Early warning method for wind driven generator in extreme environment

Abstract

The invention relates to the technical field of wind driven generators, in particular to a wind driven generator early warning method in an extreme environment, and the method comprises the steps: obtaining the operation data of a wind driven generator; preprocessing the operation data of the wind driven generator, combining parameters in the operation data of the wind driven generator into a feature vector, and establishing an early warning level label through a historical early warning record; and taking the feature vector as the input of a preset wind driven generator early warning model, taking the early warning grade label as the output of the preset wind driven generator early warning model, taking risk measurement maximization as an optimization target, inputting the feature vector into the preset wind driven generator early warning model for calculation, and outputting the early warning grade of the wind driven generator. And accurate early warning of the wind driven generator in an extreme environment is completed. According to the method, an optimization target of risk measurement maximization is introduced, and the early warning accuracy of the model to extreme weather is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbines, and in particular to an early warning method for wind turbines in extreme environments. Background Art

[0002] With the rapid development of the wind power industry, the safe operation of wind turbines under extreme weather conditions has become a key issue that needs to be solved urgently. As a large outdoor equipment, the operating status of wind turbines is complexly affected by multiple environmental factors such as wind speed, temperature, and humidity. Accurate prediction and timely warning are crucial to preventing equipment damage caused by extreme weather. At present, wind farms generally use SCADA systems to collect wind turbine operation data in real time, including key parameters such as wind speed, temperature, humidity, power, and speed. These massive multidimensional data contain rich equipment status information, providing a data basis for establishing an accurate early warning model. How to make full use of these data to establish an accurate and reliable extreme weather early warning model is an important research topic to ensure the safe and efficient operation of wind farms. At present, traditional early warning methods mainly rely on fixed threshold judgments, which cannot effectively capture the coupling relationship between multiple environmental factors, resulting in insufficient early warning accuracy. When multiple environmental parameters are close to but not exceeding the threshold at the same time, they may be in a dangerous state, but traditional early warning methods cannot issue early warnings in time; in addition, in actual applications, the same environmental conditions often have significant differences in the degree of impact on wind turbines of different types and different service times. Summary of the invention

[0003] The purpose of the present invention is to provide a wind turbine early warning method under extreme environments, which not only establishes an intelligent early warning model based on XGBoost, but also introduces the optimization goal of maximizing risk metrics, thereby realizing accurate evaluation of the operating status of wind turbines and multi-level early warning. In addition, the present invention effectively improves the model's early warning accuracy for extreme weather through feature engineering and iterative optimization, providing reliable decision-making support for the safe operation and maintenance of wind farms.

[0004] The present invention is achieved through the following technical solutions:

[0005] A wind turbine early warning method under extreme conditions, the method comprising the following steps:

[0006] Obtaining wind turbine operating data;

[0007] Preprocess the operation data of wind turbines, combine the parameters in the operation data of wind turbines into feature vectors, and establish warning level labels through historical warning records;

[0008] The feature vector is used as the input of the preset wind turbine early warning model, and the warning level label is used as the output of the preset wind turbine early warning model. At the same time, risk metric maximization is used as the optimization goal. The feature vector is input into the preset wind turbine early warning model for calculation, and the warning level of the wind turbine is output to complete the accurate early warning of the wind turbine in extreme environments.

[0009] Optionally, the operating data of the wind turbine includes: meteorological data and fan data, wherein the meteorological data includes: wind speed, temperature and humidity; the fan data includes: power and rotation speed.

[0010] Optionally, the specific composition of the characteristic vector is: wind speed, temperature, humidity, power and rotation speed within a set time period are combined into a characteristic vector X to represent the input of the wind turbine warning model, X i =[v i ,t i ,h i ,p i ,r i ], where v i is the wind speed of the ith sample, t i is the temperature of the ith sample, h i is the humidity of the i-th sample, p i is the power of the ith sample, r i is the rotation speed of the ith sample.

[0011] Optionally, the optimization target of the wind turbine early warning model is calculated as follows:

[0012]

[0013] Among them, v cut is the cut-out wind speed, t crit is the critical temperature, h max is the maximum humidity, p max is the rated power, r nom is the rated speed, and R(X) is the risk measure.

