Training method and computer equipment for wind farm icing power prediction model

Abstract

The present invention relates to the field of clean energy technology, and discloses a training method and computer equipment for a wind farm icing power prediction model. The method comprises: obtaining historical predicted power data, historical measured power data, and historical meteorological data of a target wind farm station as a feature data set; screening and grouping the feature data set for icing to obtain an icing label data set; detecting an icing start date and an icing end date for the icing label data set based on numerical features of sequence data and / or change features of adjacent sequence data to obtain an icing time period; screening the icing label data set based on the icing time period to obtain an icing data set; inputting the icing data set into a pre-built preset model for model training until the preset model converges to obtain a wind farm icing prediction model; and being able to screen the data set and perform icing prediction based on the numerical features or change features of the sequence data, taking into account obvious precursors and successors in an icing situation.

Description

Technical Field

[0001] The present invention relates to the field of clean energy technology, and in particular to a training method and computer equipment for a wind farm icing power prediction model. Background Art

[0002] As a type of clean energy, wind power generation is sensitive to changes in meteorological conditions, and its power generation process has certain instability. In winter and spring, wind turbine blades in humid climate areas often become covered with ice, which becomes the main obstacle to normal power generation in wind farms. It is necessary to predict the icing situation of wind turbine blades.

[0003] Related technologies typically use machine learning or deep learning methods to fit single-point meteorological factors and icing conditions to predict wind turbine blade icing. However, machine learning and deep learning algorithms based on single-point calculations have certain limitations. Specifically, they struggle to capture the changing characteristics of daytime meteorological conditions under extreme weather conditions, struggle to adaptively learn from intraday temperature and humidity variations, and are sensitive to high sample cleanliness requirements. Furthermore, sample imbalance is another issue with single-point machine learning and deep learning algorithms.

[0004] Therefore, there is an urgent need to propose a training method for a wind farm icing power prediction model to solve the problem of inaccurate prediction in the wind farm icing prediction method based on a single point to determine the icing status of the wind farm station in the related art. Summary of the Invention

[0005] In view of this, the present invention provides a training method and computer equipment for a wind farm icing power prediction model to solve the problem of inaccurate prediction in the wind farm icing prediction method based on a single point to determine the icing status of the wind farm station in the related art.

[0006] In a first aspect, the present invention provides a method for training a wind farm icing power prediction model, the method comprising: obtaining historical predicted power data, historical measured power data and historical meteorological data of a target wind farm station as a feature data set; performing icing screening and grouping on the feature data set to obtain an icing label data set; wherein the icing label data set comprises a plurality of groups of sequence data grouped according to a first preset sliding window and a first preset sliding step size; performing icing start date and icing end date screening on the icing label data set based on the numerical features of the sequence data and / or the change features of adjacent sequence data. Detection is performed to obtain an icing time period; the icing label data set is screened based on the icing time period to obtain an icing data set; wherein, the icing data set includes the historical meteorological data and regression label column corresponding to the icing time period, and the regression label column is determined by the historical predicted power data and the historical measured power data corresponding to the icing time period; the icing data set is input into a pre-built preset model for model training, and in the process of model training, the correspondence between the historical meteorological data and the regression label column in the icing time period is learned until the preset model converges, thereby obtaining a wind farm icing prediction model.

[0007] As an exemplary embodiment, the ice-covering start date and ice-covering end date of the ice-covering label data set are detected based on the numerical characteristics of the sequence data and / or the change characteristics of the adjacent sequence data to obtain the ice-covering time period, including: detecting the ice-covering start date of the ice-covering label data set based on the numerical characteristics of the sequence data and / or the change characteristics of the adjacent sequence data to obtain the ice-covering start date; detecting the ice-covering end date based on the numerical characteristics or change characteristics of the sequence data after the ice-covering start date to obtain the ice-covering end date matching each ice-covering start date; determining the ice-covering start time corresponding to the ice-covering start date and the ice-covering end time corresponding to the ice-covering end date based on the numerical characteristics of the historical meteorological data of each ice-covering start date and the corresponding ice-covering end date; and using the ice-covering start time and the ice-covering end time as the start time and end time of the ice-covering time period to obtain the ice-covering time period.

[0008] As an exemplary embodiment, the ice cover start date detection is performed on the ice cover label dataset based on the numerical characteristics of the sequence data to obtain multiple ice cover start dates, including: traversing and detecting the meteorological data in the sequence data based on the ice cover temperature threshold and the ice cover humidity threshold to obtain the first target ice cover start sequence data whose minimum temperature is less than the ice cover temperature threshold and whose maximum humidity is greater than the ice cover humidity threshold; wherein the minimum temperature of the adjacent sequence data adjacent to the first target ice cover start sequence data window is less than the ice cover temperature threshold and the maximum humidity is greater than the ice cover humidity threshold; and taking the first date corresponding to the first target ice cover start sequence data as the ice cover start date.

[0009] As an exemplary embodiment, the icing start date detection is performed on the icing label dataset based on the numerical characteristics of the sequence data and the change characteristics of the adjacent sequence data to obtain the icing start date, including: traversing the sequence data to obtain second target icing start sequence data, in which the sum of the minimum temperature value and the first preset temperature value is less than the icing temperature threshold and the maximum humidity value is greater than the icing humidity threshold; wherein the minimum temperature value of the adjacent sequence data adjacent to the second target icing start sequence data window decreases with the sequence sorting; and the second date corresponding to the second target icing start sequence data is used as the icing start date.

[0010] As an exemplary embodiment, the icing start date detection is performed on the icing label dataset based on the change characteristics of the sequence data to obtain multiple icing start dates, including: calculating the corresponding difference between the temperature extreme value and the humidity extreme value of each sequence data compared with the temperature extreme value and the humidity extreme value of the adjacent sequence data that is earlier in time series, to obtain the maximum humidity change value, the minimum humidity change value, the maximum temperature change value and the minimum temperature change value; taking the sequence data in which the maximum humidity change value and the minimum humidity change value are both greater than the preset humidity change value or the maximum temperature change value and the minimum temperature change value are both greater than the preset temperature change value as the third target icing start sequence data; and taking the third date corresponding to the third target icing start sequence data as the icing start date.

