A method for constructing a power generation prediction model, a prediction method and a device

Abstract

The present invention discloses a method for constructing a power generation prediction model, a prediction method and a device, and relates to the technical field of power generation prediction, so as to solve the problems of low construction efficiency and low prediction accuracy of the power generation prediction model in the prior art. The construction method includes obtaining a meteorological data set within a preset time length and a site data set of a target power station; based on the preset model prediction time length, a preset weather system scale analysis strategy is used to perform a scale analysis on the target power station to determine the weather impact geometric area of ​​the target power station; based on the meteorological reanalysis data, the site data set and the weather impact geometric area, a machine learning data set is constructed; based on the site data set, meteorological observation data and meteorological reanalysis data, the meteorological characteristic variables affecting the target power station are determined; and a preset meteorological characteristic variable time extraction strategy is used to construct a target power generation prediction model corresponding to the target power station. In this way, the power generation prediction model has high construction efficiency and high prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of electric field power prediction, and in particular to a method for constructing a power generation prediction model, a prediction method and a device. Background Art

[0002] The output power prediction technology for wind and photovoltaic power generation can be divided into two categories based on the use of meteorological data: one is the prediction method based on the historical meteorological observation data of the station, and the other is the prediction method based on numerical weather forecasting.

[0003] The wind and photovoltaic power generation prediction algorithm based on the historical meteorological observation data of the station uses the meteorological observation data of the station over the past period of time (wind speed, wind direction, temperature, humidity, air pressure, etc.) and the output power of the wind and photovoltaic power stations to establish a mapping relationship to obtain the power generation prediction.

[0004] The wind and photovoltaic power generation prediction algorithm based on numerical weather forecast uses the meteorological element grid forecast output by the numerical weather forecast model and performs spatial interpolation in combination with the location of the wind farm to obtain the meteorological element forecast (wind speed, wind direction, temperature, air pressure, humidity and radiation) for the future time period of the wind farm. At the same time, the real-time observation data of the electric field is considered, and the wind and photovoltaic power generation prediction is obtained through the established power prediction model.

[0005] The wind and photovoltaic power prediction algorithm based on the historical meteorological observation data of the station makes predictions by capturing the characteristics of the evolution of meteorological elements in the past period of a single station that affect the power generation of wind and photovoltaic power plants. Therefore, it is difficult to identify changes in weather flow fields. As the forecast time increases, the accuracy of power prediction will decrease significantly. The generally effective prediction time is 6-8 hours.

[0006] The wind and photovoltaic power generation prediction algorithm based on numerical weather forecast can increase the prediction time lead time to 3-10 days according to the forecast time of the weather forecast model used, and can identify the impact of the evolution of the weather system on the future power output of the station. However, the accuracy of this method depends on the function and performance of the weather forecast model used. The coarser the resolution of the model, the greater the error when performing spatial interpolation for the station location, which requires numerical forecast products with the highest possible spatial resolution. However, higher-resolution numerical forecast products require huge computing resources, and the model runs longer. Therefore, the reality is often that it is impossible to provide meteorological element forecasts with high resolution (such as horizontal resolution less than 3 kilometers and vertical resolution less than 10 meters) and fast product generation for station power prediction. Summary of the invention

[0007] The object of the present invention is to provide a method for constructing a power generation prediction model, a prediction method and a device, so as to improve the construction efficiency and prediction accuracy of the power generation prediction model of a power station.

[0008] In order to achieve the above object, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides a method for constructing a power generation prediction model, comprising:

[0010] Acquire a meteorological data set within a preset time period and a site data set of a target power station; the meteorological data set at least includes meteorological observation data;

[0011] Based on a preset model prediction time, a preset weather system scale analysis strategy is used to perform scale analysis on the target power station to determine a weather impact geometric area of ​​the target power station;

[0012] Constructing a machine learning dataset based on the meteorological reanalysis data, the station dataset and the weather impact geometric area;

[0013] Determining meteorological characteristic variables affecting the target power station based on the station data set, the meteorological observation data and the meteorological reanalysis data;

[0014] A preset meteorological characteristic variable time extraction strategy is adopted to extract multiple groups of meteorological characteristic variable training data from the machine learning data set, train a preset power prediction model, and construct a target power generation prediction model corresponding to the target power station.

[0015] In an optional embodiment, the meteorological data set also includes model background field data to be assimilated;

[0016] Before constructing the machine learning dataset based on the meteorological reanalysis data, the station dataset and the weather impact geometric area, the method further includes:

[0017] The meteorological observation data and the model background field data to be assimilated are assimilated to obtain the meteorological reanalysis data.

