Ultra-short-term forecast method of DNI based on ground-based cloud images

Abstract

The present invention discloses a DNI ultra-short-term forecasting method based on ground-based cloud maps, which relates to the technical field of DNI forecasting, and includes the following steps: integrating DNI historical meteorological observation data and pollutant data, and establishing a DNI attenuation ratio model based on pollutant concentration after preprocessing; calculating the DNI relative attenuation value as the dependent variable through meteorological data under clear sky and cloud conditions; obtaining different pollutants and combining them, and identifying the initial pollutant combination using variance analysis; training the DNI attenuation ratio model based on the initial combination, and gradually increasing the pollutant combination to construct a pollutant training output set. Acceptable and general pollutant combinations are screened out through data analysis, providing a solid foundation for DNI prediction; calculating the attenuation value of clouds to DNI. Based on the DNI prediction sub-models of different pollutant types, the initial DNI prediction value is corrected to obtain the final DNI prediction result. This method improves the accuracy and robustness of DNI prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of ultra-short-term prediction, and more specifically, to an ultra-short-term DNI prediction method based on ground-based cloud images. Background Art

[0002] Solar energy is a fluctuating power source. It is affected by the geometric position of the sun and weather processes, and has the characteristics of seasonality, randomness, and intermittency. Therefore, DNI exhibits randomness and volatility. This fluctuation will cause uneven heating of the heat absorber of the CSP power station, affect the flow rate of the lava in the pipeline, and directly affect the safe operation of the CSP power station. Predicting the changes in DNI in advance and taking measures in advance will help ensure the safety of power generation production. Therefore, through the study of DNI, accurately predicting DNI changes has important practical significance for the safe production of power generation systems. However, modeling the nonlinear relationship between pollutant concentration and DNI is still a challenge, especially when dealing with the interaction of multiple pollutants in the atmosphere. In addition, the current traditional DNI ultra-short-term forecast does not take into account the absorption and reflection of DNI by clouds, aerosols, water vapor and other substances during the process of DNI passing through the atmosphere, and ignores the influence of clouds on DNI values, resulting in inaccurate DNI value forecasts. In response to the above problems, the present invention proposes a solution. Summary of the invention

[0003] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a DNI ultra-short-term forecasting method based on ground-based cloud maps, which solves the problem of inaccurate traditional DNI ultra-short-term forecasting.

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A DNI ultra-short-term forecasting method based on ground-based cloud maps includes the following steps: obtaining DNI historical meteorological observation data and pollutant data and preprocessing them, wherein the DNI historical meteorological data includes meteorological observation data under historical clear sky conditions, meteorological observation data under historical cloud conditions, historical DNI observation data and extraterrestrial radiation data; calculating the DNI relative attenuation value based on a preset clear sky atmosphere basic DNI attenuation proportional model, and using the DNI relative attenuation value as a dependent variable and each pollutant concentration as an independent variable to construct a DNI attenuation proportional model based on pollutant concentration; obtaining different pollutant data, and arranging and combining the pollutant data to obtain several pollutant combinations, calculating the combined evaluation value of the pollutant combination based on variance analysis, and performing data analysis to obtain an initial pollutant combination; inputting the initial pollutant combination into The DNI attenuation ratio model based on pollutant concentration is trained, the pollutant combinations for training are randomly and gradually increased from several pollutant combinations, and the DNI attenuation ratio models of several pollutant concentrations are used to construct a pollutant training output set; data analysis is performed on the pollutant training output set to screen out an acceptable general-purpose DNI attenuation ratio model based on pollutant concentration; meteorological observation data under historical cloud conditions are input into the general-purpose DNI attenuation ratio model based on pollutant concentration to obtain the DNI attenuation ratio value of the clear sky atmosphere, and the cloud-to-DNI attenuation ratio value is obtained based on the attenuation ratio calculation method; DNI historical meteorological observation data is input into the DNI prediction submodel to obtain the initial DNI prediction value, and the initial DNI prediction value is corrected based on the cloud-to-DNI attenuation ratio value to obtain the final DNI prediction value.

[0006] In a preferred embodiment, the preset clear sky atmosphere basic DNI attenuation ratio model is specifically: based on the SMARTS clear sky radiation transmission model, an atmospheric model is selected and the spectral range of the model and the geographical location parameters of the observation point are set; meteorological observation data and historical DNI observation data under historical clear sky conditions are input into the atmospheric model for training and verification to obtain the clear sky atmosphere basic DNI attenuation ratio model.

[0007] In a preferred embodiment, the DNI relative attenuation value is calculated based on a preset clear sky atmosphere basic DNI attenuation ratio model, specifically: DNI historical meteorological observation data is input into a preset clear sky atmosphere basic DNI attenuation ratio model to obtain a baseline DNI value; an attenuation factor for each pollutant is set, and based on the DNI calculation formula under pollutant conditions, the DNI value under pollutant-containing conditions is obtained; based on the baseline DNI value and the DNI value under pollutant-containing conditions, the DNI relative attenuation value is calculated.

