A power prediction system

Abstract

The application discloses a power prediction system capable of adaptively adjusting a prediction frequency and improving prediction accuracy. The power prediction system comprises a photovoltaic data collector, a satellite ground station receiver and a data processing device. The photovoltaic power generation system comprises photovoltaic inverters and photovoltaic subarrays, and a plurality of photovoltaic inverters are connected with one or more photovoltaic inverters respectively. The photovoltaic data collector is used for collecting operation data of the plurality of photovoltaic inverters; the satellite ground station receiver is used for receiving satellite meteorological data of a geographical space to which the photovoltaic power generation system belongs; and the data processing device is used for constructing a virtual cloud picture according to the operation data of the plurality of photovoltaic inverters and the satellite meteorological data, obtaining node features corresponding to a plurality of nodes of the virtual cloud picture, and adjusting a prediction frequency of power prediction to a target prediction frequency according to the node features.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power generation technology, and in particular to a power prediction system. Background Technology

[0002] Photovoltaic power generation, as an important form of clean and renewable energy, has led to the widespread application of photovoltaic power generation systems. Among these systems, photovoltaic power forecasting is a crucial link in ensuring the participation of photovoltaic power generation systems in electricity market transactions and guaranteeing their compliant and stable grid connection, thus determining the economic benefits and operational safety of the photovoltaic power generation system.

[0003] However, the prediction frequency of power prediction models is usually fixed and cannot be adjusted to follow changes in the environment in which the photovoltaic power generation system is located, which reduces the prediction accuracy and makes it difficult to meet the high-precision power prediction requirements in practical applications. Summary of the Invention

[0004] This application provides a power prediction system that can adaptively adjust the prediction frequency, thereby improving prediction accuracy.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions: A first aspect of this application provides a power prediction system for power prediction of a photovoltaic (PV) power generation system. The power prediction system includes a PV data acquisition unit, a satellite ground station receiver, and a data processing device. The PV power generation system includes PV inverters and PV subarrays, with multiple PV inverters connected to one or more PV inverters respectively. The PV data acquisition unit is used to collect operating data from the multiple PV inverters; the satellite ground receiver is used to receive satellite meteorological data from the geographic space to which the PV power generation system belongs; the data processing device is used to construct a virtual cloud map based on the operating data of the multiple PV inverters and the satellite meteorological data, obtain node characteristics corresponding to multiple nodes in the virtual cloud map, with each node corresponding one-to-one with multiple PV subarrays, and adjust the prediction frequency of the power prediction to the target prediction frequency based on the node characteristics.

[0006] Based on this scheme, a virtual cloud map is constructed using the operating data of the photovoltaic inverter collected by the photovoltaic data acquisition device and the satellite meteorological data of the geographic space of the photovoltaic power generation system received by the satellite ground station receiver. This allows the node characteristics of each node in the virtual cloud map to accurately represent the environmental and operational status of the corresponding photovoltaic subarray area, thus solving the problem of single data in the virtual cloud map and its inability to reflect local environmental differences within the photovoltaic subarray, thereby improving the perception accuracy of environmental changes at the photovoltaic subarray level. Furthermore, after obtaining the virtual cloud map, the prediction frequency for power forecasting is adjusted to the target prediction frequency based on the node characteristics of multiple nodes in the virtual cloud map. This allows the prediction frequency to adaptively adjust with changes in the operational status of the photovoltaic subarray and meteorological conditions. For example, the prediction frequency is reduced when meteorological fluctuations are stable to reduce computing power consumption, and increased when meteorological fluctuations are severe to capture environmental changes and improve prediction accuracy, thus achieving a balance between prediction accuracy and computing power efficiency.

[0007] In conjunction with the first aspect, in one implementation, the node features include mean brightness, variance brightness, and gradient brightness. Satellite meteorological data includes irradiance data and satellite cloud imagery data of the geographic space to which the photovoltaic power generation system belongs. The data processing equipment is further configured to: obtain the mean brightness, variance brightness, and gradient brightness from the node features based on the irradiance data and the satellite cloud imagery data.

[0008] Based on this scheme, the mean brightness, variance brightness, and gradient brightness are calculated using irradiance data and satellite cloud image data. This quantifies the physical states of each photovoltaic subarray, such as illumination conditions, cloud fluctuations, and spatial meteorological differences, into specific characteristic parameters. Specifically, the mean brightness reflects the illumination conditions of the area where multiple photovoltaic subarrays are located, the variance brightness reflects the severity of irradiance changes caused by cloud fluctuations in the area, and the gradient brightness reflects the degree of spatial meteorological differences between adjacent photovoltaic subarrays. In this way, the node features can characterize the meteorological conditions of the corresponding photovoltaic subarray from three dimensions: illumination intensity, temporal fluctuations, and spatial differences. This improves the accuracy of the virtual cloud image in depicting the spatial distribution characteristics of the photovoltaic power generation system and provides a calculable and comparable data basis for the adaptive adjustment of the subsequent prediction frequency.

[0009] In conjunction with the first aspect, in one implementation, the target prediction frequency includes a first frequency, a second frequency, and a third frequency, with the prediction intervals corresponding to the first frequency, the second frequency, and the third frequency decreasing sequentially.

[0010] Based on this scheme, by setting prediction frequency levels with progressively decreasing prediction intervals for the first, second, and third frequencies, a multi-level selectable prediction frequency configuration is provided for the power prediction system. This allows the power prediction system to select a matching prediction frequency from multiple prediction frequency levels based on the meteorological conditions of multiple photovoltaic subarrays. This provides a tiered basis and flexible selection space for prediction frequency adjustment, avoiding the problem of a single prediction frequency being unable to adapt to various meteorological scenarios. Furthermore, the progressively decreasing prediction intervals for the three frequency levels are suitable for three scenarios: stable meteorological changes, moderate meteorological fluctuations, and severe meteorological fluctuations, respectively. The first frequency is used when meteorological changes are stable to reduce system computing power consumption, while the third frequency is used when meteorological fluctuations are severe to capture rapid changes, achieving differentiated prediction frequency configurations for different meteorological scenarios.

[0011] In conjunction with the first aspect, in one embodiment, the data processing device is further configured to: adjust the target prediction frequency to a first frequency if the brightness variance of multiple nodes is less than or equal to a first variance threshold and the brightness gradient is less than or equal to a first gradient threshold; adjust the target prediction frequency to a second frequency if the brightness variance of multiple nodes is greater than the first variance threshold and less than or equal to a second variance threshold, and the brightness gradient is greater than the first gradient threshold and less than or equal to the second gradient threshold; and adjust the target prediction frequency to a third frequency if the brightness variance corresponding to any node among the multiple nodes is greater than the second variance threshold and / or the corresponding brightness gradient is greater than the second gradient threshold.

[0012] Based on this scheme, by setting two thresholds for brightness variance and brightness gradient, the degree of meteorological fluctuation is divided into three levels: stable, moderate, and severe. Power prediction of the photovoltaic power generation system is then performed using a first frequency, a second frequency, and a third frequency, respectively, to achieve precise matching between the prediction frequency and the degree of meteorological fluctuation, thereby improving prediction accuracy. Specifically, when the brightness variance and brightness gradient are less than or equal to the first variance threshold and the first gradient threshold, respectively, the degree of meteorological fluctuation is stable, and the first frequency with the largest prediction interval is used for prediction to reduce computational power consumption. When the brightness variance of multiple nodes is greater than the first variance threshold and less than or equal to the second variance threshold, and the brightness gradient is greater than the first gradient threshold and less than or equal to the second gradient threshold, the degree of meteorological fluctuation is moderate, and the second frequency with a medium prediction interval is used for prediction to balance prediction accuracy and computational power. When the brightness variance of any node is greater than the second variance threshold, and / or the brightness gradient is greater than the second gradient threshold, the degree of meteorological fluctuation is severe, and the third frequency with the smallest prediction interval is used for prediction to capture drastic changes and improve prediction accuracy.

[0013] In conjunction with the first aspect, in one embodiment, the power prediction system further includes a meteorological data acquisition device and a panoramic sky camera. The meteorological data acquisition device is used to collect surface meteorological data of the geographic space to which the photovoltaic power generation system is located. The panoramic sky camera is used to collect panoramic sky image data of the geographic space to which the photovoltaic power generation system is located; the data processing device is also used to obtain the mean brightness, variance brightness, and gradient brightness of the node features based on the surface meteorological data, panoramic sky image data, irradiance data, and satellite cloud image data.

