Wind power low output event discrimination and power prediction model training method and device

Abstract

The invention provides a wind power low output event discrimination and power prediction model training method and device, and relates to the technical field of wind power prediction, and the method comprises the steps: carrying out the principal component analysis of historical wind power data, clustering the principal component analysis result, determining the data corresponding to a wind power low output event, and forming a sample set; labeling the wind power low output event type for at least one meteorological element in the sample set and the wind power point corresponding to the at least one meteorological element; and training the wind power low-output event classification model based on the sample set, and training the wind power low-output power prediction models corresponding to different wind power low-output events based on the sample set, so that prediction can be performed according to different wind power low-output events in a subsequent power prediction scene, the precision of wind power short-term power prediction is improved, and the prediction efficiency is improved. And the positive significance of the wind power prediction result on power dispatching is improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of wind power prediction, and in particular to a method and device for identifying low-output wind power events and training a power prediction model. Background Art

[0002] Short-term wind power forecasting can provide wind power for the next 1 to 3 days, which provides a basis for rotating reserve regulation, peak and frequency regulation optimization, and load economic dispatch within a short time scale. Common wind power forecasting methods include physical methods, statistical methods, and artificial intelligence forecasting. Short-term wind power forecasting is often based on numerical weather prediction (NWP) as the input of the power forecasting model, establishing a nonlinear mapping model of meteorology-power, and achieving stable forecasting results.

[0003] In the future power system, the proportion of renewable energy will continue to increase, and its intermittency and volatility will bring huge challenges to the operation planning of the power system. In the ultra-short time scale, low wind power output events have little impact on the dispatch side, which is mainly due to the accurate prediction of wind power. However, in the short time scale, low wind power output events in unknown periods will seriously affect the reliability of regional power supply and cause system power imbalance in a short period of time.

[0004] Currently, most of the research based on short-term wind power forecasting focuses on weather type classification and data model mechanism driving, ignoring the definition of wind power low output events and the risks and challenges to power system planning and operation. At the same time, there is little research on the impact of wind power low output events on day-ahead wind power forecasting. Summary of the invention

[0005] The present disclosure provides a method and device for identifying low-output wind power events and training a power prediction model, so as to at least solve the above technical problems existing in the prior art.

[0006] According to a first aspect of the present disclosure, a method for identifying a wind power low output event and training a power prediction model is provided, comprising: Conduct principal component analysis on historical wind power data, cluster the results of principal component analysis, determine the data corresponding to low wind power output events, and form a sample set; Marking the type of wind power low output event for at least one meteorological element in the sample set and at least one wind power point corresponding to the meteorological element; Based on at least one meteorological element in the sample set and the wind power point corresponding to the at least one meteorological element, determine a wind power pixel map; input the wind power pixel map into a feature extraction network included in a wind power low output event classification model to obtain wind power data features after dimensionality reduction; input the wind power data features after dimensionality reduction into a classification network included in the wind power low output event classification model to determine the predicted wind power low output event type corresponding to the wind power pixel map; based on the annotation of the wind power low output event type corresponding to the wind power pixel map and the predicted wind power low output event type, adjust the weight of the wind power low output event classification model; At least one meteorological element corresponding to the first wind power low output event in the sample set and the wind power point corresponding to the at least one meteorological element are input into the first wind power low output power prediction model to obtain the power prediction results of the first wind power low output power prediction model for the at least one meteorological element and the at least one meteorological element; based on the power prediction results and the actual wind power point, the parameters of the first wind power low output power prediction model are adjusted.

[0007] According to a second aspect of the present disclosure, a method for identifying low-output wind power events and predicting power is provided, which is implemented based on a wind power low-output event classification model and a first wind power low-output power prediction model trained by the above method, and the method includes: Based on the wind power data to be predicted, determining a predicted wind power pixel map; Inputting the predicted wind power pixel map into the wind power low output event classification model to determine the wind power low output event classification result of the wind power data to be predicted; Determining a corresponding first wind power low output power prediction model based on the classification result of the wind power low output event of the wind power data to be predicted; The wind power data to be predicted is input into the first wind power low-output power prediction model to obtain the power in a future time period corresponding to the wind power data to be predicted.

[0008] The method for distinguishing wind power low output events and training a power prediction model disclosed in the present invention performs principal component analysis on historical wind power data, clusters the results of the principal component analysis, determines the data corresponding to the wind power low output events, and forms a sample set; labels the type of wind power low output event for at least one meteorological element in the sample set and at least one wind power point corresponding to the meteorological element; trains a wind power low output event classification model based on the sample set, and trains wind power low output power prediction models corresponding to different wind power low output events based on the sample set. In this way, in subsequent power prediction scenarios, predictions can be made based on different wind power low output events, thereby increasing the accuracy of short-term wind power predictions and improving the positive significance of wind power prediction results for power dispatching.

[0009] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present disclosure will become readily understood. In the accompanying drawings, several embodiments of the present disclosure are shown in an exemplary and non-limiting manner, wherein: In the accompanying drawings, the same or corresponding reference numerals represent the same or corresponding parts.

[0011] Figure 1 A schematic diagram of a low wind power output event in the related art is shown; Figure 2 A first optional flow chart of the method for identifying low-output wind power events and training a power prediction model provided by an embodiment of the present disclosure is shown; Figure 3 A first optional flow chart of the method for identifying low-output wind power events and predicting power provided by an embodiment of the present disclosure is shown; Figure 4 A second optional flow chart of the method for identifying low-output wind power events and training a power prediction model provided by an embodiment of the present disclosure is shown; Figure 5 A schematic diagram of the structure of a classification network in an embodiment of the present disclosure is shown; Figure 6 A second optional flow chart of the method for identifying low-output wind power events and predicting power provided by an embodiment of the present disclosure is shown; Figure 7 A third optional flow chart of the method for identifying low-output wind power events and training a power prediction model provided by an embodiment of the present disclosure is shown; Figure 8 A schematic diagram of wind power low output event identification and power prediction provided by an embodiment of the present disclosure is shown; Fig. 9 The duration and frequency distribution of wind power low output events; Fig.10 This is the KPCA principal component contribution rate diagram; Fig.11 The classification result of wind power low output event - wind power stable low output event; Fig.12 The classification result of wind power low output event - wind power fluctuation low output event; Fig.13 This is the classification result of wind power low output event - wind power hybrid low output event I; Fig.14This is the classification result of wind power low output event - wind power hybrid low output event II; Fig.15 Pie-shaped heat map showing the accuracy of preliminary identification of low wind power output events; Fig.16 The statistical results of low output event discrimination for each model; Fig.17 is the grey correlation between meteorological factors and power series under each low output event; Fig.18 It is the trend diagram of power sequence and wind speed sequence under normal low output event; Fig.19 Direct forecast results of wind power for one week; Fig. 20 It is the power prediction result of different scenarios for three non-consecutive different output events; Fig.21 It is the error correction result of three non-consecutive different output events - fluctuating low output events; Fig. 22 It is the error correction result of three non-consecutive different output events - mixed low output event I; Fig.23 It is the error correction result of three non-consecutive different output events - mixed low output event II; Fig.24 It is the error correction result of three non-consecutive different output events - normal output event; Fig.25 Comparison of error evaluation indicators before and after power prediction error correction in various low-output scenarios; Fig.26 Comparison of error evaluation indicators before and after power prediction error correction in each scenario; Fig. 27 An optional structural schematic diagram of a wind power low output event identification and power prediction model training device provided by an embodiment of the present disclosure is shown; Fig.28 An optional structural schematic diagram of a wind power low output event identification and power prediction device provided by an embodiment of the present disclosure is shown; Fig.29 A schematic diagram of the structure of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0012] In order to make the purpose, features, and advantages of the present disclosure more obvious and easy to understand, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.

[0013] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it can be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0014] In the following description, the terms "first\second" involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It can be understood that "first\second" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present disclosure described herein can be implemented in an order other than that illustrated or described herein.

