Fine-grained photovoltaic power output prediction method and terminal

Abstract

The application discloses a fine-grained photovoltaic output prediction method and a terminal, collects in-station monitoring data of a meteorological station and a photovoltaic power station, and constructs a coarse-grained database containing meteorological monitoring data and operating conditions of the photovoltaic power station; fine-grained feature construction based on time window moving average is constructed to construct a fine-grained database; then, coarse-grained and fine-grained prediction models are respectively established, a first photovoltaic output value output by the coarse-grained prediction model and a photovoltaic output error output by the fine-grained prediction model are used to obtain a predicted photovoltaic output value. Therefore, by constructing fine-grained features of photovoltaic output influencing quantities, information density of input data of the prediction model can be improved, so that the prediction accuracy of the photovoltaic output is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical energy metering, and in particular to a photovoltaic output prediction method and terminal based on fine-grainedness. Background Art

[0002] PV output forecasting typically uses meteorological data or PV output numerical data as features. However, such features are coarse-grained and cannot reflect the trend changes in PV output under complex weather patterns. Prediction models based on such coarse-grained features inherently lack information density, leaving significant room for improvement in accuracy and other aspects of model performance. With the widespread integration of clean energy sources, accurate PV output forecasting methods are urgently needed to ensure stable grid operation when PV output is integrated.

[0003] Therefore, there is an urgent need for a prediction model that can construct fine-grained features of photovoltaic output, thereby improving the accuracy of photovoltaic output prediction methods. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a photovoltaic output prediction method and terminal based on fine-grainedness, which can solve the problem of insufficient photovoltaic output prediction accuracy.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A photovoltaic output prediction method based on fine-grainedness includes the following steps:

[0007] Acquire monitoring data from a weather station and a photovoltaic power station, and construct a coarse-grained database based on the monitoring data;

[0008] Performing time window-based sliding processing on the coarse-grained database to obtain time window trend features and construct a fine-grained database;

[0009] Dividing the data in the coarse-grained database into a training set, a validation set, and a test set, and constructing a coarse-grained prediction model to calculate a first photovoltaic output value;

[0010] Dividing the data in the fine-grained database into a training set, a validation set, and a test set, and constructing a fine-grained prediction model to calculate the photovoltaic output error;

[0011] The predicted photovoltaic output value is obtained by combining the first photovoltaic output value and the photovoltaic output error.

[0012] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0013] A photovoltaic output prediction terminal based on fine-grainedness includes:

[0014] A data acquisition module is used to obtain monitoring data from the weather station and the photovoltaic power station, and to construct a coarse-grained database based on the monitoring data;

[0015] A fine-grained construction module is used to perform time-window-based sliding processing on the coarse-grained database to obtain time-window trend features and construct a fine-grained database;

[0016] a coarse-grained model construction module, configured to divide the data in the coarse-grained database into a training set, a validation set, and a test set, and to construct a coarse-grained prediction model to calculate a first photovoltaic output value;

[0017] A fine-grained result correction module is used to divide the data in the fine-grained database into a training set, a validation set, and a test set, and to construct a fine-grained prediction model to calculate the photovoltaic output error;

[0018] A prediction module is configured to obtain a predicted photovoltaic output value by combining the first photovoltaic output value and the photovoltaic output error.

[0019] The beneficial effects of the present invention are as follows: Monitoring data from meteorological stations and photovoltaic power plants is collected to construct a coarse-grained database containing both meteorological monitoring data and photovoltaic power plant operating conditions; a fine-grained database is constructed based on fine-grained features derived from a time window sliding average; and coarse- and fine-grained prediction models are then established, respectively, to obtain a predicted photovoltaic output value based on the first photovoltaic output value output by the coarse-grained prediction model and the photovoltaic output error output by the fine-grained prediction model. Therefore, by constructing fine-grained features for the photovoltaic output influencing variables, the information density of the prediction model's input data can be increased, thereby improving the accuracy of photovoltaic output predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flowchart of a photovoltaic output prediction method based on fine granularity according to an embodiment of the present invention;

[0021] Figure 2 Schematic diagram of a photovoltaic output prediction terminal based on fine granularity according to an embodiment of the present invention;

[0022] Description of labels:

[0023] 1. A photovoltaic output prediction terminal based on fine-grainedness; 2. Memory; 3. Processor. DETAILED DESCRIPTION

[0024] To illustrate the technical content, achieved objectives and effects of the present invention in detail, the following description is given in conjunction with the embodiments and accompanying drawings.

