Wind field forecast data processing method and system based on deep learning

Abstract

The invention discloses a wind field forecast data processing method and system based on deep learning, and belongs to the technical field of weather forecast big data analysis. According to the invention, various meteorological data resources can be fully utilized and can be effectively processed according to the integration indication. A previous observation element embedding vector is obtained through observation element mining, and meteorological observation information features can be deeply extracted through feature pooling and vector mining. Meteorological element derivation and trend deduction are carried out based on past cross-modal space mapping vectors, and the connotation of meteorological information is expanded. The subsequently generated weather forecast reference vector improves the vector accuracy. A wind field forecasting training result is obtained through discriminant analysis, forecasting confidence features are obtained, and the credibility of the forecasting result is improved. And the data assimilation thermal feature set locates the target interaction variable to determine the target wind field forecasting processing network, finally, the wind field forecasting result can be accurately generated according to the current information, and the accuracy, reliability and practicability of wind field forecasting are integrally improved.

Description

Technical Field

[0001] The present application belongs to the technical field of weather forecast big data analysis, and specifically relates to a wind field forecast data processing method and system based on deep learning. Background Art

[0002] In the field of meteorological forecasting, the accuracy and reliability of wind field forecasts have always been key issues. Traditional wind field forecasting methods often have many limitations. On the one hand, the use of meteorological observation information is not sufficient, and only relying on data from a single source or a small number of sources cannot fully and accurately reflect the wind field conditions. For example, relying solely on ground meteorological station data and lacking multi-source data fusion such as satellites and high-altitude detection will miss a lot of information on key meteorological elements. On the other hand, traditional methods lack the ability to deeply mine and integrate when processing meteorological data. There are complex relationships between meteorological elements, and the ability to extract features, derive elements, and deducing trends is insufficient, making it difficult to accurately grasp the internal mechanism of wind field changes. In addition, there is a lack of effective evaluation and adjustment mechanisms in the forecasting process, making it difficult to determine the credibility of the forecast results, and it is impossible to optimize and adjust for different forecast situations. Summary of the invention

[0003] The present application provides a wind field forecast data processing method and system based on deep learning, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] The embodiment of the present application provides a wind farm forecast data processing method based on deep learning, which is applied to a wind farm forecast data processing system. The method includes: obtaining past multi-source meteorological observation information and past data integration indication information; in an initial wind farm forecast processing network, using the past data integration indication information to perform observation element mining on the past multi-source meteorological observation information to obtain past observation element embedding vectors; the observation element mining is used to perform feature pooling and vector mining; and using the past data integration indication information to perform cross-modal spatial mapping on the past observation element embedding vector to obtain a past cross-modal spatial mapping vector, and using the original interaction variable to map the past observation element embedding vector with the past cross-modal spatial The mapping vector performs vector interaction and feature optimization to obtain a meteorological forecast reference vector; the cross-modal spatial mapping is used to derive meteorological elements and deduce meteorological trends; the meteorological forecast reference vector is discriminated and analyzed by the initial wind field forecast processing network to obtain a wind field forecast training result, and the forecast confidence feature of the wind field forecast training result is obtained; a data assimilation thermal feature set is generated by the original interaction variables and the forecast confidence feature, a target interaction variable is located from the data assimilation thermal feature set, and the initial wind field forecast processing network including the target interaction variable is determined as a target wind field forecast processing network; the target wind field forecast processing network is used to generate a wind field forecast result according to current data integration indication information.

[0005] An embodiment of the present application provides a wind farm forecast data processing system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor performs the above method.

[0006] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0007] This application can make full use of a variety of meteorological data resources and process them effectively according to the integration instructions. Past observation element embedding vectors are obtained through observation element mining, and feature pooling and vector mining can deeply extract meteorological observation information features. Meteorological element derivation and trend deduction based on past cross-modal space mapping vectors expand the connotation of meteorological information. The subsequent generated meteorological forecast reference vector improves the accuracy of the vector. Discriminant analysis obtains wind field forecast training results and obtains forecast confidence features, which increases the credibility of the forecast results. The data assimilation thermal feature set locates the target interactive variables to determine the target wind field forecast processing network, which can eventually accurately generate wind field forecast results based on current information, and improve the accuracy, reliability and practicality of wind field forecasts as a whole. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of a wind field forecast data processing method based on deep learning provided in an embodiment of the present application.

[0009] Figure 2 A schematic diagram of the structure of a wind farm forecast data processing system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0011] The terms "first", "second", etc. in this application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually a class, and the number of objects is not limited. For example, the first object can be one or more. In addition, in this application, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0012] Figure 1 A wind farm forecast data processing method based on deep learning is shown, which is applied to a wind farm forecast data processing system. The method includes the following steps 202-208.

[0013] Step 202: Obtain past multi-source meteorological observation information and past data integration instruction information.

[0014] Step 204: In the initial wind field forecast processing network, the past data integration indication information is used to perform observation element mining on the past multi-source meteorological observation information to obtain a past observation element embedding vector; the observation element mining is used to perform feature pooling and vector mining; and, the past data integration indication information is used to perform cross-modal spatial mapping on the past observation element embedding vector to obtain a past cross-modal spatial mapping vector, and the original interaction variables are used to perform vector interaction and feature optimization on the past observation element embedding vector and the past cross-modal spatial mapping vector to obtain a meteorological forecast reference vector; the cross-modal spatial mapping is used to perform meteorological element derivation and meteorological trend deduction.

[0015] Step 206: Perform discriminant analysis on the meteorological forecast reference vector through the initial wind farm forecast processing network to obtain a wind farm forecast training result, and acquire a forecast confidence feature of the wind farm forecast training result.

[0016] Step 208: Generate a data assimilation thermal feature set through the original interaction variables and the forecast confidence features, locate the target interaction variables from the data assimilation thermal feature set, and determine the initial wind field forecast processing network including the target interaction variables as the target wind field forecast processing network; the target wind field forecast processing network is used to generate a wind field forecast result according to the current data integration indication information.

[0017] In this application, the wind field forecast data processing system plays an important role in building a high-resolution mesoscale numerical weather forecast system. First, in step 202, the wind field forecast data processing system obtains past multi-source meteorological observation information and past data integration indication information. Past multi-source meteorological observation information contains meteorological data from multiple sources, such as wind speed, wind direction, temperature, air pressure and other data observed by ground meteorological stations, cloud map information obtained by satellite remote sensing, and atmospheric vertical structure information obtained by high-altitude detection balloons. These data come from a wide range of sources, and the formats and precisions may vary. Past data integration indication information is used to guide the subsequent processing of multi-source meteorological observation information. It can contain some information about data types, data quality identifiers, relationships between different data sources, etc. For example, some specific indication information may indicate that the weight of data from a high-precision meteorological station in wind field forecasts should be higher, or the fusion method of data from different sources under specific weather conditions.

[0018] Then enter step 204, and carry out a series of complex operations in the initial wind field forecast processing network. Use the past data integration indication information to mine the observation elements of the past multi-source meteorological observation information. Taking wind speed data as an example, the feature pooling operation may summarize the data with similar wind speed characteristics at different times and locations. Vector mining is to further explore the hidden relationships behind these wind speed data, such as the potential connection between the trend of wind speed changes over time and the surrounding environmental factors in a specific geographical area, so as to obtain the embedding vector of past observation elements. The observation element embedding vector is a deep feature representation of wind field-related observation elements in multi-source meteorological observation information.

[0019] Then, the past data integration indication information is used to perform cross-modal spatial mapping on the embedding vector of past observation elements. In terms of meteorological element derivation, for example, the original wind speed and wind direction data can be used to derive other meteorological elements related to the wind field, such as wind energy density, through cross-modal spatial mapping. For meteorological trend deduction, taking the monsoon in a certain area as an example, based on the long-term wind speed and wind direction observation data, the development trend of the monsoon can be deduced through cross-modal spatial mapping, whether it is gradually strengthening or weakening, etc., so as to obtain the past cross-modal spatial mapping vector. Then, the original interaction variables are used to perform vector interaction and feature optimization on the embedding vector of past observation elements and the past cross-modal spatial mapping vector. The original interaction variables can promote information exchange between the two. For example, the original interaction variables may adjust the weights of derived elements such as wind energy density in the entire vector according to the wind speed and wind direction data, perform feature optimization, and obtain a meteorological forecast reference vector. This meteorological forecast reference vector integrates the results of deep mining, mapping, interaction and optimization of multi-source meteorological observation information, and is an important basis for subsequent wind field forecasts.

[0020] Then in step 206, the weather forecast reference vector is discriminated by the initial wind field forecast processing network. The discriminant analysis evaluates the weather forecast reference vector according to some pre-set rules and models. For example, for wind field forecast, the discriminant analysis may judge the accuracy of the current forecast based on the information such as wind speed, wind direction and related derivative elements in the weather forecast reference vector, combined with the wind field conditions under similar meteorological conditions in history. The wind field forecast training results are obtained through this discriminant analysis, and this result can include the forecast conditions of wind fields in different time periods and different regions. At the same time, the forecast confidence feature of the wind field forecast training result is obtained. The forecast confidence feature reflects the reliability of the wind field forecast training result. For example, if in a certain area, according to historical data and current meteorological conditions, the confidence of wind speed forecast is high, while the confidence of wind direction forecast is relatively low, then the forecast confidence feature will clearly reflect this difference.

[0021] Finally, in step 208, a data assimilation thermal feature set is generated through the original interaction variables and the forecast confidence features. The data assimilation thermal feature set is a collection of multiple information. For example, it can include information about the relationship between different meteorological elements in the original interaction variables, and information about the reliability of wind field forecasts in the forecast confidence features. The target interaction variable is located from this data assimilation thermal feature set. The target interaction variable is a very critical part of the entire wind field forecast data processing system, which determines how the initial wind field forecast processing network generates wind field forecast results based on the current data integration indication information. For example, the target interaction variable may be a variable closely related to wind speed and wind direction in a specific area and under specific meteorological conditions. The initial wind field forecast processing network including the target interaction variable is determined as the target wind field forecast processing network. This target wind field forecast processing network can generate accurate wind field forecast results based on the current data integration indication information, such as the current meteorological observation data type, quality identification, etc.

[0022] Let's further explore some specific situations in the initial wind forecast processing network. This network can adopt a specific deep learning neural network architecture, such as some variants of convolutional neural network (CNN) or recurrent neural network (RNN). Taking convolutional neural network as an example, its convolution layer can effectively extract local features in multi-source meteorological observation information. In wind farm forecast, it can extract local features of data such as wind speed and wind direction at different spatial positions. The pooling layer can compress and simplify these features, reducing the amount of data while retaining key information. When mining observation elements, the convolutional neural network can adaptively learn the complex relationship between different meteorological observation elements by continuously adjusting the parameters of the convolution kernel, thereby obtaining a more accurate embedding vector of past observation elements.

[0023] In the process of cross-modal spatial mapping, for example, a method based on an attention mechanism is used. The attention mechanism can focus on the most important parts of meteorological element derivation and meteorological trend deduction. For example, when deriving meteorological elements, for meteorological elements related to wind fields, the attention mechanism will pay more attention to the interaction between key factors such as wind speed, wind direction, and atmospheric vertical structure, so as to more accurately derive elements such as wind energy density. In terms of meteorological trend deduction, the attention mechanism can focus on factors that have a greater impact on wind field trends, such as terrain, land and sea distribution, etc., based on historical data and current meteorological conditions, so as to better deduce the development trend of wind fields.

