Wind turbine power curve prediction method and device

Abstract

The application provides a wind turbine power curve prediction method and device, and relates to the technical field of wind turbines.The actual power, generator speed and environmental data of the wind turbine are collected first to provide a basis for subsequent analysis.Then, the power generation data of the previous moment is taken as input, the actual power of the next moment is taken as a label, a multiple linear regression model is trained, and the influence of the current moment is focused on, while the cumulative influence of the past is easily ignored.Then, the power generation data of the previous multiple moments is taken as input, the actual power of the next moment is taken as a label, a long short-term memory network model is trained, and the influence of the time series data is focused on, which is suitable for dynamic changing data.The generator speed, wind speed, temperature and humidity data are analyzed to obtain a fluctuation adjustment coefficient, the actual power predicted by the two models at the current moment is corrected by using the fluctuation adjustment coefficient, and the corrected actual power is applied to the wind turbine power curve.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind turbine generator sets, in particular to a wind turbine generator set power curve prediction method and device. BACKGROUND

[0002] Wind power generation, as an important renewable energy source, has become increasingly prominent in the global energy structure in recent years. With the deepening of global concern for sustainable development, the use of traditional fossil fuels is facing increasing environmental pressure, leading to problems such as greenhouse gas emissions and air pollution. Wind energy, due to its clean and renewable characteristics, has become one of the alternative energy sources actively promoted by governments around the world, helping to reduce carbon emissions, protect the environment, and improve energy security.

[0003] Wind turbine generator set power curve prediction is a crucial step in wind energy utilization and management. The power curve describes the power generation capacity of a wind turbine generator set at different wind speeds, and accurate prediction can effectively improve the efficiency of wind power generation and the stability of the power grid. The following is the background technology related to wind turbine generator set power curve prediction. First, the establishment of the power curve is usually based on the relationship between wind speed and output power. The power output of a wind turbine generator set is affected by wind speed, which usually exhibits nonlinear characteristics. Traditional power curve prediction methods rely heavily on statistical analysis of historical data, including regression analysis and time series models. However, these methods often fail to provide sufficient accuracy when faced with complex and dynamic wind speed changes. Therefore, in recent years, more and more research has begun to use machine learning and data-driven methods to improve prediction accuracy. These methods analyze a large amount of historical data to extract features and patterns, allowing the prediction model to better adapt to changes in wind speed and improve the model's generalization ability.

[0004] The prior art has the following shortcomings:

[0005] The existing wind turbine power curve prediction technology has many shortcomings, mainly including the neglect of the nonlinear relationship between wind speed and actual power, poor dynamic change adaptability, insufficient model generalization ability, and lack of real-time performance, etc., which may lead to reduced prediction accuracy and reliability. Many traditional prediction methods mainly use linear regression models, assuming that the relationship between wind speed and power generation is linear. However, in fact, this relationship is often complex and nonlinear, especially when the wind speed changes greatly, the linear model is prone to increase the prediction error and cannot accurately reflect the actual power generation. In addition, traditional methods have poor adaptability in dealing with rapidly changing environmental conditions, which makes it difficult for them to capture the impact of changes in wind speed, temperature, and other factors on power generation in a timely manner, resulting in lagging prediction results. Furthermore, many models perform well on specific data sets, but lack generalization ability in different wind farms or different geographical conditions, limiting their application to a wide range of situations, restricting their practical application. Some traditional prediction methods lack real-time performance and cannot be dynamically updated according to the latest data, which makes the model unable to reflect the actual state of the current wind turbine, affecting the timeliness and effectiveness of scheduling decisions.

[0006] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0007] The purpose of the present application is to provide a wind turbine power curve prediction method and device to solve the problems raised in the background.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0009] A wind turbine power curve prediction method, the specific steps comprising:

[0010] Step 1: Collecting historical power generation data of the wind turbine, the historical power generation data including actual power, generator speed and environmental data at each time point in the power curve of the wind turbine, the environmental data including wind speed, temperature and humidity;

[0011] Step 2: Taking the actual power, generator speed and environmental data of the previous time point in the historical power generation data as input, and the actual power of the next time point as label, a multiple linear regression model is constructed and trained;

[0012] Step 3: Taking the historical power generation data of the previous time point in the historical power generation data as input, and the actual power of the next time point as label, a power prediction model is constructed and trained based on a long short-term memory network model;

[0013] Step 4: Calculate the fluctuation adjustment coefficient according to the generator speed, wind speed, temperature and humidity data at the current time;

[0014] Step 5: Collect the power generation data at the current time and the previous time, input them into the multiple linear regression model and the power prediction model, obtain the actual power at the next time predicted by the two models, and then calculate the actual power corrected by the fluctuation adjustment coefficient, and apply the corrected actual power to the wind turbine power curve.

