Multi-source multi-attribute photovoltaic power prediction time series data set construction method

Abstract

The invention provides a multi-source multi-attribute photovoltaic power prediction time series data set construction method. The method comprises the following steps: acquiring urban meteorological data, wind cloud static meteorological satellite secondary cloud picture data and photovoltaic power station micro-meteorological data; traversing a wind cloud satellite secondary cloud picture with high photovoltaic power generation association degree, and realizing space registration by unifying longitude and latitude; extracting each attribute parameter of a pixel subjected to spatial registration at each corresponding time point in each secondary cloud picture file, and exporting corresponding wind cloud satellite secondary cloud picture time sequence data; carrying out abnormal value detection and processing on the multi-dimensional meteorological attribute time series data; performing time registration operation on the meteorological time sequence data to realize alignment and synchronization of multi-dimensional heterogeneous data; meteorological characteristic attribute parameters having great influence on photovoltaic power generation are screened, and photovoltaic power prediction multi-dimensional characteristic vectors under each timestamp are formed; and constructing a time sequence meteorological data set according to the target multi-dimensional feature vector, and completing normalization processing, thereby constructing a multi-source multi-attribute photovoltaic power prediction time sequence data set.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of photovoltaic power prediction, and in particular to a multi-source multi-attribute photovoltaic power prediction time series dataset construction method. BACKGROUND

[0002] The uncertainty and volatility of photovoltaic power generation have always been a serious challenge to the safe and stable operation of the distribution network. Accurate short-term prediction of photovoltaic power output and rapid warning of abnormal changes are key support technologies for distribution network dispatching, energy storage coordination and new energy consumption. Based on the secondary cloud data of Fengyun geostationary meteorological satellite and urban meteorological data, as well as the multi-source fusion of photovoltaic power station ground micro-meteorological observation data, the large-scale information of satellite cloud evolution can be captured from the spatial scale, and the high-frequency observation data of photovoltaic power station and urban meteorological data can be complemented, so as to significantly improve the short-term prediction accuracy and warning timeliness of photovoltaic power generation.

[0003] The short-term prediction technology of photovoltaic power generation not only can promote the assimilation of spatio-temporal multi-source data and short-term time series modeling research, but also has important theoretical value. It has clear engineering application value and can directly serve intelligent distribution network real-time dispatching and economic operation optimization, and has good industrialization prospect and platform promotion value. The current ground micro-meteorological station observation data can truly reflect the local meteorological situation of the photovoltaic power station, but only represents a small range of the monitoring point and cannot reflect the dynamics of the cloud several kilometers away. The spatial representativeness is weak. Although the urban meteorological data has wide coverage and long time series, it can reflect the weather change trend of the entire region, but the observation point is often far away from the photovoltaic power station and the environmental difference is obvious, so the micro-response is not sensitive. The satellite cloud map widely used in photovoltaic power generation power prediction often extracts the overall meteorological features of a regional cloud map, and the cloud map has wide spatial coverage, but the resolution is too low, and the area represented by one or several pixels is often wider than the area of the photovoltaic power station, so it is difficult to accurately use satellite cloud map data. The present application determines the meteorological attribute corresponding to the pixel of the Fengyun satellite secondary cloud map through the accurate longitude and latitude coordinates of the photovoltaic power station, so as to realize the accurate use of satellite cloud map data. On this basis, the present application fuses three types of meteorological data, i.e. the secondary cloud map of Fengyun satellite, the ground micro-meteorological station observation data and the urban meteorological data, realizes the multi-dimensional compensation of meteorological data in space-time-physical, and finally obtains a dataset which can not only reflect the macro weather trend, but also express the local meteorological response and accurately express the dynamic characteristics of the cloud layer, providing high spatio-temporal resolution, strong physical consistency and information complementary meteorological input feature support for photovoltaic power station power prediction model training. SUMMARY

[0004] The application provides a multi-source multi-attribute photovoltaic power prediction time series dataset construction method, and solves the problem that it is difficult to construct city meteorological data, photovoltaic power station micro-meteorological data and secondary cloud image data of a Fengyun stationary meteorological satellite into a time series dataset for high-precision photovoltaic power generation power prediction deep learning model training.

[0005] To achieve the above object, the application provides the following technical scheme.