[0018] In an optional implementation, the method of performing scale analysis on the target power station based on a preset model prediction duration and using a preset weather system scale analysis strategy to determine a weather impact geometric area of ​​the target power station includes:

[0019] If the model prediction duration is the first prediction time advance, a scale analysis of the mesoscale weather system is performed on the target power station to determine a first spatial radius of weather factors affecting future power changes of the target power station; and the first prediction time advance is greater than or equal to a preset time advance threshold;

[0020] Determine a first square area with the target power station as the center and the first space radius as the side length;

[0021] The first square area is determined as the weather impact geometric area.

[0022] In an optional implementation, the method of performing scale analysis on the target power station based on a preset model prediction duration and using a preset weather system scale analysis strategy to determine a weather impact geometric area of ​​the target power station includes:

[0023] If the model prediction duration is a second prediction time advance, a scale analysis of a small-scale weather system is performed on the target power station to determine a second spatial radius at which weather factors affect future power changes of the target power station; wherein the second prediction time advance is less than the first prediction time advance;

[0024] Determine a second square area with the target power station as the center and the second space radius as the side length;

[0025] The second square area is determined as the weather impact geometric area.

[0026] In an optional implementation, constructing a machine learning dataset based on the meteorological reanalysis data, the station dataset and the weather impact geometric area includes:

[0027] Selecting target meteorological reanalysis data covering the weather impact geometric area from the meteorological reanalysis data;

[0028] The target meteorological reanalysis data and the site data set are constructed as the machine learning data set.

[0029] In an optional implementation, the using of a preset meteorological characteristic variable time extraction strategy to extract multiple sets of meteorological characteristic variable training data from the machine learning data set includes:

[0030] If the model prediction duration is the first prediction time advance, a set of meteorological characteristic variable training data is extracted from the machine learning data set every first preset time period to obtain the multiple sets of meteorological characteristic variable training data.

[0031] In an optional implementation, the using of a preset meteorological characteristic variable time extraction strategy to extract multiple sets of meteorological characteristic variable training data from the machine learning data set includes:

[0032] If the model prediction duration is the second prediction time advance, a group of meteorological characteristic variable training data is extracted from the machine learning data set every second preset time period to obtain the multiple groups of meteorological characteristic variable training data; wherein the second preset time period is smaller than the first preset time period.

[0033] In an optional implementation, the data in the site data set is collected once every third preset time period; wherein the third preset time period is less than or equal to the second preset time period.

[0034] In a second aspect, the present invention further provides a power generation prediction method, which uses the target power generation prediction model constructed by the above-mentioned power generation prediction model construction method to predict the power generation of a target power station.

[0035] Compared with the prior art, the construction method and prediction method of the power generation prediction model provided by the present invention first use meteorological reanalysis data (meteorological reanalysis field) and power station power generation data to build a model. Compared with the traditional wind and solar power generation prediction algorithm based on historical meteorological observation data of a single station, it takes into account the impact of changes in the upstream weather system in the atmospheric flow field on the power generation of the downstream power station, and has a better prediction effect; secondly, the spatial radius of the meteorological factors affecting the future power prediction changes of the station is determined according to the scale characteristics of the weather system, and the size of the affected space is determined by the impact radius, thereby narrowing the spatial range covered by the power prediction modeling data, which can greatly reduce the time of prediction model training and improve training efficiency; finally, the machine learning feature variable time extraction strategy is used to enable the model to better learn relevant features, thereby achieving better power prediction results. In summary, the power generation prediction model constructed using the above method can take into account the impact of the evolution of the weather system on the power generation, as well as the impact of the geographical environment around the site and the status of the power generation equipment. Therefore, the power prediction accuracy is higher and has a longer forecast time. At the same time, since the constructed target power generation prediction model is mature, it can be used directly for prediction, unlike the numerical weather forecast in the prior art, each prediction requires recalculation of the difference between all data values. Therefore, the power generation prediction model in this scheme requires less computing resources, faster calculation speed, and higher prediction efficiency.