[0008] In a preferred embodiment, the DNI relative attenuation value is used as the dependent variable and the concentrations of various pollutants are used as independent variables to construct a DNI attenuation ratio model based on pollutant concentration, specifically: based on the DNI relative attenuation value as the dependent variable and the concentrations of various pollutants as independent variables, a DNI linear regression equation is constructed to obtain a DNI linear regression model; the DNI historical meteorological observation data and pollutant data are divided into a training set and a test set according to a preset ratio; the training set is input into the DNI linear regression model for model training to obtain a trained DNI linear regression model; the trained DNI linear regression model is verified using the test set to obtain a trained DNI linear regression model.

[0009] In a preferred embodiment, different pollutant data are obtained, and the pollutant data are arranged and combined to obtain several pollutant combinations, combined evaluation values ​​of the pollutant combinations are calculated based on variance analysis, and data analysis is performed to obtain an initial pollutant combination, specifically: different pollutant data are obtained, and the pollutant data are arranged and combined to obtain several pollutant combinations; according to the number and repeatability of pollutant combinations, the ANOVA statistical method is selected to respectively calculate the between-group variance and the within-group variance between several pollutant combinations; the ratio of the between-group variance to the within-group variance is used as the combined evaluation value, and the combined evaluation values ​​are sorted in descending order to select the largest combined evaluation value; the pollutant combination corresponding to the largest combined evaluation value is used as the initial pollutant combination.

[0010] In a preferred embodiment, the data analysis of the pollutant training output set is performed to screen out an acceptable general-purpose DNI attenuation ratio model based on pollutant concentration, specifically: the model output mean and model output standard deviation corresponding to the pollutant training output set are calculated, and the model output mean and model output standard deviation are compared with a preset output mean reference value and a preset output standard deviation reference value; the pollutant combination corresponding to a model output mean greater than the preset output mean reference value and a model output standard deviation greater than the preset output standard deviation reference value is taken as a general-purpose pollutant combination; the general-purpose DNI attenuation ratio model based on pollutant concentration with the largest number among all general-purpose pollutant combinations is used to predict DNI.

[0011] In a preferred embodiment, the initial DNI prediction value is corrected based on the cloud-based DNI attenuation ratio value to obtain the final DNI prediction value, specifically: a correction coefficient is calculated based on the cloud-based DNI attenuation ratio value; based on the correction coefficient, the initial DNI prediction value is calculated using an exponential function to obtain the final DNI prediction value.

[0012] The technical effects and advantages of the neural network-based intelligent city planning decision-making auxiliary method and system of the present invention are as follows:

[0013] 1. This method systematically acquires and processes DNI historical meteorological observation data and pollutant data, and combines meteorological observations under clear sky conditions and cloud conditions to establish a DNI attenuation ratio model based on pollutant concentration. First, the DNI relative attenuation value is calculated using historical clear sky and cloud data, providing a reliable dependent variable for the model. Then, the initial pollutant combination is identified through the permutation and combination of different pollutants and variance analysis, thereby ensuring that the selected combination is statistically significant. This process provides a scientific basis for model training, gradually introduces multiple pollutant combinations and constructs a pollutant training output set, enhancing the adaptability and generalization ability of the model.

[0014] During the model training process, acceptable general-purpose pollutant combinations were screened out, making the final DNI attenuation ratio model more reliable and accurate in practical applications. In addition, the accuracy of DNI predictions was further improved by inputting historical cloud data into the model and calculating the correction of DNI attenuation based on clouds. This systematic approach not only improves the understanding of DNI changes under different atmospheric conditions, but also enhances the model's ability to respond to the impact of pollutants, providing more comprehensive and accurate DNI predictions for practical applications, which is of great significance for environmental monitoring and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a structural schematic diagram of the DNI ultra-short-term forecasting method based on ground-based cloud images of the present invention. DETAILED DESCRIPTION

[0016] Embodiment 1, Figure 1 The present invention provides a DNI ultra-short-term forecasting method based on ground-based cloud images, which is characterized by comprising the following steps:

[0017] S1, obtaining DNI historical meteorological observation data and pollutant data and preprocessing them. The DNI historical meteorological data includes historical meteorological observation data under clear sky conditions, historical meteorological observation data under cloud conditions, historical DNI observation data and extraterrestrial radiation data;

[0018] In this example, the historical meteorological observation data under clear sky conditions, the historical meteorological observation data under cloud conditions, the historical DNI observation data and the extraterrestrial radiation data are specifically:

[0019] The meteorological observation data under historical clear weather conditions include temperature, relative humidity, water vapor, and ozone concentration; the meteorological observation data under historical cloudy conditions include temperature, relative humidity, water vapor, ozone concentration, cloud amount, cloud optical thickness, and cloud type; the historical DNI observation data include DNI measurement values ​​and measurement times; the extraterrestrial radiation data refers to the amount of solar radiation reaching the outside of the earth without the influence of the atmosphere.