[0006] A multi-source multi-attribute photovoltaic power prediction time series dataset construction method comprises the following steps.

[0007] Step S1, city meteorological data and secondary cloud image data of a Fengyun stationary meteorological satellite are acquired, and photovoltaic power station micro-meteorological data is collected.

[0008] Step S2, secondary cloud image data of a Fengyun satellite in a full disc projection form stored in a Fengyun satellite cloud Figure Two Level data disc coordinate is converted into geographical coordinates of longitude and latitude by longitude and latitude grid conversion, and is accurately positioned to a pixel above a photovoltaic power station according to a spatial resolution, so as to realize spatial registration;

[0009] Step S3, time encoding information of each NetCDF file is analyzed, each meteorological attribute parameter of a pixel after spatial registration of each secondary cloud image at each corresponding time point is extracted, and corresponding secondary cloud image time series data of a Fengyun satellite is exported;

[0010] Step S4, city meteorological data, photovoltaic power station micro-meteorological data and secondary cloud image time series data of a Fengyun satellite are subjected to abnormal value processing, so as to reduce the influence of abnormal values on a photovoltaic power generation power prediction model;

[0011] Step S5, secondary cloud image time series data of a Fengyun satellite, city meteorological time series data and photovoltaic power station micro-meteorological data are subjected to time registration operation, so as to realize alignment and synchronization of multi-dimensional heterogeneous data;

[0012] Step S6, data correlation analysis is adopted, meteorological attribute parameters with an influence on photovoltaic power generation greater than a set threshold value in secondary cloud image time series data of a Fengyun satellite, city meteorological data and photovoltaic power station micro-meteorological data are screened, and a multi-dimensional meteorological feature vector for photovoltaic power prediction at each time stamp is composed;

[0013] Step S7, a time series meteorological dataset is constructed from the multi-dimensional meteorological feature vector for photovoltaic power prediction, and normalization processing is completed, and then a multi-source multi-attribute photovoltaic power prediction time series dataset is constructed.

[0014] The application has the following advantages.

[0015] The method can take into account the continuous observation ability of satellite cloud maps and the high precision and time resolution of ground meteorological data, can improve the prediction accuracy of irradiation estimation and photovoltaic power, can enhance the stability and physical consistency of the model, can realize the coordinated perception from macro cloud field change to micro meteorological response, and can provide better technical support for photovoltaic power prediction. BRIEF DESCRIPTION OF DRAWINGS

[0016] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application, illustrate the preferred embodiments of the application and assist in the explanation of the application. In the drawings, the same reference numbers represent the same elements throughout the several views of the drawings:

[0017] Figure One is a flowchart of the present application;

[0018] Figure Two is a flowchart of the present application;

[0019] Figure Three is a schematic diagram of the Spearman correlation matrix obtained by the Spearman correlation analysis;

[0020] Figure Four is a schematic diagram of the Spearman correlation coefficient;

[0021] Figure Five is a schematic diagram of the present application; DETAILED DESCRIPTION

[0022] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. In addition, the technical features involved in each embodiment of the present application described below can be combined with each other as long as they do not conflict with each other. In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows.

[0023] It can be found from the background art that none of the existing photovoltaic power prediction methods simultaneously uses urban meteorological data, photovoltaic power station micro-meteorological data and Fengyun satellite secondary cloud map time series data. Generally, the coverage range of satellite cloud maps is too large, which leads to the problem that the real meteorological characteristics above the photovoltaic power station cannot be correctly represented for photovoltaic power prediction. In order to solve this problem, in a typical embodiment of the present application, a multi-source multi-attribute photovoltaic power prediction time series data set construction method is provided.