[0036] In a third aspect, the present invention further provides a device for constructing a power generation prediction model, comprising:

[0037] An acquisition module, used to acquire a meteorological data set within a preset time period and a site data set of a target power station; the meteorological data set at least includes meteorological observation data;

[0038] A scale analysis module, configured to perform scale analysis on the target power station based on a preset model prediction duration and using a preset scale analysis strategy to determine a weather-affected geometric area of ​​the target power station;

[0039] A first construction module is used to construct a machine learning dataset based on the meteorological reanalysis data, the station dataset and the weather impact geometric area;

[0040] A determination module, configured to determine the meteorological characteristic variables affecting the target power station based on the station data set, the meteorological observation data and the meteorological reanalysis data;

[0041] The second construction module is used to adopt a preset meteorological characteristic variable time extraction strategy, extract multiple groups of meteorological characteristic variable training data from the machine learning data set, train a preset power prediction model, and construct a target power generation prediction model corresponding to the target power station. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0043] Figure 1 One of the flow charts of a method for constructing a power generation prediction model provided by an embodiment of the present invention;

[0044] Figure 2 A second flow chart of a method for constructing a power generation prediction model provided by an embodiment of the present invention;

[0045] Figure 3 It is a structural schematic diagram of a device for constructing a power generation prediction model of the present invention. DETAILED DESCRIPTION

[0046] In order to clearly describe the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, words such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and their order is not limited. Those skilled in the art can understand that words such as "first" and "second" do not limit the quantity and execution order, and words such as "first" and "second" do not necessarily limit them to be different.

[0047] It should be noted that, in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0048] In the present invention, "at least one" means one or more, "more than one" means two or more. "And / or" describes the association relationship of the associated objects, indicating that three types of relationships may exist.

[0049] like Figure 1 As shown, an embodiment of the present invention provides a method for constructing a power generation prediction model, which may include:

[0050] Step 110: Acquire a meteorological data set within a preset time period and a site data set of a target power station; the meteorological data set at least includes meteorological observation data;

[0051] Specifically, the target power station includes at least one power station.

[0052] For example, the target power station may be a wind farm, or a photovoltaic power station, or multiple power stations such as a wind farm and a photovoltaic power station.

[0053] Specifically, the meteorological data set includes at least meteorological observation data and model background field data to be assimilated.

[0054] The observation channels for meteorological observation data may include ground automatic stations, soundings, aircraft reports, ships, offshore buoys, wind profiler radars, weather radars, and satellite cloud wind guides.

[0055] The data interval of the meteorological observation data may be at least once every hour.

[0056] The horizontal resolution of the model background field data can be no less than 3 kilometers, and the time interval can be no less than 1 hour. The meteorological elements must at least include the ground layer zonal wind U, meridional wind V, temperature T, specific humidity q, air pressure P, solar shortwave radiation, cloud amount, cloud thickness and zonal wind U and meridional wind V of 13 conventional high-altitude pressure layers.

[0057] After obtaining the meteorological data set, the meteorological observation data and the model background field data to be assimilated can be assimilated to obtain meteorological reanalysis data. Among them, the regional three-dimensional variational assimilation system (GrapesMeso 3DVAR) and cloud analysis system of the Global / Reginal Assimilation and Prediction System (Grapes) can be used to perform data assimilation analysis on the meteorological observation data and the model background field data to be assimilated to obtain meteorological reanalysis data.

[0058] It should be noted that the meteorological elements corresponding to the meteorological reanalysis data include at least two categories: the first category is high-altitude elements: such as U wind, V wind, temperature, potential height, specific humidity, etc.

[0059] The meteorological reanalysis data corresponding to the high-altitude elements can include the data corresponding to 13 vertical standard atmospheric pressure layers: 50hPa, 100hPa, 150hPa, 200hPa, 250hPa, 300hPa, 400hPa, 500hPa, 600hPa, 700hPa, 850hPa, 925hPa, and 1000hPa.

[0060] The second category is ground elements: for example, temperature T2 and humidity q2 at a height of 2 meters, latitudinal wind U10 and longitudinal wind V10 at a height of 10 meters, and sea level pressure.

[0061] The process of forming the site data set of the target power station includes: collecting a variety of power station related data including at least basic information of the target power station, power generation data and shutdown period data.

[0062] If the target power station is a wind farm, then the power generation data and shutdown period data of the entire wind farm, for example, every 15 minutes, can be collected. The data time length is equal to the time length of the meteorological reanalysis data. Basic information of the wind farm is also required, such as the coordinates of the center point of the wind farm, wind turbine model, number of wind turbines, installed capacity, hub height, cut-in wind speed, cut-out wind speed and rated wind speed.

[0063] If the target power station is a photovoltaic power station, then the power generation data and outage period data of the entire photovoltaic power station, for example, every 15 minutes, can be collected. The data time length is equal to the time length of the meteorological reanalysis data. The basic information of the photovoltaic power station is also required, such as the coordinates of the center point of the photovoltaic power station, the model of the component, the number of components, the safety inclination angle, the installation azimuth angle, the distance between the front and rear rows, the installed capacity, etc.

