Wind power output prediction method, device, system and computer program product

Abstract

The application discloses a wind power output prediction method, device, system and computer program product, and relates to the technical field of wind power. The method comprises the following steps: obtaining historical operation data of a wind turbine generator; and performing output feature prediction on the historical operation data by using a target extreme gradient boosting model to obtain predicted output information of the wind turbine generator. The historical operation data comprises a plurality of data samples, the target extreme gradient boosting model is obtained by training based on a target loss function, the target loss function is constructed by an experience loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the experience loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples by using a target control coefficient, and the regularization constraint term is used to limit the complexity of the target extreme gradient boosting model within a preset fitting range. The application solves the technical problems of low accuracy and poor adaptability of the wind power output prediction scheme in the related art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wind power, in particular to a wind power output prediction method, device, system and computer program product. BACKGROUND

[0002] With the development of new energy power generation technology, wind power generation technology has developed rapidly. Since wind power generation has intermittency and randomness, accurate prediction of wind power output can help power dispatching agencies arrange power grid operation in advance, ensure power supply and demand balance, and avoid power grid instability caused by wind power output fluctuations.

[0003] In related technologies, wind power output is mainly predicted based on a time series prediction model. A linear model is established using historical data provided by a wind turbine to fit the trend of time series data. However, since different data in the historical data have different physical meanings, the related technology cannot effectively fit data with different physical meanings, resulting in low accuracy and poor adaptability of the prediction result of wind power output, which is difficult to meet the operation and dispatching requirements of the wind farm.

[0004] To address the above problems, no effective solutions have been proposed so far. SUMMARY

[0005] The embodiments of the present application provide a wind power output prediction method, device, system and computer program product to at least solve the technical problems of low accuracy and poor adaptability of the wind power output prediction scheme in related technologies.

[0006] According to an aspect of an embodiment of the present application, a wind power output prediction method is provided, comprising: obtaining historical operation data of a wind turbine; predicting output characteristics of the historical operation data using a target extreme gradient boosting model to obtain predicted output information of the wind turbine; wherein the historical operation data comprises a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an experience loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the experience loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples through a target control coefficient, and the regularization constraint term is used to limit the complexity of the target extreme gradient boosting model within a preset fitting range.

[0007] Optionally, the plurality of feature parameters in each data sample comprise wind speed, wind direction angle, ambient temperature, generator speed, and generator power; the historical operation data of the wind turbine generator is obtained by: obtaining sampling data corresponding to each feature parameter of the wind turbine generator in a historical time domain; performing a data preprocessing operation on the sampling data to update the sampling data, wherein the data preprocessing operation comprises data cleaning, missing value processing, and outlier processing; and performing mean value calculation on the sampling data corresponding to each feature parameter based on a preset value range to obtain the plurality of data samples.

[0008] Optionally, the output feature analysis of the historical operation data by the target extreme gradient boosting model to obtain the predicted output information of the wind turbine generator comprises: inputting a target data sample in the plurality of data samples into the target extreme gradient boosting model to construct a plurality of base learners, wherein each base learner in the plurality of base learners is configured to analyze and predict the plurality of feature parameters in the target data sample to obtain a parameter prediction result; and performing summation calculation on the parameter prediction result to obtain the predicted output information.

[0009] Optionally, the base learner is a classification and regression tree model, and the regularization constraint term is configured to constrain leaf node parameters of the classification and regression tree model.

[0010] Optionally, constructing a target base learner in the plurality of base learners comprises: determining a tree split node based on a preset error function, a preset split criterion, and a feature value range of the target data sample; and constructing the target base learner according to the tree split node.

[0011] Optionally, the empirical loss function is obtained by accumulating a plurality of sub-functions corresponding to the plurality of data samples, a target sub-function in the plurality of sub-functions is configured to determine a target error between the target data sample and a parameter prediction result corresponding to the target data sample, the target error is a square error constrained by a target control coefficient, and the target error is configured to determine whether the target data sample is an abnormal sample.

[0012] Optionally, the regularization constraint term is obtained by accumulating a plurality of regularization terms corresponding to the plurality of base learners, and a target regularization term in the plurality of regularization terms is configured to constrain a number of leaf nodes and a leaf node weight coefficient corresponding to the target base learner.

[0013] According to another aspect of the embodiments of the present application, a wind power output prediction device is also provided, comprising: an acquisition module configured to acquire historical operation data of a wind turbine generator; and a prediction module configured to perform output feature prediction on the historical operation data by using a target extreme gradient boosting model to obtain predicted output information of the wind turbine generator; wherein the historical operation data comprises a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an empirical loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the empirical loss function is configured to control a sensitivity of the target extreme gradient boosting model to abnormal samples by using a target control coefficient, and the regularization constraint term is configured to limit a complexity of the target extreme gradient boosting model within a preset fitting range.

[0014] According to another aspect of the embodiments of the present application, a wind power output prediction system is also provided, comprising a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to perform the wind power output prediction method of any one of the above.