[0015] Further, the historical power generation data of the wind turbine generator set specifically includes:

[0016] The current, voltage and generator speed data of the wind turbine generator set are collected through the power sensor and the speed sensor equipped on the wind turbine generator set, the power factor of the wind turbine generator set is obtained through the calculation type electric energy meter, and the formula for calculating the actual power of the wind turbine generator set is:

[0017]

[0018] wherein, P is the actual power, V is the voltage, I is the current, PF is the power factor;

[0019] An environmental data collection point is deployed on the upper part of the wind turbine tower, and a temperature sensor, a humidity sensor and a wind speed sensor are installed to collect temperature, humidity and wind speed data at the location of the wind turbine generator set;

[0020] The collected historical power generation data is preprocessed, including removing outliers, missing values and normalization processing, calculating the Z-score of each parameter in the historical power generation data, and when the absolute value of the Z-score of the parameter is , the parameter is regarded as an outlier and is removed, and the parameters after removing outliers are filled with the mean of the previous time and the next time parameters. After the processing of outliers and missing values, the Min-Max method is used for normalization processing, and all parameters are mapped to the interval [0, 1]. The historical power generation data after preprocessing is sorted into a historical data set according to the time stamp.

[0021] Further, the training of the multiple linear regression model specifically includes:

[0022] The actual power, generator speed, wind speed, temperature and humidity data at the previous time in the historical data set are taken as input, and the actual power at the next time is taken as label, a multiple linear regression model is selected, and the multiple linear regression model is represented as:

[0023] wherein, is actual power at the moment, is rotational speed of the wind turbine at the moment, is actual power at the moment, is wind speed at the moment, is temperature at the moment, is humidity at the moment, is an intercept, is a parameter influence coefficient, is a time variable, is an error term;

[0024] The historical data set is divided into a training set and a test set, 80% for training and 20% for testing, the data in the training set is used to fit the regression model, the mean square error is selected as the loss function, the least square method is used to minimize the loss function, thereby determining the regression coefficients, the root mean square error in the training process is monitored in the training model, when the decline amplitude is less than 0.01 within 10 consecutive rounds, the training is stopped, which indicates that the model training is completed, after the model training is completed, the test set is used for model evaluation, the determination coefficient R2 is calculated to evaluate the model performance.

[0025] Further, the power prediction model is trained, and the power prediction model specifically includes:

[0026] The input sequence length of the long short-term memory network is set to time steps, 10, the historical power generation data of the first 10 moments in the training set is input, and the actual power data of the moment in the training set is taken as a label, the long short-term memory network model is selected to train the power prediction model, after the training is completed, the test set is input into the trained power prediction model for testing, the mean square error is selected as the loss function, when the validation set loss decreases by less than 0.001 within 10 consecutive rounds, it indicates that the power quality prediction model training is completed.

[0027] Further, the calculation of the fluctuation adjustment coefficient specifically includes:

[0028] The generator speed, wind speed, temperature and humidity data of the current moment and the previous moments are arranged in the order of collection time to form generator speed time series data, wind speed time series data, temperature time series data and humidity time series data, and the fluctuation index of the generator speed is calculated as:

[0029] ​

[0030] wherein, is a fluctuation index of the generator speed, is a standard deviation of the generator speed time series data, is a mean of the generator speed time series data;

[0031] Similarly, the fluctuation indexes of the wind speed, temperature and humidity are respectively

[0032] The fluctuation indexes of the wind speed, temperature and humidity are analyzed to obtain a natural index, and the calculation formula is:

[0033]

[0034] wherein, is the natural index, are weight coefficients of the wind speed, temperature and humidity in the natural index, , and

[0035] The formula for calculating the fluctuation adjustment coefficient is:

[0036]

[0037] wherein, is the fluctuation adjustment coefficient. Further, the actual power corrected by the fluctuation adjustment coefficient specifically includes:

[0038] The current time power generation data of the wind turbine generator set is collected, including the generator speed, wind speed, temperature, humidity and current time actual power data, the power generation data of the previous multiple time points and the current time is input into the trained power prediction model, the actual power predicted by the model is obtained, and the actual power predicted by the multiple linear regression model and the power prediction model is corrected by the fluctuation adjustment coefficient, and the formula is:

[0039]

[0040] wherein, is the actual power corrected by the fluctuation adjustment coefficient, is the actual power predicted by the multiple linear regression model output, is the actual power predicted by the power prediction model output;

[0041] ​​The actual power corrected by the fluctuation adjustment coefficient is applied to the power curve of the wind turbine at the next moment, and the actual power corrected by the fluctuation adjustment coefficient is compared with the preset power threshold. If the actual power corrected by the fluctuation adjustment coefficient is lower than the preset power threshold, the wind turbine continues to be monitored. If the actual power corrected by the fluctuation adjustment coefficient is higher than the preset power threshold, an abnormal warning is issued.

[0042] The present invention also provides a wind turbine power curve prediction device, which is used to implement the above-mentioned wind turbine power curve prediction method, including:

[0043] The historical data acquisition module is used to collect historical power generation data of the wind turbine generator set. The historical power generation data includes the actual power, generator speed and environmental data at each moment in the generator set power curve. The environmental data includes wind speed, temperature and humidity.

[0044] The regression model building module is used to construct and train a multiple linear regression model by taking the actual power, generator speed and environmental data of the previous moment in the historical power generation data as inputs and the actual power of the next moment as the label.

[0045] The power prediction model building module is used to analyze historical power generation data from previous periods. Using historical power generation data at each moment as input and the actual power at the next moment as a label, a power prediction model is constructed and trained based on a long short-term memory network model.

[0046] The fluctuation adjustment coefficient calculation module is used to calculate the fluctuation adjustment coefficient at the current moment based on the generator speed, wind speed, temperature and humidity data at the current moment.

[0047] The real-time data prediction and correction module is used to collect data at the current time and in the previous time. The power generation data at each moment is input into a multiple linear regression model and a power prediction model to obtain the actual power predicted by the two models for the next moment. Then, the actual power is calculated after adjusting for fluctuations using the fluctuation adjustment coefficient. This adjusted actual power is then applied to the power curve of the wind turbine generator. The technical effects and advantages provided by this invention in the above technical solution are as follows:

[0048] This invention uses multiple linear regression analysis of historical power generation data to obtain the influence coefficients of various parameters on wind turbine power. Then, by constructing a Long Short-Time Memory (LSTM) network model to predict the actual power at future moments, it accurately captures the impact of generator speed and environmental data on actual power, thereby improving the accuracy and reliability of the prediction. By introducing a fluctuation adjustment coefficient to correct the actual power predicted by the two models, the final prediction results are closer to the actual power generation situation. This allows for real-time monitoring and anomaly warnings, providing strong support for wind power scheduling and management, and ultimately promoting the efficient utilization and stability of renewable energy. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0050] Figure 2 This is a schematic diagram of the device structure of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0052] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0053] Example:

[0054] Please see Figure 1 The present invention provides a technical solution:

[0055] A method for predicting the power curve of a wind turbine generator set, comprising the following steps:

[0056] Step 1: Collect historical power generation data of the wind turbine generator set. The historical power generation data includes the actual power, generator speed and environmental data at each moment in the generator set power curve. The environmental data includes wind speed, temperature and humidity.

[0057] In this embodiment, the collection of wind turbine generator set historical power generation data specifically includes:

[0058] The current, voltage and generator speed data of the wind turbine generator set are collected through the power sensor and the speed sensor equipped with the wind turbine generator set, the power factor of the wind turbine generator set is obtained through the calculation type electric energy meter, and the formula for calculating the actual power of the wind turbine generator set is:

[0059]

[0060] wherein, is the actual power, is the voltage, is the current, is the power factor, and here the calculation of the actual power represents a three-phase system;

[0061] An environmental data collection point is deployed on the upper part of the wind turbine tower, and a temperature sensor, a humidity sensor and a wind speed sensor are installed to collect temperature, humidity and wind speed data at the location of the wind turbine generator set;

[0062] The collected historical power generation data is preprocessed, including removing outliers, missing values and normalization processing, and calculating the Z-score of each parameter in the historical power generation data. When the absolute value of the Z-score of the parameter is , the parameter is regarded as an outlier and is removed. The mean value of the previous moment and the next moment of the parameter after removing the outliers is used to fill in the missing values. After the processing of outliers and missing values, the Min-Max method is used for normalization processing, and all parameters are mapped to the interval [0, 1]. The historical power generation data after preprocessing is sorted into a historical data set according to the time stamp.