[0064] Step 120: Based on the preset model prediction time, a preset weather system scale analysis strategy is used to perform scale analysis on the target power station to determine the weather impact geometric area of ​​the target power station;

[0065] Specifically, the model prediction time refers to the amount of advance time for model prediction. The target power generation prediction model needs to predict the power generation of the future model prediction time in advance. According to different model prediction times, power generation prediction models with different model prediction times can be constructed.

[0066] Optionally, the model prediction time advance may be a short-term power prediction time advance, such as 10 days.

[0067] Optionally, the model prediction time advance amount may be an ultra-short-term power prediction time advance amount, such as 4 hours.

[0068] In fact, for different model prediction durations, the preset weather system scale analysis strategy will match weather systems of different scales for scale analysis to determine the weather impact geometric area of ​​the target power station.

[0069] For example, if the model prediction time is the first prediction time advance, a scale analysis of the mesoscale weather system is performed on the target power station to determine the first spatial radius of weather factors affecting future power changes of the target power station; the first prediction time advance is greater than or equal to the preset time advance threshold; wherein the first prediction time advance can be a short-term power prediction time advance, such as 10 days; the preset time advance threshold can be one day.

[0070] A first square area is determined with the target power station as the center and the first space radius as the side length; and the first square area is determined as the weather impact geometric area.

[0071] For example, if the model prediction time is the second prediction time advance, a scale analysis of a small-scale weather system is performed on the target power station to determine a second spatial radius of weather factors affecting future power changes of the target power station; wherein the second prediction time advance is less than the first prediction time advance; wherein the second prediction time advance may be a short-term power prediction time advance, such as 4 hours;

[0072] Determine a second square area with the target power station as the center and the second space radius as the side length;

[0073] The second square area is determined as the weather impact geometry area.

[0074] According to different forecast durations, weather systems of different scales are used for scale analysis. This is to take into account the evolution of weather systems corresponding to different forecast durations. For example, if the forecast duration is 4 hours, for the mesoscale weather system, the data corresponding to the evolution of the weather system corresponding to 4 hours will change less. From the perspective of forecasting, the forecast duration is 4 hours, but the mesoscale weather system is used for scale analysis. The forecast effect is obviously not good. The characteristic of the small-scale weather system itself is that it can capture weather evolution data in a shorter time. For example, the experience time of a tornado may be only a few hours, which is difficult to reflect in the mesoscale weather system, but it can be clearly reflected in the small-scale weather system. It is obviously more appropriate to use the scale analysis of the small-scale weather system in the forecast model with a forecast duration of 4 hours. Therefore, according to different forecast durations, using weather systems of different scales for scale analysis can improve the accuracy of data analysis and thus improve the efficiency of model construction.

[0075] In addition, when determining the weather impact geometric area, a square area was selected instead of a circular area. This is mainly because the size of the target power station is generally a rectangular or square area. If a circular area is selected as the weather impact geometric area, it is not accurate enough. Selecting a square area corresponding to the size of the target power station can improve the accuracy of data analysis and thus improve the efficiency of model construction.

[0076] Step 130: construct a machine learning dataset based on the meteorological reanalysis data, the station dataset, and the weather impact geometric area;

[0077] Specifically, target meteorological reanalysis data covering the weather-affected geometric area are selected from the meteorological reanalysis data; the target meteorological reanalysis data and the station data set are constructed as a machine learning data set.

[0078] Step 140: determining meteorological characteristic variables affecting the target power station based on the station data set, meteorological observation data and meteorological reanalysis data;

[0079] Specifically, based on multi-source data such as site data sets, meteorological observation data, and meteorological reanalysis data, a variety of typical weather processes that affect the output of the target power station are selected: typical weather processes for photovoltaic power generation can include sunny days, cloudy days, and cloudy weather, etc.; typical weather processes for wind power generation can include typhoons, cold air influences, and calm weather, etc. Analyze the impact of different elements in each typical weather process on the output of the target power station, determine the key meteorological factors and their characteristic quantities that affect the target power station, that is, determine the meteorological characteristic variables that affect the target power station.

[0080] Specifically, key meteorological factors may include wind speed, wind direction, wind power density, radiation, temperature, humidity and other factors.