[0020] In this example, the pollutant data is specifically:

[0021] Pollutant data include various pollutants in the atmosphere such as PM, carbon dioxide, and ammonia.

[0022] It should be noted that obtaining pollutant data has significant advantages in DNI (direct normal radiation) prediction. First, pollutants (such as PM2.5, PM10, NO2 and SO2) can scatter and absorb solar radiation, which directly affects the intensity of DNI. Understanding the changes in the concentration of these pollutants can help researchers more accurately evaluate the optical properties of the atmosphere and quantify their attenuation of radiation. This information is crucial for optimizing the performance of solar power generation, because accurate DNI predictions can help power scheduling and system configuration. Second, real-time monitoring and analysis of pollutant data can improve the response speed and accuracy of the model, especially in environments with rapid weather changes. By combining pollutant data with meteorological information, more complex prediction models can be constructed, such as machine learning and deep learning models, which can capture more complex nonlinear relationships and thus improve prediction accuracy. In addition, changes in pollutant concentrations are often closely related to meteorological conditions. Therefore, integrating these data can better understand the impact of climate change on radiation. In summary, obtaining pollutant data can not only improve the accuracy of DNI predictions, but also provide a scientific basis for the efficient use of solar energy and the formulation of environmental policies, and promote the development of sustainable energy.

[0023] S2, based on the preset clear sky atmospheric basic DNI attenuation ratio model, calculate the DNI relative attenuation value, and use the DNI relative attenuation value as the dependent variable and the concentration of each pollutant as the independent variable to construct a DNI attenuation ratio model based on pollutant concentration;

[0024] In this example, the preset clear sky atmospheric basic DNI attenuation ratio model is as follows:

[0025] Based on the SMARTS clear-sky radiation transfer model, select an atmospheric model and set the model's spectral range and the geographical location parameters of the observation point;

[0026] The meteorological observation data under historical clear sky conditions and historical DNI observation data were input into the atmospheric model for training and verification, and the clear sky atmospheric basic DNI attenuation ratio model was obtained.

[0027] In this example, the SMARTS clear-sky radiation transfer model is:

[0028] SMARTS is an advanced model for simulating atmospheric radiation transfer, which is widely used in meteorology, environmental monitoring, remote sensing and military fields. It is based on the radiation transfer equation, takes into account factors such as gas absorption, scattering and surface reflection, and provides accurate calculation of radiation intensity in the atmosphere. The advantages of SMARTS include its high-resolution atmospheric model, flexible input parameter settings, and support for a variety of gas composition and spectral data, enabling it to provide accurate radiation transfer prediction and analysis in different applications.

[0029] It should be noted that for DNI, the main bands of solar radiation are usually of concern, including ultraviolet light of 200–400nm, visible light of 400–700nm, and near infrared of 700–2500nm, with a wavelength coverage range of 280-4000nm; the geographical location parameters of the observation point include the latitude and longitude of the observation point and the solar zenith angle.

[0030] In this example, the DNI relative attenuation value is calculated based on the preset clear sky atmospheric basic DNI attenuation ratio model, specifically:

[0031] Input DNI historical meteorological observation data into the preset clear sky atmosphere basic DNI attenuation ratio model to obtain the benchmark DNI value;

[0032] Set the attenuation factor for each pollutant, and based on the DNI calculation formula under pollutant conditions, obtain the DNI value under pollutant-containing conditions;

[0033] The DNI relative attenuation value was calculated based on the baseline DNI value and the DNI value under pollutant conditions.

[0034] In this example, the attenuation factor of each pollutant is set, and the DNI value under pollutant conditions is obtained based on the DNI calculation formula under pollutant conditions, specifically:

[0035]

[0036] in, is the DNI value under pollutant conditions, is the baseline DNI value, is the attenuation factor of the pollutant, is the concentration of the pollutant.

[0037] It should be noted that setting the attenuation factor for each pollutant is an important strategy to improve the prediction accuracy of DNI (direct normal radiation). The characteristics of each pollutant and its impact on radiation are different. For example, particulate matter such as PM2.5 and PM10 mainly reduce the solar radiation reaching the ground through scattering and absorption, while gaseous pollutants such as NO2 and SO2 may affect DNI through chemical reactions and absorption. Therefore, setting an attenuation factor for each pollutant can more accurately quantify its actual impact on DNI under specific meteorological conditions. Through this refined method, we can establish a more complex and more adaptable prediction model that can reflect the radiation attenuation effect caused by changes in pollutant concentrations in real time. At the same time, using the attenuation factor of the pollutant for calculation can enhance the interpretability of the model and help researchers better understand the different effects of various pollutants on solar radiation. In addition, this method can also be combined with meteorological data to form a dynamically adjusted prediction system to cope with changing environmental conditions. This refined processing method not only improves the prediction accuracy of DNI, but also provides a scientific basis for environmental monitoring and pollution control, and promotes the efficient use and sustainable development of renewable energy. Therefore, setting the attenuation factor for each pollutant is of great significance to improving the reliability of DNI prediction.

