Wind power data cleaning method, device and equipment and storage medium

Abstract

The embodiment of the invention provides a wind power data cleaning method and device, equipment and a storage medium, and the method comprises the steps: carrying out the preprocessing of an obtained original wind power data set, removing the missing wind speed data and power data in the original wind power data set, and obtaining a first wind power data set; global abnormal values of wind speed and power data in the first wind power data set are detected and eliminated, and a second wind power data set is obtained; detecting and eliminating local abnormal values of wind speed and power data in the second wind power data set to obtain a third wind power data set; determining a fourth wind power data set based on the second wind power data set and the third wind power data set; detecting and rejecting the fourth wind power data set to obtain a fifth wind power data set; and training a plurality of back propagation models based on the fifth wind power data set, and repairing the power abnormal data and the wind speed abnormal data based on the plurality of back propagation models to obtain a cleaned wind power data set.

Description

Technical Field

[0001] The present invention relates to the technical field of power system data preprocessing, and relates to but is not limited to a wind power data cleaning method, device, equipment and storage medium. Background Art

[0002] In the operation and maintenance of power systems, the accuracy and reliability of data are crucial. With the continuous expansion of the scale of power systems and the continuous advancement of technology, the amount of data collected by the monitoring system of centralized control stations has increased dramatically. The quality of data directly affects key links such as the evaluation of system operation status, fault diagnosis and preventive maintenance. However, in actual operation, due to various factors such as sensor failure, communication errors or environmental interference, power data often has problems such as missing, noise, and outliers. These problems not only affect the accuracy of data analysis, but may also lead to wrong decisions and even threaten the safe and stable operation of the power grid.

[0003] In order to solve the above problems and improve data quality, data cleaning technology has emerged. Data cleaning refers to the use of a series of methods and techniques to identify and correct or delete errors and inconsistencies in data to ensure the accuracy, completeness and consistency of the data. Traditional data cleaning methods mostly rely on manually set rules or simple statistical analysis, which are incapable of processing large-scale, high-dimensional, and complex data. In recent years, with the development of machine learning technology, more and more studies have begun to explore the use of machine learning algorithms for data cleaning, especially for the detection and repair of outliers. Summary of the invention

[0004] In order to solve the above technical problems existing in the prior art, the present invention provides a wind power data cleaning method, device, equipment and storage medium.

[0005] The technical method of the embodiment of the present invention is implemented as follows:

[0006] In a first aspect, an embodiment of the present invention provides a wind power data cleaning method, the method comprising:

[0007] After preprocessing the obtained original wind power data set, the missing wind speed data and power data in the original wind power data set are removed to obtain a first wind power data set;

[0008] Using an isolation forest algorithm to detect and remove global outliers in the wind speed data and the power data in the first wind power data set, to obtain a second wind power data set;

[0009] Using an outlier detection algorithm based on boundary-aware local density, local outliers of wind speed data and power data in the second wind power data set are detected and removed to obtain a third wind power data set;

[0010] Based on the second wind power data set and the third wind power data set, a fourth wind power data set is determined; by applying a density-based noise spatial clustering algorithm, the fourth wind power data set is detected and eliminated to obtain a fifth wind power data set;

[0011] A plurality of back propagation models are trained based on the fifth wind power data set, and the abnormal power data and the abnormal wind speed data are repaired based on the plurality of back propagation models to obtain a cleaned wind power data set.

[0012] In a second aspect, an embodiment of the present invention provides a wind power data cleaning device, the device comprising:

[0013] A first elimination module is used to eliminate missing wind speed data and power data from the original wind power data set after preprocessing the obtained original wind power data set, so as to obtain a first wind power data set;

[0014] A second elimination module is used to detect and eliminate global outliers of wind speed data and power data in the first wind power data set by using an isolation forest algorithm to obtain a second wind power data set;

[0015] a third elimination module, configured to detect and eliminate local outliers of the wind speed data and the power data in the second wind power data set by using an outlier detection algorithm based on boundary-aware local density, so as to obtain a third wind power data set;

[0016] a fourth elimination module, configured to determine a fourth wind power data set based on the second wind power data set and the third wind power data set; and detect and eliminate the fourth wind power data set by applying a density-based noise spatial clustering algorithm to obtain a fifth wind power data set;

[0017] The data repair module is used to train multiple back propagation models based on the fifth wind power data set, and repair the abnormal power data and abnormal wind speed data based on the multiple back propagation models to obtain a cleaned wind power data set.