[0025] Please refer to Figure 1 The embodiment of the present invention provides a photovoltaic output prediction method based on fine granularity, comprising the steps of:

[0026] Acquire monitoring data from a weather station and a photovoltaic power station, and construct a coarse-grained database based on the monitoring data;

[0027] Performing time window-based sliding processing on the coarse-grained database to obtain time window trend features and construct a fine-grained database;

[0028] Dividing the data in the coarse-grained database into a training set, a validation set, and a test set, and constructing a coarse-grained prediction model to calculate a first photovoltaic output value;

[0029] Dividing the data in the fine-grained database into a training set, a validation set, and a test set, and constructing a fine-grained prediction model to calculate the photovoltaic output error;

[0030] The predicted photovoltaic output value is obtained by combining the first photovoltaic output value and the photovoltaic output error.

[0031] As can be seen from the above description, the beneficial effects of the present invention are as follows: collecting on-site monitoring data from meteorological stations and photovoltaic power plants to construct a coarse-grained database containing meteorological monitoring data and photovoltaic power plant operating conditions; constructing a fine-grained database based on fine-grained features using a time window sliding average; then establishing coarse-grained and fine-grained prediction models, respectively, to obtain a predicted photovoltaic output value based on the first photovoltaic output value output by the coarse-grained prediction model and the photovoltaic output error output by the fine-grained prediction model. Therefore, by constructing fine-grained features of the photovoltaic output influencing variable, the information density of the prediction model's input data can be increased, thereby improving the accuracy of photovoltaic output prediction.

[0032] Furthermore, performing sliding processing based on a time window on the coarse-grained database to obtain time window trend features and construct a fine-grained database includes:

[0033] The coarse-grained database is X=[X1, X2, ..., X N ] T , the sample in the coarse-grained database is X i =[x i1 , x i2 ,…,x im ], where x ik Represents sample X i The value of the kth parameter;

[0034] The time window size w is preset, and the fine-grained time window trend characteristic value of the sample parameters in the coarse-grained database is calculated:

[0035] x ik.Trend =(X ik +X ik+1 +...Xi k+w ) / w;

[0036] The fine-grained database X is obtained based on the fine-grained time window trend feature value. Trend :

[0037] X Trend =[X 1.Trend ,X 2.Trend ,…,X N.Trend ] T , X i.Trend =[x i1.Trend , x i2.Trend ,…,x im.Trend ].

[0038] From the above description, it can be seen that based on the coarse-grained database and the time window size, the fine-grained time window trend feature is obtained, which facilitates the subsequent use of the fine-grained time window trend feature for error calculation.

[0039] Furthermore, the construction of the coarse-grained prediction model includes:

[0040] Using the temperature and humidity, solar energy data, and wind power data in the coarse-grained database as first input data, and using the instantaneous photovoltaic output power of the photovoltaic power station as a first output parameter;

[0041] The first input data and the first output parameter are fitted using a gradient boosting tree to construct a coarse-grained prediction model.

[0042] From the above description, it can be seen that based on the monitoring data of the weather station as the first input data and the instantaneous photovoltaic output power as the first output parameter, a coarse-grained prediction model can be fitted to obtain the predicted photovoltaic output power under different input conditions.

[0043] Furthermore, the building of a fine-grained prediction model includes:

[0044] Using the trend characteristic values ​​within the fine-grained time window of the temperature and humidity, solar energy data, and wind power data in the fine-grained database as second input data, and using the photovoltaic output error of the coarse-grained prediction model as a second output parameter;

[0045] The second input data and the second output parameter are fitted using a gradient boosting tree to construct a fine-grained prediction model.

[0046] From the above description, it can be seen that based on the monitoring data of the weather station as the second input data and the output error of the coarse-grained prediction model as the second output parameter, a fine-grained prediction model can be fitted to obtain the error of the predicted photovoltaic output power under different input trend conditions.

