Distributed photovoltaic power collaborative prediction system based on gridding meteorological fusion and dynamic cluster modeling

Abstract

The invention discloses a distributed photovoltaic power collaborative prediction system based on gridding meteorological fusion and dynamic cluster modeling, and the system mainly comprises a meteorological grid processing module, a dynamic cluster modeling module, a standby power demand processing module, a collaborative distribution module, a meteorological prediction processing module, and a power prediction module. The meteorological grid processing module is used for acquiring meteorological grid data of a target area and generating a photovoltaic output coefficient corresponding to each grid through a grid output calculation model; the dynamic cluster modeling module is used for generating a dynamic cluster topology matrix according to the output fluctuation similarity; the standby power demand processing module is used for generating a standby power demand priority weight through a standby power demand prediction model; the collaborative distribution module is used for calculating the output priority proportion of each photovoltaic cluster; the power prediction module is used for generating a collaborative output baseline value of each photovoltaic cluster. According to the invention, prediction precision can be improved, scheduling robustness can be enhanced, and reasonable distribution of photovoltaic resources can be realized.

Description

Technical Field

[0001] This invention relates to the field of distributed photovoltaic power generation prediction technology, and in particular to a distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling. Background Technology

[0002] In existing technologies, distributed photovoltaic power prediction typically relies on data from a single point or a small number of weather stations, directly using measured irradiance, cloud cover, and temperature to estimate the average output of the entire area or station. This approach ignores the high spatial heterogeneity and local variation of meteorological conditions, making it difficult for prediction models to accurately reflect fluctuations in actual output. Especially in scenarios with rapid movement of local cloud clusters or sudden cloudfall, prediction errors increase significantly, making it difficult to meet the needs for rapid and precise scheduling.

[0003] Meanwhile, traditional distributed photovoltaic (PV) systems often use static partitioning based on geographical location or installed capacity, lacking dynamic perception and complementary utilization of real-time output. During scheduling, this static partitioning method often leads to simultaneous load increases and decreases in each sub-cluster under similar weather conditions, failing to fully utilize the complementary advantages of PV output in different areas, thus reducing the overall system's stability and resource utilization efficiency.

[0004] Furthermore, existing multi-source factor fusion weight allocation methods mostly employ linear or empirical weighting strategies, which struggle to balance multiple objectives such as photovoltaic power output capacity, cluster complementarity, and urgent load-side demands. This leads to risks of allocation bias or weight imbalance, making it difficult to achieve fair and reasonable resource sharing while ensuring reliable power supply to critical loads. In summary, there is an urgent need for a collaborative forecasting and scheduling system that incorporates fine-grained gridded meteorological information, dynamically senses cluster complementarity relationships, and integrates game-theoretic allocation mechanisms. This would improve forecasting accuracy, enhance scheduling robustness, and achieve rational allocation of photovoltaic resources. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling, mainly comprising: The meteorological grid processing module is used to acquire meteorological grid data of the target area. The meteorological grid data includes the real-time irradiance, cloud coverage and temperature of each grid, and generates the photovoltaic output coefficient corresponding to each grid through the grid output calculation model. The dynamic cluster modeling module is used to acquire real-time power output data and geographical location data of distributed photovoltaic clusters, and generate a dynamic cluster topology matrix based on the similarity of power output fluctuations. The dynamic cluster topology matrix is ​​used to characterize the power output complementarity relationship between clusters. The backup power demand processing module is used to obtain real-time power consumption and remaining energy storage capacity data of the data center's servers, and generate backup power demand priority weights through the backup power demand prediction model. The collaborative allocation module is used to input the photovoltaic output coefficient, dynamic cluster topology matrix and backup power demand priority weight into the Shapley value game algorithm to calculate the output priority ratio of each photovoltaic cluster. The meteorological forecasting and processing module is used to build a forecasting model based on historical meteorological grid data and generate meteorological forecast data for several future periods. The meteorological forecast data includes future irradiance, cloud cover rate and temperature indicators, which are used as one of the input parameters of the power forecasting module. The power prediction module is used to generate a collaborative output baseline value for each photovoltaic cluster based on the output priority ratio and meteorological forecast data, and send the collaborative output baseline value to the power grid dispatch terminal.

[0007] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, the grid output calculation model calculates the photovoltaic output coefficient using the following formula:

[0008] in, For the first Photovoltaic output coefficient of each grid; Real-time irradiance; Cloud coverage; Real-time temperature; , These are calibration parameters; Here is the temperature decay function, as follows:

[0009] in, These are calibration parameters; This is a reference temperature value.

