Photovoltaic cluster climbing event prediction method based on satellite cloud picture

Abstract

The invention relates to a photovoltaic cluster climbing event prediction method based on a satellite cloud picture, in particular to the technical field of photovoltaic power generation power climbing prediction, and the method comprises the steps: obtaining corresponding photovoltaic power data, weather forecast data and satellite cloud picture data of a photovoltaic power station in a photovoltaic cluster within a preset collection time, the preset acquisition time is a time period before the prediction time of the photovoltaic cluster climbing event; determining a forecast meteorological factor corresponding to the photovoltaic cluster from the weather forecast data according to the association strength between the weather forecast data and the photovoltaic power of the photovoltaic power station; determining the extreme weather of the photovoltaic cluster corresponding to the prediction time according to the satellite remote sensing value of the photovoltaic cluster geographic area corresponding to the photovoltaic cluster in the satellite cloud picture data; determining the predicted power of the photovoltaic cluster corresponding to the prediction time based on the extreme weather and the forecast meteorological factor; and determining a photovoltaic cluster climbing event corresponding to the prediction time according to the climbing characteristic quantity corresponding to the prediction power.

Description

A method for predicting the ramp-up events of photovoltaic clusters based on satellite cloud images. Technical Field

[0001] This application relates to the field of photovoltaic power generation ramp-up prediction technology, and in particular to a method for predicting photovoltaic cluster ramp-up events based on satellite cloud images. Background Technology

[0002] Renewable energy sources, such as solar energy, play a crucial role in addressing energy and environmental issues. With the continuous maturation of technology and the relative reduction in power generation costs, the utilization rate of solar energy has grown exponentially over the past decade. Compared to traditional energy sources such as fossil fuels and nuclear power, renewable energy has poorer stability and controllability, making it difficult to meet the stable operation requirements of the power grid and achieve high penetration rates within the grid. Specifically, the phenomenon of significant fluctuations in solar power generation within a short period, known as photovoltaic cluster ramp-up events, poses a significant threat to the security and stability of the power system.

[0003] Currently, related technologies typically establish a mapping relationship between relevant factors and photovoltaic cluster ramp-up events, directly predicting these events. However, since the relationship between meteorological, geographical, and temporal factors and ramp-up events is nonlinear, high-dimensional, and dynamically coupled, this direct prediction method is difficult to accurately establish a mapping relationship, easily overlooking some extreme cases and leading to missed reporting of ramp-up events. Summary of the Invention

[0004] In view of this, this application provides a method, device, storage medium and electronic device for predicting photovoltaic cluster ramping events based on satellite cloud images. The main purpose is to improve the technical problem in related technologies where direct prediction methods are difficult to accurately establish mapping relationships, easily overlook some extreme cases, and cause missed reporting of ramping events.

[0005] Firstly, this application provides a method for predicting photovoltaic (PV) cluster ramp-up events based on satellite cloud images. The method includes: acquiring PV power data, weather forecast data, and satellite cloud image data corresponding to PV power plants in a PV cluster within a preset acquisition time, wherein the preset acquisition time is the period prior to the prediction time of the PV cluster ramp-up event; determining the forecast meteorological factors corresponding to the PV cluster from the weather forecast data based on the correlation strength between the weather forecast data and the PV power of the PV power plants; determining the extreme weather of the PV cluster corresponding to the prediction time based on the satellite remote sensing values ​​of the geographical area of ​​the PV cluster corresponding to the PV cluster in the satellite cloud image data; determining the predicted power of the PV cluster corresponding to the prediction time based on the extreme weather and the forecast meteorological factors; and determining the PV cluster ramp-up event corresponding to the prediction time based on the ramp-up characteristic quantity corresponding to the predicted power.

[0006] Secondly, this application provides a photovoltaic cluster ramping event prediction device based on satellite cloud images. The device includes: an acquisition module configured to acquire photovoltaic power data, weather forecast data, and satellite cloud image data corresponding to the photovoltaic power station in the photovoltaic cluster within a preset acquisition time, wherein the preset acquisition time is the time period before the prediction time of the photovoltaic cluster ramping event; and a determination module configured to determine the forecast meteorological factors corresponding to the photovoltaic cluster from the weather forecast data based on the correlation strength between the weather forecast data and the photovoltaic power of the photovoltaic power station; determine the extreme weather of the photovoltaic cluster corresponding to the prediction time based on the satellite remote sensing values ​​of the geographical area of ​​the photovoltaic cluster corresponding to the photovoltaic cluster in the satellite cloud image data; determine the predicted power of the photovoltaic cluster corresponding to the prediction time based on the extreme weather and the forecast meteorological factors; and determine the photovoltaic cluster ramping event corresponding to the prediction time based on the ramping characteristic quantity corresponding to the predicted power.

[0007] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method of the first aspect.

[0008] Fourthly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the method of the first aspect.

[0009] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method of the first aspect.

[0010] Compared with existing technologies, this application acquires photovoltaic power data, weather forecast data, and satellite cloud image data corresponding to photovoltaic power plants in a photovoltaic cluster within a preset acquisition time period. The preset acquisition time is the period before the predicted time of the photovoltaic cluster ramp-up event. Based on the correlation strength between the weather forecast data and the photovoltaic power of the photovoltaic power plant, the forecast meteorological factors corresponding to the photovoltaic cluster are determined from the weather forecast data. Based on the satellite remote sensing values ​​of the geographical area of ​​the photovoltaic cluster corresponding to the photovoltaic cluster in the satellite cloud image data, the extreme weather of the photovoltaic cluster corresponding to the predicted time is determined. Based on the extreme weather and the forecast meteorological factors, the predicted power of the photovoltaic cluster corresponding to the predicted time is determined. Based on the ramp-up characteristic quantity corresponding to the predicted power, the ramp-up event of the photovoltaic cluster corresponding to the predicted time is determined.