[0047] Furthermore, using a gradient boosting tree to fit the input data and output parameters, constructing a prediction model includes:

[0048] Combine the input data and output parameters in the training set to obtain a training sample set;

[0049] Using the root mean square error as a loss function, calculating the regression tree model parameters that minimize the loss function, and establishing an initial prediction model based on the loss function, the regression tree model parameters and the training sample set;

[0050] Calculating the negative gradient of the initial prediction model and its loss function to form a residual set of training samples and fitting a corresponding tree model, and iteratively updating the initial prediction model based on the tree model;

[0051] After the iteration end condition is met, the gradient boosting tree model is output to obtain the prediction model.

[0052] From the above description, it can be seen that by fitting the correlation between input and output data using the gradient boosting tree method and constructing a prediction model, the prediction accuracy of photovoltaic output can be improved.

[0053] Please refer to Figure 2 Another embodiment of the present invention provides a photovoltaic output prediction terminal based on fine granularity, including:

[0054] A data acquisition module is used to obtain monitoring data from the weather station and the photovoltaic power station, and to construct a coarse-grained database based on the monitoring data;

[0055] A fine-grained construction module is used to perform time-window-based sliding processing on the coarse-grained database to obtain time-window trend features and construct a fine-grained database;

[0056] a coarse-grained model construction module, configured to divide the data in the coarse-grained database into a training set, a validation set, and a test set, and to construct a coarse-grained prediction model to calculate a first photovoltaic output value;

[0057] A fine-grained result correction module is used to divide the data in the fine-grained database into a training set, a validation set, and a test set, and to construct a fine-grained prediction model to calculate the photovoltaic output error;

[0058] A prediction module is configured to obtain a predicted photovoltaic output value by combining the first photovoltaic output value and the photovoltaic output error.

[0059] As can be seen from the above description, monitoring data from meteorological stations and photovoltaic power plants is collected to construct a coarse-grained database containing both meteorological monitoring data and photovoltaic power plant operating conditions. A fine-grained database is constructed based on fine-grained features constructed using a time window sliding average. Subsequently, coarse-grained and fine-grained prediction models are established, respectively. The predicted photovoltaic output value is derived from the first photovoltaic output value output by the coarse-grained prediction model and the photovoltaic output error output by the fine-grained prediction model. Therefore, by constructing fine-grained features for the influencing variables of photovoltaic output, the information density of the prediction model's input data can be increased, thereby improving the accuracy of photovoltaic output predictions.

[0060] Furthermore, performing sliding processing based on a time window on the coarse-grained database to obtain time window trend features and construct a fine-grained database includes:

[0061] The coarse-grained database is X=[X1, X2, ..., X N ] T , the sample in the coarse-grained database is X i =[x i1 , x i2 ,…,x im ], where x ik Represents sample X i The value of the kth parameter;

[0062] The time window size w is preset, and the fine-grained time window trend characteristic value of the sample parameters in the coarse-grained database is calculated:

[0063] x ik.Trend =(X ik +X ik+1 +...Xi k+w ) / w;

[0064] The fine-grained database X is obtained based on the fine-grained time window trend feature value. Trend :

[0065] X Trend =[X 1.Trend ,X 2.Trend ,…,X N.Trend ] T , X i.Trend =[x i1.Trend , x i2.Trend ,…,x im.Trend ].

[0066] From the above description, it can be seen that based on the coarse-grained database and the time window size, the fine-grained time window trend feature is obtained, which facilitates the subsequent use of the fine-grained time window trend feature for error calculation.

[0067] Furthermore, the construction of the coarse-grained prediction model includes:

[0068] Using the temperature and humidity, solar energy data, and wind power data in the coarse-grained database as first input data, and using the instantaneous photovoltaic output power of the photovoltaic power station as a first output parameter;

[0069] The first input data and the first output parameter are fitted using a gradient boosting tree to construct a coarse-grained prediction model.

[0070] From the above description, it can be seen that based on the monitoring data of the weather station as the first input data and the instantaneous photovoltaic output power as the first output parameter, a coarse-grained prediction model can be fitted to obtain the predicted photovoltaic output power under different input conditions.

[0071] Furthermore, the building of a fine-grained prediction model includes:

[0072] Using the trend characteristic values ​​within the fine-grained time window of the temperature and humidity, solar energy data, and wind power data in the fine-grained database as second input data, and using the photovoltaic output error of the coarse-grained prediction model as a second output parameter;

[0073] The second input data and the second output parameter are fitted using a gradient boosting tree to construct a fine-grained prediction model.

