Method for constructing photovoltaic power prediction model and photovoltaic power prediction method

Abstract

The application relates to the technical field of new energy, and discloses a photovoltaic power prediction model construction method and a photovoltaic power prediction method, which are applied to a target photovoltaic power station provided with flat single-axis single-sided photovoltaic components; the photovoltaic power prediction model comprises the following steps: acquiring historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time length; calculating historical surface irradiance data of each flat single-axis single-sided photovoltaic component based on each historical measured irradiance data; dividing the historical surface irradiance data and the historical power data according to a preset time interval to obtain a historical surface irradiance point value sequence and a historical power point value sequence; taking the historical surface irradiance point value sequence as input features and taking the historical power point value sequence as target values to construct a first training data set; and inputting the first training data set into a pre-constructed photovoltaic power prediction model to perform model training, so as to obtain the photovoltaic power prediction model.

Description

Technical Field

[0001] This invention relates to the field of new energy technology, specifically to a training method for a photovoltaic power prediction model and a photovoltaic power prediction method. Background Technology

[0002] Currently, photovoltaic panels in photovoltaic power plants are usually installed using fixed photovoltaic brackets. When predicting photovoltaic power based on this installation method, the output characteristics of the photovoltaic panels are often related to the total horizontal radiation in a linear function. The photovoltaic power generation corresponding to the total horizontal radiation is obtained by using the total horizontal radiation and this linear function relationship during prediction.

[0003] In existing photovoltaic power plants, photovoltaic panels are still installed using flat single-axis photovoltaic brackets. These brackets are horizontal in the north-south direction, and the array plane rotates east-west with the sun's position along its rotation axis. Compared to fixed photovoltaic bracket installations, this method makes full use of direct sunlight in the morning and evening, resulting in better photovoltaic power generation efficiency. However, although the output curve of the flat single-axis bracket is strongly positively correlated with irradiance, the output characteristics and total radiation are no longer a linear function. Therefore, when using the traditional fixed photovoltaic bracket installation method for power prediction in photovoltaic power plants with flat single-axis single-sided photovoltaic modules, the power prediction results are inaccurate. Summary of the Invention

[0004] In view of this, the present invention provides a training method for a photovoltaic power prediction model and a photovoltaic power prediction method to solve the problem of inaccurate power prediction for photovoltaic power plants equipped with flat single-axis single-sided photovoltaic modules.

[0005] In a first aspect, the present invention provides a method for constructing a photovoltaic power prediction model, applicable to a target photovoltaic power station equipped with flat single-axis single-sided photovoltaic modules; the photovoltaic power prediction model includes: acquiring historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period; calculating historical surface irradiance data of each flat single-axis single-sided photovoltaic module based on the historical measured irradiance data; dividing the historical surface irradiance data and the historical power data according to a preset time interval to obtain a historical surface irradiance point value sequence and a historical power point value sequence; constructing a first training dataset by using the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values; inputting the first training dataset into a pre-constructed photovoltaic power prediction model for model training to obtain the photovoltaic power prediction model; and learning the correspondence between each historical surface irradiance point value sequence and the historical power point value sequence during model training.

[0006] As an exemplary embodiment, the step of using the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values ​​to construct a first training dataset includes: obtaining historical difference features of historical surface irradiance data and historical power data at each time point on a daily time scale, and constructing a time-series related historical difference feature sequence; using the historical difference feature sequence and the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values ​​to construct a first training dataset.

[0007] As an exemplary embodiment, the step of acquiring historical difference features of historical surface irradiance data and historical power data at each moment on a daily time scale, and constructing a time-related historical difference feature sequence, includes: calculating the historical change trend of the historical surface irradiance point value sequence relative to the historical power point value sequence at each moment on a daily time scale; dividing the historical change trend into a first sub-historical change trend in which the historical surface irradiance point value sequence is positively correlated with the historical power point value sequence and a second sub-historical change trend in which the historical surface irradiance point value sequence is not positively correlated with the historical power point value sequence; and constructing a time-related first historical difference feature sequence and a second historical difference feature sequence based on the first sub-historical change trend and the second sub-historical change trend, respectively.

[0008] As an exemplary embodiment, the method for constructing the photovoltaic power prediction model further includes: obtaining the first historical change period corresponding to the first historical difference feature sequence and the second historical change period corresponding to the second historical difference feature sequence; dividing the first training dataset based on the first historical change period and the second historical change period to obtain a second training dataset; wherein, the second training dataset includes a second training data subset consisting of multiple historical surface irradiance point value sequences and historical power point value sequences with different historical change trends, divided according to the first historical change period and the second historical change period; inputting each of the second training data subsets into the pre-constructed photovoltaic power prediction model for model training to obtain a first photovoltaic power prediction sub-model corresponding to each historical change period; during the model training process, learning the correspondence between each of the historical surface irradiance point value sequences and historical power point value sequences in the second training data subset corresponding to each historical change period based on each of the first photovoltaic power prediction sub-models.

[0009] As an exemplary embodiment, the method for constructing the photovoltaic power prediction model further includes: obtaining historical temperature point value sequences and / or historical wind speed point value sequences corresponding to each of the second training data subsets; adding the historical temperature point value sequences and / or historical wind speed point value sequences to the corresponding second training data subsets to obtain a third training data subset; inputting each of the third training data subsets into a pre-constructed photovoltaic power prediction model for model training to obtain a second photovoltaic power prediction sub-model corresponding to each of the historical change periods; during the model training process, learning the correspondence between the historical temperature point value sequences and / or historical wind speed point value sequences, the historical surface irradiance point value sequences, and the historical power point value sequences in the third training data subsets corresponding to each of the historical change periods based on each of the second photovoltaic power prediction sub-models.