[0081] For example, if the target power station is a wind farm, then based on the station data set of the wind farm, the corresponding meteorological observation data and the corresponding meteorological reanalysis data, a variety of typical weather processes that affect wind power output are selected, and the influence of wind speed, wind direction, wind power density, wind shear, temperature, humidity, air pressure and other factors on wind power output in each typical weather process is analyzed to determine the key meteorological factors and their characteristic quantities that affect wind power generation.

[0082] For example, if the target power station is a photovoltaic power station, then based on the site data set of the photovoltaic power station, the corresponding meteorological observation data and the corresponding meteorological reanalysis data, a variety of typical weather processes that affect photovoltaic power generation output are selected, and the influence of conventional factors such as radiation, temperature, humidity, wind speed, etc. on photovoltaic power generation output under each typical weather process is analyzed to determine the key meteorological factors and their characteristic quantities that affect photovoltaic power generation power.

[0083] Step 150: adopting a preset meteorological characteristic variable time extraction strategy, extracting multiple groups of meteorological characteristic variable training data from the machine learning data set, training a preset power prediction model, and constructing a target power generation prediction model corresponding to the target power station.

[0084] From the above content, it can be seen that the embodiment of the present invention uses a long-term series of high-precision meteorological element reanalysis fields and wind and solar power station power generation data for modeling. Compared with the traditional wind and solar power generation power prediction algorithm based on historical meteorological observation data of a single station, the data in this scheme takes into account the impact of changes in upstream weather systems in the atmospheric flow field on the power generation of downstream power stations. The target power generation prediction model constructed in this way has higher prediction accuracy.

[0085] In addition, the spatial radius of meteorological factors affecting future power forecast changes at the site is determined based on the scale characteristics of the weather system. The size of the impact space is determined based on the impact radius, and the spatial scope covered by the power forecast modeling data is narrowed, which can greatly reduce the time for model training and improve training efficiency.

[0086] In an optional implementation, step 150 may specifically include:

[0087] If the model prediction time is the first prediction time advance, a set of meteorological characteristic variable training data is extracted from the machine learning data set every first preset time period to obtain multiple sets of meteorological characteristic variable training data. The first preset time period is shorter than the first prediction time. For example, the first prediction time advance may be 10 days. For example, the second preset time period may be 3 hours.

[0088] For example, if the target power station is a wind farm, the first prediction time advance corresponds to a short-term power prediction model for wind power generation.

[0089] For example, ConvLSTM is selected as the core algorithm of the wind power prediction model. Based on the machine learning data set corresponding to the short-term power prediction of wind power generation, the characteristic variables corresponding to wind power generation and wind power generation are used as input and output variables for training. The characteristic variables of wind power generation are based on the time corresponding to the wind power generation. One group is selected every 3 hours. Then 8 groups are selected for 24 hours a day. If the prediction time is 10 days, then a total of 80 groups are selected for the ten-day forecast time. It is understandable that if the preset time corresponding to the collection of meteorological data sets is ten years, then a total of 10 training data sets are obtained in one year.

[0090] During training, the root mean square error (RMSE) between the predicted wind power and the actual wind power is used as the convergence factor to obtain the short-term wind power prediction model.

[0091] For example, if the target power station is a photovoltaic power station, the first prediction time advance corresponds to a photovoltaic power generation short-term power prediction model.

[0092] Light GBM-XGBoost based on the pre-training-fine-tuning paradigm is selected as the core algorithm of the photovoltaic power prediction model. Based on the machine learning data set corresponding to the short-term power prediction of photovoltaic power generation, the characteristic variables corresponding to photovoltaic power generation and photovoltaic power generation are used as input and output variables for training. The photovoltaic power generation characteristic variables are based on the time corresponding to the photovoltaic power generation. One group is selected every 3 hours, so a total of 8 groups are selected for 24 hours a day. If the prediction time is 10 days, then a total of 80 groups are selected for the ten-day forecast time. It is understandable that if the preset time corresponding to the collection of meteorological data sets is ten years, then a total of 10 sets of training data are obtained in one year.

[0093] During training, the root mean square error between the photovoltaic power generation prediction rate and the actual photovoltaic power generation is used as the convergence factor to obtain the photovoltaic power generation short-term power prediction model.

[0094] In an optional implementation, step 150 may specifically include:

[0095] If the model prediction time length is the second prediction time advance amount, a set of meteorological characteristic variable training data is extracted from the machine learning data set every second preset time period to obtain multiple sets of meteorological characteristic variable training data; wherein the second preset time period is shorter than the first preset time period. The length of the second preset time period is shorter than the second prediction time advance amount. For example, the second prediction time advance amount may be 4 hours. For example, the second preset time period may be 15 minutes.