[0014] Based on this scheme, ground-based meteorological data and panoramic sky imagery data are obtained from ground observations, offering high accuracy but limited coverage. Satellite meteorological data, acquired through remote sensing observations by meteorological satellites, boasts broad coverage but lower resolution and accuracy compared to ground-based observations. By combining ground-based meteorological data and panoramic sky imagery data, accurate information can be provided, compensating for the localized inaccuracies of satellite meteorological data. Thus, through the fusion and complementarity of these four data types, the mean brightness, variance brightness, and brightness gradient of node features can be accurately obtained, avoiding information gaps or biases caused by relying on a single data source. This improves the ability of node features to characterize the meteorological environment of the corresponding photovoltaic subarray.

[0015] In conjunction with the first aspect, in one implementation, the data processing device is specifically used to: extract spatial features from a virtual cloud image to obtain a spatial feature vector corresponding to the virtual cloud image; extract temporal features from irradiance data in satellite meteorological data to obtain a first temporal feature vector; and output the target prediction frequency based on the spatial feature vector and the first temporal feature vector.

[0016] Based on this scheme, spatial feature vectors are obtained by extracting spatial features from virtual cloud maps, and temporal feature vectors are obtained by extracting temporal features from irradiance data in multimodal data. This allows for the fusion of spatial topological differences between photovoltaic subarrays reflected in the virtual cloud map (such as differences in illumination between adjacent photovoltaic subarrays, cloud boundary distribution, etc.) with the temporal series change trends corresponding to irradiance data (such as short-term fluctuations and gradual trends in illumination). This enables the target prediction frequency to take into account information in both spatial and temporal dimensions, allowing for a more comprehensive perception of changes in the meteorological environment and thus outputting more accurate prediction frequencies.

[0017] In conjunction with the first aspect, in one embodiment, the data processing device is further used to extract time-series features from satellite meteorological data other than irradiation data and operational data to obtain a second time-series feature vector; and output the target prediction frequency based on the spatial feature vector, the first time-series feature vector and the second time-series feature vector.

[0018] Based on this scheme, by extracting time-series features from satellite meteorological data other than irradiance data (such as temperature, humidity, wind speed, cloud parameters, etc.) and operational data, a second time-series feature vector is obtained. This second time-series feature vector is then combined with the spatial feature vector and the first time-series feature vector (irradiance-related time-series feature vector) to output the target prediction frequency. In this way, the target prediction frequency not only considers the time-series changes in light intensity and the spatial differences between photovoltaic subarrays (spatial feature vector), but also further incorporates other meteorological data and the power generation status of photovoltaic subarrays. This allows the prediction frequency to simultaneously perceive multidimensional changes in the meteorological environment and the operating status of the equipment, thereby improving the adaptability of the prediction frequency to complex meteorological scenarios and equipment operating conditions.

[0019] In conjunction with the first aspect, in one embodiment, the data processing device is further configured to perform power prediction on the photovoltaic power generation system after the prediction frequency of the power prediction system is adjusted to the target prediction frequency.

[0020] Based on this scheme, after adjusting the prediction frequency to the target prediction frequency, power prediction is performed according to the target prediction frequency, thereby realizing closed-loop control from frequency adjustment to power prediction. This allows the prediction frequency of power prediction to be adjusted according to changes in the meteorological environment, thereby improving the accuracy of power prediction.

[0021] In conjunction with the first aspect, in one embodiment, the power prediction system further includes a display device. The data processing device is also used to render the virtual cloud map to obtain a visualized virtual cloud map; and to display the visualized virtual cloud map on the display device.

[0022] Based on this solution, by rendering each node of the virtual cloud map, a visualized virtual cloud map is generated and displayed, enabling operation and maintenance personnel to intuitively view the environmental and operational status of each photovoltaic subarray, such as illumination conditions, cloud fluctuations, operational stability, and spatial meteorological differences. This facilitates monitoring the overall status of the photovoltaic power generation system, quickly locating abnormal photovoltaic subarrays, and improving the observability and operation and maintenance efficiency of the photovoltaic power generation system.

[0023] In conjunction with the first aspect, in one implementation, in the visualized virtual cloud map, each node and its corresponding node features present the environmental and operational status of the area where the photovoltaic subarray corresponding to the node is located in at least one of the following ways: color depth, color dispersion, luminous uniformity, and color transition differences.

[0024] Based on this solution, the environmental and operational status of each photovoltaic subarray area is presented in the visualized virtual cloud map through at least one of the following methods: color depth, color dispersion, luminous uniformity, and color transition differences. For example, color depth reflects the intensity of sunlight, color dispersion reflects the intensity of cloud fluctuations, luminous uniformity reflects the operational stability of the photovoltaic inverter, and color transition differences reflect spatial meteorological differences between photovoltaic subarrays. In this way, abstract feature data is transformed into intuitive visual elements, enabling maintenance personnel to easily identify status differences and abnormal areas in each photovoltaic subarray, improving monitoring efficiency and user experience. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the composition of a photovoltaic power generation system provided in an embodiment of this application; Figure 2 This is a schematic diagram of the composition of a power prediction system provided in an embodiment of this application; Figure 3 This is a schematic diagram of another power prediction system provided in an embodiment of this application; Figure 4 A schematic diagram illustrating the process of constructing a virtual cloud map, provided as an embodiment of this application; Figure 5 This is a schematic diagram illustrating an application of frequency prediction acquisition provided in an embodiment of this application; Figure 6 This is a schematic diagram illustrating the composition of yet another power prediction system provided in an embodiment of this application; Figure 7 This is a schematic diagram of the application architecture of a power prediction system provided in an embodiment of this application. Detailed Implementation

[0026] The following sections will discuss the fabrication and use of various embodiments in detail. However, it should be understood that many applicable inventive concepts provided in this application can be implemented in a variety of specific environments. The specific embodiments discussed are merely illustrative of specific ways of implementing and using this description and technology, and do not limit the scope of this application.

[0027] Unless otherwise defined, all technical terms used herein have the same meaning as commonly known to one of ordinary skill in the art.

[0028] Each circuit or other component may be described or referred to as "for" performing one or more tasks. In this context, "for" is used to imply a structure by indicating that the circuit / component includes a structure (e.g., a circuit system) that performs one or more tasks during operation. Therefore, even when the specified circuit / component is currently inoperable (e.g., not turned on), it can still be referred to as "for performing that task." Circuits / components used with the term "for" include hardware, such as circuits that perform operations.

[0029] The technical solutions in the embodiments of this application will be described below with reference to the accompanying drawings. In this application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, or B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the related objects before and after are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can mean: a, b, c, a and b, a and c, b and c, or a, b, and c, where a, b, and c can be single or multiple. In addition, in the embodiments of this application, the words "first," "second," etc., do not limit the quantity or order.

[0030] In this application, the terms "first," "second," etc., do not limit the quantity or order. In this application, the words "exemplary" or "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of "exemplary" or "for example," etc., is intended to present the relevant concepts in a concrete manner.

[0031] Before introducing the embodiments of this application, the application scenarios involved in this application will be introduced first.

[0032] With the rapid development of new energy sources, photovoltaic power generation, as a core form of clean and renewable energy, has led to the widespread application of photovoltaic power generation systems. For example... Figure 1 The diagram shown is a schematic representation of a photovoltaic power generation system 100 provided in an embodiment of this application. The photovoltaic power generation system 100 includes a photovoltaic inverter 110. The DC side of the photovoltaic inverter 110 is used to connect to a photovoltaic subarray 200, and the AC side of the photovoltaic inverter 110 is used to connect to a load (not shown in the figure) or the low-voltage side of a transformer 300. The high-voltage side of the transformer 300 is used to connect to the power grid 400.

[0033] The photovoltaic subarray 200 includes at least one photovoltaic string, each photovoltaic string being formed by multiple photovoltaic modules connected in series. The photovoltaic modules are used to convert solar energy into direct current. The photovoltaic subarray 200 can be connected to one photovoltaic inverter or multiple photovoltaic inverters; this embodiment of the application does not impose any limitation on this.