[0015] According to another aspect of the embodiments of the present application, a wind power output prediction computer program product is also provided, comprising a computer program, the computer program is executed by a processor to implement the wind power output prediction method of any one of the above.

[0016] In the embodiments of the present application, the historical operation data of the wind turbine generator is acquired; the output feature prediction is performed on the historical operation data by using the target extreme gradient boosting model to obtain the predicted output information of the wind turbine generator; wherein the historical operation data comprises a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an empirical loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the empirical loss function is configured to control a sensitivity of the target extreme gradient boosting model to abnormal samples by using a target control coefficient, and the regularization constraint term is configured to limit a complexity of the target extreme gradient boosting model within a preset fitting range. Thus, the present application achieves the purpose of obtaining the predicted output information of the wind turbine generator more accurately, thereby realizing the technical effect of improving the accuracy and adaptability of the wind power output prediction scheme, and further solving the technical problems of low accuracy and poor adaptability of the wind power output prediction scheme in the related art. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which are included to provide a further understanding of the present application and are incorporated in and constitute a part of this application, illustrate embodiments of the present application and serve to explain the present application. In the drawings:

[0018] Figure 1is a hardware structure block diagram of an optional terminal device for a wind power output prediction method according to an embodiment of the present application;

[0019] Figure 2 is a flow chart of a wind power output prediction method according to an embodiment of the present application;

[0020] Figure 3 is a structure block diagram of a wind power output prediction device according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable persons skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by persons skilled in the art without creative work should fall within the scope of protection of the present application.

[0022] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or a chronological sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] According to an embodiment of the present application, an embodiment of a wind power output prediction method is provided. It should be noted that the steps shown in the flow chart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flow chart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0024] Figure 1 is a hardware structure block diagram of an optional terminal device for a wind power output prediction method according to an embodiment of the present application, as Figure 1As shown, the terminal device 10 can include one or more processors 102 (the processor 102 can include, but not limited to, a microcontroller unit (MCU) or a field programmable gate array (FPGA) and the like processing device), a memory 104 for storing data, and a transmission device 106 for communication function. In addition, it can also include a display device 110, an input / output device 108 (I / O device), a universal serial bus (USB) port (which can be included as one of the ports of the computer bus, not shown in the figure), a network interface (not shown in the figure), a power supply (not shown in the figure) and / or a camera (not shown in the figure). Those skilled in the art can understand that Figure 1 The structure shown is only schematic, which does not limit the structure of the terminal device 10 described above. For example, the terminal device 10 can also include more or less components than those shown in the figure, or have a different configuration from that shown in the figure. Figure 1 For example, the terminal device 10 can also include more or less components than those shown in the figure, or have a different configuration from that shown in the figure. Figure 1 For example, the terminal device 10 can also include more or less components than those shown in the figure, or have a different configuration from that shown in the figure.

[0025] It should be noted that the one or more processors 102 and / or other data processing circuits described above can be embodied in whole or in part as software, hardware, firmware or any other combination. In addition, the data processing circuit can be a single independent processing module, or all or part of any one of the other elements combined into the terminal device 10 (or mobile device).

[0026] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage means corresponding to the wind power output prediction method in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implements the wind power output prediction method described above. The memory 104 can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 can further include a memory remotely arranged with respect to the processor 102, which can be connected to the terminal device 10 through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network and a combination thereof.

[0027] The transmission device 106 is configured to receive or send data via a network. The network can include a wireless network provided by a communication provider of the terminal device 10. In an example, the transmission device 106 includes a network interface controller (NIC) configured to connect to other network devices via a base station to communicate with the Internet. In an example, the transmission device 106 can be a radio frequency (RF) module configured to communicate with the Internet via a wireless manner.

[0028] In the above operating environment, the embodiments of the present application provide a wind power output prediction method as shown in Figure 2 Figure 2 is a flowchart of a wind power output prediction method according to an embodiment of the present application, as shown in Figure 2 The method includes the following implementation steps:

[0029] Step S201, obtaining historical operation data of a wind turbine generator.

[0030] Step S202, predicting output characteristics of the historical operation data by using a target extreme gradient boosting model to obtain predicted output information of the wind turbine generator.

[0031] In the above wind power output prediction method, the historical operation data includes a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an empirical loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the empirical loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples by using a target control coefficient, and the regularization constraint term is used to limit the complexity of the target extreme gradient boosting model within a preset fitting range.

[0032] The historical operation data can be a plurality of types of data collected during the operation of the wind turbine generator in a period of time, which can include but is not limited to wind speed, wind direction angle, ambient temperature, generator speed, generator power, and generator oil temperature. By analyzing the historical data, the operation rules of the wind turbine generator under different environmental conditions can be learned, thereby providing data basis for future prediction.

[0033] The predicted output information can be predicted power output information of the wind turbine generator at a future time, which can be obtained by using the historical operation data of the wind turbine generator and the target extreme gradient boosting model.

