DNI prediction method based on all-sky imaging and clear sky DNI fitting

Abstract

The invention relates to the field of weather forecast, and discloses a DNI prediction method based on all-sky imaging and clear sky DNI fitting, which comprises the following steps: screening clear sky DNI data from DNI historical data, fitting a clear sky DNI model, and then carrying out clear sky DNI calculation by using the clear sky DNI model obtained by fitting to obtain clear sky DNI at each time point; inputting an all-sky image obtained in real time and an all-sky image obtained in the previous one minute into the trained double-picture convolutional neural network model, and predicting to obtain a clear sky index after # imgabs0 # minute; and multiplying the predicted clear sky index of the # imgabs 1 # after the minute by the corresponding clear sky DNI to obtain a predicted DNI value of the # imgabs 2 # after the minute. The method disclosed by the invention can reflect the change of actual DNI more accurately, has better prediction accuracy and strong adaptability, can be used for tower type solar thermal power generation systems in different regions, and is beneficial to reducing the operation cost of a photo-thermal power station.

Description

Technical Field

[0001] The present invention relates to the field of weather forecasting, and particularly to a DNI prediction method based on all-sky imaging and clear-sky DNI fitting. Background Art

[0002] The tower solar thermal power generation system uses heliostats to track the sun and reflect sunlight onto the absorber surface screen on the absorber tower, heating the heat medium in the absorber, and finally realizing power generation. Among the reflected sunlight, the most important component is direct normal irradiance (DNI). The sudden change of DNI will affect the reliability and power generation efficiency of the solar thermal power plant. Therefore, predicting the change of DNI in the mirror field area is crucial for protecting the absorber and improving the power generation stability.

[0003] For the method based on all-sky imager images, there are those using convolutional neural networks and those predicting DNI by predicting the movement of cloud clusters. However, since the information reflected by all-sky imaging is whether there are clouds, how many clouds, and the thickness of the clouds, the model directly predicting DNI based on all-sky imaging can only predict an average DNI. For example, for a clear day with the same solar position, due to different meteorological parameters such as aerosol concentration, sometimes the DNI can be as high as 1000 W / m 2 , and sometimes it can only reach 800 W / m 2 , then the prediction model can only give 900 W / m 2 . This brings a large prediction error and cannot be eliminated through training. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a DNI prediction method based on all-sky imaging and clear-sky DNI fitting to achieve the purpose of improving the DNI prediction accuracy.

[0005] To achieve the above purpose, the technical solution of the present invention is as follows: A DNI prediction method based on all-sky imaging and clear-sky DNI fitting, comprising the following steps: Step 1, calculate the clear-sky DNI: Screen out the clear-sky DNI data from the DNI historical data, fit the clear-sky DNI model, and then use the fitted clear-sky DNI model to calculate the clear-sky DNI to obtain the clear-sky DNI at each time point; Step 2, predict the clear-sky index: Input the real-time obtained all-sky imaging and the all-sky imaging one minute before it into the trained dual-picture convolutional neural network model to predict the minute later clear-sky index; Step 3, predict the DNI: The predicted Multiply the clearness index after a certain number of minutes by the corresponding clear-sky DNI to obtain the prediction of the DNI value after a certain number of minutes.

[0006] In the above solution, in step one, the clear-sky DNI model is as follows: ; where, represents the clear-sky DNI, represents the altitude, represents the extraterrestrial radiation intensity, represents the optical air mass, represents the turbidity, which is a parameter to be fitted; the extraterrestrial solar radiation intensity<--? is expressed as: ; where, represents the reference extraterrestrial solar radiation intensity, represents the day number in the year, that is, which day of the year this day is; the optical air mass is expressed as: ; where, represents the solar zenith angle.

[0007] In the above solution, in step one, the specific method for fitting the clear-sky DNI model is as follows: (1) Divide the DNI data of each day into data segments of a certain time length; (2) Initialize the turbidity of the parameters to be fitted in the clear-sky DNI model; Before the start of the entire fitting, the initial value of turbidity is taken as 8, and the initial value of turbidity for fitting in a time period on the first day is taken as the turbidity fitting value of the previous time period; starting from the second day, the initial value of turbidity for a time period is taken as the turbidity fitting value of the same time period on the previous day; (3) Select the first clear-sky DNI value and other clear-sky DNI values within a time period according to the solar zenith angle to obtain a list of clear-sky DNI and solar zenith angle; (4) Use the obtained list to fit the turbidity of the parameters to be fitted in the clear-sky DNI model within a time period by the least squares method; (5) Repeat steps (2)-(4) until the fitting of the clear-sky DNI model for each day and each time period is completed.

