A data processing method, device and related equipment for photovoltaic power prediction model

Abstract

The present application discloses a data processing method, device and related equipment for a photovoltaic power prediction model, which obtains the daily input data required for training the photovoltaic power prediction model, cleans the daily input data through a preset cleaning method, and obtains target training data, wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; elimination of abnormal data points of the inverter includes at least elimination of power-limiting abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shielding abnormal points of the inverter and elimination of discrete noise points of the inverter, and the target training data is used to train the photovoltaic power prediction model.

Description

Technical Field

[0001] The present application relates to the technical field of photovoltaic equipment power generation prediction training data cleaning, and more specifically, to a data processing method, device and related equipment for a photovoltaic power prediction model. Background Art

[0002] With the rapid development of new energy power generation technology, the total installed capacity of photovoltaic power generation equipment has continued to increase, and it has also driven related research in the field of photovoltaic power generation prediction. In order to improve the accuracy of photovoltaic power prediction models in predicting the power generation of photovoltaic equipment, photovoltaic power prediction models are usually established at the inverter level. The number of inverters in photovoltaic power stations ranges from dozens to thousands. In order to reduce the amount of calculation, the time series data of the inverter equipment on the day can be used to train the model.

[0003] Since the average daily power generation time of a photovoltaic power station generally does not exceed 9 hours, calculated at one data point per minute, there are only 540 data points per day for training, and the amount of effective data is even less. When the amount of abnormal data accounts for a large proportion, it is difficult to clean the data only through clustering, isolation forest and other methods. Training data containing dirty data will lead to poor training results of the photovoltaic power prediction model, and reduce the accuracy of predicting the power generation of photovoltaic equipment through the photovoltaic power prediction model.

[0004] Therefore, when the amount of data is small, how to improve the quality of the training data used in the photovoltaic power prediction model is an urgent problem to be solved in this application. Summary of the invention

[0005] In view of this, the present application discloses a data processing method, device and related equipment for a photovoltaic power prediction model, which aims to delete as many abnormal data points as possible in combination with photovoltaic business knowledge when the data volume is small. Even when the abnormal data volume accounts for a large proportion, abnormal data points can be identified, and the dependence on the hyperparameters set by the algorithm is small. Finally, the cleaned target training data is obtained, and the target training data is used to train the photovoltaic power prediction model, thereby improving the quality of the training data and improving the accuracy of the model.

[0006] In order to achieve the above purpose, the disclosed technical solution is as follows:

[0007] In a first aspect, the present application discloses a data processing method for a photovoltaic power prediction model, the method comprising:

[0008] Obtain the daily input data required to train the photovoltaic power prediction model;

[0009] The input data of the day is cleaned by a preset cleaning method to obtain target training data; wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; the elimination of abnormal data points of the inverter includes at least elimination of power limitation abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shielding abnormal points of the inverter and elimination of discrete noise points of the inverter; the target training data is used to train a photovoltaic power prediction model.

[0010] Preferably, the step of obtaining the daily input data required for training the photovoltaic power prediction model includes:

[0011] Obtain the active power measurement point time series data of all inverter equipment in the photovoltaic station on the same day, the temperature measurement point time series data of the photovoltaic station on the same day, the irradiance measurement point time series data, the parameter values ​​required for related calculations and the installed capacity of the inverter equipment;

[0012] The active power measurement point time series data of the day, the temperature measurement point time series data of the photovoltaic station of the day, the irradiance measurement point time series data, the parameter values ​​required for the relevant calculations and the installed capacity of the inverter equipment are determined as the daily input data required for training the photovoltaic power prediction model.

[0013] Preferably, the preliminary pretreatment process includes:

[0014] Deleting the null values ​​in the input data of the said day;

[0015] Normalize the irradiance time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized irradiance time series data;

[0016] Normalize the temperature time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized temperature time series data;

[0017] The normalized irradiance time series data after deleting the over-limit data and the normalized temperature time series data after deleting the over-limit data are aligned according to the timestamps of the data generation time and the null values ​​are removed to construct the independent variable data of the training data for the day;

[0018] If the amount of independent variable data of the training data for the day is less than or equal to the preset amount, the program ends;

[0019] If the amount of the independent variable data of the training data on that day is greater than the preset amount, the active power time series data of each inverter in the input data on that day is processed;

[0020] In the process of processing the active power time series data of each inverter in the input data of the day, the data volume of the time series data of each inverter in the input data of the day and the installed capacity data of each inverter are obtained;

[0021] Eliminate the time series data of each inverter in the input data of the day whose data volume is less than the preset number of time series data of the inverter devices, and obtain the time series data of each inverter whose data volume meets the requirements;

[0022] The time series data of each inverter whose data volume meets the requirements is divided by their respective installed capacity data to obtain normalized active power time series data, and the over-limit data in the normalized active power time series data is deleted to obtain the target active power time series data to complete the preliminary preprocessing process.

[0023] Preferably, the process of removing abnormal data points of the inverter includes:

[0024] For each inverter, the normalized active power time series data after deleting the over-limit data in the input data of the day is used as the dependent variable data of the training data of the day, and the independent variable data of the training data of the day and the dependent variable data of the training data of the day are aligned according to the timestamp of the time when the measurement point is generated, as the initial training data of the photovoltaic power prediction model;

[0025] Determine the point in the initial training data where the normalized active power value of the inverter is a first preset value as a shutdown fault point and remove it;

[0026] The normalized active power time series data of the inverter is used as the y value, the data grouping mark is initialized, and the y value is divided into a preset number of boxes according to the numerical value and the box number is marked to obtain the box number;

[0027] Grouping the initial training data according to the box numbers to obtain a first group;

[0028] If the amount of data in the first group is greater than a preset amount, the x values ​​are sorted in the first group, and a first difference between two adjacent x values ​​in the first group is calculated; wherein the x value is a distance value from the normalized irradiance data and the normalized temperature data to the origin on the two-dimensional plane;

[0029] If the first difference is greater than a preset threshold, the first difference is marked to obtain a marking value;

[0030] If the first difference is less than or equal to the preset threshold, it is marked as empty, until all the differences in the first group are marked, and then all empty values ​​are filled using the previous value method;

[0031] The mark values ​​of the difference between two adjacent points in the first group are grouped by a preset grouping method to obtain a second group; wherein the preset grouping method is a grouping method that groups points with the same mark value into the same group;

[0032] Calculate the amount of data in the second group, the second difference between the maximum value and the minimum value of the x value, and the third difference between the maximum value and the minimum value of the y value;

[0033] Mark the data in the second group that has a larger amount than a preset number and a ratio of the third difference to the second difference that is larger than a preset number as data points with power restriction anomaly and communication anomaly and delete them;

[0034] If the amount of remaining data is greater than a preset amount, the normalized irradiance time series data, the normalized temperature time series data, and the normalized inverter active power time series data are scaled within a preset range, and the coordinates of the remaining data points in the three-dimensional polar coordinate system are calculated using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled inverter normalized active power time series data;

[0035] Using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled inverter normalized active power time series data, a density clustering algorithm is used to perform density clustering on the remaining data points to eliminate discrete noise points of the inverter marked as a second preset value after density clustering;

[0036] The k-means algorithm is used to cluster the first angle values ​​and the second angle values ​​of the data points in the three-dimensional polar coordinate system, and the data volume, angle mean, and radius mean of each cluster are calculated;

[0037] Delete the occluded abnormal points in the cluster whose angle mean is not the maximum, whose radius mean is less than the preset mean, and whose data volume is greater than the preset number, to complete the process of eliminating abnormal data points of the inverter.

