A method and computing device for predicting long-term power in photovoltaic system

Abstract

The present invention provides a method and computing device for medium- and long-term power prediction of a photovoltaic system, which are applied to the field of photovoltaic prediction technology. The method comprises selecting multiple meteorological grid points around a target photovoltaic power station according to the distance between the target photovoltaic power station and the surrounding meteorological grid points; obtaining an actual power data set of the target photovoltaic power station; preprocessing the actual power data set; obtaining meteorological prediction data of multiple meteorological grid points to form a time series data set; constructing additional meteorological features of the target photovoltaic power station and adding them to the time series data set; using the actual power data set as output and the time series data set as input to train a base learner; using the prediction results of the base learner on the validation set as secondary training data to train a meta learner; and using the trained base learner and meta learner to perform medium- and long-term power prediction of the photovoltaic system. The method of the present invention provides more reliable prediction results and realizes efficient and accurate medium- and long-term power prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic prediction, and in particular to a method and a computing device for predicting long-term power in a photovoltaic system. Background Art

[0002] Traditional power generation forecasting methods often rely on statistical models and historical data, which are difficult to handle complex nonlinear relationships. In existing technologies, the generalization ability of new energy power generation forecasting models is limited. When faced with new data or environmental changes, the models are often difficult to adapt, resulting in reduced forecasting accuracy. This limits the effectiveness and universality of forecasting models in practical applications.

[0003] Photovoltaic power generation prediction technology originated from early weather forecasting and power dispatching systems, and is mainly used for grid load balancing and energy management. However, with the increasing share of photovoltaic power generation in the global energy structure and the large-scale and distributed development of photovoltaic power generation systems, especially the widespread application in microgrids and smart grids and the rapid development of power trading, energy storage and other markets, in addition to the ultra-short-term and short-term prediction results of photovoltaic systems, future long-term power generation is also used as important decision-making information in market transactions.

[0004] Therefore, the accuracy and efficiency of photovoltaic power generation forecasts become particularly important, and machine learning technology needs to be used to improve the accuracy and stability of medium- and long-term forecasts. Summary of the invention

[0005] The present invention aims to provide a method and computing device for medium- and long-term power prediction of photovoltaic systems, which can effectively utilize multi-source data of photovoltaic systems, provide efficient and accurate medium- and long-term power prediction in complex and changeable environments, and provide strong technical support for the efficient utilization of photovoltaic power generation and the stable operation of power grids.

[0006] According to one aspect of the present invention, there is provided a method for long-term power prediction of a photovoltaic system, comprising: selecting a plurality of meteorological grid points around a target photovoltaic power station according to the distance between the target photovoltaic power station and the surrounding meteorological grid points;

[0007] Acquire an actual power data set of the target photovoltaic power station;

[0008] Preprocessing the actual power data set;

[0009] Acquire meteorological forecast data of the plurality of meteorological grid points to form a time series data set;

[0010] constructing additional meteorological features of the target photovoltaic power station and adding them to the time series data set;

[0011] Taking the actual power data set as output and the time series data set as input, training a base learner;

[0012] Using the prediction results of the base learner on the validation set as secondary training data to train the meta learner;

[0013] The trained base learner and the meta learner are used to predict the long-term power of a photovoltaic system.

[0014] The distance between the target PV power station and the surrounding meteorological grid points is calculated according to the following rules:

[0015] In the formula, is the latitude difference between two geographic coordinates, is the longitude difference between two geographic coordinate points, is the intermediate value of the calculation process, Represents the spherical distance in arcs between two geographic coordinates. is the distance between two geographic coordinates, 6371 is the radius of the earth, lon is the longitude, and lat is the latitude; 6371 is the radius of the earth.

[0016] According to some embodiments, obtaining the meteorological forecast data of the plurality of meteorological grid points to form a time series data set includes:

[0017] The first time point is taken from the daily weather forecast data of the surrounding grid points, and n values ​​of the date of the point are taken to form the data set of the first day. The data sets of each day are combined and sorted by time to form a sequence data set.

[0018] According to some embodiments, preprocessing the actual power data set includes:

[0019] The outliers of the actual power data are eliminated by the local outlier factor method, and the actual power data is corrected twice based on the actual irradiance data:

[0020]

[0021] In the formula, is the maximum actual power of the site after removing outliers using the local outlier factor method, is the actual maximum irradiance, is the ratio of maximum power to maximum irradiance, and The actual power and

[0022] The actual irradiance, and are the power and irradiance normalized by the maximum actual power and maximum actual irradiance, ranging from [0,1], is the relative measure of scaled power and irradiance at each moment, when When it is 0, is 0, The actual power data is corrected.

[0023] According to some embodiments, constructing the additional meteorological characteristics of the target photovoltaic power station includes calculating the difference between the oblique solar radiation and the irradiance of each grid point:

[0024] Calculate the theoretical irradiance on the horizontal surface using the formula ;

[0025]

[0026] In the formula, Φ is the site latitude, ω is the solar hour angle, G 0 is the solar constant, take 1368W / m2, k is the correction factor for the distance between the Sun and the Earth, N For accumulated days, δ is the solar declination, is the true solar time, in minutes. is the standard meridian longitude of the local time zone, is the local longitude;

[0027] According to the calculated theoretical irradiance of the horizontal plane, according to the inclination angle of the photovoltaic panels at the site Calculate the solar radiation of photovoltaic panels on the slope, The formula is:

[0028]

[0029] In the formula, is the solar azimuth, is the slope azimuth, Sun altitude angle.