[0024] The application provides a multi-source multi-attribute photovoltaic power prediction time series dataset construction method, including: downloading city meteorological data and Fengyun stationary meteorological satellite secondary cloud image data, and collecting photovoltaic power station micro-meteorological data; converting the wind satellite cloud Figure Two level data disc coordinates stored in the form of a full disc projection into longitude and latitude geographical coordinates, and accurately positioning to the pixel above the photovoltaic power station according to the spatial resolution, so as to realize spatial registration; by traversing the Fengyun satellite secondary cloud image NetCDF file (Network Common Data Form) with high correlation degree of photovoltaic power generation, analyzing the time coding information of each file, extracting each attribute parameter of the pixel after spatial registration at each corresponding time point of each file, and exporting the corresponding Fengyun satellite secondary cloud image time series data; performing outlier processing on the city meteorological data, the photovoltaic power station micro-meteorological data, and the Fengyun satellite secondary cloud image time series data, so as to reduce the influence of outliers on the photovoltaic power generation power prediction model; performing time registration operation on the Fengyun satellite secondary cloud image time series data, the city meteorological time series data, and the photovoltaic power station micro-meteorological time series data, so as to realize alignment and synchronization of multi-dimensional heterogeneous data; using data correlation analysis, screening attribute parameters in the Fengyun satellite secondary cloud image time series data, the city meteorological data, and the photovoltaic power station micro-meteorological data which have an influence greater than a threshold value on photovoltaic power generation, to form a multi-dimensional feature vector of photovoltaic power prediction at each time stamp; constructing a time series meteorological dataset from the target multi-dimensional feature vector, and completing normalization processing, and then constructing a multi-source multi-attribute photovoltaic power prediction time series dataset.

[0025] Optionally, the city meteorological data can be obtained from local meteorological data published by meteorological bureaus in various places, or downloaded from a meteorological data website; the Fengyun satellite secondary cloud image data is obtained from a Fengyun remote sensing data service network; and the photovoltaic power station micro-meteorological data is obtained by real-time collection of data by a micro-meteorological station arranged in an open area of the photovoltaic power station.

[0026] Optionally, the Fengyun satellite secondary cloud image data is obtained from a Fengyun FY-4B stationary satellite, and L1 radiation data is inverted into L2 product data which can be directly applied by a national satellite meteorological center. The secondary cloud image data is six types of secondary cloud image data in the L2 product, including cloud coverage rate (CFR), cloud detection (CLM), cloud type (CLT), cloud top temperature (CTT), cloud top height (CTH), and clear sky radiation (CSR).

[0027] Optionally, the wind satellite cloud Figure TwoThe conversion of the level data disc coordinates into latitude and longitude geographical coordinates is for the five L2 products of CFR, CLM, CLT, CTT and CTH, and the CSR file directly gives latitude and longitude arrays. The spatial resolution of the six types of L2 product data pixels of the Fengyun FY-4B satellite is 4km, which can be positioned to the pixel above the photovoltaic power station by inputting the accurate latitude and longitude coordinates of the photovoltaic power station, so as to realize spatial registration.

[0028] Optionally, the time encoding information of each NetCDF file is parsed, which refers to 14-bit time parameters of year, month, day, hour, minute and second. If there is no missing data, each attribute parameter of the pixel after spatial registration at each corresponding time point of each file is extracted, and the corresponding Fengyun satellite secondary cloud image time series data is exported, and if there is missing data, an empty timestamp meteorological feature attribute vector is output as the time series data export.

[0029] Optionally, the isolated forest method is used to detect the abnormal values of urban meteorological data, photovoltaic power station micro-meteorological data and Fengyun satellite secondary cloud image time series data, and then the random sampling consistency algorithm is used to repair the missing data, thereby effectively improving the quality of the multi-dimensional meteorological attribute time series data of the three sources.

[0030] Optionally, the alignment of the multi-dimensional heterogeneous data is to align the urban meteorological data, the photovoltaic power station micro-meteorological data and the exported Fengyun satellite secondary cloud image time series data according to the 15-minute time series. For time series data with a time resolution greater than 15 minutes, a cubic spline interpolation method is used to obtain time series parameter values at intervals of 15 minutes, thereby realizing synchronization.

[0031] Optionally, the attribute parameter with an influence on photovoltaic power generation greater than a threshold value is obtained by screening each dimension of the Fengyun satellite secondary cloud image time series data, the urban meteorological time series data and the photovoltaic power station micro-meteorological time series data through correlation analysis, thereby forming a multi-dimensional feature vector of photovoltaic power prediction at each timestamp.