[0015] Unless otherwise defined, all technical and scientific terms used in this disclosure have the same meaning as those commonly understood by those skilled in the art to which this disclosure belongs. The terms used in this disclosure are only for the purpose of describing the embodiments of this disclosure and are not intended to limit this disclosure.

[0016] It should be understood that in the various embodiments of the present disclosure, the size of the serial number of each implementation process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure.

[0017] Before further describing the embodiments of the present disclosure in detail, the nouns and terms involved in the embodiments of the present disclosure are described. The nouns and terms involved in the embodiments of the present disclosure are subject to the following interpretations.

[0018] 1. Wind power low output event.

[0019] Figure 1 A schematic diagram of a low wind power output event in the related art is shown.

[0020] As an intermittent energy source, the volatility of wind power will lead to high / low wind power output scenarios. High wind power output events are unlikely to cause power supply risks to the power system. On the contrary, low wind power output events, especially extreme low wind power output events, will significantly affect the safe and stable operation of the system and even cause serious accidents. Existing studies rarely consider the impact of low wind power output events on the power system, and the definition of low wind power output events is not clear. Therefore, in the embodiments of the present disclosure, it is defined that within a certain duration range, the wind power output level is continuously below a certain power threshold, which is considered to be a low wind power output event. Figure 1 As shown, Indicates the lower limit of the minimum output threshold of wind power low output events, where Indicates the highest historical power value of the wind farm. Indicates the start time of the low wind power output event, Indicates the end time of the low wind power output event. The duration of represents the entire duration of the low wind power output event, while The wind power sequence between or greater than part).

[0021] 2. Division of wind power low output power sequence. Even though low wind power output events can be identified based on the initial discriminant model, the fluctuation of wind speed will lead to the existence of multiple types of low wind power output events. After studying a large number of sample sequences of low wind power output events, it is found that low wind power output events can be roughly divided into three types: 1) Low output type: During the entire wind power low output event, the wind power output is always at a low and relatively stable state; it includes four types: mixed low output I, mixed low output II, fluctuating low output and stable low output.

[0022] 2) Fluctuation type: Wind power fluctuates significantly during the period of low wind power output event. 3) Hybrid type: During the entire wind power low output event, the wind power includes both continuous low-level and stable output periods and fluctuating output processes.

[0023] Figure 2 A first optional flow chart of the method for identifying low-output wind power events and training a power prediction model provided by an embodiment of the present disclosure is shown, and will be described according to each step.

[0024] Step S201 , performing principal component analysis on historical wind power data, clustering the principal component analysis results, determining data corresponding to wind power low output events, and forming a sample set.

[0025] In some embodiments, the historical wind power data includes meteorological elements and wind power points corresponding to each wind power low output event. Specifically, the meteorological elements include at least one of air pressure, wind speed, wind direction, temperature and humidity; the wind power points include points obtained by sampling the power of the duration of the wind power low output event at a first frequency. For example, if a certain wind power low output event lasts for 4 hours, it can be sampled at a frequency of 15 minutes to obtain 16 power points. The various wind power low output events include low output type wind power low output events, fluctuating wind power low output events and hybrid wind power low output events.

[0026] In some embodiments, the historical wind power data may include multiple types of wind power low output events. Before model training, a carrier (hereinafter referred to as the carrier) that implements the wind power low output event discrimination and power prediction model training method first needs to classify the data in the historical wind power data. Specifically, in the disclosed embodiment, the kernel functions corresponding to the power segments corresponding to multiple wind power low output events in the historical wind power data are calculated; a first kernel matrix is ​​constructed based on the kernel functions corresponding to the power segments corresponding to the multiple wind power low output events; the eigenvalues ​​of each column of elements in the first kernel matrix are calculated; based on the eigenvalues ​​of each column of elements, the corresponding eigenvectors are determined; the eigenvectors are clustered, and based on the clustering results, the data corresponding to the wind power low output events are determined to form a sample set.

[0027] In specific implementation, based on any wind power low output event, Power point at moment and Power point at moment , calculate the kernel function . Determine the first kernel matrix based on the kernel function Among them, the power point and power point Power segments belonging to the same wind power low output event. , .

[0028] In some embodiments, the carrier calculates the eigenvalue of each column in the first kernel matrix and sorts them from large to small; determines the eigenvector corresponding to each column eigenvalue ,in, Characterization The eigenvalues ​​corresponding to the eigenvectors ( ), after normalizing all feature vectors, we get the sample set in the first Coordinates under principal components:

[0029] in, Characterizes the number of samples in the sample set, Characterization The first feature vector values, among which, The dimension of is consistent with the total sample size in the sample set. For the principal components.

[0030] In some embodiments, the carriers are sorted in descending order of characteristic values, at least one principal component in the front is determined as the input of the clustering model, and the output of the clustering model is determined as the data corresponding to the wind power low output event to form a sample set. Specifically, the carriers form a sample set based on meteorological elements and wind power points corresponding to different types of wind power low output events.

[0031] In some embodiments, the carrier can be a computer program, electronic circuit, database, mobile application, electronic device, cloud computing platform, distributed system, artificial intelligence framework, mathematical model, automation tool and microcontroller, etc., which can implement software or hardware of algorithm and method flow.

[0032] Step S202: marking the type of wind power low output event for at least one meteorological element in the sample set and at least one wind power point corresponding to the meteorological element.

[0033] In some embodiments, the carrier labels the data in the sample set to clarify the type of wind power low output event corresponding to each meteorological element and wind power point.

[0034] Step S203: determine a wind power pixel map based on at least one meteorological element in a sample set and a wind power point corresponding to the at least one meteorological element; input the wind power pixel map into a feature extraction network included in a wind power low output event classification model to obtain reduced-dimensional wind power data features; input the reduced-dimensional wind power data features into a classification network included in the wind power low output event classification model to determine a predicted wind power low output event type corresponding to the wind power pixel map; adjust the weight of the wind power low output event classification model based on the annotation of the wind power low output event type corresponding to the wind power pixel map and the predicted wind power low output event type.

[0035] In some embodiments, the wind power low output event identification and power prediction model includes a wind power low output event classification model and at least one wind power low output power prediction model.

[0036] In some embodiments, the carrier first trains a wind power low output event classification model; the wind power low output event classification model may include a feature extraction network and a classification network; the feature extraction network may be implemented by CNN, and the classification network may be implemented by BiGRU.

[0037] In some embodiments, the carrier constructs a wind power data matrix based on at least one meteorological element in a sample set and a wind power point corresponding to the at least one meteorological element, determines each element in the wind power data matrix as a pixel point, and constructs a wind power pixel map.

[0038] In some embodiments, the carrier inputs the wind power pixel map into a feature extraction network, extracts wind power data features of the wind power pixel map, reduces the dimension of the wind power data features, and obtains the reduced wind power data features; the reduced wind power data features are expressed in the form of a sequence or a vector; the reduced wind power data features are input into the classification network to obtain the output of the classification network; and it is determined that the type with the highest probability in the output of the classification network is the predicted wind power low output event type corresponding to the wind power pixel map.

[0039] In some embodiments, the carrier determines a first loss function of the wind power low output event classification model based on the annotation of the wind power low output event type corresponding to the wind power pixel map and the predicted wind power low output event type, and adjusts the weight of the wind power low output event classification model based on the first loss function. Optionally, the carrier may determine the first loss function based on cross entropy.

[0040] Step S204: input at least one meteorological element corresponding to the first wind power low output event in the sample set and the wind power point corresponding to the at least one meteorological element into the first wind power low output power prediction model to obtain the power prediction result of the first wind power low output power prediction model for the at least one meteorological element and the at least one meteorological element; adjust the parameters of the first wind power low output power prediction model based on the power prediction result and the actual wind power point.

[0041] In some embodiments, the carrier trains different wind power low output power prediction models based on different wind power low output event types. In the disclosed embodiment, taking the training of the first wind power low output power prediction model as an example, the first wind power low output power prediction model can be a model corresponding to any wind power low output event type.