[0018] In some embodiments, the second elimination module is further used to randomly select β sample points from the first wind power data set as subsamples and put them into the root node of an isolated tree; randomly specify a dimension q, and randomly generate a cutting point p within the current node data range, where p is generated between the maximum value and the minimum value of the specified dimension; the cutting point p generates a hyperplane, and the sample points with the specified dimension less than p are placed in the left child node, and those greater than or equal to p are placed in the right child node; determine whether all leaf nodes have only one sample point reaching the specified height, if so, the construction of a single isolated tree is completed, if not, return to the previous step; repeat the above steps t times, so as to obtain t isolated trees and generate an isolation forest model; then, substitute each sample point x into the isolation forest model, and calculate the abnormality score, that is:

[0019] ;

[0020] In the formula, h(x) is the height of sample x in the isolated tree, and c(β) is the average path length; if the c(β) anomaly score is close to 1, it is an abnormal sample; if the c(β) anomaly score is much less than 0.5, it is a normal sample; if the c(β) anomaly score is about 0.5, it means that there is no abnormal sample.

[0021] In some embodiments, the third elimination module is further used to determine the first wind power data set as the data to be detected; calculate the current data point x in the first wind power data set i The sum of the squares of the distances of the k nearest neighbors is used to obtain the local density; the distance between the current data point and all other data points in the first wind power data set is calculated to obtain a distance matrix; based on the local density, the sum of the number of elements in the distance matrix that meet the preset conditions is calculated; the preset condition indicates that the element value in the distance matrix is ​​less than the local density value, so as to determine whether it is an abnormal point; based on the neighborhood density and deviation parameter of each data point in the first wind power data set, the neighborhood boundary of each data point is determined; each data point is compared with the corresponding neighborhood boundary, and the data points located outside the neighborhood boundary and with lower neighborhood density are determined as outliers; the abnormal factors are screened based on the preset threshold to obtain the third wind power data set.

[0022] In some embodiments, the fourth elimination module is also used to integrate the common data in the second wind power data set and the third wind power data set to obtain the fourth wind power data set; and set the radius and the minimum number of sample points; traverse the fourth wind power data set to determine whether each data point in the fourth wind power data set is clustered; if not, the core points within the radius neighborhood of the core object are grouped and marked; traverse all core points of the current class and mark them; output all clusters, core points, boundary points and outliers in the fourth wind power data set, and eliminate abnormal data to obtain the fifth wind power data set.

[0023] In some embodiments, the data repair module is also used to train the multiple back propagation models using the fifth wind power data set and the back propagation neural network; wherein the first back propagation model is used to represent the relationship between wind speed and power at time t; the second back propagation model is used to represent the relationship between power and wind speed at time t; the third back propagation model is used to represent the relationship between power at time tk, ..., t-1 and power at time t; the fourth back propagation model is used to represent the relationship between wind speed at time tk, ..., t-1 and wind speed at time t; using the first back propagation model, the abnormal power data at time t is repaired based on the wind speed data at time t; using the second back propagation model, the abnormal wind speed data at time t is repaired based on the power data at time t; if the wind speed data and power data at time t are both eliminated, the third back propagation model is used to repair the power data at time t according to the power data at time tk, ..., t-1, and the fourth back propagation model is used to repair the wind speed data at time t according to the wind speed data at time tk, ..., t-1, to obtain the cleaned wind power data set.

[0024] In a third aspect, an embodiment of the present invention provides an electronic device, comprising: a memory for storing executable instructions; and a processor for implementing the above-mentioned wind power data cleaning method when executing the executable instructions stored in the memory.

[0025] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing executable instructions for causing a processor to execute the executable instructions to implement the above-mentioned wind power data cleaning method.