[0024] In the discriminant analysis phase, a decision tree-based model may be used. The decision tree model classifies and evaluates the weather forecast reference vector through a series of conditional judgments. For example, taking wind speed forecast as an example, the root node of the decision tree may be the air pressure value under the current meteorological conditions. The data is divided into different subsets according to the size of the air pressure value, and then further judgments are made in each subset based on other meteorological factors such as temperature and humidity, and finally the accuracy evaluation of the wind field forecast training results and the forecast confidence characteristics are obtained.

[0025] The generation of the data assimilation thermal feature set can also involve a method based on statistical analysis. For example, statistical analysis is performed on the data in the original interaction variables and the forecast confidence features, and statistical indicators such as correlation and variance between them are calculated. The data assimilation thermal feature set is constructed based on these statistical indicators so that this feature set can comprehensively reflect various information in the wind farm forecast. When locating the target interaction variable, a method based on cluster analysis can be used. The variables in the data assimilation thermal feature set are clustered according to their similarities, and then the variables that are most relevant to the wind farm forecast and contribute most to improving the forecast accuracy are found in each cluster as the target interaction variable.

[0026] In practical applications, the wind field forecast in a coastal area is taken as an example. The past multi-source meteorological observation information includes wind speed, wind direction, temperature, humidity and other data from multiple meteorological stations along the coast, as well as sea surface temperature, cloud map and other information obtained by satellite remote sensing. The past data integration indication information shows that in this coastal area, the sea surface temperature has a greater impact on the wind field, and the data from high-precision meteorological stations has a higher weight. In the initial wind field forecast processing network, the wind speed data of different meteorological stations are feature pooled and vector mined according to their geographical distribution and time series through observation element mining to obtain the past observation element embedding vector. In the cross-modal spatial mapping, according to the sea surface temperature and wind speed and wind direction data, the method based on the attention mechanism is used to derive meteorological elements such as the distribution of marine wind energy resources, and the development trend of sea breeze is deduced to obtain the past cross-modal spatial mapping vector. The original interaction variables are used to interact and optimize the features of these two vectors to obtain the meteorological forecast reference vector. The wind field forecast training results and forecast confidence features are obtained through discriminant analysis based on decision trees. For example, the forecast results show that the wind speed of the sea breeze in a certain period of time ranges from 5 to 10 meters per second, and the wind direction is southeast, while the forecast confidence characteristics indicate that the confidence of the wind speed forecast is 80%, and the confidence of the wind direction forecast is 70%. Then, a data assimilation thermal feature set is generated based on a statistical analysis method, and then the target interactive variables are located based on cluster analysis, such as variables closely related to sea surface temperature and sea breeze speed, and finally the target wind field forecast processing network is determined, thereby generating more accurate wind field forecast results, providing accurate wind field forecast support for wind energy development, maritime shipping and other activities in the coastal area.

[0027] It can be seen that through the above steps, the wind field forecast data processing system makes full use of past multi-source meteorological observation information and data integration indication information, and with the help of various technical means in the initial wind field forecast processing network, it finally determines the target wind field forecast processing network to generate wind field forecast results, providing a more accurate and effective wind field forecast solution for the field of meteorological forecasting.

[0028] It should be noted that in the wind farm forecast data processing system, the past observation element embedding vector, the past cross-modal space mapping vector, the meteorological forecast reference vector, the original interaction variable and the target interaction variable are the key elements for constructing an accurate wind farm forecast.

[0029] 1. Past Observation Elements Embedding Vector

[0030] The embedding vector of past observation elements is a vector representation obtained after in-depth mining of past multi-source meteorological observation information. Past multi-source meteorological observation information comes from a wide range of sources, including ground meteorological stations, satellites, high-altitude detection equipment, etc., and contains multiple meteorological elements such as wind speed, wind direction, temperature, and air pressure.

[0031] Take a certain area as an example, where there are multiple ground weather stations. At a specific moment, the meteorological elements measured by each weather station are as follows: wind speed ranges between 2-15 m / s, wind direction covers 0°-360°, temperature is between -10°C and 30°C, and air pressure is between 950-1050 hPa.

[0032] When mining observational elements, the feature pooling operation groups wind speed data according to geographical regions. For example, the wind speeds of weather stations in valleys are generally low, such as weather stations A, B, and C, with wind speeds of 2-5 m / s, and they are grouped together; the wind speeds of weather stations D, E, and F in open plains are relatively high, at 8-15 m / s, and are grouped together.

[0033] Vector mining further explores the relationship behind the grouped data. For example, in the low wind speed group (weather stations A, B, C), when the temperature is -5℃-0℃ and the air pressure is 1030-1050 hPa, the wind speed is more likely to be maintained at 2-3 m / s, which can be expressed as a vector form [2-3, -5-0, 1030-1050], where the first component represents the wind speed range, the second component represents the temperature range, and the third component represents the air pressure range; when the temperature rises to 0℃-5℃ and the air pressure drops slightly to 1020-1030 hPa, the wind speed may increase to 3-5 m / s, which can be expressed as [3-5, 0-5, 1020-1030]. There is a similar relationship for the high wind speed group (weather stations D, E, and F). For example, when the temperature is between 15°C and 20°C and the air pressure is between 980 and 1000 hPa, the wind speed is 10-12 m / s, which can be expressed as [10-12, 15-20, 980-1000].

[0034] By integrating the relationship information of various meteorological elements under different conditions, we can get the embedding vector of past observation elements. This vector may contain multiple sub-vectors similar to the above, which comprehensively represents the wind speed characteristics of different meteorological stations under different meteorological conditions and the relationship between various meteorological elements, laying the foundation for subsequent processing.

[0035] 2. Past cross-modal space mapping vector

[0036] The past cross-modal spatial mapping vector is obtained through cross-modal spatial mapping based on the past observation element embedding vector, which is of great significance in the derivation of meteorological elements and the deduction of meteorological trends.

[0037] 1. Derivation of meteorological elements

[0038] Taking the wind field and temperature field in a certain area as an example, the average wind speed data embedded in the vector of past observation elements is 8 meters per second, the wind direction is southwest, and the temperature is between 15°C and 20°C. Meteorological elements are derived according to aerodynamic principles and meteorological laws. It is known that the air density is, for example, 1.2 kilograms per cubic meter. According to the kinetic energy formula, the kinetic energy value when the wind speed is 8 meters per second is 0.5×1.2×8^2=38.4. This derived kinetic energy value is related to the original observation elements such as wind speed and temperature. The numerical representation in the past cross-modal space mapping vector can be [8, 15-20, 38.4], where the first component represents the wind speed, the second component represents the temperature range, and the third component represents the kinetic energy value.

[0039] (II) Weather Trend Deduction

[0040] Consider the meteorological data sequence of a coastal area over the past month. At the beginning of the month, the wind speed is 3-5 m / s, the wind direction is northerly, and the temperature is about 10°C; at the end of the month, the wind speed reaches 8-10 m / s, the wind direction is southerly, and the temperature rises to 15°C-20°C. The embedding vector of the past observation elements contains the wind speed, wind direction, and temperature information at these different times.

[0041] Through cross-modal spatial mapping, meteorological trends are deduced and analysis shows that when the temperature rises, the temperature difference between land and sea increases, leading to stronger sea breezes and changes in wind direction. For example, if the temperature is predicted to continue to rise in the future, according to past trend relationships, the wind speed may continue to increase to 10-12 m / s, and the wind direction will further shift southward to closer to the south. In the past cross-modal spatial mapping vector, this trend information can be expressed as [3-5-8-10-10-12, north-south-more south, 10-15-20], where the first component represents the trend of wind speed, the second component represents the trend of wind direction, and the third component represents the trend of temperature.

[0042] Past cross-modal spatial mapping vectors not only include new elements derived from original meteorological elements, but also cover meteorological trend forecast information. They are more comprehensive and in-depth vector representations of meteorological phenomena, providing more basis for meteorological forecasts.

[0043] 3. Weather Forecast Reference Vector

[0044] The weather forecast reference vector is obtained by vector interaction and feature optimization of the original interaction variables based on the past observation element embedding vectors and the past cross-modal space mapping vectors.

[0045] The original interaction variables play a key role in connecting and adjusting in this process. For example, in wind field forecast, the weight of wind speed is 0.4, the weight of wind direction is 0.3, the weight of temperature is 0.2, and the weight of air pressure is 0.1.

[0046] Taking a certain area as an example, the average wind speed data in the past observation element embedding vector is 6 meters per second, and the wind speed kinetic energy data in the past cross-modal space mapping vector is 30 (for example, according to the specific value calculated previously). The original interaction variable adjusts the interaction between the two according to the weight. Assuming that there is a quantitative relationship between wind speed and wind speed kinetic energy, if the wind speed kinetic energy is proportional to the wind speed, the wind speed related value after the original interaction variable is adjusted is 6×0.4+30×0.4=14.4.

[0047] In terms of feature optimization, if an abnormal wind speed value, such as 18 m / s (significantly deviating from the normal range), appears at a certain weather station in the past observation element embedding vector due to instrument failure, the original interaction variable can be corrected based on other normal weather station data and the meteorological trend information in the past cross-modal spatial mapping vector. For example, based on the wind speed and meteorological trend of surrounding weather stations, the normal wind speed in the area should be 4-6 m / s, and the original interaction variable corrects this abnormal value to 6 m / s (or reduces its influence in the vector).

[0048] After such vector interaction and feature optimization, the meteorological forecast reference vector combines the advantages of past observation element embedding vectors and past cross-modal space mapping vectors. For example, the meteorological forecast reference vector may be expressed numerically as [14.4 (adjusted wind speed related value), a certain direction (weighted wind direction related value), a certain temperature value (adjusted temperature related value), a certain air pressure value (adjusted air pressure related value)], which more accurately reflects the relationship between meteorological elements and the nature of meteorological phenomena, and provides a reliable basis for wind field forecast discrimination analysis.

[0049] 4. Original Interaction Variables

[0050] The original interaction variables are an important part of the wind farm forecast data processing system, connecting and adjusting the relationships between different vectors and elements during the forecast process.

[0051] 1. Relationship between wind speed and wind direction based on geographical regions

[0052] The values ​​and definitions of the original interaction variables are based on a deep understanding of the meteorological system and a large amount of statistical analysis of data. For example, in a specific geographical area such as a mountainous area, wind direction has a significant impact on wind speed. When the wind direction is perpendicular to the direction of the mountain range, the wind speed may increase significantly; when it is parallel to the direction of the mountain range, the wind speed decreases.

[0053] For example, in a mountainous area, when the wind direction is perpendicular to the direction of the mountain, the influence coefficient of wind direction on wind speed is set to 1.5; when it is parallel to the direction of the mountain, the influence coefficient is 0.5. If the angle between the wind direction and the direction of the mountain is 30° at a certain moment (for example, the relationship between the angle and the influence coefficient can be quantified), the influence coefficient of wind direction on wind speed at this time is calculated to be about 1.3 based on the trigonometric function relationship. This 1.3 is the value in the original interaction variable that reflects the special relationship between wind direction and wind speed.

[0054] 2. Interaction variables related to spatiotemporal scales

[0055] The original interaction variables are related to the temporal and spatial scales of meteorological elements. For short-term (hourly) wind forecasts, in urban environments, building height and density affect wind speed and direction. For example, there are tall buildings in cities, with a height of 50-300 meters and a density of 10-50 buildings per square kilometer. The original interaction variables set the values ​​of the interaction variables related to wind speed and wind direction based on these values.