[0010] Secondly, the present invention provides a photovoltaic power prediction method, applied to a target photovoltaic power station equipped with flat single-axis single-sided photovoltaic modules; the photovoltaic power prediction method includes: acquiring predicted irradiance data of the target photovoltaic power station; calculating predicted surface irradiance data of each of the flat single-axis single-sided photovoltaic modules based on the predicted irradiance data; dividing the predicted surface irradiance data according to a preset time interval to obtain a predicted surface irradiance point value sequence; inputting the predicted surface irradiance point value sequence into a pre-trained photovoltaic power prediction model for power prediction to obtain a power prediction result; wherein, the pre-trained photovoltaic power prediction model is obtained by constructing a first training dataset using historical surface irradiance point value sequences as input features and historical power point value sequences as target values, and the historical surface irradiance point value sequences and the historical power point value sequences are obtained by acquiring historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period, calculating historical surface irradiance data of each of the flat single-axis single-sided photovoltaic modules based on each of the historical measured irradiance data, and dividing the data according to a preset time interval.

[0011] As an exemplary embodiment, the photovoltaic power prediction method further includes: acquiring historical surface irradiance data and historical power data of the target photovoltaic power station at various times on a daily time scale to construct a time-series related historical difference feature sequence; selecting the historical difference feature sequence corresponding to the historical surface irradiance data with a similarity greater than a preset degree to the predicted surface irradiance point value sequence as the predicted difference feature sequence and inputting it into a pre-trained photovoltaic power prediction model for power prediction to obtain the photovoltaic power prediction result.

[0012] As an exemplary embodiment, the step of acquiring historical surface irradiance data and historical power data of the target photovoltaic power station at each moment on a daily time scale to construct a time-series related historical difference feature sequence includes: calculating the historical change trend of the historical surface irradiance point value sequence relative to the historical power point value sequence at each moment on a daily time scale; dividing the historical change trend into a first sub-historical change trend in which the historical surface irradiance point value sequence and the historical power point value sequence are positively correlated and a second sub-historical change trend in which the historical surface irradiance point value sequence and the historical power point value sequence are not positively correlated; and constructing a time-series related first historical difference feature sequence and a second historical difference feature sequence based on the first sub-historical change trend and the second sub-historical change trend, respectively.

[0013] As an exemplary embodiment, the photovoltaic power prediction method further includes: obtaining the predicted change period corresponding to each predicted surface irradiance point value sequence on a daily time scale; based on the predicted change period, inputting the predicted surface irradiance point value corresponding to the predicted change period into a first photovoltaic power prediction sub-model corresponding to a historical change period that matches the predicted change period for power prediction, thereby obtaining a photovoltaic power prediction result; wherein, the first photovoltaic power prediction sub-model is trained by a second training data subset consisting of a historical surface irradiance point value sequence and a historical power point value sequence corresponding to the historical change period, and learning the correspondence between the historical surface irradiance point value sequence and the historical power point value sequence in the second training data subset corresponding to each historical change period during the training process.

[0014] As an exemplary embodiment, the photovoltaic power prediction method further includes: obtaining the predicted temperature point value sequence and / or predicted wind speed point value sequence corresponding to the predicted surface irradiance point value for each predicted change period; inputting the predicted surface irradiance point value and the predicted temperature point value sequence and / or predicted wind speed point value sequence into a pre-trained second photovoltaic power prediction sub-model, wherein the second photovoltaic power prediction sub-model is a third training data subset composed of the predicted temperature point value sequence and / or predicted wind data point value sequence, the predicted surface irradiance point value sequence and the predicted surface irradiance point value sequence according to the predicted change trend under the predicted change period, and learning the correspondence between the predicted temperature point value sequence and / or predicted wind data point value sequence, the predicted surface irradiance point value sequence and the predicted power point value sequence during the training process.

[0015] This invention provides a method for constructing a photovoltaic power prediction model, applicable to a target photovoltaic power station equipped with flat single-axis single-sided photovoltaic modules. The photovoltaic power prediction model includes: acquiring historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period; calculating historical surface irradiance data of each flat single-axis single-sided photovoltaic module based on the historical measured irradiance data; dividing the historical surface irradiance data and the historical power data according to a preset time interval to obtain a historical surface irradiance point value sequence and a historical power point value sequence; constructing a first training dataset using the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values; inputting the first training dataset into a pre-constructed photovoltaic power prediction model for model training to obtain the photovoltaic power prediction model; and learning the correspondence between each historical surface irradiance point value sequence and the historical power point value sequence during model training.

[0016] As an exemplary embodiment, the step of using the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values ​​to construct a first training dataset includes: obtaining historical difference features of historical surface irradiance data and historical power data at each time point on a daily time scale, and constructing a time-series related historical difference feature sequence; using the historical difference feature sequence and the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values ​​to construct a first training dataset.

[0017] As an exemplary embodiment, the step of acquiring historical difference features of historical surface irradiance data and historical power data at each moment on a daily time scale, and constructing a time-related historical difference feature sequence, includes: calculating the historical change trend of the historical surface irradiance point value sequence relative to the historical power point value sequence at each moment on a daily time scale; dividing the historical change trend into a first sub-historical change trend in which the historical surface irradiance point value sequence is positively correlated with the historical power point value sequence and a second sub-historical change trend in which the historical surface irradiance point value sequence is not positively correlated with the historical power point value sequence; and constructing a time-related first historical difference feature sequence and a second historical difference feature sequence based on the first sub-historical change trend and the second sub-historical change trend, respectively.

[0018] As an exemplary embodiment, the method for constructing the photovoltaic power prediction model further includes: obtaining the first historical change period corresponding to the first historical difference feature sequence and the second historical change period corresponding to the second historical difference feature sequence; dividing the first training dataset based on the first historical change period and the second historical change period to obtain a second training dataset; wherein, the second training dataset includes a second training data subset consisting of multiple historical surface irradiance point value sequences and historical power point value sequences with different historical change trends, divided according to the first historical change period and the second historical change period; inputting each of the second training data subsets into the pre-constructed photovoltaic power prediction model for model training to obtain a first photovoltaic power prediction sub-model corresponding to each historical change period; during the model training process, learning the correspondence between each of the historical surface irradiance point value sequences and historical power point value sequences in the second training data subset corresponding to each historical change period based on each of the first photovoltaic power prediction sub-models.