[0030] According to some embodiments, additional meteorological characteristics of the target photovoltaic power plant are constructed, including saturated water vapor pressure characteristics, wherein:

[0031] The characteristic formula for constructing saturated water vapor pressure is:

[0032]

[0033] In the formula, e s0 =6.1078, is the saturated water vapor pressure at 0℃, t represents the temperature in℃;

[0034] e represents water vapor pressure, RH represents relative humidity, represents the saturated vapor pressure;

[0035] When t>-15℃, a=17.269, b=35.86; when t<-40℃, a=21.8746, b=7.66.

[0036] According to some embodiments, constructing additional meteorological characteristics of the target photovoltaic power station includes constructing daily temperature difference and humidity difference characteristics, wherein:

[0037] The training set data is grouped by day, and the lowest temperature and relative humidity values ​​of each day are selected. The sequence composed of the other temperature and relative humidity values ​​within the day minus the lowest temperature and relative humidity values ​​is selected as a feature and entered into the data set.

[0038] According to some embodiments, constructing the additional meteorological characteristics of the target photovoltaic power station includes constructing the photovoltaic panel temperature mutual characteristics according to the following formula:

[0039]

[0040] In the formula, is the operating temperature of the constructed photovoltaic cell, which refers to the temperature at an irradiance of 800 W·m -2 , temperature 20℃, wind speed 1 m·s -1 The temperature of the photovoltaic cell under the external environmental conditions;

[0041] It indicates the increase in panel temperature due to the light-heat conversion caused by solar radiation;

[0042] t represents the temperature in °C, represents the irradiance, Represents wind speed, unit is m / s.

[0043] According to some embodiments, the training process of the base learner uses Bayesian parameter optimization, and defines the root mean square accuracy as an evaluation function in the optimization process.

[0044] According to some embodiments, the root mean square accuracy is defined as:

[0045]

[0046] In the formula, Cap represents the maximum installed capacity of the station, P Mk represents the actual power value of the kth point, P Pk Represents the predicted value of the kth point.

[0047] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein the computer program implements any of the above methods when executed by a processor.

[0048] According to another aspect of the present invention, there is provided a computing device, comprising:

[0049] Processor; and

[0050] A memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform any of the aforementioned methods.

[0051] According to an example embodiment of the present invention, by integrating the data of multiple meteorological grid points, the meteorological conditions of the area where the photovoltaic power station is located can be more comprehensively reflected. The data of the meteorological grid points closer to the target photovoltaic power station can better represent the actual meteorological conditions of the photovoltaic power station, reducing the errors caused by geographical location differences and thus improving the accuracy of power prediction; by preprocessing the actual power data set, not only the data quality can be improved, but also additional meteorological features can be added to construct a richer feature set, thereby improving the prediction ability of the model; by constructing the meteorological prediction data into a time series data set, the dynamic characteristics of the time series data can be better captured, thereby improving the performance of time series prediction; by selecting a base learner and a meta-learner for long-term power prediction of the photovoltaic system, the advantages of different models can be utilized to improve the robustness and generalization ability of the overall model.

[0052] The system for predicting medium- and long-term power generation of renewable energy based on machine learning of the present invention can ensure that the selected meteorological grid point data can more accurately reflect the actual situation of the photovoltaic power station, thereby improving the performance of the power prediction model. This can not only improve the accuracy of the prediction, but also enhance the robustness of the model, and better cope with the power prediction challenges under various weather conditions.

[0053] It is to be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for describing the embodiments are briefly introduced below.

[0055] Figure 1 A flow chart showing a method for medium- and long-term power prediction in a photovoltaic system according to an example embodiment is shown.

[0056] Figure 2 A flow chart of a method for preprocessing multi-source data for medium- and long-term power prediction of a photovoltaic system according to an example embodiment is shown.

[0057] Figure 3 A flow chart showing a method for constructing additional meteorological features of a target photovoltaic power plant in medium- and long-term power prediction of a photovoltaic system according to an example embodiment.

[0058] Figure 4A schematic diagram showing construction of a grid point forecast weather data set according to an example embodiment.

[0059] Figure 5 A block diagram of a computing device according to an example embodiment of the invention is shown. DETAILED DESCRIPTION

[0060] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that the present invention will be comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0061] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, those skilled in the art will appreciate that the technical solution of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0062] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0063] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0064] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another component. Therefore, the first component discussed below can be referred to as the second component without departing from the teachings of the present inventive concept. As used herein, the term "and / or" includes any one of the associated listed items and all combinations of one or more.

[0065] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0066] Those skilled in the art will appreciate that the drawings are merely schematic diagrams of example embodiments, and the modules or processes in the drawings are not necessarily necessary for implementing the present invention, and therefore cannot be used to limit the protection scope of the present invention.

[0067] Photovoltaic power generation forecasting usually uses meteorological data and historical power generation data for modeling, and the forecasting methods include physical models, statistical models, and machine learning models. Physical models are based on photovoltaic equipment parameters and environmental parameters such as solar radiation and temperature. They are mainly used for short-term forecasting and have high requirements for data quality and real-time performance. Statistical models such as time series analysis and autoregressive moving average model (ARIMA) are suitable for ultra-short-term and short-term forecasting, but have limited processing capabilities for nonlinear and complex data, and have poor forecasting performance and medium- and long-term forecasting accuracy. Machine learning models such as decision trees, support vector machines, and random forests can handle complex nonlinear relationships by training on a large amount of historical data, and have made significant progress in photovoltaic power generation forecasting in recent years. However, these models still face challenges in computing efficiency and forecasting accuracy when processing real-time data and large-scale distributed power generation systems.