[0011] By applying the technical solution of this application, this application first collects photovoltaic power data, weather forecast data, and satellite cloud image data of each photovoltaic power station in the photovoltaic cluster within a preset collection time before the prediction time. By detecting the correlation strength between weather forecast data and photovoltaic power, the forecast meteorological factors are determined to screen and fuse the meteorological data corresponding to the photovoltaic cluster, effectively capturing the strong correlation between meteorological changes and power fluctuations. Then, image recognition is performed on the satellite cloud image based on satellite remote sensing values ​​of the satellite cloud image data. Based on the predicted extreme weather and the screened forecast meteorological factors, the predicted power of the photovoltaic cluster is predicted. Finally, based on the predicted power and the ramp-up characteristics of the photovoltaic cluster, the ramp-up event of the photovoltaic cluster corresponding to the prediction time is determined. By predicting photovoltaic power and the ramp-up event of the photovoltaic cluster under the consideration of extreme weather, the risk of sudden changes in photovoltaic power caused by extreme weather is effectively reduced, the underreporting of ramp-up events is reduced, and key decision support is provided for grid safety dispatch.

[0012] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0013] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 shows a flowchart of a photovoltaic cluster ramping event prediction method based on satellite cloud imagery provided in an embodiment of this application; Figure 2 shows a flowchart of an example provided in an embodiment of this application; Figure 3 shows a structural schematic diagram of a photovoltaic cluster ramping event prediction device based on satellite cloud imagery provided in an embodiment of this application. Detailed Implementation

[0016] To better understand the above-mentioned objectives, features, and advantages of this application, the solution of this application will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0017] In related technologies, although various measures have been proposed to mitigate the negative impact of photovoltaic cluster ramp-up events, there is still a lack of low-cost solutions in practice. The most effective approach remains to improve the prediction accuracy of photovoltaic cluster ramp-up events and provide early warnings of their occurrence.

[0018] Currently, the prediction methods for photovoltaic (PV) cluster ramp-up events can be broadly categorized into indirect and direct prediction. Indirect prediction first predicts power generation and then the ramp-up event; direct prediction, on the other hand, does not predict power but directly establishes a mapping relationship between relevant factors and the ramp-up event. This approach has lower prediction accuracy for PV cluster ramp-up events. This embodiment focuses on indirect prediction methods. Although existing indirect ramp-up prediction methods have achieved relatively superior results in their respective experimental scenarios, most prediction methods tend to ignore extreme samples (such as extreme weather), resulting in the loss of some ramp-up information and leading to missed ramp-up events. Given the rapid development of PV power generation, accurately predicting PV cluster ramp-up events has become one of the urgent problems to be solved.

[0019] To address the technical problem in related technologies where direct prediction methods struggle to accurately establish mapping relationships and easily overlook extreme cases, leading to missed reporting of ramp-up events, this embodiment provides a photovoltaic cluster ramp-up event prediction method based on satellite cloud images. As shown in Figure 1, the method includes: Step 101, acquiring photovoltaic power data, weather forecast data, and satellite cloud image data corresponding to the photovoltaic power stations in the photovoltaic cluster within a preset acquisition time. The preset acquisition time is the period preceding the prediction time of the photovoltaic cluster ramp-up event.

[0020] The photovoltaic cluster can be a collection of multiple photovoltaic power stations that are geographically close, have the same grid connection level, and have a strong correlation in output. The prediction time can be a future time period in which it is necessary to determine whether a climbing event will occur in the photovoltaic cluster, such as T+4h or T+30min, where T is the prediction start time. The preset collection time can be a historical data collection window before the prediction time, such as T-24h or T-60min, and can be collected at collection intervals (such as 15min). The photovoltaic power data can be the active power sequence reported by each photovoltaic power station in the photovoltaic cluster. The weather forecast data can be numerical weather forecast data from the Numerical Weather Prediction (NWP) model, i.e., NWP data, which can include data such as total irradiance, direct / scattered components, cloud cover (low / medium / high clouds), cloud water path, temperature, humidity, and wind speed. The satellite cloud image data can be remote sensing images from meteorological satellites, which can include data such as visible light, infrared, and water vapor bands.

[0021] In some embodiments, the prediction time of the photovoltaic cluster to be predicted can be determined first, and then the preset collection time corresponding to the prediction time can be determined. For example, if the current time is the prediction start time, the prediction is to predict the photovoltaic cluster ramping event within the next 4 hours corresponding to the current time. Photovoltaic power data, numerical weather forecast data and high spatiotemporal resolution satellite cloud image data of each photovoltaic power station within the past 24 hours corresponding to the current time can be collected to support subsequent meteorological factor screening, extreme weather identification and ramping trend prediction.

[0022] Step 102: Based on the correlation strength between weather forecast data and photovoltaic power of photovoltaic power station, determine the forecast meteorological factors corresponding to photovoltaic cluster from the weather forecast data.

[0023] Among them, the forecast meteorological factors can be key meteorological variables that can be selected from meteorological variables in weather forecast data and used to predict photovoltaic output, such as total irradiance (GHI), direct irradiance, and diffuse irradiance.

[0024] In some embodiments, forecast meteorological factors with high correlation to photovoltaic power can be automatically selected based on different regions (such as plateaus and coastal areas). The correlation strength between weather forecast data and photovoltaic power can be re-analyzed periodically to adapt to long-term trends such as seasonal changes or power plant aging. For example, the correlation strength between each meteorological variable in the weather forecast data and the photovoltaic power sequence can be calculated based on Pearson correlation coefficient, Spearman rank correlation, mutual information, etc. According to the correlation strength ranking, meteorological variables with correlation strength higher than a preset threshold or ranked in the top N are selected as forecast meteorological factors corresponding to the photovoltaic cluster, forming a set of forecast meteorological factors.