[0034] ​Specifically, a plurality of data samples of the wind turbine is acquired, a target loss function is constructed according to an experience loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, and a target extreme gradient boosting model is trained based on the plurality of data samples of the wind turbine and the target loss function by using a deep learning method.

[0035] In an exemplary application scenario, the target loss function can be as shown in formula (1).

[0036]

[0037] In formula (1), represents an experience loss function of the model, n represents a sample quantity, Ω(f s ) represents a regularization constraint term of the s-th model.

[0038] It is easy to understand that, in the embodiments of the present application, by introducing the target control coefficient and the regularization constraint term, on the one hand, the abnormal values in the wind turbine operation data can be effectively processed, and the predicted power output information of the wind turbine can be stably predicted even in the case that abnormal values appear in the wind turbine operation data, thereby improving the accuracy and reliability of the predicted power output information; on the other hand, the target extreme gradient boosting model can be prevented from overfitting, thereby improving the accuracy and stability of the predicted power output information. In addition, the embodiments of the present application can make full use of the information in the historical operation data, can improve the prediction effect of the target extreme gradient boosting model, and improve the accuracy of the predicted power output information.

[0039] The above method of the embodiments of the present application will be further introduced below.

[0040] Optionally, the plurality of feature parameters in each data sample include: wind speed, wind direction angle, environment temperature, generator speed, and generator power; and in the step S201 of acquiring the historical operation data of the wind turbine, the following execution steps can be further included:

[0041] In step S211, sampling data corresponding to each feature parameter of the wind turbine in the historical time domain is acquired.

[0042] In step S212, a data preprocessing operation is performed on the sampling data, and the sampling data is updated, wherein the data preprocessing operation includes: a data cleaning operation, a missing value processing operation, and an abnormal value processing operation.

[0043] In step S213, based on a preset value interval, mean calculation is performed on the sampling data corresponding to each feature parameter to obtain a plurality of data samples.

[0044] The data cleaning operation described above can be used to remove noise and irrelevant data in the data set. The missing value processing operation described above can be used to process records containing missing values due to device measurement interruption, data transmission failure, etc., such as filling missing values with average values, predicting filling with machine learning algorithms, etc. The outlier processing operation described above can be used to identify and process data points that are significantly different from normal values, such as filtering out abnormal data points due to measurement errors or device failure.

[0045] Specifically, the data acquisition and monitoring system of the wind turbine will periodically collect and record the sampling data of various sensors, including wind speed, wind direction angle, environmental temperature, generator speed, generator power and other characteristic parameters, to record the operating state of the wind turbine under different environmental conditions. By regularly collecting the sampling data of each characteristic parameter of the wind turbine in the historical time domain, sufficient training data can be provided for subsequent model training. Further, the sampling data is sequentially subjected to data cleaning operation, missing value processing operation and outlier processing operation to realize data preprocessing operation on the sampling data, which can improve the quality of the data set and ensure the effectiveness of the model training data. Further, in order to simplify the data and reduce the volatility in time series prediction, the sampling data is divided into a preset value interval, and the mean value of the sampling data corresponding to each characteristic parameter is calculated to obtain a plurality of data samples. By calculating the mean value of all sampling data, a data sample representing the average state in this time period can be obtained, which can better reflect the average operating state of the wind turbine in a certain time range, and is helpful for the learning of the prediction model and the improvement of the prediction accuracy.

[0046] In an exemplary application scenario, a data acquisition and monitoring (e.g., Supervisory Control And Data Acquisition, SCADA) system installed in a wind farm is used. The wind farm has a total of 22 wind turbines, and each wind turbine records sampling data (denoted as SCADA data) in the middle of each month (i.e., from the 11th to the 20th of each month) with a recording interval of 10 minutes. The SCADA data can include wind speed v1, wind direction angle ω1, environmental temperature temp, generator speed v2, generator power and other characteristic data. In this application scenario, the preset value interval (denoted as t) is set to 30 minutes, and each characteristic data sample value in the value interval t is represented by the mean value of the corresponding characteristic data sampling value in the SCADA data in the value interval t.

[0047] Each characteristic data sample value in the above-mentioned value interval t can be calculated by formula (2).

[0048]

[0049] In formula (2), x t represents each feature data sample value in the value interval t, x t,1 represents the first sample value of the feature data in the corresponding SCADA data in the value interval, x t,2 represents the second sample value of the feature data in the corresponding SCADA data in the value interval, x t,3 represents the third sample value of the feature data in the corresponding SCADA data in the value interval.

[0050] It is easy to understand that by the above steps S211 to S213, the embodiment of the application can improve the quality of the sampling data by obtaining detailed historical operation data, performing data preprocessing, and calculating the mean value of the sampling data based on a preset time interval, ensure the effectiveness of the model sampling data, and obtain data that better reflects the average operation state of the wind turbine in a certain time range.