[0074] From the above description, it can be seen that based on the monitoring data of the weather station as the second input data and the output error of the coarse-grained prediction model as the second output parameter, a fine-grained prediction model can be fitted to obtain the error of the predicted photovoltaic output power under different input trend conditions.

[0075] Furthermore, using a gradient boosting tree to fit the input data and output parameters, constructing a prediction model includes:

[0076] Combine the input data and output parameters in the training set to obtain a training sample set;

[0077] Using the root mean square error as a loss function, calculating the regression tree model parameters that minimize the loss function, and establishing an initial prediction model based on the loss function, the regression tree model parameters and the training sample set;

[0078] Calculating the negative gradient of the initial prediction model and its loss function to form a residual set of training samples and fitting a corresponding tree model, and iteratively updating the initial prediction model based on the tree model;

[0079] After the iteration end condition is met, the gradient boosting tree model is output to obtain the prediction model.

[0080] From the above description, it can be seen that by fitting the correlation between input and output data using the gradient boosting tree method and constructing a prediction model, the prediction accuracy of photovoltaic output can be improved.

[0081] The fine-grained photovoltaic output prediction method and terminal described above are applicable to photovoltaic output prediction. By constructing fine-grained features of photovoltaic output influencing variables, the information density of input parameters is improved to solve the problem of insufficient photovoltaic output prediction accuracy. The following is an explanation of the specific implementation method:

[0082] Example 1

[0083] Please refer to Figure 1 , a photovoltaic output prediction method based on fine-grainedness, comprising the steps of:

[0084] S1. Acquire monitoring data from a weather station and a photovoltaic power station, and construct a coarse-grained database based on the monitoring data.

[0085] Specifically, monitoring data from meteorological stations and photovoltaic power plants is collected to construct a multidimensional, coarse-grained database containing meteorological data and photovoltaic power plant operating conditions. In this embodiment, meteorological data includes temperature, relative humidity, global solar radiation, diffuse solar radiation, wind direction, and wind speed; photovoltaic power plant operating conditions include instantaneous photovoltaic output power data within the plant.

[0086] Among them, the coarse-grained database is X=[X1,X2,…,X N ] T , the sample in the coarse-grained database is X i =[x i1 , x i2 ,…,x im ], x ik Represents sample X i The value of the kth parameter, m is the number of multidimensional parameter types.

[0087] S2. Performing time window-based sliding processing on the coarse-grained database to obtain time window trend features and construct a fine-grained database.

[0088] S21. Preset the time window size w and calculate the fine-grained time window trend characteristic value of the sample parameters in the coarse-grained database:

[0089] x ik.Trend =(X ik +X ik+1 +...X ik+w ) / w.

[0090] S22, obtaining a fine-grained database X according to the fine-grained time window trend feature value Trend :

[0091] X Trend =[X 1.Trend ,X 2.Trend ,…,XN.Trend ] T , X i.Trend =[x i1.Trend , x i2.Trend ,…,x im.Trend ].

[0092] S3. Divide the data in the coarse-grained database into a training set, a validation set, and a test set, and construct a coarse-grained prediction model to calculate and obtain a first photovoltaic output value.

[0093] S31 . Using the temperature and humidity, solar energy data, and wind power data in the coarse-grained database as first input data, and using the instantaneous photovoltaic output power of the photovoltaic power station as a first output parameter.

[0094] In this embodiment, temperature, relative humidity, global solar radiation, diffuse solar radiation, wind direction and wind speed are used as the first input data, and instantaneous photovoltaic output power in the station is used as the first output parameter.

[0095] S32: Use a gradient boosting tree to fit the first input data and the first output parameter to construct a coarse-grained prediction model, thereby outputting predicted photovoltaic output power under different input conditions.

[0096] S4. Divide the data in the fine-grained database into a training set, a validation set, and a test set, and construct a fine-grained prediction model to calculate the photovoltaic output error.

[0097] S41. Using the trend characteristic values ​​within the fine-grained time window of the temperature and humidity, solar energy data, and wind power data in the fine-grained database as second input data, and using the photovoltaic output error of the coarse-grained prediction model as a second output parameter.

