A multi-level observation station cooperative photovoltaic power prediction method based on satellite internet and related device

Abstract

The application provides a kind of multi-level observation station cooperative photovoltaic power prediction method based on satellite internet and related device, it is related to photovoltaic power prediction technical field.Based on multi-level observation station cooperative network, through real-time acquisition target photovoltaic field station, large photovoltaic field station and micro observation station photovoltaic power and meteorological data, combine high-resolution remote sensing cloud picture and multi-source meteorological data, a high-precision photovoltaic power prediction model is constructed.Through deep learning method and physical modeling, model realizes long time scale trend prediction and short time scale fluctuation prediction respectively, simultaneously, using adaptive weight adjustment mechanism, the data weight of observation station is dynamically optimized, the prediction accuracy is improved, especially in complex weather conditions, the response ability to sudden weather change is improved.Under the layout of large-scale photovoltaic field station, spatial dependence and local weather change are effectively captured, high-precision photovoltaic power prediction is provided for power grid dispatching, and the safe and stable operation of power grid is ensured.

Description

Technical Field

[0001] This application relates to the field of photovoltaic power prediction technology, and in particular to a method and related apparatus for photovoltaic power prediction based on multi-level observation stations using satellite internet. Background Technology

[0002] With the rapid deployment of large-scale photovoltaic power plants, the uncertainty and volatility of their output have placed higher demands on grid dispatch and operation safety. Photovoltaic power prediction is increasingly important as a key support for improving the controllability of new energy and the resilience of system operation. Accurate power prediction is very important for the dispatch and operation of distribution networks with high proportion of distributed photovoltaic penetration in the future and for the local consumption of distributed photovoltaic. Although photovoltaic prediction technology has made significant progress in recent years, it can be divided into model-based prediction methods and data-based prediction methods. However, it still faces many bottlenecks in practical applications, mainly in the following aspects: (1) The output of photovoltaic power is greatly affected by weather factors (such as cloud cover, temperature, wind speed, etc.), especially in local areas. The accuracy of weather change prediction directly affects the accuracy of photovoltaic power prediction. Existing weather data has problems such as low spatiotemporal resolution and delay in supporting photovoltaic power prediction, and cannot effectively provide high-precision short-term photovoltaic power prediction. (2) Photovoltaic power is affected by both spatial and temporal factors. Especially in the case of large-scale photovoltaic power plant layout, the modeling of spatial dependence is crucial. Most current prediction models focus on the prediction of a single station or a small number of stations, ignoring the role of multi-station collaboration in improving prediction accuracy.

[0003] Therefore, how to extract effective spatiotemporal features from large-scale photovoltaic power plants and achieve multi-station collaboration remains an urgent problem to be solved. Summary of the Invention

[0004] To extract effective spatiotemporal features from large-scale photovoltaic power plants and achieve multi-station collaboration, this application provides a method and related apparatus for multi-level observation station collaborative photovoltaic power prediction based on satellite internet.

[0005] Firstly, the photovoltaic power prediction method based on satellite internet and involving multi-level observation stations, provided in this application, adopts the following technical solution:

[0006] A multi-level observation station collaborative photovoltaic power prediction method based on satellite internet includes:

[0007] A multi-level collaborative observation network is constructed, including the target photovoltaic power station, large photovoltaic power stations, and micro observation stations deployed around the target photovoltaic power station; the micro observation stations are equipped with photovoltaic modules, environmental sensors, and satellite communication modules, and realize data backhaul through low-orbit satellites;

[0008] Real-time collection of photovoltaic power data, meteorological data, and remote sensing cloud image data from target photovoltaic power plants, large photovoltaic power plants, and micro observation stations;

[0009] The data transmission method is selected based on the terrestrial network coverage, and the transmission method includes: terrestrial communication network and satellite internet;

[0010] The spatiotemporal correlation between observation stations at all levels and the target photovoltaic power station is calculated based on the spatiotemporal correlation model. The spatiotemporal correlation includes historical power correlation and cloud map evolution trend correlation.

[0011] The information gain weights of each level of observation station are dynamically allocated based on the spatiotemporal correlation.

[0012] Based on the information gain weight, the site with the highest correlation is determined and the photovoltaic power prediction result is output through a dual time scale prediction model.

[0013] The prediction accuracy is optimized by dynamically adjusting the weights of the observation station data based on the prediction error feedback.

[0014] Optionally, selecting the data transmission method based on terrestrial network coverage includes:

[0015] Data is transmitted back through terrestrial communication networks in areas covered by terrestrial networks, including fiber optics, 4G communication networks, and 5G communication networks.

[0016] In areas without terrestrial network coverage, data is transmitted back to the data center via low-Earth orbit satellites through satellite communication modules.

[0017] Optionally, the step of calculating the spatiotemporal correlation between observation stations at all levels and the target photovoltaic power station based on the spatiotemporal correlation model includes:

[0018] Calculation of large-scale photovoltaic power plants Spatiotemporal correlation with the target photovoltaic power station o:

[0019]

[0020] Computational micro-observation station Spatiotemporal correlation with the target photovoltaic power station o:

[0021]

[0022] in, Indicates large-scale photovoltaic power station Correlation with the historical power of the target photovoltaic power station Indicates large-scale photovoltaic power station Correlation between cloud map evolution and target photovoltaic power station o Indicates a miniature observation station Correlation between the historical power of the target photovoltaic power station o and the actual power output.

