Spatial-temporal correlation-based wind-solar power prediction method and device

Abstract

The invention discloses a wind-solar power prediction method and device based on spatial-temporal correlation, and relates to the technical field of field power prediction. According to the main technical scheme, the method comprises the following steps: selecting a to-be-tested node object from all node objects in a new energy field; a future micro-scale numerical weather forecast corresponding to the node object to be measured is obtained, the future micro-scale numerical weather forecast is obtained after a future mesoscale numerical weather forecast corresponding to the node object to be measured is subjected to localization downscaling processing, and the future mesoscale numerical weather forecast is output in a local numerical forecasting mode of the new energy field area; and inputting the future micro-scale numerical weather forecast into the power prediction model corresponding to the to-be-measured node object to obtain the future prediction power corresponding to the to-be-measured node object. The method mainly aims at reducing the influence of actual working conditions with complex landforms and changeable weather conditions on new energy field power prediction in a new energy field and improving the power prediction precision.

Description

Technical Field

[0001] The present application relates to the technical field of field power prediction, and particularly to a wind-solar power prediction method and device based on spatio-temporal correlation. Background Art

[0002] With the increasing global demand for renewable energy, wind energy and solar energy, as important components of clean energy, have been widely applied and developed. Power prediction of new energy fields (such as wind farms and photovoltaic power stations) is crucial for the dispatching, operation, and optimization of power systems. Accurate wind-solar power prediction can not only improve the stability and reliability of the power grid, but also effectively reduce operating costs and promote the efficient utilization of renewable energy.

[0003] Currently, the wind-solar power prediction in the prior art mainly combines meteorological data with historical power data and constructs a prediction model by mining the statistical laws therein to carry out the prediction of wind power. Although this method is computationally simple, the meteorological data used is mostly medium-scale numerical weather prediction, and it is difficult for medium-scale numerical weather prediction to reflect the local microclimate characteristics formed by the micro-scale terrain undulation and vegetation cover difference inside the new energy field, resulting in an obvious deviation between the prediction and the actual power generation. Moreover, when constructing the power prediction model, the spatio-temporal characteristics in the data are not fully and systematically extracted and utilized, resulting in limited generalization ability and prediction accuracy of the power prediction model. Especially when the new energy field faces the actual working conditions of complex terrain and diverse weather conditions, the accuracy of power prediction will have a large deviation. Summary of the Invention

[0004] In view of the above problems, the present application provides a wind-solar power prediction method and device based on spatio-temporal correlation, and the main purpose is to reduce the influence of the actual working conditions of complex terrain and diverse weather conditions on the power prediction of the new energy field and improve the accuracy of power prediction.

[0005] To solve the above technical problems, the present application proposes the following solutions:

[0006] In the first aspect, the present application provides a wind-solar power prediction method based on spatio-temporal correlation, and the method includes:

[0007] Select a node object to be measured from all node objects in the new energy field;

[0008] Obtain the future micro-scale numerical weather forecast corresponding to the node object to be measured, where the future micro-scale numerical weather forecast is obtained by locally downscaling the future medium-scale numerical weather forecast corresponding to the node object to be measured, and the future medium-scale numerical weather forecast is output by using the local numerical forecast model of the new energy field;

[0009] Input the future micro-scale numerical weather forecast into the power prediction model corresponding to the node object to be measured, and obtain the future predicted power corresponding to the node object to be measured. The power prediction model is trained according to the time series features, spatial correlation features, and seasonal features extracted by a sliding time window from the historical micro-scale numerical weather forecast and historical power performance data corresponding to the node object to be measured.

[0010] In a second aspect, the present application provides a wind-solar power prediction device based on spatio-temporal correlation. The device includes:

[0011] A selection unit for selecting a node object to be measured from all node objects in the new energy field area;

[0012] A first acquisition unit for acquiring the future micro-scale numerical weather forecast corresponding to the node object to be measured. The future micro-scale numerical weather forecast is obtained by locally downscaling the future meso-scale numerical weather forecast corresponding to the node object to be measured, and the future meso-scale numerical weather forecast is output by using the local numerical forecast model of the new energy field area;

[0013] A first processing unit for inputting the future micro-scale numerical weather forecast into the power prediction model corresponding to the node object to be measured, and obtaining the future predicted power corresponding to the node object to be measured. The power prediction model is trained according to the time series features, spatial correlation features, and seasonal features extracted by a sliding time window from the historical micro-scale numerical weather forecast and historical power performance data corresponding to the node object to be measured.

[0014] To achieve the above object, according to a third aspect of the present application, there is provided a storage medium. The storage medium includes a stored program. When the program runs, it controls the device where the storage medium is located to execute the wind-solar power prediction method based on spatio-temporal correlation in the first aspect above.

[0015] To achieve the above object, according to a fourth aspect of the present application, there is provided a processor. The processor is used to run a program. When the program runs, it executes the wind-solar power prediction method based on spatio-temporal correlation in the first aspect above.

[0016] With the above technical solution, a method and device for predicting wind and light power based on spatio-temporal correlation provided by the present application, when it is necessary to predict wind and light power in a new energy field area based on spatio-temporal correlation, first, a node object to be measured is selected from all node objects in the new energy field area, and then, future micro-scale numerical weather forecasts of the node object to be measured are obtained. The future micro-scale numerical weather forecasts are obtained after local downscaling of the corresponding future meso-scale numerical weather forecasts of the node object to be measured. The future meso-scale numerical weather forecasts are output by using the local numerical weather prediction model of the new energy field area. Finally, the future micro-scale numerical weather forecasts are input into the power prediction model corresponding to the node object to be measured to obtain the future predicted power corresponding to the node object to be measured. The power prediction model is trained according to the time series features, spatial correlation features, and seasonal features extracted by the sliding time window from the historical micro-scale numerical weather forecasts and historical power performance data corresponding to the node object to be measured. The technical solution provided by the present application selects the node object to be measured from multiple node objects, which can perform subsequent power prediction in a targeted manner. Moreover, the future micro-scale numerical weather forecasts corresponding to the node object to be measured are obtained by local downscaling of the meso-scale numerical weather forecasts, which can more accurately approximate the actual meteorological conditions of the node object to be measured. Especially in complex terrain areas, such as mountainous and valley terrains, it can more accurately simulate the changes of key meteorological factors such as wind speed and temperature, significantly improving the ability to capture the local microclimate characteristics under complex terrain conditions. At the same time, the power prediction model adopts the sliding time window method to comprehensively explore the correlations among time series features, spatial correlation features, and seasonal features, considering not only the changing trends in time but also the distribution characteristics in space, enabling the power prediction model to better adapt to complex meteorological and terrain conditions, enhancing the generalization ability of the model, and enabling it to maintain a high prediction accuracy when facing different meteorological conditions and topographies. By improving the ability to capture local microclimate characteristics and making full use of spatio-temporal correlation, the present application is particularly suitable for actual working conditions with complex terrain and changing weather, ensuring the accuracy and reliability of wind and light power prediction in the new energy field area.