[0034] Specifically, see Figure 1As shown, a photovoltaic subarray 200 is connected to a photovoltaic inverter 110. The photovoltaic inverter 110 is a centralized inverter that converts the direct current generated by the entire photovoltaic subarray 200 into alternating current. When a photovoltaic subarray 200 is connected to multiple photovoltaic inverters 110, these multiple photovoltaic inverters 110 can be string inverters, with each photovoltaic inverter 110 corresponding to a photovoltaic string in the photovoltaic subarray 200.

[0035] The photovoltaic inverter 110 may include a DC bus, a DC-DC conversion circuit, and an inverter circuit. The DC bus is used to connect to the photovoltaic string through the DC-DC conversion circuit. The DC bus is also used to connect to a load or transformer 300 through the photovoltaic inverter 110. The DC-DC conversion circuit is used to convert the DC power from the photovoltaic string into voltage. The DC bus is used to transmit the voltage-converted DC power to the inverter circuit. The inverter circuit is used to convert the DC power into AC power and transmit it to the grid 400 or the load.

[0036] During the operation of the photovoltaic (PV) power generation system 100, power generation forecasting is a key technology for ensuring efficient system management, stable grid connection, and optimized dispatch. Specifically, power forecasting is a core component of the PV system 100's participation in electricity trading and compliant operation. Based on the power forecast results, the PV system 100 submits its power generation plan to dispatching or regulatory authorities and is subject to assessments of indicators such as power forecast deviation. If the power forecast deviation exceeds the allowable range, it will face fines or power curtailment penalties. Therefore, power forecasting directly affects the economic benefits and compliant operation of the PV power generation system.

[0037] However, the prediction frequency of power prediction models is usually fixed and cannot be adjusted to follow changes in the environment in which the photovoltaic power generation system is located, which reduces the prediction accuracy and makes it difficult to meet the high-precision power prediction requirements in practical applications.

[0038] Based on this, this application provides a power prediction system. This system utilizes the operating data of the photovoltaic inverter collected by a photovoltaic data acquisition device and the satellite meteorological data of the geographic space to which the photovoltaic power generation system belongs, received by a satellite ground station receiver, to construct a virtual cloud map. This allows the node characteristics corresponding to each node in the virtual cloud map to accurately represent the environmental and operational status of the area where the corresponding photovoltaic subarray is located, thereby solving the problem of the virtual cloud map data being singular and unable to reflect local environmental differences in the photovoltaic subarray, and improving the perception accuracy of environmental changes at the photovoltaic subarray level. Furthermore, after obtaining the virtual cloud map, the prediction frequency of the power prediction is adjusted to the target prediction frequency based on the node characteristics corresponding to multiple nodes in the virtual cloud map. This allows the prediction frequency to adaptively adjust with changes in the operational status of the photovoltaic subarray and meteorological conditions. For example, the prediction frequency is reduced when meteorological fluctuations are stable to reduce computing power consumption, and the prediction frequency is increased when meteorological fluctuations are severe to capture environmental changes and improve prediction accuracy, thereby achieving a balance between prediction accuracy and computing power efficiency.

[0039] In one implementation, such as Figure 2 The diagram shows the composition of a power prediction system 500 provided in an embodiment of this application. The power prediction system 500 is used for power prediction of a photovoltaic power generation system 100. The power prediction system 500 includes a photovoltaic data acquisition unit 511 and a satellite ground station receiver 512.

[0040] The photovoltaic data acquisition unit 511 is used to collect operating data from multiple photovoltaic inverters 110. For example, it collects operating data such as output power, operating current, operating voltage, and operating status indicators (e.g., whether the output of the photovoltaic inverter 110 is stable) of multiple photovoltaic subarrays 200 connected to them at set time intervals. This data reflects the operating status and power generation characteristics of the photovoltaic subarrays 200, thus providing data support for power prediction of the photovoltaic power generation system 100. The photovoltaic data acquisition unit 511 can be located within the photovoltaic inverter 110, or it can be a separate device from the photovoltaic inverter 110; there is no limitation on this.

[0041] The satellite ground station receiver 512 is used to receive satellite meteorological data from the geographic space where the photovoltaic power generation system 100 is located. This satellite meteorological data is standardized data that has undergone geolocation and radiometric calibration preprocessing. It reflects the irradiance distribution and cloud cover conditions in the geographic space where the photovoltaic power generation system 100 is located. For example, the satellite meteorological data includes irradiance data and satellite cloud image data. Irradiance data is used to determine the irradiance variation trend in the area where multiple photovoltaic subarrays 200 are located, and satellite cloud image data is used to determine the cloud cover variation trend in the area where multiple photovoltaic subarrays 200 are located. Changes in cloud cover variation trends directly affect the irradiance received by the photovoltaic subarrays 200.

[0042] The geographical space to which the aforementioned photovoltaic power generation system 100 belongs may refer to the area where the photovoltaic sub-arrays 200 and other components of the photovoltaic power generation system 100 are arranged.

[0043] In one implementation, see Figure 2 The power prediction system 500 also includes a data processing device 520.

[0044] The data processing device 520 is used to: construct a virtual cloud map based on the operating data of multiple photovoltaic inverters 110 and satellite meteorological data, obtain the node characteristics corresponding to multiple nodes of the virtual cloud map, with multiple nodes corresponding one-to-one with multiple photovoltaic subarrays, and adjust the prediction frequency of power prediction to the target prediction frequency based on the node characteristics.

[0045] The aforementioned virtual cloud map includes multiple nodes and node features corresponding to each node. Each node represents a photovoltaic inverter 110 connected to a photovoltaic subarray 200 in a photovoltaic power generation system. The node features represent the environmental and operational status of the area where the corresponding photovoltaic subarray 200 is located.

[0046] When constructing the virtual cloud map, multiple nodes in the virtual cloud map are built based on the photovoltaic subarray topology information. The photovoltaic subarray topology information is used to indicate the spatial distribution relationship between multiple photovoltaic subarrays 200 in the photovoltaic power generation system 100 and the connection relationship between each photovoltaic subarray 200 and the photovoltaic inverter 110. Specifically, it includes the geographical location, arrangement, adjacency relationship between adjacent photovoltaic subarrays 200, and the number of inverters connected to each photovoltaic subarray 200. For example, when a photovoltaic subarray 200 is connected to a photovoltaic inverter 110, the operating data of the photovoltaic inverter 110 is bound to the node corresponding to the connected photovoltaic subarray 200. Or, when a photovoltaic subarray 200 is connected to multiple photovoltaic inverters 110, the operating data of the multiple photovoltaic inverters 110 is aggregated and bound to the node corresponding to the connected photovoltaic subarray 200. In this way, each node in the virtual cloud map can be accurately associated with the geographical region of the corresponding photovoltaic subarray 200 and the inverter connection relationship, solving the problem that the virtual cloud map cannot reflect the local differences of the photovoltaic subarray 200 due to the lack of photovoltaic subarray topology information.

[0047] The data processing device 520 analyzes the characteristic changes of each node in the virtual cloud map to comprehensively judge the current meteorological fluctuation level of the photovoltaic power generation system 100, thereby dynamically determining the target prediction frequency adapted to the current scenario. For example, when the meteorological fluctuation is stable, the target prediction frequency is lower to reduce unnecessary computing power consumption; when the meteorological fluctuation is severe, the target prediction frequency is higher to capture short-term environmental changes and improve prediction accuracy. After determining the target prediction frequency, the data processing device 520 performs power generation prediction according to that frequency.

[0048] In one embodiment, the data processing device 520 is further configured to: perform power prediction on the photovoltaic power generation system 100 after the prediction frequency of the power prediction system 500 is adjusted to the target prediction frequency. Thus, by adjusting the prediction frequency to the target prediction frequency and then performing power prediction according to the target prediction frequency, a closed-loop control from frequency adjustment to power prediction is achieved, enabling the prediction frequency of power prediction to be adjusted according to changes in meteorological conditions, thereby improving the accuracy of power prediction.