[0068] Most of the existing photovoltaic power generation prediction technologies focus on short-term prediction. However, for medium- and long-term prediction, the prediction accuracy and methods are still subject to much improvement due to the limitations of the accuracy of meteorological data and the generalization ability of the model. At the same time, a single model has limited prediction effect when facing complex and changeable meteorological conditions. Although some studies have adopted integrated models such as Bagging and Boosting, these methods still have certain limitations in processing heterogeneous data and multi-source data fusion.

[0069] Artificial intelligence and big data mining technologies have not yet been widely used in the field of photovoltaic power generation prediction. In particular, how to build an efficient and intelligent photovoltaic power generation prediction system under various meteorological conditions and complex power grid environments lacks specific technical solutions and mature application cases.

[0070] In summary, the existing technologies for forecasting renewable energy power generation have problems such as insufficient consideration of the instability of renewable energy power generation, insufficient handling of the uncertainty of weather data, lack of effective methods for medium- and long-term forecasts, limited model generalization capabilities, and insufficient precision in data processing and feature selection.

[0071] In view of the challenges faced by renewable energy power generation prediction technology in the prior art, the present invention proposes a method for medium- and long-term power prediction of photovoltaic systems, which integrates multi-source data with a deep learning stacking integrated model, fully utilizes the advantages of multi-source data, and improves the accuracy and reliability of prediction.

[0072] Before describing the embodiments of the present invention, some terms or concepts involved in the embodiments of the present invention are explained.

[0073] SMOTE (Synthetic Minority Over-sampling Technique) is a technology used to deal with data imbalance problems in machine learning classification tasks. It synthesizes new minority class samples to balance the number of samples in each category in the training data set, thereby improving the performance of the classifier.

[0074] SVM (Support Vector Machine, SVM), support vector machine is a supervised learning model, mainly used for classification and regression tasks. In the prediction of renewable energy power generation, SVM can be used to build a prediction model, especially when the data set is not very large or has less noise.

[0075] Stacking is an ensemble learning method that improves predictive performance by combining multiple different machine learning models.

[0076] SWVP (Saturated Water Vapor Pressure) is an important parameter that describes the relationship between water vapor content and temperature in the atmosphere. It reflects the maximum amount of water vapor that air can hold at a certain temperature.

[0077] Exemplary embodiments of the present invention are described below with reference to the accompanying drawings.

[0078] Figure 1 A flow chart showing a method for medium- and long-term power prediction in a photovoltaic system according to an example embodiment is shown.

[0079] According to some example embodiments, since the actual meteorological conditions of a photovoltaic power station may deviate from the data of a single meteorological station, the use of data from multiple meteorological grid points can better reflect the meteorological conditions of the area where the photovoltaic power station is located. The method of this example embodiment can not only improve the prediction accuracy, but also have strong adaptability in changing meteorological conditions and complex power grid environments, providing technical support for the efficient use of photovoltaic power generation and the stable operation of the power grid.

[0080] See also Figure 1In S101, a plurality of meteorological grid points around the target photovoltaic power station are selected according to the distance between the target photovoltaic power station and the surrounding meteorological grid points.

[0081] The geographical location information of the target photovoltaic power station is obtained, and multiple meteorological grid points around the target photovoltaic power station are selected. The distance between the target photovoltaic power station and the surrounding meteorological grid points is calculated according to the following rules:

[0082]

[0083] in, is the latitude difference between two geographic coordinates (in radians);

[0084] is the longitude difference between two geographic coordinate points (in radians);

[0085] is the intermediate value of the calculation process;

[0086] Represents the spherical distance in arcs between two geographic coordinates;

[0087] is the distance between two geographic coordinates, and 6371 is the radius of the earth, that is, the distance from the center of the earth to the North Pole or the South Pole.

[0088] According to the example embodiment, the distance between each grid and the target photovoltaic power plant coordinates is calculated by traversing through a formula. The longitude and latitude of the target photovoltaic power plant A are known to be denoted as lon and lat, respectively. In the multi-source prediction meteorological grid data set, the four grid points closest to the target photovoltaic power plant A are selected. The grid points are determined not by directly selecting the longitude and latitude of the grid points, but by calculating the distance between each site and the longitude and latitude of the target photovoltaic power plant grid through a formula rule, and the four points with the smallest absolute value of the distance are selected according to the calculation results.

[0089] In S103, an actual power data set of the target photovoltaic power station is obtained, and the actual power data set is preprocessed.

[0090] According to some embodiments, the purpose of preprocessing is to clean data, fill missing values, standardize and / or normalize data, and construct useful features to improve the performance of the model.

[0091] This embodiment obtains the historical actual power data of the target photovoltaic power station, checks whether there are abnormal values ​​in the data, such as extreme values ​​or unreasonable values, through feature construction and transformation, and deletes or replaces these abnormal values ​​according to actual conditions.

[0092] In S105, meteorological forecast data of the plurality of meteorological grid points are obtained to form a time series data set, and additional meteorological features of the target photovoltaic power station are constructed and added to the time series data set.

[0093] In order to improve the accuracy of PV power prediction, in addition to using the actual power data of the target PV power station, the influence of meteorological factors needs to be considered. Therefore, the meteorological forecast data of multiple meteorological grid points are obtained and constructed into a time series dataset, and then the additional meteorological features of the target PV power station are constructed and added to the time series dataset.

[0094] According to the example embodiment, meteorological data is extracted from the four grid points selected in step S101, the first time point is taken from the forecast data of the meteorological data of each day, and the 96 values ​​of the date of the point are taken to form the data set t1 of the first day, the 96 points of the date of t1 plus 1 day are the t2 data set, and the t10 data set is taken in sequence. The t1 data of each day are combined and sorted by time to form the T1 sequence data set, and the sequence data sets of T2~T10 are obtained in the same way.