[0063] A comprehensive database is provided by collecting historical power generation data of wind turbines, including actual power, speed and environmental data (wind speed, temperature and humidity) of the wind turbine. These data are crucial for understanding the operating characteristics of wind turbines and their response to environmental changes. By deploying environmental data collection points on the upper part of the wind turbine tower, wind speed, temperature and humidity data can be obtained in real time. This real-time monitoring method, combined with the operating status of the wind turbine, increases the response capability to external environmental changes and more accurately reflects the impact of environmental factors on power generation performance. By using the Z-score method to detect and remove outliers, the reliability of the data is ensured, and missing values are filled by the mean value to avoid unstable model training caused by missing data. This process improves the integrity and quality of the data set, providing a more robust data foundation for subsequent model training. Through Min-Max normalization processing, all parameters are mapped to the [0, 1] interval, which not only eliminates the influence between different dimensions, but also improves the convergence speed of model training and the stability of prediction, so that different features can be compared and analyzed on the same scale.

[0064] Step 2: The actual power, generator speed and environmental data of the previous time in the historical power generation data are taken as input, and the actual power of the next time is taken as label to build and train a multiple linear regression model.

[0065] In this embodiment, training the multiple linear regression model specifically includes:

[0066] The actual power, generator speed, wind speed, temperature and humidity data of the previous time in the historical data set are taken as input, and the actual power of the next time is taken as label. A multiple linear regression model is selected, and the multiple linear regression model is represented as:

[0067] wherein, is the actual power at time t, is the speed of the wind turbine at time t, is the actual power at time t, is the wind speed at time t, is the temperature at time t, is the humidity at time t, is the intercept, is the influence coefficient of each parameter, is the time variable, is the error term.

[0068] ​​​​​​The historical data set is divided into a training set and a test set, 80% for training and 20% for testing. The data in the training set is used to fit the regression model, the mean square error is selected as the loss function, the least square method is used to minimize the loss function, and the regression coefficients are determined. The root mean square error is monitored during the training of the model. When the decrease is less than 0.01 in 10 consecutive rounds, the training is stopped, indicating that the model training is completed. After the model training is completed, the test set is used for model evaluation, and the determination coefficient R<2> value is calculated to evaluate the model performance. The smaller the R<2> value, the better the model performance. The smaller the R<2> value, the better the model performance.

[0069] By selecting the generator speed and environmental data (wind speed, temperature, humidity) as features, using the actual power at the previous moment as input, and the target being the actual power at the current moment. This selection conforms to the physical characteristics of wind turbine generators and can accurately capture key factors affecting power generation, helping to improve model effectiveness. The multiple linear regression model is a classic statistical method suitable for handling linear relationships between features and targets. By building this model, the influence of each feature on actual power can be clearly understood, which can help optimize the operation and management of wind turbine generators. The historical data set is divided into a training set and a test set (80% for training and 20% for testing), which ensures the generalization ability of the model on different data sets. In the training process, the mean square error is used as the loss function, and the least square method is used to optimize the regression coefficients to ensure that the model can effectively fit the historical data. In addition, by monitoring the root mean square error (RMSE) and setting the condition for stopping training, overfitting is avoided, and the robustness of the model is improved. The R<2> value is used to evaluate the model performance, and the closer the R<2> value is to 1, the better the model fits the data. Appropriate evaluation criteria help to quantify model performance and ensure that the final model can accurately and reliably predict power generation.

[0070] Step 3: Use the historical power generation data at the previous moment as input and the actual power at the next moment as label to build and train a power prediction model based on the long short-term memory network model.