[0032] Optionally, the normalization processing is to solve the problem of non-uniformity of the dimensions of photovoltaic power generation and meteorological attributes, which avoids the gradient distortion of the photovoltaic power prediction model training caused by the imbalance of the dimensions, and thereby the training process converges slowly, oscillates or even diverges. The normalization processing adopts a linear normalization method:

[0033] ;

[0034] In the formula, x represents the actual value of a certain attribute parameter at the current time, x represents the actual value of the variable sequence, x represents the actual value of the variable sequence, x represents the actual value of the variable sequence, and x represents the normalized data of the variable. After data normalization, a multi-source multi-attribute photovoltaic power prediction time series data set is constructed from the target multi-dimensional feature vector.

[0035] According to another aspect of the present application, a multi-source multi-attribute photovoltaic power prediction time series dataset construction device is also provided, comprising a computing module, a storage module, a signal acquisition module, and a communication module.

[0036] Further, the computing module is configured to implement operations such as outlier detection, missing data repair, correlation calculation, data alignment and synchronization, etc.

[0037] Further, the storage module is configured to store city meteorological data, photovoltaic power station micro-meteorological data, and Fengyun satellite secondary cloud image time series data, as well as various algorithms used by the computing module.

[0038] Further, the signal acquisition module is configured to implement real-time acquisition and download of photovoltaic power station micro-meteorological data.

[0039] Further, the communication module is configured to download city meteorological data and Fengyun satellite secondary cloud image time series data.

[0040] More specifically, in an embodiment of the present application, a multi-source multi-attribute photovoltaic power prediction time series dataset construction method is provided. Figure One is a flowchart of the multi-source multi-attribute photovoltaic power prediction time series dataset construction method according to an embodiment of the present application, as shown in Figure One The method comprises the following steps:

[0041] Step S1, city meteorological data and Fengyun stationary meteorological satellite secondary cloud image data are obtained, and photovoltaic power station micro-meteorological data is acquired.

[0042] Step S2, by traversing the Fengyun satellite secondary cloud image NetCDF file (Network Common Data Form) with high correlation degree of photovoltaic power generation, the Fengyun satellite cloud image data stored in the form of full disc projection is converted from disc coordinates to geographical coordinates through latitude and longitude grid conversion, and the spatial resolution is accurately positioned to the pixel above the photovoltaic power station according to the spatial resolution, to realize spatial registration. Figure Two

[0043] Step S3, the time encoding information of each file is parsed, the attribute parameters of each pixel of the secondary cloud image at each corresponding time point after spatial registration are extracted, and the corresponding Fengyun satellite secondary cloud image time series data is exported.

[0044] Step S4, the city meteorological data, photovoltaic power station micro-meteorological data, and Fengyun satellite secondary cloud image time series data are subjected to outlier detection and processing, to reduce the influence of outliers on the photovoltaic power generation power prediction model.

[0045] ​Step S5, time registration operation is performed on the Fengyun satellite secondary cloud image time series data, urban meteorological time series data, and photovoltaic power station micro-meteorological time series data, so as to realize alignment and synchronization of multi-dimensional heterogeneous data.

[0046] Step S6, data correlation analysis is adopted to screen attribute parameters in the Fengyun satellite secondary cloud image time series data, urban meteorological data, and photovoltaic power station micro-meteorological data that have an influence on photovoltaic power generation greater than a threshold value, so as to form a multi-dimensional feature vector of photovoltaic power prediction under each time stamp.

[0047] Step S7, a time series meteorological data set is constructed from the target multi-dimensional feature vector, and normalization processing is completed, and then a multi-source multi-attribute photovoltaic power prediction time series data set is constructed.

[0048] Through the above steps, according to respective advantages of the urban meteorological data, the photovoltaic power station micro-meteorological data, and the Fengyun satellite secondary cloud image time series data for photovoltaic power prediction, the embodiment of the application can realize coordinated perception of a photovoltaic power prediction system from macro weather situation to micro irradiation change from the aspects of space-time scale and physical consistency, and provide more physically meaningful and generalizable data support for training and testing of a short-time power generation prediction model of a photovoltaic power station and deployment of a future photovoltaic prediction model.