[0056] For example, when the building height is 200 meters and the density is 30 buildings per square kilometer, after analysis and statistics, it is found that the attenuation coefficient of wind speed near the ground is 0.6 (this 0.6 is the value contained in the original interaction variable), which accurately reflects the impact of the urban environment on the wind field.

[0057] 3. Original interaction variables in the interaction between different meteorological elements

[0058] For the interaction between the wind field and the temperature field, the original interaction variables include values ​​related to heat exchange, such as the thermal conductivity coefficient. In coastal areas, the sea water temperature is 20°C, the air temperature on land is 25°C, and the wind blows from the land to the ocean. If the thermal conductivity coefficient is large (for example, 0.8, indicating faster heat exchange), the temperature of the wind changes faster as it blows toward the ocean, which in turn affects changes in wind speed and direction. The original interaction variables accurately adjust the interactions between the various elements in the wind field forecast by including the relevant values ​​of these different meteorological elements.

[0059] 5. Target Interaction Variables

[0060] The target interactive variables are located from the data assimilation thermal characteristics set, which is crucial to determine the target wind field forecast processing network.

[0061] The data assimilation thermal feature set is generated by the original interaction variables and the forecast confidence features, including the relationship information of meteorological elements in the original interaction variables and the forecast reliability information in the forecast confidence features. For example, when the confidence level of wind speed forecast in a certain area is 80%, the relationship values ​​of wind speed, wind direction, temperature, air pressure and other meteorological elements are included in it, which are the manifestations of the original interaction variables under a specific forecast confidence level.

[0062] Taking the wind field forecast of a large wind farm as an example, the stability of the wind field and the accuracy of wind speed in this area are crucial to the power generation efficiency. There may be multiple variables related to the wind field in the data assimilation thermal feature set, such as the relationship between wind speed and terrain, the relationship between wind speed and atmospheric stability, and the relationship between wind direction and season.

[0063] The power plant is located in the channel between the valley and the mountain. The terrain has a significant impact on the wind farm. The terrain slope is between 5°-15°, and the wind speed enhancement coefficient is 1.2-1.5 (based on actual measurements and statistics). Through analysis, it is found that the terrain factor is the most critical. Variables related to wind speed and terrain, such as the wind speed enhancement coefficient of 1.3 when the terrain slope is 10°, will be positioned as target interaction variables.

[0064] Once the target interaction variable is determined, the initial wind farm forecast processing network containing the variable is determined as the target wind farm forecast processing network. This network can integrate the indicative information (such as meteorological observation data, data quality information, etc.) based on the current data, focusing on the relationship between meteorological elements represented by the target interaction variable, thereby generating more accurate wind farm forecast results. For example, in the wind farm forecast of a wind farm, the target wind farm forecast processing network can more accurately forecast wind speed and wind direction based on the terrain enhancement factor of 1.3 for wind speed and current meteorological conditions, providing reliable wind farm information for wind farm operation and management, and ensuring maximum power generation efficiency.

[0065] It can be seen that the past observation element embedding vector, the past cross-modal space mapping vector, the meteorological forecast reference vector, the original interaction variable and the target interaction variable are interrelated and interact with each other in the wind farm forecast data processing system, and jointly construct a complete wind farm forecast technical solution, providing a more accurate and effective wind farm forecast method for the meteorological forecast field. This technical solution can make full use of various meteorological observation data and information, deeply explore the complex relationship between meteorological elements, and improve the accuracy and reliability of wind farm forecasts through continuous adjustment and optimization, so as to meet the needs of wind farm forecasts in different fields (such as wind power generation, aerospace, navigation, etc.).

[0066] In a replaceable technical idea, the acquisition of past multi-source meteorological observation information and past data integration indication information includes: obtaining an initial wind farm forecast processing network, obtaining target forecast training annotations and a deep learning network debugging set corresponding to the initial wind farm forecast processing network; the initial wind farm forecast processing network is a deep learning network that meets pre-debugging conditions; obtaining past multi-source meteorological observation information containing observation elements corresponding to the target forecast training annotations from the deep learning network debugging set, and generating past data integration indication information according to the target forecast training annotations.

[0067] This alternative technical approach involves a specific method of obtaining past multi-source meteorological observation information and past data integration indication information.

[0068] First, obtain the initial wind farm forecast processing network. This initial wind farm forecast processing network is a deep learning network that has reached the pre-debugging conditions. The deep learning network has powerful data analysis and processing capabilities. In wind farm forecasting, it can deeply mine and analyze complex meteorological data. For example, in an actual wind farm forecast scenario, this deep learning network may have undergone a certain degree of pre-training, and its internal neuron structure and parameters have been preliminarily adjusted to a state where it can perform basic processing of meteorological data.

[0069] Next, obtain the target forecast training annotations and the deep learning network debugging set corresponding to the initial wind farm forecast processing network. The target forecast training annotations are very critical data, which provides a clear goal orientation for the entire wind farm forecast processing. For example, the target forecast training annotations may contain the ideal forecast results of the wind farm in a specific area, such as the accurate numerical range of wind speed and the precise direction of wind direction in a specific geographical area over a period of time. These annotated data provide a reference for subsequent processing. The deep learning network debugging set is a collection of various data, which can be used to debug and optimize the deep learning network.

[0070] Then, the past multi-source meteorological observation information containing the observation elements corresponding to the target forecast training annotations is obtained from the deep learning network debugging set. For example, if the target forecast training annotations focus on the two observation elements of wind speed and wind direction in a certain area, then the past multi-source meteorological observation information containing the relevant data of these two elements is screened out from the deep learning network debugging set. Past multi-source meteorological observation information comes from a wide range of sources, including ground meteorological stations, satellites, and high-altitude detection equipment. For example, the wind speed measured by the ground meteorological station at a certain moment is between 3-10 meters per second, and the wind direction is between 0°-180°. These values ​​are all part of the past multi-source meteorological observation information. Satellites may provide meteorological element information in a more macroscopic area, and high-altitude detection equipment can provide information related to the vertical structure of the atmosphere. These data from different sources together constitute a wealth of past multi-source meteorological observation information.

[0071] Finally, the past data integration instruction information is generated according to the target forecast training annotation. Taking wind speed forecast as an example, if the target forecast training annotation has high requirements for wind speed accuracy, such as requiring the error to be within ±1 m / s, then the past data integration instruction information will provide integration instructions for data from different sources based on this requirement. For ground meteorological station data with higher accuracy, a higher weight may be given, while satellite data may need further processing or weight adjustment to meet the requirements of the target forecast training annotation for wind speed forecast accuracy. If the target forecast training annotation indicates that wind direction is more critical to wind field forecast under certain specific meteorological conditions (such as air pressure between 980-1000 hPa), then the past data integration instruction information will emphasize the integration method of wind direction related data, such as special feature extraction of wind direction data during data processing or specific correlation analysis with other meteorological elements.

[0072] Applying the above-mentioned embodiments, firstly, through clear target forecast training annotations, accurate target guidance is provided for wind farm forecasting, making the entire forecasting process more targeted. For example, in the forecast of wind speed and wind direction, the data can be accurately processed according to the annotation requirements to improve the accuracy of the forecast. Secondly, by acquiring past multi-source meteorological observation information from the deep learning network debugging set, the existing data resources can be fully utilized. These data contain rich meteorological information. By screening the observation elements related to the target forecast training annotations, the interference of irrelevant data can be avoided. Finally, according to the target forecast training annotations, the past data integration indication information is generated, and the data from different sources can be reasonably integrated according to different forecast requirements, and the different data can be reasonably weighted and the processing method can be selected, thereby improving the performance of the entire wind farm forecasting system and providing more reliable wind farm forecast results for meteorological forecasting.

[0073] Based on the above technical ideas, the past multi-source meteorological observation information containing the observation elements corresponding to the target forecast training annotations is obtained from the deep learning network debugging set, including any one of the following methods.

[0074] Method 1: Identify past forecast training annotations of the multi-source observation data examples included in the deep learning network debugging set, obtain the annotation feature commonality value between the past forecast training annotations of the multi-source observation data examples and the target forecast training annotations, and determine the multi-source observation data examples corresponding to the past forecast training annotations whose annotation feature commonality value is not less than the forecast annotation commonality threshold as past multi-source meteorological observation information.

[0075] Method 2: Determine the selected wind field targeted by the wind field forecast event based on the target forecast training annotation, parse the selected wind field from the multi-source observation data examples included in the deep learning network debugging set, and determine the multi-source observation data examples of the selected wind field obtained by parsing as the past multi-source meteorological observation information.

[0076] In the wind farm forecast data processing system, based on the technical ideas mentioned above, there are two methods, method 1 and method 2, for obtaining past multi-source meteorological observation information containing observation elements corresponding to the target forecast training annotations from the deep learning network debugging set. These two methods will be explained in detail below.

[0077] Let's first look at method 1. The deep learning network debugging set contains many multi-source observation data examples, and each multi-source observation data example has its corresponding past forecast training annotation. Identifying the past forecast training annotations of these multi-source observation data examples is the first step in this method. For example, in the deep learning network debugging set, there is a multi-source observation data example, whose past forecast training annotations include information such as the wind speed in a specific area ranges from 5 to 10 meters per second, the wind direction is in several specific angle ranges, and the temperature is between 15 and 20 degrees Celsius. The target forecast training annotation may include a more precise range of wind speed in the area, such as 6 to 8 meters per second, a more detailed angle range of wind direction, and a similar temperature range.

[0078] Then, the annotation feature commonality value between the past forecast training annotations and the target forecast training annotations of the multi-source observation data example is obtained. The determination of this annotation feature commonality value requires a detailed comparative analysis of each meteorological element. Taking wind speed as an example, if the wind speed range in the past forecast training annotations of the multi-source observation data example has a large overlap with the wind speed range in the target forecast training annotations, then the wind speed element has a higher annotation feature commonality value. For example, the wind speed range of the past forecast training annotations of the multi-source observation data example is 5-10 meters per second, and the wind speed range of the target forecast training annotations is 6-8 meters per second, and the two have a certain overlap. Similar comparative analysis is also performed for other meteorological elements such as wind direction and temperature, and finally the annotation feature commonality value is comprehensively obtained.

[0079] For example, the forecast annotation commonality threshold is set to 0.6 (this value is determined based on the system's requirements for data and past experience), and the multi-source observation data examples corresponding to the past forecast training annotations whose annotation feature commonality values ​​are not less than this forecast annotation commonality threshold are determined as past multi-source meteorological observation information. If a multi-source observation data example calculates the commonality value of the target forecast training annotations in meteorological elements such as wind speed, wind direction, and temperature to be 0.7, which is greater than the forecast annotation commonality threshold of 0.6, then this multi-source observation data example will be determined as past multi-source meteorological observation information. This means that the meteorological observation data in this multi-source observation data example has a high correlation with the target forecast training annotations, and can provide valuable data for wind farm forecasts.

[0080] Let's look at method 2. Determine the selected wind field for the wind field forecast event based on the target forecast training annotation. For example, the target forecast training annotation clearly states that the wind field in a coastal area is forecasted, and this coastal area is the selected wind field. This selected wind field has its own unique geographical characteristics, meteorological environment and other factors.

[0081] The selected wind farm is parsed from the multi-source observation data examples included in the deep learning network debugging set. The multi-source observation data examples in the deep learning network debugging set include meteorological observation data from different data sources, such as data from ground meteorological stations, satellites, etc. In the parsing process, it is necessary to find data related to the selected wind farm from these large amounts of data. For example, from the data of the ground meteorological station, the wind speed, wind direction, temperature, air pressure and other data observed by the meteorological stations located in the coastal area and within a certain range around it are screened out. From the satellite data, the cloud map information covering the coastal area, the atmospheric temperature distribution and other data related to the wind farm are extracted. The multi-source observation data examples of the selected wind farm obtained by parsing are determined as past multi-source meteorological observation information. These determined past multi-source meteorological observation information are closely related to the selected wind farm targeted by the target forecast training annotation, and can provide an accurate data basis for wind farm forecasting.