[0011] As an exemplary embodiment, the training method of the wind farm icing power prediction model also includes: traversing each of the icing start dates to obtain the first minimum temperature value of the sequence data corresponding to the icing start date; obtaining the second minimum temperature value of the adjacent sequence data whose windows are adjacent and whose time sequence is later than the icing start date; and eliminating the icing start date whose second minimum temperature value is greater than the first minimum temperature value, and the sum of the second minimum temperature value and the second preset temperature value is greater than the icing temperature threshold.

[0012] As an exemplary embodiment, the icing end date detection is performed based on the numerical characteristics or change characteristics of the sequence data after the icing start date to obtain the icing end date matching each of the icing start dates, including: obtaining the actual climate type of the target wind farm station; when the actual climate type meets the first preset climate type, traversing the numerical characteristics of the sequence data after the icing start date in sequence, and obtaining the sequence data with a minimum temperature greater than the icing temperature threshold and / or a maximum humidity less than the icing humidity threshold as the target icing end sequence data; when the actual climate type meets the second preset climate type, traversing the numerical characteristics and / or change characteristics of the sequence data after the icing start date in sequence, and obtaining the first target sequence data with a maximum temperature greater than the icing temperature threshold, a temperature change value greater than the preset temperature change value, and a humidity change value greater than the preset humidity change value, or the second target sequence data with a humidity change value greater than the preset humidity change value as the target icing end sequence data; and taking the date corresponding to the target icing end sequence data as the icing end date.

[0013] As an exemplary embodiment, the determining of the ice-covering start time corresponding to the ice-covering start date and the ice-covering end time corresponding to the ice-covering end date based on the numerical characteristics of the historical meteorological data of each of the ice-covering start dates and each of the corresponding ice-covering end dates includes: sliding backtracking and sliding forward of the sequence data corresponding to the ice-covering start time and the ice-covering end time according to a second preset sliding window with a second preset sliding step size to obtain a plurality of first sub-ice-covering sequence data and second sub-ice-covering sequence data; wherein the window length of the second preset sliding window is smaller than the window length of the first preset sliding window; traversing and screening the first sub-ice-covering sequence data to obtain target first sub-ice-covering sequence data whose temperature values ​​are all smaller than the ice-covering temperature threshold and whose humidity values ​​are all greater than the preset humidity value; traversing and screening the second sub-ice-covering sequence data to obtain second target sub-ice-covering sequence data whose temperature values ​​are all greater than the ice-covering temperature threshold; taking the start time of the target first sub-ice-covering sequence data as the ice-covering start time; and taking the start time of the second target sub-ice-covering sequence data as the ice-covering end time.

[0014] As an exemplary embodiment, before determining the icing start time corresponding to the icing start date and the icing end time corresponding to the icing end date based on the numerical characteristics of the historical meteorological data of each of the icing start dates and the corresponding icing end dates, the training method of the wind farm icing power prediction model also includes: obtaining multiple overlapping icing time periods containing the same time interval; merging the overlapping icing time periods based on the icing start dates of each of the overlapping icing time periods, and obtaining the minimum target icing time period on the icing start day as the icing time period.

[0015] In a second aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to execute the training method of the wind farm icing power prediction model as described in any of the above embodiments.

[0016] The present invention provides a training method and computer equipment for an icing power prediction model for a wind farm, the method comprising: obtaining historical predicted power data, historical measured power data and historical meteorological data of a target wind farm as a feature data set; screening and grouping the feature data set for icing to obtain an icing label data set; wherein the icing label data set comprises a plurality of groups of sequence data grouped according to a first preset sliding window and a first preset sliding step; detecting the icing start date and the icing end date of the icing label data set based on the numerical features of the sequence data and / or the change features of adjacent sequence data to obtain an icing time period; screening the icing label data set based on the icing time period to obtain an icing data set; wherein the icing data set comprises the historical meteorological data and a regression label column corresponding to the icing time period, and the regression label column is determined by the historical predicted power data and the historical measured power data corresponding to the icing time period; inputting the icing data set into a pre-built preset model for model training, and learning the historical meteorological data within the icing time period during the model training process. The correspondence between the image data and the regression label column is calculated until the preset model converges to obtain a wind farm icing prediction model; in the above method, the numerical features of the sequence data can reflect the characteristic representation of the meteorological conditions at the time scale with the time scale of the first preset sliding window as the time unit; the change characteristics of the numerical features of adjacent sequence data can reflect the changing process of the meteorological conditions under the time period determined by the adjacent first preset time window and the first preset sliding step. For icing weather under extreme weather conditions of continuous processes, a detection method for the icing time period is used to detect the icing start date and icing end date of the icing label dataset based on the numerical features of the sequence data and / or the change characteristics of the adjacent sequence data. Compared with sample point by sample point, it can better represent the reliability of icing prediction under extreme weather conditions and can take into account the obvious precursor and successor of icing conditions; further, when the icing dataset obtained by screening using the above method is used for model training, it has a higher hit rate than the traditional icing prediction method, can effectively mark the icing period of the wind farm, reduce the occurrence of icing omissions, and reduce the power prediction deviation caused by icing. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 1 is a flow chart of a method for training a wind farm icing power prediction model according to an embodiment of the present invention;

[0019] Figure 2 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0020] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0021] As a type of clean energy, wind power generation is sensitive to changes in meteorological conditions, and its power generation process has certain instability. In winter and spring, wind turbine blades in humid climate areas often become covered with ice, which becomes the main obstacle to normal power generation in wind farms. It is necessary to predict the icing situation of wind turbine blades.

[0022] When it comes to predicting ice cover periods, machine learning and deep learning algorithms based on single-point computation have certain limitations. Specifically, they struggle to capture the changing characteristics of meteorological features during the day under extreme weather conditions, struggle to adaptively learn from intraday temperature and humidity fluctuations, and are sensitive to high sample cleanliness requirements. Furthermore, sample imbalance is another issue with single-point computation-based machine learning and deep learning algorithms.

[0023] In view of this, the present invention provides an embodiment of a training method for a wind farm icing power prediction model. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0024] In this embodiment, a training method for a wind farm icing power prediction model is provided. Figure 1FIG. 1 is a flow chart of a method for training a wind farm icing power prediction model according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0025] Step S101 : acquiring historical predicted power data, historical measured power data, and historical meteorological data of a target wind farm as a feature data set.

[0026] In this embodiment, in order to construct historical training data for the icing prediction model, historical predicted power data, historical measured power data and historical meteorological data are obtained as feature data sets; wherein, the historical meteorological data may include temperature, relative humidity, wind speed, wind direction, precipitation, etc.

