A data-driven photovoltaic energy storage system control method and system

Abstract

The application discloses a kind of photovoltaic energy storage system control method and system based on data driving, it is related to power grid power control technical field, including, acquisition photovoltaic environment data, and pretreatment;Satellite cloud picture data is handled by pre-training light spot prediction model, the light spot coverage track is predicted, and output fusion feature dataset;Based on the light spot influence time window and current deviation value in fusion feature dataset, power loss prediction curve is constructed, power influence conversion is carried out to the coverage rate of shelter, and power correction amount is generated;Photovoltaic energy storage is carried out double-channel compensation, and compensation instruction set is generated;Power adjustment is carried out to photovoltaic energy storage adjacent to the position of shelter, and control completion signal is output;After power sensor receives control completion signal, the effect of photovoltaic energy storage control is judged.The application realizes the dynamic accurate calibration of photovoltaic shading compensation by the cooperation of satellite cloud picture driven light spot coverage track modeling and spatial verification of string current data.

Description

Technical Field

[0001] This invention relates to the field of power grid control technology, and in particular to a data-driven photovoltaic energy storage system control method and system. Background Technology

[0002] In recent years, photovoltaic (PV) energy storage system control technology has gradually evolved towards multi-source sensing and data-driven collaboration. Mainstream solutions rely on satellite cloud imagery data to construct sliver trajectory prediction models, combine image recognition technology to locate shading areas of components, and verify prediction reliability using current sensors to achieve early warning of PV power output fluctuations. Among these, power correction mechanisms based on historical contamination data have significantly improved the accuracy of local shading compensation. Breakthrough progress has been made in the application of reinforcement learning in charging and discharging strategy optimization. Dynamic compensation commands coordinate the suppression of both sliver and shading disturbances, driving the evolution of PV energy storage control from static threshold response to predictive regulation. The deep integration of Supervisory Control and Data Acquisition (SCADA) with power sensors provides a closed-loop verification foundation for grid-connected power stability.

[0003] However, existing technologies have shortcomings in terms of data fusion accuracy and compensation coordination. Traditional spot prediction models lack a spatiotemporal verification mechanism for current attenuation values, leading to increased trajectory prediction errors in scenarios with sudden cloud changes. Static power correction strategies fail to establish a dynamic mapping relationship between shading coverage and historical pollution data, resulting in a mismatch between local compensation amounts and actual power losses. Furthermore, the command execution and power monitoring processes lack a real-time feedback loop, and strategy regeneration cannot be triggered when compensation deviations exceed thresholds, making it difficult to meet the stringent requirements of grid frequency regulation in scenarios with a high proportion of renewable energy access. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a data-driven photovoltaic energy storage system control method to solve the problems of deficiencies in data fusion accuracy and compensation coordination.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a data-driven control method for a photovoltaic energy storage system, comprising: collecting photovoltaic environmental data and preprocessing it; the photovoltaic environmental data including satellite cloud image data, photovoltaic array images, string current data, and historical contamination data; processing the satellite cloud image data using a pre-trained spot prediction model to predict the spot coverage trajectory and identify the coverage rate of obstructions in the photovoltaic array image; verifying the reliability of the spot coverage trajectory and obstruction coverage rate using the current attenuation value in the string current data, and outputting a fused feature dataset; and based on the spot influence time window in the fused feature dataset... Based on the current deviation value, a power loss prediction curve is constructed, and the power impact of the shading coverage rate is converted to generate a power correction amount. The photovoltaic energy storage is compensated in two channels using the power loss prediction curve and the power correction amount, generating a compensation instruction set. The monitoring and control center adjusts the discharge power of the energy storage converter according to the compensation instruction set and adjusts the power of photovoltaic energy storage adjacent to the shading location, outputting a control completion signal. After receiving the control completion signal, the power sensor continuously monitors the actual power value of the photovoltaic energy storage, compares the actual power value with the target power value in the compensation instruction set, and judges the effect of photovoltaic energy storage control.

[0008] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the specific steps of processing satellite cloud image data through a pre-trained spot prediction model to predict the spot coverage trajectory are as follows:

[0009] Satellite cloud image data is input into a pre-trained spot prediction model. The convolutional neural network of the spot prediction model is used to extract cloud motion vector information and generate a cloud motion vector map.

[0010] Based on the speed and direction information in the cloud motion vector map, a linear interpolation algorithm is used to obtain the cloud movement path for a fixed period in the future. The cloud movement paths are then integrated to generate a cloud coverage trajectory map.

[0011] Convert the latitude and longitude coordinates in the cloud cover trajectory map into the planar coordinates of the photovoltaic array;

[0012] The component areas of the shaded photovoltaic array are marked according to the planar coordinates, and the component areas are integrated according to the time sequence of the cloud movement path to output the light spot coverage trajectory.

[0013] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the specific steps for identifying the coverage rate of obstructions in the photovoltaic array image are as follows:

[0014] Extract the bounding box coordinates of the shading objects from the photovoltaic array image, align the bounding box coordinates of the shading objects with the pixel coordinate system of the photovoltaic array image, and generate shading data;

[0015] Map the occlusion data to the area covered by the occlusion;

[0016] The system calculates the shading status of the area covered by obstructions relative to the total area of ​​the photovoltaic array and outputs the shading coverage rate.

[0017] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the following steps are taken: The reliability verification of the light spot coverage trajectory and obstruction coverage rate is performed using the current attenuation value in the string current data, and a fused feature dataset is output.

[0018] The current attenuation value in the string current data is matched with the light spot coverage trajectory by timestamp to generate the occlusion-current relationship table.

[0019] Identify component regions whose current attenuation values ​​are consistent with the changing trend of the light spot coverage trajectory from the shading-current relationship table, and mark them as light spot regions;

[0020] The component association information of current attenuation value is spatially compared with the component area in the shading coverage to obtain the spatial distribution pattern of current attenuation value and shading coverage.

[0021] Clustering algorithms are used to identify component regions in the coverage of obstructions that have a spatial distribution pattern consistent with the current attenuation value, and these regions are marked as obstructed areas.

[0022] The light spot coverage trajectory and occlusion coverage are integrated based on the light spot region and the occlusion region to output a fused feature dataset.