[0019] As an exemplary embodiment, the method for constructing the photovoltaic power prediction model further includes: obtaining historical temperature point value sequences and / or historical wind speed point value sequences corresponding to each of the second training data subsets; adding the historical temperature point value sequences and / or historical wind speed point value sequences to the corresponding second training data subsets to obtain a third training data subset; inputting each of the third training data subsets into a pre-constructed photovoltaic power prediction model for model training to obtain a second photovoltaic power prediction sub-model corresponding to each of the historical change periods; during the model training process, learning the correspondence between the historical temperature point value sequences and / or historical wind speed point value sequences, the historical surface irradiance point value sequences, and the historical power point value sequences in the third training data subsets corresponding to each of the historical change periods based on each of the second photovoltaic power prediction sub-models.

[0020] This invention provides a method for constructing a photovoltaic power prediction model, applicable to a target photovoltaic power station equipped with flat single-axis single-sided photovoltaic modules. The photovoltaic power prediction model includes: acquiring historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period; calculating historical surface irradiance data for each of the flat single-axis single-sided photovoltaic modules based on the historical measured irradiance data; dividing the historical surface irradiance data and the historical power data according to a preset time interval to obtain a historical surface irradiance point value sequence and a historical power point value sequence; constructing a first training dataset using the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values; inputting the first training dataset into a pre-constructed photovoltaic power prediction model for model training to obtain the photovoltaic power prediction model; and in the model... During training, the correspondence between the historical surface irradiance point value sequences and the historical power point value sequences is learned. In the construction method of the photovoltaic power prediction model of this application, the first training dataset is constructed by using the historical surface irradiance point value sequences as input features and the historical power point value sequences as target values. The historical surface irradiance point value sequences are calculated based on the historical measured irradiance data, taking into account the changes in irradiance caused by the east-west rotation of the photovoltaic panel array plane with the position of the sun along the rotation axis. Therefore, the photovoltaic power prediction model of this embodiment can learn the impact of the changes in photovoltaic panel surface irradiance caused by the installation form of the photovoltaic panel with the flat single-axis photovoltaic bracket on photovoltaic power prediction during the training process, and can adaptively predict photovoltaic power for photovoltaic panels with the flat single-axis photovoltaic bracket installation form. Attached Figure Description

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a graph showing the irradiance and power curves of a photovoltaic panel mounted on a single-axis photovoltaic bracket in related technologies.

[0023] Figure 2 This is a flowchart of a method for constructing a photovoltaic power prediction model according to an embodiment of the present invention;

[0024] Figure 3 This is a flowchart of a photovoltaic power prediction method according to an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In related technologies, photovoltaic panels in photovoltaic power plants are usually installed using fixed photovoltaic brackets. When predicting photovoltaic power based on this installation method, the output characteristics of the photovoltaic panels are often related to the total horizontal radiation in a linear function. The photovoltaic power generation corresponding to the total horizontal radiation is obtained by using the total horizontal radiation and this linear function relationship during prediction.

[0027] However, some photovoltaic (PV) panels in PV power plants are installed using a single-axis flat PV bracket. This bracket is horizontal in the north-south direction, and the array plane rotates east-west around its axis according to the sun's position. Compared to fixed PV bracket installations, this method makes full use of direct sunlight in the morning and evening, resulting in better PV power generation efficiency. However, although the output curve of this single-axis bracket has a strong positive correlation with irradiance, the output characteristics and total radiation are no longer a linear function. Specifically, Figure 1 This is a graph showing the output characteristics of fixed-mounted photovoltaic panels and flat-mounted single-axis photovoltaic panels in related technologies, such as... Figure 1 As shown, the first output characteristic curve 11 is the curve of the output characteristic of photovoltaic panels installed on a flat single-axis photovoltaic bracket, and the second output characteristic curve 12 is the curve of the output characteristic of fixed photovoltaic panels. The trend of fixed photovoltaic panels basically changes with the trend of solar irradiance. However, the irradiance received by flat single-axis photovoltaic panels changes continuously with the rotation of the bracket and the azimuth angle, elevation angle and incident angle over time. This results in the final output characteristic showing that the output is highest in the morning and evening and "concave" at noon. On a daily time scale, it can be divided into the first output characteristic events 13 and 15, where solar irradiance and power change positively, and the second output characteristic event 14, where solar irradiance and power change non-positively. Therefore, for photovoltaic power plants with flat single-axis single-sided photovoltaic modules, the power prediction results are inaccurate when the traditional prediction method of fixed photovoltaic bracket installation is still used for power prediction.

[0028] To address the aforementioned problems, in a first aspect, according to an embodiment of the present invention, a method for constructing a photovoltaic power prediction model is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0029] This embodiment provides a method for constructing a photovoltaic power prediction model. Figure 2 This is a flowchart of a method for constructing a photovoltaic power prediction model according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps:

[0030] Step S101: Obtain historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period.

[0031] For example, historical measured irradiance data can be obtained by sensors that detect solar irradiance. For instance, it can be obtained by installing a total solar radiation sensor at the target photovoltaic power station and reading historical data from the sensor; it can also be obtained from historical weather forecasts of the target photovoltaic power station; historical power data can be obtained by reading historical data from the target photovoltaic power station.

[0032] In this invention, the expected power generation of the target photovoltaic power station can be the ultra-short-term power generation; wherein, the prediction period of the ultra-short-term power generation is usually 24 hours, and the time resolution of the predicted power data is 15 minutes per day; therefore, in this invention, the preset duration is a duration greater than 24 hours.

[0033] Step S102: Calculate the historical surface irradiance data of each flat single-axis single-sided photovoltaic module based on the historical measured irradiance data.

[0034] As mentioned above, the irradiance received by the photovoltaic panels installed on the horizontal single-axis photovoltaic support changes continuously with the rotation of the support and the azimuth, elevation, and incident angles over time, resulting in the final output characteristics being highest in the morning and evening, and "dipping" at midday. Therefore, after obtaining historical measured irradiance data, the irradiance of the horizontal single-axis photovoltaic module (Monofacial PV with Horizontal Single-axis Tracking, MHST) can be calculated according to the latitude, longitude, elevation, incident angle, and other factors of the target photovoltaic power station using the following formula. Specifically, the MHST can be calculated using the formula (1):

[0035]

[0036] In equation (1), represents the tilt angle of the single-axis photovoltaic module relative to the horizontal plane, which varies with the solar hour angle ω, and ρ represents ground reflection. Q is the local latitude, α is the solar altitude angle (the angle between the solar rays and the horizontal plane), and δ is the solar declination angle; where Q DNI Horizontal direct radiation irradiance, Q DHI Q represents the horizontal diffuse irradiance. GRI Q represents the ground reflected irradiance. MHST The historical surface irradiance data for each flat single-axis single-sided photovoltaic module obtained from the calculation.