[0017] The above description is only an overview of the technical solution of the present application. In order to be able to understand the technical means of the present application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of the present application more obvious and understandable, the following specifically gives the specific implementation manners of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiments. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present application. Also, throughout the drawings, the same reference numerals are used to represent the same components. In the drawings:

[0019] Figure 1 Shows a flowchart of a method for predicting wind and light power based on spatio-temporal correlation provided by an embodiment of the present application;

[0020] Figure 2 Shows a flowchart of another method for predicting wind and light power based on spatio-temporal correlation provided by an embodiment of the present application;

[0021] Figure 3 Shows a block diagram of the composition of a device for predicting wind and light power based on spatio-temporal correlation provided by an embodiment of the present application;

[0022] Figure 4 Shows a block diagram of the composition of another device for predicting wind and light power based on spatio-temporal correlation provided by an embodiment of the present application. Detailed Embodiments

[0023] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0024] An embodiment of the present application provides a method for predicting wind and light power based on spatio-temporal correlation. Through this method, the influence of actual working conditions such as complex terrain and changeable weather on power prediction in a new energy field area can be reduced, and the accuracy of power prediction can be improved. The specific implementation steps are as Figure 1 shown and include:

[0025] 101. Select a node object to be measured from all node objects in the new energy field area.

[0026] Among them, the node objects are pre-divided according to the prediction scale requirements corresponding to the new energy plant area.

[0027] In this step, the node object to be measured is one or more of the multiple node objects pre-divided according to the prediction scale requirements in the new energy plant area in advance. A node object represents a power prediction target, which can be each wind turbine or each group of adjacent photovoltaic panels, or the new energy plant area can be divided into multiple sub-areas with equal areas according to the field terrain and equipment layout, or it can be a terrain unit or functional area in the new energy plant area. Specifically, it can be determined according to the power prediction type corresponding to the prediction scale requirements, including but not limited to ultra-short-term prediction, short-term prediction, and medium-term prediction. For example, when the time scale is less than 6 hours (ultra-short-term prediction), each wind turbine or each group of adjacent photovoltaic panels is used as a node for fine division in space to capture locally fast-changing information. When the time scale is between 6 and 24 hours (short-term prediction), combined with the field terrain and equipment layout, the field is divided into several areas with roughly equal areas as nodes. When the time scale is greater than 24 hours (medium-term prediction), larger terrain units or functional areas (such as valley areas, mountaintop areas, etc.) are used as nodes for division. For the node objects divided by different power prediction types above, they can be stored in an efficient data management system (such as a relational database, a NoSQL database, a distributed file system). A hierarchical data structure is adopted to organize data according to different prediction scale requirements (ultra-short-term, short-term, medium-term) and spatial division methods (wind turbine level, sub-area level, functional area level). At the same time, an index is created for each node object, and a retrieval interface is generated to facilitate users to select the node object to be measured according to the current operating state and scheduling requirements. For example, in a wind farm, those wind turbines that are about to enter the high wind speed area can be selected as the node objects to be measured.

[0028] 102. Obtain the future micro-scale numerical weather forecast of the node object to be measured.

[0029] Among them, the future micro-scale numerical weather forecast is obtained after local downscaling of the future meso-scale numerical weather forecast corresponding to the node object to be measured, and the future meso-scale numerical weather forecast is output by using the local numerical forecast model of the new energy plant area.

[0030] In this step, obtain the meso-scale numerical weather forecast data of the new energy plant area for a future period from a global-scale or regional-scale numerical weather forecast system (such as GFS, ECMWF), including key meteorological variables such as wind speed, temperature, and humidity, and extract the meso-scale numerical weather forecast corresponding to the node object to be measured. It should be noted that the numerical weather forecast model corresponding to this numerical weather forecast system can be the original numerical forecast model or a local numerical forecast model localised according to data such as the topography and landform of the new energy plant area. In this regard, this embodiment does not make a limitation.

[0031] Obtain a high - resolution digital elevation model (DEM) of the new - energy field area through satellite remote sensing, UAV aerial photography, or other geographic information systems (GIS), etc., with a resolution of 30 meters or higher, as the high - resolution terrain data of the new - energy field area. And set up a suitable mesoscale - microscale coupling model (such as WRF - CALME, WRF - ARW), which can convert mesoscale numerical weather forecast data into microscale numerical weather forecasts applicable to the specific new - energy field area. The microscale numerical weather forecast is used to characterize the local microclimate characteristics corresponding to different microscale terrains. At the same time, according to the actual geographical location and climate characteristics of the new - energy field area, set the initial conditions and boundary conditions of the coupling model.

[0032] Extract the high - resolution terrain data of the node object to be measured from the high - resolution terrain data of the new - energy field area, and input the high - resolution terrain data of the node object to be measured and the corresponding future mesoscale numerical weather forecast of the node object to be measured into the mesoscale - microscale coupling model. Run the mesoscale - microscale coupling model, and then the future microscale numerical weather forecast corresponding to the node object to be measured can be generated.

[0033] Furthermore, in order to make the mesoscale numerical weather forecast data more suitable for the actual working conditions of the new - energy field area with complex terrain and changing weather, it is necessary to optimize and adjust the original numerical forecast model of the new - energy field area so that the mesoscale numerical weather forecast data is more in line with the actual situation of the new - energy field area. The specific implementation process is as follows: optimize the three - dimensional grid of the original numerical forecast model according to the high - resolution terrain data, meteorological observation data, and the three - dimensional scales of the main weather systems in the new - energy field area; obtain the physical schemes corresponding to the optimized original numerical forecast model and the arrangement order of the physical schemes, and use the multi - criterion decision - making analysis algorithm to determine the optimal physical scheme for each season in turn according to the arrangement order; use the non - linear optimal perturbation algorithm to determine the sensitive parameters in each optimal physical scheme, and adjust the sensitive parameters according to the meteorological observation data to obtain the local numerical forecast model corresponding to the new - energy field area.