[0027] For example, after obtaining historical predicted power data, historical measured power data and historical meteorological data, these sets of data are aligned and integrated using inner join according to the time column, and the resulting data is used as historical training data for wind farm icing period prediction.

[0028] Statistics show that most ice coverage occurs between November and March of the following year. Therefore, in this embodiment, the feature data set can select data from November to March of the following year in the past three years as training samples. The above data set selection method can, on the one hand, downsample the non-ice-covered samples, and on the other hand, prevent the subsequent percentile screening threshold from being affected by noise and causing large deviations.

[0029] Furthermore, wind farm units are subject to abnormal operating conditions, such as unit shutdowns (planned or faulty), grid curtailment, and sensor failures. Operating data under these abnormal conditions is often distorted, which interferes with accurate predictions of the overall freezing period and status of the wind farm during icing forecasting, directly impacting the reliability of the icing prediction model and the accuracy of wind farm icing period forecasts. Therefore, it is necessary to collect, mark, and eliminate these abnormal operating data.

[0030] For example, abnormal data in the station's historical records are directly discarded.

[0031] For example, for unknown abnormal data, statistics-based threshold selection methods and unsupervised learning methods such as clustering (K-means++) and isolation forest are used to perform anomaly detection, and then the abnormal data is eliminated after evaluation.

[0032] Step S102 , performing ice screening and grouping on the feature data set to obtain an ice label data set; wherein the ice label data set includes multiple groups of sequence data grouped according to a first preset sliding window and a first preset sliding step size.

[0033] Illustratively, after obtaining the feature data set, the feature data set is screened and grouped based on ice coverage.

[0034] In the process of ice cover screening of feature data sets, the meteorological elements contained in the historical meteorological data are first screened to obtain multiple ice cover meteorological factors whose impact on ice cover is greater than the preset impact level; specifically, temperature, humidity, wind speed, etc. are selected as core features based on the dual verification of physical theoretical mechanisms and data-driven statistics.

[0035] From a meteorological perspective, the mechanism of temperature is to control the phase transition threshold. Ice formation requires the phase transition condition of supercooled water droplets in contact with a low-temperature object. Temperature directly determines the efficiency of the conversion of liquid water to solid ice. Humidity serves as a material basis. When the relative humidity is too high, the concentration of supercooled water droplets in the air increases significantly, providing an ample source of liquid water for ice accumulation. In low-humidity environments, even if the temperature meets the standard, the ice accumulation rate will drop sharply due to insufficient water. Wind speed can serve as a dynamic balance factor, reflecting the speed at which supercooled water droplets carried by the wind hit the blades, thereby affecting the ice accumulation rate.

[0036] Furthermore, from a data-driven statistical perspective, the importance of candidate climate features was quantified and ranked for selection. Specifically, the data was first standardized, with mean-variance normalization performed on candidate indicators such as temperature, humidity, and wind speed to eliminate the impact of dimensional differences.

[0037] Secondly, in terms of statistical methods, the Grey Relational Analysis (GRA) method is used to quantitatively calculate the correlation values ​​between multiple indicator factors and the icing status of wind farms, determine the importance of the influencing factors, and sort them by the degree of influence. The features with greater influence are selected to form a key factor sequence.

[0038] Among them, random forest and XGBoost modeling are used in machine learning methods. The importance score of influencing factors is determined through the built-in feature importance scoring strategy (Gini importance, gain contribution), and the contribution of features to the model output is quantified through SHAP value to determine the final selection of key ice cover features.

[0039] Furthermore, due to wind farm maintenance and other uncontrollable factors, the actual icing data of most wind farms are incomplete, fragmented and difficult to obtain in batches. Therefore, it is necessary to further screen the key icing features obtained and calculate the icing threshold features.

[0040] Specifically, first, the first fuzzy screening is performed based on the key icing features and feature data sets to mark the icing sample points, and the sample points with zero actual power and a small ratio of actual power to predicted power are screened out.

[0041] Furthermore, based on the first fuzzy screening to mark the ice-covered sample points, a second fuzzy screening is performed to mark the ice-covered sample points; specifically, according to meteorological theory, meteorological boundary values ​​that are conducive to freezing and icing are manually preset, such as the temperature boundary value T_b and the humidity boundary value H_b. When the temperature of the sample point is less than T_b and the humidity is greater than H_b, the sample point is retained; the second fuzzy screening aims to screen out reasonable ice-covered sample points according to meteorological theory and minimize the impact of abnormal noise point data.

[0042] Furthermore, the unsupervised learning DBSCAN algorithm was used to perform cluster analysis to separate the ice cover cluster from other cause clusters and retain the ice cover cluster.

[0043] Finally, from the perspective of statistics and probability density distribution, the temperature and humidity quantiles within the ice-covered cluster are calculated to help identify high-density areas of ice-covered samples, aiming to screen out more refined and representative ice-covered sample points; the minimum temperature quantile within the ice-covered cluster is recorded as the ice-covered temperature threshold T_tsh, and the maximum humidity quantile within the cluster is recorded as the ice-covered humidity threshold H_tsh. They are important hyperparameters of the ice-covered period prediction model, which affect the accuracy of ice-covered prediction based on daily changes and sliding windows.

[0044] In order to solve the limitations of machine learning and deep learning algorithms based on single-point calculations, which are manifested in that it is not only difficult to capture the changing process of meteorological characteristics during the day under extreme weather conditions, but also difficult to adaptively learn the changes in temperature and humidity during the day, in this embodiment, the ice-covered clusters are divided and grouped by sliding windows to obtain multiple groups of sequence data; specifically, after obtaining the ice-covered clusters, the ice-covered clusters are grouped according to a first preset sliding window and a first preset sliding step size to obtain an ice-covered label data set consisting of multiple groups of sequence data grouped according to the first preset sliding window and the first preset sliding step size.

[0045] Exemplarily, the time scale of the first preset window is no greater than 24 hours.

[0046] In one embodiment, the first preset window is 96 sample points, and the corresponding time scale is 24 hours; the first preset sliding step is 96 sample points, and the first preset sliding step of 96 sample points and the first preset sliding window are used to obtain daily variation series data with a day as the time scale as the ice cover label data set.

[0047] In one embodiment, the first preset window is 12 sample points, and the first preset sliding step is 12 sample points.

[0048] In one embodiment, the first preset window is 8 sample points, and the first preset sliding step is 8 sample points.