[0071] In this embodiment, training the power prediction model specifically includes:

[0072] The input sequence length of the long short-term memory network is set to 10 time steps, 10, the first the actual power data of the next moment in the training set as the label, a long short-term memory network model is selected to train the power prediction model, after the training is completed, the test set is input into the trained power prediction model for testing, the mean square error is selected as the loss function, when the loss of the validation set decreases by less than 0.001 in 10 consecutive rounds, it indicates that the power quality prediction model training is completed. Here, the training set and the test set are the training set and the test set divided in step 2.

[0073] The long short-term memory network (LSTM) is a deep learning model specially designed to process sequence data, which can effectively capture long-term dependencies in time series. By taking the historical power generation data of the previous moment as input, LSTM can learn the patterns hidden in the historical data, thereby enhancing the predictive ability of the model. When processing sequence data, LSTM can effectively overcome the gradient vanishing and explosion problems of traditional recurrent neural networks (RNN) in long sequence learning. This feature makes LSTM very suitable for wind power prediction, as it can capture both long-term and short-term information related to time. Selecting 10 time steps as the input length is a reasonable choice, as it can capture the trends of wind speed, temperature, and rotational speed in a short period of time. According to the actual characteristics of the data and the training effect of the model, this parameter can be further optimized, but the selection of 10 time steps usually performs well in practice. Selecting the mean square error (MSE) as the loss function can effectively measure the difference between the predicted value and the actual value of the model. During the training process, setting the decrease of the validation set loss by less than 0.001 in 10 consecutive rounds as the stopping condition ensures that the model can achieve good convergence and generalization ability during the training process, thereby improving the accuracy and robustness of the prediction.

[0074] Step 4: Calculate the fluctuation adjustment coefficient of the current moment based on the generator speed, wind speed, temperature and humidity data of the current moment;

[0075] In this embodiment, the calculation of the fluctuation adjustment coefficient specifically includes:

[0076] The generator speed, wind speed, temperature and humidity data of the current moment and the previous moment are arranged in the order of collection time to form generator speed time series data, wind speed time series data, temperature time series data and humidity time series data, and the fluctuation index of the generator speed is calculated as:

[0077]

[0078] wherein, is the fluctuation index of the generator speed, is the standard deviation of the generator speed time series data, The mean value of the generator speed time series data;

[0079] Similarly, the fluctuation indicators of wind speed, temperature, and humidity are obtained as ;

[0080] The fluctuation indicators of wind speed, temperature, and humidity are analyzed to obtain the natural indicator, and the calculation formula is:

[0081]

[0082] wherein, is the natural indicator, are the weight coefficients of wind speed, temperature, and humidity in the natural indicator, , and ;

[0083] The natural indicator is a quantity that comprehensively considers the fluctuations of wind speed, temperature, and humidity. Its purpose is to quantify the degree of influence of environmental factors on the power output of wind generators. Wind speed, temperature, and humidity are key environmental factors that affect the operating efficiency and power output of wind generators. Different environmental conditions can have a significant impact on the extraction efficiency of wind energy, so the fluctuations of these factors need to be considered. Wind speed is one of the most critical factors in wind power generation, and changes in wind speed directly affect the availability of wind energy, so its fluctuation indicator has a relatively large impact on power generation. Since wind speed has a huge impact on power generation, the larger the wind speed, the larger the natural indicator, and the larger the fluctuation adjustment coefficient, indicating that in the case of large environmental fluctuations, it relies more on the power prediction model, so its weight is the largest. Temperature affects the efficiency and performance of wind generators, including air density, which in turn affects the efficiency of wind energy conversion. Although the impact of temperature changes on power output is smaller than that of wind speed, it is still important, and its degree of influence is lower than that of wind speed, so its weight is set lower than that of wind speed. Humidity has a relatively small impact on wind power generation, and only in specific cases (such as high humidity environments) can it affect the operation of the equipment. Among the three environmental factors, humidity has the smallest impact on the natural indicator, so its weight is set to be the smallest.

[0084] The formula for calculating the fluctuation adjustment coefficient is expressed as:

[0085]

[0086] wherein, is the fluctuation adjustment coefficient.

[0087] When the environmental fluctuations are large, it means that the changes in the environment have a significant impact on the power, and at this time will have a higher value, relying more on the power prediction model; when the generator speed is large, it means that the system is in a relatively unstable state, and the current time has a greater impact, at this time The value of the fluctuation adjustment coefficient is smaller, and more dependent on the multiple linear regression model.