[0082] Furthermore, by comparing the commonality values ​​of the annotation features with the commonality thresholds of the forecast annotations, method 1 can accurately screen out examples of multi-source observation data that are highly relevant to the target forecast training annotations as past multi-source meteorological observation information. For example, by determining the commonality values ​​through detailed comparison of meteorological elements such as wind speed and wind direction, the mixing of irrelevant data is avoided, thereby improving the pertinence and effectiveness of the data. Method 2 determines the past multi-source meteorological observation information based on the analysis of the selected wind field, and can focus on a specific wind field area and extract data related to the wind field from a large number of multi-source observation data examples. For example, in the forecast of wind fields in coastal areas, the meteorological observation data of coastal areas can be accurately obtained, thereby providing accurate and targeted data support for wind field forecasts and improving the accuracy and reliability of wind field forecasts.

[0083] Based on the above technical ideas, the obtaining of past multi-source meteorological observation information containing observation elements corresponding to the target forecast training annotations from the deep learning network debugging set also includes any one of the following methods.

[0084] Method 3: Obtaining an original meteorological observation information example containing the observation elements corresponding to the target forecast training annotations from the deep learning network debugging set, and determining the original meteorological observation information example as past multi-source meteorological observation information.

[0085] Method 4: Obtaining an original meteorological observation information example containing the observation elements corresponding to the target forecast training annotations from the deep learning network debugging set, preprocessing the original meteorological observation information example, and obtaining the past multi-source meteorological observation information.

[0086] Method 5: Obtain an original meteorological observation information example containing the observation elements corresponding to the target forecast training annotations from the deep learning network debugging set, obtain the observation status label of the original meteorological observation information example, and determine the past multi-source meteorological observation information from the original meteorological observation information example based on the observation status label.

[0087] In the technical framework of the wind farm forecast data processing system, based on the technical ideas described above, there are different methods such as method 3, method 4 and method 5 for obtaining past multi-source meteorological observation information containing observation elements corresponding to the target forecast training annotations from the deep learning network debugging set, which will be explained in detail below.

[0088] Method 3 is relatively direct. It obtains examples of original meteorological observation information containing observation elements corresponding to the target forecast training annotations from the deep learning network debugging set, and then determines them as past multi-source meteorological observation information. The deep learning network debugging set contains a wealth of original meteorological observation information examples, which cover observation data of various meteorological elements. For example, the observation elements corresponding to the target forecast training annotations are wind speed and wind direction in a certain area. In the deep learning network debugging set, there are some original meteorological observation information examples, which clearly record the wind speed value range (such as 3-10 m / s) and the specific angle range of wind direction (such as 0°-180°) in the area. Directly determine these original meteorological observation information examples containing the observation elements of interest in the target forecast training annotations as past multi-source meteorological observation information. This method simply and directly obtains data related to the target forecast training annotations, providing a basic data source for subsequent wind field forecast processing.

[0089] Method 4 adds a preprocessing step based on Method 3. First, the original meteorological observation information examples containing the observation elements corresponding to the target forecast training annotations are also obtained from the deep learning network debugging set. For example, the wind speed data of a certain area in the original meteorological observation information example has certain fluctuations, the wind speed value is between 2-12 meters per second, and the wind direction angle range is relatively wide, from 0° to 360°, and also contains data on other meteorological elements such as temperature and air pressure.

[0090] Then the original meteorological observation information example is preprocessed. The purpose of preprocessing is to improve the data quality and make it more suitable for wind farm forecasting. For example, for wind speed data, there may be some outliers caused by the accuracy of the measuring equipment or environmental interference. If the wind speed recorded at a certain moment is 15 meters per second, which is significantly higher than the normal wind speed range in the area, it can be corrected by preprocessing means (such as data smoothing, outlier removal, etc.). For example, after processing, the abnormal 15 meters per second wind speed is corrected to 8 meters per second, which is in line with the normal range of the area. For wind direction data, if the records are messy, they can be sorted out by certain algorithms (such as angle averaging algorithms, etc.) to make the wind direction angle range more accurate. After such preprocessing, the past multi-source meteorological observation information obtained has been improved in terms of data accuracy and availability, which is more conducive to the subsequent wind farm forecast processing process.

[0091] Method 5 also obtains the original meteorological observation information examples containing the observation elements corresponding to the target forecast training annotations from the deep learning network debugging set, but then obtains the observation status labels of the original meteorological observation information examples, and determines the past multi-source meteorological observation information from the original meteorological observation information examples based on the observation status labels.

[0092] After obtaining the original meteorological observation information examples, each example is accompanied by an observation status label. These observation status labels reflect some characteristics of the original meteorological observation information examples. For example, the observation status label may contain information such as the quality level of the data (such as high quality, medium quality, and low quality), the reliability of the data source (such as data from high-precision weather stations marked as high reliability, and data from some simple observation equipment marked as low reliability), and the meteorological environment status at the time of observation (such as normal meteorological environment, observation under complex meteorological environment).

[0093] Taking the wind speed and wind direction observations in a certain area as an example, the wind speed data in the original meteorological observation information example is between 4-10 m / s, the wind direction is between 90°-270°, and its observation status label shows that the data source is a medium-reliability meteorological station, and the data is observed in a complex meteorological environment, and the quality level is medium quality. Based on these observation status labels, if the wind farm forecasting system has high requirements for data quality, the original meteorological observation information example may be screened according to preset rules (such as giving priority to high-reliability sources, data in normal meteorological environments and high quality). If the original meteorological observation information example does not meet the requirements, it may not be determined as past multi-source meteorological observation information; if there is another original meteorological observation information example, its wind speed is between 3-9 m / s, the wind direction is between 120°-240°, and the observation status label shows that the data source is a high-reliability meteorological station, observed in a normal meteorological environment, and the quality level is high quality, then this original meteorological observation information example is more likely to be determined as past multi-source meteorological observation information. Through this screening method based on observation status labels, it is possible to ensure that the determined past multi-source meteorological observation information has high quality and reliability, thereby improving the accuracy of wind field forecasts.

[0094] In combination with the above, method 3 directly obtains the original meteorological observation information examples as past multi-source meteorological observation information, which simply and efficiently provides the data basis related to the target forecast training annotation. Method 4 can correct the outliers in the data, improve the accuracy and availability of the data by preprocessing the original meteorological observation information examples, so that the data based on the subsequent wind farm forecast is more reliable. Method 5 determines the past multi-source meteorological observation information from the original meteorological observation information examples based on the observation status label, and can screen out higher quality and more reliable data to avoid the impact of low-quality or unreliable data on wind farm forecasts. Combining these three methods, it is possible to obtain higher-quality, more accurate, and more targeted past multi-source meteorological observation information from the deep learning network debugging set, providing solid data support for wind farm forecasts and improving the accuracy and reliability of wind farm forecasts.

[0095] In a preferred example, the past observation element embedding vector includes an observation element dynamic framework vector, a physical process parameterization vector and a mesoscale weather pattern vector. Based on this, in the initial wind field forecast processing network, the observation element mining of the past multi-source meteorological observation information is performed using the past data integration indication information to obtain the past observation element embedding vector, including: in the initial wind field forecast processing network, the multi-source meteorological observation element vector corresponding to the past multi-source meteorological observation information is subjected to dynamic framework feature mapping to obtain the observation element dynamic framework vector; the dynamic framework feature mapping is used to strengthen the observation elements under the dynamic framework state; the observation element dynamic framework vector is subjected to feature quantization to obtain the physical process parameterization vector; the physical process parameterization vector is used to represent the meteorological physical process state corresponding to the past multi-source meteorological observation information; the past integration indication vector corresponding to the past data integration indication information is obtained, and the physical process parameterization vector is weightedly integrated with the past integration indication vector to obtain the mesoscale weather pattern vector; the mesoscale weather pattern vector is used to represent the meteorological joint description features of the past multi-source meteorological observation information and the past data integration indication information.

[0096] Based on the above example, the method also includes: performing a convolution operation on the past multi-source meteorological observation information to obtain a past meteorological observation convolution vector; generating a past observation offset feature based on the vector size of the past meteorological observation convolution vector; and performing weighted integration of the past meteorological observation convolution vector and the past observation offset feature to obtain the multi-source meteorological observation element vector.

[0097] In the wind forecast data processing system, the technical solution of this preferred example revolves around the construction of past observation element embedding vectors and related operations, and this process includes multiple complex and interrelated links. First, for the construction of past observation element embedding vectors, it includes three important parts: observation element dynamic framework vector, physical process parameterization vector and mesoscale weather pattern vector. The initial step of this construction process involves the processing of past multi-source meteorological observation information.

[0098] In the past, multi-source meteorological observation information contained a rich variety of meteorological data, which came from different observation sources, such as numerous ground meteorological stations, satellite observation equipment, and high-altitude detection instruments. The data provided by these data sources cover multiple meteorological elements such as wind speed, wind direction, temperature, air pressure, humidity, etc., and each meteorological element has different numerical records at different observation points and times.

[0099] In the initial wind field forecast processing network, to obtain the embedding vector of past observation elements, it is necessary to first perform convolution operation on the past multi-source meteorological observation information to obtain the convolution vector of past meteorological observations. Convolution operation is a technical means widely used in data processing and is also of great significance in meteorological data processing. Take the meteorological observation data of a certain area as an example. For example, there are 10 ground meteorological stations distributed in the area, and each meteorological station observes the four meteorological elements of wind speed, wind direction, temperature, and air pressure once every hour. At a certain moment, the wind speed values ​​observed by these meteorological stations may fluctuate between 0.5-15 m / s, the wind direction covers all possible directions from 0° to 360°, the temperature varies between -10℃ and 30℃, and the air pressure is between 950-1050 hPa. When performing convolution operation on these multi-source meteorological observation information, the convolution kernel will slide on the data according to certain rules, thereby extracting the local features of the data and obtaining the convolution vector of past meteorological observations. This vector incorporates local relationships between different weather stations and different meteorological elements, for example it may highlight a special correlation between wind speed and temperature in a small area.

[0100] Based on the vector size of the past meteorological observation convolution vector, the past observation offset feature is generated. The vector size of the past meteorological observation convolution vector reflects the structural characteristics of the data after the convolution operation. For example, if the size of the past meteorological observation convolution vector is [20, 10] (the numerical value here represents the size of the vector in two dimensions), this size information will be used to generate the past observation offset feature. The generated past observation offset feature is related to the vector size, and it may adjust the data in the convolution vector according to a specific algorithm to better adapt to subsequent processing. This adjustment is designed based on a deep understanding of the structure and characteristics of meteorological data, and aims to dig out deeper relationships in meteorological data.

[0101] Then, the convolution vector of past meteorological observations and the offset features of past observations are weightedly integrated to obtain the multi-source meteorological observation element vector. Weighted integration is a method that comprehensively considers the information of two vectors. For example, the value of an element in the convolution vector of past meteorological observations is 5 (this value may represent the value of the wind speed of a certain meteorological station after the convolution operation), and the corresponding element value in the offset features of past observations is 3. If the weighting coefficients are set to 0.6 and 0.4 respectively (these weighting coefficients are determined according to the importance of the system to different information), then the corresponding element value in the multi-source meteorological observation element vector obtained after weighted integration is 5×0.6+3×0.4=4.2. In this way, the multi-source meteorological observation element vector integrates the information of the convolution vector of past meteorological observations and the offset features of past observations, providing a basis for the subsequent construction of the dynamic framework vector of the observation element.