[0038] The second aspect of the present application discloses a data processing device for a photovoltaic power prediction model, the device comprising:

[0039] An acquisition unit, used for acquiring the daily input data required for training the photovoltaic power prediction model;

[0040] A data cleaning unit is used to clean the input data of the day by a preset cleaning method to obtain target training data; wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; the elimination of abnormal data points of the inverter includes at least elimination of power limitation abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shielding abnormal data points of the inverter and elimination of discrete noise points of the inverter; the target training data is used to train a photovoltaic power prediction model.

[0041] Preferably, the acquisition unit includes:

[0042] The first acquisition module is used to obtain the current day active power measurement point time series data of all inverter equipment in the photovoltaic station, the current day temperature measurement point time series data of the photovoltaic station, the irradiance measurement point time series data, the parameter values ​​required for related calculations and the installed capacity of the inverter equipment;

[0043] The first determination module is used to determine the current day's active power measurement point time series data, the current day's temperature measurement point time series data of the photovoltaic station, the current day's irradiance measurement point time series data, the parameter values ​​required for the related calculations, and the installed capacity of the inverter equipment as the current day's input data required for training the photovoltaic power prediction model.

[0044] Preferably, the preliminary preprocessing data cleaning unit includes:

[0045] A first deletion module, used to delete null values ​​in the input data of the day;

[0046] A first normalization module is used to normalize the irradiance time series data in the input data of the day after deleting the null value, and delete the over-limit data of the normalized irradiance time series data;

[0047] The second normalization module is used to normalize the temperature time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized temperature time series data;

[0048] A construction module is used to align the normalized irradiance time series data after deleting the over-limit data and the normalized temperature time series data after deleting the over-limit data according to the timestamp of the data generation time and remove the null value to construct the independent variable data of the training data of the day;

[0049] An end module, used to end the program if the amount of independent variable data of the training data on that day is less than or equal to a preset amount;

[0050] A processing module, for processing the active power time series data of each inverter in the input data of the day if the data amount of the independent variable data of the training data of the day is greater than a preset amount, and in the process of processing the active power time series data of each inverter in the input data of the day, obtaining the data amount of the time series data of each inverter in the input data of the day and the installed capacity data of each inverter;

[0051] The second deletion module is used to remove the time series data of each inverter in the input data of the day whose data volume is less than the time series data of the inverter device which is preset, and obtain the time series data of each inverter whose data volume meets the requirements;

[0052] The third deletion module is used to divide the timing data of each inverter whose data volume meets the requirements by their respective installed capacity data to obtain normalized active power timing data, and delete the over-limit data in the normalized active power timing data to obtain the target active power timing data to complete the preliminary preprocessing process.

[0053] Preferably, the data cleaning unit for removing abnormal data points of the inverter includes:

[0054] An alignment module is used for, for each inverter, using the normalized active power time series data after deleting the over-limit data in the input data of the day as the dependent variable data of the training data of the day, and aligning the independent variable data of the training data of the day and the dependent variable data of the training data of the day according to the timestamp of the measurement point generation time as the initial training data of the photovoltaic power prediction model;

[0055] A fourth deletion module, used to determine and delete a point in the initial training data where the normalized active power value of the inverter is a first preset value as a shutdown fault point;

[0056] The first marking module is used to take the normalized active power time series data of the inverter as the y value, initialize the data grouping mark, divide the y value into a preset number of boxes according to the numerical value, mark the box number, and obtain the box number;

[0057] A first grouping module, used for grouping the initial training data according to the box numbers to obtain a first group;

[0058] A sorting calculation module, used for sorting the x values ​​in the first group if the amount of data in the first group is greater than a preset amount, and calculating a first difference between two adjacent x values ​​in the first group; wherein the x value is a distance value from the normalized irradiance data and the normalized temperature data to the origin on a two-dimensional plane;

[0059] A second marking module, configured to mark the first difference value to obtain a marking value if the first difference value is greater than a preset threshold;

[0060] A marking and filling module, used for marking as empty if the first difference is less than or equal to a preset threshold, until all the difference values ​​in the first group are marked, and then filling all empty values ​​using the previous value method;

[0061] A second grouping module is used to group the mark values ​​of the difference between two adjacent points in the first group by a preset grouping method to obtain a second group; wherein the preset grouping method is a grouping method that groups points with the same mark value into the same group;

[0062] A first calculation module is used to calculate the amount of data in the second group, a second difference between the maximum value and the minimum value of the x value, and a third difference between the maximum value and the minimum value of the y value;

[0063] a fifth deletion module, configured to mark and delete data points having a data volume in the second group greater than a preset number and a ratio of the third difference to the second difference greater than a preset number as data points having power restriction anomalies and communication anomalies;

[0064] A second calculation module is used to scale the normalized irradiance time series data, the normalized temperature time series data, and the normalized active power time series data of the inverter within a preset range if the amount of remaining data is greater than a preset amount, and calculate the coordinates of the remaining data points in the three-dimensional polar coordinate system by using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled normalized active power time series data of the inverter;

[0065] A clustering elimination module is used to perform density clustering on the remaining data points using a density clustering algorithm through the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled inverter normalized active power time series data, so as to eliminate discrete noise points of the inverter marked as a second preset value after density clustering;

[0066] A clustering calculation module, used for clustering the first angle value and the second angle value of the data points in the three-dimensional polar coordinate system using a k-means algorithm, and calculating the data volume, angle mean and radius mean of each cluster;

[0067] The sixth deletion module is used to delete the occluded abnormal points in the cluster whose angle mean is not the maximum, whose radius mean is less than the preset mean, and whose data volume is greater than the preset number, so as to complete the process of eliminating abnormal data points of the inverter.

[0068] A third aspect of the present application discloses a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the data processing method of the photovoltaic power prediction model as described in any one of the first aspects.

[0069] The fourth aspect of the present application discloses an electronic device, comprising a memory and one or more instructions, wherein the one or more instructions are stored in the memory and are configured to be executed by one or more processors as the data processing method of the photovoltaic power prediction model as described in any one of the first aspects.

[0070] It can be seen from the above technical scheme that the present application discloses a data processing method, device and related equipment for a photovoltaic power prediction model, which obtains the daily input data required for training the photovoltaic power prediction model, cleans the daily input data through a preset cleaning method, and obtains target training data; wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; elimination of abnormal data points of the inverter includes at least elimination of power limitation abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shading abnormal points of the inverter and elimination of discrete noise points of the inverter; the target training data is used to train the photovoltaic power prediction model.

[0071] Through the above scheme, the time series data, equipment information, parameter values ​​and other input data of the day required for training the photovoltaic power prediction model are obtained. According to the reasons for the abnormal data points caused by accidental factors such as power restriction, communication abnormality, shading, cloudy days, etc., the data characteristics of the abnormal data points are analyzed, so as to clean the time series data of each inverter equipment, the irradiance and temperature data of the photovoltaic station on the same day and other input data of the day. In the process of cleaning the input data of the day, the preliminary preprocessing of the time series data of the measuring point and the elimination of abnormal data points of the inverter are performed. In the case of a small amount of data, combined with photovoltaic business knowledge, as many abnormal data points as possible are deleted. Even if the amount of abnormal data accounts for a large proportion, the abnormal data points can be identified, and the dependence on the hyperparameters set by the algorithm is small. Finally, the cleaned target training data is obtained, and the target training data is used to train the photovoltaic power prediction model, thereby improving the quality of the training data and improving the accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0073] Figure 1 A schematic flow chart of a data processing method for a photovoltaic power prediction model disclosed in an embodiment of the present application;

[0074] Figure 2 The effect diagram of data cleaning at each stage disclosed in the embodiment of this application;

[0075] Figure 3 A schematic flow chart of another data processing method for a photovoltaic power prediction model disclosed in an embodiment of the present application;

[0076] Figure 4A schematic diagram of the structure of a data processing device for a photovoltaic power prediction model disclosed in an embodiment of the present application;

[0077] Figure 5 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0078] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0079] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0080] As can be seen from the background technology, since the average daily power generation time of photovoltaic power stations generally does not exceed 9 hours, calculated at one data point per minute, there are only 540 data points for training per day, and the amount of effective data is even less. When the amount of abnormal data accounts for a large proportion, it is difficult to clean the data only through clustering, isolation forest and other methods. Training data containing dirty data will lead to poor training results of photovoltaic power prediction models, and reduce the accuracy of photovoltaic power prediction models in predicting the power generation of photovoltaic equipment. Therefore, when the amount of data is small, how to improve the quality of training data used by photovoltaic power prediction models is a problem that needs to be solved urgently in this application.

