Internet of things-based real-time monitoring and prediction system for new energy photovoltaic power

By combining an Internet of Things (IoT) system with generative adversarial networks and photovoltaic (PV) array topology analysis, the problem of inaccurate PV power prediction caused by PV panel coating aging was solved, enabling accurate PV power monitoring and prediction during periods of dramatic spectral distribution changes.

WO2026232923A1PCT designated stage Publication Date: 2026-11-12DATANG HYDROPOWER SCI & TECH RES INST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2026/080630
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-06
Filing Date
2026-02-28
Publication Date
2026-11-12

AI Technical Summary

Technical Problem

The aging of existing photovoltaic panel coatings leads to differential attenuation of absorption rates across different wavelengths. Traditional photovoltaic power prediction methods cannot accurately decouple spectral absorption rate attenuation, resulting in large prediction errors, especially during periods of dramatic spectral distribution changes.

Method used

A real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things is adopted. The system collects spectral incident data of the photovoltaic array area through multispectral sensors, utilizes the split-segment response of the generative adversarial network, and performs hierarchical recursive analysis in combination with the photovoltaic array topology to calculate the global attenuation degree and achieve accurate photovoltaic power prediction.

Benefits of technology

It enables precise monitoring and prediction of photovoltaic power based on the differentiated attenuation of absorptivity in different light bands under the aging of photovoltaic panel coatings, thereby improving the accuracy of photovoltaic power prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2026080630_12112026_PF_FP_ABST
    Figure CN2026080630_12112026_PF_FP_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of photovoltaic monitoring, and in particular to an Internet of things-based real-time monitoring and prediction system for new energy photovoltaic power. The system comprises: a data collection module, configured to collect spectral incident data and electrical parameters of a photovoltaic array region; an energy density conversion module, configured to perform energy density conversion on the spectral incident data to obtain spectral‑band energy density distribution data; an attenuation identification module, configured to obtain effective spectral energy values of respective spectral bands on the basis of the spectral-band energy density distribution data and an installation angle of photovoltaic panels, and load the effective spectral energy values into a photovoltaic attenuation identification model to obtain spectral attenuation factor vectors; a global fitting module, configured to perform hierarchical recursive analysis on the basis of the topology of a photovoltaic array to obtain a global contribution matrix and to obtain a global attenuation degree on the basis of the global contribution matrix and the spectral attenuation factor vectors; and a power prediction module, configured to perform power prediction on the basis of the global attenuation degree and target spectral-band energy density distribution data to obtain an overall predicted power of the photovoltaic array. More accurate real-time monitoring and prediction of photovoltaic power are realized.
Need to check novelty before this filing date? Find Prior Art

Description

A Real-time Monitoring and Prediction System for New Energy Photovoltaic Power Based on the Internet of Things Technical Field

[0001] This invention relates to the field of photovoltaic monitoring technology, and in particular to a real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things. Background Technology

[0002] During long-term operation, the nano-coating of photovoltaic (PV) panels ages due to ultraviolet (UV) radiation, leading to a continuous decrease in the absorption rate of the PV panel for specific wavelengths and consequently, a hidden degradation in power generation. Traditional PV power prediction methods typically employ current-voltage curve analysis, which struggles to decouple spectral absorption rate degradation, resulting in significant errors in power prediction. Some teams use linear prediction models based on full-spectrum irradiance and module temperature for PV power prediction, but these models cannot distinguish the differentiated degradation of UV, visible, and infrared light, making predictions inaccurate under conditions of dramatic spectral variations, such as large errors during dawn and dusk due to significant spectral changes. Therefore, achieving more accurate real-time monitoring and prediction of PV power under conditions of coating aging-induced decreases in the absorption rate of PV panels for specific wavelengths has become an urgent problem to be solved. Summary of the Invention

[0003] This invention provides a real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things, in order to solve the problem of inaccurate photovoltaic power prediction caused by the differential attenuation of the absorption rate of photovoltaic panels in different bands due to coating aging.

[0004] To achieve the above objectives, the present invention provides a real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things. The system includes: a data acquisition module, an energy density conversion module, an attenuation identification module, a global fitting module, and a power prediction module, wherein the modules are connected in sequence.

[0005] The data acquisition module is used to collect spectral incident data of the photovoltaic array area in real time through multispectral sensors arranged in a preset area of ​​the photovoltaic array, and to collect the current and voltage parameters of each photovoltaic panel in real time through current and voltage sensors to obtain the power of each photovoltaic panel.

[0006] The energy density conversion module is used to divide the spectral incident data into light segment datasets according to light segment types and perform energy density conversion to obtain light segment energy density distribution data; the light segment types include ultraviolet light segment, visible light segment and near-infrared light segment.

[0007] The attenuation identification module is used to calculate the effective spectral energy value of the optical segment based on the energy density distribution data of the optical segment and the installation angle of the photovoltaic panel. The effective spectral energy value of the optical segment and the power data of each photovoltaic panel are loaded into the photovoltaic attenuation identification model to obtain the spectral attenuation factor vector of each photovoltaic panel.

[0008] The global fitting module is used to perform hierarchical recursive analysis based on the photovoltaic array topology to obtain the global contribution matrix, and to calculate the global attenuation based on the global contribution matrix and the spectral attenuation factor vector of each photovoltaic panel.

[0009] The power prediction module is used to acquire weather index prediction data for the target prediction time point and determine the similarity of historical weather indexes to obtain a reference time point. The energy density distribution data of the optical segment corresponding to the reference time point is used as the energy density distribution data of the target optical segment. Based on the global attenuation degree and the energy density distribution data of the target optical segment, the overall predicted power of the photovoltaic module is obtained through power prediction.

[0010] Furthermore, the energy density conversion module includes a data preprocessing module, an energy density calculation module, a discrete integration module, and a data integration module connected in sequence.

