Model training method and device, electronic equipment and storage medium

Abstract

The invention discloses a model training method and device, electronic equipment and a storage medium, and the method comprises the steps: determining a static feature according to the historical operation data and historical meteorological data of a wind turbine generator, and determining a dynamic feature and an interaction feature according to the static feature of the wind turbine generator in a period time; according to the static characteristics, the dynamic characteristics and the interaction characteristics, determining icing always time corresponding to the wind turbine generator; inputting the icing data set, the virtual icing data set and the climate characteristics corresponding to the target wind turbine generator into a pre-training model to adjust encoder parameters of the pre-training model, and determining that training of the pre-training model is ended based on the icing always time to obtain an icing detection model; the icing data set comprises historical operation data and historical meteorological data of the target wind turbine generator, the virtual icing data set is a data set obtained by performing icing physical simulation on the target wind turbine generator, and the pre-training model is a model pre-trained based on historical data of a wind power plant to which a non-target wind turbine generator belongs.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of power technology, and in particular to a model training method, apparatus, electronic device and storage medium. Background Technology

[0002] As an important component of clean energy, the accuracy of wind power forecasting directly impacts grid dispatch efficiency and operational stability. During cold snaps and low temperatures, icing on wind turbine blades can lead to decreased aerodynamic performance, significantly reduced power output, and even shutdown.

[0003] Currently, icing detection for wind turbines relies solely on simple meteorological factors such as wind speed, temperature, and humidity, and is performed using models based on small sample scenarios. However, since icing is a low-probability event, models trained under these conditions have poor generalization ability, leading to significant biases in icing detection. Summary of the Invention

[0004] This invention provides a model training method, apparatus, electronic device, and storage medium, which increases the number of samples and makes the trained model more accurate.

[0005] In a first aspect, embodiments of the present invention provide a model training method, the method comprising:

[0006] Static characteristics are determined based on historical operating data and historical meteorological data of wind turbines, and dynamic and interactive characteristics are determined based on the static characteristics of wind turbines within a cycle time.

[0007] The duration of icing for wind turbines is determined based on static, dynamic, and interactive characteristics.

[0008] The icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine are input into the pre-trained model to adjust the encoder parameters of the pre-trained model. The training end of the pre-trained model is determined based on the icing start and end time, resulting in an icing detection model. The icing dataset includes historical operating data and historical meteorological data of the target wind turbine. The virtual icing dataset is a dataset integrating icing data obtained after performing icing physical simulation on the target wind turbine. The pre-trained model is a model obtained after pre-training based on historical data from wind farms that do not belong to the target wind turbine.

[0009] The model training method provided in this invention, after obtaining the wind turbine's operational and meteorological data, determines static characteristics and, based on these static characteristics over a period of time, determines dynamic and interactive characteristics. It utilizes simple collected data to analyze and expand the data, obtaining its dynamic and interactive properties. This not only provides accurate and diversified data for subsequent training of the icing detection model but also solves the problem of relying solely on simple meteorological elements for icing detection, laying the foundation for improving the accuracy and generalization ability of icing detection using diversified data. By using multiple datasets representing wind farms requiring the icing detection model as the training sample set and incorporating the climate characteristics corresponding to the target wind turbine, the final trained icing detection model becomes more adaptable to icing detection of all wind turbines in the target wind farm. Furthermore, by using a virtual icing dataset and simulating the data, more simulation data from all wind turbines in the target wind farm are added, further enhancing the generalization ability of the trained model for the wind turbines in that wind farm. This not only increases the number of samples but also improves the accuracy of the trained model.

[0010] Secondly, embodiments of the present invention also provide a model training apparatus, the apparatus comprising:

[0011] The feature determination module is used to determine static features based on historical operating data and historical meteorological data of the wind turbine, and to determine dynamic and interactive features based on the static features of the wind turbine within a cycle time.

[0012] The time determination module is used to determine the duration of icing for wind turbines based on static, dynamic, and interactive characteristics.

[0013] The training module is used to input the icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine into the pre-trained model to adjust the encoder parameters of the pre-trained model, and to determine the end of training of the pre-trained model based on the icing start and end time to obtain the icing detection model. The icing dataset includes the historical operating data and historical meteorological data of the target wind turbine, the virtual icing dataset is a dataset integrated from the icing data obtained after performing icing physical simulation on the target wind turbine, and the pre-trained model is a model obtained after pre-training based on historical data of wind farms that do not belong to the target wind turbine.

[0014] Thirdly, embodiments of the present invention also provide an electronic device, the electronic device comprising:

[0015] At least one processor; and

[0016] A memory that is communicatively connected to at least one processor; wherein,

[0017] The memory stores a computer program that can be executed by at least one processor, such that the at least one processor is able to perform the model training method of any embodiment of the present invention.

[0018] Fourthly, embodiments of the present invention also provide a computer-readable storage medium storing computer instructions that are used to cause a processor to execute the model training method of any embodiment of the present invention.

[0019] Fifthly, embodiments of the present invention also provide a computer program product, including a computer program that, when executed by a processor, implements the model training method of any embodiment of the present invention.

[0020] It should be noted that the aforementioned computer instructions may be stored, in whole or in part, on a computer-readable storage medium. This computer-readable storage medium may be packaged together with the processor of the model training device, or it may be packaged separately from the processor of the model training device; this application does not impose any limitations on this.

