aggregation are performed using a spatial enhanced attention module to obtain enhanced feature representations. The enhanced feature representation is processed using a memory retrieval feedforward network module to obtain the power reconstruction sequence of the photovoltaic system.

4. The photovoltaic power generation inversion method based on plant growth status according to claim 3, characterized in that, The spatial enhanced attention module includes a feature embedding unit, a two-dimensional rotational position encoding unit, and an attention weighting computation unit; The feature embedding unit performs a linear mapping on the multimodal features in the fixed-length total surface irradiance time series. The two-dimensional rotational position encoding unit encodes the temporal and spatial location information of the mapped multimodal features in the total surface irradiance time series. The attention weighting calculation unit performs joint attention aggregation in the spatial and temporal dimensions to obtain the enhanced feature representation. The specific formula is as follows: ； in, Represents the query matrix. Represents the key matrix, Represents a value matrix, Both represent points in time. This represents the total number of points in time. Represents a mapping function. The dimension representing the key / query vector. Spatial representation enhances attention. Indicates the first Time-point query matrix Indicates the first Key matrix at time points, Indicates the first The value matrix at time points, Indicates the first Key matrix at time points.

5. The photovoltaic power generation inversion method based on plant growth status according to claim 3, characterized in that, The memory retrieval feedforward network module includes a memory storage unit, a retrieval matching unit, and a dynamic expert combination unit; The enhanced feature representation is stored as key-value pairs in the memory storage unit of the memory retrieval feedforward network module. The retrieval matching unit in the memory retrieval feedforward network module calculates the similarity between the stored enhanced feature representation and the key vectors of all historical patterns in the memory bank. The highest matching score is then selected using the Softmax function. The historical experience representation vector is obtained by weighted summation of each memory unit, using the following formula: ； ； in, Indicates the first Attention weights for each memory unit The feature vector representing the time series of total surface irradiance. express transpose, Indicates the first The key vector of each memory unit, Indicates the first The key vector of each memory unit, Represents a vector of historical experience. Indicates the first A vector of values ​​for each memory unit; After concatenating the enhanced feature representation with the historical experience representation vector, the gating weights of each expert are dynamically calculated using a dynamic expert combination unit. All gating weights are then weighted and fused to obtain the power reconstruction sequence of the photovoltaic system. The specific formula is as follows: ； in, For the first The gating weight of each expert, Indicates the first A network of experts This represents the power reconstruction sequence of a photovoltaic system.

6. The photovoltaic power generation inversion method based on plant growth status according to claim 2, characterized in that, In step S4, a photovoltaic power reconstruction curve is plotted based on the power reconstruction sequence of the photovoltaic system; calculations are performed. The probability range of power estimates in the posterior covariance matrix is ​​used to obtain the error confidence interval; the photovoltaic power output model is used to predict the irradiance and power of the missing period, and the credibility of the supplementary values ​​is evaluated in combination with the error confidence interval, so as to obtain a missing period supplementary report containing power estimates and accuracy indicators.

7. A system applied to the photovoltaic power generation inversion method based on plant growth status as described in claim 1, characterized in that, include: The feature acquisition module is used to acquire remote sensing image data and hyperspectral image data of the target area at a reference time point through satellite remote sensing, UAV multispectral image and ground hyperspectral sensor, acquire plant growth status index data based on the remote sensing image data, extract the features of the hyperspectral image data, and obtain enhanced plant status features. The time series acquisition module is used to acquire the light energy utilization of plants through photosynthesis based on enhanced plant state characteristics and auxiliary environmental data, through a plant-photosynthetic response sub-model, and to map the light energy utilization into a total surface irradiance time series using an illumination inversion sub-model. The inversion module is used to input the total surface irradiance time series into the photovoltaic power output model to obtain the power reconstruction sequence of the photovoltaic system; based on the power reconstruction sequence of the photovoltaic system, the photovoltaic power reconstruction curve, error confidence interval and missing measurement period supplement report are obtained to complete the inversion of photovoltaic power generation.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable 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 plant growth status as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, The computer program, when executed by the processor, performs the photovoltaic power generation inversion method based on plant growth status as described in any one of claims 1 to 6.

Images