[0095] According to some embodiments, it is assumed that each grid point can obtain weather forecast data for the next 10 days every day (the process of constructing the T1 sequence data set is shown in Figure 4 ), take the prediction results of the first day in each day's prediction file date, concatenate them using time as the index, and take them sequentially up to the 10th day.

[0096] For example, each grid point provides weather forecasts for the next 10 days. The weather forecast for each future day starts at 00:00 and ends at 23:45. The time resolution is 15 minutes, and each day contains 96 values.

[0097] The forecast file generated on January 1, 2020 for the grid point in the figure contains the weather forecast results for 10 days from January 2, 2020 to January 11, 2020. For the forecast file generated on January 1, 2020, the 96 values ​​(00:00-23:45) on January 2, 2020 are defined as t1, …, and the 96 values ​​(00:00-23:45) on January 11, 2020 are defined as t10. Similarly, for the forecast file on January 2, 2020, the 96 values ​​(00:00-23:45) on January 3, 2020 are t1, …, and the 96 values ​​(00:00-23:45) on January 12, 2020 are t10.

[0098] Take out t1 in each prediction file, splice them together, and sort them by time to form a T1 sequence data set, and then take them to T10 in sequence.

[0099] After forming the data set of each grid point T1~T10, The sequence data set is added to the data set in the form of a new column based on the order of site distances, constructing a sequence data set containing the predicted meteorology of all grid points. The sequence data set based on site distances can make the meteorology with stronger correlation with the site enter the model first.

[0100] For each The sequence data set and the data set after power cleaning are concatenated with time as the index, and 80% of each data set is divided into the training set and 20% into the test set.

[0101] The cleaned training set is equally divided into 5 categories according to the percentage of the actual power generation (Power). These 5 categories are 0 < Power <= 0.2 * max(Power), 0.2 < Power <= 0.4 * max(Power), 0.4 < Power <= 0.6 * max(Power), 0.6 < Power <= 0.8 * max(Power), and 0.8 < Power <= 1.0 * max(Power).

[0102] The oversampling method (SMOTE) is used to balance the data volume of each category for data augmentation. The SMOTE method generates new minority-class samples by interpolating between minority-class samples, thus increasing the number of minority-class samples. For example, two minority-class samples and generate a new sample :

[0103]

[0104] where is a random number.

[0105] Imbalanced data is a common challenge in machine learning. Model algorithms often have a bias towards the majority class and ignore the minority class. For a photovoltaic power station, the model may be more likely to identify weather conditions with concentrated distributions and ignore the cases of partial cloudy and partial sunny days. Through the SMOTE method, the model bias caused by the scarcity of minority-class samples can be effectively reduced, enabling the model to better identify and classify minority-class samples, and improving the prediction performance and generalization ability of the model.

[0106] In S107, the actual power data set is used as the output, the time series data set is used as the input to train the base learner, and the prediction result of the base learner on the validation set is used as the secondary training data to train the meta-learner. The trained base learner and meta-learner are used for medium- and long-term power prediction of the photovoltaic system.

[0107] Through the method of this example embodiment, multiple meteorological grid points can be selected according to the distance between the target photovoltaic power station and the surrounding meteorological grid points, and the data of these grid points can be merged to obtain a data set that is closer to the actual meteorological conditions of the photovoltaic power station, which is more helpful to improve the accuracy and reliability of photovoltaic power prediction. This not only helps to improve the prediction performance of the model, but also ensures that the model can better adapt to the actual photovoltaic power station operating environment.

[0108] The new energy medium- and long-term power generation forecasting system based on machine learning in this example embodiment can provide an efficient, accurate and easy-to-use solution for new energy power generation forecasting through comprehensive data collection, high-quality data preprocessing, efficient forecasting model construction and intuitive user interaction. The system of the present invention can not only improve the forecasting accuracy, but also help enterprises and governments make better decisions, thereby promoting the sustainable development of the new energy industry.

[0109] Figure 2 A flow chart showing a method for preprocessing multi-source data for medium- and long-term power prediction in a photovoltaic system according to an example embodiment.

[0110] In photovoltaic power stations, there is a strict correspondence between actual irradiance and actual power. However, due to situations such as power restrictions and equipment failures that may occur in actual operation, various outliers may exist in the actual power data. For obvious outliers, outlier detection methods can be used to eliminate them; however, for monitoring anomalies and small fluctuations caused by local equipment failures, traditional detection methods are often difficult to accurately identify. Through irradiance-based detection and reconstruction methods, these outliers can be identified more accurately, while ensuring the integrity of site power data and providing sufficient data support for model training.

[0111] In S201, outliers in actual power data are eliminated by using a local outlier factor method.

[0112] According to an example embodiment, the outlier value of the actual power data is removed by using the Local Outlier Factor (LOF) method, and the outlier factor calculation formula is:

[0113]

[0114] in,

[0115] Representative Points p of k The number of nearest neighbors;

[0116] is the local reachability density of point o;

[0117] is the local reachability density of point p.

[0118] Defined as:

[0119]

[0120] reach d ist k (p, o) The formula for reachable distance is:

[0121]

[0122] in,

[0123] Yes To its Distance to nearest neighbor;

[0124] Point p and point The distance between.

[0125] If the proportion of abnormal data in the power data is set to 10%, the 10% of data with the largest outlier factor will be identified as outliers and deleted. The setting value of the outlier factor is adjustable.

[0126] This embodiment uses the LOF algorithm to identify and remove outliers in the actual power data, thereby ensuring the cleanliness and quality of the data set. In addition to the LOF algorithm, other outlier detection methods such as isolation forest and distance-based anomaly detection can also be used to process abnormal data.

[0127] In S203, the power data is corrected for the second time.