[0025] In this way, the correlation between weather forecast data and photovoltaic power can be quantified, so as to identify and screen key forecast meteorological factors that have a significant impact on changes in photovoltaic output from multiple meteorological variables in the weather forecast data, and use them as core input features for subsequent power prediction and early warning of ramp-up events.

[0026] Step 103: Based on the satellite remote sensing values ​​of the geographical area of ​​the photovoltaic cluster corresponding to the photovoltaic cluster in the satellite cloud image data, determine the extreme weather of the photovoltaic cluster corresponding to the prediction time.

[0027] In some embodiments, the geographical region (cluster area, target area) of the photovoltaic cluster can be determined first in each satellite cloud image based on the location information of the photovoltaic cluster. Then, based on the satellite remote sensing values ​​corresponding to the geographical distribution area of ​​the photovoltaic cluster in the satellite cloud image data, such as brightness temperature, albedo, cloud optical thickness, etc., cloud features are extracted and weather is identified. The shape, temperature, movement trend and development status of the cloud cluster are identified, and it is predicted whether the photovoltaic cluster will encounter extreme weather events within the predicted time. Extreme weather may include meteorological phenomena that may cause a sudden drop in photovoltaic output, such as dense cloud cover, strong convective cloud system passage, sandstorms, and haze.

[0028] Step 104: Based on extreme weather and forecast meteorological factors, determine the predicted power of the photovoltaic cluster corresponding to the forecast time.

[0029] In some embodiments, a power prediction model can be constructed. This model integrates extreme weather prediction results with selected key forecast meteorological factors to generate a predicted power output sequence of the photovoltaic cluster at the predicted time, thereby obtaining the predicted power of the photovoltaic cluster. This is used to reflect the comprehensive impact of extreme weather disturbances, such as a nonlinear drop in photovoltaic output under cloud cover. By identifying extreme weather and combining it with forecast meteorological factors to predict photovoltaic power, prediction bias can be effectively corrected, and overall prediction robustness can be improved. This allows for the prediction of photovoltaic power and photovoltaic cluster ramp-up events under extreme weather conditions, effectively reducing the risk of sudden changes in photovoltaic power caused by extreme weather, reducing the underreporting of ramp-up events, and providing key decision support for grid safety dispatch.

[0030] Step 105: Determine the photovoltaic cluster ramping event corresponding to the prediction time based on the ramping characteristic quantity corresponding to the predicted power.

[0031] In some embodiments, based on the calculation rules of the ramp-up characteristic quantity corresponding to the predicted power within the prediction time, it is possible to identify whether a significant increase or decrease in photovoltaic output occurs within the prediction time, thereby determining the photovoltaic cluster ramp-up event and guiding different response strategies such as energy storage charging and discharging and thermal power regulation. The ramp-up characteristic quantity can be the power change rate between adjacent moments or adjacent time periods. If the ramp-up characteristic quantity exceeds a preset threshold, it can be determined that a photovoltaic cluster ramp-up event has occurred within the prediction time range. Correspondingly, the probability of the ramp-up event can be output based on the ramp-up characteristic quantity corresponding to the predicted power, achieving risk level early warning.

[0032] Compared with related technologies, this embodiment first collects photovoltaic power data, weather forecast data, and satellite cloud image data of each photovoltaic power station in the photovoltaic cluster within a preset collection time before the prediction time. By detecting the correlation strength between weather forecast data and photovoltaic power, forecast meteorological factors are determined to screen and fuse the meteorological data corresponding to the photovoltaic cluster, effectively capturing the strong correlation between meteorological changes and power fluctuations. Then, image recognition is performed on the satellite cloud image based on satellite remote sensing values ​​of the satellite cloud image data. Based on the predicted extreme weather and the screened forecast meteorological factors, the predicted power of the photovoltaic cluster is predicted. Finally, based on the predicted power and the ramp-up characteristics of the photovoltaic cluster, the ramp-up event of the photovoltaic cluster corresponding to the prediction time is determined. By predicting photovoltaic power and the ramp-up event of the photovoltaic cluster under the consideration of extreme weather, the risk of sudden changes in photovoltaic power caused by extreme weather is effectively reduced, the underreporting of ramp-up events is reduced, and key decision support is provided for grid safety dispatch.

[0033] Furthermore, as a refinement and extension of the specific implementation of the above embodiments, in order to fully illustrate the implementation of this embodiment, step 102 may optionally include: using the Pearson correlation coefficient to determine the correlation strength between weather forecast data and photovoltaic power of photovoltaic power plants, and screening out forecast meteorological factors from the weather forecast data.

[0034] In some embodiments, since not all meteorological variables in the NWP data are strongly correlated with photovoltaic power, inputting all meteorological factors into the power prediction model may cause data redundancy. Therefore, the Pearson correlation coefficient can be used. We select meteorological variables that are strongly correlated with photovoltaic power, and their calculation formula can be expressed as: In the formula, and These represent the average values ​​of power output curve A of a certain power station and power output curve B of another power station, respectively.

[0035] Correspondingly, a preset number of forecast meteorological factors can be selected based on the Pearson correlation coefficient to form a forecast meteorological factor set M.