[0096] For example, if the target power station is a wind farm, the second prediction time advance corresponds to an ultra-short-term power prediction model for wind power generation.

[0097] For example, ConvLSTM is selected as the core algorithm of the wind power prediction model. Based on the machine learning data set corresponding to the ultra-short-term power prediction of wind power generation, the characteristic variables corresponding to wind power generation and wind power generation are used as input and output variables for training. The wind power generation characteristic variables are based on the time corresponding to the wind power generation. One group is selected every 15 minutes, so 16 groups are selected in 4 hours, and 96 groups are selected in one day. It is understandable that if the preset time corresponding to the collection of meteorological data sets is ten years, then a total of 100 training data sets are obtained in one year.

[0098] During training, the root mean square error (RMSE) between the predicted wind power and the actual wind power is used as the convergence factor to obtain the ultra-short-term wind power prediction model.

[0099] For example, if the target power station is a photovoltaic power station, the second prediction time advance corresponds to an ultra-short-term power prediction model for photovoltaic power generation.

[0100] Light GBM-XGBoost based on the pre-training-fine-tuning paradigm is selected as the core algorithm of the photovoltaic power prediction model. Based on the machine learning data set corresponding to the short-term power prediction of photovoltaic power generation, the characteristic variables corresponding to photovoltaic power generation and photovoltaic power generation are used as input and output variables for training. The photovoltaic power generation characteristic variables are based on the time corresponding to the photovoltaic power generation. One group is selected every 15 minutes, so 16 groups are selected in 4 hours, and 96 groups are selected in one day. It is understandable that if the preset time corresponding to the collection of meteorological data sets is ten years, then a total of 100 training data sets are obtained in one year.

[0101] During training, the root mean square error between the predicted photovoltaic power generation and the actual photovoltaic power generation is used as the convergence factor to obtain the photovoltaic power generation ultra-short-term power prediction model.

[0102] From the above content, it can be seen that the embodiment of the present invention adopts a machine learning feature variable time extraction strategy for different power forecast lead times. For the demand of short-term power generation forecast lead time of, for example, 10 days, a set of feature variables is extracted every, for example, 3 hours; for the demand of ultra-short-term power generation forecast lead time of, for example, 4 hours, a set of feature variables is extracted every, for example, 15 minutes. Through the above feature variable extraction strategy in time, the model can better learn the relevant features, and thus have better power forecast results.

[0103] In an optional implementation, the data in the site data set is collected once every third preset time period; wherein the third preset time period is less than or equal to the second preset time period.

[0104] The embodiment of the present invention establishes a new power generation prediction method based on a long-term series of high-precision meteorological element reanalysis fields, targeting wind power generation and photovoltaic power generation, and facing the short-term and ultra-short-term power prediction needs. Compared with existing statistical methods and artificial intelligence methods, this method takes into account the impact of the evolution of weather systems on power generation and the spatial range of its impact, and is more advanced in technology.

[0105] In a specific embodiment, in combination Figure 2 As shown, the construction process of the wind power generation prediction model with different prediction time advance amounts and the photovoltaic power generation prediction model with different prediction time advance amounts are briefly described.

[0106] Collect various data, including: collect meteorological data, including meteorological observation data and model background field data, assimilate meteorological data to form a meteorological reanalysis field. Collect wind farm data and photovoltaic power station data.

[0107] Based on the above data, the short-term power prediction dataset of the station and the ultra-short-term power prediction dataset of the station are constructed respectively.

[0108] In addition, based on the above data, the characteristic variables of wind power generation and photovoltaic power generation are analyzed and obtained.

[0109] Based on the characteristic variables of photovoltaic power generation and the short-term power prediction data set of the site, a short-term power prediction model for wind power generation is constructed.

[0110] Based on the wind power generation characteristic variables and the ultra-short-term power forecasting data set of the site, an ultra-short-term power forecasting model for wind power generation is constructed.

[0111] Based on the characteristic variables of photovoltaic power generation and the short-term power prediction data set of the site, a short-term power prediction model of photovoltaic power generation is constructed.

[0112] Based on the characteristic variables of wind power generation and the ultra-short-term power prediction data set of the site, an ultra-short-term power prediction model for photovoltaic power generation is constructed.

[0113] An embodiment of the present invention may also provide a power generation prediction method, which uses a target power generation prediction model constructed by the method for constructing a power generation prediction model in any of the above embodiments to predict the power generation of a target station.