[0034] Among them, the local numerical weather prediction model is used to output the local mesoscale numerical weather forecast for the new energy field area. The mesoscale numerical weather forecast includes the future mesoscale numerical weather forecast for prediction and the historical mesoscale numerical weather forecast for model training. The meteorological observation data refers to the historical meteorological observation data covering many years, including key meteorological variables such as wind speed, temperature, and humidity. Analyze the three-dimensional scales of the weather systems (such as fronts, cyclones, anticyclones, etc.) affecting this area to understand their impacts on the original numerical weather prediction model in the numerical weather prediction system. According to the size and complexity of the new energy field area, select an appropriate initial grid resolution, such as 1 km, 5 km, etc. Larger grids are suitable for large-scale weather systems, while smaller grids are suitable for capturing local microclimate characteristics. Set the boundary conditions of the initial grid to ensure compatibility with the data of the global-scale or regional-scale numerical weather prediction systems (such as GFS, ECMWF). First, use the genetic algorithm to optimize the grid resolution and layout. By simulating the natural selection process, find the optimal grid configuration. Then, use the particle swarm optimization algorithm to search for the optimal combination of grid parameters based on the principle of swarm intelligence. Next, adopt the adaptive grid refinement technology to dynamically adjust the grid resolution so that higher resolutions are available in areas with complex terrain or where important meteorological phenomena occur. Compare and verify the numerical weather prediction results generated by the optimized grid with the historical meteorological observation data, and adjust the grid parameters according to the verification results so that the optimized three-dimensional grid configuration can accurately reflect the meteorological characteristics of the new energy field area. Select multiple candidate physical schemes from the existing numerical weather prediction models, including microphysical schemes, combinations of atmospheric boundary layer schemes + surface layer schemes, and combinations of longwave radiation schemes + shortwave radiation schemes. Specifically, it can be based on seasons, and they are sorted in the order of: microphysical scheme → combination of atmospheric boundary layer + surface layer schemes → combination of longwave + shortwave radiation schemes. Use multi-criteria decision analysis methods (such as TOPSIS, AHP, etc.), comprehensively consider different evaluation indicators such as forecast accuracy, computational efficiency, and stability. According to the local climate characteristics, divide a year into several seasons (such as spring, summer, autumn, winter), and select the most suitable combination of physical schemes for each season, that is, determine the optimal physical scheme corresponding to each season. Conduct a sensitivity analysis on each physical scheme to identify the scheme that performs best under different seasonal conditions and make corresponding adjustments. After determining the optimal physical scheme, use the nonlinear optimal perturbation algorithm (CNOP) to identify the sensitive parameters in each optimal physical scheme. For example, the ice crystal formation parameter in the cloud microphysical process, the turbulent mixing coefficient in the atmospheric boundary layer scheme, etc.By calculating the optimal perturbation, the initial condition or parameter perturbation that can maximize the change in the forecast result is found, so as to identify the most critical sensitive parameters, which include but are not limited to microphysical parameters (cloud droplet condensation nucleus concentration, ice crystal formation threshold, etc.), radiation transfer parameters (aerosol optical depth, cloud water path, etc.), surface process parameters (soil moisture, vegetation coverage, etc.), and boundary layer parameters (roughness length, friction velocity, etc.). Then, according to the long-term meteorological observation data, the identified sensitive parameters are adjusted until the output of the local numerical prediction model is as consistent as possible with the meteorological observation data. The optimized three-dimensional grid, optimal physical scheme, and adjusted sensitive parameters are integrated into a complete numerical prediction model to form a local numerical prediction model applicable to the new energy field area, making it more in line with the actual situation of the new energy field area and ensuring the accuracy of the mesoscale numerical weather forecast data.

[0035] After determining the object of the node to be measured, the mesoscale numerical weather forecast of the object of the node to be measured for a future period of time can be determined through the local numerical prediction model, and then it is downscaled through the high-resolution terrain data and the mesoscale-microscale coupling model to obtain the microscale numerical weather forecast for a future period of time, that is, the future microscale numerical weather forecast corresponding to the object of the node to be measured, so as to execute step 103.

[0036] 103. Input the future microscale numerical weather forecast into the power prediction model corresponding to the object of the node to be measured to obtain the future predicted power corresponding to the object of the node to be measured.

[0037] Among them, the power prediction model is trained according to the time series features, spatial correlation features, and seasonal features extracted by the sliding time window from the historical microscale numerical weather forecast and historical power performance data corresponding to the object of the node to be measured.

[0038] In this step, the historical mesoscale numerical weather forecast can be obtained in advance by using the local numerical prediction model obtained in Step 102, and it is downscaled by high-resolution terrain data and a mesoscale-microscale coupling model to obtain the historical microscale numerical weather forecast. At the same time, historical power performance data is extracted from on-site measurement devices such as the wind turbine SCADA system and the photovoltaic power station monitoring system in the new energy field area. It should be noted that the historical microscale numerical weather forecast and the historical power performance data are time-sequentially corresponding. Due to the actual working conditions of complex terrain and variable weather in some new energy field areas, the distribution of light and wind speed is extremely uneven in different regions. Even within the same wind farm or photovoltaic power station, the resources received by wind turbines and photovoltaic panels at different positions will vary greatly. Moreover, the meteorological conditions in local areas change complexly, and there may be local severe weather such as strong winds, heavy rains, and low temperatures, which also have a significant impact on the power generation of wind turbines and photovoltaic panels. Therefore, in order to better adapt to the uneven resource distribution caused by this complex terrain and establish more appropriate power predictions for the meteorological characteristics of different regions, better cope with the diversity and uncertainty of meteorological conditions, and thus improve the accuracy of power prediction, local power prediction can be carried out for each node object in the new energy plant area. Spatiotemporal features are extracted from the historical microscale numerical weather forecast and the historical power performance data corresponding to each node object to obtain the corresponding time series features (such as mean, variance, autocorrelation coefficient, spectral analysis results, etc.), spatial correlation features (considering the spatial correlation between adjacent node objects, such as distance weights, spatial autocorrelation coefficients, covariance matrices, etc.), and seasonal features (such as annual cycle, monthly cycle, daily cycle, etc.). Based on this, a power prediction model corresponding to each node object is trained separately. The power prediction model can select long short-term memory network (LSTM), gated recurrent unit (GRU), random forest (Random Forest), gradient boosting tree (GBDT), etc. During the training process, methods such as Bayesian optimization and genetic algorithms can be used to optimize the hyperparameters of the model to ensure the optimal performance of the model on the training set. The extracted time series features, spatial correlation features, and seasonal features are used as inputs, combined with the historical power performance data as output labels to train the power prediction model.

[0039] Based on the node object to be measured in Step 101, the power prediction model corresponding to the node object to be measured is called, and the microscale numerical weather forecast data for a future period in Step 102 is input into the power prediction model, and the predicted power of the node object to be measured for a future period can be output, that is, the future predicted power.