[0011] The data preprocessing module is used to divide the spectral incident data into discrete wavelength points according to the optical segment type to obtain the irradiance data of the discrete wavelength points contained in each optical segment as the optical segment dataset.

[0012] The energy density calculation module is used to calculate the corresponding energy density value based on the irradiance data of discrete wavelength points. for:

[0013] .

[0014] in, Discrete wavelength point Irradiance, It is Planck's constant. It's the speed of light. Discrete wavelength point wavelength, Discrete wavelength point The energy density value.

[0015] The discrete integration module is used to perform discrete integration on the energy density values ​​at discrete wavelength points within a specific optical band to obtain the optical band energy density value. for:

[0016] .

[0017] in, It refers to the spectral resolution of the multispectral sensor. It is the energy density value of the optical segment.

[0018] The data integration module is used to summarize the energy density values ​​of the ultraviolet, visible, and near-infrared light bands to obtain the light band energy density distribution data.

[0019] Furthermore, the attenuation identification module includes an angle conversion module, an energy analysis module, and a generative adversarial network module connected in sequence.

[0020] The angle conversion module is used to obtain solar position data from regional weather stations via an API interface, and to obtain the incident angle of sunlight based on the solar position data and the installation angle of the photovoltaic panel. for:

[0021] .

[0022] in, It is the solar altitude angle. It is the azimuth of the sun. It refers to the installation angle of the photovoltaic panels. It is the azimuth angle of the photovoltaic panel. It is the angle of incidence of light.

[0023] The energy analysis module is used to obtain the effective spectral energy value of the optical segment based on the optical segment energy density value and the incident angle of the light in the optical segment energy density distribution data. for:

[0024] .

[0025] in, It refers to the area of ​​a single photovoltaic panel. It is the effective spectral energy value of the optical segment.

[0026] A generative adversarial network module is used to load the effective spectral energy value of the optical segment and the power data of each photovoltaic panel into the photovoltaic attenuation identification model established by the generative adversarial network algorithm to obtain the spectral attenuation factor vector of each photovoltaic panel.

[0027] Furthermore, the generative adversarial network module includes an input unit, a beam splitting response unit, a power reconstruction unit, and a discrimination unit connected in sequence.

[0028] The input unit is used to generate fused features by taking the effective spectral energy value of the optical band and the power data of each photovoltaic panel as input through the fully connected layer. for:

[0029] .

[0030] in, It refers to the power of the photovoltaic panel. and These are the weight matrix and bias terms of the fully connected layer. It is a feature of fusion.

[0031] The spectral segment response unit is used to set up independent sub-networks for the ultraviolet, visible, and near-infrared light segments to perform spectral segment response on the fused features. The spectral attenuation factor vector of each photovoltaic panel is obtained by LeakyReLU activation. for:

[0032] .

[0033] in, It is the spectral attenuation factor vector. and These are the weight matrices and bias terms of each optical segment network, and LeakyReLU is the activation function.

[0034] The power reconfiguration unit is used to reconfigure the photovoltaic panel's power using a fully connected regression network based on the panel's spectral attenuation factor vector, thus obtaining the reconfigured power. The expression is:

[0035] .

[0036] and These are the weight matrix and bias terms of a fully connected regression network. It is the power reconfiguration of the photovoltaic panel.

[0037] The discrimination unit is used to input the absolute error between the reconstructed power of the photovoltaic panel and the power data of the photovoltaic panel into the multilayer perceptron to obtain the attenuation discrimination probability. When the attenuation discrimination probability is less than the preset discrimination threshold, the spectral attenuation factor vector of the photovoltaic panel is determined to be valid and stored. Otherwise, the parameters are updated and the spectral attenuation factor vector is regenerated through backpropagation of the loss function of the photovoltaic attenuation generator.

[0038] The multilayer perceptron uses the Sigmoid function, expressed as:

[0039] ;

[0040] in, , Here are the weight matrix and bias terms for the photovoltaic degradation discriminator. It is the attenuation of the discrimination probability;

[0041] Furthermore, the training steps for the photovoltaic attenuation identification model are as follows:

[0042] The effective spectral energy values ​​of the optical segment and the corresponding photovoltaic power data within the historical operating cycle are used as training data.

[0043] The parameters of the photovoltaic degradation generator and the photovoltaic degradation discriminator are initialized using the He initialization method. The initial learning rate of the photovoltaic degradation generator and the photovoltaic degradation discriminator is set, and the learning rate is updated using the cosine annealing learning rate scheduling strategy.

[0044] The photovoltaic degradation generator and photovoltaic degradation discriminator are iteratively trained using an alternating training strategy. Each training iteration includes:

[0045] The parameters of the photovoltaic degradation generator are fixed, and the photovoltaic degradation discriminator is continuously trained. The parameters of the photovoltaic degradation discriminator are updated according to the loss function of the photovoltaic degradation discriminator, and the parameters are updated by the optimizer with the learning rate of the photovoltaic degradation discriminator.

[0046] .

[0047] in, It is the decay of the discrimination probability. It is the loss function value of the photovoltaic degradation discriminator.

[0048] With fixed parameters for the photovoltaic degradation discriminator, a photovoltaic degradation generator is trained. The parameters of the photovoltaic degradation generator are updated according to its loss function. The Adam optimizer is used to update the parameters with the learning rate of the photovoltaic degradation generator. The loss function of the photovoltaic degradation generator is:

[0049] .

[0050] in, It is the reconfiguration power of the photovoltaic panel. These are the reconstruction error weighting coefficients. It refers to the power of the photovoltaic panel. It is the loss function value of the photovoltaic degradation generator.

[0051] The reconstruction error weighting coefficient is used to balance adversarial loss and power accuracy. The initial value of the reconstruction error weighting coefficient is set to 0.5, and it is gradually increased to 1 through linear updates. The linear updates are as follows:

[0052] .