[0081] In order to solve the above problems, the present application discloses a data processing method, device and related equipment for a photovoltaic power prediction model, obtains the time series data, equipment information, parameter values ​​and other daily input data required for training the photovoltaic power prediction model, analyzes the data characteristics of the abnormal data points according to the causes of abnormal data points generated by accidental factors such as power restriction, communication abnormality, shading, cloudy days, etc., and cleans the time series data of each inverter device, the irradiance and temperature data of the photovoltaic station on the same day, and in the process of cleaning the input data on the same day, performs preliminary preprocessing of the time series data of the measurement point and removes the abnormal data points of the inverter, and in the case of a small amount of data, combined with photovoltaic business knowledge, deletes as many abnormal data points as possible, even when the amount of abnormal data accounts for a large proportion, the abnormal data points can be identified, and the dependence on the hyperparameters set by the algorithm is small, and finally the cleaned target training data is obtained, thereby improving the quality of the training data and improving the accuracy of the model. The specific implementation method is specifically described by the following embodiments.

[0082] refer to Figure 1 As shown, a data processing method for a photovoltaic power prediction model disclosed in an embodiment of the present application is shown, and the data processing method for the photovoltaic power prediction model mainly includes the following steps:

[0083] S101: Obtaining the daily input data required for training the photovoltaic power prediction model.

[0084] In S101, first, the active power measurement point time series data of the day, the temperature measurement point time series data of the day, the irradiance measurement point time series data of the day, the parameter values ​​required for related calculations and the installed capacity of the inverter equipment of all inverter equipment in the photovoltaic station are obtained, and then the active power measurement point time series data of the day, the temperature measurement point time series data of the day, the irradiance measurement point time series data of the day, the parameter values ​​required for related calculations and the installed capacity of the inverter equipment are determined as the daily input data required for training the photovoltaic power prediction model.

[0085] It should be noted that the theoretical power generation of a station is linearly related to the total solar radiation. The calculation formula for the power generation of a power station is shown in formula (1).

[0086] (1)

[0087] in, The amount of electricity used for Internet access; is the total horizontal solar irradiance; is the installed capacity; is the irradiance under standard conditions; is the comprehensive coefficient.

[0088] The power generation of photovoltaic equipment is mainly related to the solar irradiance and temperature at that moment, and the relationship is linear. In the actual production process, even if it is interfered by various external factors, it can generally be approximated by a linear relationship. Therefore, the input data of the training data used for the prediction of photovoltaic equipment power generation is temperature data and irradiance data, and the output is the active power of the inverter equipment.

[0089] It should be noted that in order to make the three types of time series data (the above-mentioned irradiance data, temperature data and active power data) have the same order of magnitude, the irradiance data and temperature time series data need to be scaled so that their maximum values ​​are on the order of 10E0. Therefore, the value of the irradiance data is divided by 1000 and the value of the temperature time series data is divided by 100.

[0090] S102: Clean the input data of the day by a preset cleaning method to obtain target training data; wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; elimination of abnormal data points of the inverter includes at least elimination of power limitation abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shielding abnormal points of the inverter and elimination of discrete noise points of the inverter; the target training data is used to train the photovoltaic power prediction model.

[0091] In S102, during the data cleaning process of the input data of the day, the input data of the day is preliminarily preprocessed and abnormal data points of the inverter are eliminated to obtain the target training data required for training the photovoltaic power prediction model; wherein, the preliminary preprocessing is used to delete the null values ​​in all time series data, scale the time series data of temperature, irradiance or other meteorological measurement points, normalize the inverter active power time series data according to the installed capacity of the equipment and eliminate the over-limit data.

[0092] The specific process of constructing the independent variable data and preliminary preprocessing of the training data for the day is shown in A1-A10.

[0093] A1: Deletes the null values ​​in the input data of the current day.

[0094] In A1, delete all the null values ​​in the time series data of the input data for the current day.

[0095] A2: Normalize the irradiance time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized irradiance time series data.

[0096] Irradiance data refers to the radiation energy per unit area per unit time that reaches the solid earth's surface after solar radiation is absorbed, scattered, and reflected by the atmosphere. Its maximum value is approximately 1373W / ㎡. Therefore, data that is not in the range of 0~1400 does not conform to the laws of physics and should be eliminated.

[0097] In A2, the irradiance time series data in the input data of the day is processed, the irradiance time series value is divided by 1000 to obtain the normalized irradiance time series data, and the out-of-limit data of the normalized irradiance time series data is deleted, such as the normalized irradiance time series data greater than 1.4 or less than 0.

[0098] This application does not specifically limit the numerical value of the out-of-limit data of the normalized irradiance time series data.

[0099] A3: Normalize the temperature time series data in the input data of the day after deleting the null values, and delete the over-limit data in the normalized temperature time series data.

[0100] Temperature data refers to the temperature of the photovoltaic string equipment itself. If it cannot be provided, the ambient temperature can also be used. The temperature range is related to the season, geographical location, weather, etc. During the daytime power generation period, the temperature is generally between -50℃ and 100℃. Temperatures outside this range should be excluded.

[0101] In A3, the temperature time series data values ​​in the input data of the day are divided by 100 to obtain normalized temperature time series data, and out-of-limit data of the normalized temperature time series data are deleted, such as data greater than 1 or less than -0.5.

[0102] Among them, the numerical value of the out-of-limit data of the normalized temperature time series data is not specifically limited in this application.

[0103] A4: Align the normalized irradiance time series data after deleting the out-of-limit data and the normalized temperature time series data after deleting the out-of-limit data according to the timestamps of the data generation time and remove the null values ​​to construct the independent variable data of the training data for the day.

[0104] The independent variable data of the training data for the day refers to the scaled irradiance time series data and the scaled temperature time series data after time alignment for the day.

[0105] After aligning the scaled irradiance time series data with the scaled temperature time series data at time, and removing the null values, the distance from the scaled irradiance and temperature data to the origin on the two-dimensional plane is calculated and recorded as the "x value".

[0106] A5: If the amount of independent variable data in the training data for the day is less than or equal to the preset amount, the program ends.

[0107] The data volume judgment is to judge the overall situation of the time series data of each measurement point on the day, including the active power measurement point of photovoltaic power generation equipment and the temperature, irradiance or other measurement points of meteorological monitoring equipment. When the amount of time series data is less than a certain value, it means that the amount of training data is too small, which may easily lead to underfitting of the model and is not suitable for training the photovoltaic power prediction model.

[0108] The preset number may be 60, 65, etc. The preset data is set according to the actual situation, and this application does not make any specific limitation. The preset number of this application is preferably 60.

[0109] A6: If the amount of the independent variable data of the training data on that day is greater than the preset amount, the active power time series data of each inverter in the input data on that day is processed.

[0110] It should be noted that if the number of independent variable data of the input data on that day is greater than the preset number, the time series data of the inverter whose data amount meets the requirement is determined.

[0111] A7: In the process of processing the active power time series data of each inverter in the input data of the day, the data volume of the time series data of each inverter in the input data of the day and the installed capacity data of each inverter are obtained.

[0112] A8: Eliminate the time series data of each inverter in the input data of the day whose data volume is less than the preset number of time series data of the inverter devices, and obtain the time series data of each inverter whose data volume meets the requirements.