[0026] The wind power data cleaning method, device, equipment and storage medium provided by the embodiment of the present invention first pre-process the original wind power data set, and then remove the missing wind speed data and power data in the original wind power data set to obtain a first wind power data set; use the isolation forest algorithm to detect and remove the global outliers of the wind speed data and power data in the first wind power data set to obtain a second wind power data set; use the outlier detection algorithm based on boundary-aware local density to detect and remove the local outliers of the wind speed data and power data in the second wind power data set to obtain a third wind power data set; based on the second wind power data set and the third wind power data set, determine a fourth wind power data set; use the density-based noise spatial clustering algorithm to detect and remove the fourth wind power data set to obtain a fifth wind power data set; train multiple back propagation models based on the fifth wind power data set, and repair the abnormal power data and abnormal wind speed data based on the multiple back propagation models to obtain a cleaned wind power data set. Thus, the present invention provides a multi-mode machine learning data preprocessing method consisting of multiple outlier detection algorithms and BPNN that can realize abnormal data repair. By systematically combining them, their respective advantages are fully utilized to perform more accurate outlier detection and data repair; thereby meeting the requirements for real-time, accuracy, integrity, consistency, scalability and automatic alarm of wind power data cleaning algorithms in the intelligent upgrade of the functions of the centralized control station monitoring system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a structural schematic diagram of a wind power data cleaning system provided by an embodiment of the present invention;

[0028] Figure 2 It is a flow chart of a wind power data cleaning method provided by an embodiment of the present invention;

[0029] Figure 3 It is a flow chart of a multi-mode machine learning data preprocessing method for wind power data provided by an embodiment of the present invention;

[0030] Figure 4 is an abnormal data distribution diagram of data sets #1-#5 provided in an embodiment of the present invention;

[0031] Figure 5 (a) is the distribution diagram of data set #5 containing abnormal data; Figure 5 (b) is the distribution diagram of the data set after iForest removes global outliers and BLDOD algorithm identifies local outliers; Figure 5 (c) is the distribution diagram of the data set after removing distribution abnormal data; Figure 5 (d) is the distribution diagram of the data set after cleaning;

[0032] FIG6 (a) is a result diagram of repairing the wind speed value by using the wind speed sequence change; FIG6 (b) is a result diagram of repairing the power value by using the wind power sequence change;

[0033] Figure 7 It is a schematic diagram of the structure of a wind power data cleaning device provided by an embodiment of the present invention;

[0034] Figure 8 It is a schematic diagram of the composition structure of the wind power data cleaning device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0036] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is 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. Unless otherwise defined, all technical and scientific terms used in the embodiments of the present invention have the same meaning as those commonly understood by those skilled in the art to which the embodiments of the present invention pertain. The terms used in the embodiments of the present invention are only for the purpose of describing the embodiments of the present invention and are not intended to limit the present invention.

[0037] The following describes an exemplary application of the wind power data cleaning device of the embodiment of the present invention. The wind power data cleaning device provided by the embodiment of the present invention can be implemented as a terminal or a server. In one implementation, the wind power data cleaning device provided by the embodiment of the present invention can be implemented as various types of terminals such as laptops, tablet computers, desktop computers, mobile devices, etc.; in another implementation, the wind power data cleaning device provided by the embodiment of the present invention can also be implemented as a server, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms. The terminal and the server can be directly or indirectly connected by wired or wireless communication, which is not limited in the embodiment of the present invention. The following describes an exemplary application of the wind power data cleaning device when it is implemented as a server.

[0038] See also Figure 1 , Figure 11 is a schematic diagram of the structure of a wind power data cleaning system 10 provided in an embodiment of the present invention. In order to clean the abnormal values ​​appearing in the original wind power data set, an embodiment of the present invention may provide a wind power data cleaning platform, which may be implemented as a wind power data cleaning application. The wind power data cleaning system 10 provided in an embodiment of the present invention includes a terminal 110, a network 120 and a server 130, wherein the server 130 is a server of the wind power data cleaning application. The server 130 may constitute a wind power data cleaning device in an embodiment of the present invention. The terminal 110 is connected to the server 130 via the network 120, and the network 120 may be a wide area network or a local area network, or a combination of the two.

[0039] In some embodiments, please refer to Figure 1 When cleaning wind power data, the terminal 110 sends the original wind power data set to the server 130 through the network 120. The server 130 receives the original wind power data set sent by the terminal 110, pre-processes the obtained original wind power data set, and removes the missing wind speed data and power data in the original wind power data set to obtain a first wind power data set; uses an isolation forest algorithm to detect and remove global outliers of the wind speed data and power data in the first wind power data set to obtain a second wind power data set; uses an outlier detection algorithm based on boundary-aware local density to detect and remove local outliers of the wind speed data and power data in the first wind power data set to obtain a third wind power data set; based on the second wind power data set and the third wind power data set, a fourth wind power data set is determined; by applying a density-based noise spatial clustering algorithm, the fourth wind power data set is detected and removed to obtain a fifth wind power data set; based on the fifth wind power data set, multiple back propagation models are trained, and power abnormal data and wind speed abnormal data are repaired based on the multiple back propagation models to obtain a cleaned wind power data set. After obtaining the cleaned wind power data set, the server 130 sends the cleaned wind power data set to the terminal 110 through the network 120 .