[0049] The power prediction system 500 provided in this application embodiment utilizes the operating data of the photovoltaic inverter 110 collected by the photovoltaic data acquisition device 511 and the satellite meteorological data of the geographic space to which the photovoltaic power generation system 100 belongs, received by the satellite ground station receiver 512, to construct a virtual cloud map. This allows the node characteristics corresponding to each node in the virtual cloud map to accurately represent the environmental and operating status of the corresponding photovoltaic subarray area, thereby solving the problem of the virtual cloud map data being singular and unable to reflect the local environmental differences of the photovoltaic subarray 200, and improving the perception accuracy of environmental changes at the photovoltaic subarray 200 level. Furthermore, after obtaining the virtual cloud map, the prediction frequency of power prediction is adjusted to the target prediction frequency according to the node characteristics corresponding to multiple nodes in the virtual cloud map. This allows the prediction frequency to adaptively adjust with changes in the operating status of the photovoltaic subarray 200 and meteorological conditions. For example, the prediction frequency is reduced when the meteorological fluctuations are stable to reduce computing power consumption, and the prediction frequency is increased when the meteorological fluctuations are severe to capture environmental changes and improve prediction accuracy, thereby achieving a balance between prediction accuracy and computing power efficiency.

[0050] In one implementation, the node features include mean brightness, variance brightness, and gradient brightness. The data processing device 520 is further configured to: obtain the mean brightness, variance brightness, and gradient brightness from the node features using irradiance data and satellite cloud image data.

[0051] Among them, the mean brightness reflects the illumination conditions of the geographical area where the photovoltaic subarray 200 is located, and the variance of brightness reflects the intensity of cloud fluctuations in the geographical area where the photovoltaic subarray 200 is located. The brightness gradient reflects the degree of spatial meteorological difference between the current photovoltaic subarray 200 and its neighboring photovoltaic subarrays 200. The calculation of the mean brightness, variance of brightness, and brightness gradient is explained below.

[0052] The average brightness is calculated based on the average solar irradiance within a preset time-series sliding window. A higher value indicates better local illumination conditions for the photovoltaic subarray 200. Specifically, the calculation formula is as follows:

[0053] In the formula: Let be the average brightness of the node corresponding to the i-th photovoltaic subarray 200; is the solar irradiance of the i-th photovoltaic subarray at the current moment; N is the number of sampling points in the preset time-series sliding window.

[0054] Brightness variance is calculated based on the temporal variance of solar irradiance within a sliding window. A larger variance indicates more pronounced local cloud cover and fluctuations in illumination, resulting in poorer temporal stability of the weather pattern. Specifically, the calculation formula is as follows:

[0055] In the formula: Let be the brightness variance of the node corresponding to the i-th photovoltaic subarray 200, and the other parameters are defined as above.

[0056] The brightness gradient is calculated based on the difference in average illumination between the current photovoltaic subarray 200 and its geographically adjacent photovoltaic subarray 200. A larger gradient indicates a more significant local cloud boundary crossing of the photovoltaic subarray 200, more pronounced uneven illumination at the site, and more severe abrupt changes in spatial weather. Specifically, the calculation formula is as follows:

[0057] In the formula: Let be the brightness gradient of the node corresponding to the i-th photovoltaic subarray 200; This is the set of geographically adjacent subarrays of the current photovoltaic subarray 200; This represents the number of adjacent photovoltaic subarrays. The average brightness of the adjacent photovoltaic subarray 200.

[0058] When calculating the aforementioned node characteristics, the irradiance data is based on surface irradiance information retrieved from satellite cloud imagery data. When the cloud layer, as represented by the satellite cloud imagery data, thickens or moves faster, the solar irradiance of the corresponding photovoltaic subarray 200 decreases and fluctuates more significantly, resulting in changes in the mean and variance of brightness. The brightness gradient is calculated based on the difference in mean brightness between the current photovoltaic subarray 200 and its neighboring subarrays. When the satellite cloud imagery data indicates spatial differences in cloud distribution across different subarrays, the brightness gradient increases accordingly.

[0059] In one implementation, the node characteristics also include node attribute consistency, and the data processing device 520 is further used to obtain node attribute consistency from the operating data. Node attribute consistency characterizes the overall operational stability of the photovoltaic inverters 110 connected to the photovoltaic subarray 200. It is calculated based on the dispersion of the electrical output parameters of the photovoltaic inverters 110 connected to the photovoltaic subarray 200. The closer the value is to 1, the higher the output consistency of the photovoltaic inverters 110 connected to the photovoltaic subarray 200, and the stronger the operational stability. Specifically, the calculation formula is as follows:

[0060] In the formula: To ensure the consistency of node attributes corresponding to the i-th photovoltaic subarray 200; Standard deviation of output power of photovoltaic inverter 110 connected to photovoltaic subarray 200; The average output power of the photovoltaic inverter 110 connected to the photovoltaic subarray 200.

[0061] In this way, the consistency of node attributes can be used to determine whether the power fluctuations in a photovoltaic power generation system originate from equipment malfunctions or weather changes. When the photovoltaic inverter 110 malfunctions or has an abnormal output, even if the weather conditions are stable, the power generation may fluctuate. Misjudging this fluctuation as a weather change could trigger high-frequency prediction, causing unnecessary computing power consumption and improving the accuracy of prediction frequency adjustment.

[0062] After obtaining the parameters of the node features, a relational graph reflecting the spatial structure of the photovoltaic power generation system 100 is constructed, using the photovoltaic inverters 110 connected to the photovoltaic subarray 200 as nodes and the spatial adjacency relationships between the photovoltaic subarrays 200 as edges. The mean brightness, variance brightness, node attribute consistency, and brightness gradient corresponding to each node are used as the four-dimensional feature attributes of that node, jointly representing the environmental and operational states of the corresponding photovoltaic subarray 200. This forms a spatial graph that simultaneously contains spatial relationship information and state attribute information, providing structured input data for subsequent graph neural network processing.

[0063] After obtaining the parameters of the node features, each node in the virtual cloud map corresponds to a photovoltaic inverter 110 connected to a photovoltaic subarray 200. The edges between nodes represent the spatial adjacency relationships between photovoltaic subarrays, thus forming a relationship graph that reflects the spatial structure of the photovoltaic power generation system. Each node is represented by four-dimensional features: mean brightness, brightness variance, node attribute consistency, and brightness gradient. These four-dimensional features collectively characterize the environmental and operational states of the area where the corresponding photovoltaic subarray is located, and serve as the attribute information of that node in the graph. Thus, the virtual cloud map integrates the spatial topology of the photovoltaic power generation system with the state characteristics of each photovoltaic subarray, forming a spatial graph that simultaneously contains spatial relationship information and state attribute information.

[0064] The power prediction system 500 provided in this application calculates the mean brightness, variance brightness, and gradient brightness using irradiance data and satellite cloud image data. It quantifies the physical states of each photovoltaic subarray 200, such as illumination conditions, cloud fluctuations, and spatial meteorological differences, into specific feature parameters. This enables the node features to characterize the meteorological conditions of the corresponding photovoltaic subarray 200 from three dimensions: illumination intensity, temporal fluctuations, and spatial differences. This improves the accuracy of the virtual cloud image in depicting the spatial distribution characteristics of the photovoltaic power generation system 100 and provides a calculable and comparable data basis for the adaptive adjustment of the prediction frequency.

[0065] In one implementation, such as Figure 3 The diagram shown is a schematic representation of another power prediction system 500 provided in an embodiment of this application. The power prediction system 500 also includes a meteorological data acquisition device 513 and a panoramic sky camera 514.

[0066] The meteorological data acquisition device 513 is used to collect surface meteorological data of the geographic space to which the photovoltaic power generation system is located. The surface meteorological data is used to represent the meteorological parameters of the ground where the photovoltaic power generation system is located, including at least one of wind speed, wind direction, temperature, humidity and solar irradiance of the geographic space to which the photovoltaic power generation system 100 is located at different times (e.g., hourly, daily, monthly or different seasons).

[0067] The meteorological data acquisition device 513 mentioned above can be a meteorological instrument, and multiple meteorological instruments can be installed in the geographic space to which the photovoltaic power generation system 100 belongs. By collecting ground meteorological data from multiple points, the spatial distribution differences of meteorological parameters in the geographic space to which the photovoltaic power generation system 100 belongs can be more comprehensively reflected, avoiding local errors caused by single-point collection, thereby improving the accuracy and representativeness of ground meteorological data.

[0068] The sky panoramic camera 514 is used to collect panoramic sky image data of the geographic space where the photovoltaic power generation system is located. This panoramic sky image data is used to capture local cloud cover conditions above the photovoltaic power generation system. For example, it can obtain cloud data such as cloud images, cloud height, cloud thickness, and cloud movement speed in the geographic space where the photovoltaic power generation system 100 is located, in order to monitor cloud changes in the geographic space where the photovoltaic power generation system 100 is located, thereby enabling the prediction of future weather conditions for the photovoltaic power generation system 100.