[0113] If the amount of the daily time series data of each inverter is less than the preset amount, for example, if the amount of the daily time series data of each inverter is less than 60, it is determined that the quality of the daily time series data of the inverter is poor and should not be retained as training data.

[0114] A9: Divide the time series data of each inverter whose data volume meets the requirements by their respective installed capacity data to obtain normalized active power time series data, and delete the over-limit data in the normalized active power time series data to obtain the target active power time series data to complete the preliminary preprocessing process.

[0115] The reason for normalizing the inverter timing data is that the installed capacity of different photovoltaic power generation equipment is different. This value is related to the installed capacity and number of strings under it. The installed capacity represents the maximum active power that the device can output. Although the actual active power range output by different devices is different, the ratio of actual active power to installed capacity has the same value range, both between 0 and 1. Values ​​less than 0 and greater than 1 are data that do not conform to the laws of physics. Therefore, it is necessary to eliminate data whose active power values ​​are not in the range of 0~1 after normalization.

[0116] In A9, for the time series data of the inverter whose data volume meets the requirements, divide it by its own installed capacity to obtain the normalized active power time series data, and delete the excessive data in the normalized active power time series data, such as data greater than 1 or less than 0, to obtain the target active power time series data.

[0117] A10: For each inverter, the normalized active power time series data after removing the over-limit data from the input data of the day is used as the dependent variable data of the training data of the day, and the independent variable data of the training data of the day and the dependent variable data of the training data of the day are aligned according to the timestamp of the measurement point generation time, as the initial training data of the photovoltaic power prediction model after preliminary preprocessing.

[0118] A scatter plot (such as a two-dimensional scatter plot, a three-dimensional scatter plot, etc.) can be drawn for the merged training data. When drawing a two-dimensional scatter plot, irradiance data and active power data can be used. When drawing a three-dimensional scatter plot, irradiance data, temperature data, and active power data are used. If the inverter device has no abnormal points, the scatter plot will show an approximate straight line with a slope close to 1, which is a plane straight line in a two-dimensional scatter plot and a spatial straight line in a three-dimensional scatter plot.

[0119] The specific process of eliminating abnormal data points of the inverter is shown in B1-B13.

[0120] B1: Determine the points in the initial training data where the normalized active power value of the inverter is a first preset value as shutdown fault points and eliminate them.

[0121] Among them, the first preset value can be 0.

[0122] Eliminate shutdown fault points, that is, points where the active power value of the equipment is 0 after normalization.

[0123] When photovoltaic power generation equipment is affected by various internal and external factors, such as power cuts, abnormal communication, cloudy or overcast weather, shading, non-shutdown faults, etc., some data points will be far away from the straight line. The specific power cuts, abnormal communication, cloudy or overcast weather, shading, non-shutdown faults, etc. are explained as follows:

[0124] When the equipment is affected by power limitation and communication anomaly, the power generation is the given constant value of power limitation or the last value before communication anomaly. This value does not change with the change of irradiance and presents a horizontal straight line feature that is independent of irradiance.

[0125] On cloudy or overcast days, the irradiance data fluctuates dramatically in the time dimension, and the active power data fluctuates accordingly. Affected by the hysteresis of equipment power regulation, the active power values ​​of the same equipment under the same irradiance value may be different, and the data points are relatively scattered;

[0126] Occlusion mainly occurs in the morning and evening, and data points at times of high irradiance are generally not affected by occlusion;

[0127] Non-shutdown faults generally will not be eliminated in a short time. When the equipment is blocked or a non-shutdown fault occurs, the power generation is lower than the normal value and increases with the increase of irradiance. The slope of the straight line composed of data points in the blocking or non-shutdown fault time period is smaller than the slope in normal circumstances, or even 0.

[0128] B2: Use the normalized active power time series data of the inverter as the y value, initialize the data grouping mark, and divide the y value into a preset number of boxes according to the numerical value and mark the box number to obtain the box number.

[0129] For a set of training data points , the horizontal axis is the modulus of the vector of the input data of the training data at the i-th moment, that is, ; is the value of the zth meteorological observation point at the ith moment; the ordinate is the value of the output data of the training data at the i-th moment; n is the number of data;

[0130] Mark the y values ​​with box numbers, that is, when hour, Marked as k, that is belongs to group k; for The maximum value of; m is the number of boxes, there is .

[0131] For example, the normalized time series data of the inverter is used as the y value, and the y value is divided into 150 boxes, that is, m=150, and each data point is marked with a box number, and the box number of the first point (that is, the grouping mark) is k=1.

[0132] B3: Group the initial training data according to the box number to obtain the first group.

[0133] The training data is grouped according to the box numbers marked in B2, which is recorded as the first group (i.e., group 1).

[0134] B4: If the amount of data in the first group is greater than a preset amount, the x values ​​in the first group are sorted, and a first difference between two adjacent x values ​​in the first group is calculated.

[0135] After sorting the x values ​​within the group, the coordinates of the data points within the kth group are , u is the number of data points in the kth group, and ; Calculate in sequence , to ;

[0136] Among them, the preset number is set according to the actual situation and this application does not make any specific limitation.

[0137] For example, if the amount of data in a group 1 is greater than 10, the first difference between two adjacent x values ​​after sorting in the first group is calculated, otherwise the data in the next group 1 is calculated.

[0138] B5: If the first difference is greater than a preset threshold, the first difference is marked to obtain a marking value.

[0139] If the first difference is less than the preset threshold, the points are considered to be clustered together and the two points are marked the same; if the first difference is greater than the preset threshold, the points are considered to be not clustered together and the two points are marked differently;

[0140] The tag value is represented by j, and the initial value of j can be set to 1.

[0141] If the first difference is greater than the preset threshold, it is marked as j, and j=j+1.

[0142] B6: If the first difference is less than or equal to the preset threshold, it is marked as empty. After all the differences in the first group are marked, all empty values ​​are filled using the previous value method.

[0143] For the points that are clustered together, the marks of these points are the same; for the points that are not clustered together, different marks are given to the two adjacent points. After filling with the previous value method, the marks a of all the data points in group 1 are obtained. If there are v different marks, and the marks of the first two points are the same, there is ,have ;

[0144] In B6, if the first difference is less than the threshold, it is marked as empty. After all the differences in group 1 are marked, all empty values ​​are filled using the previous value method.

[0145] B7: Grouping the mark values ​​of the first difference between two adjacent points in the first group by a preset grouping method to obtain a second group; wherein the preset grouping method is a grouping method that groups points with the same mark value into the same group.

[0146] Among them, grouping is performed again according to the mark value j of the first difference between two adjacent points in group 1 to obtain the second group (ie, group 2).

[0147] B8: Calculate the amount of data in the second group, the second difference between the maximum and minimum values ​​of the x value, and the third difference between the maximum and minimum values ​​of the y value.

[0148] In B8, the amount of data in each group 2 is calculated, the difference between the maximum and minimum values ​​of the x value is calculated to obtain the second difference, which can be represented by B, and the difference between the maximum and minimum values ​​of the y value is calculated to obtain the third difference, which can be represented by A.

[0149] In the group with k as the first group and j as the second group, the difference between the maximum and minimum values ​​of x and y are B= and A= ;

[0150] B9: Mark the data in the second group that has a data volume greater than a preset number and a ratio of the third difference to the second difference greater than a preset number as data points with power restriction anomaly and communication anomaly and delete them.

[0151] calculate ,like When the value is less than the threshold (the threshold is set according to the actual situation and is not specifically limited in this application) or the number of points under this mark is small, it is marked as normal data, otherwise it is marked as an abnormal point.

[0152] Delete all data points with power restriction anomalies and communication anomalies to obtain the remaining data volume.

[0153] For example, if the amount of data in group 2 is more than 10 and B / A is greater than 10, the data in this group are marked as abnormal points, otherwise they are marked as normal points.