[0040] Furthermore, due to the actual working conditions of some new energy fields, such as complex terrain and diverse weather conditions, after the division based on node objects, the local microclimate is still variable. Therefore, in order to adapt to the changing meteorological conditions and equipment states, further ensure the accuracy of the power prediction model, and thus improve the accuracy and reliability of wind and light power prediction in the new energy field, the corresponding future actual power can be compared with it, and corresponding processing measures can be taken based on the comparison result to avoid deviation in the future predicted power output by the power prediction model. The specific implementation process is as follows: Obtain the future actual power corresponding to the node object to be measured; Compare the future actual power with the future predicted power, and determine whether the power prediction model corresponding to the node object to be measured meets the update condition according to the comparison result. The update condition is used to represent whether the number of qualified error indicators between the future actual power and the future predicted power exceeds the preset number threshold; If not, mark the power prediction model corresponding to the node object to be measured as an unavailable state, and trigger the retraining operation corresponding to the power prediction model corresponding to the node object to be measured, and then switch the unavailable state to an available state after the retraining operation is completed.

[0041] After a period of time in the future for the node object to be measured, the actual power data of each node object to be measured can be collected in real time from on-site measurement devices such as the SCADA system of the wind farm or the monitoring system of the photovoltaic power station, or the actual power data uploaded by the new energy field can be obtained through the power dispatching system, that is, the future actual power. Referring to this period of time, align the future actual power and the future predicted power according to the same time stamp so that the two can be directly compared. Multiple error indicators and their respective index thresholds are preset. The error indicators include but are not limited to mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and relative error, etc. The index threshold can be an error magnitude threshold or a duration threshold, and no limit is imposed on this. Calculate the mean square error (MSE), mean absolute error (MAE), root mean square error (RMSE), and relative error of the two respectively, and compare them with their respective index thresholds. If it exceeds, it means that the error indicator is unqualified, and if it does not exceed, it means that the error indicator is qualified. Count the number of qualified error indicators, and set a reasonable number threshold to compare with the number of qualified indicators. If it exceeds, it is considered that the model needs to be updated, and if it does not exceed, it is considered that the model does not need to be updated.

[0042] When it is considered that the model needs to be updated, since the model error is large and it is not suitable for use, the power prediction model corresponding to the node object to be measured can be marked as unavailable, the prediction output of the model can be suspended to avoid providing inaccurate prediction results, and a notice can be sent to relevant personnel to inform that the model has entered the unavailable state and the retraining process can be started. At the same time, the update operation of the power prediction model is automatically triggered. Specifically, the latest micro-scale numerical weather forecast and actual power performance data can be collected, and the corresponding time series features, spatial correlation features, and seasonal features can be extracted as the input for retraining to obtain the updated power prediction model for the node object to be measured, and the model status can be switched back to the available state to resume normal prediction output.

[0043] It should be noted that before automatically triggering the update operation of the power prediction model, an artificial review link can also be set to check and confirm whether the model really needs to be updated, so as to avoid wasting computing resources and ensure that the model can be called at any time.

[0044] Based on the above Figure 1As can be seen from the implementation method, a wind-solar power prediction method based on spatio-temporal correlation provided by this application is as follows: when wind-solar power prediction for a new energy field area needs to be carried out based on spatio-temporal correlation, first, a node object to be measured is selected from all node objects in the new energy field area. Then, future micro-scale numerical weather forecasts of the node object to be measured are obtained. The future micro-scale numerical weather forecasts are obtained after local downscaling of the corresponding future meso-scale numerical weather forecasts of the node object to be measured, and the future meso-scale numerical weather forecasts are output by using the local numerical prediction model of the new energy field area. Finally, the future micro-scale numerical weather forecasts are input into the power prediction model corresponding to the node object to be measured to obtain the corresponding future predicted power of the node object to be measured. The power prediction model is trained according to the time series features, spatial correlation features, and seasonal features extracted by the sliding time window from the corresponding historical micro-scale numerical weather forecasts and historical power performance data of the node object to be measured. The technical solution provided by this application selects the node object to be measured from multiple node objects, which can perform subsequent power prediction in a targeted manner. Moreover, the future micro-scale numerical weather forecasts corresponding to the node object to be measured are obtained by local downscaling of the meso-scale numerical weather forecasts, which can more accurately approximate the actual meteorological conditions of the node object to be measured. Especially in complex terrain areas, such as mountainous and valley terrains, it can more accurately simulate the changes of key meteorological factors such as wind speed and temperature, significantly improving the ability to capture the local microclimate characteristics under complex terrain conditions. At the same time, the power prediction model adopts the sliding time window method to comprehensively explore the correlations among time series features, spatial correlation features, and seasonal features, considering not only the changing trends in time but also the distribution characteristics in space, enabling the power prediction model to better adapt to complex meteorological and terrain conditions and enhancing the generalization ability of the model, so that it can still maintain a high prediction accuracy when facing different meteorological conditions and topographies. By improving the ability to capture local microclimate characteristics and making full use of spatio-temporal correlation, this application is particularly suitable for actual working conditions with complex terrain and changing weather, ensuring the accuracy and reliability of wind-solar power prediction in the new energy field area.

[0045] Furthermore, the preferred embodiment of this application is a detailed description of the specific construction process of the power prediction model based on the above Figure 1 and its specific steps are as Figure 2 shown, including:

[0046] 201. Divide the new energy plant area according to the prediction scale requirements to obtain multiple node objects.

[0047] In this step, the prediction scale requirements of the user can be collected in advance. The prediction scale requirements are used to characterize the power prediction types at different time scales, including at least one of ultra-short-term prediction, short-term prediction, and medium-term prediction. Among them, if the power prediction type is ultra-short-term prediction, each wind turbine or each group of adjacent photovoltaic panels in the new energy plant area is taken as a node object; if the power prediction type is short-term prediction, the new energy plant area is divided into multiple sub-areas of equal area according to the field terrain and equipment layout, and each sub-area is taken as a node object; if the power prediction type is medium-term prediction, each terrain unit or functional area in the new energy plant area is taken as a node object.

[0048] For the node objects divided by different power prediction types as described above, they can be stored in an efficient data management system (such as a relational database, a NoSQL database, a distributed file system). Adopting a hierarchical data structure, the data is organized according to different prediction scale requirements (ultra-short-term, short-term, medium-term) and spatial partitioning methods (wind turbine level, sub-area level, functional area level). At the same time, an index is created for each node object, and a retrieval interface is generated to facilitate the user to select the node object to be measured according to the current operating status and scheduling requirements.