[0049] Embodiment one:

[0050] In the embodiment of the application, the urban meteorological data can be obtained from local meteorological data published by meteorological bureaus in various places, or downloaded from meteorological data websites such as Datashareclub data network and the meteorological data of the National Meteorological Science Data Center; the photovoltaic power station micro-meteorological data is obtained by real-time collection of data by micro-meteorological stations arranged in open areas of the photovoltaic power station, and the micro-meteorological stations can deploy PT100 thermometers, capacitive hygrometers, three-cup or ultrasonic anemometers, thermoelectric pile type radiation meters, tipping bucket rain gauges, and barometric pressure sensors, etc. The Fengyun satellite secondary cloud image data is obtained from the Fengyun remote sensing data service network, and access permission needs to be obtained after being approved by the National Satellite Meteorological Center.

[0051] In the embodiment of the application, the Fengyun satellite secondary cloud image data is obtained from the Fengyun FY-4B geostationary satellite, and L1 radiation data is inverted into L2 product data that can be directly applied by the National Satellite Meteorological Center. The secondary cloud image data is six types of secondary cloud image data in the L2 product, including cloud coverage rate (CFR), cloud detection (CLM), cloud type (CLT), cloud top temperature (CTT), cloud top height (CTH), and clear sky radiation (CSR).

[0052] In the embodiment of the application, the Fengyun satellite cloud image data stored in the form of a full disc projection is converted into a grid through latitude and longitude grid conversion. Figure TwoThe conversion of the level data disc coordinates into the latitude and longitude geographical coordinates is for the five L2 products of CFR, CLM, CLT, CTT and CTH, and the CSR file directly gives the latitude and longitude array. The spatial resolution corresponding to the L2 product data pixels of the six categories of the Fengyun FY-4B satellite is 4km, which can be positioned to the pixel above the photovoltaic power station by inputting the accurate latitude and longitude coordinates of the photovoltaic power station, so as to realize spatial registration.

[0053] In the embodiment of the application, by traversing the Fengyun satellite secondary cloud map NetCDF file (Network Common Data Form) with high correlation degree of photovoltaic power generation, the time coding information of each file is parsed, the spatially registered pixel attribute parameters at each corresponding time point of each file are extracted, and the corresponding Fengyun satellite secondary cloud map time series data is exported. Figure Two The flow chart of the secondary cloud map multi-dimensional attribute parameter extraction method according to the embodiment of the application is used to store all secondary cloud map NetCDF format data in the same file directory, unify the data storage path, and then construct the Fengyun satellite secondary cloud map time series data according to the model data set specification format, which specifically includes the following steps:

[0054] Step S201, setting the target latitude and longitude coordinates of the photovoltaic power station, the time interval, the NetCDF format data storage path and the storage location;

[0055] Step S202, reading the unprocessed NetCDF file, if the read Fengyun FY-4B satellite official secondary cloud map product data is legal data meeting the requirements in terms of naming specification and type, the timestamp and product type of the cloud map are extracted;

[0056] Step 203, judging whether it is a CSR cloud map, if yes, jumping to step 204. If not, projecting the target latitude and longitude to the full disc coordinate system, obtaining the pixel index, and querying the meteorological attribute parameters corresponding to the target pixel (i.e. the photovoltaic power station position) according to the pixel index, and then executing step 205;

[0057] Step 204, obtaining the nearest neighbor index according to the target latitude and longitude, and querying the meteorological attribute parameters of the target pixel (i.e. the photovoltaic power station position) according to the nearest neighbor index, and then executing step 205;

[0058] Step 205, storing the meteorological attribute parameters of the corresponding pixel (i.e. the photovoltaic power station position) at the target latitude and longitude obtained by the current product type data into the dictionary with the timestamp as the primary key;

[0059] Step 206, judge whether the current data storage path still exists unread NetCDF files. If there are unread NetCDF files, jump to step 202, and repeat the above steps; if there is no unread NetCDF file, convert the dictionary into a data table, and sort the extraction results according to the timestamp. In order to facilitate multi-source data time alignment and data synchronization operation, missing data rows are inserted according to the time interval here;

[0060] Step 207, output the data table and store it to the storage location set in step 201.

[0061] In the embodiment of the present application, in order to solve the problem that the original urban meteorological data, photovoltaic power station micro-meteorological data, and Fengyun satellite secondary cloud time series data are prone to produce abnormal values or missing values in the process of collection due to various interference and noise, the isolated forest method is used to detect abnormal values and repair missing data. First, an isolated forest is constructed, second, meteorological data points are input into the isolated forest, and an anomaly score is calculated to reflect the abnormality degree of each data point, then abnormal value determination is completed according to the comparison of the anomaly score and the set threshold, finally, the random sample consensus algorithm (RANSAC) is used to repair missing data. After the abnormal value detection and missing value repair operation, the quality of the urban meteorological data, photovoltaic power station micro-meteorological data, and Fengyun satellite secondary cloud time series data can be effectively improved, providing a reliable data basis for subsequent photovoltaic power generation power prediction model training and testing.