[0010] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, the generation of a dynamic cluster topology matrix based on the power output fluctuation similarity includes the following steps: Obtain the output data of each photovoltaic node in the distributed photovoltaic cluster within the most recent T time periods, and construct an output time series matrix. ; Based on the time series matrix Using Pearson correlation coefficient, dynamic time warping, or mutual information methods, the output similarity between any two photovoltaic nodes is calculated. Form a similarity matrix ; For the similarity matrix Perform threshold filtering or clustering operations to obtain the dynamic cluster topology matrix. ,in:

[0011] At preset time intervals, the output similarity is recalculated based on the updated output time series, and the topology matrix is ​​dynamically updated.

[0012] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, the generation of backup power demand priority weights through the backup power demand prediction model includes the following steps: Real-time collection of server cluster power consumption data in the data center and the remaining capacity of the energy storage system ; Calculate the total power consumption fluctuation coefficient of the data center ; Analysis of the rate of change of remaining energy storage capacity based on sliding time window : Establish a backup power demand urgency assessment function :

[0013] in, This indicates the urgency assessment value of backup power, reflecting the current system's urgent demand for electricity. As the benchmark volatility coefficient, This is the power fluctuation weighting coefficient. This is the weighting coefficient for remaining energy storage capacity. The weighting coefficient for the energy storage change trend is... ; Describes the minimum capacity warning threshold set for the energy storage system; urgency assessment value Mapping to weight interval Generate priority weights That is, the weights are generated using a linear normalization formula, as follows:

[0014] in, Priority weight; This is the lower bound of the priority weight; This represents the upper limit of priority weight; This represents the minimum value of the urgency assessment, used for normalization. This represents the maximum value of the urgency assessment, used for normalization.

[0015] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, wherein: in the backup power demand urgency assessment function: When the remaining energy storage capacity is below the warning threshold At that time, the weighting factor of the remaining energy storage capacity will be automatically increased. Up to 1.5 times the original value; When server cluster power consumption data suddenly increases, the power volatility weighting coefficient Use the maximum value within the sliding window.

[0016] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, the following steps are included: Calculating the output priority ratio of each photovoltaic cluster using the Shapley value game algorithm: Define the marginal value function for each photovoltaic cluster, which is composed of a weighted combination of the following factors; Based on all possible combinations of photovoltaic clusters, the marginal contribution of each cluster is weighted and summed using the Shapley value calculation formula to obtain the Shapley value of each photovoltaic cluster. The Shapley values ​​of each photovoltaic cluster are normalized to obtain the output priority ratio of each photovoltaic cluster, which is used to generate the collaborative output baseline value in the power prediction module.

[0017] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, the multiple factors include the photovoltaic output coefficient of the photovoltaic cluster, the output complementarity information represented in the dynamic cluster topology matrix, and the backup power priority weight of the corresponding data center.

[0018] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, wherein: in the power prediction module, the generation of the collaborative output baseline value includes: Correcting the gridded photovoltaic output coefficient based on meteorological forecast data Calculate the predicted output value for each grid; Based on the output priority ratio of the photovoltaic power distribution, the original output of each cluster is weighted and corrected to generate the most predictable collaborative output baseline value for photovoltaic power generation.

[0019] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, the weighted correction adopts a segmented strategy: When meteorological forecast data indicates that the cloud coverage rate will be greater than 50% in the next hour, the output weight of low-priority clusters will be reduced first. When the priority weight of backup power demand Exceeding the limit At that time, the output ratio of the corresponding cluster will be forcibly increased by at least 20%.

[0020] As a preferred embodiment of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling described in this invention, it further includes a feedback optimization module, which is used to dynamically adjust the parameters of the grid output calculation model, the backup power demand prediction model, or the Shapley value game algorithm based on the deviation between the actual output data of the power grid dispatch terminal and the collaborative output baseline value.

[0021] The beneficial effects of this invention are: 1. This invention establishes a meteorological grid processing module to construct a multi-dimensional grid data system based on real-time irradiance, cloud cover, and temperature. Combined with a grid output calculation model, it generates a photovoltaic output coefficient for each grid, achieving a spatially distributed expression of photovoltaic power output capacity. Compared to traditional regional averaging models, this approach significantly improves prediction accuracy and adaptability to abrupt weather changes, helps capture the impact of local meteorological differences on photovoltaic output, and provides a more reliable foundation for subsequent collaborative forecasting.