[0023] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the specific steps for constructing a power loss prediction curve based on the light spot influence time window and current deviation value in the fused feature dataset are as follows:

[0024] Extract the time window of the photovoltaic module's light spot influence and the corresponding current deviation value from the fused feature dataset;

[0025] The time windows of multiple light spots affecting the same photovoltaic module are sorted by timestamp to generate a deviation sequence;

[0026] Map the current deviation values ​​in the deviation sequence to power loss values ​​and output power loss data;

[0027] The power loss data is integrated according to the timestamps of the photovoltaic array to generate a power loss sequence;

[0028] A continuous curve is plotted along the time axis of the power loss sequence to serve as the power loss prediction curve.

[0029] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the specific steps of converting the power impact of the obstruction coverage rate to generate a power correction amount are as follows:

[0030] Set up a loss mapping table based on historical soiling data;

[0031] Extract component information labeled as occlusion regions from the fused feature dataset, calculate the proportion of occlusion area in each component information, and output the occlusion proportion data;

[0032] The power loss value is output by matching and looking up the loss mapping table and the occlusion percentage data.

[0033] The power loss value is combined with the rated power of the photovoltaic module to generate a power correction value.

[0034] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the step of performing dual-channel compensation of photovoltaic energy storage using power loss prediction curves and power correction amounts to generate a compensation instruction set includes the following specific steps:

[0035] Extract the power loss value from the power loss prediction curve and overlay it with the power correction value in the power correction amount according to the timestamp to generate a comprehensive compensation sequence;

[0036] Based on the total power compensation demand in the comprehensive compensation sequence and combined with the control protocol of photovoltaic energy storage, a compensation instruction set is generated.

[0037] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the power regulation of photovoltaic energy storage adjacent to the location of the obstruction refers to increasing the discharge power of photovoltaic energy storage adjacent to the obstruction area, making up for the power loss value of the obstruction area, and completing the power regulation.

[0038] As a preferred embodiment of the data-driven photovoltaic energy storage system control method of the present invention, the step of comparing the actual power value with the target power value in the compensation instruction set to determine the effect of photovoltaic energy storage control refers to calculating the power difference between the actual power value and the target power value using the difference method; if the power difference exceeds a preset power threshold, the compensation instruction set is dynamically corrected and a correction instruction set is output; if the power difference does not exceed the preset power threshold, the photovoltaic energy storage control effect is determined to be satisfactory.

[0039] Secondly, the present invention provides a data-driven photovoltaic energy storage system control system, comprising,

[0040] The data acquisition module is used to collect photovoltaic environmental data and perform preprocessing; the photovoltaic environmental data includes satellite cloud image data, photovoltaic array images, string current data, and historical pollution data.

[0041] The prediction module is used to process satellite cloud image data through a pre-trained spot prediction model, predict the spot coverage trajectory, and identify the coverage of obstructions in the photovoltaic array image; the reliability of the spot coverage trajectory and obstruction coverage is verified by the current attenuation value in the string current data, and the fused feature dataset is output.

[0042] The conversion module is used to construct a power loss prediction curve based on the light spot influence time window and current deviation value in the fused feature dataset, perform power influence conversion on the shading coverage, and generate power correction amount;

[0043] The instruction module is used to perform dual-channel compensation for photovoltaic energy storage based on the power loss prediction curve and the power correction amount, and generate a compensation instruction set.

[0044] The adjustment module is used to monitor and control the center to adjust the discharge power of the energy storage converter according to the compensation instruction set, and to issue local compensation instructions to photovoltaic energy storage adjacent to the location of the obstruction, and output control completion signal;

[0045] The judgment module is used to continuously monitor the actual power value of photovoltaic energy storage after the power sensor receives the control completion signal, compare the actual power value with the target power value in the compensation instruction set, and judge the effect of photovoltaic energy storage control.

[0046] The beneficial effects of this invention are as follows: By combining satellite cloud image-driven light spot coverage trajectory modeling with spatial verification of string current data, dynamic and accurate calibration of photovoltaic shading compensation is achieved. Cloud motion vectors are extracted using a convolutional neural network and a planar coordinate mapping is constructed, dynamically transforming the three-dimensional cloud layer into a spatiotemporal sequence of photovoltaic module shading. This provides high-resolution trajectory data for power loss prediction. By comparing the spatial distribution of current attenuation values ​​with shading data through clustering, misjudged areas are eliminated and shading boundaries are corrected, generating a fused feature dataset. This ensures multidimensional consistency between light spot prediction and actual shading status. A dual-channel compensation strategy is used to adjust discharge power, achieving dynamic power balance between the shading area and adjacent energy storage units. Ultimately, this achieves continuous and stable output of photovoltaic energy storage and optimal allocation efficiency of energy storage resources under fluctuating illumination. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of a data-driven photovoltaic energy storage system control method.

[0049] Figure 2 This is a schematic diagram of a data-driven photovoltaic energy storage system control system.

[0050] Figure 3 A flowchart for generating power correction values.

[0051] Figure 4 Flowchart for generating compensation instruction set. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0055] Reference Figures 1-4 This is one embodiment of the present invention, which provides a data-driven photovoltaic energy storage system control method, including the following steps:

[0056] S1. Collect photovoltaic environmental data and perform preprocessing.

[0057] Photovoltaic environmental data includes satellite cloud imagery, photovoltaic array images, string current data, and historical contamination data;

[0058] Specifically, satellite cloud imagery data is acquired through cloud images provided by meteorological satellites, containing information on cloud distribution, density, and location, reflecting the potential shading effects on the photovoltaic array. The satellite cloud imagery data is in high-resolution format, including latitude and longitude coordinates and timestamps. Photovoltaic array images are collected by cameras installed at the photovoltaic power station, recording visual information about the array surface, including any dust, bird droppings, or other obstructions. The photovoltaic array images are represented in pixel coordinates and include a timestamp. String current data is collected from the inverters or current sensors of the photovoltaic array, recording the real-time current value of each array, reflecting its power generation performance. String current data is stored in time-series format, including string number and timestamp. Historical contamination data is collected from the maintenance records of the photovoltaic power station, including the type, area, and impact of surface contamination on the photovoltaic array over a past period. Historical contamination data is stored in tabular format, including timestamps, photovoltaic module numbers, and contamination descriptions.

[0059] Preprocessing includes data cleaning, missing value imputation, and normalization.