[0038] In this example, the DNI relative attenuation value is calculated based on the baseline DNI value and the DNI value under pollutant conditions, specifically:

[0039]

[0040] in, is the relative attenuation value of DNI, is the DNI value under pollutant conditions, is the benchmark DNI value.

[0041] It should be noted that by calculating the relative attenuation value of DNI based on the baseline DNI value and the DNI value under pollutant conditions, the actual impact of pollutants on solar radiation can be accurately evaluated. The advantage of this method is that it quantifies the degree of attenuation of radiation by different pollutants by comparing the impact of pollutants with the standard DNI value under pollution-free conditions, thus providing a concise and efficient analysis method. This not only makes DNI predictions more accurate, but also can reflect the specific impact of changes in pollutant concentrations on DNI in real time, which helps to optimize the utilization of solar energy. In addition, calculating relative attenuation values ​​can also help researchers better understand the relationship between atmospheric conditions and pollutants, and provide a scientific basis for environmental monitoring and policy making.

[0042] In this example, the relative attenuation value of DNI is used as the dependent variable, and the concentration of each pollutant is used as the independent variable to construct a DNI attenuation ratio model based on pollutant concentration, specifically:

[0043] Based on the relative attenuation value of DNI as the dependent variable and the concentration of each pollutant as the independent variable, the DNI linear regression equation was constructed to obtain the DNI linear regression model;

[0044] The DNI historical meteorological observation data and pollutant data are divided into training sets and test sets according to the preset ratio;

[0045] The training set is input into the DNI linear regression model to optimize the parameters for model training, and the trained DNI linear regression model is obtained;

[0046] The trained DNI linear regression model was tested using the test set to verify the trained DNI linear regression model.

[0047] S3, obtaining different pollutant data, and arranging and combining the pollutant data to obtain several pollutant combinations, calculating the combined evaluation value of the pollutant combination based on variance analysis, and performing data analysis to obtain the initial pollutant combination, specifically:

[0048] Obtain different pollutant data, and arrange and combine the pollutant data to obtain several pollutant combinations;

[0049] According to the number and repeatability of pollutant combinations, the ANOVA statistical method was selected to calculate the inter-group variance and intra-group variance of several pollutant combinations;

[0050] The ratio of the inter-group variance to the intra-group variance is used as a combined evaluation value, and the combined evaluation values ​​are sorted in descending order to select the largest combined evaluation value;

[0051] The pollutant combination corresponding to the maximum combined evaluation value is used as the initial pollutant combination.

[0052] In this example, based on the number and repeatability of pollutant combinations, the ANOVA statistical method is selected to calculate the inter-group variance and intra-group variance of several pollutant combinations, specifically:

[0053] List all permutations and combinations, check whether there are repeated pollutants in each combination, or whether there are multiple factors (i.e. pollutants) that affect DNI at the same time, and then calculate the number of permutations and combinations of each pollutant;

[0054] Different ANOVAs were selected based on different numbers of pollutant permutations and combinations.

[0055] It should be noted that ANOVA (Analysis of Variance) is a statistical method used to compare whether the differences in means between multiple groups are significant. It determines the degree of influence of different factors or combinations on the response variable by analyzing the variance between different groups (between-group variance) and the variance within the group (within-group variance). In variance analysis, a statistic (such as F value) is usually calculated, which represents the ratio of the between-group variance to the within-group variance. If the F value is large, it means that the difference between groups is significantly greater than the difference within the group, indicating that the mean difference between different groups is unlikely to be caused by random factors and is therefore statistically significant. This method is very useful in processing experimental data and multifactor analysis. It is widely used in fields such as agriculture, industry, medicine, and environmental science to analyze the impact of multiple factors on a certain outcome variable.

[0056] Based on different numbers of pollutant combinations, different ANOVAs are selected. When only one pollutant affects DNI at different levels (such as different concentrations), one-way ANOVA can be used to test the impact of the pollutant level on DNI; when the combination contains two different pollutants (such as PM2.5 and NO2) and there are no repeated measurements, two-way ANOVA can be used to analyze their main effects and interactions; when the combination contains three or more different pollutants (such as PM2.5, NO2, O3) and there are no repeated measurements, multi-way ANOVA can be used to analyze the main effects and interactions of multiple pollutants.