[0062] In the embodiment of the present application, the alignment of the multi-dimensional heterogeneous data is to align the urban meteorological data, photovoltaic power station micro-meteorological data, and derived Fengyun satellite secondary cloud time series data according to 15-minute time series. For time series data with a time resolution greater than 15 minutes, a cubic spline interpolation method is used to obtain time series parameter values with an interval of 15 minutes, and then synchronization is realized. Compared with linear interpolation, the cubic spline interpolation method can better approximate the meteorological original data, and the generated interpolation curve is smooth and continuous. For each interval of meteorological data, a cubic polynomial is fitted, and its general form is:

[0063] ;

[0064] wherein, and are the known observation time points of the adjacent two meteorological data, is the specific time point that needs to be interpolated, and the output result is the interpolated meteorological data value, and the interpolation function is the coefficient , , , By solving the three bending moment equation sets.

[0065] In the embodiment of the present application, the attribute parameter with an influence on photovoltaic power generation greater than a threshold value is obtained by performing correlation analysis on the wind cloud satellite secondary cloud image time series data, urban meteorological time series data and each dimension attribute parameter of photovoltaic power station micro-meteorological time series data, and then forming a multi-dimensional feature vector of photovoltaic power prediction under each timestamp. The correlation analysis adopts the Spearman method, and assumes that the time series data of photovoltaic power generation is , is the total number of time series samples, is the power value at the moment. The dimensional meteorological attribute time series parameter is , , indicates the observation value of the dimensional meteorological attribute parameter at the moment. Then, for each meteorological attribute feature , first, the rank conversion is performed, and the numbers of and are replaced by the ranked ranks, denoted as and

[0066] ; then the correlation coefficient is calculated:

[0067] ; and are the rank means of the photovoltaic power generation power and the meteorological feature, respectively; finally, the features passing the significance test are retained, and the meteorological feature attributes with high values are taken according to the size of . The correlation analysis process quantifies the correlation between the meteorological attribute features and the photovoltaic power, and selects the meteorological feature attributes with high influence on the photovoltaic power, and reduces the dimension of the photovoltaic prediction time series data set for training on the basis of ensuring the prediction model accuracy.

[0068] Figure Three ​​The Spearman correlation matrix shown indicates a strong positive correlation between irradiance and photovoltaic (PV) power in micro-meteorological data for PV power plants. The correlation is slightly lower for irradiance in urban meteorological data, reflecting the continued value of large-scale meteorological data for power prediction, although its accuracy is limited. Temperature in micro-meteorological data for PV power plants shows a moderate positive correlation with PV power, while the correlation is relatively stronger in urban meteorological data, indicating a greater impact of temperature on output power. Wind speed in micro-meteorological data for PV power plants shows a certain positive correlation with PV power, while the correlation is weaker in urban meteorological data, suggesting a relatively minor overall impact of wind speed on power. Air pressure in micro-meteorological data for PV power plants shows a weak positive correlation with PV power, and the correlation is also very weak in urban meteorological data, indicating a limited direct impact of air pressure on PV power. Humidity in micro-meteorological data for PV power plants shows a negative correlation with PV power, as does precipitation in both urban and rural meteorological data, suggesting a negative impact of precipitation on PV power. Humidity in urban meteorological data is also negatively correlated with photovoltaic power, and humidity often accompanies cloud movement, potentially affecting output power indirectly by influencing irradiance. Based on the above analysis results, meteorological attribute parameters of urban meteorological data and meteorological attribute parameters of photovoltaic power plant micrometeorological data with absolute values ​​of correlation calculation results with photovoltaic power greater than a set threshold (the threshold set empirically).