[0021] The descriptions of the second, third, fourth, and fifth aspects in this application can be referred to the detailed description of the first aspect; and the beneficial effects of the descriptions of the second, third, fourth, and fifth aspects can be referred to the analysis of the beneficial effects of the first aspect, which will not be repeated here.

[0022] In this application, the name of the aforementioned model training device does not limit the device or functional module itself. In actual implementation, these devices or functional modules may appear under other names. As long as the function of each device or functional module is similar to that of this application, it falls within the scope of the claims of this application and its equivalents.

[0023] These or other aspects of this application will become more readily apparent in the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 A schematic flowchart of the model training method provided in an embodiment of the present invention;

[0026] Figure 2 A flowchart illustrating another model training method provided in an embodiment of the present invention;

[0027] Figure 3 This is a schematic flowchart of a model training device provided in an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0030] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three situations: A exists alone, A and B exist simultaneously, and B exists alone.

[0031] The terms "first" and "second," etc., used in the specification and drawings of this application are used to distinguish different objects or to distinguish different treatments of the same object, rather than to describe a specific order of objects.

[0032] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus.

[0033] Before discussing the exemplary embodiments in more detail, it should be noted that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the figures. The process can correspond to a method, function, procedure, subroutine, subroutine, etc. Moreover, embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0034] It should be noted that in the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0035] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0036] Figure 1 This is a flowchart illustrating the model training method provided in an embodiment of the present invention. This embodiment is applicable to training an icing detection model. The method can be executed by a model training device, which can be implemented using software and / or hardware. In this embodiment, the model training device is located in an electronic device, which can be a computer or a server. Specifically, it includes the following steps:

[0037] S101. Determine the static characteristics based on the historical operating data and historical meteorological data of the wind turbine, and determine the dynamic and interactive characteristics based on the static characteristics of the wind turbine within the cycle time.

[0038] In this embodiment, the wind turbine refers to any wind turbine within any wind farm. Historical operating data refers to the operating data of the wind turbine during its historical operation. Historical meteorological data refers to the regional meteorological data of the wind farm to which the wind turbine belongs during historical periods. Static features are used to characterize the static properties of the data. The period can be preset or determined according to actual conditions. Dynamic features are used to characterize the dynamic properties of the data. Interactive features are used to characterize the interaction between data.

[0039] Specifically, historical operating data of multiple wind turbines and regional meteorological data of the wind farms to which these turbines belong during historical periods can be obtained, i.e., historical meteorological data, forming historical operating data and historical meteorological data for each wind turbine. Furthermore, based on the historical operating data and historical meteorological data of each wind turbine, the corresponding static characteristics can be determined. Additionally, since each wind turbine has its corresponding operating time and operating status, the cycle time can be determined based on this, and using the cycle time as a boundary, dynamic characteristics can be determined based on the corresponding static characteristics of the wind turbine, thereby determining the interaction characteristics.

[0040] In this embodiment, after obtaining the wind turbine's operating data and meteorological data, not only are the static characteristics determined, but also the dynamic and interactive characteristics are determined based on the static characteristics of the wind turbine within a cycle time. This enables the analysis and expansion of data using simple collected data to obtain the dynamic and interactive characteristics of the data. This not only provides accurate and diversified data for subsequent training of the icing detection model, but also solves the problem of relying solely on simple meteorological elements for icing detection. It provides a foundation for improving the accuracy and generalization ability of icing detection by using diversified data.

[0041] S102. Determine the icing duration of the wind turbine based on static characteristics, dynamic characteristics, and interactive characteristics.

[0042] The icing start and end times are the calculated start and end times of icing during the historical operation of the wind turbine.

[0043] Specifically, since the icing duration can be used to determine the accuracy of the output (predicted icing time) of the subsequently trained model, the accurate icing duration for wind turbines can be calculated based on static, dynamic, and interactive features that reflect historical operational and meteorological data. That is, since static features most directly reflect whether operational and meteorological data can cause wind turbine icing, the initial icing time range for wind turbines can be determined first using static features. Furthermore, dynamic features reflect the operational changes of wind turbines during operation; therefore, the initial icing time range can be adjusted based on dynamic features. Finally, since interactive features reflect the correlation between pairs of parameters, the adjusted initial icing time range can be verified based on interactive features, ultimately yielding the corresponding icing duration for the wind turbines.

[0044] In this embodiment, not only were static features, dynamic features, and interactive features determined based on the collected data, expanding the wind turbine's operating data and meteorological data, but the icing duration corresponding to the wind turbine was also calculated based on these features, providing accurate comparative verification indicators for subsequent training of the icing detection model.

[0045] S103. Input the icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine into the pre-trained model to adjust the encoder parameters of the pre-trained model, and determine the end of training of the pre-trained model based on the icing start and end time to obtain the icing detection model.