[0037] After obtaining the historical measured irradiance data, Q can be calculated for each historical measured irradiance data using formula (1). MHST .

[0038] The method described in this embodiment calculates the historical surface irradiance data of each flat single-axis single-sided photovoltaic module based on historical measured irradiance data. It can take into account the changes in irradiance caused by the east-west rotation of the photovoltaic array plane of the flat single-axis photovoltaic support along the rotation axis with the position of the sun. In this way, the model can consider the impact of changes in irradiance on the output power of the photovoltaic panel during subsequent model training.

[0039] Step S103: Divide the historical surface irradiance data and historical power data according to a preset time interval to obtain the historical surface irradiance point value sequence and the historical power point value sequence;

[0040] For example, in order to match the time scale of the input data with the time scale of the predicted power data during model training, the preset time interval can be 24h; in this embodiment, the historical surface irradiance data and historical power data are divided into historical surface irradiance point value sequences and historical power point value sequences with a period of 24h and a time resolution of 15min.

[0041] Step S104: Use the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values ​​to construct the first training dataset.

[0042] Step S105: Input the first training dataset into the pre-built photovoltaic power prediction model for model training to obtain the photovoltaic power prediction model; during the model training process, learn the correspondence between the historical surface irradiance point value sequence and the historical power point value sequence.

[0043] In this embodiment, the photovoltaic power prediction model can be a convolutional neural network (CNN) model; wherein, the input feature of the CNN model is a sequence of surface irradiance point values, and the output feature is a sequence of power point values.

[0044] The photovoltaic power prediction model construction method of this application uses historical surface irradiance point value sequences as input features and historical power point value sequences as target values ​​to construct the first training dataset. The historical surface irradiance point value sequences are calculated based on the changes in irradiance caused by the east-west rotation of the photovoltaic panel array plane with the position of the sun along the rotation axis, considering the historical measured irradiance data. Therefore, the photovoltaic power prediction model of this embodiment can learn the impact of changes in photovoltaic panel surface irradiance caused by the installation form of photovoltaic panel installation with flat single-axis photovoltaic brackets on photovoltaic power prediction during the training process, and can adaptively predict photovoltaic power for photovoltaic panels with flat single-axis photovoltaic bracket installation.

[0045] As an exemplary embodiment, constructing a first training dataset by using a historical surface irradiance point value sequence as input features and a historical power point value sequence as target values ​​includes: obtaining historical difference features of historical surface irradiance data and historical power data at each time point on a daily time scale, and constructing a time-related historical difference feature sequence; using the historical difference feature sequence and the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values ​​to construct a first training dataset.

[0046] The irradiance received by a flat-axis photovoltaic system changes continuously over time due to the rotation of the support structure and the variations in azimuth, elevation, and incident angles. This results in the power output being highest in the morning and evening, with a dip at midday. To ensure that the model can consider the differences between the aforementioned power output characteristics and existing technologies where irradiance and power are linearly related during model training, this embodiment constructs a historical difference feature sequence based on the historical differences between historical surface irradiance data and historical power data at various times on a daily timescale. This historical difference feature sequence represents the power output characteristics suitable for flat-axis photovoltaic systems. Specifically, the historical difference features may include the historical trend of the historical surface irradiance point value sequence relative to the historical power point value sequence at each time point. This historical trend can be the slope or difference in change of the historical surface irradiance point value sequence relative to the historical power point value sequence.

[0047] Furthermore, a time-related historical difference feature sequence is constructed; the historical difference feature sequence and the historical surface irradiance point value sequence are used as input features, and the historical power point value sequence is used as the target value to construct the first training dataset. The model is trained based on the first training dataset so that the model can consider the correspondence between the historical surface irradiance point value sequence and the power sequence corresponding to the historical difference feature sequence reflecting the difference in output characteristics brought about by the installation form of the phase-flat single-axis photovoltaic.

[0048] Based on this, as an exemplary embodiment, historical difference characteristics of historical surface irradiance data and historical power data at each time point on a daily time scale are obtained, and a time-series related historical difference characteristic sequence is constructed, including: calculating the historical change trend of the historical surface irradiance point value sequence at each time point on a daily time scale relative to the historical power point value sequence; dividing the historical change trend into a first sub-historical change trend in which the historical surface irradiance point value sequence and the historical power point value sequence are positively correlated and a second sub-historical change trend in which the historical surface irradiance point value sequence and the historical power point value sequence are not positively correlated; and constructing a time-series related first historical difference characteristic sequence and a second historical difference characteristic sequence based on the first sub-historical change trend and the second sub-historical change trend, respectively.

[0049] In this embodiment, the historical trend is divided into a first sub-historical trend in which the historical surface irradiance point value sequence and the historical power point value sequence are positively correlated, and a second sub-historical trend in which the historical surface irradiance point value sequence and the historical power point value sequence are not positively correlated. Then, a time-series-related first historical difference feature sequence and a second historical difference feature sequence are constructed based on the first and second sub-historical trend. The construction method of the above difference feature sequence is such that the first historical difference feature sequence can be used to describe the state in which irradiance and power are positively correlated in the installation mode of flat single-axis photovoltaic, and the second historical difference feature can be used to describe the state in which irradiance and power are not positively correlated in the installation mode of flat single-axis photovoltaic.

[0050] Furthermore, the historical difference feature sequence and the historical surface irradiance point value sequence are used as input features, and the historical power point value sequence is used as the target value to construct the first training dataset. When the first training dataset is input into the pre-constructed photovoltaic power prediction model for model training, the model can determine the output state of the currently input historical irradiance point value sequence through the aforementioned first and second historical difference feature sequences. In this way, it can adaptively learn the positive or non-positive correlation between the historical power point value sequence and the historical irradiance point value sequence to obtain a power prediction model that is adapted to the photovoltaic output characteristics of the flat single-axis photovoltaic installation method.