[0098] In this embodiment, the fine-grained time window trend characteristics of temperature, relative humidity, total solar radiation value, scattered solar radiation value, wind direction and wind speed are used as the second input data, and the photovoltaic output error of the coarse-grained feature prediction model is used as the second output parameter.

[0099] S42. Use a gradient boosting tree to fit the second input data and the second output parameter to construct a fine-grained prediction model, thereby outputting the error of the predicted photovoltaic output power under different input trend conditions.

[0100] S5. Obtain a predicted photovoltaic output value by combining the first photovoltaic output value and the photovoltaic output error.

[0101] Specifically, the test set data is predicted, and the result P1 of the coarse-grained prediction model is fused with the result △P of the fine-grained prediction model to form a fusion model, which outputs the actual output photovoltaic power P2 = P1 + △P.

[0102] Example 2

[0103] The difference between this embodiment and the first embodiment is that the method of fitting the association relationship of the input and output trees using the gradient boosting tree method is further limited. Specifically:

[0104] Step 1: For the input data sample Z used to train the model, i ,…Z N ] and the output parameter sample y=[y1,…y N ] are combined, N is the number of data samples, and a training sample set D = {Z, y} is formed.

[0105] Step 2: Use the root mean square error as the loss function L, calculate the regression tree model parameter γ that minimizes the loss function L(y,γ), and use it as the initial model f0(Z i ):

[0106]

[0107] Step 3. Let C be the number of iterations. For the cth iteration, c = 1, 2, ... C, repeat steps 31 to 34:

[0108] Step 31: Sample Z i , i = 1, 2, ... N, calculate the negative gradient r of the current model loss function and the model ic , where f c-1 (Z) is the model obtained after c-1 iterations.

[0109]

[0110] Step 32, r ic As sample Z i The output parameter samples of the training sample residual set [(Z i ,r ic ),i=1,2,…N].

[0111] Step 33: Set the training sample residual set [(Z i ,r ic ),i=1,2,…N] as new training data, fit the c-th tree model. The model consists of multiple leaf nodes R jc Composition, j = 1, 2, ... J, J is the number of leaf nodes of the regression tree model. jc , j=1,2,…J, calculate its subordinate sample Z i ∈R jc In the current model f c-1 The fitted value γ with the minimum loss function in (Z) jc .

[0112]

[0113] Where, f c-1 (Z i ) is the sample Z i The output value of the model at the c-1th iteration.

[0114] Step 34: Update the strong learning model of the cth iteration:

[0115]

[0116] Among them, I(Z i ∈R jc ) is the indicator function, that is, when the sample Z i Belongs to leaf node R jc The function outputs 1 when , otherwise 0.

[0117] Step 4: After C iterations, output the gradient boosting tree model f c (Z i ), the model fits the input data sample Z=[Z1,…,Z i ,…Z N ] and the output parameter sample y=[y1,…y N ]’s related relationship:

[0118]

[0119] Example 3

[0120] Please refer to Figure 2 , a photovoltaic output prediction terminal based on fine-grainedness, including:

[0121] The data acquisition module is used to obtain monitoring data from the weather station and the photovoltaic power station, and to build a coarse-grained database based on the monitoring data.

[0122] The fine-grained construction module is used to perform time-window-based sliding processing on the coarse-grained database to obtain time-window trend features and construct a fine-grained database.

[0123] Specifically, the coarse-grained database is X=[X1, X2, ..., X N ] T , the sample in the coarse-grained database is X i =[x i1 , x i2 ,…,x im ], where x ik Represents sample X i The value of the kth parameter;

[0124] The time window size w is preset, and the fine-grained time window trend characteristic value of the sample parameters in the coarse-grained database is calculated:

[0125] x ik.Trend =(X ik +X ik+1 +...X ik+w ) / w;

[0126] The fine-grained database X is obtained based on the fine-grained time window trend feature value. Trend :

[0127] X Trend =[X 1.Trend ,X 2.Trend ,…,X N.Trend ] T , X i.Trend =[x i1.Trend , x i2.Trend ,…,x im.Trend ].

[0128] The coarse-grained model construction module is used to divide the data in the coarse-grained database into a training set, a validation set and a test set, and to construct a coarse-grained prediction model to calculate the first photovoltaic output value.