[0014] Optionally, the calculation formula of the warning level label is:

[0015]

[0016] Among them, L(x i ) is the warning level, and θ1, θ2, θ3, and θ4 are the warning level thresholds respectively.

[0017] Optionally, the specific training process of the wind turbine early warning model is as follows:

[0018] Obtain historical operating data of wind turbines;

[0019] After preprocessing the historical operating data of wind turbines, feature vectors are constructed and risk metrics are calculated;

[0020] Based on the calculated risk metric, the historical operation data of wind turbines are divided into five warning levels by preset thresholds to form a training set (X train ,Y train ) and validation set (X val ,Y val );

[0021] The wind turbine early warning model takes the maximum risk metric as the objective function and optimizes the parameters of the wind turbine early warning model through an iterative training process. In the tth round of iteration, the wind turbine early warning model is based on the current prediction probability P t (k|x) calculates the gradient Construct a new decision tree f t (x) and add it to the wind turbine warning model: F t (x) = F t-1 (x)+f t (x);

[0022] After each round of iteration, the multi-classification error rate is calculated on the validation set. When the maximum number of iterations is reached, the training of the wind turbine warning model is completed.

[0023] Optionally, the historical operating data of the wind turbine is preprocessed, which specifically includes: outlier detection and missing value processing.

[0024] Optionally, the wind turbine early warning model is specifically an XGBoost model.

[0025] The technical solution of the present invention has at least the following advantages and beneficial effects:

[0026] The present invention not only establishes an intelligent early warning model based on XGBoost, but also introduces the optimization goal of maximizing risk metrics, thereby realizing accurate evaluation of the operating status of wind turbines and multi-level early warning. In addition, the present invention effectively improves the model's early warning accuracy for extreme weather through feature engineering and iterative optimization, thus providing reliable decision-making support for the safe operation and maintenance of wind farms. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A schematic flow chart of a wind turbine early warning method under extreme environments provided by the present invention. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, what is described is only a part of the present invention, not all of it. Generally, the components of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0029] like Figure 1 As shown, the present invention provides one embodiment: a wind turbine generator early warning method under extreme environment, the method comprising the steps of:

[0030] Obtaining wind turbine operating data;

[0031] Preprocess the operation data of wind turbines, combine the parameters in the operation data of wind turbines into feature vectors, and establish warning level labels through historical warning records;

[0032] The feature vector is used as the input of the preset wind turbine early warning model, and the warning level label is used as the output of the preset wind turbine early warning model. At the same time, risk metric maximization is used as the optimization goal. The feature vector is input into the preset wind turbine early warning model for calculation, and the warning level of the wind turbine is output to complete the accurate early warning of the wind turbine in extreme environments.

[0033] In this embodiment, the operation data of the wind turbine generator includes: meteorological data and fan data, wherein the meteorological data includes: wind speed, temperature and humidity; the fan data includes: power and rotation speed.

[0034] In a specific implementation, the sampling period of all data included in the above wind turbine operation data is 10 minutes.

[0035] In the specific application of this embodiment, the specific composition of the characteristic vector is: the wind speed, temperature, humidity, power and speed within a set time period are combined into a characteristic vector X to represent the input of the wind turbine warning model, X i =[v i ,t i ,h i ,p i ,r i ], where v i is the wind speed of the ith sample, t i is the temperature of the ith sample, h i is the humidity of the i-th sample, p i is the power of the ith sample, r i is the rotation speed of the ith sample.

[0036] The calculation formula of the warning level label is:

[0037]

[0038] Among them, L(x i ) is the warning level, and θ1, θ2, θ3, and θ4 are the warning level thresholds respectively.

[0039] Specifically, in this embodiment, 0 is a normal operating state; 1 is a blue warning; 2 is a yellow warning; 3 is an orange warning; and 4 is a red warning. It can be understood that in this embodiment, the larger the warning level number, the higher the warning level. According to different warning levels, the standard warning plan for wind turbines in the corresponding area is selected for execution.

[0040] In a further implementation of this embodiment, the optimization target of the wind turbine early warning model is calculated as follows:

[0041]

[0042] Among them, v cut is the cut-out wind speed, t crit is the critical temperature, h max is the maximum humidity, p max is the rated power, r nom is the rated speed, and R(X) is the risk measure.