[0042] In some embodiments, the carrier inputs the sample data corresponding to the first wind power low output event into the first wind power low output power prediction model, and the first wind power low output power prediction model generates a mixed kernel function based on the kernel matrix, random matrix, radial basis function and polynomial kernel function of the sample data corresponding to the first wind power low output event; and updates the parameters of the first wind power low output power prediction model based on the vector weighted average algorithm.

[0043] In this way, through the wind power low output event identification and power prediction model training method provided by the embodiment of the present disclosure, principal component analysis is performed on historical wind power data, and the principal component analysis results are clustered to determine the data corresponding to the wind power low output event to form a sample set; at least one meteorological element in the sample set and at least one wind power point corresponding to the meteorological element are labeled with the type of wind power low output event; a wind power low output event classification model is trained based on the sample set, and a wind power low output power prediction model corresponding to different wind power low output events is trained based on the sample set. In this way, in subsequent power prediction scenarios, predictions can be made based on different wind power low output events, thereby increasing the accuracy of short-term wind power predictions and improving the positive significance of wind power prediction results for power dispatching.

[0044] Figure 3 A first optional flow chart of the method for identifying low-output wind power events and predicting power provided by an embodiment of the present disclosure is shown, and will be described according to each step.

[0045] Step S301: determining a predicted wind power pixel map based on wind power data to be predicted.

[0046] In some embodiments, a matrix of wind power data to be predicted is constructed based on the wind power data to be predicted, and each element in the matrix represents the meteorological element of the wind power data to be predicted, or the wind power data that has been collected; each element in the matrix of wind power data to be predicted is used as a pixel point, and a predicted wind power pixel map is determined based on the matrix of wind power data to be predicted.

[0047] Step S302: input the predicted wind power pixel map into the wind power low output event classification model to determine the wind power low output event classification result of the wind power data to be predicted.

[0048] Step S303: determining a corresponding first wind power low output power prediction model based on the classification result of the wind power low output event of the wind power data to be predicted.

[0049] For the convenience of explanation, it is assumed here that the classification result of the wind power low output event of the wind power data to be predicted is the same as the wind power low output event type of the first wind power low output power prediction model. In actual applications, different wind power low output event types correspond to different wind power low output power prediction models, and the wind power low output power prediction model for power prediction should be determined based on the wind power low output event discrimination result (or classification result).

[0050] Step S304: input the wind power data to be predicted into the first wind power low-output power prediction model to obtain the power of the future time period corresponding to the wind power data to be predicted.

[0051] In some embodiments, the power of the future time period corresponding to the predicted wind power data obtained based on the first wind power low-output power prediction model may be biased, that is, different from the actual power of the future time period. The method may also include: obtaining at least two historical wind power data of the same type corresponding to the wind power low-output event classification result of the wind power data to be predicted; matching the wind power data to be predicted with the at least two historical wind power data of the same type to obtain at least one historical wind power data of the same type with a high matching value; based on the at least one historical wind power data of the same type, compensating the power of the future time period corresponding to the wind power data to be predicted, and determining the compensation result as the target power of the future time period corresponding to the wind power data to be predicted.

[0052] For example, after determining the classification result of the wind power low output event of the wind power data to be predicted based on step S302, at least two historical wind power data of the same type corresponding to the wind power low output event classification are obtained; the historical wind power data of the same type at least includes meteorological elements and at least includes power segments of the future time period.

[0053] During specific implementation, the grey correlation coefficient of each feature in the wind power data to be predicted is calculated; the grey correlation coefficients of all features are weighted based on the weight coefficient to obtain a first weighted grey correlation coefficient; the grey correlation coefficient of each feature in the at least two historical wind power data of the same type is calculated respectively, and weighted based on the weight coefficient to obtain at least one second weighted grey correlation coefficient; the first weighted grey correlation coefficient and the at least one second weighted grey correlation coefficient are matched to determine at least one historical wind power data of the same type having a high matching value with the wind power data to be predicted.

[0054] In some embodiments, an average error value is determined based on the at least one piece of historical wind power data of the same type; and power in a future time period corresponding to the wind power data to be predicted is compensated based on the average error value.

[0055] In this way, through the wind power low output event identification and power prediction method provided by the embodiment of the present disclosure, different wind power low output types have different error characteristics, and wind power sequences with similar types of wind power low output segments have certain similarities in error characteristics. Similar segments can be matched through the meteorological elements of the wind power sequence, and samples of similar scenes (or wind power low output event types) are selected by calculating the weighted grey correlation coefficient between the meteorological element matrix corresponding to the wind power low output event in the time period to be predicted and the historical sample set, so as to achieve effective error compensation. At the same time, there are a large number of scenes below the cut-in wind speed under the wind power low output event. At this time, for these scenes, the power can be set to zero to achieve secondary correction.

[0056] Figure 4A second optional flow chart of the method for identifying low-output wind power events and training a power prediction model provided by an embodiment of the present disclosure is shown, and will be described according to each step.

[0057] Step S401, determining a sample set.

[0058] In some embodiments, the first kernel matrix is ​​calculated by a kernel function , where the elements of the kernel matrix are: ,in , is the historical wind power data, is the corresponding kernel function; Calculate the kernel matrix n eigenvalues ​​(one eigenvalue is calculated for each column) and sorted from large to small. Find the eigenvector corresponding to each eigenvalue , Indicates The eigenvector corresponding to the eigenvalue is normalized; the original sample is obtained at the Coordinates under principal components: (1) in, Indicates the number of samples, ranging from , Indicates The first feature vector values, among which The dimensions of are consistent with the total sample size. For the corresponding The first few principal components after principal component feature extraction are used as the input of the clustering model to classify wind power low output events, and samples of different wind power low output event types are obtained to form a sample set.

[0059] Step S402: training a wind power low output event classification model.

[0060] In some embodiments, the carrier classifies (or discriminates) wind power low output events based on a wind power low output event classification model (CNN-BiGRU). Specifically, since there are two valley periods and two peak periods in a load (power) sequence in a day, for different peak periods, a method of separately correcting the peak period and the valley period is used to reduce harmful errors, wherein based on an analysis of a year's load data, the time when the load peak and valley period of each day in a year appear in a concentrated manner is taken as the load peak and valley moment, including load valley period I, load peak period I, load valley period II and load peak period II; the low output scenario involved in the wind power low output event is closely related to meteorological elements, especially wind speed. Based on the differences in meteorological elements, a preliminary wind power low output event discrimination can be achieved, and on this basis, the type of wind power low output event can be further analyzed. The disclosed embodiment adopts a feature extraction network (Convolutional Neural Network, CNN) to first extract implicit information of the wind power pixel map (the meteorological element matrix of the wind power sequence), and adopts a classification network (Bidirectional Gate Recurrent Unit (BiGRU)) to classify wind power low output events. This is mainly because wind power low output events have a chronological relationship in the time series. The use of BiGRU with a bidirectional memory characteristic can mine this temporal relationship to achieve accurate scene discrimination.

[0061] The principle of the feature extraction network (CNN network) is that the convolutional neural network has good feature extraction capabilities. It mainly includes input layer, convolution layer, pooling layer, fully connected layer and output layer. The convolution layer and pooling layer are alternately stacked to achieve the functions of feature extraction and dimensionality reduction. The specific process of two-dimensional convolution can be expressed as: (2) in, Indicates The pixels of the layer feature map, represents the activation function, and Respectively represent The input and output of the layer convolution kernel, represents the convolution operation, and Respectively represent Layer weights and thresholds.

[0062] Figure 5 A schematic diagram of the structure of a classification network in an embodiment of the present disclosure is shown.

[0063] In some embodiments, Figure 5As shown in the figure, BiGRU consists of two GRU stacks, and its output is determined by the two network structures. The classification network BiGRU network is an improvement and simplification of the GRU network based on the LSTM forget gate, which makes the GRU network better than the Long and Short-Term Memory networks (LSTM) network in performance. GRU merges the internal state vector and the output vector into a unified state vector , there are only two gates left, namely the reset gate and the update gate, which can save information from a long time ago. It can be expressed as: (3) Update Gate It can be expressed as: (4) Candidate hidden layer states It can be expressed as: (5) Hidden State It can be expressed as: (6) in, , Represents the reset gate and update gate in Output at the moment, and denote the parameter matrices of the reset gate and the update gate respectively, and Represent the residual vectors of the reset gate and update gate respectively, represents the activation function, is the hyperbolic activation function, express The input of the time model, Representation Model Network output at that moment.