[0023] Optionally, the calculation of historical power correlation includes:

[0024] For any observation station i and the target photovoltaic power station o, the historical power correlation Calculated using the Pearson correlation coefficient:

[0025]

[0026] in, and These are the power data of the i-th observation station and the target photovoltaic power station o at time t, respectively. and are the average power data of the i-th observation station and the target photovoltaic power station o, respectively.

[0027] Optionally, the calculation of the correlation of cloud map evolution trends includes:

[0028] For large-scale photovoltaic power plants The correlation between the cloud map evolution and the target photovoltaic power station o is expressed as:

[0029]

[0030] in, and They represent the i-th large-scale photovoltaic power station. The cloud map feature data of the target photovoltaic power station at time t. and They represent the i-th large-scale photovoltaic power station. The average value of cloud map feature data of the target photovoltaic power station within the historical time window.

[0031] Optionally, the step of dynamically allocating information gain weights for each level of observation station based on the spatiotemporal correlation includes:

[0032] The contribution of spatiotemporal correlation to the overall network prediction accuracy is normalized and calculated, defining the role of each large-scale photovoltaic power station. and miniature observation stations The information gains are respectively , :

[0033]

[0034]

[0035] Where m and n are the total number of large-scale photovoltaic power plants and micro-observation stations in the selected area, respectively. Indicates large-scale photovoltaic power station The spatiotemporal correlation between the target photovoltaic power station o and the target photovoltaic power station o. This represents the spatiotemporal correlation between the micro-observation station Mj and the target photovoltaic power station o.

[0036] Optionally, before the step of determining the most relevant site based on the information gain weight and outputting the photovoltaic power prediction result through the dual-time-scale prediction model, the method further includes:

[0037] Construct a photovoltaic power prediction model system with dual time scales;

[0038] In the dual-timescale photovoltaic power prediction model system, the XGBoost model is used to fuse multi-source NWP data for long-term trend prediction, and the TCN model combined with a multi-head attention mechanism is used for short-term fluctuation prediction.

[0039] Secondly, this application provides a multi-level observation station collaborative photovoltaic power prediction device based on satellite internet, comprising:

[0040] The network construction module is used to build a multi-level collaborative network of observation stations, including the target photovoltaic power station, large photovoltaic power stations, and micro observation stations deployed around the target photovoltaic power station; the micro observation stations are equipped with photovoltaic modules, environmental sensors, and satellite communication modules, and realize data backhaul through low-orbit satellites;

[0041] The data acquisition module is used to collect photovoltaic power data, meteorological data, and remote sensing cloud image data of the target photovoltaic power station, large photovoltaic power station, and micro observation station in real time;

[0042] The data transmission module is used to select the data transmission method according to the terrestrial network coverage, and the transmission method includes: terrestrial communication network and satellite Internet;

[0043] The calculation module is used to calculate the spatiotemporal correlation between observation stations at all levels and the target photovoltaic power station based on the spatiotemporal correlation model. The spatiotemporal correlation includes historical power correlation and cloud map evolution trend correlation.

[0044] The allocation module is used to dynamically allocate the information gain weights of each level of observation station according to the spatiotemporal correlation.

[0045] The prediction module is used to determine the most relevant sites based on the information gain weights and output photovoltaic power prediction results through a dual-time-scale prediction model.

[0046] The optimization module is used to dynamically adjust the weights of observation station data based on prediction error feedback to optimize prediction accuracy.

[0047] Thirdly, this application provides a computer device, the device comprising: a memory and a processor, wherein the processor, when executing computer instructions stored in the memory, performs the method described above.

[0048] Fourthly, this application provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the method described above.

[0049] In summary, this application, based on a multi-level observation station collaborative network, constructs a high-precision photovoltaic power prediction model by real-time acquisition of photovoltaic power and meteorological data from target photovoltaic power plants, large-scale photovoltaic power plants, and micro-observation stations, combined with high-resolution remote sensing cloud images and multi-source meteorological data. Through deep learning methods and physical modeling, the model achieves both long-term trend prediction and short-term fluctuation prediction. Simultaneously, an adaptive weight adjustment mechanism dynamically optimizes the data weights of the observation stations, improving prediction accuracy, especially enhancing the response capability to sudden meteorological changes under complex weather conditions. In a large-scale photovoltaic power plant layout, it effectively captures spatial dependencies and local meteorological changes, providing high-precision photovoltaic power prediction for power grid dispatch and ensuring the safe and stable operation of the power grid. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiments of this application;

[0051] Figure 2 This is a flowchart illustrating the first embodiment of the multi-level observation station collaborative photovoltaic power prediction method based on satellite internet in this application;

[0052] Figure 3 This is a schematic diagram illustrating the application scenario of the first embodiment of the multi-level observation station collaborative photovoltaic power prediction method based on satellite internet in this application;

[0053] Figure 4 This is the multi-timescale prediction process of the first embodiment of this application;

[0054] Figure 5 This is a flowchart of the data fusion process based on XGboost according to the first embodiment of this application;

[0055] Figure 6 This is a schematic diagram of the TCN principle structure in the short-timescale fluctuation prediction of the first embodiment of this application;

[0056] Figure 7 This is a simulation result diagram of the first embodiment of this application;

[0057] Figure 8 This is a structural block diagram of the first embodiment of the multi-level observation station collaborative photovoltaic power prediction device based on satellite internet, as described in this application. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0059] Reference Figure 1 , Figure 1 This is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiments of this application.