[0088] The power generation capacity of a wind turbine is affected by various environmental factors, including wind speed, temperature, humidity, etc. The volatility of these factors at different times and conditions can be large. By introducing a fluctuation adjustment coefficient , the prediction model can be adjusted in real time according to the current environmental conditions, thereby enhancing the adaptability of the model to different working environments. In practical applications, the operating state and environmental conditions of the wind turbine are not always smooth. When the environment fluctuates greatly (such as when the wind speed changes dramatically), the traditional multiple linear regression model may not be able to capture the impact of these changes well. On the contrary, the long short-term memory network (LSTM) has an advantage in handling time series data and capturing dynamic patterns. By appropriately reducing the weight of the multiple linear regression model (i.e., increasing the weight of the LSTM model), more accurate predictions can be obtained in high fluctuation situations. Each model has its limitations. The multiple linear regression model assumes a linear relationship between input and output, and is sensitive to outliers; while the LSTM model, although capable of handling complex nonlinear relationships, may also perform poorly when the data volume is insufficient or the feature selection is inappropriate. The steps of implementing the fluctuation adjustment coefficient can effectively combine the advantages of both models, select the more suitable model under different environmental conditions, and overcome the limitations of a single model. The introduction of the fluctuation adjustment coefficient can help reduce the prediction error caused by changes in environmental factors. When the fluctuation is large, the system can preferentially rely on the LSTM model, which is sensitive to dynamic changes, to reduce the inaccuracy of the prediction caused by dramatic changes in the environment; while in a relatively stable environment, relying on the multiple linear regression model can provide more stable prediction results. This flexibility allows the overall prediction error to be reduced.

[0089] Step 5: Collect the power generation data at the current time and the previous times, input them into the multiple linear regression model and the power prediction model, obtain the actual power predicted by the two models at the next time, and then calculate the actual power corrected by the fluctuation adjustment coefficient, and apply the corrected actual power to the wind turbine power curve.

[0090] In this embodiment, the calculation of the actual power corrected by the fluctuation adjustment coefficient specifically includes:

[0091] Collect the power generation data of the wind turbine at the current time, including the generator speed, wind speed, temperature, humidity, and actual power data at the current time, input the power generation data of the previous times and the current time into the trained power prediction model, obtain the actual power predicted by the model, and then correct the actual power predicted by the multiple linear regression model and the power prediction model through the fluctuation adjustment coefficient, which is represented by the formula:

[0092]

[0093] wherein, is the actual power corrected by the fluctuation adjustment coefficient, is the actual power predicted by the multiple linear regression model, is the actual power predicted by the power prediction model;

[0094] The actual power corrected by the fluctuation adjustment coefficient is applied to the power curve of the wind turbine at the next time, and the actual power corrected by the fluctuation adjustment coefficient is compared with the preset power threshold, if the actual power corrected by the fluctuation adjustment coefficient is lower than the preset power threshold, the wind turbine continues to monitor, if the actual power corrected by the fluctuation adjustment coefficient is higher than the preset power threshold, an abnormal early warning is sent.

[0095] The multiple linear regression model (usually performs well in the case of little change in environmental conditions, depends on the linear relationship of historical data, and the power prediction model (such as deep learning model like LSTM) is more suitable for dynamic and complex environmental changes, and can capture more complex nonlinear relationships, therefore, the outputs of the two models are weighted and fused, which can fully utilize the advantages of each other and improve the accuracy of prediction. The introduction of the fluctuation adjustment coefficient is to dynamically adjust the weights of the two models to adapt to different environmental fluctuation states, when the environmental fluctuation is large, is large, so that the weight of the power prediction model is larger, and the flexibility and accuracy of prediction are improved, when the environmental fluctuation is small, is small, which increases the stability of prediction, this mechanism can meet the prediction demand under different environmental conditions such as wind speed, temperature and humidity, and achieve more accurate power prediction.

[0096] Please refer to Figure 2 , the application further provides a wind turbine power curve prediction device, the wind turbine power curve prediction device is used for realizing the wind turbine power curve prediction method, and includes:

[0097] A historical data acquisition module is configured to acquire historical power generation data of a wind turbine, wherein the historical power generation data includes actual power at each time in a power curve of a generator, generator speed and environmental data, and the environmental data includes wind speed, temperature and humidity.