[0102] Next, the dynamic framework feature mapping is performed on the multi-source meteorological observation element vector to obtain the observation element dynamic framework vector. The dynamic framework is the basic framework for describing the movement and change of the atmosphere in meteorology. The dynamic framework feature mapping is to map the data in the multi-source meteorological observation element vector according to the requirements of the dynamic framework. For example, in the dynamic framework of the atmosphere, there is a complex dynamic relationship between wind speed and wind direction, which is affected by multiple factors such as terrain and pressure gradient. In the multi-source meteorological observation element vector, the values ​​of wind speed and wind direction may be simple observation values, such as wind speed of 8 m / s and wind direction of 120°. Through the dynamic framework feature mapping, the wind speed and wind direction are adjusted taking into account the terrain characteristics of the area (for example, whether it is a plain area or a mountainous area) and the pressure gradient (for example, the pressure gradually decreases from east to west). After the dynamic framework feature mapping, the wind speed in the observation element dynamic framework vector may become 9 m / s and the wind direction may become 125°. This adjusted value can more accurately reflect the true relationship between wind speed and wind direction in the dynamic framework state, and strengthen the observation elements in the dynamic framework state.

[0103] Then, the dynamic framework vector of the observation element is characterized and quantified to obtain the parameterized vector of the physical process. The meteorological physical process is very complex, involving heat exchange, water vapor transport, energy conversion and other aspects. The purpose of the parameterized vector of the physical process is to express the state of the meteorological physical process corresponding to the past multi-source meteorological observation information in a quantitative form. For example, in the dynamic framework vector of the observation element, in addition to wind speed and wind direction, it also contains a temperature of 18°C ​​and an air pressure of 1010 hPa. In the physical process, there is an interactive relationship between temperature and air pressure and wind speed and wind direction. Through characteristic quantification, the values ​​of these meteorological elements are converted into a specific quantitative representation. For example, the value of wind speed after characteristic quantification is 10 (this value is obtained based on the wind speed of 9 m / s and a specific quantization rule), the wind direction is 15 (the quantized value corresponding to 125°), the temperature is 15 (the quantized value corresponding to 18°C), and the air pressure is 12 (the quantized value corresponding to 1010 hPa). These quantized values ​​constitute the parameterized vector of the physical process, which can accurately express the relationship between the various elements in the meteorological physical process.

[0104] Then obtain the past integration indicator vector corresponding to the past data integration indicator information, and perform weighted integration of the physical process parameterization vector and the past integration indicator vector to obtain the mesoscale weather pattern vector. The past data integration indicator information contains various indicator information for integrating past multi-source meteorological observation information, such as the weight allocation of data from different data sources, the importance identification of different meteorological elements in the integration process, etc. The past integration indicator vector represents these indicator information in the form of a vector. For example, the weights of wind speed, wind direction, temperature, and air pressure in the past integration indicator vector are set to 0.3, 0.2, 0.3, and 0.2, respectively (these weight values ​​are determined based on the needs of meteorological forecasts and the evaluation of the importance of different meteorological elements). The quantified values ​​of wind speed in the physical process parameterization vector are 10, wind direction is 15, temperature is 15, and air pressure is 12. When performing weighted integration, the corresponding component values ​​in the mesoscale weather pattern vector are calculated according to the weights, for example, the wind speed component is 10×0.3=3, the wind direction component is 15×0.2=3, the temperature component is 15×0.3=4.5, and the pressure component is 12×0.2=2.4. The mesoscale weather pattern vector is used to represent the meteorological joint description characteristics of the past multi-source meteorological observation information and the past data integration indication information. It integrates the meteorological observation data and the integrated indication information, providing a more comprehensive and accurate feature description for the wind field forecast.

[0105] Therefore, the entire technical solution constructs the embedding vector of past observation elements through a series of rigorous operations, providing a comprehensive and accurate foundation for wind farm forecasting. Starting from the initial convolution operation, the local features of the original meteorological observation data are gradually integrated, so that the data can better reflect the intrinsic relationship between meteorological elements in subsequent processing. The dynamic framework feature mapping strengthens the relationship between meteorological elements under the dynamic framework state, so that the relationship between elements such as wind speed and wind direction is more in line with the actual situation of atmospheric movement, and improves the accuracy of the description of meteorological dynamic processes. The physical process parameterization vector obtained by feature quantization can accurately represent the state of meteorological physical processes, which helps to deeply understand the interaction mechanism between meteorological elements. Finally, the mesoscale weather pattern vector obtained by weighted integration combines observation information with integrated indication information. This joint description feature provides more comprehensive and detailed information for wind farm forecasting, which helps to improve the accuracy and reliability of wind farm forecasting, thereby better meeting the needs of wind farm forecasting in related fields such as meteorological forecasting.

[0106] In an optional embodiment, the past data integration indication information is used to perform cross-modal spatial mapping on the past observation element embedding vector to obtain a past cross-modal spatial mapping vector, and the original interaction variables are used to perform vector interaction and feature optimization on the past observation element embedding vector and the past cross-modal spatial mapping vector to obtain a weather forecast reference vector, including:

[0107] (1) When u is a positive integer greater than 1 and not greater than X, in the u-th spatial mapping kernel of the initial wind farm forecast processing network, the past data integration indication information is used to perform cross-modal spatial mapping on the u-1th meteorological forecast reference vector to obtain the u-th past cross-modal spatial mapping vector, and the X-u+1th past observation element embedding vector generated by the X-u+1th element embedding kernel is obtained. The u-th original interaction variable in the u-th spatial mapping kernel is used to perform vector interaction on the X-u+1th past observation element embedding vector and the u-th past cross-modal spatial mapping vector to obtain the u-th past interaction vector. If u is X, the X-th past interaction vector is feature optimized to obtain the meteorological forecast reference vector; X is the number of spatial mapping kernels included in the initial wind farm forecast processing network;

[0108] (2) When u is 1, in the first spatial mapping kernel of the initial wind field forecast processing network, the past data integration indication information is used to perform cross-modal spatial mapping on the Xth past observation element embedding vector to obtain the first past cross-modal spatial mapping vector, and the first original interaction variable is used to perform vector interaction on the Xth past observation element embedding vector and the first past cross-modal spatial mapping vector to obtain the first past interaction vector.

[0109] In this optional embodiment, the technical solution revolves around processing the vector using past data integration indication information and original interaction variables to obtain a weather forecast reference vector. First of all, it is necessary to clarify that the entire processing process is carried out under the framework of the initial wind field forecast processing network. This network contains X spatial mapping cores, and the value of X is a key parameter for determining the number of spatial mapping cores. Each spatial mapping core has a unique and indispensable role in the entire process.

[0110] When u is a positive integer greater than 1 and not greater than X, for example, X is 10, when u=3, a series of operations are started in the third spatial mapping core of the initial wind field forecast processing network.

[0111] In this spatial mapping kernel, the second meteorological forecast reference vector must first be cross-modal spatially mapped using the past data integration indication information. The past data integration indication information is a rich information set that covers various integration-related instruction information for meteorological data. For example, for the meteorological element of wind speed, the past data integration indication information may specify the integration weights of wind speed data from different observation sources (such as ground meteorological stations, satellite observations, etc.) in a specific geographical area or under specific meteorological conditions. Similar provisions are also available for other meteorological elements such as wind direction, temperature, and air pressure. When performing cross-modal spatial mapping, the information in the second meteorological forecast reference vector will be transformed based on these indication information. For example, the second meteorological forecast reference vector contains some features of wind speed, which are expressed as wind speeds in a certain area in the range of 5-8 m / s and wind directions in a specific angle range. The third past cross-modal spatial mapping vector obtained after cross-modal spatial mapping may establish a new relationship between this wind speed range and other meteorological elements in different modes (such as different spatial dimensions or modes related to different meteorological physical processes). For example, the wind speed may be linked to the temperature gradient in the vertical direction of the atmosphere to form a new vector representation, in which there is a specific correspondence between the wind speed range and the temperature gradient value, which is mined during the cross-modal space mapping process.

[0112] At the same time, the X-u+1th past observation element embedding vector generated by the X-u+1th element embedding kernel is obtained. When X=10, u=3, the 8th past observation element embedding vector generated by the 8th element embedding kernel is obtained. The past observation element embedding vector is the result of the element embedding kernel processing, which contains the feature information of the meteorological observation data after deep mining. Taking wind speed as an example, the wind speed-related features in this past observation element embedding vector may not only be a simple numerical range, but may also include the change trend information of wind speed in different time series, as well as the correlation information between wind speed and other meteorological elements (such as temperature, air pressure, etc.) observed by surrounding meteorological stations. For example, in this past observation element embedding vector, the change of wind speed at certain specific times shows a trend of increasing first and then decreasing, and this trend is related to the temperature change of surrounding meteorological stations. This correlation is represented in the vector through a specific numerical value or encoding method.

[0113] Then, the u-th original interaction variable in the u-th spatial mapping kernel is used to perform vector interaction between the embedding vector of the X-u+1th past observation element and the u-th past cross-modal spatial mapping vector. The original interaction variables play a crucial bridging and adjustment role in this process. The original interaction variables are based on a deep understanding of the meteorological system and a large amount of statistical analysis of meteorological data. For example, for the two meteorological elements of wind speed and wind direction, the third original interaction variable in the third spatial mapping kernel may specify the interaction coefficient between wind speed and wind direction under a specific meteorological environment (such as a specific pressure range and temperature range). For example, in this meteorological environment, the air pressure is between 980-1000 hPa, the temperature is between 15-20℃, and the interaction coefficient between wind speed and wind direction is specified as 0.6 (this value is obtained based on a large amount of statistical analysis of historical data, indicating that the influence of wind direction on wind speed under such meteorological conditions is 0.6). When performing vector interaction, if the wind speed-related eigenvalue in the eighth past observation element embedding vector is a certain value, and the wind direction-related eigenvalue is a specific angle, and the eigenvalues ​​related to wind speed and wind direction in the third past cross-modal space mapping vector are transformed, the original interaction variable will perform weighted summation or other forms of interaction operations on the wind speed and wind direction-related features in the two vectors according to this interaction coefficient, thereby obtaining the third past interaction vector. This past interaction vector integrates the information of the two vectors and accurately reflects the relationship between meteorological elements through the original interaction variables, so that the relationship between meteorological elements is more accurately reflected in the vector.

[0114] When u is X, taking X=10 as an example, when u=10, the 10th past interaction vector is feature optimized to obtain the weather forecast reference vector. In this feature optimization process, the 10th past interaction vector is comprehensively checked and adjusted. For example, the 10th past interaction vector may contain feature representations of multiple meteorological elements, and there may be some situations between the feature values ​​of elements such as wind speed, wind direction, temperature, and air pressure that do not conform to the basic principles of meteorology or the statistical laws of historical data. For example, in this vector, the wind speed in a certain area corresponding to the wind speed feature value is very high, but at the same time, the air pressure feature value does not change significantly, which is contrary to the general relationship between air pressure and wind speed in meteorology. In the feature optimization process, these feature values ​​will be adjusted according to meteorological principles and historical data statistical laws. The normal value range of air pressure under similar wind speed conditions in history may be referred to to adjust the air pressure feature value to a more reasonable value. At the same time, similar checks and adjustments will be made to the relationships between other meteorological elements, such as whether the relationship between wind direction and temperature is reasonable. Through such a feature optimization process, the final weather forecast reference vector can more accurately reflect the real relationship between meteorological elements and is more suitable for weather forecasting.