[0022] 2. This invention acquires the output characteristics and geographical information of each photovoltaic cluster through a dynamic cluster modeling module, and generates a dynamic cluster topology matrix based on the similarity of output fluctuations to characterize the output complementarity relationship between clusters. This topology matrix can dynamically reflect the power synergy potential of each cluster under different operating conditions, enabling the system to fully utilize output differences to achieve load balancing during scheduling, thereby improving the overall system stability and resource utilization efficiency.

[0023] 3. The collaborative allocation module of this invention calculates the output priority ratio of each photovoltaic cluster by inputting the photovoltaic output coefficient, dynamic cluster topology matrix, and backup power demand priority weights into the Shapley value game algorithm, which serves as an important input basis for power prediction. This method, considering multiple factors such as meteorological conditions, complementarity, and backup power demand, achieves fairness and rationality in resource allocation through game theory, avoiding the limitations of traditional weighted methods that are prone to imbalance, and improving the effectiveness and accuracy of collaborative output prediction. Attached Figure Description

[0024] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. 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. Wherein: Figure 1 This is a structural block diagram of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling of the present invention.

[0025] Figure 2 This is a flowchart of the prediction process of the distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling of the present invention. Detailed Implementation

[0026] 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.

[0027] 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.

[0028] 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.

[0029] Secondly, the present invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of the present invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not according to the usual scale. Furthermore, the schematic diagrams are merely examples and should not limit the scope of protection of the present invention. In addition, actual fabrication should include three-dimensional spatial dimensions of length, width, and depth.

[0030] Example 1 Reference Figure 1-2 As an embodiment of the present invention, a distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling is provided, mainly comprising: The meteorological grid processing module is used to acquire meteorological grid data of the target area. The meteorological grid data includes the real-time irradiance, cloud coverage and temperature of each grid, and generates the photovoltaic output coefficient corresponding to each grid through the grid output calculation model.

[0031] It should be noted that meteorological grid data refers to dividing a target geographic area into several meteorological grid units, each corresponding to a specific geographic location and number. For example, dividing the area into several regular grid units according to latitude and longitude coordinates, such as forming a meteorological grid unit every 0.05° × 0.05°. Real-time irradiance represents the solar radiation power / intensity received per unit area per unit time, usually expressed in watts per square meter (W / m²). 2 Cloud cover rate represents the proportion of the sky obscured by clouds, usually expressed as a percentage; temperature represents the ambient temperature at the current grid location, in degrees Celsius.

[0032] The aforementioned meteorological grid data can be generated by fusing satellite remote sensing data, ground meteorological station observation data, and meteorological forecasting models. It features real-time performance and high spatial resolution, effectively depicting the dynamic changes in the environment surrounding photovoltaic modules within the region, and providing accurate support for subsequent calculations of the photovoltaic power output coefficient. Photovoltaic modules are the basic units for solar power generation, and their output capacity is significantly affected by meteorological conditions.

[0033] Specifically, the grid-based power output calculation model estimates the relative photovoltaic (PV) output capacity of each meteorological grid under current environmental conditions based on multiple key environmental parameters (such as real-time irradiance, cloud cover, and temperature) from the meteorological grid data. The final output is represented as the PV output coefficient of that grid, used to quantify its expected output level per unit capacity. The grid-based power output calculation model calculates the PV output coefficient using the following formula:

[0034] in, For the first Photovoltaic output coefficient of each grid For real-time irradiance, Cloud coverage For real-time temperature, , For calibration parameters, Let be the temperature decay function. , For calibration parameters, This is a reference temperature value. Setting it to 0.2 indicates the normalized conversion efficiency of output per unit irradiance. A setting of 0.8 indicates that power output will decrease by approximately 80% under complete cloud cover. A setting of 0.004 means that for every 1°C increase, the output decreases by approximately 0.4%. The temperature is set to 25℃, which is the reference temperature under standard testing conditions in the photovoltaic industry. The above calibration parameters can be set by regression fitting based on historical operating data, or based on the performance parameters provided by the photovoltaic module manufacturer.

[0035] For example: when , , Reference temperature value At that time, and if the real-time meteorological data for a certain grid area is: irradiance Cloud coverage ,temperature According to The formula is used to calculate the photovoltaic power output coefficient for this region. .

[0036] The dynamic cluster modeling module is used to acquire real-time power output data and geographical location data of distributed photovoltaic clusters, and generate a dynamic cluster topology matrix based on the similarity of power output fluctuations. The dynamic cluster topology matrix is ​​used to characterize the power output complementarity relationship between clusters.