[0036] Optionally, before step 103, the method of this embodiment may further include: acquiring historical cluster data of the photovoltaic cluster, including historical photovoltaic power data, historical weather forecast data, and historical satellite cloud image data; dividing the historical cluster data into a training set and a test set corresponding to a preset meteorological prediction model according to a preset division ratio; using weather tags to mark extreme weather pixel areas and non-extreme weather pixel areas of the historical satellite cloud image data in the training set, the weather tags including extreme weather tags and non-extreme weather tags; using a multi-scale convolutional feature fusion model to train the model on the training set, and using the trained multi-scale convolutional feature fusion model as the preset meteorological prediction model, the preset meteorological prediction model is used to identify extreme weather on the test set, the input of the multi-scale convolutional feature fusion model is satellite remote sensing data, and the output is the weather tags of the photovoltaic cluster.

[0037] For example, the historical photovoltaic power data (historical output data) of n photovoltaic power plants within the cluster area corresponding to a photovoltaic cluster can be represented as: In the formula, This represents the historical power output of photovoltaic power station i from time t0 to time t0+T. The historical output of photovoltaic power station i at time t (t ranges from t0 to t0+T) can be represented by the photovoltaic power data of the most recent 3 years, with a time interval of 15 minutes. T is the total number of historical output samples, T=365×24×60 / 15×3=105120.

[0038] In some embodiments, NWP can be applied The historical weather forecast data (historical NWP data) for time t within the cluster area may include, but is not limited to, forecast irradiance, forecast temperature, forecast wind speed, forecast wind direction, and forecast humidity, where t∈(t0, t0+T). Correspondingly, historical satellite cloud image data SL for the cluster area can be obtained from the satellite remote sensing data service network, where SL(t) represents the satellite image of the target area at time t, where t∈(t0, t0+T). It should be noted that the historical photovoltaic power data, historical NWP data, and historical satellite image data have the same time span and time resolution.

[0039] In some embodiments, weather labels may include extreme weather labels and non-extreme weather labels, such as dust storm labels and non-dust storm labels; the preset meteorological prediction model may include a multi-scale convolutional feature fusion (MSCFF) model. For example, MSCFF can be used to train the model on the training set. The input of MSCFF can be the multi-time, multi-channel satellite remote sensing numerical matrix C of each pixel, and the output of MSCFF is the dust storm label Label[0,1]. Then, the trained model is used to identify whether extreme weather has occurred in the test set.

[0040] In some embodiments, all datasets corresponding to historical cluster data can be divided into a training set Data_train (t∈(t0, t0+Time_train)) and a test set Data_test (t∈(t0+Time_train, t0+T)) in an 8:2 ratio. For a preset collection time (training set period) t∈(t0, t0+T1) corresponding to the training set of the preset weather prediction model, satellite remote sensing values ​​of the area photographed during that period are obtained using satellite cloud image data. For areas experiencing sandstorm weather, the sandstorm weather label of the corresponding pixel area in the satellite cloud image data can be set to 1, and the label of areas without sandstorm weather can be set to 0.

[0041] Optionally, step 103 may specifically include: using a preset meteorological prediction model, based on the satellite remote sensing values ​​of pixels in the satellite cloud image data at different acquisition times and in different channels, to obtain the weather tag of the photovoltaic cluster corresponding to the prediction time, and the weather tag is used to determine the extreme weather of the photovoltaic cluster.

[0042] For example, for clustered areas experiencing extreme dust storms, the satellite remote sensing values ​​of the corresponding pixels in satellite images can reflect the dust storm phenomenon. Therefore, if the multi-channel satellite remote sensing values ​​of this pixel at multiple times are used as input feature data, it will be more effective to determine whether dust storms have occurred at that pixel. For example, assuming the single-time multi-channel satellite remote sensing data of a single pixel can be represented as Ct={C1, C2, ..., Cnum}, its multi-channel satellite remote sensing data at 2k+1 consecutive times can be represented as C={Ct-k, ..., Ct, ..., Ct+k}; where C1 represents channel 1, num can be the maximum number of channels, such as 14; each channel can be represented as Channel1, Channel2, ..., Channeln, where the multi-time satellite remote sensing value corresponding to Channel1 can be represented as... ... The multi-time satellite remote sensing values ​​corresponding to Channel 2 can be expressed as follows: ... The corresponding multi-time satellite remote sensing values ​​can be expressed as: Correspondingly, the arrangement of the multi-time, multi-channel satellite remote sensing numerical matrix C is as follows: .

[0043] In some embodiments, if the collected satellite cloud image data is Fengyun satellite data, each image element has 14 satellite remote sensing data channels. The names of each channel and the corresponding wavelengths are shown in Table 1. Data from 2k+1 consecutive time points constitute two-dimensional data of [14, (2k+1)]. Considering the duration of sandstorms approximately once per hour and the calculation time, k can be taken as 4, and the time span is approximately 2 hours.

[0044] Table 1

[0045] Furthermore, particulate matter (PM10) is a major characteristic factor of dust storms; therefore, when assessing the impact of dust storms, the hourly mass concentration variation of PM10 is primarily considered. Dust storms are typically accompanied by a sharp and rapid increase in PM10 mass concentration and a sharp decrease in the PM2.5 / PM10 mass concentration ratio. Simultaneously, the variation characteristics of PM2.5 are also considered, and particulate matter monitoring data for time periods exhibiting significant external dust intrusion characteristics are analyzed to determine the start and end times of the dust storm's impact.

[0046] Specifically, MSCFF is used to train the model on Data_train. The model input is the multi-time, multi-channel satellite remote sensing numerical matrix C for each pixel, and the model output is the dust storm weather label Label[0,1]. Then, the trained model is used to identify whether extreme weather has occurred in the test set. MSCFF is based on a convolutional neural network and is used to extract multi-scale, multi-level high-dimensional features of pixels. The MSCFF model consists of two parts: an encoder and a decoder and a multi-scale feature fusion module. The model uses the encoder to extract and map features from pixels, and then the decoder and multi-scale feature fusion module arrange the generated feature maps into a time series in chronological order and perform comprehensive feature extraction, ultimately achieving the classification of dust storm pixels and non-dust storm pixels in satellite cloud images.