[0154] The above B1-B9 are used to remove abnormal points during power outages and communication anomalies. The data characteristics of abnormal points caused by these two reasons are the same, which are seemingly continuous data points, with a large range of changes in the horizontal axis and a small range of changes in the vertical axis. The vertical axis is sometimes exactly the same value, sometimes fluctuates within a small range, and even the point set is a slant line with a very small slope. When cleaning the data, these clustered points should be removed, but the point sets that are almost overlapping should not be deleted by mistake. For discontinuous points with similar y values, since it is impossible to confirm which one is a normal data point, they should all be retained.

[0155] B10: If the amount of remaining data is greater than the preset amount, scale the normalized irradiance timing data, the normalized temperature timing data, and the inverter normalized active power timing data within the preset range, and calculate the coordinates of the remaining data points in the three-dimensional polar coordinate system using the scaled normalized irradiance timing data, the scaled normalized temperature timing data, and the scaled inverter normalized active power timing data.

[0156] Among them, the preset range is [0.001, 1].

[0157] In B10, if the amount of remaining data is more than 60, the irradiance, temperature data, and inverter normalized time series data are scaled to the range of [0.001, 1], and the scaled data are used to calculate the coordinates of the data point in the three-dimensional polar coordinate system.

[0158] B11: Using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled inverter normalized active power time series data, a density clustering algorithm is used to perform density clustering on the remaining data points to eliminate the noise points of the inverter time series data marked as the second preset value after density clustering.

[0159] The second preset value may be -1.

[0160] For example, if the number of remaining data points is more than 60, the scaled data is used for density clustering, and the hyperparameters are set to eps=0.04, min_samples=20, and the noise points of the inverter timing data marked as -1 are removed.

[0161] When using the density clustering algorithm, the two hyperparameters of the algorithm, the minimum number of cluster points and the cluster radius, can be set more loosely. After clustering, the data points with a class label of -1 are abnormal points.

[0162] The above B10-B11 is used to remove noise points generated by cloudy, overcast or other accidental factors. The noise data is relatively discrete, the number is small, and the distribution of noise data has no regularity, so the density clustering algorithm is used.

[0163] B12: Use the k-means algorithm to cluster the first angle values ​​and second angle values ​​of the data points in the three-dimensional polar coordinate system, and calculate the data volume, angle mean, and radius mean of each cluster.

[0164] The coordinates of the data point in the three-dimensional polar coordinate system are (radius, first angle value, and second angle value).

[0165] Use the k-means algorithm to cluster the first angle value and the second angle value in the polar coordinates, set the hyperparameter n_clusters to 6, and calculate the data volume, angle mean, and radius mean of each cluster.

[0166] B13: Delete the occluded abnormal points in the cluster whose angle mean is not the maximum, whose radius mean is less than the preset mean, and whose data volume is greater than the preset number, so as to complete the process of eliminating abnormal data points of the inverter.

[0167] The data in the clusters with small angle mean and small radius mean are outliers.

[0168] In B13, data points in clusters whose angle mean is not the maximum, whose radius mean is less than a preset mean, such as 0.4, and whose data volume is greater than 10 are deleted, and the remaining data is output by the program as training data for the inverter device.

[0169] The above B12-B13 is used to eliminate abnormal data points generated under occlusion conditions, namely, occlusion abnormal points. The slope of the curve composed of these abnormal data points is significantly smaller than that of normal data points, and usually occurs when the irradiance is low. Therefore, the k-means clustering algorithm is used to cluster the polar coordinates of the data points.

[0170] After the abnormal points analyzed in the above steps are removed, the remaining data points can be brought into the model as target training data to train the model.

[0171] The hyperparameters used in the above technical solutions can be optimized based on experience or by using algorithms such as particle swarm. The hyperparameters involved include but are not limited to: the number of bins m when binning by y value, The threshold of , the number of clusters selected for k-means clustering, the threshold of the radius mean of clusters judged as occluded, the minimum number of cluster points and cluster radius for density clustering.

[0172] The target training data processed by the above steps can be brought into the photovoltaic equipment power prediction model for training to obtain the power prediction model of each device on that day. Using this model to predict the given prediction data, the theoretical power generation power of the device on that day can be obtained.

[0173] The data cleaning results of the above stages are as follows Figure 2 As shown, the horizontal axis is the scaled daily irradiance data, and the vertical axis is the normalized active power data. Sub-figure (a) is a scatter plot after preliminary data preprocessing, the green points in sub-figure (b) are the identified power outage and communication interruption data, and the blue points are the data retained after power outage and communication interruption cleaning, the blue points in sub-figure (c) are the identified noise data, and the green points are the data retained after noise removal, the blue points in sub-figure (d) are the data retained after occlusion removal, and the points of other colors are the identified occlusion data.

[0174] In order to facilitate the understanding of the data processing process of the photovoltaic power prediction model, combined with Figure 3 To illustrate with an example, Figure 3 A flow chart of another data processing method for a photovoltaic power prediction model is shown.

[0175] Figure 3 , get the input data and remove the null values;

[0176] The input data includes the active power measurement point time series data of all inverter equipment under the photovoltaic station on the same day, the irradiance measurement point time series data of the same day, the temperature measurement point time series data of the same day, and the installed capacity of all inverter equipment, and the calculation of the relevant parameter values ​​required.

[0177] The relevant parameter values ​​required for calculation are mainly the parameters used in the data cleaning process, such as various thresholds, hyperparameters of the model used in cleaning, etc. See all the values ​​mentioned in the examples for details.

[0178] Preprocess the input data and construct the independent variable data of the training data for that day. The specific details of the preprocessing process are as follows:

[0179] Divide the irradiance data by 1000 as the normalized irradiance, and retain the data in the range of 0 to 1.4;

[0180] The temperature data is divided by 100 as the normalized temperature, and the data in the range of -0.5 to 1 is retained;

[0181] Align the normalized irradiance time series data and the normalized temperature time series data by time, remove the null values, and use them as independent variable data;

[0182] Determine whether the amount of independent variable data is greater than or equal to 60;

[0183] If the number of independent variable data is less than 60, no results will be output and the program will end;

[0184] If the amount of independent variable data is greater than or equal to 60, calculate the distance from the normalized irradiance and temperature to the origin on the two-dimensional plane as the x value;

[0185] Determine whether there is unprocessed inverter data;

[0186] If there is no unprocessed inverter data, merge and output the training data of all inverters, and end the program;

[0187] If there is unprocessed inverter data, filter the timing data and installed capacity of the next inverter, and initialize the data grouping mark i=1;

[0188] Determine whether the number of time series data of the device is less than 60;

[0189] If the number of the device time series data is less than 60, return to the step of determining whether there is unprocessed inverter data;

[0190] If the number of time series data of the device is greater than or equal to 60, divide the time series data by the installed capacity to obtain the normalized power, record the normalized power as the y value, and retain the data in the range of (0, 1];

[0191] Align the y value with the independent variable data at time to obtain the training data;

[0192] Divide the y values ​​of the training data into 150 boxes and mark the box numbers. Group the training data by box number and record it as "Group 1";

[0193] Determine whether there is unprocessed group 1 (group 1) and whether the amount of data in the group is greater than 10;

[0194] If there is an unprocessed group 1 and the amount of data in the group is greater than 10, sort the data in group 1 by x value and calculate the difference between the x values ​​of two adjacent points;

[0195] Determine whether the difference is less than 0.05;

[0196] If the difference is less than 0.05, mark the difference as empty;

[0197] If the difference is greater than or equal to 0.05, mark the difference as j, then j=j+1, and use the previous value method to fill the empty difference;

[0198] Group the data by difference mark, i.e. group 2;

[0199] Calculate the amount of data in each group 2, the difference between the maximum and minimum x values ​​(B), and the difference between the maximum and minimum y values ​​(A);

[0200] Determine whether the amount of data in group 2 is greater than 10, and whether B / A is greater than 10;