[0049] For example, after obtaining the ice label dataset, for each sequence data in the ice label dataset, the minimum temperature value of each sequence data is extracted and recorded as T_min, and the maximum temperature value is recorded as T_max; the maximum humidity value is extracted and recorded as H_max, and the minimum temperature value of the i-th sequence data is further recorded as D i [T_min], the maximum humidity value of the i-th sequence data is recorded as D i [H_max].

[0050] Step S103 : detecting the ice cover start date and ice cover end date of the ice cover label dataset based on the numerical features of the sequence data and / or the change features of adjacent sequence data to obtain an ice cover time period.

[0051] In this embodiment, the ice covering start date and ice covering end date of the ice covering label dataset are detected by using the numerical features of the sequence data and / or the change features of the adjacent sequence data to obtain the ice covering time period.

[0052] In one embodiment, the ice cover start date and ice cover end date may be detected based solely on the numerical features of the sequence data to obtain the ice cover time period.

[0053] For example, the ice cover start date and ice cover end date can be detected based on the relationship between the temperature extreme value and the ice cover temperature threshold and the relationship between the humidity extreme value and the ice cover humidity threshold of each sequence data to obtain the ice cover time period.

[0054] In one embodiment, the ice cover start date and ice cover end date may be detected based on the numerical features of adjacent sequence data to obtain the ice cover time period.

[0055] In one embodiment, the ice covering start date and ice covering end date of the ice covering label dataset may be detected based on the numerical features of the sequence data and the change features of adjacent sequence data to obtain the ice covering time period.

[0056] In the above embodiment, the numerical characteristics of the sequence data can reflect the characteristic representation of the meteorological conditions at the time scale of the first preset sliding window as the time unit; the change characteristics of the numerical characteristics of adjacent sequence data can reflect the change process of the meteorological conditions under the time period determined by the adjacent first preset time window and the first preset sliding step. For icing weather under extreme weather conditions in a continuous process, a detection method for the icing time period is adopted to detect the icing start date and icing end date of the icing label data set based on the numerical characteristics of the sequence data and / or the change characteristics of the adjacent sequence data. This method can better represent the reliability of icing prediction under extreme weather conditions than sample point by sample point, and can take into account the obvious predecessor and successor characteristics of the icing situation.

[0057] Step S104: Filter the ice label data set based on the ice time period to obtain an ice data set; wherein, the ice data set includes the historical meteorological data and regression label column corresponding to the ice time period, and the regression label column is determined by the historical predicted power data and the historical measured power data corresponding to the ice time period.

[0058] Illustratively, after obtaining the ice-covered time period, the ice-covered label dataset is filtered according to the ice-covered time period to obtain an ice-covered dataset; the ice-covered dataset includes the historical meteorological data and regression label columns corresponding to the ice-covered time period.

[0059] For example, the data sample labels in the ice cover dataset are of the one-hot encoding type, representing the ice cover status of each sample point.

[0060] For example, after obtaining the icing data set, the regression label column is determined based on the comparison of the historical predicted power data and the historical measured power data corresponding to the icing time period to represent the icing coefficient of the sample point during the icing period, which is used to subsequently optimize the predicted power of each point during the icing period.

[0061] Step S105: input the icing dataset into a pre-built preset model for model training. During the model training process, the correspondence between the historical meteorological data in the icing time period and the regression label column is learned until the preset model converges to obtain a wind farm icing prediction model.

[0062] In this embodiment, the icing dataset is input into a pre-built preset model for model training. During the model training process, the correspondence between the historical meteorological data in the icing time period and the regression label column is learned until the preset model converges to obtain a wind farm icing prediction model.

[0063] Among them, as a possible implementation method, the preset model can be constructed based on an integrated learning model and a deep learning model; specifically, the integrated learning model can be a LightGBM model, and the deep learning model can be a convolutional neural network (CNN).

[0064] Among them, the trained wind farm icing prediction model is used to output the icing coefficient corresponding to each point in the icing period based on the input meteorological data and the corresponding relationship, and multiply the predicted power by the icing coefficient according to each icing point to output the predicted power under the influence of final icing.

[0065] The training method of the wind farm icing power prediction model provided by the present invention comprises: obtaining historical predicted power data, historical measured power data and historical meteorological data of the target wind farm station as a feature data set; performing icing screening and grouping on the feature data set to obtain an icing label data set; wherein the icing label data set comprises a plurality of groups of sequence data grouped according to a first preset sliding window and a first preset sliding step; performing icing start date and icing end date detection on the icing label data set based on the numerical characteristics of the sequence data and / or the change characteristics of adjacent sequence data to obtain an icing time period; screening the icing label data set based on the icing time period to obtain an icing data set; wherein the icing data set comprises the historical meteorological data and regression label column corresponding to the icing time period, and the regression label column is determined by the historical predicted power data and the historical measured power data corresponding to the icing time period; inputting the icing data set into a pre-built preset model for model training, and in the process of model training, learning the icing time period. The correspondence between the historical meteorological data and the regression label column is obtained until the preset model converges to obtain a wind farm icing prediction model; in the above method, the numerical characteristics of the sequence data can reflect the characteristic representation of the meteorological conditions at the time scale with the time scale of the first preset sliding window as the time unit; the change characteristics of the numerical characteristics of adjacent sequence data can reflect the meteorological change process under the time period determined by the adjacent first preset time window and the first preset sliding step size, and the icing time period detection method is used to detect the icing start date and icing end date of the icing label data set based on the numerical characteristics of the sequence data and / or the change characteristics of the adjacent sequence data. Compared with each sample point, it can better represent the reliability of icing prediction under extreme weather conditions, and can take into account the obvious predecessors and successors under icing conditions; further, when the icing data set obtained by screening using the above method is used for model training, it has a higher hit rate than the traditional icing prediction method, can effectively mark the icing period of the wind farm, reduce the occurrence of icing omissions, and reduce the power prediction deviation caused by icing.

[0066] As an exemplary embodiment, the ice-covering start date and ice-covering end date of the ice-covering label data set are detected based on the numerical characteristics of the sequence data and / or the change characteristics of the adjacent sequence data to obtain the ice-covering time period, including: detecting the ice-covering start date of the ice-covering label data set based on the numerical characteristics of the sequence data and / or the change characteristics of the adjacent sequence data to obtain the ice-covering start date; detecting the ice-covering end date based on the numerical characteristics or change characteristics of the sequence data after the ice-covering start date to obtain the ice-covering end date matching each ice-covering start date; determining the ice-covering start time corresponding to the ice-covering start date and the ice-covering end time corresponding to the ice-covering end date based on the numerical characteristics of the historical meteorological data of each ice-covering start date and the corresponding ice-covering end date; and using the ice-covering start time and the ice-covering end time as the start time and end time of the ice-covering time period to obtain the ice-covering time period.