[0069] Ground-based meteorological data and panoramic sky imagery are obtained from ground observations, offering high accuracy but limited coverage. Satellite meteorological data, acquired through remote sensing by meteorological satellites, provides broad coverage but suffers from relatively lower resolution and accuracy compared to ground-based observations. By combining ground-based meteorological data and panoramic sky imagery, accurate information can be provided, compensating for the localized limitations of satellite meteorological data.

[0070] The aforementioned photovoltaic data acquisition device 511, meteorological acquisition equipment 513, sky panoramic camera 514, and satellite ground station receiver 512 can be collectively referred to as data acquisition equipment. Satellite meteorological data, sky panoramic image data, operational data, and ground meteorological data can be referred to as multimodal data.

[0071] In one implementation, after acquiring data from multiple devices, the data processing device 520 further performs data standardization. For example, it performs anomaly detection on the operational data, removing data points that exceed reasonable ranges or have incorrect formats to prevent abnormal data from contaminating subsequent processing; it fills in missing data due to device failure or communication interruption using interpolation algorithms to ensure the continuity of time-series data; it unifies multimodal data with different sampling frequencies and time axes onto the same time axis through resampling or timestamp alignment to ensure time synchronization of multimodal data; and it normalizes data with different dimensions, mapping them to a unified numerical range to avoid the impact of dimensional differences on model processing accuracy. After the above processing, standardized data is output. This dataset is complete, continuous, time-aligned, and dimensionally unified, avoiding the impact of data heterogeneity on virtual cloud map construction and prediction frequency inference accuracy.

[0072] In one implementation, the data processing device 520 is further used to obtain the mean brightness, variance brightness, and gradient brightness of the node features based on ground meteorological data, panoramic sky image data, irradiance data, and satellite cloud image data.

[0073] By combining ground meteorological data, panoramic sky image data, irradiance data, and satellite cloud image data, the mean brightness, variance brightness, and gradient brightness of multiple nodes in the virtual cloud image are determined. This enables the node features to more accurately reflect the illumination conditions, cloud fluctuations, and spatial meteorological differences of the geographical areas where each photovoltaic subarray 200 is located, thereby improving the accuracy of the node features in representing the actual meteorological environment of the photovoltaic subarray 200.

[0074] In one implementation, such as Figure 4 The diagram shown is a schematic representation of a virtual cloud map construction process provided in an embodiment of this application. Figure 4 As shown, the satellite ground station receiver 512, the sky panoramic camera 514, the photovoltaic data acquisition device 511, and the meteorological acquisition device 513 are used to collect satellite meteorological data, sky panoramic image data, operational data, and ground meteorological data of the photovoltaic power generation system 100, respectively. The photovoltaic subarray topology information and multimodal data are used together to construct a virtual cloud map. In this way, the data processing device 520 can construct nodes based on the photovoltaic subarray topology information and bind multimodal data according to the geographical location of each photovoltaic subarray 200, thereby achieving accurate mapping of the independent environmental and operational states of each photovoltaic subarray 200 in the virtual cloud map. This solves the problem of the virtual cloud map data being singular and unable to reflect the local differences of the photovoltaic subarray 200, improving the utilization efficiency of multimodal data and the perception accuracy of environmental changes at the photovoltaic subarray 200 level.

[0075] The power prediction system 500 provided in this application embodiment can accurately obtain the mean brightness, variance brightness, and gradient brightness in the node features through the fusion and complementation of multimodal data, avoiding information loss or bias caused by relying on a single data source, thereby improving the ability of node features to characterize the meteorological environment of the corresponding photovoltaic subarray 200.

[0076] In one implementation, the target prediction frequency includes a first frequency, a second frequency, and a third frequency, with the prediction intervals corresponding to the first frequency, the second frequency, and the third frequency decreasing sequentially.

[0077] Among them, the first frequency has the largest prediction interval and the lowest prediction frequency; the second frequency has a moderate prediction interval and a medium prediction frequency; and the third frequency has the smallest prediction interval and the highest prediction frequency. The first, second, and third frequencies correspond to different frequency ranges. One frequency from the three frequency ranges is selected as the target prediction frequency because it matches the current meteorological environment. Within the same frequency range, it can be fine-tuned according to the actual scenario and is not limited to a fixed frequency value.

[0078] For example, the first frequency is once every 15 minutes, suitable for scenarios with stable weather changes, to save computing power. The second frequency is once every 5 minutes, suitable for scenarios with moderate weather fluctuations, such as local small fluctuations, to balance prediction accuracy and computing power. The third frequency is 1 to 2 times per minute, suitable for scenarios with drastic weather fluctuations, such as local drastic fluctuations, to capture sudden changes with high-frequency sampling and avoid prediction bias.

[0079] In one embodiment, the data processing device 520 is further configured to adjust the target prediction frequency to a first frequency if the brightness variance of multiple nodes is less than or equal to a first variance threshold and the brightness gradient is less than or equal to a first gradient threshold; adjust the target prediction frequency to a second frequency if the brightness variance of multiple nodes is greater than the first variance threshold and less than or equal to a second variance threshold, and the brightness gradient is greater than the first gradient threshold and less than or equal to the second gradient threshold; and adjust the target prediction frequency to a third frequency if the brightness variance corresponding to any node among the multiple nodes is greater than the second variance threshold and / or the corresponding brightness gradient is greater than the second gradient threshold.

[0080] The aforementioned first variance threshold and first gradient threshold are used to distinguish between stable and moderate meteorological fluctuations, while the second variance threshold and second gradient threshold are used to distinguish between moderate and severe meteorological fluctuations. Specifically, when the brightness variance and brightness gradient are less than or equal to the first variance threshold and first gradient threshold, respectively, the meteorological fluctuation level is stable, and the first frequency with the largest prediction interval is used for prediction to reduce computing power consumption. When the brightness variance of multiple nodes is greater than the first variance threshold and less than or equal to the second variance threshold, and the brightness gradient is greater than the first gradient threshold and less than or equal to the second gradient threshold, the meteorological fluctuation level is moderate, and the second frequency with a medium prediction interval is used for prediction to balance prediction accuracy and computing power. When the brightness variance of any node is greater than the second variance threshold, and / or the brightness gradient is greater than the second gradient threshold, the meteorological fluctuation level is severe, and the third frequency with the smallest prediction interval is used for prediction to capture drastic changes and improve prediction accuracy. The above thresholds can be determined based on statistical analysis of the historical operating data of the photovoltaic power generation system 100, or can be dynamically adjusted according to actual operating conditions and prediction accuracy requirements, without limitation.

[0081] For example, the first variance threshold is 0.15, the first gradient threshold is 0.12, the second variance threshold is 0.55, and the second gradient threshold is 0.5. When the brightness variance corresponding to most nodes in the virtual cloud map is less than or equal to 0.15 and the brightness gradient is less than or equal to 0.12, the degree of meteorological fluctuation is determined to be stable; when the brightness variance corresponding to any node in the virtual cloud map is greater than 0.55 or the brightness gradient is less than 0.50, the degree of meteorological fluctuation is determined to be severe; when neither of the above two situations applies, the degree of meteorological fluctuation is determined to be moderate.

[0082] In this way, by setting two thresholds for brightness variance and brightness gradient respectively, the degree of meteorological fluctuations can be divided into three levels: stable, moderate and severe. The power prediction of the photovoltaic power generation system is then performed using the first frequency, the second frequency and the third frequency respectively, so as to achieve accurate matching between the prediction frequency and the degree of meteorological fluctuations and improve the prediction accuracy.

[0083] The power prediction system 500 provided in this application provides a multi-level selectable prediction frequency configuration by setting prediction frequency levels with sequentially decreasing prediction intervals for a first frequency, a second frequency, and a third frequency. In this way, the power prediction system 500 can select a matching prediction frequency from multiple prediction frequency levels as the target prediction frequency based on the meteorological conditions of multiple photovoltaic subarrays. This provides a tiered basis and flexible selection space for adjusting the prediction frequency, avoiding the problem that a single prediction frequency cannot adapt to multiple meteorological scenarios. Furthermore, the sequentially decreasing prediction intervals of the three frequency levels are suitable for three scenarios: stable meteorological changes, moderate meteorological fluctuations, and severe meteorological fluctuations, respectively. The first frequency is used when meteorological changes are stable to reduce system computing power consumption, while the third frequency is used when meteorological fluctuations are severe to capture rapid changes, achieving differentiated prediction frequency configurations for different meteorological scenarios.