[0114] It can be known that the existing method of predicting power generation based on the forecast of meteorological elements in numerical weather forecast models requires large computing resources and long operation time. Generally, for a 10-day forecast covering the Chinese region, it takes about 9,600 core hours to calculate. However, by using the power generation model constructed in the invention for prediction, only an ordinary server is needed to complete daily calculations, and the calculation time is about a few minutes, which can fully meet the service demand of updating the forecast data every 15 minutes in the ultra-short-term power forecast assessment.

[0115] The following is a description of the construction device of the power generation prediction model provided by the present invention. The construction device of the power generation prediction model described below and the construction method of the power generation prediction model described above can be referred to each other. Figure 3 As shown, the device for constructing the power generation prediction model may include:

[0116] The acquisition module 310 is used to acquire a meteorological data set within a preset time period and a site data set of a target power station; the meteorological data set at least includes meteorological observation data;

[0117] The scale analysis module 320 is used to perform scale analysis on the target power station based on the preset model prediction time and using the preset weather system scale analysis strategy to determine the weather impact geometric area of ​​the target power station;

[0118] A first construction module 330 is used to construct a machine learning dataset based on meteorological reanalysis data, station datasets and weather impact geometric areas;

[0119] A determination module 340 is used to determine the meteorological characteristic variables affecting the target power station based on the station data set, the meteorological observation data and the meteorological reanalysis data;

[0120] The second construction module 350 is used to adopt a preset meteorological characteristic variable time extraction strategy, extract multiple groups of meteorological characteristic variable training data from the machine learning data set, train the preset power prediction model, and construct a target power generation prediction model corresponding to the target power station.

[0121] The meteorological data set also includes model background field data to be assimilated;

[0122] Before building a machine learning dataset based on meteorological reanalysis data, station datasets, and weather impact geometric areas, the following are also included:

[0123] Assimilate the meteorological observation data and the model background field data to be assimilated to obtain meteorological reanalysis data;

[0124] The scale analysis module 320 is specifically used to: if the model prediction time is the first prediction time advance, perform a scale analysis of the mesoscale weather system on the target power station to determine the first spatial radius of the weather factor affecting the future power change of the target power station; the first prediction time advance is greater than or equal to a preset time advance threshold;

[0125] Determine a first square area with the target power station as the center and the first space radius as the side length;

[0126] The first square area is determined as the weather impact geometric area.

[0127] The scale analysis module 320 is specifically used to: if the model prediction time is the second prediction time advance, perform a scale analysis of the small-scale weather system on the target power station to determine the second spatial radius of the weather factor affecting the future power change of the target power station; wherein the second prediction time advance is less than the first prediction time advance;

[0128] Determine a second square area with the target power station as the center and the second space radius as the side length;

[0129] The second square area is determined as the weather impact geometry area.

[0130] The first construction module 330 is specifically used to: select target meteorological reanalysis data covering the weather impact geometric area from the meteorological reanalysis data;

[0131] The target meteorological reanalysis data and station datasets are constructed as machine learning datasets.

[0132] The second construction module 350 is specifically used to: if the model prediction time length is the first prediction time advance, extract a set of meteorological characteristic variable training data from the machine learning data set every first preset time period to obtain multiple sets of meteorological characteristic variable training data.

[0133] The second construction module 350 is specifically used to: if the model prediction time length is the second prediction time advance, extract a group of meteorological characteristic variable training data from the machine learning data set every second preset time period to obtain multiple groups of meteorological characteristic variable training data; wherein the second preset time period is smaller than the first preset time period.

[0134] The data in the site data set is collected once every third preset time period; wherein the third preset time period is less than or equal to the second preset time period.

[0135] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0136] Although the present invention has been described in conjunction with specific features and embodiments thereof, it is apparent that various modifications and combinations may be made thereto without departing from the spirit and scope of the present invention. Accordingly, this specification and the accompanying drawings are merely exemplary illustrations of the present invention as defined by the appended claims and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present invention. Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, the present invention is intended to include such modifications and variations if they fall within the scope of the claims of the present invention and their equivalents.