[0049] It should be noted that since the meteorological conditions such as wind speed and temperature at different locations may vary significantly, especially under complex terrain conditions (such as mountains and valleys), this difference is particularly obvious. By dividing the new energy plant area into multiple node objects, the local microclimate characteristics corresponding to each wind turbine or each group of adjacent photovoltaic panels, each sub-area, or each terrain unit can be better captured. Each node object can be finely modeled according to its unique meteorological and geographical characteristics, avoiding the rough estimation of the entire field area by a single model, improving the accuracy of power prediction, and reducing the overfitting risk that may exist in a single model, ensuring that the performance of the model on unseen data is more stable and reliable. Moreover, dividing the complex new energy plant area into multiple relatively simple node objects can decompose the originally complex global problem into several local problems that are easy to handle. The prediction models of each node object can be independently developed and optimized, reducing the complexity of the overall system. And once an abnormal situation is found (such as a large deviation between the actual power and the predicted value of a certain node), corresponding measures can be quickly taken for targeted adjustment. At the same time, according to the prediction scale requirements (such as ultra-short-term, short-term, medium-term prediction), different node partitioning strategies can be selected. For example, ultra-short-term prediction can be refined to each wind turbine or each group of adjacent photovoltaic panels, while medium-term prediction can be divided according to terrain units or functional areas to ensure the prediction accuracy at different time scales.

[0050] 202. Obtain the historical mesoscale numerical weather forecasts corresponding to each node object based on the local numerical weather prediction model, and obtain the historical power performance data corresponding to each node object based on the on-site measurement devices in the new energy field area.

[0051] In this step, obtain the historical mesoscale numerical weather forecast data for a past period from the local numerical weather prediction model, including key meteorological variables such as wind speed, temperature, humidity, and air pressure. Select data covering at least one year to ensure that different seasonal meteorological conditions are covered. The time resolution of the data is usually hourly or shorter. Obtain the historical power performance data from on-site measurement devices such as the SCADA system of the wind farm or the monitoring system of the photovoltaic power station, and ensure that the time resolution of the historical power performance data matches that of the historical mesoscale numerical weather forecast data, usually at the minute or hourly level. Clean outliers, align the time and space dimensions, fill in missing values, and perform standardization / normalization processing to eliminate the influence of dimensions and ensure the comparability between different variables.

[0052] 203. Use the high-resolution terrain data and the preset mesoscale-microscale coupling model to downscale the historical mesoscale numerical weather forecasts corresponding to each node object to obtain the historical microscale numerical weather forecasts corresponding to each node object.

[0053] Among them, the historical microscale numerical weather forecasts are used to characterize the local microclimate characteristics corresponding to the node objects.

[0054] In this step, select WRF-CALME and WRF-ARW as the mesoscale-microscale coupling models in this embodiment. The mesoscale-microscale coupling model is used to convert the mesoscale numerical weather forecast data into microscale numerical weather forecasts applicable to the specific new energy field area. The initial conditions and boundary conditions of the coupling model can be set according to the actual geographical location and climate characteristics of the new energy field area, including terrain, land cover type, vegetation distribution, etc. Since the node objects have been divided, the corresponding high-resolution terrain data and historical mesoscale numerical weather forecast data can also be determined. Therefore, input the high-resolution terrain data and historical mesoscale numerical weather forecast data corresponding to each node object into this mesoscale-microscale coupling model and run this mesoscale-microscale coupling model to generate the historical microscale numerical weather forecasts corresponding to each node object. Randomly select several test points in each node object, compare the obtained historical microscale numerical weather forecasts with the actual meteorological observation data to verify the downscaling effect. If the evaluation result meets the accuracy requirements, the finally generated historical microscale numerical weather forecasts applicable to the new energy field area are used for subsequent spatio-temporal feature extraction and model training.

[0055] 204. Extract spatio-temporal features according to the sliding time window based on the historical micro-scale numerical weather forecasts and historical power performance data corresponding to each node object, and obtain the time series features, spatial correlation features, and seasonal features corresponding to each node object.

[0056] In this step, extract the corresponding time series features, spatial correlation features, and seasonal features from the historical micro-scale numerical weather forecasts and historical power performance data corresponding to each node object according to the sliding time window. The sliding time window can be 1 hour, 6 hours, 24 hours, etc., which corresponds to the power prediction type, that is, different power prediction types correspond to different sliding time window sizes, so as to ensure the effectiveness of feature extraction. The time series features include mean, variance, autocorrelation coefficient, spectral analysis results, etc. It can capture the dynamic changes of meteorological conditions and power output in a short period of time (such as minute level, hour level), which is particularly important for ultra-short-term prediction. It can accurately reflect the impact of instantaneous wind speed, light intensity, etc. on the power generation, and by analyzing the time series data over a long time span (such as day, month, year), long-term trends and patterns can be identified, helping the prediction model to better adapt to factors such as climate change and equipment aging, thereby improving the prediction accuracy. The spatio-temporal features consider the spatial correlation between adjacent node objects, including distance weight, spatial autocorrelation coefficient, covariance matrix, etc. It can fully consider the topographic differences at different positions in the new energy field area, such as altitude, slope, vegetation cover, etc. These factors have a significant impact on the local microclimate. And combined with the equipment layout (such as the distance between wind turbines, the orientation of photovoltaic panels), the interaction between each node object can be more accurately simulated, improving the reliability of the prediction, avoiding the rough estimation of the entire field area by a single model, reducing the risk of overfitting, and making the performance of the model on unseen data more stable and reliable. The seasonal features include annual cycle, monthly cycle, daily cycle, etc., ensuring that the model can adapt to the changes in different seasons. It can capture the periodic changes of meteorological conditions and power output throughout the year, such as low temperature and low light in winter, high temperature and high light in summer, etc., ensuring that the model performs consistently in different seasons. And by considering the seasonal features, the model can better adapt to long-term climate change and environmental factors, maintaining the stability and reliability of the prediction results, and can also help identify abnormal situations, such as extreme weather events or equipment failures, and take measures in advance for adjustment and maintenance.

[0057] Furthermore, there are many meteorological factors in historical microscale numerical weather forecasts. However, not all meteorological factors have a significant impact on wind and solar power, and for different node objects, the meteorological factors that have a significant impact on wind and solar power may also be different. Therefore, in order to reduce noise interference, focus on strongly relevant factors in a targeted manner, and adapt to changing meteorological conditions and equipment states, thereby improving the accuracy of prediction, the key meteorological factor set for each node object can be determined through correlation analysis. The specific process is as follows: For each node object, calculate the correlation coefficient between each meteorological factor in the historical microscale numerical weather forecast and the wind and solar power in the historical power performance data, and determine the key meteorological factor set corresponding to the wind and solar power based on the correlation coefficient. The key meteorological factor set is composed of meteorological factors that are strongly correlated with the wind and solar power; based on the key meteorological factor set, extract spatio-temporal features from the historical microscale numerical weather forecast and historical power performance data corresponding to each node object according to a sliding time window, and obtain the time series features, spatial correlation features, and seasonal features corresponding to each node object.