[0043] This embodiment is specifically based on the iterative strategy of gradient boosting. Starting from the maximization of the risk metric R(X), the accuracy of early warning is improved by continuously adjusting the model parameters. Specifically, in each round of iteration, the predicted probability distribution of the sample is first calculated based on the current model state, and then the gradient of the risk metric R(X) to the predicted probability is calculated. This gradient reflects the direction and magnitude of the model prediction result that needs to be adjusted. Next, a new decision tree is constructed to fit this gradient direction so that the prediction performance of the model in this direction is improved. After the new tree is constructed, it is added to the existing model to form an updated cumulative model. It can be understood that in order to prevent overfitting, the multi-classification error rate is calculated on the validation set after each iteration. When the performance of the validation set is no longer improved for 10 consecutive rounds, the early stopping mechanism is triggered, and the model is considered to have reached the optimal state. This optimization process is essentially to find the optimal solution in the risk metric space, guide the search direction through gradient information, and capture the complex relationship between features with the help of the nonlinear characteristics of the decision tree, and finally obtain an early warning model that can accurately predict the risk of extreme weather. The entire optimization process controls the magnitude of each update step by using a learning rate of η = 0.1, and limits the complexity of a single tree by setting a maximum depth.

[0044] Furthermore, the specific training process of the wind turbine early warning model is as follows:

[0045] Obtain historical operating data of wind turbines;

[0046] After preprocessing the historical operating data of wind turbines, feature vectors are constructed and risk metrics are calculated;

[0047] Based on the calculated risk metric, the historical operation data of wind turbines are divided into five warning levels by preset thresholds to form a training set (X train ,Y train ) and validation set (X val ,Y val );

[0048] The wind turbine early warning model takes the maximum risk metric as the objective function and optimizes the parameters of the wind turbine early warning model through an iterative training process. In the tth round of iteration, the wind turbine early warning model is based on the current prediction probability P t (k|x) calculates the gradient Construct a new decision tree f t (x) and add it to the wind turbine warning model: F t (x) = F t-1 (x)+f t (x);

[0049] After each round of iteration, the multi-classification error rate is calculated on the validation set. When the maximum number of iterations is reached, the training of the wind turbine warning model is completed.

[0050] In implementation, the model starts with the original data {wind speed v, temperature t, humidity h, power p, speed r}, and after outlier detection and missing value processing, constructs the feature vector X i =[v i ,t i ,h i ,p i ,r i ], and calculate the risk metric:

[0051]

[0052] Based on the calculated R(X) value, the samples are divided into 5 warning levels by using the preset thresholds {θ1, θ2, θ3, θ4} to form a training set (X train ,Y train ) and validation set (X val ,Y val ).

[0053] Next, the XGBoost model optimizes the model parameters through an iterative training process with the objective function of maximizing the risk metric R(X). In the tth iteration, the model optimizes the model parameters based on the current predicted probability P t (k|x) calculates the gradient Construct a new decision tree f t (x) and add it to the model: F t (x) = F t-1 (x)+f t (x). After each round of iteration, the multi-classification error rate is calculated on the validation set. When the performance of the validation set does not improve after 10 consecutive rounds, the training process stops early and the optimal model is selected. Finally, for the newly input real-time monitoring data, the model calculates its risk metric value and predicts the warning level through the optimal model, thereby achieving real-time warning of extreme weather.

[0054] In an application example, a wind farm in a certain place is taken as an example. The wind farm is located in a mountainous area, with an installed capacity of 100MW and a total of 50 wind turbines with a single unit capacity of 2MW. The wind farm is located in a monsoon climate zone, with strong winds in winter and thunderstorms in summer, which pose a serious threat to the safe operation of wind turbines. Based on historical operation data, this embodiment collects SCADA data from January 2022 to December 2023, with a sampling interval of 10 minutes, including key parameters such as wind speed, temperature, humidity, power and speed. In the data preprocessing stage, outliers are identified by the box plot method, and the time series interpolation method is used to process missing data, and finally about 52,000 valid samples are obtained. Among them, the data for the whole year of 2022 is selected as the training set, and the data for 2023 is selected as the verification set. During the model training process, the warning level thresholds are set: θ1=0.3, θ2=0.5, θ3=0.7, θ4=0.9. After 100 rounds of iterative training, the multi-classification error rate of the model on the validation set dropped to 0.08, indicating that the warning accuracy rate reached 92%. Especially for extreme weather events, such as a severe convective weather process in August 2023, the model issued a red warning signal (level 4) 30 minutes in advance under extreme conditions of wind speed reaching 20m / s, temperature rising to 35°C, and humidity of 90%, allowing the operation and maintenance personnel to take shutdown protection measures in time.