[0115] When u is 1, in the first spatial mapping kernel of the initial wind field forecast processing network, the past data integration indication information is used to perform cross-modal spatial mapping on the Xth past observation element embedding vector. When X = 10, the 10th past observation element embedding vector is subjected to cross-modal spatial mapping to obtain the first past cross-modal spatial mapping vector. This process is also based on the past data integration indication information. The 10th past observation element embedding vector contains rich meteorological observation element information. In the cross-modal spatial mapping process, the past data integration indication information will guide the conversion of this information to different modes and mine new relationships between meteorological elements. For example, the wind speed and wind direction observed on the ground may be combined with the atmospheric cloud map information observed by the satellite (established through some conversion and wind speed and wind direction) to obtain the first past cross-modal spatial mapping vector. The information in this vector is no longer limited to the traditional meteorological element values, but contains more cross-modal related information. Then, the first original interaction variable is used to perform vector interaction between the embedding vector of the Xth past observation element and the first past cross-modal spatial mapping vector to obtain the first past interaction vector. The first original interaction variable plays an initial adjustment role in this process. For example, the first original interaction variable may perform preliminary weighting or adjustment operations on the eigenvalues ​​of wind speed and wind direction in the two vectors, so that the first past interaction vector can accurately reflect the interaction relationship between meteorological observation elements in the initial stage, laying the foundation for subsequent operations in other spatial mapping kernels.

[0116] It can be seen that by performing a series of complex operations in a specific order in multiple spatial mapping kernels in the initial wind farm forecast processing network, the information of past observation element embedding vectors and past cross-modal spatial mapping vectors is effectively integrated. The operations in each spatial mapping kernel are closely dependent on the past data integration indication information and the original interaction variables. This design enables the relationship between meteorological elements to be deeply mined, accurately adjusted and optimized. From the operation process of different spatial mapping kernels, when u is a different value, the operation is carried out in sequence, and the relationship between meteorological elements in different modes is gradually integrated and optimized, which can better adapt to the complexity and diversity of meteorological data. The meteorological forecast reference vector is obtained by optimizing the features of the past interaction vector, which improves the accuracy and reliability of this vector, thereby providing a more accurate basis for wind farm forecasting, which helps to improve the accuracy and stability of wind farm forecasting and the ability to cope with complex meteorological conditions.

[0117] Under a possible technical idea, the method of obtaining the forecast confidence feature of the wind farm forecast training result includes: obtaining the relationship influence coefficient between the wind farm forecast training result and the past data integration indication information, and obtaining the forecast inversion label of the wind farm forecast training result; using a forecast opinion evaluation network to perform a forecast opinion evaluation on the wind farm forecast training result to obtain the forecast accuracy feature of the wind farm forecast training result; combining the relationship influence coefficient, the forecast inversion label and the forecast accuracy feature to obtain the forecast confidence feature of the wind farm forecast training result.

[0118] In addition, the method further includes: obtaining a quantitative discrimination interval, obtaining an expected index for network debugging, obtaining a global discrimination score from the quantitative discrimination interval according to the expected index for network debugging; determining a target discrimination score according to the global discrimination score, and converting the target discrimination score into an inversion matching score and a forecast accuracy score. Then, the forecast confidence feature of the wind farm forecast training result is obtained by combining the relationship influence coefficient, the forecast inversion label, and the forecast accuracy feature, including: using the global discrimination score, the inversion matching score, and the forecast accuracy score, combining the relationship influence coefficient, the forecast inversion label, and the forecast accuracy feature, to obtain the forecast confidence feature of the wind farm forecast training result.

[0119] In the wind farm forecast data processing system, the above technical ideas involve obtaining the forecast confidence features of the wind farm forecast training results and the operations related thereto. First, the process of obtaining the forecast confidence features of the wind farm forecast training results includes multiple key operations. First, the relationship influence coefficient between the wind farm forecast training results and the past data integration indication information is obtained. The past data integration indication information contains various instructions and guiding information for the integration of past multi-source meteorological observation information. The wind farm forecast training results are the results of wind farm forecast obtained after a series of processing (such as the various vector operations mentioned above in the initial wind farm forecast processing network). The relationship influence coefficient reflects the degree of correlation between the two. For example, the wind speed forecast part of the wind farm forecast training results may be closely related to the weight distribution of the wind speed data source in the past data integration indication information, the integration method of wind speed data in different regions, etc. If a high-precision wind speed observation source is given a higher weight in the past data integration indication information, and the wind speed forecast in the wind field forecast training results is highly consistent with the data of the high-precision observation source, then the relationship influence coefficient may be relatively high, such as 0.8.

[0120] Then obtain the forecast inversion label of the wind farm forecast training result. The forecast inversion label is an identifier of the wind farm forecast training result, which may contain a variety of attribute information about the forecast result. For example, the forecast inversion label may indicate that the wind farm forecast training result is obtained under specific meteorological conditions (such as a specific air pressure range, temperature range, etc.), or is a forecast result for a specific area (such as a wind farm in a certain geographical area). For example, the forecast inversion label indicates that the wind farm forecast training result is for a coastal area and is obtained under meteorological conditions of an air pressure of 980-1000 hPa and a temperature of 15-20°C.

[0121] Then, the forecast viewpoint evaluation network is used to evaluate the wind farm forecast training results, and the forecast accuracy characteristics of the wind farm forecast training results are obtained. The forecast viewpoint evaluation network is a network structure specifically used to evaluate the accuracy of wind farm forecasts. It is built based on a large amount of meteorological data, meteorological principles, and historical forecasting experience. When evaluating the forecast viewpoint, it will comprehensively consider the forecast of each meteorological element in the wind farm forecast training results. For example, for wind speed forecast, if the wind farm forecast training results predict that the wind speed in a certain area is between 5-8 meters per second, and the actual meteorological observation data (which can be accurate data in the verification data set) shows that the wind speed in the area is between 6-7 meters per second, then the forecast accuracy feature will give a corresponding evaluation result based on this difference. For example, if this evaluation result is in the range of 0-1, according to the accuracy of the above wind speed forecast, the forecast accuracy feature that may be obtained is 0.8 (indicating that the forecast is relatively accurate).

[0122] Then, the forecast confidence feature of the wind farm forecast training result is obtained by combining the relationship influence coefficient, forecast inversion label and forecast accuracy feature. This process comprehensively considers the impact of multiple factors on the forecast confidence. The relationship influence coefficient reflects the close correlation between the wind farm forecast training result and the past data integration indication information, the forecast inversion label provides background information about the forecast result, and the forecast accuracy feature directly reflects the accuracy of the forecast. For example, if the relationship influence coefficient is 0.8, the forecast inversion label indicates that the forecast result is obtained under relatively stable meteorological conditions (in this case, the forecast reliability is usually higher), and the forecast accuracy feature is 0.8. After combining these factors, the forecast confidence feature obtained may be a comprehensive value, such as 0.85 (the calculation here is an illustration after comprehensive consideration of various factors, not a simple weighted average or other formula calculation).

[0123] In addition, this technical solution also involves other related operations. First, obtain the quantitative discrimination interval. The quantitative discrimination interval is a numerical range interval for discriminating and evaluating the wind farm forecast results. For example, this interval may be set to [0, 1], where 0 indicates that the forecast is completely inaccurate and 1 indicates that the forecast is completely accurate. At the same time, obtain the network debugging expected index. The network debugging expected index is an index determined based on the target requirements and debugging experience of the wind farm forecast system. Obtain a global discrimination score from the quantitative discrimination interval based on the network debugging expected index. For example, the network debugging expected index requires a higher forecast accuracy. According to this index, the global discrimination score that may be obtained from the quantitative discrimination interval of [0, 1] is 0.7.

[0124] The target discrimination score is determined based on the global discrimination score. This target discrimination score is a more targeted score further determined on the basis of the global discrimination score. For example, according to certain specific rules in the wind farm forecast system or the requirements of specific meteorological elements, the global discrimination score of 0.7 is further adjusted to a target discrimination score of 0.65. The target discrimination score is then converted into an inversion matching score and a forecast accuracy score. This conversion process is based on specific conversion rules, which are formulated based on meteorological principles and the characteristics of the wind farm forecast system. For example, a target discrimination score of 0.65 may be converted into an inversion matching score of 0.3 and a forecast accuracy score of 0.35 according to the conversion rules (the specific conversion rules will vary depending on the complexity of the system).

[0125] Finally, when combining the relationship influence coefficient, forecast inversion label and forecast accuracy characteristics to obtain the forecast confidence characteristics of the wind farm forecast training results, the global discriminant score, the inversion matching score and the forecast accuracy score are used. For example, when comprehensively considering the aforementioned relationship influence coefficient of 0.8, forecast inversion label (indicating a forecast under stable meteorological conditions), forecast accuracy feature of 0.8, global discriminant score of 0.7, inversion matching score of 0.3 and forecast accuracy score of 0.35, the final forecast confidence characteristic is obtained through a complex comprehensive mechanism (this mechanism is based on a comprehensive understanding of the wind farm forecast system and a large amount of data verification, not a simple linear combination), for example, this final forecast confidence characteristic is 0.82.

[0126] It can be seen that by obtaining the relationship influence coefficient between the wind farm forecast training results and the past data integration indication information, the correlation between the forecast results and the data integration indication information can be considered to ensure the scientific nature of the forecast. The forecast inversion label provides background information for the forecast results, which helps to comprehensively evaluate the reliability of the forecast. The forecast accuracy characteristics obtained by using the forecast viewpoint evaluation network directly reflect the accuracy of the forecast. Combining these factors with the global discriminant score, inversion matching score and forecast accuracy score to obtain the forecast confidence feature can comprehensively and accurately measure the confidence of the wind farm forecast from multiple angles, thereby improving the credibility and availability of the wind farm forecast results and providing a more reliable basis for meteorological forecasting decisions.

[0127] On the basis of steps 202 to 208, the method further includes: obtaining a first feature enhancement variable interval and a second feature enhancement variable interval, arbitrarily generating Y first original enhancement variables within the first feature enhancement variable interval, and arbitrarily generating Y second original enhancement variables within the second feature enhancement variable interval; Y is a positive integer, the first original enhancement variable is used to feature enhance the past observation element embedding vector, and the second original enhancement variable is used to feature enhance the past cross-modal space mapping vector; using the Y first original enhancement variables and the Y second original enhancement variables, Y original interaction variables are obtained; each original interaction variable includes a first original enhancement variable and a second original enhancement variable.

[0128] In the present application, based on the technical framework constructed by steps 202-208, this embodiment further introduces a new operation to obtain the original interaction variable. This process includes multiple key links and is closely related to each other. First, the first feature enhancement variable interval and the second feature enhancement variable interval are obtained. These two intervals are key elements for determining the value range of the enhancement variable. Their determination is based on the requirements of the wind field forecast data processing system for the feature enhancement of the past observation element embedding vector and the past cross-modal space mapping vector. The first feature enhancement variable interval is a value range set to generate variables for feature enhancement of the past observation element embedding vector, and the second feature enhancement variable interval is a feature enhancement variable for the past cross-modal space mapping vector. For example, the first feature enhancement variable interval may be set to [1, 10]. The setting of this interval takes into account the characteristic value range of the meteorological elements in the past observation element embedding vector and the degree of enhancement required in subsequent processing. For the second feature enhancement variable interval, for example, it is set to [5, 15], which is determined based on the characteristics and enhancement requirements of the past cross-modal space mapping vector.

