Wind power prediction method and system based on meteorological data

Abstract

The invention discloses a wind power prediction method and system based on meteorological data, and relates to the wind power prediction technology, and the method comprises the steps: training a large model in advance, and training an LSTM model; obtaining regional position description information of a target wind power plant, and constructing a fine tuning training set based on the key fields; performing fine tuning on the trained large model by utilizing the fine tuning training set; obtaining current meteorological prediction data of a target wind power plant coverage area, and outputting wind power plant prediction information of a plurality of distribution positions based on the fine-tuned large model; based on the wind field prediction information, utilizing a trained LSTM model to execute prediction so as to obtain power generation prediction information of a plurality of distribution positions; and calculating a power generation prediction result of the target wind power plant based on the power generation prediction information of the plurality of distribution positions. According to the method, the large model and the LSTM are combined, the meteorological data are trained and predicted through the large model, the LSTM is used for predicting the power generation data of the wind field, and the accuracy of wind power prediction is improved.

Description

Technical Field

[0001] The present application relates to the technical field of wind power prediction, and in particular to a method and system for wind power prediction based on meteorological data. Background Art

[0002] To meet energy needs, wind energy has received extensive attention from researchers, and wind power generation technology is becoming increasingly mature. However, due to the strong randomness and volatility of wind energy, it poses challenges to the safe and stable operation of the power system. High-precision wind power forecasting can not only improve the utilization level of wind energy, but also effectively alleviate the impact of wind power uncertainty on the system.

[0003] With the rapid development of artificial intelligence technology, deep learning models have gradually been introduced into wind power prediction. However, a single model has limitations in wind power prediction and often has preferences in data feature extraction, resulting in incomplete captured wind power data features. Summary of the invention

[0004] The embodiment of the present application provides a wind power prediction method and system based on meteorological data, which combines a large model and LSTM, trains and predicts meteorological data through a large model, and uses LSTM to predict wind farm power generation data, thereby improving the accuracy of wind power prediction.

[0005] The present application embodiment proposes a wind power prediction method based on meteorological data, comprising the following steps: Obtain historical meteorological forecast data and historical actual meteorological data for areas covered by different wind farms in advance, and train a large model. Also, train an LSTM model based on the historical meteorological forecast data, historical actual meteorological data, and power generation data from multiple distribution locations of different wind farms. Acquire the geographical location description information of the target wind farm, and filter out key fields related to wind power generation from the description information to construct a fine-tuning training set based on the key fields; Using the fine-tuning training set, fine-tuning the trained large model; Obtain current weather forecast data for the target wind farm coverage area, and output wind farm forecast information for multiple distribution locations based on the fine-tuned large model; Based on the wind farm prediction information, using the trained LSTM model to perform prediction to obtain power generation prediction information for multiple distribution locations; The power generation prediction result of the target wind farm is calculated based on the power generation prediction information of multiple distribution locations.

[0006] Optionally, historical meteorological forecast data and historical actual meteorological data of different wind farm coverage areas are obtained in advance, and a large model is trained, including: Use crawler technology to obtain historical weather forecast data for different wind farm coverage areas from weather websites; Determining the central area of ​​the wind farm in the historical meteorological forecast data, and calculating the distance deviation between the central area of ​​the wind farm and the areas covered by the different wind farms; The historical meteorological forecast data is marked based on the distance deviation, and the historical meteorological forecast data and the historical actual meteorological data are used as a set of training data at set time intervals to perform full parameter training on the large model.

[0007] Optionally, training the LSTM model based on historical meteorological forecast data, historical actual meteorological data, and power generation data of multiple distribution locations of different wind farms includes: Acquire the terrain relationship of the different wind farms, and determine the positions of all wind turbines in the different wind farms based on the terrain relationship; terrain features that determine the location of each wind turbine; According to the positions of all wind turbines in different wind farms and the terrain features, wind turbines at multiple positions of the required number of wind turbines are selected, wherein the selected wind turbine positions include a position with the largest deviation in the terrain features and multiple positions with dispersed distances from the position with the largest deviation; According to the wind turbines in multiple selected locations, their historical power generation data are obtained to train the LSTM model.

[0008] Optionally, obtaining the geographical location description information of the target wind farm, and filtering out key fields related to wind power generation from the description information to construct a fine-tuning training set based on the key fields includes: Obtaining description information related to the geographical location of the target wind farm from relevant web pages of the target wind farm and an authoritative geographic platform, and performing word segmentation processing on the description information; Based on the segmentation result, the edit distance between the segmentation result and the pre-configured candidate segmentation result is calculated; According to the calculation result of the edit distance, determining key fields related to wind power generation from the description information; A fine-tuning training set is constructed based on the identified key fields related to wind power generation.