[0060] Specifically, data cleaning targets satellite cloud imagery, photovoltaic array images, string current data, and historical contaminated data, removing outliers and invalid data. For satellite cloud imagery, high-resolution images are checked to ensure they contain complete cloud information, and blurry or blank images caused by transmission errors are removed. For photovoltaic array images, black or overexposed images caused by camera malfunctions are deleted. For string current data, data points with current values ​​outside the normal range (such as negative values) are removed. For historical contaminated data, entries with incomplete records or missing timestamps are removed.

[0061] Missing value imputation is handled separately for each data type of photovoltaic environmental data: for missing pixels in satellite cloud image data, linear interpolation is performed using cloud images with adjacent timestamps; for missing frames in photovoltaic array images, the average pixel values ​​of the preceding and following frames are used for imputation; for missing values ​​in string current data, linear interpolation is performed based on the current values ​​of the same string at preceding and following times; for missing records in historical contamination data, the most recently recorded contamination information is used for imputation.

[0062] Normalization processing converts all photovoltaic environmental data into a unified dimension: the pixel grayscale values ​​of satellite cloud image data are mapped to the range of 0 to 1; the pixel values ​​of photovoltaic array images are standardized to the range of 0 to 1; string current data are normalized to the range of 0 to 1 according to the string rated current; and the proportion of polluted area in historical pollution data is normalized to the range of 0 to 1.

[0063] S2. The satellite cloud image data is processed by a pre-trained spot prediction model to predict the spot coverage trajectory and identify the coverage of obstructions in the photovoltaic array image. The reliability of the spot coverage trajectory and obstruction coverage is verified by the current attenuation value in the string current data, and the fused feature dataset is output.

[0064] Satellite cloud image data is input into a pre-trained spot prediction model. The convolutional neural network of the spot prediction model is used to extract cloud motion vector information and generate a cloud motion vector map.

[0065] Specifically, the pre-trained spot prediction model is trained based on historical satellite cloud image data, which includes cloud distribution and corresponding timestamps. The training process uses a convolutional neural network architecture, with continuous time-series satellite cloud image data as input and cloud motion vector information as output. The training data generates cloud motion vector information by labeling cloud movement trajectories and recording the positional changes of clouds in continuous satellite images. The cloud movement trajectories are compared with the cloud motion vector information, and the backpropagation algorithm is used to adjust the cloud motion vector information through gradient descent to generate the trained spot prediction model.

[0066] Each satellite cloud image is input into a trained spot prediction model, which is built on a convolutional neural network (CNN). The CNN extracts cloud edge, shape, and movement features through multiple convolutional operations. During the CNN processing, the pixel grayscale values ​​of the satellite cloud image data are scanned layer by layer to identify continuous changes in cloud regions and generate feature vectors describing the direction and speed of cloud movement. The feature vectors describing the direction and speed of cloud movement are differentially processed, and the displacement of cloud regions at adjacent time points is compared to generate cloud motion vector information. The cloud motion vector information includes the movement speed and direction of each cloud region. Based on the original resolution of the satellite cloud image data, the cloud motion vector information is mapped to two-dimensional image coordinates to generate a cloud motion vector map.

[0067] Based on the speed and direction information in the cloud motion vector map, a linear interpolation algorithm is used to obtain the cloud movement path for a fixed period in the future. The cloud movement paths are then integrated to generate a cloud coverage trajectory map.

[0068] Specifically, the process extracts the speed and direction information of each cloud region from the cloud motion vector map. Speed ​​represents the distance the cloud moves per second, and direction represents the angle of movement. Starting from the current timestamp, a fixed future time period (e.g., one hour) is defined, and the cloud motion vector map is analyzed sequentially over time. A linear interpolation algorithm is used to process the speed and direction information at consecutive time points, statistically analyzing the positional changes of cloud regions within the fixed future time period. The specific operation is as follows: for each cloud region, based on the latitude and longitude coordinates and motion vector at the current time point, the latitude and longitude coordinates and motion vector at the next time point are compared. This comparison is repeated until the entire fixed time period is covered, generating a cloud movement path. The cloud movement path includes a timestamp and corresponding latitude and longitude coordinates. The cloud movement paths of all cloud regions are sorted by timestamp and integrated into a continuous trajectory, generating a cloud coverage trajectory map.

[0069] Convert the latitude and longitude coordinates in the cloud cover trajectory map into the planar coordinates of the photovoltaic array;

[0070] The process involves acquiring the geographic information of a photovoltaic (PV) power station, including the actual latitude and longitude range and physical layout of the PV array. The physical layout is represented by planar coordinates, which are a two-dimensional grid. Each grid point corresponds to a physical location of the PV array. A mapping relationship is established between latitude and longitude coordinates and the planar coordinates of the PV array. This mapping relationship is based on the geographic boundaries of the PV power station, defined by its minimum and maximum latitude and longitude. For each cloud region in the cloud cover trajectory map, the latitude and longitude coordinates are read, mapped, and the planar coordinates of the PV array are output.

[0071] The component areas of the shaded photovoltaic array are marked according to the planar coordinates. The component areas are integrated according to the time sequence of the cloud movement path, and the predicted light spot coverage trajectory is output.

[0072] Traverse the planar coordinates in the cloud coverage trajectory map, identify grid points that overlap with the planar coordinates of the photovoltaic array, mark the overlapping grid points as the areas of the photovoltaic array components that are shaded, and mark them with the photovoltaic component number and the corresponding timestamp; organize all the areas of the photovoltaic array components that are shaded according to the timestamp order of the cloud movement path, and generate the predicted light spot coverage trajectory.

[0073] Extract the bounding box coordinates of the shading objects from the photovoltaic array image, align the bounding box coordinates of the shading objects with the pixel coordinate system of the photovoltaic array image, and generate shading data;

[0074] Specifically, the process involves identifying obstructions from the photovoltaic array image. Obstructions are areas with a different color from the background of the photovoltaic panel, such as dust or bird droppings. An edge detection algorithm is applied to scan the photovoltaic array image to extract the outlines of the obstructions and generate bounding box coordinates. The physical layout of the photovoltaic array is then obtained, represented by a pixel coordinate system. The pixel coordinate system is a two-dimensional grid, and the grid points correspond to the physical locations of the photovoltaic array. The bounding box coordinates of the obstructions are aligned with the pixel coordinate system of the photovoltaic array image. The alignment process involves scaling and displacement to ensure that the bounding box coordinates are consistent with the physical locations of the photovoltaic array. The aligned bounding box coordinates record the position and extent of the obstructions, generating obstruction data. The obstruction data includes the bounding box coordinates of the obstructions, the corresponding pixel coordinates, and a timestamp.