[0037] It should be noted that a distributed photovoltaic (PV) cluster refers to a collection of PV power generation units deployed in multiple physical locations within a target area, possessing a degree of independence but also having the potential for coordinated scheduling. Each PV power generation unit can be referred to as a PV node, and it has the ability to independently collect and upload data.

[0038] Output data refers to the power generation or power density of each photovoltaic node per unit time, which can be collected in real time and uploaded to the system platform by equipping each photovoltaic node with devices such as power meters and current and voltage monitoring modules.

[0039] Geographic location refers to the GPS latitude and longitude coordinates of each photovoltaic node during the construction phase. This information can be automatically collected through on-site communication modules or back-end equipment filing, or it can be manually entered by the system administrator.

[0040] Output fluctuation similarity refers to an index used to quantify the similarity of the behavior of multiple photovoltaic nodes in changing photovoltaic output over time within a certain time window. It is a key basis in the dynamic cluster modeling process and is used to guide cluster partitioning and the generation of dynamic cluster topology matrix.

[0041] Specifically, generating a dynamic cluster topology matrix based on the similarity of output fluctuations includes the following steps: Obtain the output data of each photovoltaic node in a distributed photovoltaic cluster over the most recent T time periods (e.g., the past 24 hours), and construct an output time series matrix. For example:

[0042] in Indicates the first Each photovoltaic node in time The processing value.

[0043] Based on the aforementioned time series matrix, the output similarity between any two photovoltaic nodes is calculated using methods such as Pearson correlation coefficient, dynamic time warping (DTW), or mutual information. Form a similarity matrix ; For the similarity matrix Perform threshold filtering or clustering operations (such as DBSCAN, spectral clustering, etc.) to obtain the dynamic cluster topology matrix. ,in:

[0044] At preset time intervals, the output similarity is recalculated based on the updated output time series, and the topology matrix is ​​dynamically updated.

[0045] Example: Assume the target area contains 5 distributed photovoltaic (PV) nodes, numbered Node 1 to Node 5. The system's analysis time window is set to the past 24 hours, with sampling every hour, resulting in a total of T=24 time points. The system acquires the actual power output data of each PV node during these 24 hours and constructs a power output time series matrix. The following is a simplified version:

[0046] The system then uses the Pearson correlation coefficient to calculate the output time series between any two nodes in the above matrix, obtaining the output similarity between each pair, forming a 5×5 similarity matrix. Assume the similarity matrix is ​​as follows:

[0047] Next, the system analyzes the similarity matrix. Threshold filtering is performed, setting the similarity threshold to 0.85. At that time, it is considered that the node With nodes There is a strong correlation between them, and they can be classified into the same category. This refers to the same dynamic sub-cluster. This generates a dynamic cluster topology matrix. Its form is as follows:

[0048] corresponding topological matrix for:

[0049] The topology matrix shows that the power output fluctuations among nodes 1, 2, and 5 are highly similar, forming one sub-cluster; nodes 3 and 4 form another sub-cluster. This topology matrix is ​​then fed into the system allocation module to guide the grouping modeling of photovoltaic power output priorities in the Shapley value game algorithm.

[0050] In addition, to adapt to changes in output patterns caused by weather and load variations, the system is set to update dynamically every hour. That is, output data from the past 24 hours is recollected every hour, and the similarity matrix and topology matrix are regenerated according to the steps described above to ensure the timeliness and accuracy of cluster partitioning.

[0051] The backup power demand processing module is used to obtain real-time power consumption and remaining energy storage capacity data of the data center's servers, and generate backup power demand priority weights through the backup power demand prediction model.

[0052] It should be noted that the backup power demand forecasting model adopts a hybrid modeling structure based on ARIMA and LSTM neural networks. First, based on the collected historical power consumption data of the server cluster, wavelet transform (preferably db4 mother wavelet, three-level decomposition) is used to denoise the original time series, extracting load features such as average load, volatility, and kurtosis. Then, combining the remaining energy storage capacity, SOH (sustainable state of energy), discharge efficiency, and temperature data, a capacity-power coupling equation is constructed to characterize the actual available energy storage capacity's support capability for backup power scheduling. The model as a whole adopts an LSTM predictor structure, optimized using a strategy combining offline training and online updates: in the offline phase, network parameters are trained using three consecutive years of historical data to form a stable initial prediction model; in the online phase, fine-tuning is performed every 15 minutes based on the latest SOH and temperature data. If a decrease in discharge efficiency exceeding a set threshold (e.g., 5%) is detected, the model is automatically reconstructed. Finally, the model outputs the backup power demand priority weights, which are then input into the collaborative allocation module.