[0051] Optionally, in the step S202, the output feature analysis of the historical operation data by the target extreme gradient boosting model to obtain the predicted output information of the wind turbine can further include the following steps:

[0052] In step S221, input the target data sample in the plurality of data samples into the target extreme gradient boosting model to construct a plurality of base learners, wherein each base learner in the plurality of base learners is used to analyze and predict a plurality of feature parameters in the target data sample to obtain a parameter prediction result.

[0053] In step S222, sum the parameter prediction results to obtain the predicted output information.

[0054] The parameter prediction result can be a predicted value related to the predicted output information predicted by each base learner according to the input plurality of feature parameters.

[0055] Specifically, the target data sample in the plurality of data samples is input into the target extreme gradient boosting model as a training data set, and a plurality of base learners are gradually constructed by iteration. In the construction process of each base learner, the value range of the feature parameters is divided to find the optimal split point to reduce the prediction error. Each base learner constructed will predict the feature parameters in the data sample. In each step of model training, a new base learner is added to the existing model to correct the prediction error of the current model. Each base learner will give a predicted value for the plurality of feature parameters in the target data sample. Further, the predicted values output by each base learner are summed to calculate the predicted output information.

[0056] It should be noted that the target data sample can be any one of the plurality of data samples, and the target data sample can include a plurality of feature parameters. By inputting each data sample in the plurality of data samples as a target data sample into the target extreme gradient boosting model, a plurality of base learners corresponding to each data sample can be constructed.

[0057] In an exemplary application scenario, the target extreme gradient boosting model can be an eXtreme Gradient Boosting (XGBoost) algorithm model. The XGBoost algorithm model is improved. The target data sample in the plurality of data samples is input into the improved XGBoost algorithm model, and a plurality of base learners are constructed. Each base learner analyzes and predicts a plurality of feature parameters in the target data sample to obtain a parameter prediction result corresponding to the base learner. The parameter prediction results of each base learner are summed to obtain prediction output information.

[0058] The prediction output information can be obtained by formula (3).

[0059]

[0060] In formula (3), pred represents the prediction output information, S represents the number of base learners, and f represents the prediction result of the i-th feature parameter of the s-th base learner. s (x i )

[0061] It is easy to understand that by the steps S221 to S222, the embodiments of the present application not only consider the time series characteristics of each feature data, but also use the multi-dimensional feature data of the wind turbine. By constructing and combining a plurality of base learners, the prediction output information of the wind turbine can be more accurately and comprehensively obtained.

[0062] Optionally, in the wind power output prediction method, the base learner is a classification and regression tree model, and the regularization constraint term is used to constrain the leaf node parameters of the classification and regression tree model.

[0063] The classification and regression tree model can be used to solve classification and regression problems. The classification and regression tree model can be constructed by recursive partitioning, purity maximization, and pruning. The recursive partitioning can start from the root node of the tree model. Each time, the optimal feature and split point are selected to divide the data set into two subsets, and the recursion is gradually performed until the stop condition is met (such as reaching the preset maximum depth of the tree, or the number of samples in the subset is less than a certain threshold). The purity maximization can use the square error to obtain the minimum error of the node. The pruning can be used to prevent model overfitting.

[0064] The leaf node can be the end node of the tree model, which can be used to make a final decision or prediction. The leaf node parameters can be determined through learning of the model on the training data set.

[0065] Specifically, the base learner is set as a CART model, and a regularization constraint term is introduced to constrain the leaf node parameters in order to control the complexity of the model and improve the generalization ability.

[0066] It is easy to understand that by taking the CART model as the base learner and introducing the regularization constraint term to constrain the leaf node parameters, the improved XGBoost model can not only learn the complex patterns in the data when processing the wind power prediction problem, but also effectively control the model complexity and improve the prediction accuracy and stability.

[0067] Optionally, in the wind power output prediction method, the target base learner in the plurality of base learners can further include the following method steps:

[0068] Step S231, determining a tree split node based on a preset error function, a preset split criterion, and a feature value range of the target data sample;

[0069] Step S232, constructing the target base learner according to the tree split node.

[0070] The preset error function can be an index for measuring the difference between the model prediction result and the actual result, which can guide the learning process of the model. The preset split criterion can be a rule for deciding how to split the node during the construction of the decision tree. The tree split node can be completed by calculating the split gain value of each possible split, which can be used to represent the change of the loss function before and after the split under the constraint of the regularization term.

[0071] Specifically, by traversing the value range of all features, each possible split point is evaluated, the change of the loss function before and after the split is calculated, and the tree split node is determined. Further, according to the tree split node, the feature value range of the target data sample is divided to obtain a plurality of sample subspaces, so as to construct the target base learner.

[0072] In an exemplary application scenario, the preset error function can be a square loss function, and when splitting a split point (such as an s split point) of a feature (denoted as j) of a target data sample, the input target data sample is divided into two sample subspaces according to the s split point, and further, the nodes of the same layer are sequentially split according to the preset splitting criterion to generate nodes of a new layer, until the depth of the tree reaches the preset maximum depth.

[0073] The square loss function is shown in formula (4).