[0067] In one embodiment, when performing icing start date detection, the icing start date may be determined solely based on the numerical features of the sequence data.

[0068] Specifically, as a possible implementation method, for general extreme weather conditions, when multiple sequence data adjacent to the window all meet the requirement that the minimum temperature is less than the icing temperature threshold and the maximum humidity is greater than the icing humidity threshold, the first date corresponding to the sequence data corresponding to the middle window is used as the icing start date.

[0069] Based on this, as an exemplary embodiment, the ice cover start date detection is performed on the ice cover label dataset based on the numerical characteristics of the sequence data to obtain multiple ice cover start dates, including: based on the ice cover temperature threshold and the ice cover humidity threshold, the meteorological data in the sequence data is traversed and detected to obtain the first target ice cover start sequence data whose minimum temperature is less than the ice cover temperature threshold and whose maximum humidity is greater than the ice cover humidity threshold; wherein the minimum temperature of the adjacent sequence data adjacent to the first target ice cover start sequence data window is less than the ice cover temperature threshold, and the maximum humidity is greater than the ice cover humidity threshold; and the first date corresponding to the first target ice cover start sequence data is used as the ice cover start date.

[0070] In one embodiment, multiple ice cover start dates are obtained by detecting the numerical characteristics of the sequence data whose window adjacent number is 1; that is, when D i Minimum temperature D i [T_min], D i-1 Minimum temperature D i-1 [T_min] and D i+1 Minimum temperature D i+1 When [T_min] is less than the ice temperature threshold T_tsh, and Di The maximum humidity D i [H_max], D i-1 The maximum humidity D i-1 [H_max] and D i+1 The maximum humidity D i+1 [H_max] is greater than the ice humidity threshold H_tsh, then it is determined that D i The corresponding date is the ice cover start date; it should be understood that the window adjacent number can also be other positive integers.

[0071] Since the freezing state is more sensitive to temperature, as another possible implementation method, when determining the start date of icing based solely on the numerical characteristics of the sequence data, when the minimum temperature value of the sequence data is less than the icing temperature threshold minus the first preset value, and the maximum humidity value is greater than the icing humidity threshold, the date corresponding to the sequence data is used as the start date of icing.

[0072] Exemplarily, the first preset value may be 2, that is, D i Minimum temperature D i [T_min] is less than the ice temperature threshold T_tsh-2, and D i The maximum humidity D i When [H_max] is greater than the ice humidity threshold H_tsh, it is determined that D i The corresponding date is the ice cover start date.

[0073] In one embodiment, when detecting the ice cover start date, the ice cover start date may be determined based on the numerical characteristics of the sequence data and the change characteristics of adjacent sequence data.

[0074] As a possible implementation method, considering the influence of continuous time periods, it is judged whether the temperature continuously decreases to below the temperature threshold T_tsh to determine whether there is ice covering; specifically, as an exemplary embodiment, the ice covering start date is detected on the ice covering label dataset based on the numerical characteristics of the sequence data and the change characteristics of the adjacent sequence data to obtain the ice covering start date, including: traversing the sequence data to obtain the second target ice covering start sequence data whose sum of the minimum temperature value and the first preset temperature value is less than the ice covering temperature threshold and the maximum humidity value is greater than the ice covering humidity threshold; wherein the minimum temperature value of the adjacent sequence data adjacent to the second target ice covering start sequence data window decreases with the sequence sorting; and the second date corresponding to the second target ice covering start sequence data is used as the ice covering start date.

[0075] In one embodiment, multiple ice cover start dates are obtained by detecting numerical features of sequence data and change features of adjacent sequence data with a window adjacent number of 1.

[0076] Exemplarily, the first preset temperature value may be 0.5.

[0077] That is, when D i [T_min] is less than D i-1 [T_min], D i+1 [T_min] is less than D i [T_min], and D i [T_min] is less than the temperature threshold T_tsh-0.5, D i When [H_max] is greater than the humidity threshold H_tsh, it is determined that D i The day when the ice cover period begins.

[0078] In one embodiment, when detecting the ice accumulation start date, the ice accumulation start date may be determined based on the change characteristics of the adjacent sequence data.

[0079] As a possible implementation method, considering the situation that drastic meteorological changes and strong cold air between consecutive time periods may cause icing, multiple icing start dates are obtained by comparing the corresponding differences between the temperature extreme values ​​and humidity extreme values ​​of the sequence data and the temperature extreme values ​​and humidity extreme values ​​of the adjacent sequence data that precede the time series; specifically, as an exemplary embodiment, the icing start date detection is performed on the icing label data set based on the change characteristics of the sequence data to obtain multiple icing start dates, including: calculating the corresponding differences between the temperature extreme values ​​and humidity extreme values ​​of each of the sequence data and the temperature extreme values ​​and humidity extreme values ​​of the adjacent sequence data that precede the time series to obtain the maximum humidity change value, the minimum humidity change value, the maximum temperature change value and the minimum temperature change value; taking the sequence data in which the maximum humidity change value and the minimum humidity change value are both greater than the preset humidity change value or the maximum temperature change value and the minimum temperature change value are both greater than the preset temperature change value as the third target icing start sequence data; and taking the third date corresponding to the third target icing start sequence data as the icing start date.

[0080] In one embodiment, the maximum humidity change value is recorded as H_max_dif, the minimum humidity change value is recorded as H_min_dif, the maximum temperature change value is recorded as T_max_dif, and the minimum temperature change value is recorded as T_min_dif.

[0081] For example, the preset humidity change value may be 20.

[0082] For example, the preset temperature change value may be -10.

[0083] That is, for situations where drastic daytime weather changes and strong cold air may cause icing, when the maximum humidity change value H_max_dif is greater than 20, or the minimum humidity change value H_min_dif is greater than 20, or the maximum temperature change value T_max_dif is less than -10, or the minimum temperature change value T_min_dif is less than -10, then D is determined. i The day when the ice cover period begins.