[0047] For example, the MSCFF model is trained using a training set. The model input can be a multi-time, multi-channel satellite remote sensing numerical matrix C for the two hours preceding the predicted time T1, and the output can be a dust weather label for the predicted time T1. After training, the predicted multi-time, multi-channel satellite remote sensing numerical matrix C corresponding to the test set is input into the model to obtain the dust weather label LabelData_test for the test set period, thereby determining whether the weather of the photovoltaic cluster corresponding to the predicted time is dust weather.

[0048] Optionally, after acquiring historical cluster data of the photovoltaic cluster, the process may further include: determining whether there are missing data periods in the historical acquisition time of the historical cluster data; if it is determined that there are missing data periods in the historical acquisition time of the historical cluster data, then processing the historical cluster data according to the missing data periods and a preset interval threshold; processing the historical cluster data according to the missing data periods and the preset interval threshold includes: if it is determined that the missing time interval of the missing data periods exceeds a preset time interval, then deleting the historical cluster data of the acquisition period corresponding to the missing data periods; if it is determined that the missing time interval of the missing data periods does not exceed the preset time interval, then supplementing the historical cluster data corresponding to the missing time intervals based on linear interpolation.

[0049] In some embodiments, after collecting historical cluster data of photovoltaic clusters, all collected data can undergo data processing such as outlier removal and missing value supplementation. The processed historical cluster data is then divided into training and testing sets for a preset weather prediction model. For example, if the missing data period of historical cluster data collected on a certain day exceeds 2 hours (i.e., more than 8 time intervals), all historical cluster data collected on that day is deleted. If the missing data period of historical cluster data collected on a certain day does not exceed 8 time intervals, linear interpolation can be used to supplement the data, facilitating subsequent model training and improving model robustness.

[0050] Optionally, step 104 may specifically include: using a power prediction model to predict the power of the photovoltaic cluster based on weather tags, forecast meteorological factors, and photovoltaic power data corresponding to preset collection times.

[0051] In some embodiments, the power prediction model may include a Transformer. The input of the power prediction model may be a set of forecast meteorological factors, historical photovoltaic power data, and predicted weather labels. The output may be the predicted power of the photovoltaic cluster at the predicted time. The power prediction model may be trained and tested based on historical cluster data. The training set may be used to train the model, and the input of the test set may be used to predict the photovoltaic power at the corresponding time.

[0052] For example, during the training process of the power prediction model, the input of the power prediction model can be the set of forecast meteorological factors MData_train, photovoltaic power data PData_train, and dust weather label LabelData_train for the day before the prediction time T2. The output of the power prediction model can be the photovoltaic power data for the next 24 hours before the prediction time T2. The model is trained and the parameters are saved.

[0053] The Transformer model is a neural network model architecture applied to machine translation. The Transformer architecture models the global dependency between the source and target language sequences entirely through an attention mechanism. The Transformer consists of an encoder and a decoder, each composed of several basic Transformer blocks, and each Transformer block receives a vector sequence. As input, it outputs a vector sequence of equal length as output. Here, xi and yi correspond to the representation of a word in the text sequence. yi is the output of the current Transformer block after further integrating the contextual semantics of the input xi. The semantic abstraction process from input to output mainly involves the following modules: Attention layer: A multi-head attention mechanism is used to integrate contextual semantics, allowing the dependency relationship between any two words in the sequence to be directly modeled without relying on traditional loop structures, thus better addressing long-range dependencies in the text; Position-aware feedforward layer: A fully connected layer performs a more complex transformation on the representation of each word in the input text sequence; Residual connection: Corresponding to the Add part, it is a direct connection acting on the two sub-layers mentioned above, used to connect their input and output. This makes information flow more efficient and beneficial for model optimization; Layer normalization: Corresponding to the Norm part, it acts on the output representation sequence of the two sub-layers, performing layer normalization on the representation sequence, also playing a role in stabilization and optimization.

[0054] During training, the input set InputData_train=[MData_train, PData_train, LabelData_train] from the day before training time T2 is input into the Transformer model, so that its output is the photovoltaic power data corresponding to the next 4 hours after training time T2. In this way, the power prediction model can learn the relationship between extreme weather, forecast meteorological factors and photovoltaic power, thereby realizing power prediction based on extreme weather.

[0055] Correspondingly, the trained power prediction model can be used to predict the power during the test set period. The model input is MData_test from the day before the prediction time T3, photovoltaic power data PData_test, and weather label LabelData_test. Finally, the photovoltaic power prediction result for the next 4 hours before the prediction time T3 is obtained.

[0056] Optionally, step 105 may specifically include: determining the weather sensitivity coefficient corresponding to the photovoltaic cluster based on the weather label corresponding to the satellite cloud image data; predicting the climbing characteristic quantity corresponding to the photovoltaic cluster based on different prediction time scales and weather sensitivity coefficients corresponding to the climbing events of the photovoltaic cluster; predicting the photovoltaic cluster climbing events corresponding to the photovoltaic cluster by comparing the climbing characteristic quantity with a preset climbing threshold, wherein the preset climbing threshold is determined based on the total installed capacity of the photovoltaic power station corresponding to the photovoltaic cluster.