[0009] Optionally, it also includes repeatedly obtaining descriptive information related to the geographical location of the target wind farm from relevant web pages of the target wind farm and an authoritative geographic platform according to a preset cycle, and updating the fine-tuning training set to fine-tune the trained large model when there are changes in key fields.

[0010] Optionally, obtaining current weather forecast data for the target wind farm coverage area and outputting wind farm forecast information for multiple distribution locations based on the fine-tuned large model include: The current meteorological forecast data of the target wind farm coverage area is obtained, and tags are added to the current meteorological forecast data based on the keywords of the target wind farm and input into the fine-tuned large model, so as to use the fine-tuned large model to output wind farm forecast information of multiple distribution locations.

[0011] Optionally, calculating the power generation prediction result of the target wind farm based on the power generation prediction information of multiple distribution locations includes: Dividing all wind turbine locations of the target wind farm into a plurality of sub-areas, wherein any sub-area contains at least two predicted distribution locations; Calculating power generation prediction information of other locations within the sub-area based on the power generation prediction information of the divided sub-area; The power generation prediction information of each item is summed up to obtain the power generation prediction result of the target wind farm.

[0012] The embodiment of the present application further proposes a wind power prediction system based on meteorological data, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the wind power prediction method based on meteorological data as described above are implemented.

[0013] The method of the present application combines a large model with LSTM, trains and predicts meteorological data through a large model, and uses LSTM to predict wind farm power generation data, thereby improving the accuracy of wind power prediction.

[0014] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings: Figure 1 This is a basic flow chart of this embodiment. DETAILED DESCRIPTION

[0016] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0017] The present application embodiment proposes a wind power prediction method based on meteorological data, such as Figure 1 As shown, the following steps are included: In step S101, historical meteorological forecast data and historical actual meteorological data of areas covered by different wind farms are obtained in advance, and a large model is trained, and an LSTM model is trained based on the historical meteorological forecast data, historical actual meteorological data and power generation data of multiple distribution locations of different wind farms.

[0018] In some embodiments, pre-acquiring historical meteorological forecast data and historical actual meteorological data of different wind farm coverage areas and training a large model includes: Use crawler technology to obtain historical meteorological forecast data for areas covered by different wind farms from meteorological websites. For example, you can use relevant technologies to obtain historical meteorological forecast data for areas covered by different wind farms from authoritative meteorological websites, or you can use nearby meteorological monitoring stations and other methods to obtain historical meteorological forecast data for areas covered by different wind farms.

[0019] Determine the central area of ​​the wind farm in the historical meteorological forecast data, and calculate the distance deviation between the central area of ​​the wind farm and the areas covered by the different wind farms. In some examples, the meteorological forecast data is not only for different wind farm coverage areas, but for the entire range. In this example, the distance deviation between the central area of ​​the wind farm and the areas covered by the different wind farms is further calculated. In specific applications, the central area of ​​the wind farm does not only include, for example, the center of the typhoon, but also includes the starting point of possible wind factors, etc. The method of the present application is aimed at the fact that most wind farms cover a large area, but the meteorological forecast data only predicts a large area, which may cause prediction deviations. The embodiment of the present application calculates the distance deviation and derives prediction data that is more in line with the geographical location of the wind turbine in subsequent examples based on the reasoning ability of the large model. By selecting different wind farms to perform training, the large model can learn more regional weather information, and through subsequent fine-tuning, it can be adapted to the target wind farm.

[0020] The historical meteorological forecast data is marked based on the distance deviation, and the historical meteorological forecast data and the historical actual meteorological data are used as a set of training data at a set time interval to perform full parameter training on the large model. In some examples, the time interval can be set according to the location of the actual wind farm, for example, according to the periodicity of seasonal climate.

[0021] In step S102, the geographical location description information of the target wind farm is obtained, and key fields related to wind power generation are screened out from the description information to construct a fine-tuning training set based on the key fields.

[0022] In step S103, the trained large model is fine-tuned using the fine-tuning training set. By fine-tuning the large model, a large model of the target wind farm that is more closely matched to the target wind farm is further obtained.

[0023] In step S104, the current weather forecast data of the target wind farm coverage area is obtained, and the wind farm forecast information of multiple distribution locations is output based on the fine-tuned large model. By setting the distribution location, a large-scale forecast of the wind farm is achieved through multiple local representative locations, reducing the amount of calculation and improving the calculation efficiency.

[0024] In step S105, based on the wind farm prediction information, prediction is performed using the trained LSTM model to obtain power generation prediction information for multiple distribution locations.