[0053] Specifically, generating backup power demand priority weights through a backup power demand forecasting model includes the following steps: Real-time collection of server cluster power consumption data in the data center and the remaining capacity of the energy storage system ; Calculate the total power consumption fluctuation coefficient of the data center The calculation formula is as follows:

[0054] in, Indicates the first The power consumption value of the server cluster at the current moment. The average power consumption of all server clusters. The total number of server clusters participating in the evaluation; Analysis of the rate of change of remaining energy storage capacity based on sliding time window :

[0055] in, Current moment The remaining capacity of the energy storage system This indicates the energy storage capacity at the time of the previous sampling period. The sampling interval; Establish a backup power demand urgency assessment function :

[0056] in, This indicates the urgency assessment value of backup power, reflecting the current system's urgent demand for electricity. The baseline fluctuation coefficient can be set based on historical operational statistics. This is the power fluctuation weighting coefficient. This is the weighting coefficient for remaining energy storage capacity. The weighting coefficient for the energy storage change trend is... Based on experience, it can be set It is 0.4. It is 0.5. It is 0.1; Describe the minimum capacity warning threshold set for the energy storage system (which can be set according to load requirements); In the backup power demand urgency assessment function: When the remaining energy storage capacity is below the warning threshold At that time, the weighting factor of the remaining energy storage capacity will be automatically increased. Up to 1.5 times the original value. The purpose of this setting is to prevent situations where the remaining energy storage capacity falls below a warning threshold. In cases of power outages, the system automatically amplifies the impact of energy storage capacity on the urgency of backup power to ensure rapid power supply and avoid system downtime.

[0057] When server cluster power consumption data suddenly increases, the power volatility weighting coefficient The maximum value within the sliding window is used. This is because when power consumption suddenly increases, using the mean or smoothed value can easily lead to a lag in weight estimation. Using the maximum value within the sliding window can quickly capture the severity of the "sudden event" and improve the evaluation function. Power fluctuation weighting coefficient This contributes to raising the overall emergency power reserve level.

[0058] urgency assessment value Mapping to weight interval Generate priority weights That is, the weights are generated using a linear normalization formula, as follows:

[0059] in, Priority weight; This is the lower bound of the priority weight, for example, 0; This represents the upper limit of priority weight, for example, 1; This represents the minimum value of the urgency assessment, used for normalization. This represents the maximum value of the urgency assessment, used for normalization.

[0060] in addition, and This can be determined through historical data statistics, that is, by collecting all calculated urgency scores within a set assessment time window (such as the last 7 days). After removing outliers, the 5th and 95th percentiles of the distribution were taken as... and .

[0061] For example: Suppose a data center is connected to three physically isolated server clusters (denoted as clusters A, B, and C), which are respectively connected to three distributed photovoltaic (PV) power stations (denoted as PV-A, PV-B, and PV-C). Now, based on the power consumption characteristics of each server cluster and the status of its corresponding energy storage system, the output priority weight of its corresponding PV cluster needs to be determined. , , This is to guide the coordinated allocation of photovoltaic power.

[0062] S1: Real-time data collection of power consumption and energy storage system capacity for each server cluster, with a collection time of t. Assumptions: The total power consumption of all cluster servers is: , , The remaining capacity of the energy storage system is: , , S2: Calculate the total power consumption fluctuation coefficient of the server cluster Assuming data is collected at one point every 10 minutes, six sample data points from each cluster were obtained over the past hour:

[0063] The average power consumption can be calculated from the data obtained in the table above: , , ,but:

[0064]

[0065]

[0066] S3: Calculate the rate of change of remaining energy storage capacity by =30min as the window, assuming: , ,

[0067]

[0068]

[0069] but: , , S4: Calculate the backup power urgency assessment value set up: Standard volatility; ;parameter: , , Substitute into the formula:

[0070]

[0071]

[0072] S5: Normalization mapping to weights set up: , , , Using linear normalization:

[0073] calculate:

[0074]

[0075]

[0076] Therefore, the generated priority weights are: Photovoltaic cluster PV-A→ Photovoltaic cluster PV-B→ Photovoltaic cluster PV-C→ This priority will be used to guide the clusters with higher weights in the system's collaborative allocation module to obtain more photovoltaic power output resources in order to prioritize their backup power needs.