[0051] To enable the model to learn the correspondence between irradiance and power at different time periods, as an exemplary embodiment, the method for constructing a photovoltaic power prediction model further includes: obtaining the first historical change period corresponding to the first historical difference feature sequence and the second historical change period corresponding to the second historical difference feature; dividing the first training dataset based on the first historical change period and the second historical change period to obtain a second training dataset; wherein, the second training dataset includes a subset of second training data consisting of multiple historical surface irradiance point value sequences and historical power point value sequences with different historical change trends, divided according to the first historical change period and the second historical change period; inputting each subset of second training data into the pre-constructed photovoltaic power prediction model for model training to obtain a first photovoltaic power prediction sub-model corresponding to each historical change period; during the model training process, learning the correspondence between each historical surface irradiance point value sequence and historical power point value sequence in the second training data subset corresponding to each historical change period based on each first photovoltaic power prediction sub-model.

[0052] In this embodiment, by obtaining the first historical change period corresponding to the first historical difference feature sequence and the second historical change period corresponding to the second historical difference feature, the first training dataset is further divided based on the first historical change period and the second historical change period to obtain the second training dataset. The first historical change period can be used to describe the time when irradiance and power are positively correlated in the installation mode of flat single-axis photovoltaic, and the second historical difference feature can be used to describe the time when irradiance and power are not positively correlated in the installation mode of flat single-axis photovoltaic.

[0053] Furthermore, the first training dataset is divided into two parts based on the first and second historical change periods to obtain the second training dataset. Each subset of the second training dataset is then input into a pre-built photovoltaic power prediction model for model training, resulting in a first photovoltaic power prediction sub-model corresponding to each historical change period. Each first photovoltaic power prediction sub-model can determine the correspondence between the output state and time of the currently input historical irradiance point value sequence through the aforementioned first and second historical change periods. In turn, it can adaptively learn the time periods to which the historical power point value sequence and the historical irradiance point value sequence belong, or the time periods to which they belong, based on a positive correlation or a non-positive correlation, to obtain a power prediction model that is adapted to the photovoltaic output characteristics of the flat single-axis photovoltaic installation method.

[0054] As an exemplary embodiment, the method for constructing a photovoltaic power prediction model further includes: obtaining historical temperature point value sequences and / or historical wind speed point value sequences corresponding to each second training data subset; adding the historical temperature point value sequences and / or historical wind speed point value sequences to the corresponding second training data subsets to obtain a third training data subset; inputting each third training data subset into a pre-constructed photovoltaic power prediction model for model training to obtain a second photovoltaic power prediction sub-model corresponding to each historical change period; during the model training process, learning the correspondence between historical temperature point value sequences and / or historical wind speed point value sequences, historical surface irradiance point value sequences, and historical power point value sequences in the third training data subsets corresponding to each historical change period based on each second photovoltaic power prediction sub-model.

[0055] Based on the theoretical power generation curve of photovoltaic panels, the actual power generation of photovoltaic panels is affected by temperature. Under the same irradiance, the power generation of photovoltaic panels first increases and then decreases with increasing temperature. As a possible implementation method, in order to consider the impact of temperature on the power generation of photovoltaic panels, the historical temperature point value sequence corresponding to each second training data subset is obtained. The historical temperature point value sequence is added to the corresponding second training data subset to obtain the third training data subset. Each third training data subset is input into a pre-built photovoltaic power prediction model for model training to obtain the second photovoltaic power prediction sub-model corresponding to each historical change period. During the model training process, the correspondence between the historical temperature point value sequence, the historical surface irradiance point value sequence, and the historical power point value sequence in the third training data subset corresponding to each historical change period is learned based on each second photovoltaic power prediction sub-model.

[0056] Furthermore, in the installation of a single-axis photovoltaic (PV) bracket, the bracket is horizontal in the north-south direction, and the array plane rotates east-west around the axis of rotation with the position of the sun. The rotation angle of the array plane may be affected by the wind force in the east-west direction. As a possible implementation method, in order to consider the impact of wind speed on the power generation of the PV panels, the historical wind speed point value sequence corresponding to each second training data subset is obtained. The historical wind speed point value sequence is added to the corresponding second training data subset to obtain the third training data subset. Each third training data subset is input into a pre-built PV power prediction model for model training to obtain the second PV power prediction sub-model corresponding to each historical change period. During the model training process, based on each second PV power prediction sub-model, the correspondence between the historical wind speed point value sequence, the historical surface irradiance point value sequence, and the historical power point value sequence in the third training data subset corresponding to each historical change period is learned.

[0057] For example, the historical wind speed point value sequence corresponding to each second training data subset can be the wind speed value that actually affects the east-west rotation of the array plane with the position of the sun along the rotation axis; in this embodiment, the historical wind speed point value sequence can be the wind speed component along the east-west direction of the historical wind speed obtained by historical wind speed, so as to consider that the rotation angle of the east-west rotating array plane may be affected by the wind force in the east-west direction.

[0058] As another exemplary embodiment, in order to simultaneously consider the impact of temperature and wind speed on the power generation of photovoltaic panels, the method for constructing a photovoltaic power prediction model further includes: obtaining historical temperature point value sequences and historical wind speed point value sequences corresponding to each second training data subset; adding the historical temperature point value sequences and historical wind speed point value sequences to the corresponding second training data subsets to obtain a third training data subset; inputting each third training data subset into a pre-constructed photovoltaic power prediction model for model training to obtain a second photovoltaic power prediction sub-model corresponding to each historical change period; during the model training process, learning the correspondence between historical temperature point value sequences, historical wind speed point value sequences, historical surface irradiance point value sequences, and historical power point value sequences in the third training data subsets corresponding to each historical change period based on each second photovoltaic power prediction sub-model.

[0059] Secondly, the present invention provides a photovoltaic power prediction method, which is applied to a target photovoltaic power station equipped with flat single-axis single-sided photovoltaic modules; Figure 3 A photovoltaic power prediction method according to an embodiment of the present invention includes:

[0060] Step S201: Obtain the predicted irradiance data of the target photovoltaic power station.