[0053] in, It is the first Reconstruction error weighting coefficients for the next iteration These are the initial reconstruction error weighting coefficients. These are the final reconstruction error weighting coefficients. It is the iteration round. It represents the maximum number of iterations.

[0054] The iteration ends when the training reaches the preset maximum number of iterations.

[0055] Furthermore, the global fitting module includes a hierarchical analysis module, a contribution analysis module, and a decay analysis module connected in sequence.

[0056] The hierarchy resolution module is used to determine the series and parallel levels of each photovoltaic panel based on the photovoltaic array topology, and to obtain the voltage contribution coefficient of the photovoltaic panel based on its series level. for:

[0057] .

[0058] in, It is a photovoltaic panel The number of photovoltaic panels in the series connection level It is a photovoltaic panel The voltage contribution coefficient.

[0059] The current contribution coefficient of the photovoltaic panel is obtained based on the parallel connection level of the photovoltaic panel. for:

[0060] .

[0061] in, It is a photovoltaic panel The number of photovoltaic panels in the parallel layer, It is a photovoltaic panel The current contribution coefficient.

[0062] The contribution analysis module is used to determine the contribution of the photovoltaic panel based on its voltage contribution coefficient and current contribution coefficient. for:

[0063] .

[0064] in, It is a photovoltaic panel The degree of contribution.

[0065] The attenuation analysis module is used to summarize the contributions of all photovoltaic panels to obtain a global contribution matrix. It then sequentially retrieves the contribution of each photovoltaic panel from the global contribution matrix, multiplies it by its corresponding spectral attenuation factor vector, and sums the results to obtain the global attenuation. for:

[0066] .

[0067] in, It is the total number of photovoltaic panels in the photovoltaic system. It is the first in the spectral attenuation factor vector The spectral attenuation factor corresponding to each optical segment It is the first Global attenuation of each optical segment.

[0068] Furthermore, the power prediction module includes a similarity determination module, a first prediction unit, a second prediction unit, and a third prediction unit connected in sequence.

[0069] The similarity determination module is used to obtain the weather index prediction data of the target prediction time point, calculate the difference coefficient between the weather index prediction data of the target prediction time point and the historical weather index data, select the time point with the smallest difference coefficient as the reference time point, and retrieve the light band energy density distribution data corresponding to the reference time point as the target light band energy density distribution data.

[0070] The first prediction unit is used to obtain solar position prediction data for the target prediction time point from the regional meteorological station, and to obtain the predicted value of the incident angle of light based on the solar position prediction data and the installation angle of the photovoltaic panel.

[0071] The predicted angle of light incidence is obtained based on the predicted solar position data and the installation angle of the photovoltaic panel. for:

[0072] .

[0073] in, This is the predicted value of the solar altitude angle. This is the predicted value of the solar azimuth angle. It refers to the installation angle of the photovoltaic panels. It is the azimuth angle of the photovoltaic panel. It is the predicted value of the angle of incidence of light.

[0074] The second prediction unit is used to obtain the total effective spectral energy prediction value based on the optical segment energy density values ​​of each optical segment and the predicted light incidence angle in the target optical segment energy density distribution data. for:

[0075] .

[0076] in, It is the first The target optical energy density of the segment optical band. It is the first Total predicted effective spectral energy for the segment.

[0077] The third prediction unit is used to calculate the overall predicted power of the photovoltaic module based on the global attenuation degree and the total predicted effective spectral energy. for:

[0078] .

[0079] in, It is the first The energy conversion coefficient of the optical segment, This is the overall predicted power of the photovoltaic module.

[0080] Furthermore, the weather indicators include: solar irradiance, cloud cover, temperature, and visibility; the similarity determination module includes an indicator normalization unit and a difference coefficient calculation unit connected in sequence.

[0081] The index normalization unit is used to normalize the predicted weather index data for each target prediction time point with the corresponding historical weather index using the maximum-minimum value normalization method to obtain normalized weather index data.

[0082] The difference coefficient calculation unit is used to calculate the difference coefficient by performing Euclidean distance calculation on the normalized weather index data according to the Euclidean distance calculation method.

[0083] Compared with the prior art, the beneficial effects of this invention are as follows: by discretely integrating the energy density of discrete wavelength points within a specific optical band, the energy density distribution data of the optical band is obtained; by generating an adversarial network to split the optical band response, the spectral attenuation factor vector is accurately extracted in an adversarial manner; and by further combining the hierarchical recursive analysis of the photovoltaic array topology, the global attenuation degree is obtained. Thus, it is possible to achieve more accurate monitoring and prediction of photovoltaic power under the condition of different attenuation of the photovoltaic panel's absorption rate in different optical bands. Attached Figure Description

[0084] Figure 1 is a schematic diagram of the module composition of a real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to Embodiment 1 of the present invention. Embodiments of the present invention

[0085] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0086] Example 1: As shown in Figure 1, this example provides a real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things. The system includes: a data acquisition module, an energy density conversion module, an attenuation identification module, a global fitting module, and a power prediction module, which are connected in sequence.

[0087] The data acquisition module is used to collect spectral incident data of the photovoltaic array area in real time through multispectral sensors arranged in a preset area of ​​the photovoltaic array, and to collect the current and voltage parameters of each photovoltaic panel in real time through current and voltage sensors to obtain the power of each photovoltaic panel.

[0088] For example: A photovoltaic system in a certain area has 50 photovoltaic panels, each with an area of ​​1.92m². 2 The photovoltaic array consists of five parallel strings of 10 photovoltaic panels connected in series, forming a 10×5 matrix structure. An Ocean Optics STS-UV multispectral sensor (wavelength range 280-1100nm, spectral resolution 2nm) is positioned at the center of the array to collect spectral incident data. ABB-VSN800 current and voltage sensors are installed on the photovoltaic panels to collect current and voltage parameters. Taking the first photovoltaic panel as an example, at a certain time, the current parameter of the panel is 8.73A and the voltage parameter is 38.21V. Multiplying the current and voltage parameters gives the power of this panel as 333.57W.