[0058] In this step, extract the time series data of each meteorological factor (such as wind speed, temperature, humidity, air pressure, etc.) from the historical microscale numerical weather forecast, and extract the time series data of the wind and solar power from the historical power performance data. Any one of the correlation algorithms such as Pearson correlation coefficient, Spearman rank correlation coefficient, and mutual information can be used to calculate the correlation coefficient between each meteorological factor and the wind and solar power. A correlation threshold can be set for comparison. If it exceeds, it is considered a strongly relevant meteorological factor; if it does not exceed, it is considered a non-strongly relevant meteorological factor. Alternatively, multiple of the correlation algorithms such as Pearson correlation coefficient, Spearman rank correlation coefficient, and mutual information can be used to calculate the correlation coefficient between each meteorological factor and the wind and solar power respectively, and the proportion exceeding the correlation threshold can be statistically calculated. A proportion threshold can be set for comparison. If it exceeds, it is considered a strongly relevant meteorological factor; if it does not exceed, it is considered a non-strongly relevant meteorological factor. In this regard, this embodiment does not make a limitation. The strongly relevant meteorological factors are summarized into a set, that is, the key meteorological factor set corresponding to the node object to be measured. Referring to this key meteorological factor set, extract the corresponding time series features, spatial correlation features, and seasonal features from the historical microscale numerical weather forecast and historical power performance data corresponding to each node object according to a sliding time window, that is, comprehensively capture the key features that significantly affect the wind and solar power from the three dimensions of time, space, and season, so as to improve the accuracy and generalization ability of the prediction model.

[0059] Furthermore, to simultaneously consider the interactions between multiple meteorological factors, capture more complex non-linear relationships, and take into account the time lag effect of meteorological factors on wind and solar power, thereby improving the accuracy of subsequent predictions. Specifically, for each node object, using the multiple correlation algorithm, calculate the multiple correlation coefficient corresponding to the strongest lag period within the preset time lag range for each meteorological factor and wind and solar power; compare each correlation coefficient in the multiple correlation coefficients with its respective preset correlation threshold, and count the proportion of coefficients exceeding the correlation threshold; regard the meteorological factors with a coefficient proportion greater than the preset proportion as key meteorological factors to construct a set of key meteorological factors.

[0060] In this step, a reasonable time lag range is preset to capture the lag effect of meteorological factors on wind and solar power. For example, it can be set to 6 hours, that is, consider the lag period from -6 to +6 hours. The multiple correlation algorithm consists of Pearson correlation coefficient, Spearman rank correlation coefficient, and mutual information, and calculates the correlation coefficient between each meteorological factor and wind and solar power within the preset time lag range. For each meteorological factor x i and wind and solar power y j , traverse all lag periods τ∈[-T,T], and calculate the correlation coefficient ρ ij (τ) for each lag period, find the lag period τ * that makes the correlation coefficient the largest, and record the strongest correlation coefficient corresponding to this lag period

[0061] The specific expression is:

[0062]

[0063] where is the multiple correlation coefficient, representing the correlation between the i-th meteorological factor and wind and solar power at different lag periods, x i,t represents the observed value of the i-th meteorological factor at time t, y j,t represents the observed value of wind and solar power at time t, τ is the time lag, T is the preset time lag range, and n is the number of samples.

[0064] Set a preset correlation threshold θ i for each meteorological factor, and a preset proportion P of exceeding the correlation threshold. Compare the strongest correlation coefficients of each meteorological factor with their respective preset correlation thresholds θ i respectively, and calculate the proportion P i of coefficients exceeding the correlation threshold. The specific expression is:

[0065]

[0066] The coefficient ratio P i Regarding meteorological factors greater than the preset ratio P as key meteorological factors, and constructing a set S of key meteorological factors. The specific expression is:

[0067] S = {i | P i > P};

[0068] Combining the selected key meteorological factors into a set S for subsequent spatio-temporal feature extraction and prediction model training.

[0069] 205. Respectively perform model training based on the time series features, spatial correlation features, and seasonal features corresponding to each node object to obtain a power prediction model corresponding to each node object.

[0070] Among them, the power prediction model is used to predict future power performance according to future microscale numerical weather forecasts.

[0071] In this step, a suitable model architecture is preselected. Specifically, long short-term memory network (LSTM), gated recurrent unit (GRU), random forest (Random Forest), or gradient boosting tree (GBDT) can be selected. Using the extracted time series features, spatial correlation features, and seasonal features as inputs, combined with historical power performance data as output labels, train the power prediction model. In this process, appropriate loss functions (such as mean squared error MSE, mean absolute error MAE) and optimizers (such as Adam, SGD) can be used for model training to ensure that the model converges to the best state. Use batch training (BatchTraining) or online learning (Online Learning) methods to gradually update the model parameters and improve the training efficiency. Use an independent test set to verify the performance of the trained model to ensure that it meets the expected requirements. Save the trained power prediction model for prediction when future microscale numerical weather forecasts are input. Specifically, the model can be deployed to the production environment to ensure that it can operate efficiently in real-time application scenarios and be maintained and updated regularly.

[0072] Furthermore, as an implementation of the above Figure 1-2 method embodiment shown, the embodiment of the present application provides a spatio-temporal correlation-based wind-solar power prediction device. This device is used to reduce the impact of the actual working conditions of complex terrain and changing weather in the new energy field area on power prediction in the new energy field area and improve the accuracy of power prediction. The embodiment of this device corresponds to the foregoing method embodiment. For the convenience of reading, the details in the foregoing method embodiment will not be repeated one by one in this embodiment. However, it should be clear that the device in this embodiment can correspondingly implement all the content in the foregoing method embodiment. Specifically as Figure 3 shown, this device includes:

[0073] A selection unit 301 is configured to select a node object to be measured from all node objects in the new energy field area;

[0074] A first acquisition unit 302 is configured to acquire a future microscale numerical weather forecast corresponding to the node object to be measured. The future microscale numerical weather forecast is obtained by locally downscaling the future mesoscale numerical weather forecast corresponding to the node object to be measured, and the future mesoscale numerical weather forecast is output by using a local numerical prediction model of the new energy field area;

[0075] A first processing unit 303 is configured to input the future microscale numerical weather forecast into a power prediction model corresponding to the node object to be measured to obtain a future predicted power corresponding to the node object to be measured. The power prediction model is trained according to time series features, spatial correlation features, and seasonal features extracted by a sliding time window from the historical microscale numerical weather forecast and historical power performance data corresponding to the node object to be measured.

[0076] Further, as Figure 4 shown, the apparatus further includes:

[0077] An optimization unit 304 is configured to optimize a three-dimensional grid of an original numerical prediction model according to high-resolution terrain data, meteorological observation data, and three-dimensional scales of weather systems in the new energy field area before determining a node object to be measured from all node objects in the new energy field area;

[0078] A determination unit 305 is configured to acquire a physical scheme corresponding to the optimized original numerical prediction model and an arrangement order of the physical scheme, and sequentially determine an optimal physical scheme corresponding to each season by using a multi-criteria decision analysis algorithm according to the arrangement order;

[0079] An adjustment unit 306 is configured to determine sensitive parameters in each of the optimal physical schemes by using a non-linear optimal perturbation algorithm, and adjust the sensitive parameters according to the meteorological observation data to obtain a local numerical prediction model of the new energy field area. The local numerical prediction model is used to output a locally downscaled mesoscale numerical weather forecast of the new energy field area, and the mesoscale numerical weather forecast includes the future mesoscale numerical weather forecast and the historical mesoscale numerical weather forecast.