[0046] The icing dataset includes historical operating data and historical meteorological data of the target wind turbine. The virtual icing dataset is an integrated dataset of icing data obtained after performing icing physical simulation on the target wind turbine. The pre-trained model is a model obtained after pre-training based on historical data from wind farms other than the target wind turbine. In this embodiment, the icing dataset includes historical operating data and historical meteorological data of all wind turbines in the wind farm to which the target wind turbine belongs. The virtual icing dataset is a collection of virtual icing data generated by simulating the icing physical process using a simulation model based on meteorological data from the wind farm to which the target wind turbine belongs. The target wind turbine is any wind turbine in a wind farm. Climate characteristics are used to characterize the climate characteristics of a region. Optionally, the wind farm to which the target wind turbine belongs can be the wind farm to which the icing detection model needs to be applied in this embodiment.

[0047] Specifically, in this embodiment, three sample sets will be prepared to train the model. The first set is a general sample set, which includes historical data from wind farms other than the target wind turbine. This sample set includes historical operating data and historical meteorological data of all wind turbines in the wind farms other than the target wind turbine, and optionally also includes icing time records. The data in the general sample set can be data from icing scenarios or non-icing scenarios. Based on the general sample set, the basic model can be pre-trained. The goal of the pre-training is to minimize the mean squared error. After the pre-training achieves the goal, the pre-trained model is obtained. The second set of sample sets is an icing dataset, which includes historical operating data and historical meteorological data of all wind turbines in the wind farm where the target wind turbine belongs, but the data in the icing dataset is all from icing scenarios. The third sample set is a virtual icing dataset. The icing data in this dataset is derived from meteorological data of the wind farm to which the target wind turbine belongs, simulating the icing physical processes of different wind turbines within the farm. Based on the icing dataset and the virtual icing dataset, after obtaining the pre-trained model, both are used as sample sets to input into the model for further training, adjusting the encoder parameters of the pre-trained model. After each training iteration, the model outputs a predicted icing time based on the input data. The predicted icing time can then be compared with the actual icing time to determine if the model's prediction accuracy meets the requirements. Finally, once the model's prediction accuracy meets the requirements, an icing detection model that accurately adapts to the wind farm to which the target wind turbine belongs can be obtained.

[0048] In this embodiment, multiple datasets representing wind farms requiring icing detection models are used as the training set for the model. Climate features corresponding to the target wind turbines are also incorporated, making the final trained icing detection model more suitable for icing detection of all wind turbines in the target wind farm. Furthermore, a virtual icing dataset is used, and simulation data of all wind turbines in the target wind farm is added through simulation. This further enhances the generalization ability of the trained model to wind turbines in that wind farm, increasing both the number of samples and the accuracy of the trained model.

[0049] The model training method provided in this invention, after obtaining the wind turbine's operational and meteorological data, determines static characteristics and, based on these static characteristics over a period of time, determines dynamic and interactive characteristics. It utilizes simple collected data to analyze and expand the data, obtaining its dynamic and interactive properties. This not only provides accurate and diversified data for subsequent training of the icing detection model but also solves the problem of relying solely on simple meteorological elements for icing detection, laying the foundation for improving the accuracy and generalization ability of icing detection using diversified data. By using multiple datasets representing wind farms requiring the icing detection model as the training sample set and incorporating the climate characteristics corresponding to the target wind turbine, the final trained icing detection model becomes more adaptable to icing detection of all wind turbines in the target wind farm. Furthermore, by using a virtual icing dataset and simulating the data, more simulation data from all wind turbines in the target wind farm are added, further enhancing the generalization ability of the trained model for the wind turbines in that wind farm. This not only increases the number of samples but also improves the accuracy of the trained model.

[0050] Figure 2 This is a flowchart illustrating another model training method provided by an embodiment of the present invention. This embodiment elaborates on the steps of determining dynamic features and interaction features, determining the icing duration, and obtaining the icing detection model based on the above embodiments. In this embodiment, the method may include:

[0051] S201. Determine the historical operating status of the wind turbine based on historical operating data, and determine the static operating characteristics corresponding to the cycle time and historical operating status based on the historical operating status.

[0052] In this embodiment, historical operating data includes historical measured power data, historical predicted power data (which may be obtained during model training based on other methods or this method), and shutdown / maintenance records, etc. Historical operating status is used to reflect the operating status of the wind turbine unit in the historical process.

[0053] Specifically, since historical operating data includes shutdown / maintenance records, the historical operating status of the wind turbine can be determined based on this, such as when it started operating, when it stopped, and when it underwent maintenance. Furthermore, one operating cycle of the wind turbine can be considered a period, and the static operating characteristics corresponding to the historical operating status within that period can be determined. For example, if one operating cycle of a wind turbine is from 10:00 to 19:00, the maximum and minimum power values ​​can be determined based on the measured power data of the wind turbine from 10:00 to 19:00, and these maximum and minimum power values ​​can be used as static operating characteristics. Optionally, the period in this embodiment can also be set based on wind turbine operating experience.

[0054] It is worth noting that the examples given in this embodiment are all optional solutions, and the specific methods for determining the cycle time and static operating characteristics can be selected and changed according to the actual situation. For example, a cycle time can be defined as the wind turbine operating under a preset cycle. Static operating characteristics can also be median, average, or standard values, etc.

[0055] S202. Determine the static meteorological characteristics corresponding to the historical meteorological data based on the historical meteorological data, determine the dynamic characteristics based on the static operating characteristics and static meteorological characteristics within the period, and determine the interactive characteristics based on the dynamic characteristics.