[0060] like Figure 1 As shown, the computer device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The memory 1005 may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk drive. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0061] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0062] like Figure 1 As shown, the memory 1005, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a multi-level observation station collaborative photovoltaic power prediction program based on satellite internet.

[0063] exist Figure 1In the computer device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in this application can be set in the computer device. The computer device calls the satellite Internet-based multi-level observation station collaborative photovoltaic power prediction program stored in the memory 1005 through the processor 1001, and executes the satellite Internet-based multi-level observation station collaborative photovoltaic power prediction method provided in the embodiment of this application.

[0064] This application provides a method for predicting photovoltaic power using a multi-level observation station collaborative method based on satellite internet, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the multi-level observation station collaborative photovoltaic power prediction method based on satellite internet in this application.

[0065] In this embodiment, the multi-level observation station collaborative photovoltaic power prediction method based on satellite internet includes the following steps:

[0066] Step S10: Construct a multi-level observation station collaborative network, including the target photovoltaic power station, large photovoltaic power stations, and micro observation stations deployed around the target power station.

[0067] It should be noted that the terminology explained in this embodiment includes:

[0068] Multi-level observation stations: The multi-level observation stations mentioned in this case include the large photovoltaic power stations (large stations) that have already been deployed and the micro observation stations (small stations) proposed in this case.

[0069] Miniature weather stations: Deployed at different directions and distances around the target site, equipped with photovoltaic modules, environmental sensors, and satellite communication modules. The photovoltaic modules collect irradiance data, the environmental sensors monitor local meteorological parameters in real time, and the satellite communication modules ensure data is transmitted back to the data center via low-Earth orbit satellites, providing integrated energy-communication capabilities. Leveraging high-frequency data acquisition capabilities and small-volume data transmission characteristics, miniature weather stations reflect local meteorological changes in real time, improving forecast accuracy under complex weather conditions.

[0070] XGBoost: XGBoost (Extreme Gradient Boosting) is a highly efficient gradient boosting tree algorithm widely used for processing structured data, especially in regression and classification tasks. It improves prediction accuracy by integrating multiple weak classifiers. In this invention, XGBoost is used for long-term trend prediction of photovoltaic power, fusing multi-source NWP data to provide high-quality irradiance predictions, helping the model predict the photovoltaic power trend for the next 24 hours.

[0071] TCN (Temporal Convolutional Network) is a deep learning model for processing sequential data. Based on a convolutional neural network (CNN) architecture, it excels at capturing temporal features in long-term sequences. In this invention, TCN is combined with a multi-head attention mechanism for short-term fluctuation prediction, capturing power fluctuations caused by rapid weather changes and improving the accuracy of short-term photovoltaic power prediction. Furthermore, in this invention, TCN is also used to fuse and correct trend prediction and fluctuation prediction results, achieving collaborative prediction across multiple time scales.

[0072] In practice, a multi-level observation station collaborative network is constructed, including the target photovoltaic power station, existing large photovoltaic power stations in the region, and newly added micro observation stations. Observation stations at all levels collect photovoltaic power and meteorological data in real time.

[0073] The three types of stations in this embodiment specifically include:

[0074] 1) Target photovoltaic power station (target station): Located in the core area of ​​photovoltaic power generation, it provides real-time photovoltaic power data and historical trend data, provides benchmark data for prediction models, and supports the accuracy of power prediction for the whole network.

[0075] 2) Large-scale photovoltaic power stations (large stations): cover a large area, provide regional trend baselines, supplement target station data, and enhance regional forecasting capabilities.

[0076] 3) Micro-observation Stations (Small Stations): These micro-observation stations are equipped with photovoltaic modules, environmental sensors, and satellite communication modules, enabling data transmission via low-Earth orbit (LEO) satellites. Deployed at different directions and distances around the target site, the photovoltaic modules collect irradiance data, the environmental sensors monitor local meteorological parameters in real time, and the satellite communication modules ensure data transmission back to the data center via LEO satellites, providing integrated "energy-communication" capabilities. Leveraging their high-frequency data acquisition capabilities and small-volume data transmission characteristics, these micro-observation stations reflect local meteorological changes in real time, improving forecast accuracy under complex weather conditions.

[0077] When distributing micro-monitoring stations, the possibility of local weather changes must be considered to ensure coverage of potential monitoring blind spots of the target stations. Through a reasonable station layout, micro-observatories can respond quickly to factors such as cloud movement and local temperature fluctuations, adjust their data output in real time, and improve the overall accuracy of photovoltaic power prediction.