[0098] A regression model construction module is configured to use the actual power at a previous time, the generator speed and the environmental data in the historical power generation data as input, and the actual power at a next time as label, to construct and train a multiple linear regression model.

[0099] The power prediction model construction module is configured to construct and train a power prediction model based on a long short-term memory network model, with historical power generation data at a previous time as input and actual power at a next time as a label.

[0100] The fluctuation adjustment coefficient calculation module is configured to calculate a fluctuation adjustment coefficient at the current time according to the generator speed, wind speed, temperature and humidity data at the current time.

[0101] The real-time data prediction and correction module is configured to collect power generation data at the current time and at a previous time, input the data into a multiple linear regression model and the power prediction model, obtain actual power at a next time predicted by the two models, calculate actual power after correction of the fluctuation adjustment coefficient in combination with the fluctuation adjustment coefficient, and apply the corrected actual power to a power curve of the wind turbine generator.

[0102] The above formulas are all dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the most recent real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0103] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of the examples described in connection with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0104] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0105] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application.​​

Claims

1. A method of predicting power curve of a wind turbine generator system, characterized in that, The specific steps include: Step 1: Collecting historical power generation data of the wind turbine generator set, the historical power generation data including actual power at each time point in the power curve of the generator set, generator speed and environmental data, the environmental data including wind speed, temperature and humidity; Step 2: Taking the actual power, generator speed and environmental data at the previous time point in the historical power generation data as input, and the actual power at the next time point as label, a multiple linear regression model is constructed and trained; Step 3: Extract the previous historical power generation data Using historical power generation data at each moment as input and the actual power at the next moment as a label, a power prediction model is constructed and trained based on a long short-term memory network model. Step 4: According to the generator speed, wind speed, temperature and humidity data at the current time point, the fluctuation adjustment coefficient at the current time point is calculated; Step 5: collect the current and previous time's power generation data, input into the multiple linear regression model and the power prediction model, obtain the next time's actual power predicted by the two models, and then combine the fluctuation adjustment coefficient to calculate the fluctuation adjustment coefficient corrected actual power, and apply the corrected actual power to the wind turbine power curve. Step 5: collect the current and previous time's power generation data, input into the multiple linear regression model and the power prediction model, obtain the next time's actual power predicted by the two models, and then combine the fluctuation adjustment coefficient to calculate the fluctuation adjustment coefficient corrected actual power, and apply the corrected actual power to the wind turbine power curve. The calculation of the fluctuation adjustment coefficient specifically includes: Use the current time and the previous time The generator speed, wind speed, temperature, and humidity data at each moment are arranged in chronological order of collection time to form time-series data for generator speed, wind speed, temperature, and humidity. The fluctuation index of generator speed is calculated as follows: wherein, is an index of the fluctuation of the generator speed, is a standard deviation of the generator speed time series data, is a mean value of the generator speed time series data; By the same token, the fluctuation indexes of wind speed, temperature and humidity are respectively ; The fluctuation index of wind speed, temperature and humidity is analyzed to obtain a natural index, and the calculation formula is: wherein, is a natural index, are weight coefficients of wind speed, temperature and humidity in the natural index, respectively, , and ; The calculation formula of the fluctuation adjustment coefficient is: wherein is a fluctuation adjustment coefficient.

2. A method of predicting power curve of a wind turbine generator system as claimed in claim 1, wherein, The collection of the historical power generation data of the wind turbine generator set specifically includes: The current, voltage and generator speed data of the wind turbine generator set are collected through the power sensor and the speed sensor equipped on the wind turbine generator set, the power factor of the wind turbine generator set is obtained through the calculation type electric energy meter, and the formula for calculating the actual power of the wind turbine generator set is: wherein, is the actual power, is the voltage, is the current, is the power factor; Environmental data collection points are deployed on the upper part of the wind turbine tower, temperature sensors, humidity sensors and wind speed sensors are installed to collect temperature, humidity and wind speed data at the location of the wind turbine generator set; The collected historical power generation data is preprocessed, including removing outliers, missing values and normalization processing, calculating the Z-score of each parameter in the historical power generation data, when the absolute value of the Z-score of the parameter is When the absolute value of the Z-score of the parameter is When the absolute value of the Z-score of the parameter is After the processing of outliers and missing values, the Min-Max method is used for normalization processing, and all parameters are mapped to the interval [0, 1]. The historical power generation data after preprocessing is sorted into a historical data set according to the time stamp.