[0057] In this example, the ANOVA statistical method is selected to calculate the inter-group variance and intra-group variance of several pollutant combinations, specifically:

[0058]

[0059]

[0060]

[0061] in, is the between-group variance, is the number of pollutant combinations, is the number of samples of the ith pollutant combination, is the overall average DNI attenuation value of all pollutant combinations, is the average DNI attenuation value of the i-th pollutant combination, is the within-group variance, is the DNI attenuation value of the jth sample in the ith pollutant combination, and R is the combined evaluation value.

[0062] It should be noted that by calculating the ratio of the inter-group variance to the intra-group variance between pollutant combinations as the combined evaluation value, the difference in the impact of each pollutant combination on DNI (direct normal irradiance) can be scientifically and effectively evaluated. The advantage of this method is that combinations with larger inter-group variances indicate that different pollutant conditions have a significant impact on DNI, while smaller intra-group variances indicate that the response of DNI under the same combination conditions changes less, so that pollutant combinations with significant impact on DNI and high data consistency can be clearly identified. Selecting the combination with the largest combined evaluation value as the initial pollutant combination helps to improve the accuracy and robustness of the prediction model. Since this method has optimized the significance of the impact on DNI in the selection of the initial combination, the subsequent model has better adaptability and interpretability when dealing with the impact of atmospheric pollution on DNI. This variance ratio-based method provides a scientific basis for the effective screening of pollutant combinations and significantly improves the reliability and practicality of the prediction.

[0063] S4, training the DNI attenuation ratio model based on pollutant concentration based on the initial pollutant combination, randomly and gradually adding the trained pollutant combination from a plurality of pollutant combinations, and constructing a pollutant training output set from the output of the DNI attenuation ratio model of the outputted plurality of pollutant concentrations;

[0064] It should be noted that by training the DNI attenuation ratio model based on the initial pollutant combination and gradually adding other pollutant combinations for training, this method can dynamically enrich the training data, making the model more adaptable and generalizable to changes in pollutant combinations. In the process of gradually introducing different pollutant combinations, the model gradually learns the attenuation characteristics of diverse pollutants on DNI, which helps the model better capture the nonlinear relationship under complex atmospheric conditions. The outputs of multiple trained DNI attenuation ratio models are combined into a pollutant training output set, so that the model can not only handle the known pollutant impacts, but also infer the impact of unseen combinations on DNI, thereby improving the comprehensiveness and stability of the prediction. In addition, while avoiding model overfitting, this method can continuously accumulate diverse data features, so that the model has higher prediction accuracy and wider applicability for new pollutant combinations in future practical applications.

[0065] S5, performing data analysis on the pollutant training output set, and screening out the DNI attenuation ratio model of pollutant concentration corresponding to the pollutant combination with acceptable generality;

[0066] In this example, data analysis is performed on the pollutant training output set to screen out the DNI attenuation ratio model of pollutant concentration corresponding to the pollutant combination with acceptable generality, specifically:

[0067] Calculate the model output mean value and the model output standard deviation in the pollutant training output set, and compare the model output mean value and the model output standard deviation with a preset output mean value reference value and a preset output standard deviation reference value;

[0068] The pollutant combination corresponding to the model output average value greater than the preset output average value reference value and the model output standard deviation greater than the preset output standard deviation reference value is regarded as the pollutant combination with acceptable general applicability;

[0069] The DNI attenuation ratio model of the pollutant concentration corresponding to the acceptable general pollutant combination with the largest number of pollutant combinations among all acceptable general pollutant combinations is used to predict DNI.

[0070] It should be noted that this method calculates the mean and standard deviation of the output of each model in the pollutant training output set and compares them with the preset reference value to identify pollutant combinations with high versatility. The advantage of this process is that the dual measurement standards of mean and standard deviation not only ensure that the numerical level of the model output is close to the target reference range, but also ensure the stability and consistency of the model under different pollutant combinations. In particular, the pollutant combination with high output mean and standard deviation is regarded as a combination of "acceptable versatility". This screening method makes the model more suitable for various complex pollutant environments and improves the model's wide adaptability to DNI attenuation. In addition, the pollutant combination with the largest number of combinations in acceptable versatility is selected for DNI prediction to ensure that the combination selected by the model has the largest coverage, thereby further improving the robustness and reliability of the model prediction. This method effectively combines accuracy and generalization ability, providing more reliable data support and adaptability for DNI prediction.

[0071] S6, input the meteorological observation data under historical cloud conditions into the DNI attenuation ratio model based on pollutant concentration to obtain the DNI attenuation ratio value of the clear sky atmosphere and obtain the DNI attenuation ratio value of the cloud based on the attenuation ratio calculation method.