[0075] Map the occlusion data to the area covered by the occlusion;

[0076] The process iterates through the bounding box coordinates of shading objects in the shading data. For each bounding box coordinate, the coverage area in the pixel coordinate system is determined by identifying the pixels within the bounding box. The coverage area of ​​the shading object bounding box coordinates is compared with the pixel coordinate system of the photovoltaic array. Grid points within the coverage area are marked, and these grid points correspond to the module areas of the photovoltaic array, generating shading object coverage areas. Shading object coverage areas represent the parts of the photovoltaic array module areas covered by dust, bird droppings, or other shading objects. The shading object coverage areas are then organized, and the photovoltaic module number, pixel coordinates, and timestamp for each module area are recorded to ensure consistency with the timestamps in the shading data.

[0077] The shading status of the area covered by obstructions relative to the total area of ​​the photovoltaic array is statistically analyzed, and the shading coverage rate is output.

[0078] For each shading area, extract the pixel coordinates of the coverage area, count the number of pixels within the coverage area, and calculate the proportion of the covered pixels to the total pixel area of ​​the corresponding component area. Traverse the shading areas by timestamp, and for each photovoltaic module number, summarize the area percentage at the corresponding time point to generate a shading status. The shading status describes the degree of shading for each component area at a specific time point, including the photovoltaic module number, area percentage, and timestamp. Compile the shading status of all component areas to generate the shading coverage rate, which is tabular data.

[0079] The formula for the coverage rate of obstructions is:

[0080] ;

[0081] in, This indicates the coverage rate of the area covered by the obstruction. This indicates the number of pixels in the area covered by the occluder. This represents the total pixel area of ​​the component region, i.e., the total number of pixels.

[0082] The current attenuation value in the string current data is matched with the light spot coverage trajectory by timestamp to generate the occlusion-current relationship table.

[0083] Specifically, the current attenuation value of each module region is extracted from the string current data; the photovoltaic module number, timestamp, and corresponding shading status recorded in the spot coverage trajectory are read synchronously; the current attenuation value and the spot coverage trajectory are sorted by the timestamp field, and the current attenuation value of the module region under the same timestamp is associated with the shading status in the spot coverage trajectory. If there is a discrepancy in the timestamp, the time alignment is corrected by linear interpolation; the matching results are organized into a table to form a shading-current relationship table.

[0084] Identify component regions whose current attenuation values ​​are consistent with the changing trend of the light spot coverage trajectory from the shading-current relationship table, and mark them as light spot regions;

[0085] Traverse the records in the shading-current relationship table, compare the current attenuation value sequence of each component region with the shading state sequence in the spot coverage trajectory over time, obtain the changing trend of the current attenuation value and the spot coverage trajectory. If the changing trend is consistent, it is determined that there is a significant correlation between the current attenuation value of the component region and the spot shading state; mark the significantly correlated component regions as spot regions.

[0086] The component association information of current attenuation value is spatially compared with the component area in the shading coverage to obtain the spatial distribution pattern of current attenuation value and shading coverage.

[0087] The correspondence between photovoltaic module numbers and current attenuation values ​​is extracted from the shading-current relationship table. Simultaneously, the photovoltaic module numbers, area percentages, and timestamps recorded in the shading coverage table are read. A mapping index for photovoltaic module numbers is established, aligning the current attenuation values ​​of the same photovoltaic module number with the area percentages of the shading coverage by timestamp. The spatial location of the current attenuation value and area percentage for each module region is compared. By comparing the physical locations of the module regions in the photovoltaic array's planar coordinate system, the spatial distribution overlap is obtained. The propagation paths and density differences of current attenuation values ​​and area percentages between adjacent module regions are statistically analyzed. A spatial distribution pattern describing the spatial correlation between module association information and the spatial relationship between module regions in the shading coverage is generated. The spatial distribution pattern includes the grid coordinates of the module region, the current attenuation gradient, the shading coverage gradient, and the spatial correlation coefficient.

[0088] Clustering algorithms are used to identify component regions in the coverage of obstructions that have a spatial distribution pattern consistent with the current attenuation value, and these regions are marked as obstructed areas.

[0089] Based on spatial distribution patterns, the current attenuation gradient, shading coverage gradient, and spatial correlation coefficient of the component regions are extracted as clustering features. Density-based clustering algorithms (such as DBSCAN) are used to group the component regions. During the clustering process, the spatial correlation coefficient is used as the core indicator. A minimum neighborhood density threshold (e.g., 5 adjacent component regions) and a maximum distance threshold (e.g., 3 meters) are set. The minimum neighborhood density threshold and the maximum distance threshold are set based on the physical layout characteristics of the photovoltaic array and the spatial propagation law of current attenuation. Component regions with consistent spatial distribution patterns in the clustering results are marked as shading regions. The marking information includes the photovoltaic component number, cluster number, shading type, and timestamp.

[0090] The light spot coverage trajectory and occlusion coverage are integrated based on the light spot region and the occlusion region to output a fused feature dataset;

[0091] The marked areas of light spots in the light spot coverage trajectory are associated with the marked areas of shading in the shading coverage table according to the photovoltaic module number and timestamp. Logical judgment is performed on the light spot shading status and shading coverage status of the same module area at the same timestamp. If both light spot shading and shading coverage exist simultaneously, it is marked as a double-shading area; if only light spot shading or only shading coverage exists, the original markings are retained respectively. The final marked status of all module areas, the corresponding current attenuation value, the shading coverage area ratio, and the timestamp are integrated into a multidimensional table. The timestamp field in the multidimensional table is uniformly sorted to ensure that the data is arranged continuously in time series. The final fused feature dataset includes photovoltaic module number, timestamp, light spot shading status, shading coverage type, current attenuation value, shading coverage area ratio, and spatial correlation coefficient, which are used for subsequent photovoltaic power prediction or operation and maintenance decision analysis.