[0129] Next, Y first original reinforcement variables are arbitrarily generated in the first feature reinforcement variable interval, and Y second original reinforcement variables are arbitrarily generated in the second feature reinforcement variable interval, where Y is a positive integer. Taking Y=3 as an example, three first original reinforcement variables are generated in the first feature reinforcement variable interval [1, 10], for example, the values ​​of these three variables are 3, 6, and 9 respectively. The generation of these first original reinforcement variables is random, but within the specified interval, they will be used to perform feature reinforcement on the embedding vectors of past observation elements. For the second feature reinforcement variable interval [5, 15], three second original reinforcement variables are also generated, for example, their values ​​are 7, 10, and 13. These second original reinforcement variables will act on the feature reinforcement of the past cross-modal space mapping vectors.

[0130] Then, Y first original enhanced variables and Y second original enhanced variables are used to obtain Y original interaction variables, each of which includes a first original enhanced variable and a second original enhanced variable. For example, the first original interaction variable is composed of the first first original enhanced variable (value is 3) and the first second original enhanced variable (value is 7). This combination method is the core of constructing the original interaction variable, which links the variables that enhance the features of the past observation element embedding vector and the variables that enhance the features of the past cross-modal space mapping vector. The original interaction variables constructed in this way can comprehensively consider the feature enhancement requirements of the two vectors, and play an important role in the subsequent wind field forecast processing.

[0131] In the wind field forecast data processing system, the embedding vector of past observation elements contains the result of deep processing of rich meteorological observation element information, such as the relationship between meteorological elements such as wind speed, wind direction, temperature, and air pressure, as well as their characteristic representations under different conditions. The first original enhancement variable enhances its characteristics, which can adjust the representation of the relationship between these meteorological elements or highlight certain key features. For example, in the embedding vector of past observation elements, there is a certain relationship between wind speed and wind direction. When the first original enhancement variable acts on the vector, it may change the weight of this relationship representation or enhance its correlation, so that the true relationship between wind speed and wind direction can be more accurately reflected in the subsequent forecast processing.

[0132] The past cross-modal spatial mapping vector contains the relationship information of meteorological elements in different modes (such as different spatial dimensions or modes related to different meteorological physical processes). The second original reinforcement variable can further explore and highlight the key information in these cross-modal relationships by enhancing its features. For example, the past cross-modal spatial mapping vector contains the cross-modal relationship between the wind speed in the vertical direction of the atmosphere and the temperature gradient in the horizontal direction. The second original reinforcement variable may strengthen certain features of this relationship, so that this cross-modal relationship information can be more accurately transmitted when interacting with the past observation element embedding vector.

[0133] The original interaction variable combines these two reinforcement variables with different functions, which is of great significance in the entire wind farm forecast processing flow. In subsequent vector interaction, feature optimization and other operations, it can accurately adjust the interaction relationship between the past observation element embedding vector and the past cross-modal space mapping vector according to the characteristics of the first original reinforcement variable and the second original reinforcement variable in the combination, thereby affecting the generation of the meteorological forecast reference vector and ultimately affecting the wind farm forecast results. For example, in the process of vector interaction, if the first original reinforcement variable in the original interaction variable specifically strengthens the wind speed characteristics in the past observation element embedding vector, and the second original reinforcement variable also strengthens the wind speed-related cross-modal relationship in the past cross-modal space mapping vector, then in the interaction process, the two pieces of information about wind speed can be more accurately integrated, thereby improving the accuracy of wind speed forecasting by the entire wind farm forecast system.

[0134] In this design, by obtaining the first original reinforcement variable and the second original reinforcement variable within a specific interval to construct the original interaction variable, the past observation element embedding vector and the past cross-modal space mapping vector can be more accurately enhanced. This method takes into account the characteristics and reinforcement requirements of the two vectors, and combines the reinforcement variables of the two into the original interaction variable, so that in the wind field forecast processing, the information of the two vectors can be better integrated and the interaction relationship between them can be accurately adjusted. Thereby improving the accuracy of the generation of meteorological forecast reference vectors, further improving the accuracy and reliability of wind field forecast results, and providing more effective data support for meteorological forecasts.

[0135] In an expandable embodiment, the number of the original interaction variables is Y, the number of the past data integration indication information is Z, Y is a positive integer, and Z is a positive integer; the number of the wind farm forecast training results is Y×Z; the data assimilation thermal feature set is generated by the original interaction variables and the forecast confidence features, and the target interaction variables are located from the data assimilation thermal feature set, including: based on the record information of the forecast confidence features corresponding to the Z past data integration indication information under each original interaction variable, determining the initial feature enhancement thermal values ​​corresponding to the Y original interaction variables respectively; combining the Y original interaction variables and the initial feature enhancement thermal values ​​corresponding to the Y original interaction variables to form Y feature enhancement binary group; each feature enhancement binary group includes an original interaction variable and the initial feature enhancement thermal value corresponding to the original interaction variable; a data assimilation thermal feature set is generated according to the Y feature enhancement binary groups; if the target feature enhancement thermal value in the data assimilation thermal feature set meets the stability requirement, the interaction variable corresponding to the target feature enhancement thermal value in the data assimilation thermal feature set is determined as the target interaction variable; if the target feature enhancement thermal value in the data assimilation thermal feature set does not meet the stability requirement, the interaction adjustment variable associated with the target feature enhancement thermal value is obtained from the data assimilation thermal feature set, and the forecast confidence characteristics of the Z past data integration indication information under the interaction adjustment variable are obtained.

[0136] Based on this scalable embodiment, the present application revolves around the complex relationship between original interaction variables, past data integration indication information, wind field forecast training results, data assimilation thermal feature sets, and target interaction variables.

[0137] First, the number of original interaction variables is defined as Y (Y is a positive integer), the number of past data integration indication information is defined as Z (Z is a positive integer), and the number of wind farm forecast training results is defined as Y × Z. This quantitative relationship is based on the logical architecture of the entire wind farm forecast data processing system, reflecting the inherent connection between different elements.

[0138] Next, the process of generating a data assimilation thermal feature set through the original interaction variables and the forecast confidence features and locating the target interaction variables therefrom is described. Based on the record information of the forecast confidence features corresponding to the Z past data integration indication information under each original interaction variable, the initial feature enhancement thermal values ​​corresponding to the Y original interaction variables are determined. For example, Y = 3, Z = 4, for the first original interaction variable, the record information of the forecast confidence features obtained under the four past data integration indication information may be some specific numerical values ​​or state identifiers. Based on these record information, the initial feature enhancement thermal value corresponding to the first original interaction variable is determined according to a specific algorithm or logical relationship, for example, 10. Similarly, for the other two original interaction variables, their initial feature enhancement thermal values ​​are determined according to the forecast confidence feature record information under their corresponding past data integration indication information, for example, 12 and 8 respectively.

[0139] The Y original interaction variables and the initial feature enhancement thermal values ​​corresponding to the Y original interaction variables are combined into Y feature enhancement tuples. Each feature enhancement tuple includes an original interaction variable and the initial feature enhancement thermal value corresponding to the original interaction variable. Taking the example just now, the first feature enhancement tuple is (original interaction variable 1, 10), the second is (original interaction variable 2, 12), and the third is (original interaction variable 3, 8). Then, a data assimilation thermal feature set is generated based on these Y feature enhancement tuples. This data assimilation thermal feature set is a collection of original interaction variables and their related initial feature enhancement thermal values. It contains rich information, which reflects the correlation between the original interaction variables and the forecast confidence features, and is an important basis for the subsequent positioning of the target interaction variables.

[0140] Then consider two cases to locate the target interaction variable. If the target feature enhanced thermal value in the data assimilation thermal feature set meets the stability requirement, the interaction variable corresponding to the target feature enhanced thermal value in the data assimilation thermal feature set is determined as the target interaction variable. The stability requirement is a judgment criterion set based on the overall stability and accuracy requirements of the wind farm forecast system. For example, if the target feature enhanced thermal value is 12 (for example, this value is the initial feature enhanced thermal value corresponding to an element in the data assimilation thermal feature set), if the stability requirement is set to be greater than or equal to 10, then since 12 meets this requirement, the corresponding interaction variable is determined as the target interaction variable. This target interaction variable is of special importance in the entire wind farm forecast system, and it will directly affect the subsequent wind farm forecast results.

[0141] If the target feature enhanced thermal value in the data assimilation thermal feature set does not meet the stability requirements, the interactive adjustment variable associated with the target feature enhanced thermal value is obtained from the data assimilation thermal feature set, and the forecast confidence features of Z past data integration indication information under the interactive adjustment variable are obtained. For example, if the target feature enhanced thermal value is 8 (does not meet the stability requirement of, for example, greater than or equal to 10), then the interactive adjustment variable associated with this target feature enhanced thermal value is found from the data assimilation thermal feature set (this association is based on the structure and logical relationship within the data assimilation thermal feature set). Then the forecast confidence features of Z past data integration indication information (for example, Z = 4 in the previous example) under this interactive adjustment variable are obtained. These forecast confidence features will provide more information basis for further adjustment and determination of the target interactive variables, thereby ensuring that the target interactive variables finally determined can meet the accuracy and stability requirements of the wind farm forecast system.

[0142] It can be seen that by determining the initial feature enhancement thermal value based on the forecast confidence characteristics corresponding to the original interaction variables and the past data integration indication information, and then generating the data assimilation thermal feature set, this method comprehensively integrates the key information in the system. When locating the target interaction variable, different operations are performed according to whether the target feature enhancement thermal value meets the stability requirements, which can more accurately screen out the appropriate target interaction variable. When the stability requirements are met, the target interaction variable is directly determined. When it is not met, it is further adjusted by obtaining the associated interaction adjustment variable and its corresponding forecast confidence characteristics. This helps to improve the accuracy of the target interaction variable determination, thereby improving the accuracy and stability of the entire wind farm forecast system, and providing a more reliable basis for wind farm forecasting.

[0143] In summary, this application fundamentally improves the performance of the entire wind field forecasting system by means of the two key technical points of the mesoscale weather model WRF and WRFDA data assimilation. The WRFDA data assimilation system is able to assimilate various types of observation data, and this process is a key link in integrating multi-source meteorological observation information. Like a sophisticated filter and integrator, it screens and fuses meteorological observation data from different sources (such as ground meteorological stations, satellites, etc.), and incorporates data from multiple sources into a unified framework, thereby improving the accuracy of the initial field. This has laid a solid foundation for subsequent numerical simulation calculations based on the WRF model.

[0144] In the wind farm forecast processing network, the collaborative processing of past data integration indication information and multi-source meteorological observation information shows unique advantages. By mining the observation elements of past multi-source meteorological observation information, the embedding vector of past observation elements is obtained. The feature pooling and vector mining operations are like accurately mining treasures in a huge meteorological data treasure house. Feature pooling can summarize data with similar features, and vector mining deeply explores the potential connections between meteorological elements, which helps to better adapt to the advanced dynamic framework and physical process parameterization scheme of the WRF model, thereby improving the accuracy of the description of the wind farm state.

[0145] The past cross-modal spatial mapping vector is obtained by integrating the indication information of past data for cross-modal spatial mapping. The realization of meteorological element derivation and meteorological trend deduction in this process adds rich in-depth information to the wind farm forecast. This is similar to the atmospheric simulation environment constructed by the WRF model, which can not only see the current state of meteorological elements, but also predict the derivative changes of meteorological elements and future development trends based on their physical process parameterization schemes. For example, in combination with the WRF model, key information such as the energy changes of the wind field under different physical processes can be obtained through the derivation of meteorological elements. Meteorological trend deduction can more accurately predict the development trend of the wind field within the mesoscale range, which plays an irreplaceable role in improving the accuracy of wind farm forecasts and the timeliness of meteorological disaster warnings.