[0078] A collaborative observation network is constructed, with each station collecting photovoltaic power data, meteorological data (such as temperature, humidity, and wind speed), and numerical weather prediction (NWP) data in real time, according to the actual data collection frequency of each level of observation station (generally once every 15 minutes). The data collection cycle is reasonably configured based on the type and function of the observation station to ensure the continuity and real-time nature of data collection.

[0079] Step S20: Real-time acquisition of photovoltaic power data, meteorological data, and remote sensing cloud image data from target photovoltaic power plants, large photovoltaic power plants, and micro observation stations.

[0080] Step S30: Select the data transmission method according to the terrestrial network coverage, wherein the transmission method includes: terrestrial communication network and satellite Internet.

[0081] In specific implementation, the selection of data transmission method based on terrestrial network coverage includes: transmitting data back through terrestrial communication networks in areas with terrestrial network coverage, including fiber optic, 4G, and 5G communication networks; and transmitting data back to the data center via low-orbit satellite through satellite communication modules in areas without terrestrial network coverage.

[0082] It should be noted that in areas with terrestrial network coverage, stations at all levels utilize existing communication methods (fiber optic, 4G / 5G) to transmit observation data back to the data center. In areas without terrestrial network coverage but where critical data can be obtained, micro-observation stations are prioritized for deployment to improve the accuracy of photovoltaic forecasts at target sites. Especially when information gain is high, utilizing satellite internet for data collaboration and real-time data transmission can significantly improve forecast performance and ensure high reliability and low latency of observation data transmission under extreme weather conditions. Figure 3 As shown;

[0083] In areas with terrestrial network coverage, target photovoltaic (PV) power stations and large-scale PV power stations transmit collected PV power data, meteorological data, remote sensing cloud image data, and other monitoring data to the data center via fiber optic or 4G / 5G communication networks. These communication methods offer high bandwidth and low latency, enabling stable transmission of large-scale data. Miniature observation stations in these areas establish communication links with both the target stations and the data center via 4G / 5G networks, ensuring real-time and efficient data transmission.

[0084] In areas without terrestrial network coverage, such as remote regions, mountainous areas, or high-altitude areas, terrestrial communication networks cannot support stable data transmission. In these areas, miniature observation stations establish connections with low-Earth orbit satellites via satellite communication modules, ensuring that various meteorological and photovoltaic power data can be transmitted back to the data center in real time. The low latency and high reliability provided by the satellite link ensure that data transmission remains efficient and accurate even under extreme weather conditions such as unstable weather and cloud cover.

[0085] Step S40: Calculate the spatiotemporal correlation between observation stations at all levels and the target station based on the spatiotemporal correlation model. The spatiotemporal correlation includes historical power correlation and cloud map evolution trend correlation.

[0086] In specific implementation, the step of calculating the spatiotemporal correlation between observation stations at all levels and the target station based on the spatiotemporal correlation model includes:

[0087] Calculation of large-scale photovoltaic power plants Spatiotemporal relevance to target station o:

[0088]

[0089] Computational micro-observation station Spatiotemporal relevance to target station o:

[0090]

[0091] in, Indicates large-scale photovoltaic power station Correlation with historical power of the target station Indicates large-scale photovoltaic power station Correlation between the cloud map evolution and the target station o Indicates a miniature observation station Correlation with the historical power of the target station o.

[0092] It should be noted that the correlation of historical power reflects the correlation between the historical power data of the observation station and the historical power data of the target station, taking into account both temporal continuity and spatial influence. The calculation of historical power correlation includes:

[0093] For any historical power correlation between observation station i and target station o Calculated using the Pearson correlation coefficient:

[0094]

[0095] in, and Let be the power data of the i-th observation station and the target station o at time t, respectively. and are the mean values ​​of the power data of the i-th observation station and the target station o, respectively.

[0096] Understandably, considering that existing large-scale photovoltaic power stations can all acquire remote sensing cloud image data, the correlation of cloud image evolution trends between the target station and the observation station is included in the spatiotemporal correlation calculation to reflect the impact of clouds on the irradiance of the target station. The calculation of cloud image evolution trend correlation includes:

[0097] For large-scale photovoltaic power plants The correlation between the cloud map evolution and the target station o is expressed as:

[0098]

[0099] in, and They represent the i-th large-scale photovoltaic power station. The cloud map feature data of the target station o at time t. and They represent the i-th large-scale photovoltaic power station. The mean of cloud map feature data of target station o within the historical time window.

[0100] Step S50: Dynamically allocate information gain weights for each level of observation station based on the spatiotemporal correlation.

[0101] In specific implementation, the step of dynamically allocating the information gain weights of each level of observation station based on the spatiotemporal correlation includes:

[0102] The contribution of spatiotemporal correlation to the overall network prediction accuracy is normalized and calculated, defining the role of each large-scale photovoltaic power station. and miniature observation stations The information gains are respectively , :

[0103]

[0104]

[0105] Where m and n are the total number of large-scale photovoltaic power plants and micro-observation stations in the selected area, respectively. Indicates large-scale photovoltaic power station The spatiotemporal relevance of the target station o This represents the spatiotemporal correlation between the micro-observation station Mj and the target station o.

[0106] Step S60: Based on the information gain weight, determine the site with the highest correlation and output the photovoltaic power prediction result through the dual time scale prediction model.