[0092] Furthermore, based on the pre-trained spot prediction model, cloud motion vector extraction and linear interpolation algorithms, combined with the spatiotemporal dynamic characteristics of satellite cloud images, achieve high-precision prediction of cloud occlusion paths within a fixed future time period, effectively solving the prediction lag problem in scenarios with rapidly moving clouds. Secondly, by mapping the cloud coverage trajectory to the photovoltaic array plane coordinates and combining pixel-level alignment of the occlusion bounding boxes in the photovoltaic image, dual identification of spot occlusion and physical occlusion (such as dust and bird droppings) is achieved, breaking through the limitations of single occlusion type analysis. By utilizing the time series correlation analysis of string current data attenuation values ​​and occlusion states, combined with spatial distribution pattern clustering algorithms, the superimposed effects of dynamic spot occlusion and static physical occlusion can be accurately distinguished, thus providing more reliable input features for photovoltaic power prediction.

[0093] S3. Based on the light spot influence time window and current deviation value in the fused feature dataset, construct a power loss prediction curve, perform power influence conversion on the shading coverage, and generate power correction amount.

[0094] Extract the time window of the photovoltaic module's light spot influence and the corresponding current deviation value from the fused feature dataset;

[0095] Specifically, the photovoltaic module number field in the fusion feature dataset is traversed, records marked as photovoltaic modules are filtered, and the light spot impact time window field and corresponding current deviation value field of each photovoltaic module are read. The light spot impact time window field includes the timestamp range and duration, and the current deviation value field records the difference between the current decay value and the normal current value of the photovoltaic module within the same time window. The extracted light spot impact time window and current deviation value are classified according to the photovoltaic module number to ensure that the light spot impact event and current deviation data of each photovoltaic module correspond one-to-one, forming the initial dataset.

[0096] The time windows of multiple light spots affecting the same photovoltaic module are sorted by timestamp to generate a deviation sequence;

[0097] For the light spot impact time window field of each photovoltaic module, all timestamps are extracted and sorted in ascending order to ensure that the order of the time windows is consistent with the actual occurrence time. If there are overlapping or adjacent light spot impact events, they are merged or split according to the timestamp interval to generate a continuous and non-repeating deviation sequence. Each element of the deviation sequence contains the start time, end time, duration of the time window and the corresponding current deviation value. The deviation sequence is used for subsequent time series analysis of power loss values ​​to ensure that the current deviation values ​​of different time windows can be processed continuously along the time axis, providing a structured input for power loss mapping.

[0098] Map the current deviation values ​​in the deviation sequence to power loss values ​​and output power loss data;

[0099] Based on the current deviation value field in the deviation sequence, combined with the rated current and rated power parameters of the photovoltaic module, the current deviation value is converted into a power loss value. The conversion process is achieved through a linear proportional relationship, that is, the current deviation value and the rated voltage of the photovoltaic module are linearly fused to obtain the power loss value. The power loss value is merged with the time window field to generate power loss data, which includes the start time, end time, duration and corresponding power loss value of the time window.

[0100] The power loss data is integrated according to the timestamps of the photovoltaic array to generate a power loss sequence;

[0101] For the power loss data of all photovoltaic modules in the photovoltaic array, global sorting is performed by the timestamp field to ensure that the power loss events of different photovoltaic modules are arranged according to the actual occurrence time. If there are multiple power loss events under the same timestamp, they are merged in the order of photovoltaic module number to generate a time axis with a unique timestamp. The merged time axis is combined with the power loss value field to form a power loss sequence.

[0102] A continuous curve is plotted along the time axis of the power loss sequence to serve as the power loss prediction curve.

[0103] Based on the timestamp and power loss value fields in the power loss sequence, an interpolation algorithm is used to generate a continuous time axis to fill in the missing points in the timestamp intervals. The interpolation algorithm uses linear interpolation or spline interpolation methods to obtain the estimated value of the intermediate time point based on the power loss value of adjacent timestamps, ensuring the smoothness and continuity of the curve. The interpolated power loss sequence is connected in time axis order to form a power loss prediction curve, with the horizontal axis of the curve being the timestamp and the vertical axis being the power loss value.

[0104] Set up a loss mapping table based on historical soiling data;

[0105] Specifically, the photovoltaic module number, the percentage of contaminated area, and the corresponding power loss value are extracted from historical contamination data. The percentage of contaminated area records the proportion of the pixel area covered by obstructions on the surface of the photovoltaic module to the total pixel area. The power loss value is derived by mapping the current attenuation value and rated power from historical operating data. The photovoltaic module number, percentage of contaminated area, and power loss value are categorized by photovoltaic module number to generate a loss mapping table. Each photovoltaic module number in the loss mapping table corresponds to a set of mapping relationships between the percentage of contaminated area and the power loss value, ensuring that the power loss value under different percentages of contaminated area can be accurately queried.

[0106] Extract component information labeled as occlusion regions from the fused feature dataset, calculate the proportion of occlusion area in each component information, and output the occlusion proportion data;

[0107] Iterate through the photovoltaic module number field in the fused feature dataset, filter records marked as shading areas, and read the shading area percentage field for each shading area. The shading area percentage field represents the proportion of the pixel area on the photovoltaic module surface covered by shading objects such as dust and bird droppings to the total pixel area. Classify all shading areas by photovoltaic module number and corresponding shading area percentage field according to photovoltaic module number to generate shading percentage data. The shading percentage data includes photovoltaic module number and corresponding shading area percentage value.

[0108] The power loss value is output by matching and looking up the loss mapping table and the occlusion percentage data.

[0109] Based on the photovoltaic module number field in the shading ratio data, the mapping relationship between the sulphur area ratio and power loss value of the same photovoltaic module number in the loss mapping table is matched; for the module number of each shading area, the shading area ratio value is read and the corresponding power loss value is found in the loss mapping table; if the shading area ratio value is within the mapping range of the loss mapping table, the corresponding power loss value is obtained by interpolation; if the shading area ratio value is outside the mapping range, the boundary value is used as a substitute; the power loss value after matching and finding is merged with the module number field, and the power loss value is output.

[0110] The power loss value is combined with the rated power of the photovoltaic module to generate a power correction amount;

[0111] For each photovoltaic module, the output power loss value field is combined with the rated power field to calculate the power correction amount. The power correction amount represents the actual usable power of the photovoltaic module under the influence of shading. If the power loss value exceeds the rated power, the correction amount is set to zero to ensure the physical rationality of the calculation results. The correction amount is merged with the module number field to generate power correction amount data. The power correction amount data includes the module number and the corresponding corrected power value. All fields retain the timestamp accuracy of the original data.