[0025] In step S106, a power generation prediction result of the target wind farm is calculated based on the power generation prediction information of the multiple distribution locations.

[0026] The method of the present application combines a large model with LSTM, trains and predicts meteorological data through a large model, and uses LSTM to predict wind farm power generation data, thereby improving the accuracy of wind power prediction.

[0027] In some embodiments, training the LSTM model based on historical meteorological forecast data, historical actual meteorological data, and power generation data of multiple distribution locations of different wind farms includes: The terrain relationship of the different wind farms is obtained, and the positions of all wind turbines in the different wind farms are determined based on the terrain relationship. For example, the position of a wind turbine in any wind farm may be marked in the terrain relationship model.

[0028] The terrain features that determine the location of each wind turbine, such as geographical location, slope, etc., are related to power generation on the one hand, and are related factors of the wind duct of the wind farm on the other hand.

[0029] According to the positions of all wind turbines in different wind farms and the terrain features, wind turbines at multiple positions that are distributed and the number of wind turbines required to be selected are selected, wherein the selected wind turbine positions include the position with the largest deviation in the terrain features and multiple positions with dispersed distances from the position with the largest deviation. The position with the largest deviation and other distributed positions are selected as representative positions of the target wind farm for subsequent prediction.

[0030] According to the wind turbines in multiple selected locations, their historical power generation data are obtained to train the LSTM model.

[0031] In some embodiments, obtaining the geographical location description information of the target wind farm, and filtering out key fields related to wind power generation from the description information to construct a fine-tuning training set based on the key fields includes: Descriptive information related to the geographical location of the target wind farm is obtained from web pages related to the target wind farm and an authoritative geographic platform, and word segmentation processing is performed on the descriptive information.

[0032] Based on the results of word segmentation, the edit distance between the pre-configured candidate participles is calculated. For example, candidate participles related to wind turbines and candidate participles related to wind farm locations can be pre-configured. The candidate participles can be applicable to different wind farms and can also be for all-season climates. The fine-tuned large model is adapted to the overall situation of the target wind farm.

[0033] According to the calculation result of the edit distance, the key fields related to wind power generation are determined from the description information, for example, the key fields related to wind power generation are determined from the description information according to the length of the edit distance.

[0034] A fine-tuning training set is constructed based on the identified key fields related to wind power generation.

[0035] In some embodiments, the method further includes repeatedly obtaining description information related to the geographical location of the target wind farm from the target wind farm-related webpage and the authoritative geographic platform according to a preset cycle, and updating the fine-tuning training set to fine-tune the trained large model when there are changes in key fields. For example, if there are changes in terrain, seasonal climate, etc., the training large model can be updated.

[0036] In some embodiments, obtaining current meteorological forecast data for the area covered by the target wind farm, and outputting wind farm forecast information for multiple distributed locations based on the fine-tuned large model includes: obtaining current meteorological forecast data for the area covered by the target wind farm, and adding tags to the current meteorological forecast data based on keywords of the target wind farm and inputting the fine-tuned large model, so as to output wind farm forecast information for multiple distributed locations using the fine-tuned large model. The fine-tuned large model can predict the wind farm information for multiple distributed locations based on the current meteorological forecast data of the target wind farm, thereby improving the prediction accuracy, which is suitable for the characteristics of large-area coverage of wind power.

[0037] In some embodiments, calculating the power generation prediction result of the target wind farm based on the power generation prediction information of multiple distribution locations includes: Dividing all wind turbine locations of the target wind farm into a plurality of sub-areas, wherein any sub-area contains at least two predicted distribution locations; The power generation prediction information of other positions in the sub-region may be calculated based on the power generation prediction information of the divided sub-region. For example, the power generation prediction information of other positions in the sub-region may be calculated by calculating the mean, fitting, or the like.

[0038] The power generation prediction information of each item is summed up to obtain the power generation prediction result of the target wind farm.

[0039] The embodiment of the present application proposes a method for training and predicting meteorological data by combining a large model and LSTM. On the one hand, the reasoning ability of the large model is used to improve the accuracy of wind farm prediction based on historical meteorological data. On the other hand, the memory effect of LSTM is used to greatly improve the convenience and accuracy of wind power prediction.

[0040] The embodiment of the present application further proposes a wind power prediction system based on meteorological data, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the steps of the wind power prediction method based on meteorological data as described above are implemented.

[0041] In addition, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present disclosure with equivalent elements, modifications, omissions, combinations (e.g., various embodiments intersecting schemes), adaptations or changes. It is not limited to the examples described in this specification or during the implementation of this application, and its examples will be interpreted as non-exclusive.