[0089] The energy density conversion module is used to divide the spectral incident data according to the light band type to obtain a light band dataset and perform energy density conversion to obtain light band energy density distribution data; the light band types include ultraviolet light band, visible light band, and near-infrared light band; the ultraviolet light band is the wavelength in... The visible light band is a light band in the nanometer range, wherein the visible light band has a wavelength in the nanometer range. The near-infrared light segment is a wavelength in the nanometer range. Nanoscale optical bands.

[0090] The attenuation identification module is used to calculate the effective spectral energy value of the optical segment based on the energy density distribution data of the optical segment and the installation angle of the photovoltaic panel. The effective spectral energy value of the optical segment and the power data of each photovoltaic panel are loaded into the photovoltaic attenuation identification model to obtain the spectral attenuation factor vector of each photovoltaic panel.

[0091] The global fitting module is used to perform hierarchical recursive analysis based on the photovoltaic array topology to obtain the global contribution matrix, and to calculate the global attenuation based on the global contribution matrix and the spectral attenuation factor vector of each photovoltaic panel.

[0092] The power prediction module is used to acquire weather index prediction data for the target prediction time point and determine the similarity of historical weather indexes to obtain a reference time point. The energy density distribution data of the optical segment corresponding to the reference time point is used as the energy density distribution data of the target optical segment. Based on the global attenuation degree and the energy density distribution data of the target optical segment, the overall predicted power of the photovoltaic module is obtained through power prediction.

[0093] It should be noted that the energy density conversion module includes a data preprocessing module, an energy density calculation module, a discrete integration module, and a data integration module connected in sequence.

[0094] The data preprocessing module is used to divide the spectral incident data into discrete wavelength points according to the optical segment type to obtain the irradiance data of the discrete wavelength points contained in each optical segment as the optical segment dataset.

[0095] For example, the 340nm wavelength point and its corresponding irradiance are assigned to the ultraviolet light dataset, and the 460nm wavelength point and its corresponding irradiance are assigned to the visible light dataset.

[0096] The energy density calculation module is used to calculate the corresponding energy density value based on the irradiance data of discrete wavelength points. for:

[0097] .

[0098] in, Discrete wavelength point Irradiance, It is Planck's constant. It's the speed of light. Discrete wavelength point wavelength, Discrete wavelength point The energy density value.

[0099] The Planck constant is 6.626 × 10⁻⁶. -34 J·s, where the speed of light is 3×10 8 m / s.

[0100] For example, taking the 340nm wavelength point as an example, the measured irradiance I 340 =0.42W / m 2 / nm, the energy density at that wavelength can be obtained. J nm / m 2 .

[0101] The discrete integration module is used to perform discrete integration on the energy density values ​​at discrete wavelength points within a specific optical band to obtain the optical band energy density value. for:

[0102] .

[0103] in, It refers to the spectral resolution of the multispectral sensor. It is the energy density value of the optical segment.

[0104] The data integration module is used to summarize the energy density values ​​of the ultraviolet, visible, and near-infrared light bands to obtain the light band energy density distribution data.

[0105] For example: the ultraviolet light range is for Discrete integration is performed on the energy density values ​​at discrete wavelength points within the wavelength range. When the spectral resolution of the multispectral sensor is 2 nm, then... The energy density values ​​of all wavelength points within the wavelength range with a resolution of 2 nm (280 nm, 282 nm, 284 nm, up to 398 nm) are discretely integrated to obtain the energy density value of the ultraviolet band. J / m 2 Similarly, the visible light energy density can be calculated to be 435.20 J / m². 2 The energy density in the near-infrared band is 268.10 J / m². 2 The energy density distribution data for the optical band is: [83.70, 435.20, 268.10] J / m². 2 .

[0106] It should be noted that the attenuation identification module includes an angle conversion module, an energy analysis module, and a generative adversarial network module connected in sequence.

[0107] The angle conversion module is used to obtain solar position data from regional weather stations via an API interface, and to obtain the incident angle of sunlight based on the solar position data and the installation angle of the photovoltaic panel. for:

[0108] .

[0109] in It is the solar altitude angle. It is the azimuth of the sun. It refers to the installation angle of the photovoltaic panels. It is the azimuth angle of the photovoltaic panel. It is the angle of incidence of light.

[0110] The angle of incidence of sunlight is the angle between the light ray and the normal to the surface of the photovoltaic panel. It determines how much sunlight can be effectively received by the photovoltaic panel. For example, if the solar position data at a certain time point is obtained through the API interface of a regional weather station: solar altitude angle α = 68.3°, solar azimuth angle γ = 135.7°, photovoltaic panel installation angle β = 30°, and azimuth angle δ = 180° (due south), then the angle of incidence of sunlight... .

[0111] The energy analysis module is used to obtain the effective spectral energy value of the optical segment based on the optical segment energy density value and the incident angle of the light in the optical segment energy density distribution data. for:

[0112] .

[0113] in, It refers to the area of ​​a single photovoltaic panel. It is the effective spectral energy value of the optical segment.

[0114] Combining the incident angle of light and the energy density of the light band, the effective spectral energy value of the ultraviolet band Similarly, the effective spectral energy value in the visible light band is 777.09 J, and the effective spectral energy value in the near-infrared band is 478.72 J.

[0115] A generative adversarial network module is used to load the effective spectral energy value of the optical segment and the power data of each photovoltaic panel into the photovoltaic attenuation identification model established by the generative adversarial network algorithm to obtain the spectral attenuation factor vector of each photovoltaic panel.

[0116] It should be noted that the generative adversarial network module includes an input unit, a split-segment response unit, a power reconstruction unit, and a discrimination unit connected in sequence.

[0117] The input unit is used to generate fused features by taking the effective spectral energy value of the optical band and the power data of each photovoltaic panel as input through the fully connected layer. for:

[0118] .