[0008] In a further technical solution, in step (3), the method for selecting the first clear-sky DNI value within a time period is as follows: First, subtract 1 minute from the time corresponding to a DNI data point and determine according to the time 1 minute ago and the solar zenith angle one minute ago , combined with the initial turbidity value, calculate the conjectured clear sky DNI value one minute ago ; Then, according to the DNI data points to determine the solar zenith angle at the current moment , if , and it satisfies when, confirm as the first clear sky DNI value in this time period, and record the corresponding ; among them, is a threshold value between 0 and 1; If , and it satisfies when, confirm as the first clear sky DNI value in this time period, and record the corresponding , among them, is a threshold value between 0 and 1; If this DNI data point does not meet the above requirements, repeat the above steps for the next data point in this time period.

[0009] In a further technical solution, in step (3), the method for selecting other clear sky DNI values within a time period is as follows: Assume that the last selected clear sky DNI and solar zenith angle in the current time period are and respectively, and the DNI and solar zenith angle of the current data point are and respectively; if , and it satisfies when, add to the list of clear sky DNI values, and record the corresponding ; If , and it satisfies when, add to the list of clear sky DNI values, and record the corresponding ; among them, represents the time of the last point in the selected list; If this DNI data point does not meet the above requirements, do not add to the list of clear sky DNI values.

[0010] In a further technical solution, in step (3), the method for obtaining a list of clear sky DNI and solar zenith angle is as follows: ; Among them, represent the first, second, …, last time points corresponding to the data in the lists of the selected clear-sky DNI and solar zenith angle respectively, and ; represents the clear-sky DNI at the time point, represents the solar zenith angle at the time point, and so on for others.

[0011] In the above solution, the dual-image convolutional neural network model includes sub-network one, sub-network two, a fully connected layer, and an output layer. The structural parameters of sub-network one and sub-network two are the same, and both include an input layer, four alternately arranged convolutional layers and pooling layers. The input layers of sub-network one and sub-network two respectively input the all-sky imaging at the current moment and the all-sky imaging one minute ago. The output tensors of the last pooling layers of sub-network one and sub-network two are unfolded and concatenated through the fully connected layer, and finally the clear-sky index at the moment is output through the output layer; the activation function used in each convolutional layer is the RELU function.

[0012] In the above solution, the method for training the dual-image convolutional neural network model is as follows: (1) Collect and process data: Collect all-sky imaging through an all-sky imager, and then perform data cleaning to remove pictures that are saved incorrectly and cannot be opened normally; in addition, according to the sunrise and sunset times of the project location, only keep the picture data collected during the day; crop the pictures, delete the external invalid black parts, and keep the middle field of view; According to the time corresponding to each DNI data point collected from the meteorological station at the project location, use the fitted clear-sky DNI model to calculate the clear-sky DNI point by point. If the clear-sky DNI is less than 1, then delete the DNI data point, and then divide the known DNI by the clear-sky DNI to obtain the clear-sky index; (2) Train the model: Divide all the data into a training set and a validation set according to a certain ratio, and use the training set to train the dual-image convolutional neural network model. Before inputting into the model, reduce the picture size to 400 ×400, and only keep the red and blue channels; among them, use the processed all-sky imaging at the current moment t and the all-sky imaging one minute ago as the model input, and the clear-sky index at the moment as the model output; (3) Validate the model: Use the validation set to validate the dual-image convolutional neural network model. When the accuracy on the validation set does not improve significantly in consecutive multiple iterations, stop training.

[0013] In the above solution, in step three, the DNI calculation formula for prediction is as follows: ; Wherein, represents the clear sky DNI, represents the clear sky index.

[0014] Through the above technical solution, a DNI prediction method based on all-sky imaging and clear sky DNI fitting provided by the present invention has the following beneficial effects: 1. Improve prediction accuracy: The present invention predicts the clear sky index first and then combines the clear sky DNI model to predict DNI, which can more accurately reflect the change of actual DNI, especially under different meteorological conditions (such as the change of aerosol concentration). Since the historical DNI information is not input into the clear sky index prediction model, the model has better prediction accuracy for the sudden change of the clear sky index; 2. Strong adaptability: The method of the present invention does not depend on specific meteorological conditions or geographical locations, has strong versatility and adaptability, and can be used in tower solar thermal power generation systems in different regions; 3. Strong real-time performance: The present invention can process all-sky imaging data in real time and quickly predict the DNI at future moments, which helps the solar thermal power station to adjust the operation strategy in time and improve the power generation stability and efficiency; 4. High cost-effectiveness: Compared with the traditional DNI prediction method, the method of the present invention can use the common meteorological stations in the solar thermal power station, which is more economical in terms of hardware cost and data collection. At the same time, it can provide more accurate prediction results, which helps to reduce the operation cost of the solar thermal power station. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art.