[0084] Furthermore, since the frozen state is more sensitive to the subsequent temperature, for short-term warming situations, once icing begins, when the meteorological conditions temporarily do not meet the icing conditions, it is judged that the wind turbine is not iced; therefore, after obtaining the icing start date, the icing state is verified based on the subsequent sequence data that is adjacent to each icing start date window and has a later time sequence, so as to prevent misjudgment caused by short-term warming.

[0085] Based on this, as an exemplary embodiment, the training method of the wind farm icing power prediction model also includes: traversing each of the icing start dates, obtaining the first minimum temperature value of the sequence data corresponding to the icing start date; obtaining the second minimum temperature value of the adjacent sequence data whose windows are adjacent and whose time sequence is later than the icing start date; and eliminating the icing start date whose second minimum temperature value is greater than the first minimum temperature value, and the sum of the second minimum temperature value and the second preset temperature value is greater than the icing temperature threshold.

[0086] Wherein, for example, the second preset temperature value may be 1; that is, when D i+1 Minimum temperature D i+1 [T_min] is greater than D i [T_min], and is greater than the ice temperature threshold T_tsh-1, it is determined that D i To avoid ice accumulation and prevent misjudgment due to short-term temperature rise.

[0087] As an exemplary embodiment, the icing end date detection is performed based on the numerical characteristics or change characteristics of the sequence data after the icing start date to obtain the icing end date matching each of the icing start dates, including: obtaining the actual climate type of the target wind farm station; when the actual climate type meets the first preset climate type, traversing the numerical characteristics of the sequence data after the icing start date in sequence, and obtaining the sequence data with a minimum temperature greater than the icing temperature threshold and / or a maximum humidity less than the icing humidity threshold as the target icing end sequence data; when the actual climate type meets the second preset climate type, traversing the numerical characteristics and / or change characteristics of the sequence data after the icing start date in sequence, and obtaining the first target sequence data with a maximum temperature greater than the icing temperature threshold, a temperature change value greater than the preset temperature change value, and a humidity change value greater than the preset humidity change value, or the second target sequence data with a humidity change value greater than the preset humidity change value as the target icing end sequence data; and taking the date corresponding to the target icing end sequence data as the icing end date.

[0088] In the present invention, the actual climate type of the target wind farm is taken into consideration to determine the end date of icing. For the first preset climate type with normal temperature and humidity, the determination of the end of icing is more sensitive to the temperature and humidity thresholds. For the second preset climate type that often maintains a low temperature and high humidity state, the determination of the end of icing is more sensitive to the temperature and humidity.

[0089] As an exemplary embodiment, obtaining the actual climate type of the target wind farm station includes:

[0090] Obtain historical humidity data and historical temperature data of a target wind farm; calculate the proportion of the historical humidity data that is greater than the preset humidity data to obtain a first proportion result; calculate the proportion of the historical temperature data that is less than the preset temperature data to obtain a second proportion result; when the first proportion result is greater than the first preset proportion result and the second proportion result is greater than the second preset proportion result, record the actual climate type of the target wind farm as the second preset climate type.

[0091] Based on this, when the actual climate type meets the first preset climate type, the numerical characteristics of the sequence data after the ice cover start date are traversed in sequence, and the sequence data whose minimum temperature is greater than the ice cover temperature threshold and / or the maximum humidity is less than the ice cover humidity threshold is obtained as the target end sequence data.

[0092] For example, for some urban areas in Hunan and Henan, when D j Minimum temperature D j When [T_min] is greater than the ice temperature threshold T_tsh, or D j The maximum humidity Dj When [H_max] is less than the ice humidity threshold H_tsh, it is determined that D j The target end sequence data is the target icing end sequence data, and the date corresponding to the target icing end sequence data is used as the icing end date.

[0093] When the actual climate type meets the second preset climate type, the numerical characteristics and / or change characteristics of the sequence data after the ice cover start date are traversed in sequence to obtain the first target sequence data whose maximum temperature is greater than the ice cover temperature threshold, the temperature change value is greater than the preset temperature change value, and the humidity change value is greater than the preset humidity change value, or the second target sequence data whose humidity change value is greater than the preset humidity change value as the target ice cover end sequence data; and the date corresponding to the target ice cover end sequence data is used as the ice cover end date.

[0094] For example, in some urban areas of Guizhou, the actual climate type meets the second preset climate type, which often maintains a low temperature and high humidity state. The determination of the end of icing is more sensitive to the intraday statistical characteristic variables of temperature and humidity. Therefore, when the temperature change value is greater than 5, and the humidity change value is greater than 15, the maximum intraday temperature T_max is greater than the icing temperature threshold T_tsh, or when the intraday humidity change value H_in_dif is greater than 25, it is determined that D j D i The corresponding ice cover end day.

[0095] Furthermore, since there will be multiple different ice start dates and corresponding ice end dates after the ice start date and ice end date are determined, the ice time periods formed by them may overlap; for the sake of simplicity in subsequent calculations, the final ice time period is obtained by merging the overlapping time periods. Specifically, the different ice start dates SD i But the same ice cover ends on SD j The corresponding icing period P ij Merge and keep only the smallest ice start date SD i The corresponding P ij .

[0096] As an exemplary embodiment, after detecting the ice end date based on the numerical characteristics or change characteristics of the serial data after the ice start date to obtain the ice end date matching each of the ice start dates, and before determining multiple overlapping ice time periods corresponding to the ice start date that overlap the ice start acquisition period based on the numerical characteristics of the historical meteorological data of each of the ice start dates and the corresponding ice end dates, the training method of the wind farm ice power prediction model further includes: screening the ice start dates based on the overlapping ice time periods to obtain a target ice time period with the minimum ice start date; and using the target ice time period as the ice time period.

[0097] Furthermore, after obtaining the ice cover start date and the ice cover end date, the ice cover start time corresponding to the ice cover start date and the ice cover end time corresponding to the ice cover end date are determined based on the numerical characteristics of the historical meteorological data of each of the ice cover start dates and each of the corresponding ice cover end dates.

[0098] Specifically, in one embodiment, for the serial data corresponding to the day when ice cover starts, the time point of the minimum temperature and the time point of the maximum humidity in the serial data are found, and then the larger time point between them is taken as the time when ice cover starts; for the serial data corresponding to the day when ice cover ends, the time point of the maximum temperature and the time point of the minimum humidity in the serial data are found, and then the smaller time point between them is taken as the time when ice cover ends.