[0128] According to an exemplary embodiment, the power data is corrected again based on the irradiance data collected at the site, and the actual power data and the actual irradiance data are scaled to a specified range through a normalization formula. The secondary correction formula is:

[0129]

[0130] in, is the maximum actual power of the site after removing outliers using the local outlier factor method, is the actual maximum irradiance, is the ratio of maximum power to maximum irradiance, and For the actual power and actual irradiance at each moment, and are the power and irradiance normalized by the maximum actual power and maximum actual irradiance, ranging from [0,1], is the relative measure of scaled power and irradiance at each moment, when 0 o'clock (night), is 0, The actual power data is corrected.

[0131] According to an example embodiment, taking the data of a certain site within one day as an example, the irradiance and real-time power in the original data are normalized by the method, which can accurately identify outliers exceeding the threshold. The revised data set after removing the outliers can better provide data support for the model while ensuring data integrity.

[0132] For data from photovoltaic stations, the model may be more likely to identify weather conditions with concentrated distribution, while ignoring cloudy and sunny conditions. The SMOTE method or other data balancing methods such as low-frequency data repetition can effectively reduce the model bias caused by the scarcity of minority class samples, enabling the model to better identify and classify minority class samples, thereby improving the model's prediction performance and generalization ability.

[0133] In the data processing process of the present invention, the actual power is corrected twice with the actual irradiance, and the prediction performance of the model is improved by effectively preprocessing the actual power data of the target photovoltaic power station. Data preprocessing helps to ensure the quality of the data and lay a solid foundation for subsequent model training and prediction.

[0134] Figure 3 A flow chart showing a method for constructing additional meteorological features of a target photovoltaic power plant in medium- and long-term power prediction of a photovoltaic system according to an example embodiment.

[0135] At S301, a slope theoretical irradiance photovoltaic characteristic is constructed.

[0136] According to some embodiments, the photovoltaic panel is installed on an inclined surface, so it is necessary to construct a slope theoretical irradiance photovoltaic feature. Before calculating the solar radiation received on the inclined surface, the horizontal plane theoretical irradiance is first calculated. The horizontal plane theoretical irradiance (Global Horizontal Irradiance, GHI) is the total solar radiation received per unit area on a horizontal plane under ideal conditions, including direct radiation perpendicular to the direction of the sun's rays and scattered radiation from all directions of the sky.

[0137] According to an example embodiment, the theoretical irradiance formula for a horizontal plane is:

[0138] ;

[0139] in,

[0140] Φ is the site latitude, ω is the solar hour angle;

[0141] G 0 is the solar constant, take 1368 (W / m2);

[0142] k is the correction factor for the distance between the sun and the earth;

[0143] N For accumulated days, δ is the solar declination;

[0144] is true solar time (in minutes), Zone time (unit: minutes), is the time difference (in minutes), and 4 is the rotation of the earth (4 minutes / degree).

[0145] is the standard meridian (central meridian) longitude of the local time zone, It is the local longitude. The local longitude is positive when it is east of the central meridian of the time zone and negative when it is west of it.

[0146] After calculating the theoretical irradiance of the horizontal plane using the theoretical irradiance formula, the inclination angle of the photovoltaic panel at the site is Calculate the solar radiation on the slope of photovoltaic panels.

[0147] like Unknown, can be assigned by the assignment formula:

[0148]

[0149] Where lat is the latitude of the power station.

[0150] get The calculated solar radiation to the slope :

[0151]

[0152] in,

[0153] is the solar azimuth, is the slope azimuth, =0, Sun altitude angle;

[0154] It can be calculated by the following formula:

[0155]

[0156] In this exemplary embodiment, the oblique solar radiation The difference between the irradiance at each grid point is added to the dataset as a new feature, which can effectively help the model better capture the difference between the actual irradiance and the theoretical value, thereby improving the prediction accuracy.

[0157] In S303, a saturated water vapor pressure characteristic is constructed.

[0158] The saturated water vapor pressure feature is important in the photovoltaic prediction model, mainly because it is closely related to atmospheric conditions such as air humidity and scattering, which directly affect the input irradiance of photovoltaic panels and their power generation efficiency. In high humidity environments, water vapor condensation may occur on the surface of photovoltaic panels, especially at night or in the early morning, which will cause dew or mist on the surface of photovoltaic panels, thereby affecting the power generation in the morning. This is also an important factor why the actual power curve and irradiance curve have different shapes during the actual power rise and fall time periods in the morning and evening.

[0159] At the same time, when the air reaches saturation (that is, the actual water vapor pressure reaches the saturated water vapor pressure), water vapor will begin to condense to form clouds or fog. The formation and thickness of clouds directly affect the amount of solar radiation received by photovoltaic panels. In the photovoltaic prediction model, the possibility and type of cloud formation can be inferred from the saturated water vapor pressure characteristics, thereby improving the prediction of the radiation model.

[0160] According to some embodiments, when establishing a performance model of a photovoltaic power station, it is usually necessary to consider the influence of multiple meteorological parameters. By taking the saturated water vapor pressure as one of the input variables, the influence of meteorological conditions on the performance of the photovoltaic power station can be more comprehensively reflected. For example, adding SWVP data to the model can help more accurately predict the photovoltaic output power under different meteorological conditions.

[0161] According to an exemplary embodiment, the characteristic formula for saturated water vapor pressure is constructed as:

[0162]

[0163] Among them, e s0 =6.1078, is the saturated water vapor pressure at 0℃, t represents temperature (℃);

[0164] e represents water vapor pressure, RH represents relative humidity, represents the saturated vapor pressure;

[0165] When t>-15℃, a=17.269, b=35.86; when t<-40℃, a=21.8746, b=7.66.