[0084] In one embodiment, the data processing device 520 is specifically used to: extract spatial features from a virtual cloud image to obtain a spatial feature vector corresponding to the virtual cloud image; extract temporal features from irradiance data in satellite meteorological data to obtain a first temporal feature vector; and output the target prediction frequency based on the spatial feature vector and the first temporal feature vector.

[0085] The data processing device 520 extracts spatial features from the virtual cloud map to obtain spatial feature vectors that characterize the spatial topological and environmental distribution differences between the various photovoltaic subarrays 200, such as the adjacency relationship of the photovoltaic subarrays 200. It captures spatial dimension information such as the difference in illumination between adjacent photovoltaic subarrays 200, cloud boundary distribution, and local abrupt change regions. Then, it extracts temporal features from the irradiance data in the satellite meteorological data to obtain a first temporal feature vector characterizing the trend of illumination change. By fusing the spatial feature vector with the first temporal feature vector, it obtains a feature vector that simultaneously contains spatial difference information and irradiance change information, outputting the target prediction frequency so that the target prediction frequency can more comprehensively perceive changes in the meteorological environment.

[0086] In one embodiment, the data processing device 520 is further used to extract temporal features from irradiance data in satellite meteorological data, irradiance data in ground meteorological data, and irradiance data in panoramic sky image data, obtaining a first temporal feature vector. Thus, the first temporal feature vector integrates temporal information from three sources: satellite irradiance data, ground-measured irradiance data, and irradiance features obtained from image cloud changes, enabling the extracted temporal features to more comprehensively characterize changes in illumination conditions. Specifically, cloud coverage information, such as cloud coverage rate, cloud thickness, and cloud movement trend, is extracted from the panoramic sky image terminal using image processing algorithms, and this cloud coverage information is converted into irradiance data with irradiance-related information that can be used for temporal feature extraction.

[0087] In one embodiment, the data processing device 520 is further configured to extract time-series features from satellite meteorological data other than irradiance data and operational data to obtain a second time-series feature vector; and output the target prediction frequency based on the spatial feature vector, the first time-series feature vector and the second time-series feature vector.

[0088] Specifically, by extracting time-series features from satellite meteorological data other than irradiance data (such as temperature, humidity, wind speed, cloud parameters, etc.) and operational data, a second time-series feature vector is obtained. This second time-series feature vector is then combined with the spatial feature vector and the first time-series feature vector (irradiance-related time-series feature vector) to output the target prediction frequency. In this way, the target prediction frequency not only considers the time-series changes in light intensity and the spatial differences between photovoltaic subarrays (spatial feature vector), but also further incorporates other meteorological data and the power generation status of photovoltaic subarrays. This allows the prediction frequency to simultaneously perceive multidimensional changes in the meteorological environment and the operating status of equipment, thereby improving the adaptability of the prediction frequency to complex meteorological scenarios and equipment operating conditions.

[0089] In one embodiment, the data processing device 520 is further used to extract time-series features from satellite meteorological data, ground meteorological data, and panoramic sky image data (excluding irradiance data), as well as operational data, to obtain a second time-series feature vector. Thus, the second time-series feature vector integrates multi-source time-series information such as temperature, humidity, wind speed, and cloud parameters from the ground meteorological data, satellite meteorological data, and panoramic sky image data, enabling the extracted time-series features to more comprehensively characterize the multidimensional changes in the meteorological environment.

[0090] This application proposes a power prediction system 500, which extracts spatial features from a virtual cloud map to obtain a spatial feature vector, and extracts temporal features from irradiance data in multimodal data to obtain a first temporal feature vector. This system integrates the spatial topological differences between photovoltaic subarrays reflected in the virtual cloud map (such as differences in illumination between adjacent photovoltaic subarrays, cloud boundary distribution, etc.) with the temporal series change trends corresponding to the irradiance data (such as short-term fluctuations and gradual trends in illumination). This allows the target prediction frequency to take into account information in both spatial and temporal dimensions, enabling a more comprehensive perception of changes in the meteorological environment and thus outputting a more accurate prediction frequency.

[0091] The aforementioned processes of spatial feature extraction, temporal feature extraction, and feature fusion can be implemented using a graph spatial-temporal attention neural network (GSTANN), or other network models, without limitation.

[0092] In one implementation, such as Figure 5The diagram shown is an application illustration of frequency prediction acquisition provided in an embodiment of this application. Figure 5 The prediction frequency is obtained using GSTANN. The specific processing flow includes: The virtual cloud map is converted into graph structure data and input into the graph convolutional layer for spatiotemporal embedding. The spatial features of each node are extracted, corresponding to the process of obtaining the spatial feature vectors mentioned above.

[0093] The spatial feature vector and irradiance data from the multimodal data are input into the encoder. The encoder simultaneously extracts temporal features through a combined spatiotemporal attention mechanism to obtain the first temporal feature vector; and models the correlation between features from historical time steps and features from future time steps through attention transformation, mapping the hidden states of historical time steps to future time steps. This combined spatiotemporal attention mechanism models attention in both the temporal and spatial dimensions, capturing the variation patterns of irradiance data at different time steps and the spatial correlations between different nodes; the attention transformation is responsible for establishing the temporal dependency between the past and the future. The encoder outputs the first temporal feature vector after the above processing, providing transitional features for spatiotemporal fusion for the decoder.

[0094] The encoder and decoder fuse the spatial feature vector and the first temporal feature vector through a data fusion module. The fused feature vector contains both spatial difference information and irradiance variation information. Based on this, other data from the multimodal data besides irradiance data (such as temperature, humidity, wind speed, cloud parameters, etc.) are input into the data fusion module. The module extracts their temporal features to obtain a second temporal feature vector, which is then fused with the aforementioned spatial feature vector and the first temporal feature vector. This further enriches the information included in the fused feature vector and improves prediction accuracy.

[0095] in addition, Figure 5 Above the encoder is a schematic diagram of combined spatiotemporal attention processing. A, B, C, and D represent different nodes, t represents the current time, (t-1) represents the previous historical time, and (t-2) represents an even earlier historical time. Node A is the target node for attention computation, and each node can establish temporal relationships between different time steps. By constructing the correspondence between nodes and time and their interrelationships, it is possible to simultaneously mine the spatial dependencies between nodes, as well as the temporal dependencies of the output data of the same node over time.

[0096] Through the above process, GSTANN can learn the correlation between spatial topological differences, temporal variations in irradiance, temporal variations in other meteorological factors, and end-side power data from end to end, thereby outputting a prediction frequency that is adapted to the current meteorological environment.

[0097] The power prediction system 500 provided in this application embodiment integrates the spatial topological differences between photovoltaic subarrays 200 reflected in the virtual cloud map (such as the difference in illumination between adjacent photovoltaic subarrays, cloud boundary distribution, etc.) with the time series change trend reflected in the irradiance data (such as short-term fluctuations and gradual trends in illumination) through GSTANN. This allows the target prediction frequency to take into account information in both spatial and temporal dimensions, enabling a more comprehensive perception of changes in the meteorological environment and thus outputting a more accurate prediction frequency.

[0098] In one implementation, such as Figure 6 The diagram shows the composition of another power prediction system 500 provided in this application embodiment. The power prediction system 500 also includes a display device 530. The data processing device 520 is further configured to: render a virtual cloud map to obtain a visualized virtual cloud map; and display the visualized virtual cloud map on the display device 530.

[0099] In one implementation, in the visualized virtual cloud map, each node and its corresponding node features present the environmental and operational status of the area where the photovoltaic subarray corresponding to the node is located in at least one of the following ways: color depth, color dispersion, luminous uniformity, and color transition differences.