[0056] In this embodiment, historical meteorological data includes historical temperature, historical humidity, historical wind speed, historical air pressure, historical precipitation, historical radiation, and historical wind direction.

[0057] Specifically, the corresponding static meteorological characteristics can be determined based on different historical meteorological data. For example, the highest and lowest historical temperatures can be determined based on historical temperatures; the highest and lowest historical humidity can be determined based on historical humidity; and other static meteorological characteristics can be obtained similarly. Optionally, the highest and lowest values ​​can also be determined in combination with the aforementioned periodic time, for example, determining the highest and lowest values ​​within a certain period.

[0058] Furthermore, dynamic characteristics are determined based on static operational characteristics and static meteorological characteristics within the period, and interactive characteristics are determined based on the dynamic characteristics, including:

[0059] (i) Determine the dynamic operating characteristics based on the static operating characteristics within the period, and determine the dynamic meteorological characteristics based on the static meteorological characteristics within the period.

[0060] Specifically, after obtaining the static operating characteristics, the changing trend of the static operating characteristics can be fitted based on the static operating characteristics within a period. For example, the power change rate within a period can be determined based on the minimum, median, and maximum power values ​​within a period. Optionally, dynamic operating characteristics can also be determined directly based on historical operating data, and dynamic meteorological characteristics can be determined based on historical meteorological data. For example, historical operating data can be grouped, cleaned, missing value imputation, outlier removal, and data annotation to obtain processed historical operating data; the historical operating data can be sorted according to the acquisition time corresponding to it, and the change rate of the historical operating data can be determined based on this. The change rate of historical meteorological data can also be obtained in the same way. Finally, the change rate can be used as a dynamic feature; such as the temperature change rate, humidity change rate, and power change rate. Optionally, in addition to the change rate, the dynamic feature can also be a multi-order reciprocal feature.

[0061] (ii) Determine the interactive operation characteristics based on the dynamic operation characteristics, determine the interactive meteorological characteristics based on the dynamic meteorological characteristics, and determine the overall interactive characteristics based on the interactive operation characteristics and the interactive meteorological characteristics.

[0062] Specifically, the interaction relationship between at least two characteristic factors in dynamic operation characteristics can be determined. Similarly, the interaction relationship between at least two characteristic factors in dynamic meteorological characteristics can also be determined. For example, the correlation between the rate of change of wind turbine speed and the rate of change of power can be determined based on the two characteristic factors of wind turbine speed and power. The correlation between the rate of change of temperature and the rate of change of humidity can also be determined based on the two characteristic factors of temperature and humidity. Furthermore, in addition to correlation, the covariance matrix, causal relationship, and regression analysis of at least two characteristic factors can also be determined.

[0063] Furthermore, after determining the interactive operational characteristics and interactive meteorological characteristics, the overall interactive characteristics can also be determined based on these two factors. For example, based on the rate of change of temperature and humidity, and the rate of change of speed and power, the speed and power can be jointly determined as the temperature and humidity change.

[0064] It is worth noting that the above are merely illustrative examples. In practice, interactive operational characteristics, interactive meteorological characteristics, and interactive overall characteristics can be determined based on specific factors, actual needs, and at least two factors, and using different methods.

[0065] In this embodiment, by extracting multidimensional dynamic features, the static information, change information and interactive influence relationship between features can be determined, achieving comprehensive and accurate feature extraction, and providing a more accurate basis for determining the duration of icing.

[0066] S203. Determine the initial icing duration based on static characteristics and static characteristic thresholds.

[0067] Specifically, since static features can intuitively display the operation of wind turbines and current weather conditions, the icing time can be determined based on these features. For example, icing can begin when the wind turbine power is below a preset power threshold; it can also begin when the temperature is below a preset temperature and the humidity is above a preset humidity; and it can begin when the highest temperature is below the average temperature and the lowest humidity is above the average humidity. Similarly, the icing end time can be determined based on this. Finally, the initial icing duration can be determined based on the static features and their corresponding preset thresholds (static feature thresholds).

[0068] S204. Perform secondary clustering on dynamic features and interaction features to obtain double clustering results, and determine the target icing duration based on the double clustering results and the initial icing duration.

[0069] In this embodiment, the secondary clustering involves clustering the features using two different clustering methods. The biclustering result is the two clustering results obtained after clustering using the two different methods.

[0070] Specifically, since dynamic features reflect the dynamic characteristics of each feature factor, and interaction features reflect the changing interaction characteristics between several feature factors, secondary clustering can be performed on both dynamic and interaction features to obtain a bi-clustering result. Furthermore, since different clustering methods have their own clustering characteristics, the bi-clustering result can be analyzed based on the characteristics of each clustering method. Then, based on the analysis, the initial icing duration can be fine-tuned to obtain the target icing duration.

[0071] For example, a second clustering is performed on the dynamic features and interaction features to obtain a bi-clustering result, and the target icing duration is determined based on the bi-clustering result and the initial icing duration, including:

[0072] (a) Cluster the dynamic features and interaction features using the first clustering method to obtain the first clustering result.

[0073] Specifically, by clustering dynamic features and interaction features based on the first clustering method, the first clustering result can be obtained. For example, if the first clustering method is mean-based clustering, the first clustering result can be to find a cluster with high negative power deviation, temperature near freezing temperature and high humidity. All time points corresponding to this cluster are the possible time points of icing.