[0061] In this embodiment, the predicted irradiance data of the target photovoltaic power station can be obtained through weather forecasts; wherein, the predicted irradiance data includes at least horizontal direct radiation irradiance, horizontal diffuse radiation irradiance, and ground reflected radiation irradiance.

[0062] Step S202: Calculate the predicted surface irradiance data for each flat single-axis single-sided photovoltaic module based on the predicted irradiance data.

[0063] After obtaining the predicted irradiance data, the predicted surface irradiance data of each flat single-axis single-sided photovoltaic module can be calculated based on the predicted irradiance data using Equation (1).

[0064] Step S203: Divide the predicted surface irradiance data according to a preset time interval to obtain a sequence of predicted surface irradiance point values.

[0065] In this embodiment, the predicted surface irradiance data needs to be converted according to the data input format of the power prediction model; wherein, the preset time interval is 15 minutes, so as to convert the predicted surface irradiance data into a sequence of predicted surface irradiance point values ​​with a 15-minute interval.

[0066] Step S204: Input the predicted surface irradiance point value sequence into the pre-trained photovoltaic power prediction model to predict the power and obtain the power prediction result. The pre-trained photovoltaic power prediction model is trained by constructing the first training dataset with the historical surface irradiance point value sequence as the input feature and the historical power point value sequence as the target value. The historical surface irradiance point value sequence and the historical power point value sequence are obtained by acquiring the historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period, calculating the historical surface irradiance data of each flat single-axis single-sided photovoltaic module based on each historical measured irradiance data, and dividing it according to the preset time interval.

[0067] In this embodiment, the pre-trained photovoltaic power prediction model is trained using a first training dataset constructed with historical surface irradiance point value sequences as input features and historical power point value sequences as target values. The historical surface irradiance point value sequences and historical power point value sequences are obtained by acquiring historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period, calculating the historical surface irradiance data of each flat single-axis single-sided photovoltaic module based on each historical measured irradiance data, and dividing the data according to a preset time interval. The pre-trained photovoltaic power prediction model is constructed using historical surface irradiance point value sequences as input features and historical power point value sequences as target values, while the historical surface irradiance point value sequences are based on each historical measured irradiance data. The photovoltaic power prediction model in this embodiment is calculated based on the changes in irradiance caused by the east-west rotation of the photovoltaic panel array plane with the sun's position along the rotation axis. Therefore, during training, the photovoltaic power prediction model can learn the impact of changes in photovoltaic panel surface irradiance caused by the installation method of the photovoltaic panel mounted on the single-axis photovoltaic bracket on photovoltaic power prediction. It can adaptively predict photovoltaic power for photovoltaic panels with the single-axis photovoltaic bracket installation method.

[0068] As an exemplary embodiment, the photovoltaic power prediction method further includes: acquiring historical surface irradiance data and historical power data of the target photovoltaic power station at each time point on a daily time scale to construct a time-series related historical difference feature sequence; selecting the historical difference feature sequence corresponding to the historical surface irradiance data with a similarity greater than a preset degree to the predicted surface irradiance point value sequence as the predicted difference feature sequence and inputting it into the pre-trained photovoltaic power prediction model for power prediction to obtain the photovoltaic power prediction result.

[0069] In this embodiment, during model training, to enable the model to consider the differences between the aforementioned output characteristics and existing technologies where irradiance and power exhibit a linear functional relationship, a historical difference feature sequence is constructed based on the historical difference characteristics of historical surface irradiance data and historical power data at each time point on a daily timescale. This historical difference feature sequence represents the output characteristics as exhibiting the highest output in the morning and evening, with a dip in output at midday. Specifically, the historical difference feature may include the historical trend of the historical surface irradiance point value sequence relative to the historical power point value sequence at each time point. This historical trend can be represented by the historical surface irradiance... The slope and difference of the change in the power point value sequence relative to the historical power point value sequence are considered. Based on this, in this embodiment, the historical difference feature sequence corresponding to the historical surface irradiance data with a similarity greater than a preset degree to the predicted surface irradiance point value sequence is selected as the predicted difference feature sequence. This allows the actual output characteristics corresponding to the current predicted surface irradiance point value sequence to be considered during the prediction process. It also allows the influence of the change in surface irradiance of the photovoltaic panel caused by the installation form of the flat single-axis photovoltaic bracket on the photovoltaic power prediction, and enables photovoltaic power prediction to be adaptively performed for photovoltaic panels with flat single-axis photovoltaic bracket installation.

[0070] As an exemplary embodiment, the acquisition of historical surface irradiance data and historical power data of the target photovoltaic power station at each time point on a daily time scale constitutes a time-series correlated historical difference feature sequence. This includes: calculating the historical change trend of the historical surface irradiance point value sequence at each time point on a daily time scale relative to the historical power point value sequence; dividing the historical change trend into a first sub-historical change trend in which the historical surface irradiance point value sequence and the historical power point value sequence are positively correlated and a second sub-historical change trend in which the historical surface irradiance point value sequence and the historical power point value sequence are not positively correlated; and constructing a time-series correlated first historical difference feature sequence and a second historical difference feature sequence based on the first sub-historical change trend and the second sub-historical change trend, respectively.

[0071] In this embodiment, the historical trend is divided into a first sub-historical trend in which the historical surface irradiance point value sequence and the historical power point value sequence are positively correlated, and a second sub-historical trend in which the historical surface irradiance point value sequence and the historical power point value sequence are not positively correlated. Then, a time-series-related first historical difference feature sequence and a second historical difference feature sequence are constructed based on the first and second sub-historical trend. The construction method of the above difference feature sequence is such that the first historical difference feature sequence can be used to describe the feature label of the state in which irradiance and power are positively correlated in the installation mode of flat single-axis photovoltaic, and the second historical difference feature can be used to describe the feature label of the state in which irradiance and power are not positively correlated in the installation mode of flat single-axis photovoltaic.

[0072] Furthermore, the first training dataset is constructed by using historical difference feature sequences and historical surface irradiance point value sequences as input features and historical power point value sequences as target values. When selecting historical difference feature sequences corresponding to historical surface irradiance data with a similarity greater than a preset degree to the predicted surface irradiance point value sequences as predicted difference feature sequences, and inputting the predicted difference feature sequences into the pre-trained photovoltaic power prediction model for power prediction, the model can determine the output state of the currently input predicted irradiance point value sequences through the above-mentioned predicted difference features, and thus can adaptively predict power based on the output state.