[0112] The formula for power correction is:

[0113] ;

[0114] in, This indicates the power correction amount of photovoltaic modules under the influence of shading. Indicates the rated power of the photovoltaic module. This indicates the power loss of a photovoltaic module due to shading or sunspot coverage. Indicates when Less than The time correction is set to zero.

[0115] S4. Perform dual-channel compensation for photovoltaic energy storage using the power loss prediction curve and power correction amount to generate a compensation instruction set.

[0116] Extract the power loss value from the power loss prediction curve and overlay it with the power correction value in the power correction amount according to the timestamp to generate a comprehensive compensation sequence;

[0117] Specifically, based on the power loss value corresponding to each timestamp recorded in the power loss prediction curve, and the power correction value of each photovoltaic module number in the power correction dataset at the same timestamp, the power loss value and the power correction value are superimposed point by point along the time dimension. The superposition operation must strictly match the timestamp field. If there are cases where the timestamps are not completely consistent, the missing points are supplemented by a linear interpolation method to ensure the continuity of the power loss value and the power correction value on the time axis. The result of the superposition is the total power compensation demand at each timestamp. The superimposed total power compensation demand is merged according to the timestamp field to generate a comprehensive compensation sequence containing the timestamp, photovoltaic module number, and total compensation value.

[0118] Based on the total power compensation demand in the comprehensive compensation sequence, and in conjunction with the control protocol of photovoltaic energy storage, a compensation instruction set is generated;

[0119] The total power compensation demand at each timestamp is extracted from the comprehensive compensation sequence. Following the compensation priority rules defined in the photovoltaic energy storage control protocol, which determine the order of compensation execution for different photovoltaic modules when the total power compensation demand needs to be broken down into specific energy storage charging and discharging commands, the order is determined based on the operating status of the photovoltaic modules, the urgency of the compensation demand, the available capacity of the energy storage, and grid dispatch requirements. For example, at a given timestamp, if multiple photovoltaic modules have simultaneous compensation demands, the photovoltaic energy storage will prioritize processing the module with the highest power loss or the largest shading area, while also considering... The target power value is dynamically adjusted based on the current available capacity of the energy storage to maximize overall power recovery efficiency and avoid wasting energy storage resources. The compensation instruction must include the component number, timestamp, target power value, and execution time window. The target power value is determined by the ratio of the total compensation demand to the current available capacity of the photovoltaic energy storage. The execution time window is set according to the interval of the timestamp field and the dynamic response characteristics of the photovoltaic array to avoid power fluctuations caused by instruction execution delays. The decomposed instructions are sorted by component number and timestamp to generate a complete compensation instruction set. The structure of the compensation instruction set must be compatible with the photovoltaic energy storage interface to ensure the real-time performance and accuracy of the instructions.

[0120] S5. The monitoring and control center adjusts the discharge power of the energy storage converter according to the compensation instruction set, and performs power regulation on photovoltaic energy storage adjacent to the location of the obstruction, and outputs a control completion signal.

[0121] The monitoring and control center adjusts the discharge power of the energy storage converter according to the compensation instruction set;

[0122] Specifically, the component number, timestamp, target power value, and execution time window field are extracted from the compensation instruction set. The target power value is then input into the power regulation interface of the energy storage converter. The regulation process needs to be combined with the current operating status of the energy storage converter and the grid dispatch requirements. The adjustment operation is performed in segments according to the start and end times of the execution time window field to avoid equipment overload caused by excessively short timestamp intervals in the instruction set. After each adjustment, the energy storage converter needs to provide real-time feedback on the actual output power value and adjustment completion status. The monitoring and control center compares the feedback data with the target power value and generates an adjustment error log.

[0123] Power regulation of photovoltaic energy storage adjacent to the location of the obstruction refers to increasing the discharge power of photovoltaic energy storage adjacent to the obstruction area to make up for the power loss in the obstruction area and complete the power regulation.

[0124] The system filters photovoltaic (PV) module numbers marked as shading areas from the compensation instruction set and extracts the adjacent PV module number field. Adjacent PV module numbers are determined using the physical layout data of the PV array, ensuring a direct spatial connection between the shading area and adjacent areas. For each adjacent PV module number, the system reads the current discharge power value of the PV energy storage and obtains the power increment based on the corresponding target power value in the compensation instruction set. The power increment must be allocated considering the available capacity and dynamic response characteristics of the PV energy storage to avoid overheating due to excessive single adjustment. The adjustment process must synchronize the start and end times of the timestamp field in the compensation instruction set to ensure strict time alignment between the adjustment operations of adjacent modules and the compensation requirements of the shading area. After adjustment, the PV energy storage must output the actual discharge power value and adjustment status.

[0125] The adjustment results of the energy storage converter are integrated with the adjustment results of the adjacent photovoltaic modules to generate a control completion signal that includes the module number, the adjusted discharge power value, the execution time window, and the adjustment status. The control completion signal needs to be sent to the photovoltaic energy storage control terminal through the communication interface of the monitoring and control center to ensure that the timestamp field of the signal transmission and the compensation instruction set are synchronized.

[0126] S6. After receiving the control completion signal, the power sensor continuously monitors the actual power value of the photovoltaic energy storage, compares the actual power value with the target power value in the compensation instruction set, and judges the effect of photovoltaic energy storage control.

[0127] After receiving the control completion signal, the power sensor continuously monitors the actual power value of the photovoltaic energy storage.

[0128] Specifically, the photovoltaic module number, adjusted discharge power value, and execution time window field are extracted from the control completion signal, and the execution time window field is used as the time range for power monitoring. The power sensor performs supervisory control based on the photovoltaic module number and starts continuous monitoring mode according to the start and end times of the execution time window field. The supervisory control process needs to collect the actual power value of photovoltaic energy storage at fixed time intervals, and the collection frequency needs to be consistent with the accuracy of the timestamp in the compensation instruction set. The collected data includes the photovoltaic module number, timestamp, and actual power value fields, and all data is uploaded to the supervisory control center through the communication interface of the power sensor.

[0129] The power difference between the actual power value and the target power value is calculated using the difference method.

[0130] The target power value is read from the compensation instruction set and matched point by point with the actual power value uploaded by the power sensor according to the photovoltaic module number and timestamp. After the matching is completed, the power difference at each timestamp is calculated using the difference method. The power difference needs to retain the timestamp field and generate a difference sequence containing the module number, timestamp and difference.