[0119] in, It refers to the power of the photovoltaic panel. and These are the weight matrix and bias terms of the fully connected layer. It is a feature of fusion.

[0120] The spectral segment response unit is used to set up independent sub-networks for the ultraviolet, visible, and near-infrared light segments to perform spectral segment response on the fused features. The spectral attenuation factor vector of each photovoltaic panel is obtained by LeakyReLU activation. for:

[0121] .

[0122] in, It is the spectral attenuation factor vector. and These are the weight matrices and bias terms of each optical segment network, and LeakyReLU is the activation function.

[0123] For example, for photovoltaic panel #1, the system inputs the effective spectral energy values ​​[149.45, 777.09, 478.72] and the measured power of 333.57 GHz into a pre-trained photovoltaic attenuation identification model. This model first integrates the input data into a single fused feature value through a feature fusion layer. Then, it calculates the attenuation factors for each light band using three sub-networks specifically designed for different light bands: 0.35 for the ultraviolet band, 0.12 for the visible band, and 0.08 for the near-infrared band. These three values ​​form the spectral attenuation factor vector [0.35, 0.12, 0.08] for the photovoltaic panel, representing the degree of attenuation in each light band.

[0124] The power reconfiguration unit is used to reconfigure the photovoltaic panel power through a fully connected regression network based on the spectral attenuation factor vector of the photovoltaic panel to obtain the reconfigured power of the photovoltaic panel.

[0125] The expression is:

[0126] .

[0127] and These are the weight matrix and bias terms of a fully connected regression network. It is the power reconfiguration of the photovoltaic panel.

[0128] The discrimination unit is used to input the absolute error between the reconstructed power of the photovoltaic panel and the power data of the photovoltaic panel into the multilayer perceptron to obtain the attenuation discrimination probability. When the attenuation discrimination probability is less than the preset discrimination threshold, the spectral attenuation factor vector of the photovoltaic panel is determined to be valid and stored. Otherwise, the parameters are updated and the spectral attenuation factor vector is regenerated through backpropagation of the loss function of the photovoltaic attenuation generator.

[0129] The multilayer perceptron uses the Sigmoid function, expressed as:

[0130] ;

[0131] in, , Here are the weight matrix and bias terms for the photovoltaic degradation discriminator. It is the attenuation of the discrimination probability;

[0132] The weight matrices and bias term parameters in each expression are automatically iteratively optimized during model execution.

[0133] For example, to verify the effectiveness of the spectral attenuation factor vector [0.35, 0.12, 0.08], the photovoltaic attenuation discriminator reconstructed the photovoltaic panel power result as 114.34 using this spectral attenuation factor vector. The discrimination probability obtained by the photovoltaic attenuation discriminator was 0.71, which was higher than the system's preset threshold of 0.15. Therefore, parameter updates were performed to regenerate the spectral attenuation factor vector. After multiple iterations, a new spectral attenuation factor vector [0.23, 0.08, 0.06] was obtained, and the corresponding discrimination probability dropped to 0.13, which was less than the preset discrimination threshold of 0.15, indicating that the reconstruction was accurate. Therefore, the spectral attenuation factor vector [0.23, 0.08, 0.06] was determined to be effective and saved. In this model, the attenuation discrimination probability represents the degree of reconstruction error. The smaller the value, the more accurate and reliable the attenuation factor.

[0134] It should be noted that the training steps for the photovoltaic attenuation identification model are as follows:

[0135] The effective spectral energy values ​​of the optical segment and the corresponding photovoltaic power data within the historical operating cycle are used as training data.

[0136] The parameters of the photovoltaic degradation generator and the photovoltaic degradation discriminator are initialized using the He initialization method. The initial learning rate of the photovoltaic degradation generator and the photovoltaic degradation discriminator is set, and the learning rate is updated using the cosine annealing learning rate scheduling strategy.

[0137] The parameters of the photovoltaic degradation generator and the photovoltaic degradation discriminator include the weight matrix and bias terms of each layer in the photovoltaic degradation generator and the photovoltaic degradation discriminator, with the initial learning rate of the photovoltaic degradation generator set to... Set the initial learning rate of the photovoltaic degradation discriminator to .

[0138] The photovoltaic degradation generator and photovoltaic degradation discriminator are iteratively trained using an alternating training strategy. Each training iteration includes:

[0139] The parameters of the photovoltaic degradation generator are fixed, and the photovoltaic degradation discriminator is continuously trained. The parameters of the photovoltaic degradation discriminator are updated according to the loss function of the photovoltaic degradation discriminator, and the parameters are updated by the optimizer with the learning rate of the photovoltaic degradation discriminator.

[0140] .

[0141] in, It is the decay of the discrimination probability. It is the loss function value of the photovoltaic degradation discriminator.

[0142] With fixed parameters for the photovoltaic degradation discriminator, a photovoltaic degradation generator is trained. The parameters of the photovoltaic degradation generator are updated according to its loss function. The Adam optimizer is used to update the parameters with the learning rate of the photovoltaic degradation generator. The loss function of the photovoltaic degradation generator is:

[0143] .

[0144] in, It is the reconfiguration power of the photovoltaic panel. These are the reconstruction error weighting coefficients. It refers to the power of the photovoltaic panel. It is the loss function value of the photovoltaic degradation generator.

[0145] The reconstruction error weighting coefficient is used to balance adversarial loss and power accuracy. The initial value of the reconstruction error weighting coefficient is set to 0.5, and it is gradually increased to 1 through linear updates. The linear updates are as follows:

[0146] .

[0147] in, It is the first Reconstruction error weighting coefficients for the next iteration These are the initial reconstruction error weighting coefficients. These are the final reconstruction error weighting coefficients. It is the iteration round. It represents the maximum number of iterations.

[0148] For example: when the maximum number of iterations is set to 200, the reconstruction error weighting coefficient is as follows: .

[0149] The iteration ends when the training reaches the preset maximum number of iterations.