[0057] In some embodiments, photovoltaic power ramping characteristics can be defined to identify ramping events based on power prediction results. Specifically, indirect prediction methods can be used for photovoltaic ramping prediction, and the defined characteristics can include ramping rate. For example, the ramping rate calculation formula can be expressed as: In the formula, The gradient is expressed as power change per minute (MW / 15min). The predicted power value (in MW) is given at prediction time t. For time intervals, The weather sensitivity coefficient is 0.8 during sandstorms and 0.2 during normal weather.

[0058] Further, optionally, to consider the ramp rate at multiple different prediction time scales (e.g., different time intervals), the improved ramp rate calculation method can be expressed as: In the formula, This can represent the improved gradeability. It can represent the first time interval (e.g.) The gradient rate corresponding to the first time interval (which can be 15 minutes) is α, where α represents the gradient rate weight corresponding to the first time interval. It can represent a second time interval (e.g.) The gradient rate corresponding to the first time interval (which can be 1 hour) can be represented by β, which can represent the gradient rate weight corresponding to the second time interval. It can represent a third time interval (e.g.) The ramp rate (which can be 4 hours) corresponds to the ramp rate, and γ can represent the ramp rate weight corresponding to the third time interval.

[0059] Furthermore, the preset ramp-up threshold can be expressed as a percentage or a fixed value of the photovoltaic power station's installed capacity. For example, 5% of the total installed capacity of the cluster photovoltaic power station can be selected as the preset ramp-up threshold. For each prediction time T4, if the corresponding ramp rate of the device... If the power prediction result is positive, then a ramping event is considered to have occurred, thus enabling the identification of ramping events based on power prediction results and completing ramping prediction.

[0060] For example, as shown in Figure 2, a method for predicting clustered photovoltaic (PV) cluster climbing events based on satellite cloud imagery and extreme weather identification is proposed. The specific steps are as follows: S10: Obtain historical PV power data, numerical weather forecast data, satellite cloud imagery data, etc., for the target area, and perform data preprocessing such as outlier deletion and missing value supplementation; S20: Divide the dataset, label the pixels in the satellite images where extreme weather has occurred and where it has not, and use a multi-scale convolutional feature fusion model to identify extreme weather. Specifically, all datasets are divided into training and testing sets. For the training set, satellite remote sensing values ​​of the areas photographed in all time periods are obtained using satellite cloud imagery data. For pixel areas where dust storms occur in the satellite images, the dust storm weather label is set to 1, and for areas where there are no dust storms, the label is set to 0. The multi-scale convolutional feature fusion model is used for model training. The model input is the satellite remote sensing value of each pixel, and the model output is the dust storm weather label Label[0,1]. Then, the trained model is used to identify whether extreme weather events occur in the test set; S30: The Transformer is used as the power prediction model, and the Pearson correlation coefficient is used to screen out the set of forecast meteorological factors M that are strongly correlated with power. The model input is historical NMP data (or the set of forecast meteorological factors M), historical photovoltaic power data, and extreme event labels. The power data for the next 24 hours is used as the output. The model is trained using the training set, and the photovoltaic power at the corresponding time is predicted using the input of the test set; S40: The power prediction value is obtained, and photovoltaic ramp-up prediction is performed based on the relevant ramp-up features. Specifically, photovoltaic power ramp-up features can be defined, and ramp-up events are identified based on the power prediction results to complete the photovoltaic ramp-up prediction considering the impact of extreme weather.

[0061] Accordingly, this method first uses a multi-scale convolutional feature fusion model to identify extreme weather events such as sandstorms, then combines it with a Transformer power prediction model to predict photovoltaic power for the next 24 hours, and finally predicts ramp-up events based on the power prediction results. This helps grid dispatchers identify potential power surge risks in advance and formulate more scientific and reasonable operation, maintenance and dispatch strategies, thereby effectively improving the stability and operating efficiency of the power system.

[0062] Compared with related technologies, this embodiment can construct a cluster photovoltaic (PV) ramp-up prediction model based on satellite cloud images and extreme weather identification. It deeply integrates historical PV power data, historical NWP data, and historical satellite cloud image data of the cluster area, and identifies and processes missing data periods in historical cluster data to improve model robustness. It achieves accurate identification of extreme weather through multi-scale convolutional feature fusion technology, providing reliable meteorological feature input for power ramp-up event prediction. It uses Pearson correlation coefficient to screen forecast meteorological factors with strong correlation to PV power, and then models PV power data through power prediction model. Combined with weather tags and forecast meteorological factors, it effectively captures the strong correlation between meteorological changes and power fluctuations, effectively reducing the risk of sudden changes in PV power caused by extreme weather. Finally, based on different prediction time scales and weather sensitivity coefficients corresponding to PV cluster ramp-up events, it predicts the ramp-up characteristic quantities corresponding to PV clusters. By comparing the ramp-up characteristic quantities with preset ramp-up thresholds, it predicts the PV cluster ramp-up events corresponding to PV clusters, providing key decision support for grid safety dispatch, while improving the capacity for new energy consumption and contributing to the reliable and economical operation of the new power system.

[0063] Furthermore, this application provides a photovoltaic cluster ramping event prediction device based on satellite cloud imagery, as shown in Figure 3. The device includes: an acquisition module 31 and a determination module 32.

[0064] The acquisition module 31 is configured to acquire photovoltaic power data, weather forecast data, and satellite cloud image data corresponding to the photovoltaic power station in the photovoltaic cluster within a preset acquisition time. The preset acquisition time is the period before the predicted time of the photovoltaic cluster ramping event. The determination module 32 is configured to determine the forecast meteorological factors corresponding to the photovoltaic cluster from the weather forecast data based on the correlation strength between the weather forecast data and the photovoltaic power of the photovoltaic power station; determine the extreme weather of the photovoltaic cluster corresponding to the predicted time based on the satellite remote sensing values ​​of the geographical area of ​​the photovoltaic cluster corresponding to the photovoltaic cluster in the satellite cloud image data; determine the predicted power of the photovoltaic cluster corresponding to the predicted time based on the extreme weather and the forecast meteorological factors; and determine the photovoltaic cluster ramping event corresponding to the predicted time based on the ramping characteristic quantity corresponding to the predicted power.