[0107] In specific implementation, before the step of determining the most relevant site based on the information gain weight and outputting the photovoltaic power prediction result through the dual-timescale prediction model, the method further includes: constructing a dual-timescale photovoltaic power prediction model system; in the dual-timescale photovoltaic power prediction model system, using the XGBoost model to fuse multi-source NWP data for long-term trend prediction, and using the TCN model combined with a multi-head attention mechanism for short-term fluctuation prediction.

[0108] In practical implementation, a dual-timescale photovoltaic power prediction model system is constructed, consisting of long-term trend prediction and short-term fluctuation prediction. Trend prediction focuses on the long-term evolution trend of photovoltaic power, while fluctuation prediction focuses on rapid fluctuations in photovoltaic power, especially short-term fluctuations caused by weather changes, such as... Figure 4 As shown.

[0109] Long-term trend prediction uses fused multi-source NWP data and photovoltaic system parameters for physical modeling, outputting photovoltaic power prediction results for the next 24 hours; short-term fluctuation prediction uses a deep learning-based model to capture the impact of rapidly changing meteorological factors on photovoltaic power fluctuations, achieving high-precision predictions at the minute and second levels.

[0110] For long-term photovoltaic (PV) trend prediction, observational data from large PV power plants located at relatively far distances from the target site, which have high spatiotemporal correlation gain, are utilized. The power data from these plants can provide regional trend references, helping to predict future long-term PV power changes.

[0111] Specifically, XGboost is used to fuse multi-source NWP data (such as HRRR, RAP, NDFD, GFS, and NAM) to obtain timely and accurate meteorological inputs, such as... Figure 5 As shown. Let the training sample set be... ,in This represents the multi-source NWP meteorological characteristics of the i-th sample. This represents the corresponding target irradiance or power value. The prediction value of XGboost in the t-th iteration is:

[0112]

[0113] in For learning rate, To add a new regression tree for the t-th tree, This represents the function space of the regression tree.

[0114] The objective function of XGBoost is in the following form:

[0115]

[0116] in, For loss functions (such as mean squared error), regularization term Defined as:

[0117]

[0118] In the formula, T represents the number of leaf nodes in the tree. Let be the weight of the j-th leaf node. and This is the regularization coefficient.

[0119] To improve training efficiency, XGBoost performs a second-order Taylor expansion of the loss function and introduces the first and second derivatives for each sample:

[0120]

[0121] The approximate objective function becomes:

[0122]

[0123] XGBoost, as a high-efficiency data fusion module, models the nonlinear correlations in multi-source NWP data and outputs unified, high-quality irradiance prediction results. These predictions, along with system parameters (including module type, installation tilt and azimuth angles, tracking method, inverter parameters, etc.), are then input into the photovoltaic physical model to complete the irradiance-to-power conversion process, ultimately outputting the photovoltaic power output trend.

[0124] For short-term photovoltaic power fluctuation prediction, data from large photovoltaic power plants and micro-observation stations with high spatiotemporal correlation gain and close proximity to the target station are mainly utilized. Micro-observation stations provide high-frequency meteorological data, while large photovoltaic power plants in close proximity can reflect changes in photovoltaic power within a local area. Combining the two helps to more accurately capture the impact of rapid weather changes on photovoltaic power.

[0125] Specifically, the observation dataset is first processed by imputing missing values, removing outliers, aligning time series, and normalizing.

[0126] The processed meteorological sequence is divided into time periods in the form of sliding windows, forming fixed-length input tensors that serve as training samples for the Temporal Convolutional Network (TCN). Each sample contains multidimensional meteorological features from multiple spatial nodes, which are used as model input; the actual power values ​​of the target stations at the corresponding time points are used as supervision targets.

[0127] A basic TCN network is constructed using a multi-layer causal convolutional structure with residual connections, and trained using the target power value as the regression objective. The model effectively models long-term dependencies by dilating convolutions while preserving temporal order. The TCN mainly consists of stacked causal convolutions and dilated convolutions, such as... Figure 6 As shown. Let the input sequence be... The corresponding output is Then the output of the one-dimensional causal convolution at time t is:

[0128]

[0129] Where k is the convolution size, Let be the weight of the i-th convolutional kernel. To further expand the receptive field and avoid excessively deep network stacking, TCN introduces a dilation factor d-fold convolution mechanism, namely dilated convolution, whose expression is as follows:

[0130]

[0131] Where d is the expansion coefficient, which increases with the depth of the network (e.g., d=2l, l is the layer number), which can achieve an exponential expansion of the receptive field, thereby capturing long-distance dependent features such as solar radiation fluctuations and cloud movement in photovoltaic power output.

[0132] By using temporal convolutional grids, hourly and minute-level prediction results can be fused and corrected, enabling multi-scale collaborative prediction from trend evolution to local disturbances, thereby improving the model's response to sudden weather changes and overall prediction accuracy.