[0201] If the amount of data in group 2 is greater than 10 and B / A is greater than 10, the data in group 2 are marked as outliers;

[0202] If the amount of data in group 2 is less than or equal to 10, or B / A is less than or equal to 10, the data in group 2 are marked as normal points;

[0203] After marking all group 2 points in group 1, delete all points marked as abnormal in group 1;

[0204] After deleting all the points marked as abnormal in group 1, return to the step of determining whether there is unprocessed group 1 and whether the amount of data in the group is greater than 10;

[0205] If it is determined that there is no unprocessed group 1 (group 1), or the amount of data in the group is less than or equal to 10, then it is determined whether the amount of time series data for the day is less than 60;

[0206] If the amount of time series data for the day is less than 60, return to the step of determining whether there is unprocessed inverter data;

[0207] If the amount of time series data for the day is greater than or equal to 60, scale each column of the training data to the range of [0.0001, 1];

[0208] Use the scaled data to calculate the coordinate value of each data point in the three-dimensional polar coordinate system;

[0209] Use density clustering for the remaining data points and remove the data points marked as -1;

[0210] Determine whether the amount of data after removing the data points marked as -1 is less than 60;

[0211] If the amount of data after removing the data points marked as -1 is less than 60, return to the step of determining whether there is unprocessed inverter data;

[0212] If the amount of data after removing the data points marked as -1 is greater than or equal to 60, use the k-means algorithm to perform clustering calculations on the two angle values ​​in the polar coordinates, set the hyperparameter n_clusters to 6, and calculate the amount of data, angle mean, radius mean, etc. of each cluster;

[0213] The data points in the clusters with non-maximum angle mean, radius mean less than 0.4, and more than 10 data points are marked as outliers and deleted;

[0214] For each inverter device, determine whether the amount of training data on the day after deleting abnormal points is less than 60;

[0215] If the amount of the training data of the inverter device on the day after deleting the abnormal point is less than 60, return to the step of determining whether there is unprocessed inverter data;

[0216] If the amount of training data of the inverter device on that day after the abnormal point is deleted is greater than or equal to 60, the training data of the inverter device on that day is retained;

[0217] The training data of each inverter device on the same day are merged, and the training data of all inverters are output, that is, the target training data.

[0218] The data processed by the above steps can be brought into the photovoltaic equipment power prediction model for training, thereby improving the quality of training data and the accuracy of the model. The power prediction model of each device on that day is obtained. Using this model to predict the given prediction data, the theoretical power generation power of the device on that day can be obtained.

[0219] This application obtains the time series data, equipment information, parameter values, etc. required for training the optical power model; preprocesses the time series data and constructs the input and output of the training data; cleans the training data of each device in sequence according to the causes and data characteristics of abnormal data points caused by accidental factors such as power restrictions, communication anomalies, obstructions, and cloudy days; and finally outputs the cleaned target training data for model training. The above scheme can effectively eliminate a variety of abnormal data when the data volume is small, and a set of common calculation parameters can be used for inverter devices of different brands and models.

[0220] This application analyzes the characteristics of time series data generated by photovoltaic power generation equipment under various circumstances such as obstruction, failure, power restriction, communication anomaly, cloudy or overcast weather, and uses different methods to mark abnormal points for different types of data. It realizes the cleaning of training data required for photovoltaic equipment power generation prediction under the condition of small data volume (reflected in that only the time series data of the day is used for data cleaning, with a maximum of only 540 available data, and sometimes less than 100 data, while conventional data modeling uses at least thousands of data, so the amount of data before conventional data modeling cleaning can reach several thousand, and the proportion of abnormal data is small, which is convenient for removing abnormal points), combined with photovoltaic business knowledge, and the cleaning of training data required for photovoltaic equipment power generation prediction, as many abnormal data points as possible are deleted, and even when the proportion of abnormal data is large, abnormal data points can be identified, and the dependence on the hyperparameters set by the algorithm is small, and the same set of hyperparameters can be used for most devices, which is convenient for parameter management and reduces the amount of calculation when searching for parameters, and finally obtains the cleaned target training data, which is used to train the photovoltaic power prediction model, thereby improving the quality of training data and improving the accuracy of the model.

[0221] The beneficial effects of the embodiments of the present application are as follows: the time series data, equipment information, parameter values ​​and other input data of the day required for training the photovoltaic power prediction model are obtained; the data characteristics of the abnormal data points are analyzed according to the reasons for the abnormal data points generated by accidental factors such as power restriction, communication abnormality, shading, cloudy days, etc., so as to clean the time series data of each inverter device, the irradiance and temperature data of the photovoltaic station on the same day and other input data of the day; in the process of cleaning the input data of the day, preliminary preprocessing of the time series data of the measuring points and elimination of abnormal data points of the inverter are performed; when the amount of data is small, combined with photovoltaic business knowledge, as many abnormal data points as possible are deleted; even when the amount of abnormal data accounts for a large proportion, the abnormal data points can be identified, and the dependence on the hyperparameters set by the algorithm is small, and finally the cleaned target training data is obtained, and the target training data is used to train the photovoltaic power prediction model, thereby improving the quality of the training data and improving the accuracy of the model.

[0222] Based on the above embodiments Figure 1 A data processing method for a photovoltaic power prediction model is disclosed. The embodiment of the present application also discloses a data processing device for a photovoltaic power prediction model. Figure 4 As shown, the data processing device of the photovoltaic power prediction model includes:

[0223] An acquisition unit 401 is used to acquire the daily input data required for training the photovoltaic power prediction model;

[0224] The data cleaning unit 402 is used to clean the input data of the day by a preset cleaning method to obtain target training data; wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; elimination of abnormal data points of the inverter includes at least elimination of power-limiting abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shielding abnormal points of the inverter and elimination of discrete noise points of the inverter; the target training data is used to train the photovoltaic power prediction model;

[0225] Furthermore, the acquisition unit 401 includes:

[0226] The first acquisition module is used to obtain the current day active power measurement point time series data of all inverter equipment in the photovoltaic station, the current day temperature measurement point time series data of the photovoltaic station, the irradiance measurement point time series data, the parameter values ​​required for related calculations and the installed capacity of the inverter equipment;

[0227] The first determination module is used to determine the current day's active power measurement point time series data, the current day's temperature measurement point time series data of the photovoltaic station, the current day's irradiance measurement point time series data, the parameter values ​​required for related calculations, and the installed capacity of the inverter equipment as the current day's input data required for training the photovoltaic power prediction model.

[0228] The preliminary pre-processing data cleaning unit 402 includes:

[0229] The first deletion module is used to delete the null values ​​in the input data of the day;

[0230] A first normalization module is used to normalize the irradiance time series data in the input data of the day after deleting the null value, and delete the over-limit data of the normalized irradiance time series data;

[0231] The second normalization module is used to normalize the temperature time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized temperature time series data;

[0232] A construction module is used to align the normalized irradiance time series data after deleting the over-limit data and the normalized temperature time series data after deleting the over-limit data according to the timestamp of the data generation time and remove the null value to construct the independent variable data of the training data of the day;

[0233] An end module, used to end the program if the amount of independent variable data of the training data on that day is less than or equal to a preset amount;

[0234] A processing module, for processing the active power time series data of each inverter in the input data of the day if the data amount of the independent variable data of the training data of the day is greater than a preset amount, and in the process of processing the active power time series data of each inverter in the input data of the day, obtaining the data amount of the time series data of each inverter in the input data of the day and the installed capacity data of each inverter;

[0235] The second deletion module is used to remove the time series data of each inverter in the input data of the day whose data volume is less than the time series data of the inverter device which is preset, and obtain the time series data of each inverter whose data volume meets the requirements;

[0236] The third deletion module is used to divide the timing data of each inverter whose data volume meets the requirements by their respective installed capacity data to obtain normalized active power timing data, and delete the over-limit data in the normalized active power timing data to obtain the target active power timing data to complete the preliminary preprocessing process.