[0074] (ii) Use the second clustering method to cluster the dynamic features and interaction features to obtain the second clustering results.

[0075] The first clustering method differs from the second clustering method.

[0076] Specifically, a second clustering method is then used to cluster the dynamic features and interaction features to obtain the second clustering results. For example, if the second clustering method is a density-based clustering algorithm, the second clustering results are the points marked as noise or outliers that meet the icing regulation, which are the icing points, and the continuous time period corresponding to these icing points is the icing duration.

[0077] (iii) Determine the target icing duration based on the first clustering result, the second clustering result, and the initial icing duration.

[0078] Specifically, after obtaining the first clustering result and the second clustering result, it can be determined whether the icing duration of the two results is highly consistent.

[0079] If they match, the intersection or union of the two results' icing durations can be directly used as the correction value. This correction value is then compared with the initial icing duration to adjust the initial icing duration, ultimately determining the target icing duration. For example, the intersection and union can be compared with the initial icing duration, and the one with higher overlap can be selected as the correction value to adjust the initial icing duration. If they do not match, result fusion is required. For example, a high-confidence set can be established to correct for temporal continuity, ultimately obtaining an accurate icing correction value, which is then used to adjust the initial icing duration to obtain the target icing duration.

[0080] In this embodiment, firstly, the possible range of initial icing duration is quickly and accurately determined using static features that intuitively reflect the wind turbine's operating status and meteorological conditions. Secondly, different clustering methods are used with dynamic features and interaction features that are highly correlated to obtain different clustering results, enabling the determination of icing duration deviations from multiple perspectives. Finally, based on the two clustering results and the initial icing duration, the accurate target icing duration is determined. This not only integrates the clustering characteristics of the two methods, making the final icing duration more accurate, but also avoids large errors in the final icing duration caused by significant defects in the clustering results of a single method, thus achieving accurate determination of the target icing duration.

[0081] S205. Use climate features as climate labels in the icing dataset.

[0082] Climate characteristics include the climate zone or climate type to which it belongs. For example, humid climate, temperate climate, or high humidity and low temperature, dry and cold climate.

[0083] Specifically, since different regions have their own corresponding climate characteristics, in order to make the final icing detection model more targeted, the climate characteristics of the wind farm to which the target wind turbine belongs can be used as the climate label in the icing dataset.

[0084] In this embodiment, climate features are used as climate labels in the icing dataset, which can better highlight the climate conditions of the areas where the icing detection model needs to be used, and provide a foundation for the icing detection model trained later to be more suitable for the areas where it is used.

[0085] S206. Freeze the front-layer encoder parameters of the pre-trained model and input the icing dataset with climate labels into the pre-trained model to adjust the unfrozen back-layer encoder and decoder parameters of the pre-trained model to obtain the initial detection model.

[0086] In this embodiment, the frozen front-layer encoder parameters are the front-layer encoder parameters of the model set according to the specific application environment.

[0087] Specifically, the parameters of the first few layers of the pre-trained model are frozen, and the icing dataset with climate labels is input into the pre-trained model for training. During the training process, the parameters of the later layers of the pre-trained model's encoder and decoder are fine-tuned to obtain the initial detection model after parameter tuning.

[0088] In this embodiment, by freezing some encoder parameters, a pre-trained model is trained based on the icing dataset to fine-tune the unfrozen model parameters of the pre-trained model, thereby making the trained initial detection model more suitable for the specific climate characteristics and operating status of the wind farm to which the target wind turbine belongs.

[0089] S207. Simulate the icing physical process of the wind farm to which the target wind turbine belongs based on the icing dataset, obtain icing increase simulation data and icing melt simulation data, and use the icing increase simulation data and icing melt simulation data as a virtual icing dataset.

[0090] Specifically, after obtaining the initial detection model, the meteorological data of the wind farm to which the target wind turbine belongs in the icing dataset can be used, such as the historical meteorological data corresponding to each wind turbine. The icing physical process can be simulated using the icing enhancement model and the icing melting model in the simulation model to obtain icing enhancement simulation data and icing melting simulation data. The icing enhancement simulation data and the icing melting simulation data are then used as a virtual icing dataset.

[0091] For example, the ice-enhancing model can be represented by the following formula:

[0092] ;

[0093] Wherein, △M ice ρ is the icing increment; k1 is an empirical coefficient; wWhere is the density of liquid water; v is the wind speed; e is the density of liquid water. s ρ is the saturated vapor pressure; e is the actual vapor pressure; T c T represents the critical temperature for icing; T represents the actual temperature.

[0094] S208. Input the virtual icing dataset into the initial icing detection model to adjust the encoder parameters of the initial icing detection model and obtain the target detection model.

[0095] Specifically, similar to S206, a virtual icing dataset can be input into the initial icing detection model, and the encoder parameters of the initial icing detection model can be further adjusted to obtain the target detection model.

[0096] In this embodiment, in addition to the dataset obtained from actual data collection, a virtual icing dataset obtained by simulating the physical process of icing was also added to train the model. This further expanded the training set of the model, increased the sample size, and improved the accuracy of the model in detecting icing in the wind farm to which the target wind turbine belongs.