[0133] In its specific implementation, this embodiment also includes the following steps:

[0134] During the training phase, each deep learning network is divided into training and testing sets to evaluate model prediction performance. In the data preparation phase, all historical power data, cloud imagery data, high-frequency meteorological data, and multi-source NWP data are integrated to ensure data integrity and consistency. All data undergoes preprocessing through temporal and spatial alignment to ensure accurate matching of observations at each time point and spatial location. The preprocessed data is then input into the deep learning model for training. The XGBoost model is used for trend prediction, capturing the temporal relationships between different NWP data sources and generating high-quality irradiance predictions, which are further used in the irradiance-power conversion of the photovoltaic physics model. Simultaneously, a temporal convolutional network is used to train a fluctuation prediction model to capture short-term power fluctuations caused by meteorological factors.

[0135] During model training, cross-validation was used to evaluate the model's predictive performance, and mean squared error (MSE) and other evaluation metrics were used to quantify the model's performance in trend and volatility prediction. Hyperparameters were optimized and adjusted for different parts of the model to improve the prediction accuracy and generalization ability of each module.

[0136] After training, the model is validated using a test set to ensure its applicability and reliability in real-world data.

[0137] Step S70: Dynamically adjust the data weights of the observation stations based on the prediction error feedback to optimize the prediction accuracy.

[0138] In practical implementation, an adaptive weight adjustment process based on prediction error feedback is designed. By monitoring the prediction error of the target station in real time, the data weights of observation stations at all levels are dynamically adjusted to optimize prediction accuracy. In particular, the multi-directional distribution characteristics of micro-observation stations can enhance the response to local meteorological changes and further improve prediction performance.

[0139] The system monitors the prediction error of the target station in real time to determine the difference between the predicted result and the actual power output. If the prediction error exceeds a set threshold, the system will activate an error feedback mechanism to dynamically adjust the data weights of each level of observation station based on the current prediction error. The data weights of micro-observation stations and large photovoltaic power plants will be adjusted according to their contribution to the prediction accuracy of the target station. In particular, during periods of large power fluctuations, the weight of micro-observation stations will be increased because they are more sensitive to rapid changes in local weather.

[0140] By combining real-time error and historical feedback mechanisms, the data center can gradually optimize the weights of each observation station and improve the model's adaptability under different meteorological conditions.

[0141] In practical implementation, the simulation results of the multi-level observation station collaborative photovoltaic power prediction method based on satellite internet in this embodiment are as follows: Figure 7 As shown.

[0142] This embodiment establishes a multi-level collaborative observation network, including target photovoltaic (PV) power plants, large-scale PV power plants, and micro-observation stations, and optimizes the data utilization of each station based on spatiotemporal correlation. Unlike existing prediction methods that rely heavily on single-station data, this embodiment effectively utilizes the spatiotemporal characteristics of multiple observation stations within a large-scale PV power plant layout, improving prediction accuracy and stability. By fusing data from different types of stations, this embodiment can more accurately reflect local meteorological changes and reduce errors in both short-term and long-term prediction tasks. [Prediction Method Based on Multi-Level Collaborative Observation Network]

[0143] Compared to existing technologies that use meteorological data with low spatiotemporal resolution and high latency, this embodiment utilizes remote sensing cloud images and multi-source NWP data (such as HRRR, RAP, NDFD, GFS, etc.) to fuse the data through deep learning methods. This improvement enhances the timeliness and accuracy of meteorological input, enabling short-term photovoltaic power forecasts to more accurately reflect local meteorological changes. Especially under the influence of factors such as cloud evolution and temperature changes, this embodiment can provide higher-precision forecasts, avoiding the problem of insufficient weather data to support accurate short-term forecasts in existing methods. [Fusion of High-Resolution Remote Sensing Cloud Images and Multi-Source Meteorological Data]

[0144] This embodiment constructs a dual-timescale prediction model to perform trend prediction on a long timescale and fluctuation prediction on a short timescale. In existing technologies, prediction models typically focus only on one timescale, neglecting the potential for multi-timescale collaborative prediction. This embodiment, by fusing long-term trend prediction and short-term fluctuation prediction, can handle more complex meteorological changes, enhances the model's response to sudden meteorological changes, and optimizes the accuracy of photovoltaic power prediction under different meteorological conditions. [Construction of the Long-Term and Short-Term Collaborative Prediction Model]

[0145] In existing technologies, prediction models typically use fixed data weights from observation stations, failing to adjust their contributions based on real-time prediction errors. This embodiment designs an adaptive weight adjustment process based on prediction error feedback to monitor the prediction error of the target station in real time and dynamically adjust the data weights of each observation station. Especially when errors are large, the weights of micro-observation stations and large photovoltaic power plants are optimized based on their contribution to prediction accuracy, enhancing the model's adaptability to rapid weather changes and improving prediction accuracy. [Adaptive Weight Adjustment Mechanism Based on Prediction Error Feedback]

[0146] Protection points in this embodiment:

[0147] 1. Prediction method based on multi-level observation station collaborative network: By establishing a multi-level observation station collaborative network and optimizing the use of the spatiotemporal correlation of each observation station, the photovoltaic power prediction can more accurately reflect the meteorological changes and spatial dependence in the layout of large-scale photovoltaic power plants, and significantly improve the prediction accuracy.

[0148] 2. Fusion of high-resolution remote sensing cloud images and multi-source meteorological data: Fusion of high-resolution remote sensing cloud image data with multi-source meteorological data improves the timeliness and accuracy of meteorological data, thereby improving the accuracy of short-term photovoltaic power forecasting.