[0077] The collaborative allocation module is used to input the photovoltaic output coefficient, dynamic cluster topology matrix and backup power demand priority weight into the Shapley value game algorithm to calculate the output priority ratio of each photovoltaic cluster. It should be noted that the Shapley value game algorithm is an allocation mechanism derived from cooperative game theory, primarily used to fairly quantify the marginal contribution values ​​of multiple participants to the overall goal. In this system, multiple photovoltaic clusters collaboratively supply power to the data center. Due to factors such as weather conditions, spatial layout, and differences in load demand, their actual power generation value is not entirely the same. Therefore, the role of the Shapley value game algorithm is to rationally calculate and allocate the output priority ratio of each photovoltaic cluster in collaborative power output under the comprehensive influence of multiple factors, ensuring fair, efficient, and dynamically adjustable power supply allocation.

[0078] In this application, the input to the Shapley value game algorithm consists of three key variables, such as: photovoltaic power output coefficient. (Output from the meteorological grid processing module, reflecting the theoretical power generation capacity under current weather conditions), dynamic cluster topology matrix (This represents the complementary relationship of output between different clusters, which is beneficial to improving the stability of overall scheduling) and priority weights. (Output from the backup power demand forecasting model, reflecting the urgency of power supply requirements), the output of the Shapley value game algorithm is the output priority ratio for each photovoltaic cluster. photovoltaic clusters The proportion of collaborative power output tasks that should be undertaken reflects its importance and allocation priority under the current collaborative power supply strategy. This proportion will serve as one of the key parameters in the power prediction module, used to generate the collaborative power output baseline value for each photovoltaic cluster in conjunction with meteorological forecast data, and further sent to the grid dispatch terminal as the basis for dispatching.

[0079] The Shapley value game algorithm outputs the power output priority ratio for each photovoltaic cluster. Specifically, the marginal value function of each cluster is defined as the cluster's photovoltaic output coefficient and the dynamic cluster topology matrix. This involves a weighted combination of the backup power priority weights of the corresponding data centers. Based on this, the Shapley value calculation formula is used to sum the weighted marginal contributions of all possible cluster combinations to obtain the Shapley value for each cluster. Finally, the output priority ratio is obtained through normalization. This serves as a crucial basis for generating the collaborative output baseline value in the power prediction module. This method can fairly and reasonably reflect the scheduling importance of each photovoltaic cluster within the context of multi-factor fusion modeling.

[0080] The meteorological forecasting and processing module is used to build a forecasting model based on historical meteorological grid data and generate meteorological forecast data for several future periods. The meteorological forecast data includes future irradiance, cloud cover rate and temperature indicators, which are used as one of the input parameters of the power forecasting module. The power prediction module is used to generate a collaborative output baseline value for each photovoltaic cluster based on the output priority ratio and meteorological forecast data, and send the collaborative output baseline value to the power grid dispatch terminal.

[0081] Specifically, the generation of the collaborative output baseline value includes: Correcting the gridded photovoltaic output coefficient based on meteorological forecast data Calculate the predicted output value for each grid; Based on the output priority ratio of the photovoltaic power distribution, the original output of each cluster is weighted and corrected to generate the most predictable collaborative output baseline value for photovoltaic power generation.

[0082] The weighted correction adopts a segmented strategy: When meteorological forecasts indicate that cloud cover will exceed 50% in the next hour, the output weight of low-priority clusters will be reduced. The purpose of this setting is that increased cloud cover means a significant reduction in local sunlight, leading to unstable or even sudden drops in photovoltaic output. If all clusters are still maintained to output according to their original priorities at this time, it is easy to cause unstable scheduling and large deviations in power prediction. Therefore, by actively reducing the output proportion of "low-priority clusters", resources are concentrated on more critical or stable areas.

[0083] When the priority weight of backup power demand Exceeding the limit At that time, the output ratio of the corresponding cluster will be forcibly increased by at least 20%. The purpose of this setting is to adjust the priority of backup power demand. Exceeding the limit This indicates an extreme power emergency, therefore, the output ratio of the corresponding cluster should be forcibly increased by at least 20% to ensure power supply guarantee capability under extreme scenarios.

[0084] For example: There are three photovoltaic clusters: Cluster A, Cluster B, and Cluster C, which correspond to multiple meteorological grid areas.