[0080] Further, as Figure 4 shown, the apparatus further includes:

[0081] A division unit 307 is configured to divide the new energy plant area according to the prediction scale requirement to obtain a plurality of the node objects before determining a node object to be measured from all node objects in the new energy field area;

[0082] A second acquisition unit 308, configured to obtain the historical mesoscale numerical weather forecast corresponding to each of the node objects based on the local numerical weather forecast model, and obtain the historical power performance data corresponding to each of the node objects based on on-site measurement devices in the new energy field area;

[0083] A second processing unit 309, configured to perform downscaling processing on the historical mesoscale numerical weather forecast corresponding to each of the node objects by using the high-resolution terrain data and a preset mesoscale-microscale coupling model, to obtain the historical microscale numerical weather forecast corresponding to each of the node objects, where the historical microscale numerical weather forecast is used to characterize the local microclimate characteristics corresponding to the node objects;

[0084] An extraction unit 310, configured to perform spatio-temporal feature extraction on the historical microscale numerical weather forecast and the historical power performance data corresponding to each of the node objects according to a sliding time window, to obtain the time series feature, the spatial correlation feature, and the seasonal feature corresponding to each of the node objects;

[0085] A training unit 311, configured to perform model training respectively based on the time series feature, the spatial correlation feature, and the seasonal feature corresponding to each of the node objects, to obtain a power prediction model corresponding to each of the node objects, where the power prediction model is used to predict the future power performance according to a future microscale numerical weather forecast.

[0086] Further, as Figure 4 shown, the partitioning unit 307 includes:

[0087] A determination module 3071, configured to determine the power prediction type corresponding to the new energy plant area according to the prediction scale requirement;

[0088] A partitioning module 3072, configured to use each wind turbine or each group of adjacent photovoltaic panels in the new energy plant area as a node object if the power prediction type is the ultra-short-term prediction;

[0089] The partitioning module 3072 is further configured to divide the new energy plant area into multiple sub-areas with equal areas according to the terrain of the plant area and the layout of the devices, and use each of the sub-areas as a node object if the power prediction type is the short-term prediction;

[0090] The partitioning module 3072 is further configured to use a terrain unit or a functional area in the new energy plant area as a node object if the power prediction type is the medium-term prediction.

[0091] Further, as Figure 4 shown, the extraction unit 310 includes:

[0092] A calculation module 3101, for each of the node objects, calculates the correlation coefficient between each meteorological factor in the historical micro-scale numerical weather forecast and the wind-solar power in the historical power performance data, and determines a set of key meteorological factors corresponding to the wind-solar power according to the correlation coefficient. The set of key meteorological factors is composed of meteorological factors strongly correlated with the wind-solar power;

[0093] An extraction module 3102, configured to perform spatio-temporal feature extraction on the historical micro-scale numerical weather forecast and the historical power performance data corresponding to each node object according to the sliding time window based on the set of key meteorological factors, so as to obtain the time series feature, the spatial correlation feature, and the seasonal feature corresponding to each node object.

[0094] Further, as Figure 4 shown, the calculation module 3101 includes:

[0095] For each of the node objects, using a multiple correlation algorithm, calculates the multiple correlation coefficient corresponding to the strongest lag period within a preset time lag range between each meteorological factor and the wind-solar power;

[0096] Compares each correlation coefficient in the multiple correlation coefficients with its respective preset correlation threshold, and counts the proportion of coefficients exceeding the correlation threshold;

[0097] Uses the meteorological factors with the coefficient proportion greater than the preset proportion as key meteorological factors to construct the set of key meteorological factors.

[0098] Further, as Figure 4 shown, the device further includes:

[0099] A third acquisition unit 312, configured to acquire the future actual power corresponding to the node object to be measured after inputting the future micro-scale numerical weather forecast into the power prediction model corresponding to the node object to be measured to obtain the future predicted power corresponding to the node object to be measured;

[0100] A judgment unit 313, configured to compare the future actual power with the future predicted power, and judge whether the power prediction model corresponding to the node object to be measured meets the update condition according to the comparison result. The update condition is used to represent that the number of qualified error indicators between the future actual power and the future predicted power exceeds a preset number threshold;

[0101] A third processing unit 314, configured to, if not, mark the power prediction model corresponding to the node object to be measured as an unavailable state, and trigger a retraining operation corresponding to the power prediction model corresponding to the node object to be measured, until after the retraining operation is completed, switch the unavailable state mark to an available state.

[0102] Further, an embodiment of the present application further provides a storage medium for storing a computer program, wherein when the computer program runs, it controls the device where the storage medium is located to execute the above-mentioned Figure 1-2 wind and light power prediction method based on spatio-temporal correlation described therein.

[0103] Further, an embodiment of the present application further provides a processor for running a program, wherein when the program runs, it executes the above-mentioned Figure 1-2 wind and light power prediction method based on spatio-temporal correlation described therein.

[0104] In the above embodiments, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0105] It can be understood that the relevant features in the above methods and devices can be referred to each other. In addition, the "first", "second", etc. in the above embodiments are used to distinguish the respective embodiments, and do not represent the advantages and disadvantages of the respective embodiments.

[0106] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments, and will not be elaborated herein.

[0107] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The structure required to construct such a system will be apparent from the above description. In addition, the present application is not directed to any particular programming language. It should be understood that the content of the present application described herein can be implemented using various programming languages, and the description of the specific language above is to disclose the best mode of the present application.

[0108] In addition, the memory may include non-permanent memory in a computer-readable medium, random access memory (RAM), and / or non-volatile memory such as read-only memory (ROM) or flash memory (flash RAM), and the memory includes at least one storage chip.

[0109] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) that contain computer-usable program code.

[0110] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0111] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that realize the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0112] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0113] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0114] The memory may include non-permanent memory in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of a computer-readable medium.

[0115] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0116] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0117] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0118] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A wind and solar power prediction method based on spatiotemporal correlation, characterized in that: The method comprises: Select the node object to be tested from all the node objects in the new energy field; Obtaining a future microscale numerical weather forecast corresponding to the node object to be measured, wherein the future microscale numerical weather forecast is obtained after the future mesoscale numerical weather forecast corresponding to the node object to be measured is processed by localized downscaling, and the future mesoscale numerical weather forecast is output by using a local numerical forecast model of a new energy field area; The future microscale numerical weather forecast is input into the power prediction model corresponding to the node object to be measured to obtain the future predicted power corresponding to the node object to be measured. The power prediction model is trained according to the time series characteristics, spatial correlation characteristics and seasonal characteristics extracted according to the sliding time window based on the historical microscale numerical weather forecast and historical power performance data corresponding to the node object to be measured.