[0097] S209. Input the test set into the pre-trained model with adjusted encoder parameters to obtain the predicted icing time and predicted wind turbine power corresponding to the test set samples.

[0098] In this embodiment, the test set is used to verify whether the model training process has ended. The pre-trained model with adjusted encoder parameters in this embodiment can be an initial detection model or an object detection model.

[0099] Specifically, the test set samples are input into the target detection model to obtain the predicted icing time and predicted wind turbine power for each sample in the test set. Optionally, before determining the initial detection model, another test set can be used to test the pre-trained model, and the initial detection model can be determined after confirming that its accuracy meets the standards by following the steps below.

[0100] S210. Determine the icing time prediction result based on the icing start and end time and the predicted icing time of the wind turbines in the test set sample, and determine the power prediction result based on the historical operating data of the wind turbines in the test set sample and the predicted wind turbine power.

[0101] Specifically, the icing time prediction result can be determined based on the icing start and end times previously determined for each sample in the validation sample set and the predicted icing time output by the model. Simultaneously, the power prediction result can be determined based on the historical turbine power and predicted turbine power from the historical operating data corresponding to each sample's wind turbine.

[0102] S211. Determine whether the icing time prediction result and the power prediction result both meet the standard; if both meet the standard, proceed to S212; if at least one does not meet the standard, return to proceed to S205.

[0103] Specifically, both the icing time prediction result and the power prediction result can be expressed as accuracy or error rate. Therefore, time limits corresponding to the icing time prediction result and power limits corresponding to the power prediction result can be preset. Then, it is determined whether the icing time prediction result is less than or equal to the time limit, and whether the power prediction result is less than or equal to the power limit. If the icing time prediction result is greater than the time limit, and / or the power prediction result is greater than the power limit, it indicates that the model training has not met the standards. The process can return to S205 to continue adjusting the encoder and decoder parameters until they meet the standards.

[0104] S212. Determine that the training of the pre-trained model has ended, and designate the pre-trained model as the icing detection model.

[0105] Specifically, if it is determined that the icing time prediction result is less than or equal to the time result limit and the power prediction result is less than or equal to the power result limit, then the training of the pre-trained model can be determined to be completed, and the pre-trained model can be determined as the icing detection model.

[0106] Optionally, in this embodiment, after inputting the training set into the model and obtaining the "predicted wind turbine power" output by the model, this can be used as a label in the icing dataset as a regression label to reflect the degree of power loss. Simultaneously, each predicted wind turbine power in the icing dataset represents the historical predicted power.

[0107] Optionally, the pre-trained model in this embodiment can adopt a self-attention mechanism architecture and be pre-trained using a multi-wind farm dataset. This model includes a multi-layer encoder and decoder structure; the encoder is used to extract the temporal patterns of icing features, and the decoder is used to generate power prediction sequences. Simultaneously, each layer of the encoder and decoder includes a multi-head self-attention module and a feedforward neural network, equipped with residual connections and layer normalization to improve training stability. Further, during the pre-training stage, the model mainly learns the mapping relationship between general meteorological features and power output, aiming to minimize the mean square error. The optimizer uses "correct weight decay" to prevent overfitting. Further, during model training, a temporal attention module is introduced to address the temporal characteristics of icing time. By learning the weights at different time steps, the model's attention to key features of the start and end times of icing (such as sudden temperature drops and humidity surges) is enhanced. Further, during the training process optimization stage, the aforementioned optimizer, combined with cosine annealing learning rate scheduling, dynamically adjusts the learning rate during training to improve model convergence speed. An early stopping mechanism is employed, with a validation set set. Training is stopped when the validation set loss fails to decrease for several consecutive training epochs to prevent overfitting. Mini-batch training is used, leveraging the GPU for parallel computation to accelerate training, with each training session lasting less than 2 hours.

[0108] Optionally, a subset matching the regional climate characteristics can be selected from the icing dataset based on the wind farm's climate type (identified through historical data clustering or geographic information) to fine-tune the decoder parameters of the pre-trained model. During fine-tuning, the icing detection threshold is dynamically adjusted to ensure the model adapts to the regional characteristics and generates adapted prediction results.

[0109] Optionally, in the final model accuracy evaluation, indicators such as the regional power grid dual-rules standard, root mean square error, and normalized root mean square error can be combined to evaluate the deviation between the model output and the actual value.

[0110] Figure 3 This is a schematic flowchart of a model training device provided in an embodiment of the present invention, as shown below. Figure 3 As shown, the device includes:

[0111] The feature determination module 301 is used to determine static features based on the historical operating data and historical meteorological data of the wind turbine, and to determine dynamic and interactive features based on the static features of the wind turbine within a cycle time.

[0112] The time determination module 302 is used to determine the icing duration of the wind turbine based on static features, dynamic features, and interactive features.

[0113] The training module 303 is used to input the icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine into the pre-training model to adjust the encoder parameters of the pre-training model, and to determine the end of training of the pre-training model based on the icing start and end time to obtain the icing detection model. The icing dataset includes the historical operating data and historical meteorological data of the target wind turbine, the virtual icing dataset is a dataset of icing data integrated after performing icing physical simulation on the target wind turbine, and the pre-training model is a model obtained after pre-training based on historical data of wind farms that do not belong to the target wind turbine.