[0074]

[0075] In formula (4), y i represents the actual output information.

[0076] The input target data sample is divided into two sample subspaces according to the s split point, as shown in formula (5).

[0077] R L (j,s)={x|xj≤s},R R (j,s)={x|xj>s} Formula (5)

[0078] In formula (5), R L (j,s) represents the left subspace, and R L (j,s) represents the right subspace.

[0079] The preset splitting criterion can be to split the node at the corresponding feature value when the square error of the two subspaces is minimized, as shown in formula (6).

[0080]

[0081] In formula (6), y L represents the mean of the sample data in the left subspace, and y R represents the mean of the sample data in the right subspace.

[0082] It is easy to understand that through the above steps S231 to S232, in the embodiment of the application, the tree split node can be more accurately selected through the preset error function and splitting criterion, the prediction accuracy and efficiency of the model are improved, and the model can be better applied to real-time output prediction and operation scheduling adjustment of a wind farm.

[0083] Optionally, in the wind power output prediction method, the experience loss function is obtained by accumulating a plurality of sub-functions corresponding to a plurality of data samples, a target sub-function in the plurality of sub-functions is used to determine a target error between a target data sample and a parameter prediction result corresponding to the target data sample, the target error is a square error subject to a target control coefficient constraint, and the target error is used to determine whether the target data sample belongs to an abnormal sample.

[0084] The target control coefficient can be a coefficient used to adjust the balance between the model experience loss function and the regularization term, the target control coefficient can include a plurality of control coefficients, and the target control coefficient can be adjusted according to actual conditions. By adjusting the target control coefficient, the training process and the final performance of the model can be directly affected.

[0085] The target sub-function can be used to measure the error between the prediction result of the target data sample and the actual result.

[0086] In an exemplary application scenario, the experience loss function is as shown in formula (7).

[0087]

[0088] In formula (7), α represents a first control coefficient, and β represents a second control coefficient.

[0089] According to the additive model, at the s-th step, Taylor expansion is performed on in formula (1) to obtain formula (8).

[0090]

[0091] In formula (8), C represents a constant term generated in the calculation process of the model.

[0092] It is easy to understand that in the embodiments of the present application, by improving the experience loss function, the model can not only more effectively learn the normal samples in the historical operation data of the wind farm unit, but also more effectively process the abnormal samples, reduce the influence of the abnormal values on the prediction performance of the model, and thus improve the robustness and prediction accuracy of the model.

[0093] Optionally, in the wind power output prediction method, the regularization constraint term is obtained by accumulating a plurality of regularization terms corresponding to a plurality of base learners, and a target regularization term in the plurality of regularization terms is used to constrain the number of leaf nodes and the leaf node weight coefficient corresponding to a target base learner.

[0094] In an exemplary application scenario, a regularization constraint term is added to the number of leaf nodes and the parameters corresponding to the leaf nodes in formula (8) to obtain formula (9).

[0095]

[0096] In formula (9), g i represents a first derivative value of L(θ), h i represents a second derivative value of L(θ), K represents the number of leaf nodes, and r j represents a parameter corresponding to the jth leaf node, γ represents a constraint on the number of leaf nodes, and λ represents a constraint on the parameter of the leaf node.

[0097] After performing expansion calculation on all the leaf nodes in formula (9), formula (10) can be obtained.

[0098]

[0099] After formula (10) is arranged, formula (11) can be obtained.

[0100]

[0101] In formula (11), G j represents the square of the sum of the first derivatives of the jth leaf node.

[0102] After the partial derivative of formula (11) is obtained, formula (12) can be obtained.

[0103]

[0104] After the partial derivative value of formula (12) is set to 0, formula (13) can be obtained.

[0105]

[0106] After formula (13) is substituted into formula (11) and arranged, formula (14) can be obtained.

[0107]

[0108] In formula (14), T represents a value interval.

[0109] Thus, the split benefit value corresponding to each tree split node can be obtained, as shown in formula (15).

[0110]

[0111] In formula (15), G L represents the square of the sum of the first derivatives of the left child node after splitting, G R represents the square of the sum of the first derivatives of the right child node after splitting, H L represents the square of the sum of the second derivatives of the left child node after splitting, H R represents the square of the sum of the second derivatives of the right child node after splitting.

[0112] It is easy to understand that, in the embodiments of the application, by limiting the number of leaf nodes and the weight coefficient, the complexity of the model can be effectively controlled, overfitting can be avoided, and the stability and reliability of the model can be improved.

[0113] In an application scenario, in order to evaluate the prediction performance of the model of the application on predicting the predicted power information,

[0114] The mean squared error (MSE), the mean absolute error (MAE), and the improved mean absolute percentage error (MAPE) are used as evaluation indexes. The historical operation data of 22 wind turbines in 12 value intervals are obtained, and the predicted power information output by the support vector regression (SVR) model, the predicted power information output by the kernel ridge regression (KRR) model, and the predicted power information output by the XGBoost model of the application are compared and analyzed.

[0115] The MSE value can be calculated by formula (16).