[0085] Weather forecast data (from grid-based weather service): Grid 1: , , Grid 2: , , Grid 3: , , The output calculation model is as follows:

[0086] The parameters are set as follows: , , Reference temperature value Then, the grid output coefficient is corrected, and the grid predicted output is calculated as follows: The predicted output of grid 1 is:

[0087] Similarly, 135.24 ,therefore: Grid 1 → Belongs to Cluster A → 129.024 Grid 2 → Belongs to Cluster B → 135.24 Grid 3 → Belongs to cluster C → 119.258 The predicted output (unweighted) for each cluster is: 129.024 for cluster A, 135.24 for cluster C, and 119.258 for cluster C.

[0088] Then, the output priority ratio is introduced for weighted correction. Assume that the priority ratio of each cluster output by the collaborative allocation module is: Cluster A: 0.3, Cluster B: 0.5, Cluster C: 0.2; After weighted adjustment, the baseline value of the collaborative output of each cluster is: Cluster A: 129.024 × 0.3 ≈ 38.71W Cluster B: 135.24 × 0.5 = 67.62W Cluster C: 119.258 × 0.2 ≈ 23.85W The final baseline values ​​for collaborative output are: Cluster A: 38.71W, Cluster B: 67.62W, Cluster C: 23.85W; Total coordinated baseline output: 38.71 + 67.62 + 23.85 = 130.18 W. This result will be used as a reference benchmark for coordinated power generation and sent to the grid dispatch system.

[0089] The system also includes a feedback optimization module, used to dynamically adjust the parameters of the grid output calculation model, backup power demand prediction model, or Shapley value game algorithm based on the deviation between the actual output data of the power grid dispatch terminal and the collaborative output baseline value. This avoids system dispatch deviations caused by prediction errors and improves the prediction accuracy, dispatch rationality, and model robustness of the collaborative output prediction system in complex environments.

[0090] In summary, this invention establishes a meteorological grid processing module to construct a multi-dimensional grid data system based on real-time irradiance, cloud cover, and temperature. Combined with a grid output calculation model, it generates photovoltaic output coefficients for each grid, achieving a spatially distributed expression of photovoltaic output capacity. Compared to traditional regional averaging models, this approach significantly improves prediction accuracy and adaptability to weather abrupt changes, helping to capture the impact of local meteorological differences on photovoltaic output and providing a more reliable foundation for subsequent collaborative prediction. This invention also obtains the output characteristics and geographical information of each photovoltaic cluster through a dynamic cluster modeling module, and generates a dynamic cluster topology matrix based on the similarity of output fluctuations, representing the complementary output relationships between clusters. This topology matrix dynamically reflects the power synergy potential of each cluster under different operating conditions, enabling the system to fully utilize output differences to achieve load balancing during scheduling, thereby improving the overall system stability and resource utilization efficiency. Finally, the collaborative allocation module of this invention calculates the output priority ratio of each photovoltaic cluster by inputting the photovoltaic output coefficients, dynamic cluster topology matrix, and backup power demand priority weights into a Shapley value game algorithm, serving as an important input for power prediction. This method, considering multiple factors such as meteorological conditions, complementarity, and backup power demand, achieves fairness and rationality in resource allocation through game theory, avoiding the limitations of traditional weighted methods that are prone to imbalance, and improving the effectiveness and accuracy of collaborative output prediction.

[0091] 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 distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling, characterized in that, include: The meteorological grid processing module is used to acquire meteorological grid data of the target area. The meteorological grid data includes the real-time irradiance, cloud coverage and temperature of each grid, and generates the photovoltaic output coefficient corresponding to each grid through the grid output calculation model. The dynamic cluster modeling module is used to acquire real-time power output data and geographical location data of distributed photovoltaic clusters, and generate a dynamic cluster topology matrix based on the similarity of power output fluctuations. The dynamic cluster topology matrix is ​​used to characterize the power output complementarity relationship between clusters. The backup power demand processing module is used to obtain real-time power consumption and remaining energy storage capacity data of the data center's servers, and generate backup power demand priority weights through the backup power demand prediction model. The collaborative allocation module is used to input the photovoltaic output coefficient, dynamic cluster topology matrix and backup power demand priority weight into the Shapley value game algorithm to calculate the output priority ratio of each photovoltaic cluster. The meteorological forecasting and processing module is used to build a forecasting model based on historical meteorological grid data and generate meteorological forecast data for several future periods. The meteorological forecast data includes future irradiance, cloud cover rate and temperature indicators, which are used as one of the input parameters of the power forecasting module. The power prediction module is used to generate a collaborative output baseline value for each photovoltaic cluster based on the output priority ratio and meteorological forecast data, and send the collaborative output baseline value to the power grid dispatch terminal.

2. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 1, characterized in that: The grid-based power output calculation model calculates the photovoltaic power output coefficient using the following formula: in, For the first Photovoltaic output coefficient of each grid; Real-time irradiance; Cloud coverage; Real-time temperature; , These are calibration parameters; Here is the temperature decay function, as follows: in, These are calibration parameters; This is a reference temperature value.

3. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 1, characterized in that: Generating a dynamic cluster topology matrix based on the power output fluctuation similarity includes the following steps: Obtain the output data of each photovoltaic node in the distributed photovoltaic cluster within the most recent T time periods, and construct an output time series matrix. ; Based on the time series matrix Using Pearson correlation coefficient, dynamic time warping, or mutual information methods, the output similarity between any two photovoltaic nodes is calculated. Form a similarity matrix ; For the similarity matrix Perform threshold filtering or clustering operations to obtain the dynamic cluster topology matrix. ,in: At preset time intervals, the output similarity is recalculated based on the updated output time series, and the topology matrix is ​​dynamically updated.

4. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 1, characterized in that: The backup power demand priority weights are generated through the backup power demand prediction model, including the following steps: Real-time collection of server cluster power consumption data in the data center and the remaining capacity of the energy storage system ; Calculate the total power consumption fluctuation coefficient of the data center ; Analysis of the rate of change of remaining energy storage capacity based on sliding time window : Establish a backup power demand urgency assessment function : in, This indicates the urgency assessment value of backup power, reflecting the current system's urgent demand for electricity. As the benchmark volatility coefficient, This is the power fluctuation weighting coefficient. This is the weighting coefficient for remaining energy storage capacity. The weighting coefficient for the energy storage change trend is... ; Describes the minimum capacity warning threshold set for the energy storage system; urgency assessment value Mapping to weight interval Generate priority weights That is, the weights are generated using a linear normalization formula, as follows: in, Priority weight; This is the lower bound of the priority weight; This represents the upper limit of priority weight; This represents the minimum value of the urgency assessment, used for normalization. This represents the maximum value of the urgency assessment, used for normalization.

5. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 4, characterized in that: In the aforementioned backup power demand urgency assessment function: When the remaining energy storage capacity is below the warning threshold At that time, the weighting factor of the remaining energy storage capacity will be automatically increased. Up to 1.5 times the original value; When server cluster power consumption data suddenly increases, the power volatility weighting coefficient Use the maximum value within the sliding window.

6. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 1, characterized in that: The output priority ratio of each photovoltaic cluster is calculated using the Shapley value game algorithm, including the following steps: Define the marginal value function for each photovoltaic cluster, which is composed of a weighted combination of the following factors; Based on all possible combinations of photovoltaic clusters, the marginal contribution of each cluster is weighted and summed using the Shapley value calculation formula to obtain the Shapley value of each photovoltaic cluster. The Shapley values ​​of each photovoltaic cluster are normalized to obtain the output priority ratio of each photovoltaic cluster, which is used to generate the collaborative output baseline value in the power prediction module.

7. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 6, characterized in that: The factors include the photovoltaic output coefficient of the photovoltaic cluster, the output complementarity information represented in the dynamic cluster topology matrix, and the backup power priority weight of the corresponding data center.

8. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 1, characterized in that: In the power prediction module, the generation of the cooperative output baseline value includes: Correcting the gridded photovoltaic output coefficient based on meteorological forecast data Calculate the predicted output value for each grid; Based on the output priority ratio of the photovoltaic power distribution, the original output of each cluster is weighted and corrected to generate the most predictable collaborative output baseline value for photovoltaic power generation.

9. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 8, characterized in that: The weighted correction adopts a segmented strategy: When meteorological forecast data indicates that the cloud coverage rate will be greater than 50% in the next hour, the output weight of low-priority clusters will be reduced first. When the priority weight of backup power demand Exceeding the limit At that time, the output ratio of the corresponding cluster will be forcibly increased by at least 20%.

10. The distributed photovoltaic power collaborative prediction system based on gridded meteorological fusion and dynamic cluster modeling as described in claim 1, characterized in that: It also includes a feedback optimization module, which dynamically adjusts the parameters of the grid output calculation model, the backup power demand prediction model, or the Shapley value game algorithm based on the deviation between the actual output data of the power grid dispatch terminal and the collaborative output baseline value.

Images