[0129] Specifically, building a coarse-grained prediction model includes:

[0130] Using the temperature and humidity, solar energy data, and wind power data in the coarse-grained database as first input data, and using the instantaneous photovoltaic output power of the photovoltaic power station as a first output parameter;

[0131] The first input data and the first output parameter are fitted using a gradient boosting tree to construct a coarse-grained prediction model.

[0132] The fine-grained result correction module is used to divide the data in the fine-grained database into a training set, a validation set and a test set, and to build a fine-grained prediction model to calculate the photovoltaic output error.

[0133] Specifically, building a fine-grained prediction model includes:

[0134] Using the trend characteristic values ​​within the fine-grained time window of the temperature and humidity, solar energy data, and wind power data in the fine-grained database as second input data, and using the photovoltaic output error of the coarse-grained prediction model as a second output parameter;

[0135] The second input data and the second output parameter are fitted using a gradient boosting tree to construct a fine-grained prediction model.

[0136] The step of fitting the input data and output parameters using a gradient boosting tree to construct a prediction model includes:

[0137] Combine the input data and output parameters in the training set to obtain a training sample set;

[0138] Using the root mean square error as a loss function, calculating the regression tree model parameters that minimize the loss function, and establishing an initial prediction model based on the loss function, the regression tree model parameters and the training sample set;

[0139] Calculating the negative gradient of the initial prediction model and its loss function to form a residual set of training samples and fitting a corresponding tree model, and iteratively updating the initial prediction model based on the tree model;

[0140] After the iteration end condition is met, the gradient boosting tree model is output to obtain the prediction model.

[0141] A prediction module is configured to obtain a predicted photovoltaic output value by combining the first photovoltaic output value and the photovoltaic output error.

[0142] In summary, the present invention provides a fine-grained photovoltaic output prediction method and terminal. This method collects on-site monitoring data from meteorological stations and photovoltaic power plants, constructing a coarse-grained database containing meteorological monitoring data and photovoltaic power plant operating conditions. This method also constructs a fine-grained database based on fine-grained feature construction using a time window sliding average. Subsequently, coarse- and fine-grained prediction models are established, respectively, to obtain a predicted photovoltaic output value based on the first photovoltaic output value output by the coarse-grained prediction model and the photovoltaic output error output by the fine-grained prediction model. Therefore, by constructing fine-grained features for photovoltaic output influencing variables, the information density of the prediction model's input data can be increased, thereby improving the accuracy of photovoltaic output predictions.

[0143] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's description and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A photovoltaic output prediction method based on fine-grainedness, characterized in that: Including steps: Acquire monitoring data from a weather station and a photovoltaic power station, and construct a coarse-grained database based on the monitoring data; Performing time window-based sliding processing on the coarse-grained database to obtain time window trend features and construct a fine-grained database; Dividing the data in the coarse-grained database into a training set, a validation set, and a test set, and constructing a coarse-grained prediction model to calculate a first photovoltaic output value; Dividing the data in the fine-grained database into a training set, a validation set, and a test set, and constructing a fine-grained prediction model to calculate the photovoltaic output error; Combining the first photovoltaic output value and the photovoltaic output error to obtain a predicted photovoltaic output value; Performing time window-based sliding processing on the coarse-grained database to obtain time window trend features and construct a fine-grained database includes: The coarse-grained database is X=[X1,X2,…,X N ] T , the sample in the coarse-grained database is X i =[x i1 , x i2 ,…,x im ], m is the number of multidimensional parameter types; The time window size w is preset, and the fine-grained time window trend characteristic value of the sample parameters in the coarse-grained database is calculated: x ik.Trend =(x ik +x ik+1 +...x ik+w ) / w; Where x ik Represents sample X i The value of the kth parameter; The fine-grained database X is obtained based on the fine-grained time window trend feature value. Trend : X Trend =[X 1.Trend ,X 2.Trend ,…,X N.Trend ] T ,X i.Trend =[x i1.Trend ,x i2.Trend ,…,x im.Trend ]。 2. The photovoltaic output prediction method based on fine granularity according to claim 1 is characterized in that: The construction of the coarse-grained prediction model includes: Using the temperature and humidity, solar energy data, and wind power data in the coarse-grained database as first input data, and using the instantaneous photovoltaic output power of the photovoltaic power station as a first output parameter; The first input data and the first output parameter are fitted using a gradient boosting tree to construct a coarse-grained prediction model.