[0100] For example, the larger the mean brightness value, the darker the image color and the higher the luminous intensity of the corresponding node in the photovoltaic subarray; the smaller the value, the lighter the image color and the lower the luminous intensity, intuitively distinguishing the differences in light intensity among the various photovoltaic subarrays 200. The larger the brightness variance value, the wider the color divergence range of the node corresponding to the photovoltaic subarray 200, the more blurred the image edge, and the more obvious the dynamic fluctuation effect, representing violent cloud fluctuations; the smaller the value, the more regular the node image and the more convergent the color, representing a stable light condition. The larger the brightness gradient value, the more obvious the color and brightness transition difference between the current subarray node and its neighboring nodes, the clearer the image boundary and the more significant the light and dark partitioning; the smaller the value, the more uniform the color transition of all nodes, and the more consistent the overall meteorological conditions of the site.

[0101] In this way, the environmental and operational status of each photovoltaic subarray area can be presented in the visualized virtual cloud map through at least one of the following methods: color depth, color dispersion, luminous uniformity, and color transition differences. Specifically, color depth reflects the intensity of sunlight, color dispersion reflects the intensity of cloud fluctuations, luminous uniformity reflects the operational stability of the photovoltaic inverter 110, and color transition differences reflect spatial meteorological differences between photovoltaic subarrays. This transforms abstract feature data into intuitive visual elements, enabling maintenance personnel to easily identify status differences and abnormal areas in each photovoltaic subarray, improving monitoring efficiency and user experience.

[0102] The power prediction system 500 provided in this application embodiment generates and displays a visualized virtual cloud map by rendering the virtual cloud map, enabling operation and maintenance personnel to intuitively view the environmental and operational status of the area where each photovoltaic subarray is located, such as illumination conditions, cloud fluctuations, operational stability, and spatial meteorological differences. This facilitates monitoring the overall status of the photovoltaic power generation system 100, quickly locating abnormal photovoltaic subarrays 200, and improving the observability and operation and maintenance efficiency of the photovoltaic power generation system.

[0103] In one embodiment, the display device 530 is used to display an alarm signal when there is an anomaly in the data or when the prediction frequency adjustment of the power prediction system 500 fails.

[0104] When anomalies are detected in operational data, ground meteorological data, panoramic sky image data, or satellite meteorological data (such as missing data, data exceeding reasonable range, data acquisition interruption, etc.), or when prediction frequency adjustment fails (such as failing to successfully switch to the target prediction frequency after frequency adjustment), the data processing device 520 triggers an alarm signal. This alarm signal is sent to the display device 530 to present alarm information to maintenance personnel. Through this alarm mechanism, maintenance personnel can promptly detect data acquisition failures or system execution anomalies, quickly locate problems, and intervene, ensuring the operational reliability and maintainability of the power prediction system 500.

[0105] In one implementation, the power prediction system 500 also includes an application programming interface (API) module, which provides a standardized application programming interface, supports the deployment of Software as a Service (SaaS) services, and can be connected to the power grid 400 dispatch system and the power plant operation and maintenance management system to realize data interaction and command transmission.

[0106] Specifically, the API module provides interfaces for calling the following outputs: virtual cloud map, predicted frequency, power prediction results, frequency adjustment logs, and operation / schedule instructions. External platforms can obtain the above data in real time by calling the corresponding API interfaces, without needing to deploy a complete power prediction system locally.

[0107] In one embodiment, the power prediction system 500 further includes a data storage device 540. The data storage device 540 is used to store at least one of operational data, ground meteorological data, panoramic sky image data, or satellite meteorological data, and to transmit the stored data to the data processing device 520.

[0108] In this way, the data stored by the data storage device 540 is transmitted to the data processing device 520, ensuring centralized management and reliable transmission of the data, and providing complete historical data support for the data processing device 520.

[0109] In one embodiment, the data processing device 520 includes station auxiliary equipment, or the data processing device 520 includes station auxiliary equipment and a cloud server.

[0110] In one example, functions such as building a virtual cloud map and adjusting the frequency of the photovoltaic power generation system can be deployed on a cloud server for predicting the predicted frequency of the photovoltaic power generation system 100. In another example, these functions can be deployed on site auxiliary equipment, i.e., on the local server of the photovoltaic power generation system 100. In yet another example, these functions can be deployed on both site auxiliary equipment and a cloud server.

[0111] The power prediction system 500 provided in this application embodiment can flexibly deploy frequency prediction functions on the site side and the cloud side. It can complete local real-time calculations by relying on site auxiliary equipment, or achieve large-scale data processing and remote collaborative prediction with the help of cloud servers, thereby improving the applicability, flexibility and scalability of the power prediction system and meeting the deployment needs and computing power configurations in different scenarios.

[0112] Based on the above embodiments, Figure 7 This is a schematic diagram of the application architecture of a power prediction system 500 provided in an embodiment of this application. Figure 7 As shown, the power prediction system 500 provided in this application embodiment adopts a three-layer architecture, including data acquisition and storage, data processing device 520 and output application layer.

[0113] Data acquisition and storage includes a photovoltaic data acquisition unit 511, a meteorological acquisition device 513, a sky panoramic camera 514, a satellite ground station receiver 512, and a data storage device 540. The photovoltaic data acquisition unit 511 collects operational data from the photovoltaic inverters 110 connected to each photovoltaic subarray 200, such as output power, operating current, operating voltage, and operating status indicators. Multiple sky panoramic cameras 514 are distributed and installed at different locations in the geographical space where the photovoltaic power generation system 100 is located, collecting cloud images, cloud thickness, and cloud movement speed in the geographical area where the photovoltaic power generation system 100 is located. The satellite ground station receiver 512 receives satellite meteorological data such as regional cloud distribution data, cloud thickness, and cloud movement speed. The meteorological acquisition device 513 collects ground meteorological data such as solar irradiance, temperature, humidity, wind speed, and atmospheric pressure from the photovoltaic power generation system. The data storage device 540 stores multimodal data, archives historical data, and provides a data retrieval interface. In addition, the data acquisition and storage also stores photovoltaic subarray topology information, which includes the physical layout of photovoltaic subarray 200 (e.g., the coordinates, quantity, and spacing of photovoltaic subarray 200), the correspondence between photovoltaic subarray 200 and photovoltaic inverter 110, and the location of data acquisition equipment, serving as the skeleton for constructing the virtual cloud map.

[0114] The data processing device 520 includes a data preprocessing module, a virtual cloud map construction module, a cloud map analysis module, a GSTANN, and a controller. The data preprocessing module performs outlier removal, missing value completion, time-series alignment, and dimensional standardization on the collected multimodal data, outputting standardized multimodal data. The virtual cloud map construction module constructs a virtual cloud map based on the photovoltaic subarray topology information and multimodal data, with each node corresponding to a photovoltaic subarray 200. Node characteristics reflect the environmental and operational status of the area where the corresponding photovoltaic subarray is located. The cloud map analysis module extracts the node characteristics of each node in the virtual cloud map to determine the degree of meteorological fluctuations and adjusts the variance and gradient thresholds to adapt to different weather scenarios. The GSTANN is used for offline model training, online frequency prediction calculation, and adaptive update of model parameters. The controller performs frequency adjustment, triggers power prediction, provides data synchronization feedback, and triggers alarms in abnormal situations.

[0115] In addition, the display device 530 is used for cloud map visualization, weather status display, forecast frequency and power display, and supports manual intervention by maintenance personnel. The API module provides SaaS service interfaces and data interaction interfaces, supports integration with third-party systems, and uses encrypted data transmission to ensure communication security.

[0116] Furthermore, the data required by the application layer is displayed through display devices and API interfaces. Specifically, this includes virtual cloud maps, target predicted frequencies, power prediction results, frequency adjustment logs, and operation / schedule instructions.

[0117] The virtual cloud map supports real-time updates at the second level, annotates each node, and marks nodes with abnormal characteristics for display. The target prediction frequency supports adaptive switching between the first, second, and third frequencies, automatically matching the prediction frequency to the current level of meteorological fluctuations. Power prediction results include both short-term and ultra-short-term forecasts, continuously optimized based on prediction accuracy feedback. The frequency adjustment log records the time of each adjustment, the frequency level before and after the adjustment, the meteorological fluctuation level that triggered the adjustment, and the change in prediction accuracy after the adjustment. Operation and maintenance / dispatch instructions are output externally via API, including grid dispatch reference suggestions, power plant operation and maintenance guidance, and anomaly handling prompts, providing decision-making support for grid dispatchers and power plant operation and maintenance personnel.

[0118] This achieves a complete closed loop from data collection to intelligent decision-making and result output, solving the problems of insufficient data utilization and fixed prediction frequency that cannot adapt to short-term weather fluctuations, and improving prediction accuracy and computing efficiency.