[0065] In some embodiments, the determining module 32 is specifically configured to use a preset meteorological prediction model to obtain the weather tag of the photovoltaic cluster corresponding to the prediction time based on the satellite remote sensing values ​​of the pixels in the satellite cloud image data at different acquisition times and different channels. The weather tag is used to determine the extreme weather of the photovoltaic cluster.

[0066] In some embodiments, the acquisition module 31 is specifically configured to acquire historical cluster data of the photovoltaic cluster, including historical photovoltaic power data, historical weather forecast data, and historical satellite cloud image data; divide the historical cluster data into a training set and a test set corresponding to a preset meteorological prediction model according to a preset division ratio; use weather tags to mark extreme weather pixel areas and non-extreme weather pixel areas of the historical satellite cloud image data in the training set, the weather tags including extreme weather tags and non-extreme weather tags; use a multi-scale convolutional feature fusion model to train the model on the training set, and use the trained multi-scale convolutional feature fusion model as the preset meteorological prediction model. The preset meteorological prediction model is used to identify extreme weather on the test set. The input of the multi-scale convolutional feature fusion model is satellite remote sensing data, and the output is the weather tags of the photovoltaic cluster.

[0067] In some embodiments, the acquisition module 31 is further configured to determine whether there are missing data periods in the historical collection time of the historical cluster data; if it is determined that there are missing data periods in the historical collection time of the historical cluster data, then the historical cluster data is processed according to the missing data periods and a preset interval threshold; the processing of historical cluster data according to the missing data periods and the preset interval threshold includes: if it is determined that the missing time interval of the missing data periods exceeds a preset time interval, then the historical cluster data of the collection period corresponding to the missing data periods is deleted; if it is determined that the missing time interval of the missing data periods does not exceed the preset time interval, then the historical cluster data corresponding to the missing time interval is supplemented based on linear interpolation.

[0068] In some embodiments, the determining module 32 is specifically configured to use a power prediction model to predict the predicted power of a photovoltaic cluster based on weather tags, forecast meteorological factors, and photovoltaic power data corresponding to a preset collection time.

[0069] In some embodiments, the determining module 32 is specifically configured to use the Pearson correlation coefficient to determine the correlation strength between weather forecast data and photovoltaic power of a photovoltaic power plant, and to screen out forecast meteorological factors from the weather forecast data.

[0070] In some embodiments, the determining module 32 is specifically configured to determine the weather sensitivity coefficient corresponding to the photovoltaic cluster based on the weather label corresponding to the satellite cloud image data; predict the climbing characteristic quantity corresponding to the photovoltaic cluster based on different prediction time scales and weather sensitivity coefficients corresponding to the climbing events of the photovoltaic cluster; and predict the photovoltaic cluster climbing events corresponding to the photovoltaic cluster by comparing the climbing characteristic quantity with a preset climbing threshold, wherein the preset climbing threshold is determined based on the total installed capacity of the photovoltaic power station corresponding to the photovoltaic cluster.

[0071] It should be noted that other corresponding descriptions of the functional units involved in the photovoltaic cluster ramping event prediction device based on satellite cloud imagery provided in this application embodiment can be found in the corresponding descriptions in Figure 1, and will not be repeated here.

[0072] Based on the example shown in FIG1 above, the present application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the example method shown in FIG1 above.

[0073] Based on the example shown in FIG1 above, the present application also provides a computer program product, including a computer program that implements the example method shown in FIG1 when executed by a processor.

[0074] Based on this understanding, the technical solutions of the embodiments of this application can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0075] Based on the method shown in FIG1 and the virtual device embodiment shown in FIG3, in order to achieve the above objectives, this application embodiment also provides an electronic device, which includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the method shown in FIG1.

[0076] Optionally, the aforementioned electronic device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, an input unit, etc.

[0077] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0078] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0079] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware. This application constructs a clustered photovoltaic (PV) ramp-up prediction model based on satellite cloud imagery and extreme weather identification. It deeply integrates historical PV power data, historical NWP data, and historical satellite cloud imagery data for the cluster area, and identifies and processes missing data periods in historical cluster data to improve model robustness. Through multi-scale convolutional feature fusion technology, it achieves accurate identification of extreme weather, providing reliable meteorological feature inputs for power ramp-up event prediction. It utilizes Pearson correlation coefficients to screen forecast meteorological factors with strong correlation to PV power, and then models PV power data using a power prediction model. Combining weather tags and forecast meteorological factors, it effectively captures the strong correlation between meteorological abrupt changes and power fluctuations, effectively reducing the risk of sudden PV power changes due to extreme weather. Finally, based on different prediction time scales and weather sensitivity coefficients corresponding to PV cluster ramp-up events, it predicts the ramp-up characteristic quantities corresponding to the PV cluster. By comparing the ramp-up characteristic quantities with preset ramp-up thresholds, it predicts the PV cluster ramp-up events corresponding to the PV cluster, providing key decision support for grid safety dispatch, while simultaneously improving the capacity for renewable energy absorption and contributing to the reliable and economical operation of the new power system.