[0116]

[0117] In formula (16), N represents the total amount of data of the test sample.

[0118] The MAE value can be calculated by formula (17).

[0119]

[0120] The MAPE value can be calculated by formula (18).

[0121]

[0122] In order to reduce the influence of abnormal values in the test data on the MAPE value, formula (18) is improved, and the mean value of the test sample data is used instead of the true value of each test sample data, as shown in formula (19).

[0123]

[0124] In formula (19), represents the mean value of the test sample.

[0125] Table 1 is the evaluation index value of the predicted output information of 22 wind turbines predicted by the XGBoost model of the present application on 12 different value intervals, as shown below.

[0126] Table 1

[0127]

[0128] Table 2 is the evaluation index value of the predicted output information of 22 wind turbines predicted by the SVR model on 12 different scheduling window value intervals, as shown below.

[0129] Table 2

[0130]

[0131]

[0132] Table 3 is the evaluation index value of the predicted output information of 22 wind turbines predicted by the KRR model on 12 different scheduling window value intervals, as shown below.

[0133] Table 3

[0134]

[0135] As can be seen from Table 1, with the increase of the value interval, the three evaluation index values of the XGBoost model of the present application on 12 value intervals all show an overall upward trend, among which the fluctuation range of MAPE is the smallest, and the maximum difference of the error is 5.8383, while the fluctuation range of RMSE and MAE is larger, and the maximum difference of the error is 76.4349 and 86.4641, respectively. In addition, as can be seen from Table 1, when the value interval increases from 1 to 5, the evaluation index value of the XGBoost model of the present application is always increasing; with the increase of the value interval, the three evaluation indexes all decrease; when the value interval increases to 8, the evaluation index value again changes from decreasing to increasing.

[0136] As can be seen from Table 2 and Table 3, the test errors of the KRR model and the SVR model on 12 value intervals change the same as the XGBoost, and generally show an upward trend,

[0137] In addition, the evaluation index of the KRR model and the SVR model also decreases when the value interval is equal to 5, and then shows an upward trend again.

[0138] By comparing the data of the same evaluation index in the three tables, it can be seen that the corresponding RMSE value of the XGBoost model of the present application on 12 value intervals is less than the corresponding RMSE value of the SVR model and the corresponding RMSE value of the KRR model.

[0139] The MAE values corresponding to the 12 value intervals of the XGBoost model of the present application are all less than the MAE values corresponding to the KRR model, although the MAE values corresponding to the XGBoost model of the present application are greater than the MAE values corresponding to the SVR model in some value intervals, the MAE values corresponding to the XGBoost model of the present application have smaller fluctuation and better stability in the change range of the 12 value intervals.

[0140] The MAPE values corresponding to the 12 value intervals of the XGBoost model of the present application are overall less than the RMSE values corresponding to the SVR model and the MAPE values corresponding to the KRR model, and the MAPE values corresponding to the XGBoost model of the present application have smaller fluctuation and better stability in the change range of the 12 value intervals.

[0141] In the embodiment of the present application, historical operation data of the wind turbine is obtained; the target extreme gradient boosting model is used to predict the output characteristics of the historical operation data to obtain the predicted output information of the wind turbine; wherein the historical operation data includes a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an experience loss function corresponding to a plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the experience loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples through a target control coefficient, and the regularization constraint term is used to limit the complexity of the target extreme gradient boosting model within a preset fitting range. Thus, the present application achieves the purpose of more accurately obtaining the predicted output information of the wind turbine, thereby realizing the technical effect of improving the accuracy and adaptability of the wind power output prediction scheme, and further solving the technical problems of low accuracy and poor adaptability of the wind power output prediction scheme in related technologies.

[0142] In the embodiment, a wind power output prediction device is also provided, which is used to realize the above-mentioned embodiments and preferred embodiments, and will not be described again. As used below, a "module" can be a combination of software and / or hardware that realizes a predetermined function. Although the device described in the following embodiments is preferably realized in software, realization in hardware, or a combination of software and hardware is also possible and is contemplated.

[0143] Figure 3 is a structural block diagram of a wind power output prediction device according to an embodiment of the present application, as Figure 3As shown, the device comprises: an acquisition module 301, configured to acquire historical operation data of a wind turbine; a prediction module 302, configured to predict output characteristics of the historical operation data by using a target extreme gradient boosting model to obtain predicted output information of the wind turbine; wherein the historical operation data comprises a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an empirical loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the empirical loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples through a target control coefficient, and the regularization constraint term is used to limit the complexity of the target extreme gradient boosting model within a preset fitting range.

[0144] Optionally, the plurality of characteristic parameters in each data sample comprise: wind speed, wind direction angle, ambient temperature, generator speed, and generator power; the acquisition module 301 can also be configured to acquire sampling data corresponding to each characteristic parameter of the wind turbine in a historical time domain; perform a data preprocessing operation on the sampling data, and update the sampling data, wherein the data preprocessing operation comprises: a data cleaning operation, a missing value processing operation, and an outlier processing operation; perform mean value calculation on the sampling data corresponding to each characteristic parameter based on a preset value interval to obtain the plurality of data samples.