[0146] The process of obtaining the meteorological forecast reference vector by vector interaction and feature optimization of the original interaction variables between the past observation element embedding vector and the past cross-modal spatial mapping vector is a further integration of the advantages of the WRF model and WRFDA data assimilation. The original interaction variables are like an intelligent adjustment hub, which reasonably adjusts the vector according to the physical process relationship in the WRF model and the multi-source data characteristics after WRFDA data assimilation. This adjustment ensures that the meteorological forecast reference vector can fully reflect the real state of the wind field, minimize the forecast error caused by complex data sources or changeable meteorological processes, and make the forecast results more consistent with the actual atmospheric conditions simulated by the WRF model.

[0147] The initial wind forecast processing network performs discriminant analysis on the meteorological forecast reference vector to obtain the wind forecast training results and the forecast confidence features. This step is similar to the final quality inspection process based on the WRF model simulation. The forecast confidence features provide a reliable credibility assessment for the wind forecast results, which is crucial in meteorological forecast and disaster warning application scenarios. The reliability of high-resolution mesoscale weather forecast results based on the WRF model output depends on this. Decision makers can use forecast information more scientifically based on this credibility assessment, thereby avoiding making wrong decisions due to unreliable forecasts.

[0148] Finally, the data assimilation thermal feature set is generated through the original interaction variables and the forecast confidence features, and the target interaction variables are located to determine the target wind field forecast processing network to generate the wind field forecast results. This process is closely integrated with the WRF and WRFDA data assimilation systems. In the high-resolution simulation environment constructed by the WRF model, the generation of the data assimilation thermal feature set can make full use of the multi-source data information after the WRFDA data assimilation. The positioning of the target interaction variables is based on the dynamic framework and physical process characteristics in the WRF model to specifically determine the forecast processing network that is most suitable for the current meteorological conditions. This method helps to further improve the accuracy and timeliness of wind field forecasts, and can output more accurate and detailed wind field forecast results within the mesoscale range, thereby providing more powerful meteorological data support for meteorological-related fields such as meteorological forecasts and disaster warnings, and meeting the urgent needs of these fields for high-precision meteorological forecasts.

[0149] Further, Figure 2 Schematic diagram of a wind farm forecast data processing system 200 provided in an embodiment of the present application. Figure 2 The wind farm forecast data processing system 200 shown includes a processor 210, and the processor 210 can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0150] Alternatively, if Figure 2 As shown, the wind farm forecast data processing system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0151] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0152] Alternatively, if Figure 2 As shown, the wind farm forecast data processing system 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.

[0153] Optionally, the wind farm forecast data processing system 200 may implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application, which will not be described here for brevity.

[0154] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities.

[0155] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0156] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0157] It should be noted that, in this article, the terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0158] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), and includes a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0159] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the present application and the scope of protection of the present application, all of which are within the protection of the present application.

Claims

1. A wind field forecast data processing method based on deep learning, characterized in that: The method is applied to a wind farm forecast data processing system, and the method comprises: Obtain past multi-source meteorological observation information and past data integration instruction information; In the initial wind field forecast processing network, the past data integration indication information is used to perform observation element mining on the past multi-source meteorological observation information to obtain a past observation element embedding vector; the observation element mining is used to perform feature pooling and vector mining; and the past data integration indication information is used to perform cross-modal spatial mapping on the past observation element embedding vector to obtain a past cross-modal spatial mapping vector, and the original interaction variables are used to perform vector interaction and feature optimization on the past observation element embedding vector and the past cross-modal spatial mapping vector to obtain a meteorological forecast reference vector; the cross-modal spatial mapping is used to perform meteorological element derivation and meteorological trend deduction; Performing discriminant analysis on the meteorological forecast reference vector through the initial wind field forecast processing network to obtain a wind field forecast training result, and obtaining a forecast confidence feature of the wind field forecast training result; A data assimilation thermal feature set is generated through the original interaction variables and the forecast confidence features, a target interaction variable is located from the data assimilation thermal feature set, and an initial wind field forecast processing network including the target interaction variable is determined as a target wind field forecast processing network; the target wind field forecast processing network is used to generate a wind field forecast result according to current data integration indication information.

2. The method according to claim 1, characterized in that The acquisition of past multi-source meteorological observation information and past data integration instruction information includes: Acquire an initial wind farm forecast processing network, acquire target forecast training annotations, and a deep learning network debugging set corresponding to the initial wind farm forecast processing network; the initial wind farm forecast processing network is a deep learning network that meets pre-debugging conditions; Past multi-source meteorological observation information including the observation elements corresponding to the target forecast training annotations is obtained from the deep learning network debugging set, and past data integration indication information is generated according to the target forecast training annotations.

3. The method according to claim 2, characterized in that The obtaining, from the deep learning network debugging set, the past multi-source meteorological observation information containing the observation elements corresponding to the target forecast training annotations includes: Identify past forecast training annotations of multi-source observation data examples included in the deep learning network debugging set, obtain annotation feature commonality values ​​between the past forecast training annotations of the multi-source observation data examples and the target forecast training annotations, and determine the multi-source observation data examples corresponding to the past forecast training annotations whose annotation feature commonality values ​​are not less than the forecast annotation commonality threshold as past multi-source meteorological observation information; Alternatively, a selected wind field targeted by a wind field forecast event is determined based on the target forecast training annotations, the selected wind field is parsed from the multi-source observation data examples included in the deep learning network debugging set, and the multi-source observation data examples of the selected wind field obtained by parsing are determined as the past multi-source meteorological observation information.

4. The method according to claim 2, characterized in that The obtaining, from the deep learning network debugging set, the past multi-source meteorological observation information containing the observation elements corresponding to the target forecast training annotations includes: Acquire an original meteorological observation information example containing the observation element corresponding to the target forecast training annotation from the deep learning network debugging set, and determine the original meteorological observation information example as past multi-source meteorological observation information; Alternatively, an original meteorological observation information example containing the observation elements corresponding to the target forecast training annotation is obtained from the deep learning network debugging set, and the original meteorological observation information example is preprocessed to obtain the past multi-source meteorological observation information; Alternatively, an original meteorological observation information example containing the observation elements corresponding to the target forecast training annotations is obtained from the deep learning network debugging set, an observation status label of the original meteorological observation information example is obtained, and the past multi-source meteorological observation information is determined from the original meteorological observation information example based on the observation status label.

5. The method according to claim 1, characterized in that The past observation element embedding vector includes the observation element dynamic framework vector, the physical process parameterization vector and the mesoscale weather pattern vector; In the initial wind field forecast processing network, the observation element mining of the past multi-source meteorological observation information is performed using the past data integration indication information to obtain the past observation element embedding vector, including: In the initial wind field forecast processing network, a dynamic framework feature mapping is performed on the multi-source meteorological observation element vector corresponding to the past multi-source meteorological observation information to obtain the observation element dynamic framework vector; the dynamic framework feature mapping is used to strengthen the observation element in the dynamic framework state; The observation element dynamic framework vector is characterized and quantified to obtain the physical process parameterization vector; the physical process parameterization vector is used to represent the meteorological physical process state corresponding to the past multi-source meteorological observation information; Obtain a past integration indication vector corresponding to the past data integration indication information, perform weighted integration on the physical process parameterization vector and the past integration indication vector, and obtain the mesoscale weather pattern vector; the mesoscale weather pattern vector is used to represent the meteorological joint description characteristics of the past multi-source meteorological observation information and the past data integration indication information.

6. The method according to claim 5, characterized in that The method further comprises: Performing a convolution operation on the past multi-source meteorological observation information to obtain a past meteorological observation convolution vector; Generating a past observation offset feature according to the vector size of the past meteorological observation convolution vector; The past meteorological observation convolution vector and the past observation offset feature are weightedly integrated to obtain the multi-source meteorological observation element vector.

7. The method according to claim 1, characterized in that The method of using the past data integration indication information to perform cross-modal spatial mapping on the past observation element embedding vector to obtain a past cross-modal spatial mapping vector, and using the original interaction variables to perform vector interaction and feature optimization on the past observation element embedding vector and the past cross-modal spatial mapping vector to obtain a weather forecast reference vector, includes: When u is a positive integer greater than 1 and not greater than X, in the u-th spatial mapping kernel of the initial wind farm forecast processing network, the past data integration indication information is used to perform cross-modal spatial mapping on the u-1th meteorological forecast reference vector to obtain the u-th past cross-modal spatial mapping vector, and the X-u+1th past observation element embedding vector generated by the X-u+1th element embedding kernel is obtained, and the u-th original interaction variable in the u-th spatial mapping kernel is used to perform vector interaction on the X-u+1th past observation element embedding vector and the u-th past cross-modal spatial mapping vector to obtain the u-th past interaction vector. If u is X, the X-th past interaction vector is feature optimized to obtain a meteorological forecast reference vector; X is the number of spatial mapping kernels included in the initial wind farm forecast processing network; When u is 1, in the first spatial mapping kernel of the initial wind field forecast processing network, the past data integration indication information is used to perform cross-modal spatial mapping on the X-th past observation element embedding vector to obtain the first past cross-modal spatial mapping vector, and the first original interaction variable is used to perform vector interaction between the X-th past observation element embedding vector and the first past cross-modal spatial mapping vector to obtain the first past interaction vector.

8. The method according to claim 1, characterized in that The obtaining of the forecast confidence feature of the wind farm forecast training result comprises: Obtaining a relationship influence coefficient between the wind farm forecast training result and the past data integration indication information, and obtaining a forecast inversion label of the wind farm forecast training result; Using a forecast viewpoint evaluation network to perform forecast viewpoint evaluation on the wind farm forecast training result to obtain forecast accuracy characteristics of the wind farm forecast training result; Combining the relationship influence coefficient, the forecast inversion label and the forecast accuracy feature, obtaining a forecast confidence feature of the wind farm forecast training result; The method further comprises: Obtaining a quantitative discrimination interval, obtaining a network debugging expected index, and obtaining a global discrimination score from the quantitative discrimination interval according to the network debugging expected index; Determining a target discrimination score according to the global discrimination score, and converting the target discrimination score into an inversion matching score and a forecast accuracy score; The combining of the relationship influence coefficient, the forecast inversion label and the forecast accuracy feature to obtain the forecast confidence feature of the wind farm forecast training result includes: The forecast confidence feature of the wind farm forecast training result is obtained by utilizing the global discrimination score, the inversion matching score and the forecast accuracy score, in combination with the relationship influence coefficient, the forecast inversion label and the forecast accuracy feature.

9. The method according to claim 1, characterized in that The method further comprises: Obtain a first feature enhancement variable interval and a second feature enhancement variable interval, arbitrarily generate Y first original enhancement variables in the first feature enhancement variable interval, and arbitrarily generate Y second original enhancement variables in the second feature enhancement variable interval; Y is a positive integer, the first original enhancement variable is used to feature enhance the past observation element embedding vector, and the second original enhancement variable is used to feature enhance the past cross-modal space mapping vector; Using the Y first original reinforcement variables and the Y second original reinforcement variables, Y original interaction variables are obtained; each original interaction variable includes a first original reinforcement variable and a second original reinforcement variable.

10. A wind farm forecast data processing system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 9.

Images