[0016] Figure 1 is a schematic flow chart of a DNI prediction method based on all-sky imaging and clear sky DNI fitting disclosed in an embodiment of the present invention; Figure 2 is a structural diagram of a double-image convolutional neural network model; Figure 3 is an example diagram of all-sky imaging. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention.

[0018] The present invention provides a DNI prediction method based on all-sky imaging and clear-sky DNI fitting, as Figure 1 shown, which includes the following steps: Step 1, calculate the clear-sky DNI: Screen out the clear-sky DNI data from the DNI historical data, fit the clear-sky DNI model, and then use the fitted clear-sky DNI model to calculate the clear-sky DNI to obtain the clear-sky DNI at each time point.

[0019] The clear-sky DNI model is as follows: ; where, represents the clear-sky DNI, represents the altitude, represents the extraterrestrial radiation intensity, represents the optical air mass, represents the turbidity, which is a parameter to be fitted; the extraterrestrial solar radiation intensity is expressed as: ; where, represents the reference extraterrestrial solar radiation intensity, and the reference extraterrestrial solar radiation intensity generally takes 1361.2 W / m 2 ; represents the day number in the year, that is, which day of the year this day is; the optical air mass is expressed as: ; where, represents the solar zenith angle.

[0020] The specific method for fitting the clear-sky DNI model is as follows: (1) Divide the DNI data of each day into data segments of a certain time length; (2) Initialize the turbidity of the parameters to be fitted in the clear-sky DNI model; Before the start of the entire fitting, the initial value of the turbidity is taken as 8, and the initial value of the turbidity for fitting in a time period on the first day is taken as the turbidity fitting value of the previous time period; starting from the second day, the initial value of the turbidity for a time period is taken as the turbidity fitting value of the same time period on the previous day; (3) Select the first clear-sky DNI value and other clear-sky DNI values within a time period according to the solar zenith angle to obtain a list of clear-sky DNI and solar zenith angle; The method for selecting the first clear-sky DNI value within a time period is as follows: First, subtract 1 minute from the time corresponding to a DNI data point and determine according to the time 1 minute ago and the solar zenith angle 1 minute ago , combine with the initial turbidity value to calculate the conjectured clear sky DNI value 1 minute ago ; Then, determine the solar zenith angle at the current moment according to the time corresponding to the DNI data point , if , and satisfy When, will be confirmed as the first clear sky DNI value in this time period, and record the corresponding ; Among them, is a threshold value between 0 and 1, for example, it can be taken as 0.94. is a threshold value between 0 and 1, for example, it can be taken as 0.98.

[0021] If , and satisfy When, will be confirmed as the first clear sky DNI value in this time period, and record the corresponding , among which, is a threshold value between 0 and 1, for example, it can be taken as 0.98.[[ID=We]]

[0022] If the DNI data point does not meet the above requirements, repeat the above steps for the next data point in this time period.

[0023] The method to select other clear sky DNI values in a time period is as follows: Suppose the last selected clear sky DNI and solar zenith angle in the current time period are respectively and , the DNI and solar zenith angle of the current data point are respectively and ; If , and satisfy When, will be added to the list of clear sky DNI values, and record the corresponding ; If , and satisfy When, will be added to the list of clear sky DNI values, and record the corresponding ; Among them, represents the time of the last point in the selected list; If the DNI data point does not meet the above requirements, then will not be added to the list of clear sky DNI values.

[0024] Repeat the above steps for the next data point in this time period until the DNI data in this time period is processed. Finally, a list of clear sky DNI and solar zenith angle is obtained as follows: ; wherein, respectively represent the first, second, …, last time points corresponding to the data in the lists of the selected clear-sky DNI and solar zenith angle, and ; represents the clear-sky DNI at the time point, represents the solar zenith angle at the time point, and so on for others.

[0025] (4) Using the obtained lists, fit a clear-sky DNI model for a period of time by the least squares method with the parameter turbidity to be fitted; If the number of clear-sky DNI data points selected for a period of time is less than or equal to the threshold , such as 5, then the turbidity value for this period is taken as the initial value.