[0166] In S305, daily temperature difference and humidity difference features are constructed.

[0167] The daily temperature difference and humidity difference features reflect the changes in temperature and humidity every day. The training set data is grouped by day, and the lowest values ​​of temperature and relative humidity min(Temp) and min(Hum) are selected every day. The other temperature and relative humidity values ​​in the day are subtracted from the lowest values ​​of temperature and relative humidity min(Temp) and min(Hum), and the resulting sequence is selected as a feature and entered into the data set.

[0168] By using the method in this example, the daily temperature and relative humidity values ​​are subtracted from the corresponding minimum values ​​to construct a new feature sequence, which is then added to the data set. The processed data will be more conducive to model learning and prediction.

[0169] In S307 , photovoltaic panel temperature mutual characteristics are constructed.

[0170] According to some embodiments, in photovoltaic power generation prediction, photovoltaic panel temperature is an important part of the photovoltaic prediction model. The photovoltaic panel temperature directly affects the power generation efficiency and output power of the photovoltaic system. Due to the temperature effect, the power conversion efficiency of the photovoltaic panel decreases as the temperature increases. Specifically, the output power of the photovoltaic panel usually decreases by about 0.3% - 0.5% for every 1°C increase in temperature. This value is called the temperature coefficient.

[0171] PV panel temperature cross-features can form new characteristic variables by combining PV panel temperature with other environmental factors, so as to more comprehensively describe the working environment of PV panels and their impact on power generation efficiency.

[0172] According to an example embodiment, a photovoltaic panel temperature mutual characteristic formula is constructed :

[0173]

[0174] in,

[0175] is the operating temperature of the constructed photovoltaic cell, which refers to the temperature at an irradiance of 800 W·m -2 , temperature 20℃, wind speed 1m·s -1 The temperature of the photovoltaic cell under the external environmental conditions;

[0176] It indicates the increase in panel temperature due to the light-heat conversion caused by solar radiation;

[0177] t represents temperature (℃), represents the irradiance, Represents wind speed (m / s).

[0178] Constructing accurate predicted temperature of photovoltaic panels can not only improve the accuracy of photovoltaic power generation prediction, but also effectively help management and maintenance, and optimize the operating efficiency of photovoltaic systems. For example, under extreme weather conditions (such as high temperature or strong wind), temperature changes can cause greater fluctuations in power generation; in hot weather, if the temperature forecast of photovoltaic panels is too high, some measures can be taken (such as water cooling, increased ventilation, etc.) to protect photovoltaic panels and improve power generation efficiency. These factors are crucial to the promotion of renewable energy and grid stability. By constructing temperature characteristics, the model can better capture the specific impact of temperature changes on photovoltaic panel power generation, which can effectively improve the accuracy of subsequent prediction models.

[0179] The present invention can provide more useful information for the photovoltaic prediction model by constructing the theoretical irradiance characteristics of the slope, the saturated water vapor pressure characteristics, the daily temperature difference and humidity difference characteristics and the photovoltaic panel temperature mutual characteristics, and help the model better understand the impact of environmental changes on photovoltaic power generation, thereby improving the accuracy and reliability of the prediction. The additional meteorological characteristics of the target photovoltaic power station are effectively constructed and added to the time series data set, thereby providing more comprehensive and accurate data support for photovoltaic power prediction.

[0180] In this embodiment, the meteorological data set and the actual power data are model-fitted and predicted, and the medium- and long-term power prediction of the photovoltaic system is performed through the base learner and the meta learner.

[0181] According to an exemplary embodiment, a sequence data set containing the forecast weather for all grid points is prepared. , select the training data set after data processing and feature construction ( ), extract the actual power data Power as the output of the Stacking model ( ), except for the actual power data, the features are taken as model input ( ).

[0182] To train the base learner, firstly, the spatial information and temporal information of the grid point data are The input is sent to a group of base learners consisting of a convolutional neural network (CNN) for extracting spatial features, a long short-term memory network (LSTM) for extracting temporal features, and a CNN-LSTM network that combines the advantages of both. A 5-fold cross-validation is performed on each base learner, and the training data set is divided into 5 mutually exclusive subsets. In each validation, one of the subsets is selected as the validation set, and the remaining 4 subsets are used as training sets. Each base learner is trained on the 4 subsets and predicted on the remaining subset; this is repeated 5 times, with a different validation set selected each time, and eventually each base learner makes predictions on the entire training set. These prediction results are the secondary training data (meta-features), which constitute a new training data set ( ).

[0183] According to the example embodiment, the Bayesian parameter optimization method is used in each training process of the base learner to optimize important parameters such as the number of layers, number of neurons, learning rate, batch size, and number of training rounds of each base learner to ensure the training effect of a single base learner. For the photovoltaic prediction system, the custom root mean square accuracy evaluation function in the optimization process is:

[0184]

[0185] Among them, Cap represents the maximum installed capacity of the site;

[0186] P Mk represents the actual power value of the kth point;

[0187] P Pk Represents the predicted value of the kth point.

[0188] The greater the RMS accuracy, the more suitable the model parameters are for the site.

[0189] In addition, a residual connection mechanism is added during the training process of the base learner to alleviate the gradient vanishing problem and avoid the model degradation problem that may be caused by the increase in the number of layers.

[0190] According to an example embodiment, the core idea of ​​the residual connection is to introduce a direct connection between one or more layers of the network, so that the outputs of these layers include not only the features that have been nonlinearly transformed, but also the unprocessed input features. The purpose of this is to allow the network to learn the residual (i.e., the difference) between the input and output, rather than directly learning a complete mapping.