[0114] Based on the above embodiments, the feature determination module 301 is specifically used for:

[0115] Based on historical operating data, the historical operating status of the wind turbine is determined, and the static operating characteristics corresponding to the period and the historical operating status are determined based on the historical operating status. Based on historical meteorological data, the static meteorological characteristics corresponding to the historical meteorological data are determined, and the dynamic characteristics are determined based on the static operating characteristics and static meteorological characteristics within the period. The interactive characteristics are determined based on the dynamic characteristics.

[0116] Based on the above embodiments, dynamic characteristics are determined according to the static operating characteristics and static meteorological characteristics within the period, and interactive characteristics are determined according to the dynamic characteristics. The feature determination module 301 is specifically used for:

[0117] Dynamic operating characteristics are determined based on static operating characteristics within a period of time, and dynamic meteorological characteristics are determined based on static meteorological characteristics within a period of time; interactive operating characteristics are determined based on dynamic operating characteristics, interactive meteorological characteristics are determined based on dynamic meteorological characteristics, and interactive overall characteristics are determined based on interactive operating characteristics and interactive meteorological characteristics.

[0118] Based on the above embodiments, the time determination module 302 is specifically used for:

[0119] The initial icing duration is determined based on static features and static feature thresholds; secondary clustering is performed on dynamic features and interactive features to obtain bi-clustering results, and the target icing duration is determined based on the bi-clustering results and the initial icing duration.

[0120] Based on the above embodiments, a second clustering is performed on the dynamic features and interaction features to obtain a double clustering result. The target icing duration is then determined based on the double clustering result and the initial icing duration. The time determination module 302 is specifically used for:

[0121] The dynamic features and interaction features are clustered using the first clustering method to obtain the first clustering result; the dynamic features and interaction features are clustered using the second clustering method to obtain the second clustering result; the target icing duration is determined based on the first clustering result, the second clustering result, and the initial icing duration.

[0122] Based on the above embodiments, the training module 303 is specifically used for:

[0123] Climate features are used as climate labels in the icing dataset. The front-layer encoder parameters of the pre-trained model are frozen, and the icing dataset with climate labels is input into the pre-trained model to adjust the unfrozen back-layer encoder and decoder parameters of the pre-trained model, thus obtaining the initial detection model. The icing physical process of the wind farm to which the target wind turbine belongs is simulated based on the icing dataset to obtain icing enhancement simulation data and icing melting simulation data, which are used as virtual icing datasets. The virtual icing dataset is input into the initial icing detection model to adjust the encoder parameters of the initial icing detection model, thus obtaining the target detection model.

[0124] Based on the above embodiments, the training module 303 is specifically used for:

[0125] The test set is input into the pre-trained model with adjusted encoder parameters to obtain the predicted icing time and predicted wind turbine power corresponding to the test set samples. The icing time prediction result is determined based on the icing start and end times and predicted icing times of the wind turbines in the test set samples, and the power prediction result is determined based on the historical operating data and predicted wind turbine power of the wind turbines in the test set samples. If both the icing time prediction result and the power prediction result meet the standards, the training of the pre-trained model is considered complete, and the pre-trained model is designated as the icing detection model. If at least one of the icing time prediction result and the power prediction result fails to meet the standards, the process returns to the step of inputting the icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine into the pre-trained model to adjust the encoder parameters of the pre-trained model.

[0126] The model training apparatus provided in this embodiment of the invention can execute the model training method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0127] It is worth noting that in the embodiments of the above-mentioned model training device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the scope of protection of the present invention.

[0128] Figure 4This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 4 A block diagram is shown of an exemplary electronic device 11 suitable for implementing embodiments of the present invention. Figure 4 The electronic device 11 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0129] like Figure 4 As shown, the electronic device 11 is represented in the form of a general-purpose computing electronic device. The components of the electronic device 11 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0130] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0131] Electronic device 11 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 11, including volatile and non-volatile media, removable and non-removable media.

[0132] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Electronic device 11 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (… Figure 4 Not shown; usually referred to as a "hard drive"). Although Figure 4 As not shown, disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. System memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0133] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0134] Electronic device 11 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with electronic device 11, and / or with any device that enables electronic device 11 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 22. Furthermore, electronic device 11 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 20. Figure 4 As shown, network adapter 20 communicates with other modules of electronic device 11 via bus 18. It should be understood that, although... Figure 4 As not shown, other hardware and / or software modules may be used in conjunction with electronic device 11, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0135] The processing unit 16 executes various functional applications and page displays by running programs stored in the system memory 28, such as implementing the model training method provided in this embodiment. Of course, those skilled in the art will understand that the processor can also implement the technical solutions of the model training methods provided in any embodiment of this invention.

[0136] This invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements, for example, the model training method provided in this invention. The computer storage medium of this invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0137] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0138] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0139] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the model training method provided in any embodiment of this invention. The computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0140] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0141] Furthermore, the acquisition, storage, use, and processing of data in the technical solution of this invention all comply with the relevant provisions of national laws and regulations.