[0055] Based on the results of this application case, the early warning system successfully warned of 15 extreme weather events in 2023, including 8 strong wind weather, 5 thunderstorm weather and 2 extreme temperature events, with an average advance warning time of 25 minutes, providing strong support for the safe operation and maintenance of wind farms. Compared with traditional fixed threshold early warning, the false alarm rate is reduced by 60% and the missed alarm rate is reduced by 75%, which significantly improves the accuracy and reliability of early warning. Through timely early warning and active protection, the number of equipment failures caused by extreme weather in the wind farm in 2023 was reduced by 65% ​​compared with 2022, and the economic benefits were significantly improved.

[0056] The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A wind turbine early warning method in extreme environments, characterized in that: The steps of the method include: Obtaining wind turbine operating data; Preprocess the operation data of wind turbines, combine the parameters in the operation data of wind turbines into feature vectors, and establish warning level labels through historical warning records; The feature vector is used as the input of the preset wind turbine early warning model, and the warning level label is used as the output of the preset wind turbine early warning model. At the same time, risk metric maximization is used as the optimization goal. The feature vector is input into the preset wind turbine early warning model for calculation, and the warning level of the wind turbine is output to complete the accurate early warning of the wind turbine in extreme environments.

2. The wind turbine generator early warning method under extreme environment according to claim 1 is characterized in that: The operation data of the wind turbine generator includes: meteorological data and fan data, wherein the meteorological data includes: wind speed, temperature and humidity; the fan data includes: power and rotation speed.

3. The wind turbine generator early warning method under extreme environment according to claim 2 is characterized in that: The specific composition of the characteristic vector is: wind speed, temperature, humidity, power and speed within a set time period are combined into a characteristic vector X to represent the input of the wind turbine warning model, X i =[v i ,t i ,h i ,p i ,r i ], where v i is the wind speed of the ith sample, t i is the temperature of the ith sample, h i is the humidity of the i-th sample, p i is the power of the ith sample, r i is the rotation speed of the ith sample.

4. The wind turbine generator early warning method under extreme environment according to claim 3 is characterized in that: The optimization target of the wind turbine early warning model is calculated as follows: Among them, v cut is the cut-out wind speed, t crit is the critical temperature, h max is the maximum humidity, p max is the rated power, r nom is the rated speed, and R(X) is the risk measure.

5. The wind turbine generator early warning method under extreme environment according to claim 4 is characterized in that: The calculation formula of the warning level label is: Among them, L(x i ) is the warning level, and θ1, θ2, θ3, and θ4 are the warning level thresholds respectively.

6. The wind turbine generator early warning method under extreme environment according to claim 5 is characterized in that: The specific training process of the wind turbine early warning model is as follows: Obtain historical operating data of wind turbines; After preprocessing the historical operating data of wind turbines, feature vectors are constructed and risk metrics are calculated; Based on the calculated risk metric, the historical operation data of wind turbines are divided into five warning levels by preset thresholds to form a training set (X train ,Y train ) and validation set (X val ,Y val ); The wind turbine early warning model takes the maximum risk metric as the objective function and optimizes the parameters of the wind turbine early warning model through an iterative training process. In the tth round of iteration, the wind turbine early warning model is based on the current prediction probability P t (k|x) calculates the gradient Construct a new decision tree f t (x) and add it to the wind turbine warning model: F t (x) = F t-1 (x)+f t (x); After each round of iteration, the multi-classification error rate is calculated on the validation set. When the maximum number of iterations is reached, the training of the wind turbine warning model is completed.

7. The wind turbine generator early warning method under extreme environment according to claim 6 is characterized in that: The preprocessing of the historical operation data of the wind turbine generator is specifically: abnormal value detection and missing value processing.

8. The wind turbine generator early warning method under extreme environment according to claim 7 is characterized in that: The wind turbine early warning model is specifically an XGBoost model.

Images