[0150] The photovoltaic attenuation identification model was trained using the PyTorch framework and its hyperparameters were optimized using the Optuna tool.

[0151] It should be noted that the global fitting module includes a hierarchical analysis module, a contribution analysis module, and a decay analysis module connected in sequence.

[0152] The hierarchy resolution module is used to determine the series and parallel levels of each photovoltaic panel based on the photovoltaic array topology, and to obtain the voltage contribution coefficient of the photovoltaic panel based on its series level. for:

[0153] .

[0154] in, It is a photovoltaic panel The number of photovoltaic panels in the series connection level It is a photovoltaic panel The voltage contribution coefficient.

[0155] The current contribution coefficient of the photovoltaic panel is obtained based on the parallel connection level of the photovoltaic panel. for:

[0156] .

[0157] in, It is a photovoltaic panel The number of photovoltaic panels in the parallel layer, It is a photovoltaic panel The current contribution coefficient.

[0158] For example, taking photovoltaic panel number 1 as an example, since the photovoltaic array topology consists of 10 photovoltaic panels connected in series in each row, and a total of 5 parallel strings forming a 10×5 matrix structure photovoltaic array, the number of photovoltaic panels in its series layer is 10, and the number of photovoltaic panels in its parallel layer is 5. Therefore, the voltage contribution coefficient of this photovoltaic panel can be obtained. The current contribution coefficient of the photovoltaic panel .

[0159] The contribution analysis module is used to determine the contribution of the photovoltaic panel based on its voltage contribution coefficient and current contribution coefficient. for:

[0160] .

[0161] in, It is a photovoltaic panel The degree of contribution.

[0162] For example: when the voltage contribution coefficient of photovoltaic panel No. 1 Current contribution coefficient At that time, the contribution of the photovoltaic panel Similarly, the contribution of the remaining photovoltaic panels can be obtained.

[0163] The attenuation analysis module is used to summarize the contributions of all photovoltaic panels to obtain a global contribution matrix. It then sequentially retrieves the contribution of each photovoltaic panel from the global contribution matrix, multiplies it by its corresponding spectral attenuation factor vector, and sums the results to obtain the global attenuation. for:

[0164] .

[0165] in, It is the total number of photovoltaic panels in the photovoltaic system. It is the first in the spectral attenuation factor vector The spectral attenuation factor corresponding to each optical segment It is the first Global attenuation of each optical segment.

[0166] For example, taking the ultraviolet band as an example, since the above process yielded the spectral attenuation factor vector [0.23, 0.08, 0.06] for the first photovoltaic panel, the spectral attenuation factor corresponding to the ultraviolet band for that photovoltaic panel is... The contribution of photovoltaic panels Similarly, the spectral attenuation factor and contribution of the photovoltaic panel in the ultraviolet band can be obtained for the other photovoltaic panels. Therefore, the global attenuation in the ultraviolet band can be calculated as follows: If the solution result is 0.31, it means that the global attenuation of the ultraviolet band is 0.31.

[0167] It should be noted that the power prediction module includes a similarity determination module, a first prediction unit, a second prediction unit, and a third prediction unit connected in sequence.

[0168] The similarity determination module is used to obtain the weather index prediction data of the target prediction time point, calculate the difference coefficient between the weather index prediction data of the target prediction time point and the historical weather index data, select the time point with the smallest difference coefficient as the reference time point, and retrieve the light band energy density distribution data corresponding to the reference time point as the target light band energy density distribution data.

[0169] For example: the difference coefficient is calculated between the weather index prediction data at the target prediction time point and the historical weather index data to obtain the historical time point with the smallest difference coefficient, and the light band energy density distribution data at that time point is used as the target light band energy density distribution data. From the historical data, it can be seen that the light band energy density distribution data at that time point is [85.20,442.50,273.10], so the target light band energy density distribution data is [85.20,442.50,273.10].

[0170] The first prediction unit is used to obtain solar position prediction data for the target prediction time point from the regional meteorological station, and to obtain the predicted value of the incident angle of light based on the solar position prediction data and the installation angle of the photovoltaic panel.

[0171] The predicted angle of light incidence is obtained based on the predicted solar position data and the installation angle of the photovoltaic panel. for:

[0172] .

[0173] in, This is the predicted value of the solar altitude angle. This is the predicted value of the solar azimuth angle. It refers to the installation angle of the photovoltaic panels. It is the azimuth angle of the photovoltaic panel. It is the predicted value of the angle of incidence of light.

[0174] For example: obtaining the predicted solar position data for the target prediction time point from regional meteorological stations yields the predicted solar altitude angle value. =68.5°, predicted solar azimuth angle =136.2°, then the predicted angle of incidence is... .

[0175] The second prediction unit is used to obtain the total effective spectral energy prediction value based on the optical segment energy density values ​​of each optical segment and the predicted light incidence angle in the target optical segment energy density distribution data. for:

[0176] .

[0177] in, It is the first The target optical energy density of the segment optical band. It is the first Total predicted effective spectral energy for the segment.

[0178] For example: Total predicted effective spectral energy in the ultraviolet band J, the total predicted effective spectral energy in the visible light band J, Total predicted effective spectral energy in the infrared band J.

[0179] The third prediction unit is used to calculate the overall predicted power of the photovoltaic module based on the global attenuation degree and the total predicted effective spectral energy. for:

[0180] .

[0181] in It is the first The energy conversion coefficient of the optical segment, This is the overall predicted power of the photovoltaic module.

[0182] The energy conversion coefficient The energy conversion coefficients for the ultraviolet, visible, and near-infrared light bands are respectively determined through a spectral efficiency experiment of the photovoltaic panel under standard test conditions. The energy conversion coefficients for the ultraviolet, visible, and near-infrared light bands are respectively set as follows: , , The aforementioned global attenuation values ​​assume that the global attenuation is 0.31 for the ultraviolet band, 0.09 for the visible band, and 0.07 for the infrared band, resulting in a predicted overall power of the photovoltaic array.