[0072] In this example, the meteorological observation data under historical cloud conditions are input into the DNI attenuation ratio model based on pollutant concentration to obtain the clear sky atmospheric DNI attenuation ratio value, specifically:

[0073]

[0074] in, is the DNI value under sunny conditions obtained based on the DNI attenuation proportional model of pollutant concentration, is the DNI received in the outer atmosphere, is the optical thickness caused by Rayleigh scattering, is the optical depth caused by aerosol, is the optical depth caused by ozone, is the DNI attenuation ratio of the clear sky atmosphere.

[0075] It should be noted that the advantage of obtaining the clear sky atmospheric DNI attenuation ratio value based on the pollutant concentration-based DNI attenuation ratio model combined with meteorological observation data under historical cloud conditions is that this method can more accurately reflect the impact of atmospheric conditions on DNI. In actual operation, the impact of cloud conditions on DNI is very complex and changeable. Through this method, the attenuation ratio can be dynamically adjusted and optimized by using the attenuation model that has been verified under clear sky conditions and combining it with meteorological data under different cloud conditions, making the model more adaptable and accurate. This fusion enables the forecast model to not only handle clear sky conditions, but also provide more accurate DNI forecasts in cloudy conditions, thereby improving the reliability of solar resource assessment and application.

[0076] In this example, the cloud-to-DNI attenuation ratio value is obtained based on the attenuation ratio calculation method, specifically:

[0077]

[0078] in, is the cloud-to-DNI attenuation ratio in the case of clouds, It is the historical DNI observation data.

[0079] It should be noted that the main advantage of using the clear sky atmospheric DNI attenuation ratio, historical DNI observation data and extraterrestrial radiation data to calculate the cloud-induced DNI attenuation ratio in cloudy conditions is that it can accurately distinguish the different effects of the atmosphere and clouds on DNI. Using the attenuation ratio value under clear sky conditions as a benchmark, combined with historical observation data and extraterrestrial radiation data, the additional attenuation effect of clouds can be accurately quantified. This method improves the prediction ability of the forecast model under variable weather conditions, allowing the model to more accurately reflect the impact of clouds on DNI, thereby enhancing the practicality and reliability of solar power generation forecasts and meteorological analysis.

[0080] S7, inputting DNI historical meteorological observation data into the preset DNI prediction sub-models of different pollutant types to obtain an initial DNI prediction value, and correcting the initial DNI prediction value based on the cloud-to-DNI attenuation ratio value to obtain a final DNI prediction value;

[0081] In this example, the preset DNI prediction sub-models for different pollutant types are as follows:

[0082] Construct DNI prediction sub-models for different pollutant types. The DNI prediction sub-model is a Bi-GRU prediction model that integrates the CA attention mechanism and a multi-dimensional convolutional neural network.

[0083] In this example, the DNI prediction sub-model is a Bi-GRU prediction model that integrates the CA attention mechanism and the multi-dimensional convolutional neural network. Specifically:

[0084] Construct a multi-dimensional convolutional neural network, which includes a three-dimensional convolutional layer, an average pooling layer and a two-dimensional convolutional layer;

[0085] A CA attention layer is added to the GRU layer of the preset Bi-GRU prediction model to obtain the Bi-GRU prediction model after adding the CA attention mechanism;

[0086] The output of the multidimensional convolutional neural network is used as the input of the Bi-GRU prediction model after adding the CA attention mechanism.

[0087] In this example, a CA attention layer is added to the GRU layer of the preset Bi-GRU prediction model to obtain the Bi-GRU prediction model after adding the CA attention mechanism, specifically:

[0088] The DNI output feature map of the GRU layer of the preset Bi-GRU prediction model is averaged in the channel dimension to obtain an average DNI output feature map;

[0089] The average DNI output feature map is subjected to convolution operation and activation function to obtain the DNI weight coefficient;

[0090] The weighted DNI output feature map is obtained by multiplying the DNI weight coefficient with the DNI output feature map.

[0091] In this example, a multidimensional convolutional neural network is constructed. The multidimensional convolutional neural network includes a three-dimensional convolutional layer, an average pooling layer, and a two-dimensional convolutional layer. Specifically:

[0092] The three-dimensional convolution layer and the two-dimensional convolution layer can process multi-dimensional DNI feature data and simultaneously capture local features in time series and spatial distribution to extract DNI features of different dimensions.

[0093] The average pooling layer downsamples the feature map output by the three-dimensional convolutional layer to reduce the size of the feature map while retaining important features to prevent the model from overfitting.

[0094] It should be noted that the Bi-GRU prediction model is an improved LSTM network. By introducing the update gate and reset gate mechanism, it can effectively capture the dependencies in the sequence data and avoid the gradient vanishing problem caused by back propagation in the traditional RNN. Its key features are as follows:

[0095] Bidirectional structure: Bi-GRU consists of two GRU layers, one that processes the sequence from front to back (forward GRU) and the other that processes the sequence from back to front (reverse GRU). This bidirectional processing method allows the model to capture both past and future information of time series data at the same time, thereby more comprehensively understanding the characteristics of the entire sequence.