[0073] As an exemplary embodiment, the photovoltaic power prediction method further includes: obtaining the predicted change period corresponding to each predicted surface irradiance point value sequence on a daily time scale; based on the predicted change period, inputting the predicted surface irradiance point value corresponding to the predicted change period into a first photovoltaic power prediction sub-model corresponding to a historical change period that matches the predicted change period for power prediction, and obtaining the photovoltaic power prediction result; wherein, the first photovoltaic power prediction sub-model is trained by a second training data subset consisting of a historical surface irradiance point value sequence and a historical power point value sequence according to the historical change trend corresponding to the historical change period, and learning the correspondence between each historical surface irradiance point value sequence and the historical power point value sequence in the second training data subset corresponding to each historical change period during the training process.

[0074] In this embodiment, the first photovoltaic power prediction sub-model is trained using a second training data subset consisting of historical surface irradiance point value sequences and historical power point value sequences corresponding to historical change trends under historical change periods. During the training process, the model learns the correspondence between historical surface irradiance point value sequences and historical power point value sequences in the second training data subset corresponding to each historical change period. By obtaining the first historical change period corresponding to the first historical difference feature sequence and the second historical change period corresponding to the second historical difference feature, the first training dataset is further divided based on the first historical change period and the second historical change period to obtain the second training dataset. The first historical change period can be used to describe the time label of a state where irradiance and power are positively correlated in the installation mode of flat single-axis photovoltaic, and the second historical difference feature can be used to describe the time label of a state where irradiance and power are not positively correlated in the installation mode of flat single-axis photovoltaic.

[0075] As an exemplary embodiment, the photovoltaic power prediction method further includes: obtaining the predicted temperature point value sequence and / or predicted wind speed point value sequence corresponding to the predicted surface irradiance point value for each predicted change period; inputting the predicted surface irradiance point value and the predicted temperature point value sequence and / or predicted wind speed point value sequence into a pre-trained second photovoltaic power prediction sub-model, wherein the second photovoltaic power prediction sub-model is a third training data subset composed of the predicted temperature point value sequence and / or predicted wind data point value sequence, the predicted surface irradiance point value sequence and the predicted power point value sequence according to the predicted change trend under the predicted change period, and learning the correspondence between the predicted temperature point value sequence and / or predicted wind data point value sequence, the predicted surface irradiance point value sequence and the predicted power point value sequence during the training process.

[0076] Based on the theoretical power generation curve of the photovoltaic panel, the actual power generation of the photovoltaic panel is affected by temperature. Under the same irradiance, the power generation of the photovoltaic panel first increases and then decreases with the increase of temperature. As a possible implementation method, in order to consider the influence of temperature on the power generation of the photovoltaic panel, the predicted temperature point value sequence corresponding to the predicted surface irradiance point value for each predicted change period is obtained. The predicted temperature point value sequence and the predicted surface irradiance point value sequence are input into the pre-trained second photovoltaic power prediction sub-model for power prediction to obtain the power prediction result.

[0077] Furthermore, in the installation of a single-axis photovoltaic support, the support is horizontal in the north-south direction, and the array plane rotates east-west with the position of the sun around its rotation axis. The rotation angle of the array plane may be affected by the wind force in the east-west direction. As a possible implementation method, in order to consider the impact of wind speed on the power generation of the photovoltaic panel, the predicted wind speed point value sequence corresponding to the predicted surface irradiance point value for each predicted change period is obtained. The predicted wind speed point value sequence and the predicted irradiance point value sequence are input into the pre-constructed second photovoltaic power prediction sub-model for power prediction to obtain the power prediction result.

[0078] For example, the predicted wind speed point value sequence may be the wind speed values ​​that actually affect the east-west rotation of the array plane with respect to the position of the sun along the rotation axis; in this embodiment, the predicted wind speed point value sequence may be the wind speed component along the east-west direction of the historical wind speed obtained by historical wind speed, so as to take into account the influence of the east-west wind force on the rotation angle of the array plane rotating east-west.

[0079] As another exemplary embodiment, in order to simultaneously consider the impact of temperature and wind speed on the power generation of photovoltaic panels, the method for constructing a photovoltaic power prediction model further includes: obtaining the predicted temperature point value sequence and the predicted wind speed point value sequence corresponding to the predicted surface irradiance point value for each predicted change period; inputting the predicted surface irradiance point value, the predicted temperature point value sequence, and the predicted wind speed point value sequence into a pre-trained second photovoltaic power prediction sub-model, wherein the second photovoltaic power prediction sub-model is a third training data subset composed of the predicted temperature point value sequence, the predicted wind data point value sequence, the predicted surface irradiance point value sequence, and the predicted power point value sequence according to the predicted change trend under the predicted change period, and learning the correspondence between the predicted temperature point value sequence and the predicted wind data point value sequence, the predicted surface irradiance point value sequence, and the predicted power point value sequence during the training process.

[0080] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0081] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more computer devices (which may be personal computers, servers, or network devices, etc.) to execute all or part of the steps of the methods in the above embodiments.

[0082] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0083] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0084] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the solution provided in this embodiment, depending on actual needs.