[0145] Optionally, the prediction module 302 can also be configured to input a target data sample in the plurality of data samples into the target extreme gradient boosting model to construct a plurality of base learners, wherein each base learner in the plurality of base learners is used to analyze and predict the plurality of characteristic parameters in the target data sample to obtain a parameter prediction result; and perform summation calculation on the parameter prediction result to obtain the predicted output information.

[0146] Optionally, the base learner is a classification regression tree model, and the regularization constraint term is used to constrain the leaf node parameters of the classification regression tree model.

[0147] Optionally, in addition to all the above-mentioned modules, the wind power output prediction device further comprises a construction module (not shown in the figure), configured to: determine a tree splitting node based on a preset error function, a preset splitting criterion, and a characteristic value range of the target data sample; and construct a target base learner according to the tree splitting node.

[0148] Optionally, the empirical loss function is obtained by accumulating a plurality of sub-functions corresponding to the plurality of data samples, a target sub-function in the plurality of sub-functions is used to determine a target error between the target data sample and a parameter prediction result corresponding to the target data sample, the target error is a square error constrained by the target control coefficient, and the target error is used to determine whether the target data sample belongs to an abnormal sample.

[0149] Optionally, the regularization constraint term is accumulated by a plurality of regular terms corresponding to a plurality of base learners, and a target regular term in the plurality of regular terms is used to constrain the number of leaf nodes and the weight coefficient of the leaf nodes corresponding to the target base learner.

[0150] It should be noted that the above modules can be implemented by software or hardware, and for the latter, the following implementation manners can be used, but are not limited thereto: the above modules are located in the same processor; or the above modules are located in different processors in any combination.

[0151] Through the above-mentioned device technical solutions corresponding to the acquisition module and the prediction module, in the embodiment of the application, the acquisition module is used to acquire historical operation data of a wind turbine; the prediction module is used to predict output characteristics of the historical operation data by using a target extreme gradient boosting model to obtain predicted output information of the wind turbine; wherein the historical operation data includes a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an experience loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the experience loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples by using a target control coefficient, and the regularization constraint term is used to limit the complexity of the target extreme gradient boosting model within a preset fitting range.

[0152] In the above-mentioned wind power output prediction device provided by the embodiment of the application, the historical operation data of the wind turbine is acquired; the output characteristics of the historical operation data are predicted by using the target extreme gradient boosting model to obtain the predicted output information of the wind turbine; wherein the historical operation data includes a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an experience loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the experience loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples by using a target control coefficient, and the regularization constraint term is used to limit the complexity of the target extreme gradient boosting model within a preset fitting range. Thus, the purpose of obtaining the predicted output information of the wind turbine more accurately is achieved, thereby realizing the technical effect of improving the accuracy and adaptability of the wind power output prediction scheme, and further solving the technical problems of low accuracy and poor adaptability of the wind power output prediction scheme in the related art.

[0153] According to another aspect of the embodiment of the application, a wind power output prediction system is also provided, which includes a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to execute the wind power output prediction method of any one of the preceding aspects.

[0154] Optionally, in the embodiment, the processor can be configured to execute the following steps by a computer program: obtaining historical operation data of the wind turbine; and predicting output characteristics of the historical operation data by using a target extreme gradient boosting model to obtain predicted output information of the wind turbine; wherein the historical operation data comprises a plurality of data samples, the target extreme gradient boosting model is trained based on a target loss function, the target loss function is constructed by an empirical loss function corresponding to the plurality of data samples and a regularization constraint term corresponding to a plurality of base learners, the empirical loss function is used to control a sensitivity of the target extreme gradient boosting model to abnormal samples by using a target control coefficient, and the regularization constraint term is used to limit a complexity of the target extreme gradient boosting model within a preset fitting range.

[0155] Optionally, in the embodiment, specific examples in the embodiment can refer to the examples described in the above embodiments and optional implementation manners, which will not be described herein.

[0156] According to another aspect of the embodiments of the present application, a computer program product is further provided. Optionally, the computer program product can include a non-volatile computer readable storage medium, which can be used to store a computer program, and the computer program can be executed by a processor to implement the method provided by the above embodiments.

[0157] The serial numbers of the embodiments of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0158] In the above embodiments of the present application, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0159] In the several embodiments of the present application, it should be understood that the disclosed technology can be implemented in other ways. Of course, the unit described as the division is only a logical function division, and there can be other division manners in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, unit or module, and can be electrical or other forms.

[0160] 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, that is, they can be located in one place, or can be distributed on a plurality of units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment.

[0161] In addition, each function unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software function unit.

[0162] When the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a plurality of instructions to make a computer device (which can be a personal computer, a server or a network device, etc.) execute all or part of the steps of the method described in each embodiment of the present application. The foregoing storage medium includes: a U disk, a ROM, a RAM, a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0163] The above is only the preferred embodiment of the present application. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should be considered as the protection scope of the present application.