[0064] Step S403: training a wind power low output power prediction model.

[0065] In some embodiments, the Kernel Extreme Learning Machine (KELM) is based on the Extreme Learning Machine (ELM), and introduces a kernel function to replace the random mapping between its input layer and hidden layer, which has stronger nonlinear mapping capabilities based on the ELM. The Hybri Kernel Extreme Learning Machine (HKELM) is based on the KELM, and fully combines the advantages of different kernel functions to make it have stronger learning and generalization capabilities.

[0066] Based on a given input sample, the input and output of KELM can be expressed as: (7) Where: is the regularization coefficient, and are the diagonal matrix and the target vector matrix respectively, is the kernel function, is the kernel matrix.

[0067] Kernel Matrix It can be expressed as: (8) in, Represents a random matrix in ELM.

[0068] HKELM introduces radial basis kernel function (RBF) and polynomial kernel function on the basis of single-core extreme learning machine to enhance the generalization of the model. The expression of its hybrid kernel function is as follows: (9) in, represents the weight coefficient of the kernel function, represents the radial basis kernel function, represents the polynomial kernel function.

[0069] However, the kernel parameters of HKELM and the weights and regularization coefficients of the kernel matrix have a great influence on its prediction performance. Therefore, the Weighted Mean of Vectors algorithm (INFO) is used in the embodiment of the present disclosure to optimize this hyperparameter of HKELM to achieve the best prediction effect. INFO is an improved weighted averaging method that applies the idea of ​​weighted averaging to solid structures and uses three core procedures: update rules, vector combination, and local search to update the position of vectors. In engineering problems, its convergence accuracy can reach 0.99% of the global optimal solution.

[0070] In this way, through the wind power low output event identification and power prediction model training method provided by the embodiment of the present disclosure, principal component analysis is performed on historical wind power data, and the principal component analysis results are clustered to determine the data corresponding to the wind power low output event to form a sample set; at least one meteorological element in the sample set and at least one wind power point corresponding to the meteorological element are labeled with the type of wind power low output event; a wind power low output event classification model is trained based on the sample set, and a wind power low output power prediction model corresponding to different wind power low output events is trained based on the sample set. In this way, in subsequent power prediction scenarios, predictions can be made based on different wind power low output events, thereby increasing the accuracy of short-term wind power predictions and improving the positive significance of wind power prediction results for power dispatching.

[0071] Figure 6 A second optional flow chart of the method for identifying low-output wind power events and predicting power provided by an embodiment of the present disclosure is shown, and will be described according to each step.

[0072] Step S601: determining a predicted wind power pixel map based on the wind power data to be predicted.

[0073] The specific step flow of step S601 is the same as that of step S301 and will not be repeated here.

[0074] Step S602: input the predicted wind power pixel map into the wind power low output event classification model to determine the wind power low output event classification result of the wind power data to be predicted.

[0075] The specific step flow of step S602 is the same as that of step S302 and will not be repeated here.

[0076] Step S603: determining a corresponding first wind power low output power prediction model based on the classification result of the wind power low output event of the wind power data to be predicted.

[0077] The specific step flow of step S603 is the same as that of step S303 and will not be repeated here.

[0078] Step S604: input the wind power data to be predicted into the first wind power low-output power prediction model to obtain the power in the future time period corresponding to the wind power data to be predicted.

[0079] The specific step flow of step S604 is the same as that of step S304 and will not be repeated here.

[0080] Step S605 , based on at least one piece of historical wind power data of the same type, the power of the future time period corresponding to the wind power data to be predicted is compensated, and a compensation result is determined as the target power of the future time period corresponding to the wind power data to be predicted.

[0081] Since the power sequence in the future time period is used as a predicted value when the type of wind power low output event is known, it is difficult to use it as a basis for classifying wind power low output events. The meteorological elements corresponding to each type of wind power low output event are significantly different, especially reflected in the fluctuation of wind speed. Therefore, the meteorological elements corresponding to each type of wind power low output event can be analyzed to match the wind power low output events in the future time period. The embodiment of the present disclosure selects grey correlation analysis to analyze the correlation between each meteorological sequence during each low output event, and then obtains the correlation between each meteorological element of the wind power low output event type, and establishes a weighted grey correlation coefficient for identifying the wind power low output event type in the future time period.

[0082] Assume there is a low wind power output event and , and their corresponding meteorological element matrices are expressed as follows: (10) (11) in, and Represents wind power low output event and The first meteorological element of the sequence, Represents the total number of meteorological elements, and Can be expressed as: (12) (13) in, and Respectively and The time series of the same meteorological element, Indicates the length of the meteorological sequence.

[0083] Based on this, the grey correlation coefficient between the meteorological element sequences of two wind power low output events can be calculated, and its expression is as follows: (14) in, Indicates low wind power output event and No. Features in Grey correlation coefficient of time; is a constant between (0,1], usually 0.5; express and No. Features in The absolute deviation of the moment, ; For wind power low output events and The average correlation coefficient of each moment can be obtained Grey relational coefficient of features , which is expressed as follows: (15) To measure wind power low output events and Features The comprehensive correlation degree of , which is expressed as follows: (16) in, They represent the weight coefficients of the grey correlation coefficients of each feature, and their sum is 1. When classifying wind power low output events in the future period, the weighted grey correlation coefficient of the meteorological elements of the event to be classified is calculated to determine which type of historical event is closer to, and the corresponding wind power low output event type is obtained.

[0084] Since different wind power low output events have different error characteristics, wind power sequences with similar wind power low output event segments have certain similarities in error characteristics. Similar segments can be matched through the meteorological elements of the wind power sequence. This can be done by calculating the weighted grey correlation coefficient between the meteorological element matrix corresponding to the wind power low output event in the forecast period and the historical sample set to select samples of similar scenarios to achieve effective error compensation. At the same time, there are a large number of scenarios with cut-in wind speeds below zero under wind power low output events. At this time, for these scenarios, the power can be set to zero to achieve secondary correction.

[0085] In this way, through the wind power low output event identification and power prediction method provided by the embodiment of the present disclosure, different wind power low output types have different error characteristics, and wind power sequences with similar wind power low output segments have certain similarities in error characteristics. Similar segments can be matched through the meteorological elements of the wind power sequence, and samples of similar scenes are selected by calculating the weighted grey correlation coefficient of the meteorological element matrix corresponding to the wind power low output event in the time period to be predicted and the historical sample set, so as to achieve effective error compensation. At the same time, there are a large number of scenes below the cut-in wind speed under the wind power low output event. At this time, for these scenes, the power can be set to zero to achieve secondary correction.

[0086] Figure 7 A third optional flow chart of the method for identifying low-output wind power events and training a power prediction model provided by an embodiment of the present disclosure is shown. Figure 8 A schematic diagram showing the wind power low output event identification and power prediction provided by the embodiment of the present disclosure is shown. Figure 7 and Figure 8 Provide explanation.

[0087] In some embodiments, the carrier first uses KPCA to extract features from historical wind power samples of historical wind power low output events, then uses the k-medoids clustering model to divide different types of wind power low output events, and determines similar types of wind power low output events in future periods based on the weighted grey correlation coefficient.

[0088] In some embodiments, the carrier performs an initial classification of the wind power low output time type for the meteorological process in the forecast period based on CNN-BiGRU, and preliminarily distinguishes between wind power normal output events and wind power low output events.

[0089] In some embodiments, the carrier performs wind power forecasting based on the INFO-HKELM model. Specifically, based on the classification results of the wind power low output event type provided by the wind power low output event classification model, a wind power low output power prediction model corresponding to the wind power low output event type is trained to perform preliminary power forecasting.