[0026] If the number of clear-sky DNI data points selected for a period of time is greater than the threshold , then substitute the doy, solar zenith angle, and clear-sky DNI in the list into the clear-sky DNI model to fit and solve the turbidity parameter for this period.

[0027] (5) Repeat steps (2)-(4) until the fitting of the clear-sky DNI model for each day and each period is completed.

[0028] Step Two, predicting the clear-sky index: Input the real-time obtained all-sky imaging and the all-sky imaging 1 minute before it into the trained dual-image convolutional neural network model, and predict to obtain the clear-sky index after

[0029] The present invention designs a dual-image convolutional neural network model ( Figure 2 ) to predict the clear-sky index. The dual-image convolutional neural network model includes Sub-network One, Sub-network Two, a fully connected layer, and an output layer. The structural parameters of Sub-network One and Sub-network Two are the same, and both include an input layer, four alternately arranged convolutional layers and pooling layers. The input layers of Sub-network One and Sub-network Two respectively input the all-sky imaging at the current moment and the all-sky imaging 1 minute before. The output tensors of the last pooling layers of Sub-network One and Sub-network Two are unfolded and concatenated through the fully connected layer, and finally the clear-sky index at the moment is output through the output layer. The activation function used for each convolutional layer is the RELU function.

[0030] The method for training the dual-image convolutional neural network model is as follows: (1) Collecting and processing data: Collect all-sky imaging through an all-sky imager, and then perform data cleaning to remove pictures that cannot be opened normally due to storage errors. In addition, according to the sunrise and sunset times at the project location, only retain the picture data collected during the day; the middle field of view of the all-sky imaging picture ( Figure 3 ) is a circular area. Crop the picture and delete the external invalid black part.

[0031] According to the time corresponding to each DNI data point collected from the meteorological station at the project location, use the fitted clear-sky DNI model to calculate the clear-sky DNI point by point. If the clear-sky DNI is less than 1, delete the DNI data point, and then divide the known DNI by the clear-sky DNI to obtain the clear-sky index.

[0032] (2)Train the model: Divide all the data into a training set and a validation set according to a certain ratio, and use the training set to train the dual-picture convolutional neural network model. Before inputting into the model, reduce the picture size to 400 ×400, and only retain the red and blue channels; among them, use the processed all-sky imaging at the current moment t and the all-sky imaging of the previous 1 minute as the model input, and the clear-sky index at the moment as the model output; It can be 1min, 2min,..., 5min,.... Due to the limitation of the field of view of the all-sky imager, it is generally not more than 30min.

[0033] (3)Validate the model: Use the validation set to validate the dual-picture convolutional neural network model. When the accuracy on the validation set does not improve significantly in multiple consecutive iterations (such as 5 or 10), stop training.

[0034] Step three, predict DNI: Multiply the predicted clear-sky index after minutes by the corresponding clear-sky DNI to obtain the predicted DNI value after minutes.

[0035] The formula for predicting DNI is as follows: ; Among them, represents the clear-sky DNI, represents the clear-sky index.

[0036] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A DNI prediction method based on all-sky imaging and clear-sky DNI fitting, characterized in that, It includes the following steps: Step 1, calculate the clear sky DNI: Select the clear sky DNI data from the historical DNI data, fit the clear sky DNI model, and then use the fitted clear sky DNI model to calculate the clear sky DNI to obtain the clear sky DNI at each time point; Step 2, predict the clear sky index: Input the real-time obtained all-sky imaging and the all-sky imaging of the previous 1 minute into the trained dual-image convolutional neural network model, and predict to obtain the clearness index after [0000009] minutes; Step 3, predict the DNI: Multiply the predicted clearness index after minutes by the corresponding clear-sky DNI to obtain the predicted DNI value after minutes.

2. The DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 1, characterized in that, In Step 1, the clear sky DNI model is as follows: ； Among them, represents clear-sky DNI, represents altitude, represents extraterrestrial radiation intensity, represents optical air mass, represents turbidity, which is a parameter to be fitted; extraterrestrial solar radiation intensity is expressed as: ； Among them, represents the reference extraterrestrial solar radiation intensity, represents the day number in a year, that is, which day of the year this day is; the optical air quality is expressed as: ； Among them, represents the solar zenith angle.