Claims

1. A method for predicting wind power output, characterized in that, include: Obtain historical operating data of wind turbine units; The target limit gradient improvement model is used to predict the power output characteristics of the historical operating data to obtain the predicted power output information of the wind turbine. The historical running data includes multiple data samples. The target extreme gradient boosting model is trained based on a target loss function. The target loss function is constructed from the empirical loss function corresponding to the multiple data samples and the regularization constraint term corresponding to the multiple base learners. The empirical loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples through the target control coefficient. The regularization constraint term is used to limit the complexity of the target extreme gradient boosting model to a preset fitting range. The data sample contains multiple feature parameters including: wind speed, wind direction angle, ambient temperature, generator speed, and generator power. Obtaining the historical operating data of the wind turbine includes: acquiring sampled data corresponding to each feature parameter in the historical time domain; performing data preprocessing operations on the sampled data to update the sampled data, wherein the data preprocessing operations include: data cleaning, missing value handling, and outlier handling; and calculating the mean of the sampled data corresponding to each feature parameter based on a preset value range to obtain the multiple data samples. The method of using a target limit gradient boosting model to analyze the power output characteristics of the historical operating data to obtain the predicted power output information of the wind turbine includes: inputting the target data sample from the plurality of data samples into the target limit gradient boosting model to construct the plurality of base learners, wherein each base learner is used to analyze and predict multiple feature parameters in the target data sample to obtain parameter prediction results; and summing the parameter prediction results to obtain the predicted power output information. The base learner is a classification and regression tree model, and the regularization constraint term is used to constrain the leaf node parameters of the classification and regression tree model; constructing the target base learner among the multiple base learners includes: determining the tree splitting node based on a preset error function, a preset splitting criterion, and the feature value range of the target data sample; and constructing the target base learner according to the tree splitting node. The empirical loss function is obtained by summing multiple sub-functions corresponding to the multiple data samples. The target sub-function among the multiple sub-functions is used to determine the target error between the target data sample and the parameter prediction result corresponding to the target data sample. The target error is the squared error constrained by the target control coefficient. The target error is used to determine whether the target data sample belongs to the abnormal sample. The regularization constraint term is obtained by accumulating multiple regularization terms corresponding to the multiple base learners. The target regularization term among the multiple regularization terms is used to constrain the number of leaf nodes and the leaf node weight coefficients corresponding to the target base learner.

2. A wind power output prediction device, characterized in that, include: The acquisition module is used to acquire historical operating data of wind turbine units; The prediction module is used to predict the output characteristics of the historical operating data using the target limit gradient improvement model, so as to obtain the predicted output information of the wind turbine. The historical running data includes multiple data samples. The target extreme gradient boosting model is trained based on a target loss function. The target loss function is constructed from the empirical loss function corresponding to the multiple data samples and the regularization constraint term corresponding to the multiple base learners. The empirical loss function is used to control the sensitivity of the target extreme gradient boosting model to abnormal samples through the target control coefficient. The regularization constraint term is used to limit the complexity of the target extreme gradient boosting model to a preset fitting range. Each data sample contains multiple feature parameters including: wind speed, wind direction angle, ambient temperature, generator speed, and generator power; the acquisition module is also used to acquire the sampling data corresponding to each feature parameter of the wind turbine in the historical time domain; perform data preprocessing operations on the sampling data to update the sampling data, wherein the data preprocessing operations include: data cleaning operations, missing value handling operations, and outlier handling operations; and calculate the mean of the sampling data corresponding to each feature parameter based on a preset value range to obtain the multiple data samples; The prediction module is further configured to input the target data sample from the plurality of data samples into the target limit gradient boosting model to construct the plurality of base learners, wherein each base learner is configured to analyze and predict multiple feature parameters in the target data sample to obtain parameter prediction results; and to sum the parameter prediction results to obtain the predicted output information. The base learner is a classification and regression tree model, and the regularization constraint term is used to constrain the leaf node parameters of the classification and regression tree model; the wind power output prediction device also includes a construction module, which is used to determine the tree splitting node based on a preset error function, a preset splitting criterion and the feature value range of the target data sample; and construct the target base learner based on the tree splitting node. The empirical loss function is obtained by summing multiple sub-functions corresponding to the multiple data samples. The target sub-function among the multiple sub-functions is used to determine the target error between the target data sample and the parameter prediction result corresponding to the target data sample. The target error is the squared error constrained by the target control coefficient. The target error is used to determine whether the target data sample belongs to the abnormal sample. The regularization constraint term is obtained by accumulating multiple regularization terms corresponding to the multiple base learners. The target regularization term among the multiple regularization terms is used to constrain the number of leaf nodes and the leaf node weight coefficients corresponding to the target base learner.

3. A wind power output prediction system, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the wind power output prediction method of claim 1.

4. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the wind power output prediction method of claim 1.

Images