[0191] According to some embodiments, if a network layer in the model has an input of x , through the nonlinear function F(x) To get the output, in the absence of residual connection, the output of the network layer is F(x) ; When the residual connection is introduced, the output of this layer becomes F(x)+ x The input information entering the next layer of the network includes the input information of the previous layer, which prevents no output due to gradient calculation errors during the training process within the layer. F(x) +x That is, the residual output of the input itself is taken into account.

[0192] The present invention uses residual connections to directly transfer information, which helps the gradient to propagate more effectively in the deep network, effectively solving the gradient vanishing problem in the base model training process. At the same time, due to the introduction of residual connections, the base model can learn effective feature representations more quickly, accelerating the convergence speed in the training process.

[0193] Aggregate the prediction results of all base learners on the validation set to form a secondary training data set ( ), each column of the secondary training data set corresponds to a prediction result of a base learner.

[0194] Perform meta-learner training using the secondary training dataset As the input of the meta-learner (LinearSVM), the measured data of the validation set is used as the output of meta-learning and trained again.

[0195] Final prediction, for the test data First, the feature construction process of steps S201 to S207 is performed to ensure feature consistency. Secondly, the CNN, LSTM and CNN_LSTM base learners are used for prediction to generate secondary test data. The secondary test data is used as the input of the meta-learner. The meta-learner generates the final prediction result by combining the secondary test data input. The final prediction result and the measured data are used for accuracy calculation (Acc_RMSE). For models whose prediction accuracy does not reach the initial threshold, the test set data is added for retraining, and finally the model file is saved.

[0196] In this embodiment, the ensemble learning method is not limited to using a specific type of model, but can also flexibly combine a variety of different machine learning algorithms, including traditional statistical learning methods such as random forests and XGBoost. By combining an ensemble learning model of multiple different types of models for medium- and long-term power forecasting of photovoltaic systems, the advantages of multiple models can be fully utilized, thereby significantly improving the accuracy and stability of the forecast.

[0197] For the mid- and long-term photovoltaic model training and the base model of deep learning, the present invention adopts residual connection to alleviate the gradient vanishing problem and promotes the direct propagation of feature information in the network, thereby improving the network's expressive ability; at the same time, the model parameters are determined by the Bayesian optimization method, and the root mean square accuracy is customized as the evaluation function in the optimization process to ensure that the optimization result is true and effective for the photovoltaic system and improve the accuracy of the base model.

[0198] By integrating multi-source data with a deep learning integrated model, the present invention can not only effectively utilize the multi-source data of the photovoltaic system, but also provide efficient and accurate medium- and long-term power forecasts in complex and changeable environments, thereby providing strong technical support for the efficient utilization of photovoltaic power generation and the stable operation of the power grid.

[0199] During the data processing process, the present invention uses the meteorological data of the grid points near the longitude and latitude of the site as the model data to avoid the systematic error caused by the interpolation of meteorological data to the site and the loss of information on range characteristics such as irradiance and cloud cover caused by the interpolation process, and uses the actual distance as a judgment when selecting the grid points.

[0200] Compared with the traditional single prediction model, the integrated model of the present invention significantly improves the accuracy and stability of prediction by combining the advantages of multiple base learners. Under complex and changeable meteorological conditions, data fusion of forecast meteorological data from multiple surrounding grid points can better adapt to different meteorological scenes and provide more reliable prediction results. At the same time, the application of diversified processing methods and parameter optimization technology of the data processing module further improves the generalization ability and prediction accuracy of the model.

[0201] Figure 5 A block diagram of a computing device according to an example embodiment of the invention is shown.

[0202] like Figure 5 As shown, computing device 30 includes processor 12 and memory 14. Computing device 30 may also include bus 22, network interface 16, and I / O interface 18. Processor 12, memory 14, network interface 16, and I / O interface 18 may communicate with each other via bus 22.

[0203] The processor 12 may include one or more general-purpose CPUs (Central Processing Units, processors), microprocessors, or application-specific integrated circuits, etc., for executing relevant program instructions.

[0204] The memory 14 may include a machine system readable medium in the form of a volatile memory, such as a random access memory (RAM), a read-only memory (ROM) and / or a cache memory. The memory 14 is used to store one or more programs including instructions and data. The processor 12 may read the instructions stored in the memory 14 to execute the above-mentioned method according to the embodiment of the present invention.

[0205] The computing device 30 may also communicate with one or more networks via the network interface 16. The network interface 16 may be a wireless network interface.

[0206] The bus 22 may include an address bus, a data bus, a control bus, etc. The bus 22 provides a path for exchanging information between components.

[0207] It should be noted that, in the specific implementation process, the computing device 30 may also include other components necessary for normal operation. In addition, those skilled in the art may understand that the above device may only include components necessary for implementing the embodiments of this specification, and need not include all components shown in the figure.

[0208] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when the program is executed by a processor. The computer-readable storage medium may include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a micro drive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), a network storage device, a cloud storage device, or any type of medium or device suitable for storing instructions and / or data.

[0209] An embodiment of the present invention also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute part or all of the steps of any method recorded in the above method embodiments.

[0210] Those skilled in the art can clearly understand that the technical solution of the present invention can be implemented with the help of software and / or hardware. "Unit" and "module" in this specification refer to software and / or hardware that can independently complete or cooperate with other components to complete specific functions, where the hardware can be, for example, a field programmable gate array, an integrated circuit, etc.

[0211] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0212] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0213] In the several embodiments provided by the present invention, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0214] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0215] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0216] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present invention, 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, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention.

[0217] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0218] The exemplary embodiments of the present invention are specifically shown and described above. It should be understood that the present invention is not limited to the detailed structures, configurations or implementations described herein; on the contrary, the present invention is intended to cover various modifications and equivalent configurations included in the spirit and scope of the appended clauses.