[0085] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0086] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

[0087] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0088] The above are merely preferred embodiments of this application. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principles of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for constructing a photovoltaic power prediction model, characterized in that, The photovoltaic power prediction model is applied to target photovoltaic power plants equipped with flat single-axis single-sided photovoltaic modules; the photovoltaic power prediction model includes: Acquire historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period; Calculate the historical surface irradiance data of each of the aforementioned flat single-axis single-sided photovoltaic modules based on the historical measured irradiance data of each module. The historical surface irradiance data and the historical power data are divided according to a preset time interval to obtain a historical surface irradiance point value sequence and a historical power point value sequence; Using the historical surface irradiance point value sequence as input features and the historical power point value sequence as target values, a first training dataset is constructed, including: Historical surface irradiance and power data at various times on a daily timescale were obtained to identify historical differences and construct a time-series correlated sequence of historical differences, including: Calculate the historical trend of the historical surface irradiance point value sequence relative to the historical power point value sequence at each time point on a daily time scale; The historical trend is divided into a first sub-historical trend in which the historical surface irradiance point value sequence is positively correlated with the historical power point value sequence, and a second sub-historical trend in which the historical surface irradiance point value sequence is not positively correlated with the historical power point value sequence. A time-series-related first historical difference feature sequence and a second historical difference feature sequence are constructed based on the first sub-historical change trend and the second sub-historical change trend, respectively. The first training dataset is constructed by using the historical difference feature sequence and the historical surface irradiance point value sequence as input features and the historical power point value sequence as target value. The first training dataset is input into the pre-built photovoltaic power prediction model for model training to obtain the photovoltaic power prediction model; during the model training process, the correspondence between the historical surface irradiance point value sequence and the historical power point value sequence is learned. 2.The method of claim 1, wherein, The method for constructing the photovoltaic power prediction model also includes: The first historical change period corresponding to the first historical difference feature sequence and the second historical change period corresponding to the second historical difference feature sequence are obtained respectively; The first training dataset is divided into two parts based on the first historical change period and the second historical change period to obtain a second training dataset. The second training dataset includes a subset of the second training data consisting of multiple historical surface irradiance point value sequences and historical power point value sequences that have different historical change trends and are divided according to the first historical change period and the second historical change period. Each of the second training data subsets is input into a pre-constructed photovoltaic power prediction model for model training to obtain a first photovoltaic power prediction sub-model corresponding to each of the historical change periods. During the model training process, the correspondence between the historical surface irradiance point value sequence and the historical power point value sequence in each of the second training data subsets corresponding to each of the historical change periods is learned based on each of the first photovoltaic power prediction sub-models. 3.The method of claim 2, wherein, The method for constructing the photovoltaic power prediction model also includes: Obtain the historical temperature point value sequence and / or historical wind speed point value sequence corresponding to each of the second training data subsets; Add the historical temperature point value sequence and / or historical wind speed point value sequence to the corresponding second training data subset to obtain the third training data subset; Each of the third training data subsets is input into a pre-built photovoltaic power prediction model for model training to obtain a second photovoltaic power prediction sub-model corresponding to each of the historical change periods. During the model training process, the correspondence between the historical temperature point value sequence and / or historical wind speed point value sequence, the historical surface irradiance point value sequence and the historical power point value sequence in the third training data subset corresponding to each of the historical change periods is learned based on each of the second photovoltaic power prediction sub-models.

4. A photovoltaic power prediction method, characterized by, Applied to target photovoltaic power plants equipped with flat single-axis single-sided photovoltaic modules; the photovoltaic power prediction method includes: Obtain the predicted irradiance data of the target photovoltaic power station; Based on the predicted irradiance data, the predicted surface irradiance data of each of the flat single-axis single-sided photovoltaic modules is calculated; The predicted surface irradiance data is divided according to a preset time interval to obtain a sequence of predicted surface irradiance point values; The predicted surface irradiance point value sequence is input into a pre-trained photovoltaic power prediction model for power prediction to obtain the power prediction result. The pre-trained photovoltaic power prediction model is trained by constructing a first training dataset using historical surface irradiance point value sequence as input features and historical power point value sequence as target values. The historical surface irradiance point value sequence and the historical power point value sequence are obtained by acquiring historical measured irradiance data and historical power data of the target photovoltaic power station within a preset time period, calculating the historical surface irradiance data of each of the flat single-axis single-sided photovoltaic modules based on the historical measured irradiance data, and dividing them according to a preset time interval. Also includes: The historical differential feature sequence is constructed by acquiring historical surface irradiance data and historical power data of the target photovoltaic power station at various times on a daily time scale, including: Calculate the historical trend of the historical surface irradiance point value sequence relative to the historical power point value sequence at each time point on a daily time scale; The historical trend is divided into a first sub-historical trend in which the historical surface irradiance point value sequence is positively correlated with the historical power point value sequence, and a second sub-historical trend in which the historical surface irradiance point value sequence is not positively correlated with the historical power point value sequence. A time-series-related first historical difference feature sequence and a second historical difference feature sequence are constructed based on the first sub-historical change trend and the second sub-historical change trend, respectively. The historical difference feature sequence corresponding to the historical surface irradiance data with a similarity greater than a preset degree to the predicted surface irradiance point value sequence is selected as the predicted difference feature sequence and input into the pre-trained photovoltaic power prediction model to perform power prediction, thereby obtaining the photovoltaic power prediction result.

5. The photovoltaic power prediction method as described in claim 4, characterized in that, The photovoltaic power prediction method also includes: Obtain the predicted change time periods corresponding to the predicted surface irradiance point value sequences on a daily time scale; Based on the predicted change period, the predicted surface irradiance point value corresponding to the predicted change period is input into the first photovoltaic power prediction sub-model corresponding to the historical change period that matches the predicted change period for power prediction, thereby obtaining the photovoltaic power prediction result; wherein, the first photovoltaic power prediction sub-model is trained by a second training data subset consisting of a historical surface irradiance point value sequence and a historical power point value sequence according to the historical change trend corresponding to the historical change period, and the correspondence between the historical surface irradiance point value sequence and the historical power point value sequence in the second training data subset corresponding to each historical change period is learned during the training process.

6. The photovoltaic power prediction method of claim 5, wherein, The photovoltaic power prediction method also includes: Obtain the sequence of predicted temperature points and / or the sequence of predicted wind speed points corresponding to the predicted surface irradiance points for each of the predicted change periods. The predicted surface irradiance point values, the predicted temperature point value sequence, and / or the predicted wind speed point value sequence are respectively input into the pre-trained second photovoltaic power prediction sub-model. The second photovoltaic power prediction sub-model is a third training data subset composed of the predicted temperature point value sequence and / or the predicted wind data point value sequence, the predicted surface irradiance point value sequence, and the predicted surface irradiance point value sequence according to the predicted change trend under the predicted change period. The correspondence between the predicted temperature point value sequence and / or the predicted wind data point value sequence, the predicted surface irradiance point value sequence, and the predicted power point value sequence is learned during the training process.

Images