3. A method of predicting power curve of a wind turbine generator system as claimed in claim 1, wherein, The training of the multiple linear regression model specifically includes: The actual power, generator speed, wind speed, temperature and humidity data at the previous time point in the historical data set are taken as input, the actual power at the next time point is taken as label, a multiple linear regression model is selected, and the multiple linear regression model is represented as: wherein is the actual power at the moment, is the rotational speed of the wind turbine at the moment, is the actual power at the moment, is the wind speed at the moment, is the temperature at the moment, is the humidity at the moment, is the intercept, is the parameter influence coefficient, is the time variable, is the error term; The historical data set is divided into a training set and a test set, 80% for training and 20% for testing, the data in the training set is used to fit the regression model, the mean square error is selected as the loss function, the least square method is used to minimize the loss function, thereby determining the regression coefficients, the root mean square error in the training process is monitored in the training model, when the decline amplitude is less than 0.01 within 10 consecutive rounds, the training is stopped, which indicates that the model training is completed, after the model training is completed, the test set is used for model evaluation, and the determination coefficient value is calculated to evaluate the performance of the model.

4. A method of predicting power curve of a wind turbine generator system as claimed in claim 3 wherein, The training of the power prediction model specifically includes: The input sequence length of the long short-term memory network is set to time steps, Taking 10, the historical power generation data of the first 10 time points in the training set is input, the actual power data of the next time point in the training set is taken as the label, the long short-term memory network model is selected to train the power prediction model, and after the training is completed, the test set is input into the trained power prediction model for testing. The mean square error is selected as the loss function, and when the loss of the verification set decreases by less than 0.001 in the next 10 consecutive rounds, it indicates that the power quality prediction model training is completed.

5. A method of predicting power curve of a wind turbine generator system as claimed in claim 1, wherein, The actual power corrected by the fluctuation adjustment coefficient specifically includes: The current time point power generation data of the wind turbine generator set is collected, including generator speed, wind speed, temperature, humidity and actual power data at the current time point, the power generation data at the previous multiple time points and the current time point are input into the trained power prediction model, the actual power predicted by the model is obtained, and the actual power predicted by the multiple linear regression model and the power prediction model is corrected by the fluctuation adjustment coefficient, and the formula is: wherein, is the actual power corrected by the wave adjustment coefficient, is the actual power predicted by the multiple linear regression model, is the actual power predicted by the power prediction model; The actual power corrected by the fluctuation adjustment coefficient is applied to the power curve of the wind turbine generator set at the next time point, and the actual power corrected by the fluctuation adjustment coefficient is compared with the preset power threshold, if the actual power corrected by the fluctuation adjustment coefficient is lower than the preset power threshold, the wind turbine generator set continues to be monitored, if the actual power corrected by the fluctuation adjustment coefficient is higher than the preset power threshold, an abnormal early warning is issued.

6. A wind power plant power curve prediction device, characterized by, The wind turbine generator set power curve prediction device is used to realize the wind turbine generator set power curve prediction method in any one of claims 1-5, comprising: a historical data collection module for collecting historical power generation data of the wind turbine generator set, the historical power generation data including actual power at each time point in the power curve of the generator set, generator speed and environmental data, the environmental data including wind speed, temperature and humidity; The regression model construction module is configured to construct and train a multiple linear regression model by taking the actual power, the generator speed and the environmental data at a previous time in the historical power generation data as inputs and taking the actual power at a next time as a label; The power prediction model building module is used to analyze historical power generation data from previous periods. Using historical power generation data at each moment as input and the actual power at the next moment as a label, a power prediction model is constructed and trained based on a long short-term memory network model. The fluctuation adjustment coefficient calculation module is configured to calculate the fluctuation adjustment coefficient at the current time according to the generator speed, the wind speed, the temperature and the humidity data at the current time. The real-time data prediction and correction module is used to collect data at the current time and in the previous time. The power generation data at each moment is input into the multiple linear regression model and the power prediction model to obtain the actual power predicted by the two models for the next moment. Then, the actual power after the fluctuation adjustment coefficient is calculated by combining the fluctuation adjustment coefficient, and the corrected actual power is applied to the power curve of the wind turbine generator.

Images