[0237] Furthermore, the data cleaning unit 402 for removing abnormal data points of the inverter includes:

[0238] An alignment module is used for, for each inverter, using the normalized active power time series data after deleting the over-limit data in the input data of the day as the dependent variable data of the training data of the day, and aligning the independent variable data of the training data of the day and the dependent variable data of the training data of the day according to the timestamp of the time when the measurement point is generated, as the initial training data of the photovoltaic power prediction model;

[0239] A fourth deletion module, used to determine and delete a point in the initial training data where the normalized active power value of the inverter is a first preset value as a shutdown fault point;

[0240] The first marking module is used to use the normalized active power time series data of the inverter as the y value, initialize the data grouping mark, and divide the y value into a preset number of boxes according to the value size and mark the box number to obtain the box number;

[0241] A first grouping module, used for grouping the initial training data according to the box number to obtain a first group;

[0242] A sorting calculation module, used for sorting the x values ​​in the first group if the amount of data in the first group is greater than a preset amount, and calculating a first difference between two adjacent x values ​​in the first group; wherein the x value is a distance value from the normalized irradiance data and the normalized temperature data to the origin on a two-dimensional plane;

[0243] A second marking module, configured to mark the first difference value to obtain a marking value if the first difference value is greater than a preset threshold;

[0244] A marking and filling module, used for marking as empty if the first difference is less than or equal to a preset threshold, until all the difference values ​​in the first group are marked, and then filling all empty values ​​using the previous value method;

[0245] A second grouping module is used to group the mark values ​​of the difference between two adjacent points in the first group by a preset grouping method to obtain a second group; wherein the preset grouping method is a grouping method that groups points with the same mark value into the same group;

[0246] A first calculation module is used to calculate the amount of data in the second group, a second difference between the maximum value and the minimum value of the x value, and a third difference between the maximum value and the minimum value of the y value;

[0247] A fifth deletion module, configured to mark and delete data in the second group whose data volume is greater than a preset number and whose ratio of the third difference to the second difference is greater than a preset number as data points of power restriction anomaly and communication anomaly;

[0248] A second calculation module is used to scale the normalized irradiance time series data, the normalized temperature time series data, and the normalized active power time series data of the inverter within a preset range if the amount of remaining data is greater than a preset amount, and calculate the coordinates of the remaining data points in the three-dimensional polar coordinate system by using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled normalized active power time series data of the inverter;

[0249] A clustering elimination module is used to perform density clustering on the remaining data points using a density clustering algorithm through the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled inverter normalized active power time series data, so as to eliminate discrete noise points of the inverter marked as a second preset value after density clustering;

[0250] A clustering calculation module, used for clustering the first angle value and the second angle value of the data points in the three-dimensional polar coordinate system using a k-means algorithm, and calculating the data volume, angle mean and radius mean of each cluster;

[0251] The sixth deletion module is used to delete the occluded abnormal points in the cluster whose angle mean is not the maximum, whose radius mean is less than the preset mean, and whose data volume is greater than the preset number, so as to complete the process of eliminating abnormal data points of the inverter.

[0252] The beneficial effects of the embodiments of the present application are as follows: the time series data, equipment information, parameter values ​​and other input data of the day required for training the photovoltaic power prediction model are obtained; the data characteristics of the abnormal data points are analyzed according to the reasons for the abnormal data points generated by accidental factors such as power restriction, communication abnormality, shading, cloudy days, etc., so as to clean the time series data of each inverter device, the irradiance and temperature data of the photovoltaic station on the same day and other input data of the day; in the process of cleaning the input data of the day, preliminary preprocessing of the time series data of the measuring points and elimination of abnormal data points of the inverter are performed; as many abnormal data points as possible are deleted without accidentally deleting normal data points in the data set; even when the amount of abnormal data accounts for a large proportion, the abnormal data points can be identified, and the dependence on the hyperparameters set by the algorithm is small; finally, the cleaned target training data is obtained, and the target training data is used to train the photovoltaic power prediction model, thereby improving the quality of the training data and improving the accuracy of the model.

[0253] An embodiment of the present application further provides a storage medium, which includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the data processing method of the photovoltaic power prediction model as described above.

[0254] The present application also provides an electronic device, whose structural diagram is as follows: Figure 5 As shown, it specifically includes a memory 501 and one or more instructions 502, wherein the one or more instructions 502 are stored in the memory 501 and are configured to be executed by one or more processors 503 to execute the one or more instructions 502 to execute the data processing method of the above-mentioned photovoltaic power prediction model.

[0255] For the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0256] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0257] The steps in the methods of the embodiments of the present application can be adjusted in order, combined, and deleted according to actual needs.

[0258] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0259] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0260] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A data processing method for a photovoltaic power prediction model, characterized in that: The method comprises: Obtain the daily input data required to train the photovoltaic power prediction model; The input data of the day is cleaned by a preset cleaning method to obtain target training data; wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; the elimination of abnormal data points of the inverter includes at least elimination of power limitation abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shielding abnormal points of the inverter and elimination of discrete noise points of the inverter; the target training data is used to train a photovoltaic power prediction model.

2. The method according to claim 1, characterized in that The obtaining of the daily input data required for training the photovoltaic power prediction model includes: Obtain the active power measurement point time series data of all inverter equipment in the photovoltaic station on the same day, the temperature measurement point time series data of the photovoltaic station on the same day, the irradiance measurement point time series data, the parameter values ​​required for related calculations and the installed capacity of the inverter equipment; The active power measurement point time series data of the day, the temperature measurement point time series data of the photovoltaic station of the day, the irradiance measurement point time series data, the parameter values ​​required for the relevant calculations and the installed capacity of the inverter equipment are determined as the daily input data required for training the photovoltaic power prediction model.

3. The method according to claim 1, characterized in that The initial pre-processing process includes: Delete the null values ​​in the input data of the day; Normalize the irradiance time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized irradiance time series data; Normalize the temperature time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized temperature time series data; The normalized irradiance time series data after deleting the over-limit data and the normalized temperature time series data after deleting the over-limit data are aligned according to the timestamps of the data generation time and the null values ​​are removed to construct the independent variable data of the training data for the day; If the amount of independent variable data of the training data for the day is less than or equal to the preset amount, the program ends; If the amount of the independent variable data of the training data on that day is greater than the preset amount, the active power time series data of each inverter in the input data on that day is processed; In the process of processing the active power time series data of each inverter in the input data of the day, the data volume of the time series data of each inverter in the input data of the day and the installed capacity data of each inverter are obtained; Eliminate the time series data of each inverter in the input data of the day whose data volume is less than the preset number of time series data of the inverter devices, and obtain the time series data of each inverter whose data volume meets the requirements; The time series data of each inverter whose data volume meets the requirements is divided by their respective installed capacity data to obtain normalized active power time series data, and the over-limit data in the normalized active power time series data is deleted to obtain the target active power time series data to complete the preliminary preprocessing process.