[0069] Furthermore, Spearman correlation analysis was conducted on the meteorological attribute parameters extracted from six types of FY-4B satellite secondary cloud images and the measured photovoltaic power to assess the direct explanatory power of each cloud image parameter for power variations. The analyzed meteorological attribute parameters included 41 meteorological attribute data parameters from six types of secondary cloud images: cloud cover rate (CFR), cloud detection level (CLM), cloud type (CLT), cloud top temperature (CTT), cloud top height (CTH), and clear-sky radiation (CSR). Preliminary and complete analysis results are as follows: Figure Four As shown, meteorological attribute parameters whose absolute value of the correlation calculation result with photovoltaic power is greater than the set threshold (the threshold set according to experience) can be selected based on the calculation results.

[0070] In this embodiment, the normalization process is used to address the inconsistency between the dimensions of photovoltaic power generation and various meteorological attributes, preventing gradient distortion during photovoltaic power generation prediction model training due to dimensional imbalance, which in turn leads to slow convergence, oscillations, or even divergence in the training process. The normalization process employs a linear normalization method:

[0071] ;

[0072] In the formula, This represents the actual value of a certain attribute parameter at the current moment. The actual values ​​of the variable sequence. represents the data of the variable after normalization processing. After data normalization, a multi-source multi-attribute photovoltaic power prediction time series data set is constructed by the target multi-dimensional feature vector.

[0073] Embodiment two:

[0074] In the embodiments of the present application, the multi-source multi-attribute photovoltaic power prediction time series data set construction device comprises a computing module, a storage module, a signal acquisition module, and a communication module.

[0075] In the embodiments of the present application, the computing module is configured to implement operations such as anomaly value detection, missing data repair, correlation calculation, data space configuration, and time alignment and synchronization.

[0076] In the embodiments of the present application, the storage module is configured to store city meteorological data, photovoltaic power station micro-meteorological data, and Fengyun satellite secondary cloud image time series data, as well as various algorithms and data operation instructions used by the computing module.

[0077] In the embodiments of the present application, the signal acquisition module is configured to implement real-time acquisition of photovoltaic power station micro-meteorological data.

[0078] In the embodiments of the present application, the communication module is configured to download city meteorological data and Fengyun satellite secondary cloud image time series data.

[0079] Figure Five The multi-source multi-attribute photovoltaic power prediction time series data set construction device according to the embodiments of the present application is shown in FIG. 1. A multi-source multi-attribute photovoltaic power prediction time series data set construction device comprises a computing module, a storage module, a signal acquisition module, and a communication module, and is configured to implement the method and steps. Figure Five

[0080] The computing module is configured to implement operations such as anomaly value detection, missing data repair, correlation calculation, data alignment and synchronization. The storage module is configured to store city meteorological data, photovoltaic power station micro-meteorological data, and Fengyun satellite secondary cloud image time series data, as well as various algorithms used by the computing module. The signal acquisition module is configured to implement real-time acquisition and download of photovoltaic power station micro-meteorological data. The communication module is configured to download city meteorological data and Fengyun satellite secondary cloud image time series data.​

Claims

1. A method for constructing a multi-source multi-attribute photovoltaic power prediction time series dataset, characterized in that, The method comprises the following steps: Step S1, obtaining city meteorological data and FY-2 satellite second-level cloud data, collecting photovoltaic power station micro-meteorological data; Step S2, traversing the FY-2 satellite second-level cloud NetCDF file with high correlation degree of photovoltaic power generation, converting the FY-2 satellite cloud second-level data disc coordinates stored in the form of full disc projection into geographical coordinates through latitude and longitude grid conversion, and accurately positioning the pixel over the photovoltaic power station according to the spatial resolution to realize spatial registration; Step S3, analyzing the time encoding information of each NetCDF file, extracting each meteorological attribute parameter of the pixel at each corresponding time point after spatial registration of each second-level cloud, and exporting the corresponding FY-2 satellite second-level cloud time series data; Step S4, performing outlier processing on the city meteorological data, photovoltaic power station micro-meteorological data, and FY-2 satellite second-level cloud time series data to reduce the influence of outliers on the photovoltaic power generation power prediction model; Step S5, performing time registration operation on the FY-2 satellite second-level cloud time series data, city meteorological time series data, and photovoltaic power station micro-meteorological data to realize alignment and synchronization of multi-dimensional heterogeneous data; Step S6, using data correlation analysis to filter the meteorological attribute parameters in the FY-2 satellite second-level cloud time series data, city meteorological data, and photovoltaic power station micro-meteorological data that have greater influence on photovoltaic power generation than a set threshold to form a photovoltaic power prediction multi-dimensional meteorological feature vector at each time stamp; Step S7, constructing a time series meteorological data set from the photovoltaic power prediction multi-dimensional meteorological feature vector and completing normalization processing to further construct a multi-source multi-attribute photovoltaic power prediction time series data set.