[0090] In some embodiments, considering the similarity of errors in similar scenarios, wind power data (including meteorological elements and power segments) of wind power low output events with similar meteorological elements are selected to perform error prediction, and the error is used to compensate the initial power prediction result ( Figure 8 As shown in the mechanism Ⅰ), for low wind power output events, when the wind speed is lower than the wind turbine cut-in speed (3m / s), the power point is set to zero (mechanism Ⅱ) to achieve secondary correction.

[0091] The specific implementation methods are as follows: Step S701: Obtain historical wind power data corresponding to the wind power low output event from all historical data.

[0092] In some embodiments, since the duration of wind power low output events varies, the embodiments of the present disclosure analyze the duration of wind power low output events based on a large amount of historical wind power data. Existing theory holds that a sustained output level below 6% is considered an extreme wind power low output event. The wind power low output events studied in the embodiments of the present disclosure include extreme wind power output events and relatively low wind power low output events. Therefore, the embodiments of the present disclosure set the wind power low output event power threshold Set to 10% of the maximum power threshold , the duration of low wind power output events was counted, such as Fig. 9 As shown, this is the maximum power threshold at a threshold of 10%. In the case of , the duration and frequency distribution of wind power low output events obtained by the threshold segmentation method are as follows: Fig. 9 It can be concluded that the duration of most wind power low output events is concentrated in 0~40h, among which short-duration wind power low output events are relatively concentrated, especially wind power low output events of about 4h, accounting for more than 80% of the total event frequency, and some extreme wind power low output events last for more than 80~100h. In order to extract as much complete low output event power and weather process as possible, the embodiment of the present disclosure uses a sliding window method to extract a continuous 4h low output wind power sequence and corresponding wind power historical data at intervals of 15min for example verification. Among them, the historical wind power data can be in the form of a matrix, i.e., a wind power matrix, and each element in the wind power matrix represents a meteorological element or a power point.

[0093] In order to obtain a more detailed classification of wind power low output events, KPCA dimensionality reduction is performed on the power sequence of wind power low output events in historical wind power data to extract key information, such as Fig.10 As shown, the KPCA dimensionality reduction principal component contribution rate diagram of the power sequence of the continuous 4-hour low-output event is displayed, wherein the cumulative contribution rate of the first five principal components has reached more than 90%. The embodiment of the present disclosure selects the first five principal components for cluster analysis and divides them into different meteorological element types (or weather types).

[0094] Since the types of wind power low output events include stable low output events (sequence mean close to 0), fluctuating low output events, and mixed low output events, there are more types in mixed low output events, mainly including the types of fluctuating first and then stable, and stable first and then fluctuating. Since most stable sequences are sequences with all zeros in the entire period, they are first divided according to the sequence mean. Therefore, before clustering, the number of clusters is set to 3, and the following is obtained: Figures 11 to 14There are four types of low output events shown, Fig.11 Indicates a stable low output event, with an amplitude close to 0, of which all-zero sequences account for more than 80%. Fig.12 It is a fluctuating low output event, and there are significant fluctuations throughout the entire low output event duration. Fig.13 and Fig.14 All of them are mixed low output events, and their typical characteristics are as follows: during the entire low output duration, there are stable low output and large fluctuations in low output. Mixed low output event I first shows large fluctuations and then stable low output, and mixed low output event II first shows stable low output and then large fluctuations.

[0095] Based on the above classification results of wind power low output events, the disclosed embodiment establishes four INFO-HKELM wind power low output power prediction models for different wind power low output scenarios, corresponding to fluctuating low output events, mixed low output events I, mixed low output events II, and high output scenarios (normal wind power output events), and establishes a short-term prediction model with multiple inputs and multiple outputs. For stable low output events, considering that the wind power value itself is close to 0, when performing weighted grey correlation coefficient matching, the wind power matched to the stable low output event is set to zero.

[0096] Step S702: training a wind power low output event classification model.

[0097] In some embodiments, according to the above analysis, the wind power low output event classification model is trained for different types of wind power low output events, and the CNN-BiGRU is used to make a preliminary judgment on the type of wind power low output event using the meteorological elements in the future period, and the weighted grey correlation coefficient is used to further judge the type of wind power low output event. In this regard, the embodiment of the present disclosure makes a preliminary judgment on the type of wind power low output event for 1993 sliding window sampled wind power low output events and 2481 normal wind power output events in the test set. The CNN-BiGRU network structure includes 2 convolutional layers, 1 pooling layer, 3 BiGRU layers, 1 fully connected layer, and the output is connected to the softmaxLayer, and the cross entropy is used as the loss function of the network training.

[0098] To prove the effectiveness of the CNN-BiGRU model used, the disclosed embodiment compares other classification models, including CNN-BiLSTM (Convolutional Neural Networks-Bidirectional Long and Short-Term Memory networks), GRU, ELM, RF (Random Forest), etc. After training with the CNN-BiGRU model, the following results are obtained: Fig.15The pie-shaped heat map of the initial identification accuracy of wind power low output events shown in the figure includes indicators such as accuracy, precision, recall, and F1 score. The identification results based on CNN-BiGRU and CNN-BiLSTM show that the wind power low output events that are correctly identified (or correctly classified) account for the vast majority, and all indicators are significantly higher than those of other models. The identification accuracy of CNN-BiGRU is 99.98%, the accuracy of CNN-BiLSTM is 99.78%, the accuracy of GRU is 86.75%, and the accuracy of ELM and RF is 95.06%. The accuracy, precision-low output, precision-normal output, recall-low output, recall-normal output, F1-low output, and F1-normal output of the CNN-BiGRU-based low output event preliminary identification model are 5.82%, 3.63%, 6.08%, 8.94%, 3.05%, 6.96%, and 4.84% higher than those of other models, respectively.

[0099] Table 1 Statistical results of wind power low output event discrimination for each model

[0100] In order to analyze the judgment results of all wind power low output events in the test set, the embodiment of the present disclosure lists the correct and incorrect judgments of wind power low output events and wind power normal output events, as shown in Table 1. The judgment of normal output events based on the first three models (CNN-BiGRU, CNN-BiLSTM and GRU) is almost completely correct. However, almost all models have a certain degree of misjudgment in the judgment of low output events, especially GRU, ELM and RF. In this regard, the embodiment of the present disclosure analyzes the minimum, maximum, mean and other indicators of wind speed of the positive and misjudged samples of wind power low output events of these three models, as shown in Table 1. Fig.16 As shown in the figure, relatively speaking, the maximum, minimum and mean values ​​of the NWP wind speed of the misjudged samples are closer to those of the normal output events, and the three indicators are larger than those of the correctly judged wind power low output events. The minimum wind speed value of the three models' misjudged low output events is 0.59m / s higher than the minimum value of the correctly judged low output events, the maximum value is 16.13m / s higher, and the mean value is 6.35m / s higher. The difference between the minimum, maximum and mean values ​​of the wind speed of the misjudged wind power low output events and the three indicators of the normal output events are: 0.24m / s, 0.41m / s, and 2.08m / s respectively. This shows that the wind speed of the misjudged wind power low output event is closer to that of the normal wind power output event, and is larger than the wind speed of most wind power low output events, showing the characteristics of normal wind power output events. This is the main reason for the misjudgment based on general model predictions. The CNN-BiGRU model is used to mine the remaining key information except wind speed, which improves its ability to distinguish extreme abnormal low output events.

[0101] Step S703: training wind power low output power prediction models corresponding to different wind power low output events.