[0080] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0081] The above are merely specific embodiments of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to these embodiments, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for predicting photovoltaic cluster ramp-up events based on satellite cloud images, characterized in that, include: The process involves acquiring photovoltaic power data, weather forecast data, and satellite cloud image data corresponding to photovoltaic power plants in a photovoltaic cluster within a preset acquisition time period. The preset acquisition time is the period preceding the predicted time of the photovoltaic cluster's ramp-up event. Based on the correlation strength between the weather forecast data and the photovoltaic power of the photovoltaic power plants, the process determines the forecast meteorological factors corresponding to the photovoltaic cluster from the weather forecast data. Based on the satellite remote sensing values ​​of the geographical area corresponding to the photovoltaic cluster in the satellite cloud image data, the process determines the extreme weather conditions for the photovoltaic cluster at the predicted time. Based on the extreme weather conditions and the forecast meteorological factors, the process determines the predicted power of the photovoltaic cluster at the predicted time. Finally, based on the ramp-up characteristic quantities corresponding to the predicted power, the process determines the photovoltaic cluster ramp-up event at the predicted time.

2. The method according to claim 1, characterized in that, The step of determining the extreme weather of the photovoltaic cluster corresponding to the prediction time based on the satellite remote sensing values ​​of the geographical area of ​​the photovoltaic cluster in the satellite cloud image data includes: using a preset meteorological prediction model, and based on the satellite remote sensing values ​​of the pixels in the satellite cloud image data at different acquisition times and different channels, obtaining the weather tag of the photovoltaic cluster corresponding to the prediction time, wherein the weather tag is used to determine the extreme weather of the photovoltaic cluster.

3. The method according to claim 2, characterized in that, Before determining the extreme weather of the photovoltaic cluster corresponding to the prediction time based on the satellite remote sensing values ​​of the geographical area of ​​the photovoltaic cluster in the satellite cloud image data, the method further includes: acquiring historical cluster data of the photovoltaic cluster, the historical cluster data including historical photovoltaic power data, historical weather forecast data, and historical satellite cloud image data; dividing the historical cluster data into a training set and a test set corresponding to the preset meteorological prediction model according to a preset division ratio; using weather tags to mark extreme weather pixel areas and non-extreme weather pixel areas of the historical satellite cloud image data in the training set, the weather tags including extreme weather tags and non-extreme weather tags; using a multi-scale convolutional feature fusion model to train the model on the training set, and using the trained multi-scale convolutional feature fusion model as the preset meteorological prediction model, the preset meteorological prediction model is used to identify extreme weather on the test set, the input of the multi-scale convolutional feature fusion model is the satellite remote sensing value, and the output is the weather tag of the photovoltaic cluster.

4. The method according to claim 3, characterized in that, After acquiring the historical cluster data of the photovoltaic cluster, the method further includes: determining whether there are data gap periods in the historical acquisition time of the historical cluster data; if it is determined that there are data gap periods in the historical acquisition time of the historical cluster data, then processing the historical cluster data according to the data gap periods and a preset interval threshold; the processing of historical cluster data according to the data gap periods and the preset interval threshold includes: if it is determined that the time interval of the data gap periods exceeds the preset time interval, then deleting the historical cluster data of the acquisition period corresponding to the data gap periods; if it is determined that the time interval of the data gap periods does not exceed the preset time interval, then supplementing the historical cluster data corresponding to the time interval of the data gaps based on linear interpolation.

5. The method according to claim 2, characterized in that, The step of determining the predicted power of the photovoltaic cluster corresponding to the predicted time based on the extreme weather and the forecast meteorological factors includes: using a power prediction model to predict the predicted power of the photovoltaic cluster based on the weather tag, the forecast meteorological factors and the photovoltaic power data corresponding to the preset collection time.

6. The method according to claim 1, characterized in that, The step of determining the forecast meteorological factors corresponding to the photovoltaic cluster from the weather forecast data based on the correlation strength between the weather forecast data and the photovoltaic power of the photovoltaic power station includes: using the Pearson correlation coefficient to determine the correlation strength between the weather forecast data and the photovoltaic power of the photovoltaic power station, and screening out the forecast meteorological factors from the weather forecast data.

7. The method according to claim 1, characterized in that, The step of determining the photovoltaic cluster ramping event corresponding to the prediction time based on the ramping characteristic quantity corresponding to the predicted power includes: determining the weather sensitivity coefficient corresponding to the photovoltaic cluster based on the weather label corresponding to the satellite cloud image data; predicting the ramping characteristic quantity corresponding to the photovoltaic cluster based on different prediction time scales corresponding to the photovoltaic cluster ramping event and the weather sensitivity coefficient; and predicting the photovoltaic cluster ramping event corresponding to the photovoltaic cluster by comparing the ramping characteristic quantity with a preset ramping threshold, wherein the preset ramping threshold is determined based on the total installed capacity of the photovoltaic power station corresponding to the photovoltaic cluster.

8. A photovoltaic cluster ramp-up event prediction device based on satellite cloud images, characterized in that, include: The acquisition module is configured to acquire photovoltaic power data, weather forecast data, and satellite cloud image data corresponding to the photovoltaic power station in the photovoltaic cluster within a preset acquisition time, wherein the preset acquisition time is the time period before the predicted time of the photovoltaic cluster ramp-up event; the determination module is configured to determine the forecast meteorological factors corresponding to the photovoltaic cluster from the weather forecast data based on the correlation strength between the weather forecast data and the photovoltaic power of the photovoltaic power station; determine the extreme weather of the photovoltaic cluster corresponding to the predicted time based on the satellite remote sensing values ​​of the geographical area of ​​the photovoltaic cluster corresponding to the photovoltaic cluster in the satellite cloud image data; determine the predicted power of the photovoltaic cluster corresponding to the predicted time based on the extreme weather and the forecast meteorological factors; and determine the photovoltaic cluster ramp-up event corresponding to the predicted time based on the ramp-up characteristic quantity corresponding to the predicted power.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. An electronic device comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images