[0142] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A model training method, characterized in that, The method includes: Static characteristics are determined based on historical operating data and historical meteorological data of the wind turbine, and dynamic and interactive characteristics are determined based on the static characteristics of the wind turbine within a period of time. The icing duration of the wind turbine is determined based on the static features, the dynamic features, and the interactive features. The icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine are input into the pre-trained model to adjust the encoder parameters of the pre-trained model. The training of the pre-trained model ends based on the icing start and end time, resulting in an icing detection model. The icing dataset includes historical operating data and historical meteorological data of the target wind turbine. The virtual icing dataset is an integrated dataset of icing data obtained after performing icing physical simulation on the target wind turbine. The pre-trained model is a model obtained after pre-training based on historical data from wind farms that do not belong to the target wind turbine.

2. The method according to claim 1, characterized in that, The process of determining static characteristics based on historical operating data and historical meteorological data of the wind turbine, and determining dynamic and interactive characteristics based on the static characteristics of the wind turbine within a period of time, includes: The historical operating status of the wind turbine is determined based on the historical operating data, and the static operating characteristics corresponding to the cycle time and the historical operating status are determined based on the historical operating status. Based on the historical meteorological data, the static meteorological characteristics corresponding to the historical meteorological data are determined; based on the static operational characteristics and the static meteorological characteristics within a period of time, the dynamic characteristics are determined; and based on the dynamic characteristics, the interactive characteristics are determined.

3. The method according to claim 2, characterized in that, The step of determining the dynamic characteristics based on the static operational characteristics and the static meteorological characteristics within a period of time, and determining the interaction characteristics based on the dynamic characteristics, includes: Dynamic operating characteristics are determined based on the static operating characteristics within the period, and dynamic meteorological characteristics are determined based on the static meteorological characteristics within the period. Based on the dynamic operational characteristics, interactive operational characteristics are determined; based on the dynamic meteorological characteristics, interactive meteorological characteristics are determined; and based on the interactive operational characteristics and the interactive meteorological characteristics, overall interactive characteristics are determined.

4. The method according to claim 1, characterized in that, Determining the icing duration of the wind turbine based on the static features, the dynamic features, and the interactive features includes: The initial icing duration is determined based on the static features and the static feature threshold. The dynamic features and the interaction features are subjected to secondary clustering to obtain a double clustering result, and the target icing duration is determined based on the double clustering result and the initial icing duration.

5. The method according to claim 4, characterized in that, The step of performing secondary clustering on the dynamic features and the interaction features to obtain a bi-clustering result, and determining the target icing duration based on the bi-clustering result and the initial icing duration, includes: The dynamic features and the interaction features are clustered using the first clustering method to obtain the first clustering result; The dynamic features and the interaction features are clustered using a second clustering method to obtain the second clustering result; The target icing duration is determined based on the first clustering result, the second clustering result, and the initial icing duration.

6. The method according to claim 1, characterized in that, The step of inputting the icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine into the pre-trained model to adjust the encoder parameters of the pre-trained model includes: The climate features are used as climate labels in the icing dataset; The parameters of the front-layer encoder of the pre-trained model are frozen, and the icing dataset carrying climate labels is input into the pre-trained model to adjust the unfrozen parameters of the back-layer encoder and decoder of the pre-trained model to obtain the initial detection model. Based on the icing dataset, the icing physical process of the wind farm to which the target wind turbine belongs is simulated to obtain icing increase simulation data and icing melt simulation data, and the icing increase simulation data and the icing melt simulation data are used as virtual icing dataset; The virtual icing dataset is input into the initial icing detection model to adjust the encoder parameters of the initial icing detection model, thereby obtaining the target detection model.

7. The method according to claim 1, characterized in that, The step of determining the end of training of the pre-trained model based on the icing duration to obtain the icing detection model includes: The test set is input into the pre-trained model with adjusted encoder parameters to obtain the predicted icing time and predicted wind turbine power corresponding to the test set samples. The icing time prediction result is determined based on the icing duration and the predicted icing time of the wind turbines in the test set sample, and the power prediction result is determined based on the historical operating data of the wind turbines in the test set sample and the predicted wind turbine power. If both the icing time prediction result and the power prediction result meet the standard, then the training of the pre-trained model is determined to be completed, and the pre-trained model is determined to be the icing detection model. If at least one of the icing time prediction results and the power prediction results fails to meet the target, the process returns to the step of inputting the icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine into the pre-trained model to adjust the encoder parameters of the pre-trained model.

8. A model training device, characterized in that, The device includes: The feature determination module is used to determine static features based on the historical operating data and historical meteorological data of the wind turbine, and to determine dynamic features and interactive features based on the static features of the wind turbine within a period of time. The time determination module is used to determine the icing duration corresponding to the wind turbine based on the static features, the dynamic features, and the interactive features. The training module is used to input the icing dataset, the virtual icing dataset, and the climate characteristics corresponding to the target wind turbine into the pre-training model to adjust the encoder parameters of the pre-training model, and to determine the end of training of the pre-training model based on the icing start and end time to obtain the icing detection model; the icing dataset includes the historical operating data and historical meteorological data of the target wind turbine, the virtual icing dataset is a dataset integrated from the icing data obtained after performing icing physical simulation on the target wind turbine, and the pre-training model is a model obtained after pre-training based on historical data from wind farms that do not belong to the target wind turbine.

9. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the model training method as described in any one of claims 1 to 7.

10. A readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the model training method as described in any one of claims 1 to 7.

Images