[0102] In some embodiments, based on the preliminary identification of wind power low output events by the trained wind power low output event classification model, the grey correlation coefficient between the NWP feature matrix of the 4h wind power sequence in the future period and the sample category of wind power low output events in the sample set is calculated, and then the weighted grey correlation coefficient is calculated to further identify the type of wind power low output events. To determine the weighted grey correlation coefficient of meteorological elements such as wind speed, wind direction, temperature, and air pressure between different samples (including meteorological elements and power points) of the same wind power low output event The embodiment of the present disclosure analyzes the grey correlation between wind speed, wind direction, temperature, air pressure and power of four wind power low output events, such as Fig.17 As shown, it can be concluded that under each type of wind power low output event, the grey correlation between wind speed and power is relatively greater, followed by wind direction, air pressure and temperature. Therefore, when determining the grey correlation weight coefficient, the ratio of the average value of the grey correlation coefficient of each feature (one of wind speed, wind direction, temperature, air pressure and power) and power under each type of wind power low output event to the average value of the total number of features is used as the weighted grey correlation weight coefficient. The calculation formula is as follows: (17) in, Indicates The weight of the feature, Indicates The mean value of the grey correlation coefficient of each feature under each wind power low output event type, It represents the sum of the mean values ​​of the grey correlation coefficients of each feature under each type of wind power low output event.

[0103] Fig.18 The trend of power sequence and wind speed sequence under normal wind power low output event is shown. The power and wind speed under the same wind power low output event have certain common trends. When performing power forecasting, in order to accurately identify the specific type of wind power low output event, the weighted grey correlation coefficient between the feature matrix after one judgment of the event to be predicted and the meteorological matrix corresponding to the clustering center of the four wind power low output events is calculated. The model trained by the scenario with the largest matching coefficient is selected to predict the wind power of the event, and the preliminary wind power forecast result is obtained.

[0104] Firstly, the wind power in the test set is predicted based on the direct prediction method of the wind power low output event classification model. At this time, the scene division and scene discrimination of the wind power low output event are not considered. The following is obtained: Fig.19The wind power forecast results for the week shown include high wind power output scenarios and low wind power output scenarios. The predicted power sequence for future time periods has a similar trend to the actual sequence in terms of trend, which requires the numerical weather forecast to have a higher forecast accuracy. However, in certain time periods, especially in low wind power output scenarios, there are significant differences between the wind power forecast sequence and the actual value. Under the premise that it is difficult to predict the occurrence of low wind power output scenarios, it is not conducive to the dispatching side to formulate a reasonable power supply guarantee plan. In order to analyze the prediction accuracy of the direct prediction method, the embodiment of the present disclosure uses ELM, BP (Back Propagation neural network), HKELM, INFO-HKELM and other models to compare the prediction results. The prediction error evaluation index is shown in Table 2. The prediction errors of the four wind power models are relatively close, and the RMSE is between 0.09 and 0.11. However, relatively speaking, the prediction errors of BP, ELM and HKELM are larger, especially BP. The error of HKELM optimized based on INFO is relatively lower, and the goodness of fit is relatively greater. On average, it is 0.63% lower than BP, ELM and HKELM. The average is 0.58% lower and the average goodness of fit is 1.87% higher. The HKELM based on INFO optimization can achieve better wind power prediction results.

[0105] Table 2 Evaluation indicators of wind power prediction error directly predicted by each model

[0106] In this regard, the embodiment of the present disclosure adopts INFO-HKELM to establish a wind power low output power prediction model. First, based on the matching of the primary and secondary wind power low output events, four INFO-HKELM models are established for the low output fluctuation scenario, low output mixed scenario I, low output mixed scenario II and normal output scenario for scenario-by-scenario prediction, such as Fig. 20 As shown, the power prediction results of several randomly selected events under non-continuous wind power low output events are obtained. The method based on scene division can initially achieve a certain wind power prediction accuracy, especially for mixed wind power low output events. Since the fluctuating wind power low output event continues to fluctuate in the entire wind power low output event, the wind power prediction result can keep up with the trend of the power sequence, but it is difficult to completely fit the actual power curve. For stable wind power low output events, the embodiment of the present disclosure regards it as a power process below the cut-in wind speed, and the power prediction result of the entire wind power low output event is set to zero.

[0107] Step S704: correcting the power in the future time period output by the wind power low output power prediction model.

[0108] In some embodiments, under the same weather process conditions, the errors of the wind power prediction power (i.e., the power in the future time period) output by the wind power low output power prediction model have certain similarities. This similarity is mainly reflected in the fact that the same power fluctuation corresponds to similar meteorological fluctuation processes. Therefore, when performing error compensation for multiple wind power low output events, the similarity of meteorological elements of historical wind power low output events is mainly based on the similarity of meteorological elements of historical wind power low output events. The similarity of meteorological elements between the power event to be predicted and the meteorological elements of the same historical wind power low output events is calculated. The weighted grey correlation coefficient is used to select the error of the matching event with a larger weighted grey correlation coefficient in the training set to correct the power prediction error of the wind power data to be predicted. Since the matching mechanism is mainly based on meteorological similarity, there are certain deviations in the NWP forecast, which makes the error obtained from a single matching event not necessarily suitable for the sequence to be corrected. Therefore, the embodiment of the present disclosure selects the error average of multiple similar events for error compensation of wind power prediction power. Figure 21 to Figure 24 As shown, we get Fig. 20 The error compensation results of wind power forecast power obtained by preliminary matching of wind power low output events show that the wind power forecast power errors are effectively corrected under almost all wind power low output events, especially the three wind power low output events. The errors are significantly reduced after correction, and the correction curves are better fitted to the actual curves. However, for normal wind power output events, the error correction is not obvious, and there is only a certain correction effect. This is mainly because under high-power wind power output events, the power fluctuations are more severe, and it is difficult to match wind power output events with weather processes with strong similarity, so the error compensation effect is not obvious.

[0109] Fig.25 The wind power error evaluation index before and after the error compensation of multiple wind power low output events is shown. After correction, the errors under each wind power low output event are effectively corrected, especially the hybrid wind power low output event I. After correction, the NRMSE of the fluctuating wind power low output event, the hybrid wind power low output event I, and the hybrid wind power low output event II decreased by 0.004, 0.006, and 0.005, respectively, and the NMAE decreased by 0.005, 0.006, and 0.005, respectively. Fig.26The results of wind power prediction power error evaluation index in different wind power low output events are shown based on wind power low output event classification and error compensation of multiple wind power low output events in similar segments. The NRMSE and NMAE of wind power low output events are relatively small, which is mainly because the power and corresponding errors are lower in wind power low output events. The NRMSE and NMAE of wind power normal output events and the overall prediction results are relatively large. Under normal wind power output events, the NRMSE is 0.092 and the NMAE is 0.072. After error compensation for multiple wind power low output events, the NRMSE and NMAE decrease by 0.014 and 0.015 respectively. After the overall prediction result is corrected, the NRMSE and NMAE decrease by 0.013 and 0.014 respectively compared with before correction. Compared with the direct prediction method, the NRMSE and NMAE decrease by 0.019 and 0.018 respectively. The method based on wind power low output event division and error compensation of multiple wind power low output events can significantly reduce the power prediction error under wind power low output events and improve the overall accuracy of short-term wind power prediction.

[0110] Although the method of adopting the classification prediction of wind power low output events and the error compensation mechanism of multiple wind power low output events can significantly improve the accuracy of wind power prediction power, there are still miscorrected wind power low output events. For this, the embodiment of the present disclosure counts the total number of events under each type of wind power low output event and the number of miscorrected events, and defines miscorrected wind power low output events and miscorrection rates based on this. The basic definitions are as follows: 1) False correction event: In a single wind power low output event, if the mean absolute error before correction is less than the mean absolute error after correction, it is regarded as a false correction event; 2) False correction rate: The percentage of false correction events to the total number of events under a certain wind power low output event. The calculation formula is as follows: (18) in, Indicates the error correction rate of a wind power low output event, Indicates the number of events that were incorrectly corrected in a certain wind power low output event, Represents the total number of events.

[0111] Based on this, the embodiment of the present disclosure calculates the error correction rate of four types of wind power low output events in the test set, and the results are shown in Table 3, where each wind power low output event is a sliding window sampling event, and its total number is greater than the actual number. It can be concluded that almost every wind power low output event has a certain amount of error correction. This is mainly because the correction mechanism for error compensation of multiple wind power low output events is completely based on the matching of similar meteorological segments. However, this matching mechanism requires NWP to have a relatively accurate forecast accuracy. Under normal circumstances, NWP has many false alarms and missed alarms. At the same time, its key meteorological elements, especially wind speed, have significant timing advances or lags in some cases, which may cause the similar segments obtained by matching the wind power low output event to be different from the actual situation, which deteriorates the correction result. However, in general, the error compensation mechanism for multiple wind power low output events based on similar segment matching is effective in most wind power low output events.