[0099] Furthermore, after obtaining the icing time period, the icing time period is fine-tuned; since the calculation results of the icing start time and the icing end time are relatively vague and rough, in the present invention, a short-time sliding window is designed to fine-tune the icing start time and the icing end time.

[0100] Based on this, as an exemplary embodiment, the numerical characteristics of the historical meteorological data of each of the ice-covering start dates and each of the corresponding ice-covering end dates are used to determine the ice-covering start time corresponding to the ice-covering start date and the ice-covering end time corresponding to the ice-covering end date, including: sliding backtracking and sliding forward the sequence data corresponding to the ice-covering start time and the ice-covering end time according to a second preset sliding window with a second preset sliding step size to obtain multiple first sub-ice-covering sequence data and second sub-ice-covering sequence data; wherein the window length of the second preset sliding window is smaller than the window length of the first preset sliding window; traversing and filtering the first sub-ice-covering sequence data to obtain target first sub-ice-covering sequence data whose temperature values ​​are all smaller than the ice-covering temperature threshold and whose humidity values ​​are all greater than the preset humidity value; traversing and filtering the second sub-ice-covering sequence data to obtain second target sub-ice-covering sequence data whose temperature values ​​are all greater than the ice-covering temperature threshold; taking the start time of the target first sub-ice-covering sequence data as the ice-covering start time; and taking the start time of the second target sub-ice-covering sequence data as the ice-covering end time.

[0101] Specifically, when fine-tuning the ice onset time, a search is performed from past time points in each backtracking window based on the ice onset time. When the temperature of all points in the window is less than the ice temperature threshold T_tsh and the average humidity of all points in the window is greater than 75%, the time corresponding to the first point in the window is calculated as the new ice onset time. Furthermore, to refine the ice onset process, the ice onset time is calculated as T0, which represents the moment when ice first forms on the wind farm. Based on T0, w second preset sliding steps are deduced into the future to obtain time T1, which represents the ice onset time for the entire wind farm; where w is a positive integer.

[0102] When fine-tuning the start time of icing, it is difficult to estimate the relatively accurate end time by predicting meteorological conditions for different wind farms due to factors such as regional climate. In the present invention, the same second preset sliding window is used to fine-tune the end time in accordance with the same approach as refining the start time of icing. For the ice melting stage, the importance of humidity characteristics is greatly reduced, while the temperature conditions are extremely harsh. In the process of exploring future time points at the end of icing, the condition is that the temperature of all points in the window jumps out of the threshold influence range.

[0103] An embodiment of the present invention also provides a computer device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus, the memory is used to store computer programs; the processor is used to execute the method in any of the above embodiments by running the computer program stored in the memory.

[0104] Figure 2 is a structural block diagram of an optional computer device according to an embodiment of the present application, such as Figure 2 As shown, it includes a processor 10, a communication interface 20, a memory 30 and a communication bus 40, wherein the processor 10, the communication interface 20 and the memory 30 communicate with each other through the communication bus 40, wherein,

[0105] Memory 30, for storing computer programs;

[0106] The processor 10 is configured to implement the method of any of the above embodiments when executing the computer program stored in the memory 30 .

[0107] Optionally, in this embodiment, the communication bus may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus. The communication bus may be divided into an address bus, a data bus, a control bus, and the like. For ease of representation, Figure 2 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0108] The communication interface is used for communication between the above-mentioned computer device and other devices.

[0109] The memory may include RAM, or may include non-volatile memory, such as at least one disk memory. Alternatively, the memory may also be at least one storage device located away from the aforementioned processor.

[0110] The above-mentioned processor can be a general-purpose processor, which can include but is not limited to: CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be DSP (Digital Signal Processing), ASIC (Application Specific Integrated Circuit), FPGA (Field-Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0111] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and this embodiment will not be described in detail here.

[0112] It can be understood by those skilled in the art that Figure 2 The structure shown is for illustration only. The device for implementing any one of the methods in the above embodiments may be a terminal device, which may be a smart phone (such as an Android phone, an IOS phone, etc.), a tablet computer, a PDA, a mobile Internet device (MID), a PAD, or other terminal devices. Figure 2 It does not limit the structure of the above electronic device. For example, the terminal device may also include Figure 2 More or fewer components (such as network interfaces, display devices, etc.) shown in, or with Figure 2 Different configurations shown.

[0113] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, which can include: a flash drive, ROM, RAM, a magnetic disk or an optical disk, etc.

[0114] As an exemplary embodiment, the present application further provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute any one of the method steps of the present embodiment when run.

[0115] Optionally, in this embodiment, the above-mentioned storage medium can be used to execute the program code of the method steps of the embodiment of the present application.

[0116] Optionally, in this embodiment, the above-mentioned storage medium may be located on at least one network device among the multiple network devices in the network shown in the above-mentioned embodiment.

[0117] Optionally, in this embodiment, the storage medium is configured to store data for executing the method in the above embodiment.

[0118] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, which will not be described in detail in this embodiment.

[0119] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media that can store program codes, such as a USB flash drive, a ROM, a RAM, a mobile hard disk, a magnetic disk, or an optical disk.

[0120] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0121] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned computer-readable storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing one or more computer devices (such as personal computers, servers, or network devices) to execute all or part of the steps of the method in the above embodiments.

[0122] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is merely a logical functional division. In actual implementation, there may be other division methods, such as combining or integrating multiple units or components into another system, or ignoring or not implementing some features. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interface, indirect coupling or communication connection of units or modules, and may be electrical or other forms.

[0123] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected based on actual needs to achieve the purpose of the solution provided in this embodiment.

[0124] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0125] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0126] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A training method for a wind farm icing power prediction model, characterized in that: The training method of the wind farm icing power prediction model includes: Obtain historical predicted power data, historical measured power data, and historical meteorological data of the target wind farm as feature data sets; Performing ice-cover screening and grouping on the feature data set to obtain an ice-cover label data set; wherein the ice-cover label data set includes multiple groups of sequence data grouped according to a first preset sliding window and a first preset sliding step size; Detecting the ice cover start date and ice cover end date of the ice cover label dataset based on the numerical features of the sequence data and / or the change features of adjacent sequence data to obtain an ice cover time period specifically includes: Performing ice cover start date detection on the ice cover label dataset based on the numerical features of the sequence data and / or the change features of adjacent sequence data to obtain the ice cover start date; Detecting the ice cover end date based on the numerical characteristics or change characteristics of the sequence data after the ice cover start date, and obtaining the ice cover end date that matches each of the ice cover start dates; Determine the ice cover start time corresponding to the ice cover start date and the ice cover end time corresponding to the ice cover end date based on numerical characteristics of historical meteorological data of each of the ice cover start date and the corresponding ice cover end date; Taking the ice covering start time and the ice covering end time as the start time and the end time of the ice covering time period, to obtain the ice covering time period; The ice-covered label dataset is filtered based on the ice-covered time period to obtain an ice-covered dataset; wherein the ice-covered dataset includes the historical meteorological data corresponding to the ice-covered time period and a regression label column, and the regression label column is determined by the historical predicted power data and the historical measured power data corresponding to the ice-covered time period; The icing dataset is input into a pre-built preset model for model training. During the model training process, the correspondence between the historical meteorological data in the icing time period and the regression label column is learned until the preset model converges to obtain a wind farm icing prediction model.