4. The method according to claim 3, characterized in that The process of eliminating abnormal data points of the inverter includes: For each inverter, the normalized active power time series data after deleting the over-limit data in the input data of the day is used as the dependent variable data of the training data of the day, and the independent variable data of the training data of the day and the dependent variable data of the training data of the day are aligned according to the timestamp of the time when the measurement point is generated, as the initial training data of the photovoltaic power prediction model; Determine the point in the initial training data where the normalized active power value of the inverter is a first preset value as a shutdown fault point and remove it; The normalized active power time series data of the inverter is used as the y value, the data grouping mark is initialized, and the y value is divided into a preset number of boxes according to the numerical value and the box number is marked to obtain the box number; Grouping the initial training data according to the box numbers to obtain a first group; If the amount of data in the first group is greater than a preset amount, the x values ​​are sorted in the first group, and a first difference between two adjacent x values ​​in the first group is calculated; wherein the x value is a distance value from the normalized irradiance data and the normalized temperature data to the origin on the two-dimensional plane; If the first difference is greater than a preset threshold, the first difference is marked to obtain a marking value; If the first difference is less than or equal to the preset threshold, it is marked as empty, until all the differences in the first group are marked, and then all empty values ​​are filled using the previous value method; The mark values ​​of the difference between two adjacent points in the first group are grouped by a preset grouping method to obtain a second group; wherein the preset grouping method is a grouping method that groups points with the same mark value into the same group; Calculate the amount of data in the second group, the second difference between the maximum value and the minimum value of the x value, and the third difference between the maximum value and the minimum value of the y value; Mark the data in the second group that has a larger amount than a preset number and a ratio of the third difference to the second difference that is larger than a preset number as data points with power restriction anomaly and communication anomaly and delete them; If the amount of remaining data is greater than a preset amount, the normalized irradiance time series data, the normalized temperature time series data, and the normalized active power time series data of the inverter are scaled within a preset range, and the coordinates of the remaining data points in the three-dimensional polar coordinate system are calculated using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled normalized active power time series data of the inverter; Using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled inverter normalized active power time series data, a density clustering algorithm is used to perform density clustering on the remaining data points to eliminate discrete noise points of the inverter marked as a second preset value after density clustering; The k-means algorithm is used to cluster the first angle values ​​and the second angle values ​​of the data points in the three-dimensional polar coordinate system, and the data volume, angle mean, and radius mean of each cluster are calculated; Delete the occluded abnormal points in the cluster whose angle mean is not the maximum, whose radius mean is less than the preset mean, and whose data volume is greater than the preset number, to complete the process of eliminating abnormal data points of the inverter.

5. A data processing device for a photovoltaic power prediction model, characterized in that: The device comprises: An acquisition unit, used for acquiring the daily input data required for training the photovoltaic power prediction model; A data cleaning unit is used to clean the input data of the day by a preset cleaning method to obtain target training data; wherein the preset cleaning method includes preliminary preprocessing and elimination of abnormal data points of the inverter; the elimination of abnormal data points of the inverter includes at least elimination of power limitation abnormal data points of the inverter, elimination of communication abnormal data points of the inverter, elimination of shielding abnormal data points of the inverter and elimination of discrete noise points of the inverter; the target training data is used to train a photovoltaic power prediction model.

6. The device according to claim 5, characterized in that The acquisition unit comprises: The first acquisition module is used to obtain the current day active power measurement point time series data of all inverter equipment in the photovoltaic station, the current day temperature measurement point time series data of the photovoltaic station, the irradiance measurement point time series data, the parameter values ​​required for related calculations and the installed capacity of the inverter equipment; The first determination module is used to determine the current day's active power measurement point time series data, the current day's temperature measurement point time series data of the photovoltaic station, the current day's irradiance measurement point time series data, the parameter values ​​required for the related calculations, and the installed capacity of the inverter equipment as the current day's input data required for training the photovoltaic power prediction model.

7. The device according to claim 5, characterized in that The data cleaning unit for preliminary preprocessing includes: A first deletion module, used to delete null values ​​in the input data of the day; A first normalization module is used to normalize the irradiance time series data in the input data of the day after deleting the null value, and delete the over-limit data of the normalized irradiance time series data; The second normalization module is used to normalize the temperature time series data in the input data of the day after deleting the null values, and delete the over-limit data of the normalized temperature time series data; A construction module is used to align the normalized irradiance time series data after deleting the over-limit data and the normalized temperature time series data after deleting the over-limit data according to the timestamp of the data generation time and remove the null value to construct the independent variable data of the training data of the day; An end module, used to end the program if the amount of independent variable data of the training data on that day is less than or equal to a preset amount; A processing module, for processing the active power time series data of each inverter in the input data of the day if the data amount of the independent variable data of the training data of the day is greater than a preset amount, and in the process of processing the active power time series data of each inverter in the input data of the day, obtaining the data amount of the time series data of each inverter in the input data of the day and the installed capacity data of each inverter; The second deletion module is used to remove the time series data of each inverter in the input data of the day whose data volume is less than the time series data of the inverter device which is preset, and obtain the time series data of each inverter whose data volume meets the requirements; The third deletion module is used to divide the timing data of each inverter whose data volume meets the requirements by their respective installed capacity data to obtain normalized active power timing data, and delete the over-limit data in the normalized active power timing data to obtain the target active power timing data to complete the preliminary preprocessing process.

8. The device according to claim 7, characterized in that The data cleaning unit for removing abnormal data points of the inverter includes: An alignment module is used for, for each inverter, using the normalized active power time series data after deleting the over-limit data in the input data of the day as the dependent variable data of the training data of the day, and aligning the independent variable data of the training data of the day and the dependent variable data of the training data of the day according to the timestamp of the measurement point generation time as the initial training data of the photovoltaic power prediction model; A fourth deletion module, used to determine and delete a point in the initial training data where the normalized active power value of the inverter is a first preset value as a shutdown fault point; The first marking module is used to use the normalized active power time series data of the inverter as the y value, initialize the data grouping mark, and divide the y value into a preset number of boxes according to the value size and mark the box number to obtain the box number; A first grouping module, used for grouping the initial training data according to the box numbers to obtain a first group; A sorting calculation module, used for sorting the x values ​​in the first group if the amount of data in the first group is greater than a preset amount, and calculating a first difference between two adjacent x values ​​in the first group; wherein the x value is a distance value from the normalized irradiance data and the normalized temperature data to the origin on a two-dimensional plane; A second marking module, configured to mark the first difference value to obtain a marking value if the first difference value is greater than a preset threshold; A marking and filling module, used for marking as empty if the first difference is less than or equal to a preset threshold, until all the difference values ​​in the first group are marked, and then filling all empty values ​​using the previous value method; A second grouping module is used to group the mark values ​​of the difference between two adjacent points in the first group by a preset grouping method to obtain a second group; wherein the preset grouping method is a grouping method that groups points with the same mark value into the same group; A first calculation module is used to calculate the amount of data in the second group, a second difference between the maximum value and the minimum value of the x value, and a third difference between the maximum value and the minimum value of the y value; a fifth deletion module, configured to mark and delete data points having a data volume in the second group greater than a preset number and a ratio of the third difference to the second difference greater than a preset number as data points having power restriction anomalies and communication anomalies; A second calculation module is used to scale the normalized irradiance time series data, the normalized temperature time series data, and the normalized active power time series data of the inverter within a preset range if the amount of remaining data is greater than a preset amount, and calculate the coordinates of the remaining data points in the three-dimensional polar coordinate system by using the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled normalized active power time series data of the inverter; A clustering elimination module is used to perform density clustering on the remaining data points using a density clustering algorithm through the scaled normalized irradiance time series data, the scaled normalized temperature time series data, and the scaled inverter normalized active power time series data, so as to eliminate discrete noise points of the inverter marked as a second preset value after density clustering; A clustering calculation module, used for clustering the first angle value and the second angle value of the data points in the three-dimensional polar coordinate system using a k-means algorithm, and calculating the data volume, angle mean and radius mean of each cluster; The sixth deletion module is used to delete the occluded abnormal points in the cluster whose angle mean is not the maximum, whose radius mean is less than the preset mean, and whose data volume is greater than the preset number, so as to complete the process of eliminating abnormal data points of the inverter.

9. A storage medium, characterized in that: The storage medium includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the data processing method of the photovoltaic power prediction model according to any one of claims 1 to 4.

10. An electronic device, characterized in that: It comprises a memory and one or more instructions, wherein the one or more instructions are stored in the memory and are configured to be executed by one or more processors to perform the data processing method of the photovoltaic power prediction model as described in any one of claims 1 to 4.

Images