[0219] Those skilled in the art can clearly understand that the technical solution of the present invention can be implemented with the help of software and / or hardware. "Unit" and "module" in this specification refer to software and / or hardware that can independently complete or cooperate with other components to complete specific functions, where the hardware can be, for example, a field programmable gate array, an integrated circuit, etc.

[0220] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0221] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0222] In the several embodiments provided by the present invention, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0223] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0224] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0225] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present invention, 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, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention.

[0226] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0227] The exemplary embodiments of the present invention are specifically shown and described above. It should be understood that the present invention is not limited to the detailed structures, configurations or implementations described herein; on the contrary, the present invention is intended to cover various modifications and equivalent configurations included in the spirit and scope of the appended clauses.

Claims

1. A method for long-term power prediction of a photovoltaic system, characterized in that: include: Selecting a plurality of meteorological grid points around the target photovoltaic power station according to the distance between the target photovoltaic power station and the surrounding meteorological grid points; Acquire an actual power data set of the target photovoltaic power station; Preprocessing the actual power data set; Acquire meteorological forecast data of the plurality of meteorological grid points to form a time series data set; constructing additional meteorological features of the target photovoltaic power station and adding them to the time series data set; Taking the actual power data set as output and the time series data set as input, training a base learner; Using the prediction results of the base learner on the validation set as secondary training data to train the meta learner; The trained base learner and the meta learner are used to predict the long-term power of a photovoltaic system. The distance between the target PV power station and the surrounding meteorological grid points is calculated according to the following rules: In the formula, is the latitude difference between two geographic coordinates, is the longitude difference between two geographic coordinate points, is the intermediate value of the calculation process, Represents the spherical distance in arcs between two geographic coordinates. is the distance between two geographic coordinates, 6371 is the radius of the earth, lon is the longitude, and lat is the latitude; 6371 is the radius of the earth; in, Constructing additional meteorological features of the target photovoltaic power station includes constructing daily temperature difference and humidity difference features: grouping the training set data by day, selecting the lowest temperature and relative humidity values ​​of each day, and subtracting the lowest temperature and relative humidity values ​​from other temperature and relative humidity values ​​within the day. The constructed sequence is selected as a feature and entered into the data set.

2. The method according to claim 1, characterized in that Acquiring meteorological forecast data of the plurality of meteorological grid points to form a time series data set includes: The first time point is taken from the daily weather forecast data of the surrounding weather grid points, and n values ​​of the date of the point are taken to form the data set for the first day. The data sets for each day are combined and sorted by time to form a sequence data set.

3. The method according to claim 1, characterized in that Preprocessing the actual power data set includes: The outliers of the actual power data are eliminated by the local outlier factor method, and the actual power data is corrected twice based on the actual irradiance data: In the formula, is the maximum actual power of the site after removing outliers using the local outlier factor method, is the actual maximum irradiance, is the ratio of maximum power to maximum irradiance, and The actual power and The actual irradiance, p i ’ and r i ’ are the power and irradiance normalized by the maximum actual power and maximum actual irradiance, ranging from [0,1], is the relative measure of scaled power and irradiance at each moment, when When it is 0, is 0, The actual power data is corrected.

4. The method according to claim 1, characterized in that: Construct additional meteorological characteristics of the target photovoltaic power station, including calculating the difference between the slope solar radiation and the irradiance of each grid point: Calculate the theoretical irradiance on the horizontal surface using the formula ; In the formula, h is the solar altitude angle, θz is the zenith angle, Φ is the site latitude, ω is the solar hour angle, G 0 is the solar constant, 1368W / m 2 , k is the correction factor for the distance between the Sun and the Earth, N For accumulated days, δ is the solar declination, is the true solar time, in minutes. is the standard meridian longitude of the local time zone, is the local longitude; According to the calculated theoretical irradiance of the horizontal plane, according to the inclination angle of the photovoltaic panels at the site Calculate the solar radiation of photovoltaic panels on the slope, The formula is: In the formula, is the solar azimuth, is the slope azimuth, Sun altitude angle.

5. The method according to claim 1, characterized in that Construct additional meteorological characteristics of the target photovoltaic power station, including saturated water vapor pressure characteristics, wherein: The characteristic formula for constructing saturated water vapor pressure is: In the formula, e s0 =6.1078, is the saturated water vapor pressure at 0℃, t represents the temperature in℃; e represents water vapor pressure, RH represents relative humidity, represents the saturated vapor pressure; When t >-15℃, a=17.269, b=35.86; when t <-40℃, a=21.8746, b=7.

66.

6. The method according to claim 1, characterized in that Constructing the additional meteorological characteristics of the target photovoltaic power station includes constructing the photovoltaic panel temperature mutual characteristics according to the following formula: In the formula, is the operating temperature of the constructed photovoltaic cell, which refers to the temperature at an irradiance of 800 W·m -2 , temperature 20℃, wind speed 1m·s -1 The temperature of the photovoltaic cell under the external environmental conditions; It indicates the increase in panel temperature due to the light-heat conversion caused by solar radiation; t represents the temperature in °C, represents the irradiance, Represents wind speed, unit is m / s.

7. The method according to claim 1, characterized in that The training process of the base learner uses Bayesian parameter optimization, and defines the root mean square accuracy as an evaluation function in the optimization process.

8. The method according to claim 7, characterized in that The RMS accuracy is defined as: In the formula, Cap represents the maximum installed capacity of the station, P Mk represents the actual power value of the kth point, P Pk Represents the predicted value of the kth point.

9. A computing device, characterized in that include: processor; as well as A memory storing a computer program, which, when executed by the processor, enables the processor to perform the method according to any one of claims 1 to 8.

Images