[0149] 3. Dual Time Scale Collaborative Prediction Model: By constructing a collaborative prediction model for both long and short time scales, the model handles trend prediction and fluctuation prediction separately, thereby enhancing its comprehensive prediction capability for photovoltaic power changes. It has a significant advantage, especially in dealing with complex weather changes.

[0150] 4. Adaptive weight adjustment mechanism based on prediction error feedback: Real-time monitoring of prediction errors and adjustment of observation station weights, especially high-frequency data from micro-observation stations, improves prediction accuracy in the event of large local weather changes, ensuring the real-time performance and accuracy of photovoltaic power prediction.

[0151] This embodiment, based on a multi-level observation station collaborative network, constructs a high-precision photovoltaic power prediction model by real-time acquisition of photovoltaic power and meteorological data from target photovoltaic power plants, large-scale photovoltaic power plants, and micro-observation stations, combined with high-resolution remote sensing cloud images and multi-source meteorological data. Through deep learning methods and physical modeling, the model achieves both long-term trend prediction and short-term fluctuation prediction. Simultaneously, an adaptive weight adjustment mechanism dynamically optimizes the data weights of the observation stations, improving prediction accuracy, especially enhancing the response capability to sudden meteorological changes under complex weather conditions. In a large-scale photovoltaic power plant layout, it effectively captures spatial dependencies and local meteorological changes, providing high-precision photovoltaic power prediction for power grid dispatching and ensuring the safe and stable operation of the power grid.

[0152] Furthermore, embodiments of this application also propose a computer-readable storage medium storing a program for multi-level observation station collaborative photovoltaic power prediction based on satellite internet. When the program for multi-level observation station collaborative photovoltaic power prediction based on satellite internet is executed by a processor, it implements the steps of the method for multi-level observation station collaborative photovoltaic power prediction based on satellite internet as described above.

[0153] Reference Figure 8 , Figure 8 This is a structural block diagram of the first embodiment of the multi-level observation station collaborative photovoltaic power prediction device based on satellite internet in this application.

[0154] like Figure 8 As shown in the embodiments of this application, the multi-level observation station collaborative photovoltaic power prediction device based on satellite internet includes:

[0155] The network construction module is used to build a multi-level collaborative network of observation stations, including a target photovoltaic power station, a large photovoltaic power station, and micro observation stations deployed around the target power station; the micro observation stations are equipped with photovoltaic modules, environmental sensors, and satellite communication modules, and achieve data backhaul through low-orbit satellites;

[0156] The data acquisition module is used to collect photovoltaic power data, meteorological data, and remote sensing cloud image data of the target photovoltaic power station, large photovoltaic power station, and micro observation station in real time;

[0157] The data transmission module is used to select the data transmission method according to the terrestrial network coverage, and the transmission method includes: terrestrial communication network and satellite Internet;

[0158] The calculation module is used to calculate the spatiotemporal correlation between observation stations at all levels and the target station based on the spatiotemporal correlation model. The spatiotemporal correlation includes historical power correlation and cloud map evolution trend correlation.

[0159] The allocation module is used to dynamically allocate the information gain weights of each level of observation station according to the spatiotemporal correlation.

[0160] The prediction module is used to determine the most relevant sites based on the information gain weights and output photovoltaic power prediction results through a dual-time-scale prediction model.

[0161] The optimization module is used to dynamically adjust the weights of observation station data based on prediction error feedback to optimize prediction accuracy.

[0162] It should be understood that the above are merely illustrative examples and do not constitute any limitation on the technical solution of this application. In specific applications, those skilled in the art can make settings as needed, and this application does not impose any restrictions on this.

[0163] This embodiment, based on a multi-level observation station collaborative network, constructs a high-precision photovoltaic power prediction model by real-time acquisition of photovoltaic power and meteorological data from target photovoltaic power plants, large-scale photovoltaic power plants, and micro-observation stations, combined with high-resolution remote sensing cloud images and multi-source meteorological data. Through deep learning methods and physical modeling, the model achieves both long-term trend prediction and short-term fluctuation prediction. Simultaneously, an adaptive weight adjustment mechanism dynamically optimizes the data weights of the observation stations, improving prediction accuracy, especially enhancing the response capability to sudden meteorological changes under complex weather conditions. In a large-scale photovoltaic power plant layout, it effectively captures spatial dependencies and local meteorological changes, providing high-precision photovoltaic power prediction for power grid dispatching and ensuring the safe and stable operation of the power grid.

[0164] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In practical applications, those skilled in the art can select some or all of it to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0165] In addition, for technical details not described in detail in this embodiment, please refer to the method for multi-level observation station collaborative photovoltaic power prediction based on satellite Internet provided in any embodiment of this application, which will not be repeated here.