Claims

1. A method for constructing a power generation prediction model, characterized in that: include: Obtaining meteorological data sets within a preset time period and site data sets of the target power station; The meteorological data set at least includes meteorological observation data; Based on a preset model prediction time, a preset weather system scale analysis strategy is used to perform scale analysis on the target power station to determine a weather impact geometric area of ​​the target power station; Constructing a machine learning dataset based on the meteorological reanalysis data, the station dataset and the weather impact geometric area; Determining meteorological characteristic variables affecting the target power station based on the station data set, the meteorological observation data and the meteorological reanalysis data; Adopting a preset meteorological characteristic variable time extraction strategy, extracting multiple groups of meteorological characteristic variable training data from the machine learning data set, training a preset power prediction model, and constructing a target power generation prediction model corresponding to the target power station; The method of performing scale analysis on the target power station based on the preset model prediction time and using a preset weather system scale analysis strategy to determine the weather impact geometric area of ​​the target power station includes: if the model prediction time is a first prediction time advance, performing scale analysis of a mesoscale weather system on the target power station to determine a first spatial radius of weather factors affecting future power changes of the target power station; the first prediction time advance is greater than or equal to a preset time advance threshold; A first square area is determined with the target power station as the center and the first space radius as the side length; and the first square area is determined as the weather impact geometric area.

2. The method for constructing a power generation prediction model according to claim 1, characterized in that: The meteorological data set also includes model background field data to be assimilated; Before constructing the machine learning dataset based on the meteorological reanalysis data, the station dataset and the weather impact geometric area, the method further includes: The meteorological observation data and the model background field data to be assimilated are assimilated to obtain the meteorological reanalysis data.

3. The method for constructing a power generation prediction model according to claim 2, characterized in that: The method of performing scale analysis on the target power station based on the preset model prediction time and using a preset weather system scale analysis strategy to determine the weather impact geometric area of ​​the target power station includes: If the model prediction duration is a second prediction time advance, a scale analysis of a small-scale weather system is performed on the target power station to determine a second spatial radius at which weather factors affect future power changes of the target power station; wherein the second prediction time advance is less than the first prediction time advance; Determine a second square area with the target power station as the center and the second space radius as the side length; The second square area is determined as the weather impact geometric area.

4. The method for constructing a power generation prediction model according to claim 3, characterized in that: The step of constructing a machine learning dataset based on the meteorological reanalysis data, the station dataset and the weather impact geometric area includes: Selecting target meteorological reanalysis data covering the weather impact geometric area from the meteorological reanalysis data; The target meteorological reanalysis data and the site data set are constructed as the machine learning data set.

5. The method for constructing a power generation prediction model according to claim 4, characterized in that: The method of using a preset meteorological characteristic variable time extraction strategy to extract multiple sets of meteorological characteristic variable training data from the machine learning data set includes: If the model prediction duration is the first prediction time advance, a set of meteorological characteristic variable training data is extracted from the machine learning data set every first preset time period to obtain the multiple sets of meteorological characteristic variable training data.

6. The method for constructing a power generation prediction model according to claim 5, characterized in that: The method of using a preset meteorological characteristic variable time extraction strategy to extract multiple sets of meteorological characteristic variable training data from the machine learning data set includes: If the model prediction duration is the second prediction time advance, a group of meteorological characteristic variable training data is extracted from the machine learning data set every second preset time period to obtain the multiple groups of meteorological characteristic variable training data; wherein the second preset time period is smaller than the first preset time period.

7. The method for constructing a power generation prediction model according to claim 6, characterized in that: The data in the site data set is collected once every third preset time period; wherein the third preset time period is less than or equal to the second preset time period.

8. A method for predicting power generation, characterized in that: The target power generation prediction model constructed by the method for constructing a power generation prediction model as described in any one of claims 1 to 7 is used to predict the power generation of the target power station.

9. A device for constructing a power generation prediction model, characterized in that: include: An acquisition module is used to acquire a meteorological data set within a preset time period and a site data set of a target power station; The meteorological data set at least includes meteorological observation data; A scale analysis module, configured to perform scale analysis on the target power station based on a preset model prediction duration and using a preset scale analysis strategy to determine a weather-affected geometric area of ​​the target power station; A first construction module is used to construct a machine learning dataset based on meteorological reanalysis data, the station dataset and the weather impact geometric area; A determination module, configured to determine the meteorological characteristic variables affecting the target power station based on the station data set, the meteorological observation data and the meteorological reanalysis data; The second construction module is used to adopt a preset meteorological characteristic variable time extraction strategy to extract multiple groups of meteorological characteristic variable training data from the machine learning data set, train a preset power prediction model, and construct a target power generation prediction model corresponding to the target power station; The scale analysis module is specifically used to perform scale analysis of the mesoscale weather system on the target power station to determine the first spatial radius of weather factors affecting the future power change of the target power station if the model prediction duration is the first prediction time advance; the first prediction time advance is greater than or equal to the preset time advance threshold; A first square area is determined with the target power station as the center and the first space radius as the side length; and the first square area is determined as the weather impact geometric area.

Images