[0112] Table 3 Error correction rates of four types of wind power low output events

[0113] It should be noted that the embodiments of the present disclosure are not exhaustive and do not constitute a limitation on the scope of protection of the claims. Based on the inspiration gained from the embodiments of the present disclosure, those skilled in the art can think of other substantially equivalent alternatives without creative work, all of which are within the scope of protection of the present invention.

[0114] In some embodiments, the method provided by the embodiments of the present disclosure may be verified through simulation.

[0115] Simulation input: Analyze the NWP forecast data of the wind farm to determine the total installed capacity of the wind farm; Input data: Historical wind power data; Data sampling interval is 15 minutes; According to steps S701 to S704, obtain the short-term preliminary forecast results of wind power in the forecast period (i.e., the power in the future time period, the wind power forecast power); In order to measure the effectiveness of wind power forecast and error correction results, the mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate the predicted and corrected wind power forecast. The calculation formula is: (19) (20) in, Indicates the number of point predictions in the prediction sequence; Indicates the rated installed capacity.

[0116] The method for distinguishing wind power low output events and training a power prediction model provided by the embodiment of the present disclosure identifies the current wind power low output scenario based on CNN-BiGRU, and establishes multiple INFO-HKELM models according to different scenarios to perform preliminary wind power prediction; the error correction results of different wind power low output prediction scenarios are obtained according to the multi-scenario error compensation mechanism, which increases the prediction accuracy of wind power prediction power and improves the positive significance of wind power prediction results for power dispatching. The proposed method has clear physical meaning, effective prediction results and strong practicability.

[0117] Fig. 27 An optional structural schematic diagram of a wind power low output event identification and power prediction model training device provided by an embodiment of the present disclosure is shown, and will be explained according to each part.

[0118] In some embodiments, the wind power low output event identification and power prediction model training device 900 includes a sample determination unit 901 , a first training unit 902 , and a second training unit 903 .

[0119] The sample determination unit 901 is used to perform principal component analysis on historical wind power data, cluster the results of the principal component analysis, determine data corresponding to wind power low output events, and form a sample set; the sample set includes at least one meteorological element and a wind power point corresponding to the at least one meteorological element; the at least one meteorological element in the sample set and the wind power point corresponding to the at least one meteorological element are marked with the type of wind power low output event; The first training unit 902 is used to determine a wind power pixel map based on at least one meteorological element in a sample set and a wind power point corresponding to the at least one meteorological element; input the wind power pixel map into a feature extraction network included in a wind power low output event classification model to obtain wind power data features after dimensionality reduction; input the wind power data features after dimensionality reduction into a classification network included in the wind power low output event classification model to determine a predicted wind power low output event type corresponding to the wind power pixel map; adjust the weight of the wind power low output event classification model based on the annotation of the wind power low output event type corresponding to the wind power pixel map and the predicted wind power low output event type; The second training unit 903 is used to input at least one meteorological element corresponding to the first wind power low output event in the sample set and the wind power point corresponding to the at least one meteorological element into the first wind power low output power prediction model, and obtain the power prediction result of the first wind power low output power prediction model for the at least one meteorological element and the at least one meteorological element; based on the power prediction result and the actual wind power point, adjust the parameters of the first wind power low output power prediction model.

[0120] The sample determination unit 901 is specifically used to calculate the kernel functions corresponding to the power segments corresponding to the multiple wind power events in the historical wind power data; and to form a first kernel matrix based on the kernel functions corresponding to the power segments corresponding to the multiple wind power events; Calculate the eigenvalues ​​of each column element in the first kernel matrix; Determine a corresponding eigenvector based on the eigenvalues ​​of each column element; The feature vectors are clustered, and based on the clustering results, the data corresponding to the wind power low output event is determined to form a sample set.

[0121] The sample determination unit 901 is specifically used to determine the first Power point at moment To Power point at moment The corresponding kernel function . Determine the first kernel matrix based on the kernel function Among them, the power point and power point Power segments belonging to the same wind power low output event. , .

[0122] The sample determination unit 901 is specifically used to calculate the eigenvalue of each column in the first kernel matrix and sort them from large to small; determine the eigenvector corresponding to the eigenvalue of each column ,in, Characterization The eigenvalues ​​corresponding to the eigenvectors ( ), after normalizing all feature vectors, we get the sample set in the first Coordinates under principal components:

[0123] in, Characterizes the number of samples in the sample set, Characterization The first feature vector values, among which, The dimension of is consistent with the total sample size in the sample set. For the principal components; sort them in descending order according to the eigenvalues, determine at least one principal component in the front as the input of the clustering model, determine the output of the clustering model as the data corresponding to the wind power low output event, and constitute a sample set.

[0124] The first training unit 902 is specifically used to input the wind power pixel map into the feature extraction network, extract the wind power data features of the wind power pixel map, reduce the dimension of the wind power data features, and obtain the reduced wind power data features; the reduced wind power data features are expressed in the form of a sequence or a vector; the reduced wind power data features are input into the classification network to obtain the output of the classification network; and determine the type with the highest probability in the classification network, which is the predicted wind power low output event type corresponding to the wind power pixel map.

[0125] Fig.28 An optional structural schematic diagram of a wind power low output event identification and power prediction device provided by an embodiment of the present disclosure is shown, and will be explained according to each part.

[0126] In some embodiments, the wind power low output event identification and power prediction device 1000 includes a data processing unit 1001 , a first input unit 1002 , and a second input unit 1003 .

[0127] The data processing unit 1001 is used to determine a predicted wind power pixel map based on the wind power data to be predicted; The first input unit 1002 is used to input the predicted wind power pixel map into the wind power low output event classification model to determine the wind power low output event classification result of the wind power data to be predicted; based on the wind power low output event classification result of the wind power data to be predicted, determine the corresponding wind power low output power prediction model; The second input unit 1003 is used to input the wind power data to be predicted into the wind power low output power prediction model to obtain the power in the future time period corresponding to the wind power data to be predicted.

[0128] In some embodiments, the wind power low output event identification and power prediction device 1000 may further include a compensation unit 1004 .

[0129] The compensation unit 1004 is used to obtain at least two pieces of historical wind power data of the same type corresponding to the classification result of the wind power low output event of the wind power data to be predicted; Matching the wind power data to be predicted with the at least two pieces of historical wind power data of the same type, and obtaining at least one piece of historical wind power data of the same type with a high matching value; Based on the at least one piece of historical wind power data of the same type, the power of the future time period corresponding to the wind power data to be predicted is compensated, and a compensation result is determined as the target power of the future time period corresponding to the wind power data to be predicted.

[0130] The compensation unit 1004 is specifically used to calculate the grey correlation coefficient of each feature in the wind power data to be predicted; weight the grey correlation coefficients of all features based on the weight coefficient to obtain a first weighted grey correlation coefficient; Respectively calculating the grey correlation coefficient of each feature in the at least two historical wind power data of the same type, and weighting them based on the weight coefficient to obtain at least one second weighted grey correlation coefficient; The first weighted grey correlation coefficient and the at least one second weighted grey correlation coefficient are matched to determine at least one piece of historical wind power data of the same type having a high matching value with the wind power data to be predicted.

[0131] The compensation unit 1004 is specifically configured to determine an error average value based on the at least one piece of historical wind power data of the same type; The power of a future time period corresponding to the predicted wind power data is compensated based on the error average value.

[0132] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device and a readable storage medium.