3. The photovoltaic output prediction method based on fine granularity according to claim 2 is characterized in that: The constructing of the fine-grained prediction model includes: Using the trend characteristic values ​​within the fine-grained time window of the temperature and humidity, solar energy data, and wind power data in the fine-grained database as second input data, and using the photovoltaic output error of the coarse-grained prediction model as a second output parameter; The second input data and the second output parameter are fitted using a gradient boosting tree to construct a fine-grained prediction model.

4. The photovoltaic output prediction method based on fine granularity according to claim 3 is characterized in that: Using a gradient boosting tree to fit the input data and output parameters, constructing a prediction model includes: Combine the input data and output parameters in the training set to obtain a training sample set; Using the root mean square error as a loss function, calculating the regression tree model parameters that minimize the loss function, and establishing an initial prediction model based on the loss function, the regression tree model parameters and the training sample set; Calculating the negative gradient of the initial prediction model and its loss function to form a residual set of training samples and fitting a corresponding tree model, and iteratively updating the initial prediction model based on the tree model; After the iteration end condition is met, the gradient boosting tree model is output to obtain the prediction model.

5. A photovoltaic output prediction terminal based on fine granularity, characterized in that: include: A data acquisition module is used to obtain monitoring data from the weather station and the photovoltaic power station, and to construct a coarse-grained database based on the monitoring data; A fine-grained construction module is used to perform time-window-based sliding processing on the coarse-grained database to obtain time-window trend features and construct a fine-grained database; a coarse-grained model construction module, configured to divide the data in the coarse-grained database into a training set, a validation set, and a test set, and to construct a coarse-grained prediction model to calculate a first photovoltaic output value; A fine-grained result correction module is used to divide the data in the fine-grained database into a training set, a validation set, and a test set, and to construct a fine-grained prediction model to calculate the photovoltaic output error; a prediction module, configured to obtain a predicted photovoltaic output value by combining the first photovoltaic output value and the photovoltaic output error; Performing time window-based sliding processing on the coarse-grained database to obtain time window trend features and construct a fine-grained database includes: The coarse-grained database is X=[X1,X2,…,X N ] T , the sample in the coarse-grained database is X i =[x i1 , x i2 ,…,x im ], m is the number of multidimensional parameter types; The time window size w is preset, and the fine-grained time window trend characteristic value of the sample parameters in the coarse-grained database is calculated: x ik.Trend =(x ik +x ik+1 +...x ik+w ) / w; Where x ik Represents sample X i The value of the kth parameter; The fine-grained database X is obtained based on the fine-grained time window trend feature value. Trend : X Trend =[X 1.Trend ,X 2.Trend ,…,X N.Trend ] T ,X i.Trend =[x i1.Trend ,x i2.Trend ,…,x im.Trend ]。 6. The photovoltaic output prediction terminal based on fine granularity according to claim 5, characterized in that: The construction of the coarse-grained prediction model includes: Using the temperature and humidity, solar energy data, and wind power data in the coarse-grained database as first input data, and using the instantaneous photovoltaic output power of the photovoltaic power station as a first output parameter; The first input data and the first output parameter are fitted using a gradient boosting tree to construct a coarse-grained prediction model.

7. The photovoltaic output prediction terminal based on fine granularity according to claim 6, characterized in that: The constructing of the fine-grained prediction model includes: Using the trend characteristic values ​​within the fine-grained time window of the temperature and humidity, solar energy data, and wind power data in the fine-grained database as second input data, and using the photovoltaic output error of the coarse-grained prediction model as a second output parameter; The second input data and the second output parameter are fitted using a gradient boosting tree to construct a fine-grained prediction model.

8. The photovoltaic output prediction terminal based on fine granularity according to claim 7, characterized in that: Using a gradient boosting tree to fit the input data and output parameters, constructing a prediction model includes: Combine the input data and output parameters in the training set to obtain a training sample set; Using the root mean square error as a loss function, calculating the regression tree model parameters that minimize the loss function, and establishing an initial prediction model based on the loss function, the regression tree model parameters and the training sample set; Calculating the negative gradient of the initial prediction model and its loss function to form a residual set of training samples and fitting a corresponding tree model, and iteratively updating the initial prediction model based on the tree model; After the iteration end condition is met, the gradient boosting tree model is output to obtain the prediction model.

Images