[0042] The above description is intended to be illustrative rather than restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. For example, those of ordinary skill in the art may use other embodiments when reading the above description.

[0043] The above embodiments are merely exemplary embodiments of the present disclosure. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present disclosure, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims

1. A wind power prediction method based on meteorological data, characterized in that: The steps include: Obtain historical meteorological forecast data and historical actual meteorological data for areas covered by different wind farms in advance, and train a large model. Also, train an LSTM model based on the historical meteorological forecast data, historical actual meteorological data, and power generation data from multiple distribution locations of different wind farms. Acquire the geographical location description information of the target wind farm, and filter out key fields related to wind power generation from the description information to construct a fine-tuning training set based on the key fields; Using the fine-tuning training set, fine-tuning the trained large model; Obtain current weather forecast data for the target wind farm coverage area, and output wind farm forecast information for multiple distribution locations based on the fine-tuned large model; Based on the wind farm prediction information, using the trained LSTM model to perform prediction to obtain power generation prediction information for multiple distribution locations; The power generation prediction result of the target wind farm is calculated based on the power generation prediction information of multiple distribution locations.

2. The method for wind power prediction based on meteorological data according to claim 1, characterized in that: Obtain historical meteorological forecast data and historical actual meteorological data for different wind farm coverage areas in advance, and train large models including: Use crawler technology to obtain historical weather forecast data for different wind farm coverage areas from weather websites; Determining the central area of ​​the wind farm in the historical meteorological forecast data, and calculating the distance deviation between the central area of ​​the wind farm and the areas covered by the different wind farms; The historical meteorological forecast data is marked based on the distance deviation, and the historical meteorological forecast data and the historical actual meteorological data are used as a set of training data at set time intervals to perform full parameter training on the large model.

3. The method for wind power prediction based on meteorological data according to claim 2, characterized in that: The LSTM model is trained based on historical meteorological forecast data, historical actual meteorological data, and power generation data from multiple distribution locations of different wind farms, including: Acquire the terrain relationship of the different wind farms, and determine the positions of all wind turbines in the different wind farms based on the terrain relationship; terrain features that determine the location of each wind turbine; According to the positions of all wind turbines in different wind farms and the terrain features, wind turbines at multiple positions of the required number of wind turbines are selected, wherein the selected wind turbine positions include a position with the largest deviation in the terrain features and multiple positions with dispersed distances from the position with the largest deviation; According to the wind turbines in multiple selected locations, their historical power generation data are obtained to train the LSTM model.

4. The method for wind power prediction based on meteorological data according to claim 3, characterized in that: Acquiring the geographical location description information of the target wind farm, and filtering out key fields related to wind power generation from the description information to construct a fine-tuning training set based on the key fields includes: Obtaining description information related to the geographical location of the target wind farm from relevant web pages of the target wind farm and an authoritative geographic platform, and performing word segmentation processing on the description information; Based on the segmentation result, the edit distance between the segmentation result and the pre-configured candidate segmentation result is calculated; According to the calculation result of the edit distance, determining key fields related to wind power generation from the description information; A fine-tuning training set is constructed based on the identified key fields related to wind power generation.

5. The method for wind power prediction based on meteorological data according to claim 4, characterized in that: It also includes repeatedly obtaining descriptive information related to the geographical location of the target wind farm from relevant web pages of the target wind farm and an authoritative geographic platform according to a preset cycle, and updating the fine-tuning training set to fine-tune the trained large model when there are changes in key fields.

6. The method for wind power prediction based on meteorological data according to claim 4, characterized in that: Obtain the current weather forecast data for the target wind farm coverage area, and output wind farm forecast information for multiple distribution locations based on the fine-tuned large model, including: The current meteorological forecast data of the target wind farm coverage area is obtained, and tags are added to the current meteorological forecast data based on the keywords of the target wind farm and input into the fine-tuned large model, so as to use the fine-tuned large model to output wind farm forecast information of multiple distribution locations.

7. The method for wind power prediction based on meteorological data according to claim 4, characterized in that: The power generation forecast results of the target wind farm calculated based on the power generation forecast information of multiple distribution locations include: Dividing all wind turbine locations of the target wind farm into a plurality of sub-areas, wherein any sub-area contains at least two predicted distribution locations; Calculating power generation prediction information of other locations within the sub-area based on the power generation prediction information of the divided sub-area; The power generation prediction information of each item is summed up to obtain the power generation prediction result of the target wind farm.

8. A wind power prediction system based on meteorological data, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method for wind power prediction based on meteorological data as claimed in any one of claims 1 to 7 are implemented.

Images