[0166] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0167] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0168] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application. The above are only preferred embodiments of this application and do not limit the patent scope of this application. All equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A multi-level observation station cooperative photovoltaic power prediction method based on satellite internet, characterized in that, The method comprises the following steps: Constructing a multi-level observation station cooperative network, including a target photovoltaic station, a large photovoltaic station, and micro observation stations deployed around the target photovoltaic station; the micro observation stations are equipped with photovoltaic components, environmental sensors, and satellite communication modules, and data is transmitted back through low-orbit satellites; Real-time collection of photovoltaic power data, meteorological data, and remote sensing cloud image data of the target photovoltaic station, the large photovoltaic station, and the micro observation stations; Selecting a data transmission mode according to the ground network coverage, including a ground communication network and a satellite Internet; Calculating the spatio-temporal correlation between each level of observation station and the target photovoltaic station based on a spatio-temporal correlation model, including historical power correlation and cloud image evolution trend correlation; Dynamically assigning information gain weights to each level of observation station according to the spatio-temporal correlation; Determining the highest correlation station based on the information gain weights and outputting a photovoltaic power prediction result through a double-time-scale prediction model; Dynamically adjusting the observation station data weight based on the prediction error feedback to optimize the prediction accuracy.

2. The method of claim 1, wherein, The step of selecting a data transmission mode according to the ground network coverage comprises: Transmitting data back through a ground communication network in the ground network coverage area, including an optical fiber, a 4G communication network, and a 5G communication network; Transmitting data back to a data center through a low-orbit satellite in the area without ground network coverage through a satellite communication module.

3. The method of claim 1, wherein, The step of calculating the spatio-temporal correlation between each level of observation station and the target photovoltaic station based on a spatio-temporal correlation model comprises: Computing large photovoltaic farms spatiotemporal correlation with the target station o: Computing micro observatories and spatiotemporal correlation with target photovoltaic farm o: wherein, represents a large photovoltaic plant and the target photovoltaic plant o historical power correlation, represents a large photovoltaic plant and the target photovoltaic plant o evolution of the cloud cover correlation, represents a micro observation station and the target photovoltaic plant o historical power correlation.

4. The method of claim 3, wherein, The calculation of historical power correlation includes: For any one observation station i and target photovoltaic plant o historical power correlation By Pearson correlation coefficient calculation: wherein, and Pi(t) and P0(t) are the power data of the ith observation station and the target photovoltaic plant o at time t, respectively, and Pi and P0 are the mean values of the power data of the ith observation station and the target photovoltaic plant o, respectively.

5. The method of claim 3, wherein, The calculation of cloud image evolution trend correlation includes: For large photovoltaic plants and the evolution of the cloud cover of the target photovoltaic plant o, expressed as: wherein, and respectively denote the cloud map feature data of the i-th large photovoltaic plant and the target photovoltaic plant o at time instant t, and respectively denote the cloud map feature data of the i-th large photovoltaic plant and the target photovoltaic plant o in a historical time window.

6. The method of claim 1, wherein, The step of dynamically assigning information gain weights to each level of observation station according to the spatio-temporal correlation comprises: The contribution of the spatio-temporal correlation degree to the prediction accuracy of the whole network is normalized to define the information gain of each large photovoltaic station and micro observation station as , : wherein m and n are the total number of large photovoltaic plants and micro observation stations in the selected area, respectively, denotes a large photovoltaic plant and the spatio-temporal correlation of the target photovoltaic plant o, denotes a micro observation station and the spatio-temporal correlation of the target photovoltaic plant o.

7. The method of claim 1, wherein, Before the step of determining the highest correlation station based on the information gain weights and outputting a photovoltaic power prediction result through a double-time-scale prediction model, the following steps are further included: Constructing a double-time-scale photovoltaic power prediction model system; In the double-time-scale photovoltaic power prediction model system, an XGBoost model is used to fuse multi-source NWP data for long-time-scale trend prediction, and a TCN model is used to combine a multi-head attention mechanism for short-time-scale fluctuation prediction.

8. A multi-level observation station cooperative photovoltaic power prediction device based on satellite internet, characterized in that, The method comprises the following steps: A network construction module is used to construct a multi-level observation station cooperative network, including a target photovoltaic station, a large photovoltaic station, and micro observation stations deployed around the target photovoltaic station; the micro observation stations are equipped with photovoltaic components, environmental sensors, and satellite communication modules, and data is transmitted back through low-orbit satellites; A data collection module is used to real-time collection of photovoltaic power data, meteorological data, and remote sensing cloud image data of the target photovoltaic station, the large photovoltaic station, and the micro observation stations; A data transmission module is used to select a data transmission mode according to the ground network coverage, including a ground communication network and a satellite Internet; A calculation module is used to calculate the spatio-temporal correlation between each level of observation station and the target photovoltaic station based on a spatio-temporal correlation model, including historical power correlation and cloud image evolution trend correlation; An assignment module is used to dynamically assign information gain weights to each level of observation station according to the spatio-temporal correlation. A prediction module is configured to determine a highest-correlation site based on the information gain weight and output a photovoltaic power prediction result through a double-time-scale prediction model; An optimization module is configured to dynamically adjust the observation station data weight based on a prediction error feedback to optimize the prediction accuracy.

9. A computer device, comprising: The device comprises a memory and a processor, and the processor executes the method according to any one of claims 1 to 7 when running computer instructions stored in the memory.

10. A computer-readable storage medium, characterized in that, The computer program product comprises instructions, and when the instructions are run on a computer, the computer is caused to execute the method according to any one of claims 1 to 7.

Images