[0119] Based on the power prediction system 500 described above, this application embodiment also provides a power prediction method for power prediction of a photovoltaic power generation system 100. The method includes: acquiring operating data of multiple photovoltaic inverters 110 and satellite meteorological data of the geographic space to which the photovoltaic power generation system 100 belongs; constructing a virtual cloud map based on the operating data of multiple photovoltaic inverters 110 and the satellite meteorological data; obtaining node characteristics corresponding to multiple nodes of the virtual cloud map; the multiple nodes correspond one-to-one with multiple photovoltaic subarrays; and adjusting the prediction frequency of power prediction to the target prediction frequency based on the node characteristics.

[0120] In one implementation, the node features include mean brightness, variance brightness, and gradient brightness. Satellite meteorological data includes irradiance data and satellite cloud imagery data of the geographic space to which the photovoltaic power generation system is located. The method further includes obtaining the mean brightness, variance brightness, and gradient brightness from the node features based on the irradiance data and the satellite cloud imagery data.

[0121] In one implementation, the target prediction frequency includes a first frequency, a second frequency, and a third frequency, with the prediction intervals corresponding to the first frequency, the second frequency, and the third frequency decreasing sequentially.

[0122] In one implementation, adjusting the prediction frequency of power prediction to a target prediction frequency based on node characteristics includes: if the brightness variance of multiple nodes is less than or equal to a first variance threshold and the brightness gradient is less than or equal to a first gradient threshold, adjusting the target prediction frequency to a first frequency; if the brightness variance of multiple nodes is greater than the first variance threshold and less than or equal to a second variance threshold, and the brightness gradient is greater than the first gradient threshold and less than or equal to the second gradient threshold, adjusting the target prediction frequency to a second frequency; if the brightness variance corresponding to any of the multiple nodes is greater than the second variance threshold and / or the corresponding brightness gradient is greater than the second gradient threshold, adjusting the target prediction frequency to a third frequency.

[0123] In one implementation, obtaining the mean brightness, variance brightness, and gradient brightness of node features from irradiation data and satellite cloud image data includes: obtaining the mean brightness, variance brightness, and gradient brightness of node features based on ground meteorological data, panoramic sky image data, irradiation data, and satellite cloud image data.

[0124] In one embodiment, the method further includes: extracting spatial features from the virtual cloud image to obtain a spatial feature vector corresponding to the virtual cloud image; extracting temporal features from the irradiance data in the satellite meteorological data to obtain a first temporal feature vector; and outputting the target prediction frequency based on the spatial feature vector and the first temporal feature vector.

[0125] In one implementation, the target prediction frequency is output based on the spatial feature vector and the first temporal feature vector, including: extracting temporal features from satellite meteorological data (excluding irradiance data) and operational data to obtain a second temporal feature vector; and outputting the target prediction frequency based on the spatial feature vector, the first temporal feature vector, and the second temporal feature vector.

[0126] In one embodiment, after the prediction frequency of the power prediction system is adjusted to the target prediction frequency, the method further includes: performing power prediction on the photovoltaic power generation system 100.

[0127] In one embodiment, the method includes: rendering a virtual cloud map to obtain a visualized virtual cloud map; and displaying the visualized virtual cloud map.

[0128] In one implementation, in the visualized virtual cloud map, each node and its corresponding node features present the environmental and operational status of the area where the photovoltaic subarray corresponding to the node is located in at least one of the following ways: color depth, color dispersion, luminous uniformity, and color transition differences.

[0129] In one embodiment, this application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware. The program can be stored in the computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be any of the foregoing embodiments or memory. The computer-readable storage medium can also be an external storage device of the service invocation device, such as a pluggable hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the service invocation device. Further, the computer-readable storage medium can include both internal storage units of the service invocation device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the service invocation device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0130] In one embodiment, this application also provides a computer program product comprising a computer program that, when run on a computer, causes the computer to perform any of the methods described in the above embodiments.

[0131] The above detailed description of the power prediction system 500 and the analysis of its beneficial effects can be applied to power prediction methods, computer-readable storage media, and computer program products, and will not be repeated here in the embodiments of this application.

[0132] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A power prediction system, characterized in that, The power prediction system for a photovoltaic power generation system includes a photovoltaic data acquisition unit, a satellite ground station receiver, and data processing equipment; the photovoltaic power generation system includes a photovoltaic inverter and photovoltaic subarrays, with multiple photovoltaic subarrays respectively connected to one or more photovoltaic inverters. The photovoltaic data acquisition device is used to: collect operating data from multiple photovoltaic inverters; The satellite ground receiver is used to: receive satellite meteorological data of the geographic space to which the photovoltaic power generation system belongs; The data processing device is used to: construct a virtual cloud map based on the operating data of multiple photovoltaic inverters and the satellite meteorological data, obtain the node characteristics corresponding to multiple nodes of the virtual cloud map, wherein the multiple nodes correspond one-to-one with multiple photovoltaic subarrays, and adjust the prediction frequency of power prediction to the target prediction frequency based on the node characteristics.

2. The power prediction system according to claim 1, characterized in that, The node features include mean brightness, variance of brightness, and brightness gradient; the satellite meteorological data includes irradiance data and satellite cloud image data of the geographic space to which the photovoltaic power generation system belongs; The data processing device is also used to: obtain the mean brightness, variance brightness, and gradient brightness of the node features from the irradiation data and the satellite cloud image data.

3. The power prediction system according to claim 1 or 2, characterized in that, The target prediction frequency includes a first frequency, a second frequency, and a third frequency, and the prediction intervals corresponding to the first frequency, the second frequency, and the third frequency decrease sequentially.

4. The power prediction system according to claim 3, characterized in that, The data processing device is further configured to: if the brightness variance of the plurality of nodes is less than or equal to a first variance threshold and the brightness gradient is less than or equal to a first gradient threshold, adjust the target prediction frequency to a first frequency; If the brightness variance of the plurality of nodes is greater than the first variance threshold and less than or equal to the second variance threshold, and the brightness gradient is greater than the first gradient threshold and less than or equal to the second gradient threshold, the target prediction frequency is adjusted to the second frequency. If the luminance variance of any node among the plurality of nodes is greater than the second variance threshold and / or the corresponding luminance gradient is greater than the second gradient threshold, the target prediction frequency is adjusted to the third frequency.

5. The power prediction system according to claim 2, characterized in that, The power prediction system also includes meteorological data acquisition equipment and a panoramic sky camera; The meteorological data acquisition equipment is used to: collect surface meteorological data of the geographic space to which the photovoltaic power generation system belongs; The sky panoramic camera is used to: collect sky panoramic image data of the geographic space to which the photovoltaic power generation system belongs; The data processing device is also used to: obtain the mean brightness, the variance brightness, and the gradient brightness in the node features based on the ground meteorological data, the sky panoramic image data, the irradiance data, and the satellite cloud image data.

6. The power prediction system according to any one of claims 1-5, characterized in that, The data processing device is specifically used for: Spatial features are extracted from the virtual cloud map to obtain the spatial feature vector corresponding to the virtual cloud map; Temporal features are extracted from the irradiance data in the satellite meteorological data to obtain a first temporal feature vector; The target prediction frequency is output based on the spatial feature vector and the first temporal feature vector.

7. The power prediction system according to claim 6, characterized in that, The data processing device is also used for: Temporal features are extracted from the satellite meteorological data excluding irradiance data and the operational data to obtain a second temporal feature vector; The target prediction frequency is output based on the spatial feature vector, the first temporal feature vector, and the second temporal feature vector.

8. The power prediction system according to any one of claims 1-7, characterized in that, The data processing device is also used for: After the prediction frequency of the power prediction system is adjusted to the target prediction frequency, power prediction is performed on the photovoltaic power generation system.

9. The power prediction system according to any one of claims 1-8, characterized in that, The power prediction system further includes a display device; the data processing device is also used for: The virtual cloud map is rendered to obtain a visualized virtual cloud map; and the visualized virtual cloud map is displayed on the display device.

10. The power prediction system according to claim 9, characterized in that, In the visualized virtual cloud map, each node and its corresponding node features present the operating status of the photovoltaic subarray corresponding to the node and the meteorological conditions of the area in at least one of the following ways: color depth, color dispersion, luminous uniformity, and color transition differences.

Images