[0131] The formula for calculating the power difference is:

[0132] ;

[0133] in, This represents the power difference between the actual power value and the target power value. This indicates the actual power value monitored in real time by the power sensor. This represents the target power value under the corresponding timestamp in the compensation instruction set;

[0134] If the power difference exceeds the preset power threshold, the compensation instruction set will be dynamically corrected and the correction instruction set will be output.

[0135] The preset power threshold is jointly defined by the operational requirements of photovoltaic energy storage and the grid dispatch requirements. Specifically, it is set as follows: First, operational parameters are extracted from the operational requirements of photovoltaic energy storage, including maximum charging power, maximum discharging power, charging cutoff SOC, and discharging cutoff SOC. These parameters define the safe operating boundaries of photovoltaic energy storage. Second, the maximum allowable charging and discharging power limit, electricity price time period division (such as the TOU time period corresponding to peak-valley electricity prices), and grid power regulation accuracy requirements are obtained from the grid dispatch requirements, such as the grid's limit on reverse charging power. Then, the maximum charging and discharging power in the operational requirements and the power requirements allowed by grid dispatch are used as constraints to set the preset power threshold, ensuring that the preset power threshold does not exceed the physical capacity of the photovoltaic energy storage and complies with grid safety standards.

[0136] Records whose power difference exceeds a preset power threshold are selected from the difference sequence. These selected records must include the component number, timestamp, actual power value, and target power value fields. For each record exceeding the threshold, dynamic correction is performed. The target power value is adjusted based on the relationship between the actual power value and the preset power threshold. This adjustment must be tailored to the available capacity and dynamic response characteristics of the photovoltaic energy storage. The adjusted target power value must be consistent with the execution timestamp field in the original compensation instruction set to ensure that the corrected instructions are executed as planned. The adjusted target power value is then merged with the component number and timestamp fields from the original compensation instruction set to generate a correction instruction set. The structure of the correction instruction set must be completely identical to the original compensation instruction set to ensure that no additional adaptation is required in subsequent execution stages. The correction instruction set must be immediately sent to the power regulation interface of the energy storage converter after generation.

[0137] If the power difference does not exceed the preset power threshold, then the photovoltaic energy storage control effect is determined to meet the standard.

[0138] The system filters records from the difference sequence whose power difference is less than or equal to a preset power threshold. Each record must include the component number, timestamp, actual power value, and target power value fields. The filtered records indicate that the discharge power of the photovoltaic energy storage has stabilized within the target range and meets the compensation requirements. The filtered results are merged with the component number and timestamp fields in the compensation instruction set to generate a control effect compliance log. If the power difference under multiple consecutive timestamps does not exceed the preset power threshold, the compensation instruction set of the photovoltaic component is automatically archived and marked as having completed the adjustment status.

[0139] This embodiment also provides a data-driven photovoltaic energy storage system control system, including:

[0140] The data acquisition module is used to collect photovoltaic environmental data and perform preprocessing; the photovoltaic environmental data includes satellite cloud image data, photovoltaic array images, string current data, and historical pollution data;

[0141] The prediction module is used to process satellite cloud image data through a pre-trained spot prediction model, predict the spot coverage trajectory, and identify the coverage of obstructions in the photovoltaic array image; the reliability of the spot coverage trajectory and obstruction coverage is verified by the current attenuation value in the string current data, and the fused feature dataset is output.

[0142] The conversion module is used to construct a power loss prediction curve based on the light spot influence time window and current deviation value in the fused feature dataset, perform power influence conversion on the shading coverage, and generate power correction amount;

[0143] The instruction module is used to perform dual-channel compensation for photovoltaic energy storage based on the power loss prediction curve and the power correction amount, and generate a compensation instruction set.

[0144] The adjustment module is used to monitor and control the center to adjust the discharge power of the energy storage converter according to the compensation instruction set, and to issue local compensation instructions to photovoltaic energy storage adjacent to the location of the obstruction, and output control completion signal;

[0145] The judgment module is used to continuously monitor the actual power value of photovoltaic energy storage after the power sensor receives the control completion signal, compare the actual power value with the target power value in the compensation instruction set, and judge the effect of photovoltaic energy storage control.

[0146] This embodiment also provides a computer device applicable to the data-driven photovoltaic energy storage system control method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the data-driven photovoltaic energy storage system control method proposed in the above embodiment.

[0147] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0148] This embodiment also provides a storage medium storing a computer program. When executed by a processor, the program implements the data-driven photovoltaic energy storage system control method proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0149] In summary, this invention achieves dynamic and precise calibration of photovoltaic (PV) shading compensation by combining satellite cloud image-driven light spot coverage trajectory modeling with spatial verification of string current data. By extracting cloud motion vectors and constructing a planar coordinate mapping using a convolutional neural network, the three-dimensional cloud layer is dynamically transformed into a spatiotemporal sequence of PV module shading, providing high-resolution trajectory data for power loss prediction. Through clustering and comparison of current attenuation values ​​with the spatial distribution of shading data, misjudged areas are eliminated and shading boundaries are corrected, generating a fused feature dataset to ensure multidimensional consistency between light spot prediction and actual shading status. A dual-channel compensation strategy is used to adjust discharge power, achieving dynamic power balance between the shading area and adjacent energy storage units, ultimately achieving continuous and stable output of PV energy storage and optimal allocation efficiency of energy storage resources under fluctuating illumination.