[0096] Gating mechanism: The gating mechanism of GRU includes update gate and reset gate. The update gate controls the transmission of information between time steps and determines how much past information the current state should retain; the reset gate determines how much past information the current state should ignore. This mechanism enables GRU to efficiently handle long-term dependency problems.

[0097] In this example, the CA attention mechanism is added, specifically:

[0098] The CA attention mechanism (Coordinate Attention) is an attention mechanism that enhances the feature expression capability of neural networks. The CA mechanism combines spatial information and channel information to capture important information in the input feature map in a more fine-grained manner. It first performs global average pooling on the feature map in two directions to generate two attention maps in the horizontal and vertical directions. Then, after a series of transformations, these maps are used to weight the channel dimension of the original feature map. This approach not only retains the spatial information in the feature map, but also enhances the model's attention to important areas and key features.

[0099] It should be noted that in DNI prediction, adding the CA attention mechanism can significantly improve the prediction performance of the model. Since DNI is affected by many factors, including cloud distribution, aerosol concentration, and geographical location, traditional methods may find it difficult to accurately capture these complex spatial changes. The CA attention mechanism captures the spatial and channel information in the feature map more finely, enabling the model to more accurately identify and weight important feature areas related to DNI, thereby improving the accuracy and robustness of the prediction. In addition, the CA mechanism can effectively suppress noise and enhance the model's ability to adapt to environmental changes, so that it can still provide reliable DNI prediction results under changing meteorological conditions.

[0100] In this example, the initial DNI prediction value is corrected based on the cloud DNI attenuation ratio value to obtain the final DNI prediction value, which is:

[0101] Calculate the correction factor based on the cloud-to-DNI attenuation ratio;

[0102] Based on the correction coefficient, the initial DNI prediction value was calculated using an exponential function to obtain the final DNI prediction value.

[0103] In this example, the correction factor is calculated based on the cloud-to-DNI attenuation ratio, specifically:

[0104] The correction calculation formula is used to calculate the correction coefficient, where the correction formula is as follows:

[0105]

[0106] in, is the correction factor, is the cloud attenuation ratio of DNI.

[0107] It should be noted that the correction factor is a proportional factor used to adjust the baseline DNI value. By considering the actual attenuation effect of clouds on DNI, the final predicted DNI value is more in line with the actual weather conditions. Its advantage is that the correction factor can dynamically reflect the impact of cloud changes on solar radiation, such as increasing attenuation when the clouds are thicker, or reducing attenuation when the clouds are thinner. This adjustment method not only improves the accuracy of DNI prediction, but also makes the model more flexible and adaptable when facing dynamic changes in clouds, which is suitable for precise calculations in solar energy prediction and energy management.

[0108] In this example, based on the correction coefficient, an exponential function is used to calculate the initial DNI prediction value to obtain the final DNI prediction value, which is:

[0109]

[0110] in, is the final DNI prediction value, is the initial DNI prediction value, is the base of the natural logarithm, and k is the adjustment coefficient, which is used to control the strength of the correction and adjust the sensitivity of the model to cloud thickness or other attenuation factors.

[0111] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0112] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0113] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0114] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0115] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0116] Finally: The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. The DNI ultra-short-term forecasting method based on ground-based cloud images is characterized by: The following steps are involved: Obtain DNI historical meteorological observation data and pollutant data and preprocess them. DNI historical meteorological data includes historical meteorological observation data under clear sky conditions, historical meteorological observation data under cloud conditions, historical DNI observation data and extraterrestrial radiation data; Based on the preset clear sky atmospheric basic DNI attenuation proportional model, the DNI relative attenuation value is calculated, and the DNI relative attenuation value is used as the dependent variable, and the concentration of each pollutant is used as the independent variable to construct a DNI attenuation proportional model based on pollutant concentration; Obtain different pollutant data, and arrange and combine the pollutant data to obtain several pollutant combinations, calculate the combined evaluation value of the pollutant combination based on variance analysis, and perform data analysis to obtain the initial pollutant combination; Inputting the initial pollutant combination into a DNI attenuation ratio model based on pollutant concentration for training, randomly and gradually increasing pollutant combinations for training from a plurality of pollutant combinations, and constructing a pollutant training output set from the DNI attenuation ratio model of a plurality of pollutant concentrations; Conduct data analysis on the pollutant training output set and screen out acceptable and general DNI attenuation ratio models based on pollutant concentrations; Inputting meteorological observation data under historical cloud conditions into the acceptable general-purpose DNI attenuation ratio model based on pollutant concentration to obtain the clear sky atmosphere DNI attenuation ratio value, and obtaining the cloud DNI attenuation ratio value based on the attenuation ratio calculation method; The DNI historical meteorological observation data is input into the DNI prediction sub-model to obtain the initial DNI prediction value, and the initial DNI prediction value is corrected based on the cloud-to-DNI attenuation ratio value to obtain the final DNI prediction value.

2. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 1, characterized in that: The preset clear sky atmosphere basic DNI attenuation ratio model is specifically: Based on the SMARTS clear-sky radiation transfer model, select an atmospheric model and set the model's spectral range and the geographical location parameters of the observation point; The meteorological observation data under historical clear sky conditions and historical DNI observation data were input into the atmospheric model for training and verification, and the clear sky atmospheric basic DNI attenuation ratio model was obtained.

3. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 2 is characterized in that: The DNI relative attenuation value is calculated based on the preset clear sky atmosphere basic DNI attenuation ratio model, specifically: Input DNI historical meteorological observation data into the preset clear sky atmosphere basic DNI attenuation ratio model to obtain the benchmark DNI value; Set the attenuation factor for each pollutant, and based on the DNI calculation formula under pollutant conditions, obtain the DNI value under pollutant-containing conditions; The DNI relative attenuation value was calculated based on the baseline DNI value and the DNI value under pollutant conditions.

4. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 3 is characterized in that: The DNI relative attenuation value is used as the dependent variable, and the concentration of each pollutant is used as the independent variable to construct a DNI attenuation proportional model based on the pollutant concentration, specifically: Based on the relative attenuation value of DNI as the dependent variable and the concentration of each pollutant as the independent variable, the DNI linear regression equation was constructed to obtain the DNI linear regression model; The DNI historical meteorological observation data and pollutant data are divided into training sets and test sets according to the preset ratio; The training set is input into the DNI linear regression model for model training to obtain the trained DNI linear regression model; The trained DNI linear regression model was verified using the test set to obtain the trained DNI linear regression model.

5. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 4 is characterized in that: The different pollutant data are obtained, and the pollutant data are arranged and combined to obtain several pollutant combinations, the combined evaluation values ​​of the pollutant combinations are calculated based on variance analysis, and data analysis is performed to obtain the initial pollutant combination, which is specifically: Obtain different pollutant data, and arrange and combine the pollutant data to obtain several pollutant combinations; According to the number and repeatability of pollutant combinations, the ANOVA statistical method was selected to calculate the inter-group variance and intra-group variance of several pollutant combinations; The ratio of the inter-group variance to the intra-group variance is used as a combined evaluation value, and the combined evaluation values ​​are sorted in descending order to select the largest combined evaluation value; The pollutant combination corresponding to the maximum combined evaluation value is used as the initial pollutant combination.

6. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 5 is characterized in that: The pollutant training output set is subjected to data analysis to screen out an acceptable and general DNI attenuation ratio model based on pollutant concentration, specifically: Calculate the model output mean value and the model output standard deviation corresponding to the pollutant training output set, and compare the model output mean value and the model output standard deviation with a preset output mean value reference value and a preset output standard deviation reference value; The pollutant combination corresponding to the model output average value greater than the preset output average value reference value and the model output standard deviation greater than the preset output standard deviation reference value is regarded as a general-purpose pollutant combination; The DNI attenuation proportional model based on pollutant concentration with the largest number of ubiquitous pollutant combinations among all ubiquitous pollutant combinations is used to predict DNI.

7. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 6, characterized in that: The cloud-based DNI attenuation ratio value is used to correct the initial DNI prediction value to obtain the final DNI prediction value, specifically: Calculate the correction factor based on the cloud-to-DNI attenuation ratio; Based on the correction coefficient, the initial DNI prediction value was calculated using an exponential function to obtain the final DNI prediction value.

8. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 7, characterized in that: The specific formula of the DNI value under the pollutant-containing condition is: in, is the DNI value under pollutant conditions, is the baseline DNI value, is the attenuation factor of the pollutant, is the concentration of the pollutant.

9. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 8, characterized in that: The calculation of the DNI relative attenuation value is specifically as follows: in, is the relative attenuation value of DNI, is the DNI value under pollutant conditions, is the benchmark DNI value.

10. The DNI ultra-short-term forecasting method based on ground-based cloud images according to claim 9, characterized in that: The inter-group variance and intra-group variance of the response variables corresponding to the several pollutant combinations are calculated respectively, and the ratio of the inter-group variance to the intra-group variance is used as the combination evaluation value, and the pollutant combination corresponding to the largest combination evaluation value is used as the initial pollutant combination. The specific formula is expressed as follows: in, is the between-group variance, is the number of pollutant combinations, is the number of samples of the ith pollutant combination, is the overall average DNI attenuation value of all pollutant combinations, is the average DNI attenuation value of the i-th pollutant combination, is the within-group variance, is the DNI attenuation value of the jth sample in the ith pollutant combination, and R is the combined evaluation value.

Images