2. The method of claim 1, wherein, In step S1, the city meteorological data is obtained from the local meteorological data published by the local meteorological bureau or downloaded from the public meteorological data website; the FY-2 satellite second-level cloud data is obtained from the FY-2 remote sensing data service network; and the photovoltaic power station micro-meteorological data is obtained by real-time collection of data from the micro-meteorological station arranged in the open area of the photovoltaic power station.

3. The method of claim 2, wherein, The FY-2 satellite second-level cloud data is obtained from the FY-4B geostationary satellite, and the second-level cloud data includes six types of second-level cloud data, i.e., cloud coverage CFR, cloud detection CLM, cloud type CLT, cloud top temperature CTT, cloud top height CTH, and clear sky radiation CSR.

4. The method of claim 1, wherein, In step S2, the FY-2 satellite cloud second-level data disc coordinates stored in the form of full disc projection are converted into geographical coordinates through latitude and longitude grid conversion, which is for CFR, CLM, CLT, CTT, and CTH, and the CSR file directly gives latitude and longitude arrays. The spatial resolution corresponding to the pixel of the six types of FY-4B satellite second-level cloud data is 4km, and the pixel over the photovoltaic power station is positioned through input of the photovoltaic power station latitude and longitude coordinates to realize spatial registration.

5. The method of claim 1, wherein, In step S3, the time encoding information of each NetCDF file refers to 14-bit time parameters of year, month, day, hour, minute, and second. If there is no missing data, the attribute parameters of each pixel of each corresponding time point of the secondary cloud image after spatial registration are extracted, and the corresponding Fengyun satellite secondary cloud image time series data is exported. If there is missing data, an empty timestamp meteorological feature attribute vector is output as the time series data export.

6. The method of claim 1, wherein, In step S4, the isolated forest method is used to detect outliers, and then the random sample consensus algorithm is used to repair the missing data.

7. The method of claim 1, wherein, In step S5, the alignment of the multi-dimensional heterogeneous data is to align the urban meteorological data, the photovoltaic power station micro-meteorological data, and the exported Fengyun satellite secondary cloud image time series data according to 15-minute time series. For time series data with a time resolution greater than 15 minutes, a cubic spline interpolation method is used to obtain time series parameter values at intervals of 15 minutes, thereby realizing synchronization. 8.The method of claim 1, wherein, In step S6, the threshold value is an empirical value set between [0, 1] according to the results of the Spearman correlation analysis. Each value of the Spearman correlation matrix is between [-1, 1], and when the absolute value is greater than the set threshold value, the corresponding meteorological attribute parameter is considered to have a significant impact on photovoltaic power generation. The Spearman correlation analysis is used to screen each dimension of the Fengyun satellite secondary cloud image time series data, the urban meteorological time series data, and the photovoltaic power station micro-meteorological time series data, thereby forming a multi-dimensional meteorological feature vector for photovoltaic power prediction at each timestamp. 9.The method of claim 1, wherein, In step S7, the normalization processing adopts a linear normalization method, ； In the formula, represents the actual value of a certain attribute parameter at the current time, is the actual value of the variable sequence, represents the data after normalization processing of the variable.

10. The method of claim 8, wherein, The correlation analysis adopts the Spearman method, assuming that the time series data of photovoltaic power generation power is , is the total number of time series samples, is the power value at the moment; the dimensional meteorological attribute time series parameter is , , indicates the observation value of the dimensional meteorological attribute parameter at the moment; for each meteorological attribute feature , first, rank conversion is performed, and the numbers of and are replaced by the ranked ranks, denoted as and ; then, the correlation coefficient is calculated: ； wherein, and are the rank means of the photovoltaic power and the first meteorological feature attribute, respectively; finally, the features that pass the significance test are retained, and the meteorological feature attribute with the highest value is selected according to the size of .

Images