[0183] It should be noted that the weather indicators include: solar irradiance, cloud cover, temperature, and visibility; the similarity determination module includes an indicator normalization unit and a difference coefficient calculation unit connected in sequence.

[0184] The index normalization unit is used to normalize the predicted weather index data for each target prediction time point with the corresponding historical weather index using the maximum-minimum value normalization method to obtain normalized weather index data.

[0185] The difference coefficient calculation unit is used to calculate the difference coefficient by performing Euclidean distance calculation on the normalized weather index data according to the Euclidean distance calculation method.

[0186] For example: when the weather index at the target forecast time is: solar irradiance 950W / m 2 Cloud cover 15%, temperature 32℃, visibility 15km. Historical weather indicators for a specific day include: solar irradiance 920W / m². 2 Cloud cover 15%, temperature 31℃, visibility 14km, and solar irradiance ranging from 600-1100W / m² in historical data. 2 The solar irradiance at the target prediction time point is 950 W / m². 2 The normalized value is The solar irradiance at this historical point in time was 920 W / m². 2 The normalized value is When cloud cover ranges from 0-100% in historical data, the cloud cover at the target prediction time point and the cloud cover at historical time points, after normalization by 15%, are the values. Similarly, when the temperature range in historical data is 5-40℃, the normalized values ​​for the predicted temperature and the historical temperature are 0.77 and 0.74, respectively. When the visibility range in historical data is 0.1-50km, the normalized values ​​for the predicted visibility and the historical visibility are 0.30 and 0.28, respectively. The difference coefficient between the two values ​​is calculated using the Euclidean distance method.

[0187] .

[0188] If a particular date has the smallest difference coefficient (i.e., the highest similarity) among historical dates, then it is selected as the reference time point.

[0189] Finally, it should be noted that although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things, characterized in that, The system includes: a data acquisition module, an energy density conversion module, an attenuation identification module, a global fitting module, and a power prediction module, which are connected in sequence. The data acquisition module is used to collect spectral incident data of the photovoltaic array area in real time through multispectral sensors arranged in a preset area of ​​the photovoltaic array, and to collect the current and voltage parameters of each photovoltaic panel in real time through current and voltage sensors to obtain the photovoltaic power of each photovoltaic panel. The energy density conversion module is used to divide the spectral incident data according to the light segment type to obtain the light segment dataset and perform energy density conversion to obtain the light segment energy density distribution data; the light segment types include ultraviolet light segment, visible light segment and near-infrared light segment; The attenuation identification module is used to calculate the effective spectral energy value of the optical segment based on the optical segment energy density distribution data and the photovoltaic panel installation angle, and load the effective spectral energy value of the optical segment and the power data of each photovoltaic panel into the photovoltaic attenuation identification model to obtain the spectral attenuation factor vector of each photovoltaic panel; The global fitting module is used to perform hierarchical recursive analysis based on the photovoltaic array topology to obtain the global contribution matrix, and to calculate the global attenuation based on the global contribution matrix and the spectral attenuation factor vector of each photovoltaic panel. The power prediction module is used to acquire weather index prediction data for the target prediction time point and determine the similarity of historical weather indexes to obtain a reference time point. The energy density distribution data of the optical segment corresponding to the reference time point is used as the energy density distribution data of the target optical segment. Based on the global attenuation degree and the energy density distribution data of the target optical segment, the overall predicted power of the photovoltaic module is obtained through power prediction.

2. The real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to claim 1, characterized in that, The energy density conversion module also includes a data preprocessing module, an energy density calculation module, a discrete integration module, and a data integration module connected in sequence. The data preprocessing module is used to divide the spectral incident data into discrete wavelength points according to the optical segment type to obtain the irradiance data of the discrete wavelength points contained in each optical segment as the optical segment dataset. The energy density calculation module is used to calculate the corresponding energy density value based on the irradiance data of discrete wavelength points. for: ; in Discrete wavelength point Spectral irradiance, It is Planck's constant. It's the speed of light. Discrete wavelength point wavelength, Discrete wavelength point The energy density value; The discrete integration module is used to perform discrete integration on the energy density values ​​at discrete wavelength points within a specific optical band to obtain the optical band energy density value. for: ; in It refers to the spectral resolution of the multispectral sensor. It is the energy density value of the optical segment; The data integration module is used to summarize the energy density values ​​of the ultraviolet, visible, and near-infrared light bands to obtain the light band energy density distribution data.

3. The real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to claim 1, characterized in that, The attenuation identification module also includes an angle conversion module, an energy analysis module, and a generative adversarial network module connected in sequence. The angle conversion module is used to obtain solar position data from regional weather stations via an API interface, and to obtain the incident angle of sunlight based on the solar position data and the installation angle of the photovoltaic panel. for: ; in It is the solar altitude angle. It is the azimuth of the sun. It refers to the tilt angle of the photovoltaic panels. It is the azimuth angle of the photovoltaic panel. It is the angle of incidence of light; The energy analysis module is used to obtain the effective spectral energy value of the optical segment based on the optical segment energy density value and the incident angle of the light in the optical segment energy density distribution data. for: ; in It refers to the area of ​​a single photovoltaic panel. It is the effective spectral energy value of the optical segment; A generative adversarial network module is used to load the effective spectral energy value of the optical segment and the power data of each photovoltaic panel into the photovoltaic attenuation identification model established by the generative adversarial network algorithm to obtain the spectral attenuation factor vector of each photovoltaic panel.