2. The training method for the wind farm icing power prediction model according to claim 1, characterized in that: The detecting of the ice cover start date on the ice cover label dataset based on the numerical features of the sequence data to obtain multiple ice cover start dates includes: Based on an ice-covering temperature threshold and an ice-covering humidity threshold, meteorological data in the sequence data are traversed and detected to obtain first target ice-covering start sequence data, whose minimum temperature is less than the ice-covering temperature threshold and whose maximum humidity is greater than the ice-covering humidity threshold; wherein, adjacent sequence data adjacent to the first target ice-covering start sequence data window has a minimum temperature less than the ice-covering temperature threshold and a maximum humidity greater than the ice-covering humidity threshold; The first date corresponding to the first target icing start sequence data is used as the icing start date.

3. The training method for the wind farm icing power prediction model according to claim 1, characterized in that: The detecting the ice cover start date of the ice cover label dataset based on the numerical features of the sequence data and the change features of the adjacent sequence data to obtain the ice cover start date includes: The sequence data is traversed to obtain second target icing start sequence data, wherein the sum of the minimum temperature value and the first preset temperature value is less than the icing temperature threshold and the maximum humidity value is greater than the icing humidity threshold; wherein the minimum temperature values ​​of adjacent sequence data adjacent to the second target icing start sequence data window decrease as the sequence is sorted; The second date corresponding to the second target icing start sequence data is used as the icing start date.

4. The training method for a wind farm icing power prediction model according to claim 1, wherein: The detecting of the ice cover start date on the ice cover label dataset based on the change characteristics of the sequence data to obtain multiple ice cover start dates includes: Calculating the corresponding differences between the temperature extreme value and the humidity extreme value of each sequence data and the temperature extreme value and the humidity extreme value of the adjacent sequence data preceding in time sequence, to obtain the maximum humidity change value, the minimum humidity change value, the maximum temperature change value, and the minimum temperature change value; Sequence data in which both the maximum humidity change value and the minimum humidity change value are greater than the preset humidity change value or both the maximum temperature change value and the minimum temperature change value are greater than the preset temperature change value are used as third target icing start sequence data; The third date corresponding to the third target icing start sequence data is used as the icing start date.

5. The method for training a wind farm icing power prediction model according to any one of claims 2 to 4, characterized in that: The training method of the wind farm icing power prediction model further includes: Traversing each of the ice-covering start dates, and obtaining the first minimum temperature value of the sequence data corresponding to the ice-covering start date; Acquire the second minimum temperature value of the adjacent sequence data whose windows are adjacent and whose time sequence is later than the ice cover start date; Eliminate the ice-covering start date whose second minimum temperature value is greater than the first minimum temperature value, and whose sum of the second minimum temperature value and the second preset temperature value is greater than the ice-covering temperature threshold.

6. The training method for a wind farm icing power prediction model according to claim 1, wherein: The detecting of the ice cover end date based on the numerical characteristics or change characteristics of the sequence data after the ice cover start date to obtain the ice cover end date matching each of the ice cover start dates includes: Obtaining the actual climate type of the target wind farm; When the actual climate type meets the first preset climate type, sequentially traversing the numerical characteristics of the sequence data after the ice-covering start date, obtaining the sequence data whose minimum temperature is greater than the ice-covering temperature threshold and / or whose maximum humidity is less than the ice-covering humidity threshold as the target ice-covering end sequence data; When the actual climate type satisfies the second preset climate type, the numerical characteristics and / or change characteristics of the sequence data after the ice-covering start date are sequentially traversed to obtain first target sequence data having a maximum temperature greater than the ice-covering temperature threshold, a temperature change value greater than a preset temperature change value, and a humidity change value greater than a preset humidity change value, or second target sequence data having a humidity change value greater than a preset humidity change value as the target ice-covering end sequence data; The date corresponding to the target icing end sequence data is used as the icing end date.

7. The training method for a wind farm icing power prediction model according to claim 1, wherein: The determining, based on numerical characteristics of historical meteorological data of each of the ice cover start dates and each of the corresponding ice cover end dates, of the ice cover start time corresponding to the ice cover start date and the ice cover end time corresponding to the ice cover end date includes: Performing sliding backtracking and sliding advancement on the sequence data corresponding to the icing start time and the icing end time according to a second preset sliding window with a second preset sliding step size, respectively, to obtain a plurality of first sub-icing sequence data and second sub-icing sequence data; wherein the window length of the second preset sliding window is smaller than the window length of the first preset sliding window; Traversing and screening the first sub-icing sequence data to obtain target first sub-icing sequence data whose temperature values ​​are all less than the icing temperature threshold and whose humidity values ​​are all greater than the preset humidity value; Traversing and screening the second sub-icing sequence data, finding the second target sub-icing sequence data whose temperature values ​​are all greater than the icing temperature threshold; Taking the start time of the target first sub-icing sequence data as the icing start time; The start time of the second target sub-icing sequence data is used as the icing end time.

8. The training method for a wind farm icing power prediction model according to claim 1, wherein: Before determining the icing start time corresponding to the icing start date and the icing end time corresponding to the icing end date based on the numerical characteristics of the historical meteorological data of each icing start date and the corresponding icing end date, the training method of the wind farm icing power prediction model further includes: Obtain multiple overlapping ice-covered time periods containing the same time interval; The overlapping icing time periods are merged based on the icing start dates of the overlapping icing time periods to obtain a target icing time period with the smallest icing start date as the icing time period.

9. A computer device comprising: The memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the training method for the wind farm icing power prediction model according to any one of claims 1 to 8 by executing the computer instructions.

Images