3. A DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 2, characterized in that, In Step 1, the specific method for fitting the clear sky DNI model is as follows: (1) Divide the DNI data of each day into data segments of a certain time length; (2) Initialize the turbidity of the parameters to be fitted in the clear sky DNI model; Before the start of the entire fitting, the initial value of turbidity is taken as 8, and the initial value of turbidity for fitting in a time period on the first day is taken as the fitted value of turbidity in the previous time period; starting from the second day, the initial value of turbidity in a time period is taken as the fitted value of turbidity in the same time period of the previous day; (3) Select the first clear sky DNI value and other clear sky DNI values within a time period according to the solar zenith angle to obtain a list of clear sky DNI and solar zenith angle; (4) Use the obtained list to fit the turbidity of the parameters to be fitted in the clear sky DNI model within a time period by the least squares method; (5) Repeat steps (2)-(4) until the fitting of the clear sky DNI model for each day and each time period is completed.

4. The DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 3, wherein, In step (3), the method for selecting the first clear sky DNI value within a time period is as follows: First, subtract 1 minute from the time corresponding to a DNI data point to determine, based on the time 1 minute ago, and the solar zenith angle 1 minute ago . Combine with the initial turbidity value to calculate the conjectured clear sky DNI value 1 minute ago ; Then, based on the DNI data points determine the solar zenith angle at the current moment according to the corresponding time , if , and when it satisfies , set as the first clear-sky DNI value for this time period, and record the corresponding at the same time; where is a threshold value ranging from 0 to 1; If and satisfy at that time, is confirmed as the first clear sky DNI value for this time period, and the corresponding is recorded at the same time, where is a threshold value ranging from 0 to 1; If the DNI data point does not meet the above requirements, repeat the above steps for the next data point in this time period.

5. The DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 4, characterized in that: In step (3), the method for selecting other clear sky DNI values within a time period is as follows: Let the last clear sky DNI and solar zenith angle selected in the current time period be and , respectively, and the DNI and solar zenith angle of the current data point be and ; respectively. If , and it satisfies when, add to the clear sky DNI value list, and record the corresponding at the same time; If and meet the condition of at this time, add to the clear sky DNI value list, and record the corresponding at the same time; among them, represents the time of the last point in the selected list; If the DNI data point does not meet the above requirements, then is not added to the clear sky DNI value list.

6. A DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 4, characterized in that, In step (3), the list of clear sky DNI and solar zenith angle obtained is as follows: ； Among them, respectively represent the first, second,..., last time points corresponding to the data in the selected clear sky DNI and solar zenith angle lists, and ; represents the clear sky DNI at the time point, represents the solar zenith angle at the time point, and so on for the others.

7. A DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 1, characterized in that, The dual-image convolutional neural network model includes Sub-network One, Sub-network Two, a fully connected layer, and an output layer. The structural parameters of Sub-network One and Sub-network Two are the same, and both include an input layer, four alternately arranged convolutional layers and pooling layers. The input layers of Sub-network One and Sub-network Two respectively input the full-sky imaging at the current moment and the full-sky imaging one minute ago. The output tensors of the last pooling layers of Sub-network One and Sub-network Two are unfolded and concatenated through the fully connected layer, and finally output through the output layer the clear sky index at the moment; the activation function used in each convolutional layer is the RELU function.

8. A DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 1, characterized in that, The method for training the dual-image convolutional neural network model is as follows: (1) Collect and process data: Collect the all-sky imaging through the all-sky imager, and then perform data cleaning to remove the pictures that are saved incorrectly and cannot be opened normally; in addition, according to the sunrise and sunset times at the project location, only keep the picture data collected during the day; Cut the pictures, delete the external invalid black parts, and keep the middle field of view; According to the time corresponding to each DNI data point collected from the meteorological station at the project location, calculate the clear sky DNI point by point using the fitted clear sky DNI model. If the clear sky DNI is less than 1, delete the DNI data point, and then divide the known DNI by the clear sky DNI to obtain the clear sky index; (2) Train the model: Divide all the data into a training set and a validation set according to a certain ratio. Use the training set to train the dual-image convolutional neural network model. Before inputting into the model, reduce the image size to 400 400, and only retain the red and blue channels; among them, use the processed all-sky imaging at the current moment t and the all-sky imaging in the previous 1 minute as the model input, and use the clearness index at the moment as the model output; (3) Validate the model: Use the validation set to validate the dual-image convolutional neural network model. When the accuracy on the validation set does not increase significantly in consecutive multiple iterations, stop training.

9. A DNI prediction method based on all-sky imaging and clear-sky DNI fitting according to claim 1, characterized in that, In Step 3, the DNI prediction formula is as follows: ； Among them, represents clear sky DNI, represents the clear sky index.

Images