4. The real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to claim 3, characterized in that, The generative adversarial network module further includes an input unit, a beam splitting response unit, a power reconstruction unit, and a discrimination unit connected in sequence; The input unit is used to generate fused features by taking the effective spectral energy value of the optical band and the power data of each photovoltaic panel as input through the fully connected layer. for: ; in It refers to the power of the photovoltaic panel. and These are the weight matrix and bias terms of the fully connected layer. It is a feature of fusion; The spectral segment response unit is used to set up independent sub-networks for the ultraviolet, visible, and near-infrared light segments to perform spectral segment response on the fused features. The spectral attenuation factor vector of each light segment is obtained by LeakyReLU activation. for: ; in It is the spectral attenuation factor vector. and These are the weight matrices and bias terms of each optical segment subnetwork, and LeakyReLU is the activation function; The power reconfiguration unit is used to reconfigure the photovoltaic panel power through a fully connected regression network based on the spectral attenuation factor vector of the photovoltaic panel to obtain the reconfigured power of the photovoltaic panel. The discrimination unit is used to input the absolute error between the reconstructed power of the photovoltaic panel and the power data of the photovoltaic panel into the multilayer perceptron to obtain the attenuation discrimination probability. When the attenuation discrimination probability is less than the preset discrimination threshold, the spectral attenuation factor vector of the photovoltaic panel is determined to be valid and stored. Otherwise, the parameters are updated and the spectral attenuation factor vector is regenerated through backpropagation of the loss function of the photovoltaic attenuation generator.

5. A real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to claim 3, characterized in that, The training steps for the photovoltaic attenuation identification model are as follows: The effective spectral energy values ​​of the optical segment and the corresponding photovoltaic power data within the historical operating cycle are used as training data; The parameters of the photovoltaic attenuation generator and the photovoltaic attenuation discriminator are initialized using the He initialization method, the initial learning rate of the photovoltaic attenuation generator and the photovoltaic attenuation discriminator is set, and the learning rate is updated using the cosine annealing learning rate scheduling strategy. The photovoltaic degradation generator and photovoltaic degradation discriminator are iteratively trained using an alternating training strategy. Each training iteration includes: The parameters of the photovoltaic attenuation generator are fixed, the photovoltaic attenuation discriminator is continuously trained, the discriminator parameters are updated according to the photovoltaic attenuation discriminator loss function, and the parameters are updated by the optimizer with the discriminator learning rate. ; in It is the decay of the discrimination probability. It is the loss function value of the photovoltaic degradation discriminator; With fixed parameters for the photovoltaic degradation discriminator, a photovoltaic degradation generator is trained. The generator parameters are updated according to the generator loss function, and the Adam optimizer is used to update the parameters with the generator learning rate. The photovoltaic degradation generator loss function is: ; in It is the reconfiguration power of the photovoltaic panel. These are the reconstruction error weighting coefficients. It refers to the power of the photovoltaic panel. It is the value of the photovoltaic degradation generator loss function; The iteration ends when the training reaches the preset maximum number of iterations.

6. The real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to claim 4, characterized in that, The global fitting module also includes a hierarchical parsing module, a contribution analysis module, and a decay analysis module connected in sequence. The hierarchy resolution module is used to determine the series and parallel levels of each photovoltaic panel based on the photovoltaic array topology, and to obtain the voltage contribution coefficient of the photovoltaic panel based on its series level. for: ; in It is a photovoltaic panel The number of photovoltaic panels in the series connection level It is a photovoltaic panel Voltage contribution coefficient; The current contribution coefficient of the photovoltaic panel is obtained based on the parallel connection level of the photovoltaic panel. for: ; in It is a photovoltaic panel The number of photovoltaic panels in the parallel layer, It is a photovoltaic panel The current contribution coefficient; The contribution analysis module is used to determine the contribution of the photovoltaic panel based on its voltage contribution coefficient and current contribution coefficient. for: ; in It is a photovoltaic panel The degree of contribution; The attenuation analysis module is used to summarize the contributions of all photovoltaic panels to obtain a global contribution matrix. It then sequentially retrieves the contribution of each photovoltaic panel from the global contribution matrix, multiplies it by its corresponding spectral attenuation factor vector, and sums the results to obtain the global attenuation. for: ; in It is the total number of photovoltaic panels in the photovoltaic system. It is the first in the spectral attenuation factor vector The spectral attenuation factor corresponding to each optical segment It is the first Global attenuation of each optical segment.

7. A real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to claim 6, characterized in that, The power prediction module further includes a similarity determination module, a first prediction unit, a second prediction unit, and a third prediction unit connected in sequence. The similarity determination module is used to obtain the weather index prediction data for the target prediction date, calculate the difference coefficient between the weather index prediction data for the target prediction date and the historical weather index data, select the date with the smallest difference coefficient as the reference date, and retrieve the optical band energy density distribution data corresponding to the reference date as the target optical band energy density distribution data. The first prediction unit is used to obtain solar position prediction data for the target time point from the regional meteorological station, and to obtain the predicted value of the incident angle of light based on the solar position prediction data and the installation angle of the photovoltaic panel. The second prediction unit is used to obtain the total effective spectral energy prediction value based on the optical segment energy density values ​​of each optical segment and the predicted light incidence angle in the target optical segment energy density distribution data. for: ; in It is the first The target optical energy density of the segment optical band. It is the first Total predicted effective spectral energy of the optical segment; The third prediction unit is used to calculate the overall predicted power of the photovoltaic module based on the global attenuation degree and the total predicted effective spectral energy. for: ; in It is the first The energy conversion coefficient of the optical segment, This is the overall predicted power of the photovoltaic module.

8. A real-time monitoring and prediction system for new energy photovoltaic power based on the Internet of Things according to claim 7, characterized in that, The weather indicators include: solar irradiance, cloud cover, temperature, and visibility; the similarity determination module also includes an indicator normalization unit and a difference coefficient calculation unit connected in sequence. The index normalization unit is used to normalize the weather index forecast data for each target forecast date with the historical weather index of the corresponding category using the maximum and minimum value normalization method to obtain normalized weather index data. The difference coefficient calculation unit is used to calculate the difference coefficient by performing Euclidean distance calculation on the normalized weather index data according to the Euclidean distance calculation method.