[0133] Fig.29 A schematic block diagram of an example electronic device 800 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0134] like Fig.29 As shown, the electronic device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 to a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the electronic device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0135] Multiple components in the electronic device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the electronic device 800 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0136] The computing unit 801 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 801 performs the various methods and processes described above, such as a training method for wind power low output event discrimination and power prediction model and / or a wind power low output event discrimination and power prediction method. For example, in some embodiments, a wind power low output event discrimination and power prediction model training method and / or a wind power low output event discrimination and power prediction method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the wind power low output event discrimination and power prediction model training method and / or wind power low output event discrimination and power prediction method described above may be executed. Alternatively, in other embodiments, the computing unit 801 may be configured to execute the wind power low output event discrimination and power prediction model training method and / or wind power low output event discrimination and power prediction method in any other appropriate manner (e.g., by means of firmware).

[0137] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0138] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0139] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0140] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0141] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0142] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship of client and server is generated by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.

[0143] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0144] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0145] The above is only a specific embodiment of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present disclosure, which should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A method for identifying low-output wind power events and training a power prediction model, characterized in that: The method comprises: Conduct principal component analysis on historical wind power data, cluster the results of principal component analysis, determine the data corresponding to low wind power output events, and form a sample set; Marking the type of wind power low output event for at least one meteorological element in the sample set and at least one wind power point corresponding to the meteorological element; Based on at least one meteorological element in the sample set and the wind power point corresponding to the at least one meteorological element, determine a wind power pixel map; input the wind power pixel map into a feature extraction network included in a wind power low output event classification model to obtain wind power data features after dimensionality reduction; input the wind power data features after dimensionality reduction into a classification network included in the wind power low output event classification model to determine the predicted wind power low output event type corresponding to the wind power pixel map; based on the annotation of the wind power low output event type corresponding to the wind power pixel map and the predicted wind power low output event type, adjust the weight of the wind power low output event classification model; At least one meteorological element corresponding to the first wind power low output event in the sample set and the wind power point corresponding to the at least one meteorological element are input into the first wind power low output power prediction model to obtain the power prediction results of the first wind power low output power prediction model for the at least one meteorological element and the at least one meteorological element; based on the power prediction results and the actual wind power point, the parameters of the first wind power low output power prediction model are adjusted.

2. The method according to claim 1, characterized in that The principal component analysis is performed on the historical wind power data, and the principal component analysis results are clustered to determine the data corresponding to the wind power low output event to form a sample set, including: Calculate the kernel function corresponding to the power segments corresponding to multiple wind power low output events in the historical wind power data; Constructing a first kernel matrix based on kernel functions corresponding to power segments corresponding to the multiple wind power low output events; Calculate the eigenvalues ​​of each column element in the first kernel matrix; Determine a corresponding eigenvector based on the eigenvalues ​​of each column element; The feature vectors are clustered, and based on the clustering results, the data corresponding to the wind power low output event is determined to form a sample set.

3. The method according to claim 2, characterized in that The step of calculating kernel functions corresponding to power segments corresponding to a plurality of wind power low output events in historical wind power data; and forming a first kernel matrix based on the kernel functions corresponding to power segments corresponding to the plurality of wind power low output events includes: Determine the historical wind power data Power point at moment To Power point at moment The corresponding kernel function ; The kernel function Construct the first core matrix .

4. The method according to claim 2, characterized in that: The principal component analysis is performed on the historical wind power data, and the principal component analysis results are clustered to determine the data corresponding to the wind power low output event to form a sample set, including: Calculate the first kernel matrix The eigenvalues ​​of each column element in The eigenvalues ​​corresponding to the eigenvectors ; Determine the historical wind power data in Coordinates of the principal components : in, Characterizes the number of samples included in the historical wind power data, Characterization The first feature vector values, among which The dimension of is the same as the number of samples; The eigenvalues ​​are sorted in descending order, at least one principal component in the front is determined as the input of the clustering model, and the output of the clustering model is determined as the data corresponding to the wind power low output event to form a sample set.

5. The method according to claim 1, characterized in that The step of inputting the wind power pixel map into a feature extraction network included in a wind power low output event classification model to obtain the wind power data features after dimensionality reduction; and inputting the wind power data features after dimensionality reduction into a classification network included in the wind power low output event classification model to determine the predicted wind power low output event type corresponding to the wind power pixel map, including: Inputting the wind power pixel map into a feature extraction network, extracting wind power data features of the wind power pixel map, and performing dimensionality reduction on the wind power data features to obtain the wind power data features after dimensionality reduction; the wind power data features after dimensionality reduction are expressed in the form of a sequence or a vector; Inputting the dimension-reduced wind power data features into the classification network to obtain the output of the classification network; Determine the type with the highest probability in the output of the classification network as the predicted wind power low output event type corresponding to the wind power pixel map.

6. A method for identifying low-output wind power events and predicting power, characterized in that: The wind power low output event classification model and the first wind power low output power prediction model trained based on the methods of claims 1 to 5 are implemented, and the method comprises: Based on the wind power data to be predicted, determining a predicted wind power pixel map; Inputting the predicted wind power pixel map into the wind power low output event classification model to determine the wind power low output event classification result of the wind power data to be predicted; Determining a corresponding first wind power low output power prediction model based on the classification result of the wind power low output event of the wind power data to be predicted; The wind power data to be predicted is input into the first wind power low-output power prediction model to obtain the power in a future time period corresponding to the wind power data to be predicted.

7. The method according to claim 6, characterized in that The method further comprises: Acquire at least two pieces of historical wind power data of the same type corresponding to the classification result of the wind power low output event of the wind power data to be predicted; Matching the wind power data to be predicted with the at least two pieces of historical wind power data of the same type, and obtaining at least one piece of historical wind power data of the same type with a high matching value; Based on the at least one piece of historical wind power data of the same type, the power of the future time period corresponding to the wind power data to be predicted is compensated, and a compensation result is determined as the target power of the future time period corresponding to the wind power data to be predicted.

8. The method according to claim 7, characterized in that The step of matching the wind power data to be predicted with the at least two pieces of historical wind power data of the same type to obtain at least one piece of historical wind power data of the same type with a high matching value includes: Calculating the grey correlation coefficient of each feature in the wind power data to be predicted; weighting the grey correlation coefficients of all features based on the weight coefficient to obtain a first weighted grey correlation coefficient; Respectively calculating the grey correlation coefficient of each feature in the at least two historical wind power data of the same type, and weighting them based on the weight coefficient to obtain at least one second weighted grey correlation coefficient; The first weighted grey correlation coefficient and the at least one second weighted grey correlation coefficient are matched to determine at least one piece of historical wind power data of the same type having a high matching value with the wind power data to be predicted.

9. The method according to claim 7, characterized in that: The compensating the power of the future time period corresponding to the wind power data to be predicted based on the at least one piece of historical wind power data of the same type includes: Determine an error average value based on the at least one piece of historical wind power data of the same type; The power of the future time period corresponding to the wind power data to be predicted is compensated based on the error average value.

10. A wind power low output event identification and power prediction model training device, characterized in that: The device comprises: A sample determination unit is used to perform principal component analysis on historical wind power data, cluster the results of the principal component analysis, determine data corresponding to wind power low output events, and form a sample set; the sample set includes at least one meteorological element and a wind power point corresponding to the at least one meteorological element; the at least one meteorological element in the sample set and the wind power point corresponding to the at least one meteorological element are marked with the type of wind power low output event; The first training unit is used to determine a wind power pixel map based on at least one meteorological element in a sample set and a wind power point corresponding to the at least one meteorological element; input the wind power pixel map into a feature extraction network included in a wind power low output event classification model to obtain wind power data features after dimensionality reduction; input the wind power data features after dimensionality reduction into a classification network included in the wind power low output event classification model to determine a predicted wind power low output event type corresponding to the wind power pixel map; adjust the weight of the wind power low output event classification model based on the annotation of the wind power low output event type corresponding to the wind power pixel map and the predicted wind power low output event type; The second training unit is used to input at least one meteorological element corresponding to the first wind power low output event in the sample set and the wind power point corresponding to the at least one meteorological element into the first wind power low output power prediction model, and obtain the power prediction result of the first wind power low output power prediction model for the at least one meteorological element and the at least one meteorological element; based on the power prediction result and the actual wind power point, adjust the parameters of the first wind power low output power prediction model.

Images