[0150] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A data-driven based photovoltaic energy storage system control method, characterized in that: include, Photovoltaic environmental data is collected and preprocessed; the photovoltaic environmental data includes satellite cloud image data, photovoltaic array images, string current data, and historical pollution data. The system processes satellite cloud image data using a pre-trained spot prediction model to predict the spot coverage trajectory and identify the coverage rate of obstructions in the photovoltaic array image. The reliability of the spot coverage trajectory and obstruction coverage rate is verified by the current attenuation value in the string current data, and a fused feature dataset is output. Based on the light spot influence time window and current deviation value in the fused feature dataset, a power loss prediction curve is constructed, the power influence conversion of the shading coverage is performed, and a power correction amount is generated. Photovoltaic energy storage is compensated through a dual-channel system using power loss prediction curves and power correction values, generating a set of compensation instructions. The monitoring and control center adjusts the discharge power of the energy storage converter according to the compensation instruction set, and adjusts the power of photovoltaic energy storage adjacent to the location of the obstruction, and outputs a control completion signal; After receiving the control completion signal, the power sensor continuously monitors the actual power value of the photovoltaic energy storage, compares the actual power value with the target power value in the compensation instruction set, and judges the effect of photovoltaic energy storage control. The process of using a pre-trained spot prediction model to process satellite cloud image data and predict the spot coverage trajectory involves the following steps: Satellite cloud image data is input into a pre-trained spot prediction model. The convolutional neural network of the spot prediction model is used to extract cloud motion vector information and generate a cloud motion vector map. Based on the speed and direction information in the cloud motion vector map, a linear interpolation algorithm is used to obtain the cloud movement path for a fixed period in the future. The cloud movement paths are then integrated to generate a cloud coverage trajectory map. Convert the latitude and longitude coordinates in the cloud cover trajectory map into the planar coordinates of the photovoltaic array; The component areas of the shaded photovoltaic array are marked according to the planar coordinates, and the component areas are integrated according to the time sequence of the cloud movement path to output the light spot coverage trajectory. The reliability of the light spot coverage trajectory and obstruction coverage is verified by using the current attenuation value in the string current data, and a fused feature dataset is output. The specific steps are as follows: The current attenuation value in the string current data is matched with the light spot coverage trajectory by timestamp to generate a shading-current relationship table. Identify component regions whose current attenuation values ​​are consistent with the changing trend of the light spot coverage trajectory from the shading-current relationship table, and mark them as light spot regions; The component association information of current attenuation value is spatially compared with the component area in the shading coverage to obtain the spatial distribution pattern of current attenuation value and shading coverage. Clustering algorithms are used to identify component regions in the coverage of obstructions that have a spatial distribution pattern consistent with the current attenuation value, and these regions are marked as obstructed areas. The light spot coverage trajectory and occlusion coverage are integrated based on the light spot region and the occlusion region to output a fused feature dataset.

2. The data-driven based photovoltaic energy storage system control method of claim 1, wherein: The specific steps for identifying the coverage of obstructions in the photovoltaic array image are as follows: Extract the bounding box coordinates of the shading objects from the photovoltaic array image, align the bounding box coordinates of the shading objects with the pixel coordinate system of the photovoltaic array image, and generate shading data; Map the occlusion data to the area covered by the occlusion; The system calculates the shading status of the area covered by obstructions relative to the total area of ​​the photovoltaic array and outputs the shading coverage rate.

3. The data-driven based photovoltaic energy storage system control method of claim 1, wherein: The power loss prediction curve is constructed based on the time window of light spot influence and current deviation values ​​in the fused feature dataset. The specific steps are as follows: Extract the time window of the photovoltaic module's light spot influence and the corresponding current deviation value from the fused feature dataset; The time windows of multiple light spots affecting the same photovoltaic module are sorted by timestamp to generate a deviation sequence; Map the current deviation values ​​in the deviation sequence to power loss values ​​and output power loss data; The power loss data is integrated according to the timestamps of the photovoltaic array to generate a power loss sequence; A continuous curve is plotted along the time axis of the power loss sequence to serve as the power loss prediction curve.

4. The data-driven based photovoltaic energy storage system control method of claim 1, wherein: The specific steps for performing power impact conversion on the coverage rate of obstructions to generate a power correction amount are as follows: Set up a loss mapping table based on historical soiling data; Extract component information labeled as occlusion regions from the fused feature dataset, calculate the proportion of occlusion area in each component information, and output the occlusion proportion data; The power loss value is output by matching and looking up the loss mapping table and the occlusion percentage data. The power loss value is combined with the rated power of the photovoltaic module to generate a power correction value.

5. The data-driven based photovoltaic energy storage system control method of claim 1, wherein: The process involves dual-channel compensation of photovoltaic energy storage using power loss prediction curves and power correction amounts to generate a compensation instruction set. The specific steps are as follows: Extract the power loss value from the power loss prediction curve and overlay it with the power correction value in the power correction amount according to the timestamp to generate a comprehensive compensation sequence; Based on the total power compensation demand in the comprehensive compensation sequence and combined with the control protocol of photovoltaic energy storage, a compensation instruction set is generated.

6. The data-driven based photovoltaic energy storage system control method of claim 1, wherein: The power regulation of photovoltaic energy storage adjacent to the shading object refers to increasing the discharge power of photovoltaic energy storage adjacent to the shading area to make up for the power loss in the shading area and complete the power regulation.

7. The data-driven based photovoltaic energy storage system control method of claim 1, wherein: The comparison between the actual power value and the target power value in the compensation instruction set to determine the effect of photovoltaic energy storage control refers to calculating the power difference between the actual power value and the target power value using the difference method; if the power difference exceeds the preset power threshold, the compensation instruction set is dynamically corrected and the correction instruction set is output; if the power difference does not exceed the preset power threshold, the photovoltaic energy storage control effect is determined to be up to standard.

8. A data-driven photovoltaic energy storage system control system based on any one of claims 1-7, characterized in that: include, The data acquisition module is used to collect photovoltaic environmental data and perform preprocessing; the photovoltaic environmental data includes satellite cloud image data, photovoltaic array images, string current data, and historical pollution data. The prediction module is used to process satellite cloud image data through a pre-trained spot prediction model, predict the spot coverage trajectory, and identify the coverage of obstructions in the photovoltaic array image; the reliability of the spot coverage trajectory and obstruction coverage is verified by the current attenuation value in the string current data, and the fused feature dataset is output. The conversion module is used to construct a power loss prediction curve based on the light spot influence time window and current deviation value in the fused feature dataset, perform power influence conversion on the shading coverage, and generate power correction amount; The instruction module is used to perform dual-channel compensation for photovoltaic energy storage based on the power loss prediction curve and the power correction amount, and generate a compensation instruction set. The adjustment module is used to monitor and control the center to adjust the discharge power of the energy storage converter according to the compensation instruction set, and to issue local compensation instructions to photovoltaic energy storage adjacent to the location of the obstruction, and output control completion signal; The judgment module is used to continuously monitor the actual power value of photovoltaic energy storage after the power sensor receives the control completion signal, compare the actual power value with the target power value in the compensation instruction set, and judge the effect of photovoltaic energy storage control.

Images