2. The method according to claim 1, characterized in that Before selecting the node object to be tested from all the node objects in the new energy field, the method further includes: The three-dimensional grid of the original numerical forecast model is optimized according to the high-resolution terrain data of the new energy site, meteorological observation data and the three-dimensional scale of the weather system; Obtaining the physical scheme corresponding to the optimized original numerical forecast model and the arrangement order of the physical schemes, and determining the optimal physical scheme corresponding to each season in turn using a multi-criteria decision analysis algorithm according to the arrangement order; A nonlinear optimal perturbation algorithm is used to determine the sensitive parameters in each of the optimal physical solutions, and the sensitive parameters are adjusted according to the meteorological observation data to obtain a local numerical forecast model for the new energy site. The local numerical forecast model is used to output a localized mesoscale numerical weather forecast for the new energy site. The mesoscale numerical weather forecast includes the future mesoscale numerical weather forecast and the historical mesoscale numerical weather forecast.

3. The method according to claim 2, characterized in that Before selecting the node object to be tested from all the node objects in the new energy field, the method further includes: Dividing the new energy plant area according to the forecast scale requirements to obtain a plurality of node objects; Acquire the historical mesoscale numerical weather forecast corresponding to each of the node objects based on the local numerical forecast model, and acquire the historical power performance data corresponding to each of the node objects based on the on-site measurement equipment in the new energy field; Downscaling the historical mesoscale numerical weather forecast corresponding to each of the node objects using the high-resolution terrain data and a preset mesoscale-microscale coupling model to obtain a historical microscale numerical weather forecast corresponding to each of the node objects, wherein the historical microscale numerical weather forecast is used to characterize the local microclimate characteristics corresponding to the node object; Based on the historical microscale numerical weather forecast and the historical power performance data corresponding to each of the node objects, spatiotemporal feature extraction is performed according to a sliding time window to obtain the time series feature, the spatial correlation feature and the seasonal feature corresponding to each of the node objects; Model training is performed based on the time series characteristics, the spatial correlation characteristics, and the seasonal characteristics corresponding to each of the node objects to obtain a power prediction model corresponding to each of the node objects.

4. The method according to claim 3, characterized in that The new energy plant area is divided into nodes according to the forecast scale requirements, including: Determine the power prediction type corresponding to the new energy plant area according to the prediction scale requirement; If the power prediction type is ultra-short-term prediction, each wind turbine or each group of adjacent photovoltaic panels in the new energy plant is regarded as a node object; If the power prediction type is short-term prediction, the new energy plant area is divided into multiple sub-areas of equal area according to the site terrain and equipment layout, and each of the sub-areas is used as a node object; If the power prediction type is a medium-term prediction, a terrain unit or a functional area is used as a node object in the new energy plant area.

5. The method according to claim 3, characterized in that: Based on the historical microscale numerical weather forecast and the historical power performance data corresponding to each of the node objects, spatiotemporal features are extracted according to a sliding time window to obtain time series features, spatial correlation features, and seasonal features corresponding to each of the node objects, including: For each of the node objects, respectively calculate the correlation coefficients between each meteorological factor in the historical microscale numerical weather forecast and the wind and solar power in the historical power performance data, and determine a set of key meteorological factors corresponding to the wind and solar power according to the correlation coefficients, wherein the set of key meteorological factors is composed of meteorological factors that are strongly correlated with the wind and solar power; Based on the set of key meteorological factors, spatiotemporal features of the historical microscale numerical weather forecast and the historical power performance data corresponding to each of the node objects are extracted according to the sliding time window to obtain the time series features, the spatial correlation features and the seasonal features corresponding to each of the node objects.

6. The method according to claim 5, characterized in that For each of the node objects, respectively calculate the correlation coefficients between each meteorological factor in the historical microscale numerical weather forecast and the wind and solar power in the historical power performance data, and determine the key meteorological factor set corresponding to the wind and solar power according to the correlation coefficients, including: For each of the node objects, a multivariate correlation algorithm is used to calculate the multivariate correlation coefficients corresponding to the strongest lag period between each of the meteorological factors and the wind and solar power within a preset time lag range; each correlation coefficient in the multivariate correlation coefficient is compared with the respective preset correlation thresholds, and the proportion of coefficients exceeding the correlation threshold is counted; The meteorological factors whose coefficient ratio is greater than a preset ratio are taken as key meteorological factors to construct the key meteorological factor set.

7. The method according to any one of claims 1 to 6, characterized in that After inputting the future microscale numerical weather forecast into the power prediction model corresponding to the node object to be measured to obtain the future predicted power corresponding to the node object to be measured, the method further includes: Obtaining the future actual power corresponding to the node object to be measured; Compare the future actual power with the future predicted power, and determine whether the power prediction model corresponding to the node object to be tested meets the update condition according to the comparison result, wherein the update condition is used to characterize that the qualified number of error indicators between the future actual power and the future predicted power exceeds a preset number threshold; If not, the power prediction model corresponding to the node object to be tested is marked as unavailable, and a retraining operation corresponding to the power prediction model corresponding to the node object to be tested is triggered, until the retraining operation is completed and then the unavailable state is switched to an available state.

8. A wind and solar power prediction device based on spatiotemporal correlation, characterized in that: The device comprises: A selection unit, used to select a node object to be tested from all node objects in the new energy field; A first acquisition unit is used to acquire a future microscale numerical weather forecast corresponding to the node object to be measured, wherein the future microscale numerical weather forecast is obtained after the future mesoscale numerical weather forecast corresponding to the node object to be measured is processed by localized downscaling, and the future mesoscale numerical weather forecast is output by using a local numerical forecast model of a new energy field area; The first processing unit is used to input the future microscale numerical weather forecast into the power prediction model corresponding to the node object to be measured to obtain the future predicted power corresponding to the node object to be measured. The power prediction model is trained according to the time series characteristics, spatial correlation characteristics and seasonal characteristics extracted according to the sliding time window based on the historical microscale numerical weather forecast and historical power performance data corresponding to the node object to be measured.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is running, the device where the storage medium is located is controlled to execute the wind and solar power prediction method based on spatiotemporal correlation as described in any one of claims 1 to 7.

10. A processor, characterized in that: The processor is used to run a program, wherein the program, when running, executes the wind and solar power prediction method based on spatiotemporal correlation as described in any one of claims 1 to 7.

Images