[0040] The wind power data cleaning method provided in the embodiment of the present invention can also be implemented based on a cloud platform and through cloud technology. For example, the server 130 can be a cloud server. Multiple back propagation models are trained through the cloud server to repair abnormal power data and abnormal wind speed data.

[0041] It should be noted here that cloud technology refers to a hosting technology that unifies hardware, software, network and other resources in a wide area network or local area network to achieve data computing, storage, processing and sharing. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model. It can form a resource pool, which is used on demand and flexible and convenient. Cloud computing technology will become an important support. The backend services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, in the future, each item may have its own identification mark, and all need to be transmitted to the backend system for logical processing. Data of different levels will be processed separately. All kinds of industry data require strong system backing support, which can only be achieved through cloud computing.

[0042] The embodiment of the present invention provides a wind power data cleaning method, see Figure 2 , Figure 2 is a flow chart of a wind power data cleaning method provided by an embodiment of the present invention, which is combined with Figure 2 The steps shown are explained.

[0043] Step S210 , after preprocessing the obtained original wind power data set, the missing wind speed data and power data in the original wind power data set are removed to obtain a first wind power data set.

[0044] In some embodiments, the first wind power data set refers to a wind power data set obtained by removing missing wind speed data and power data from the original wind power data set.

[0045] In some embodiments, the original wind power data set obtained may contain missing or blank wind speed and power data. These missing data may be caused by sensor failure, data transmission errors, or incomplete records. In the first step of preprocessing, each row in the data set can be checked to find the missing wind speed and power values, and the rows containing missing values ​​can be removed from the data set. The purpose of this step is to ensure the integrity of the data set and avoid errors in subsequent analysis or model training.

[0046] Step S220: using an isolation forest algorithm to detect and remove global outliers in the wind speed data and power data in the first wind power data set, to obtain a second wind power data set.

[0047] In some embodiments, the second wind power data set refers to a wind power data set obtained by removing global outliers of wind speed data and power data in the first wind power data set through an isolation forest algorithm.

[0048] In some embodiments, the isolation forest algorithm refers to a tree-based anomaly detection algorithm that builds a tree by randomly selecting features and split values ​​until all data points are isolated. The isolation forest algorithm determines whether a data point is an outlier based on its depth in the tree (i.e., how easy it is to be isolated). In the second wind power data set, the isolation forest algorithm is applied to the wind speed and power data to detect and remove outliers on a global scale.

[0049] Step S230 , using an outlier detection algorithm based on boundary-aware local density, detect and remove local outliers in the wind speed data and power data in the second wind power data set to obtain a third wind power data set.

[0050] In some embodiments, the third wind power data set refers to a wind power data set obtained by removing local outliers of wind speed data and power data in the second wind power data set through an outlier detection algorithm based on boundary-aware local density.

[0051] In some embodiments, the boundary-aware local density-based outlier detection algorithm is an algorithm that considers the local density of data points and the distance from other points. It identifies outliers in a local range by analyzing the local neighborhood of the data points. In the second wind power data set, this algorithm will be applied to the wind speed and power data to detect and remove local outliers.

[0052] Step S240: determining a fourth wind power data set based on the second wind power data set and the third wind power data set; and detecting and eliminating the fourth wind power data set by applying a density-based noise spatial clustering algorithm to obtain a fifth wind power data set.

[0053] In some embodiments, the fourth wind power data set refers to a wind power data set obtained by integrating common data in the second wind power data set and the third wind power data set.

[0054] In some embodiments, the fifth wind power data set refers to a wind power data set obtained by removing abnormal data in the fourth wind power data set by applying a density-based noise spatial clustering algorithm.

[0055] In some embodiments, after obtaining the second wind power data set (after global outliers are removed) and the third wind power data set (after local outliers are removed), a unified fourth wind power data set needs to be determined. This can be achieved by taking the intersection or merging of the two. Then, the density-based spatial clustering algorithm with noise (DBSCAN) is applied to the fourth wind power data set. The DBSCAN algorithm divides the data points into different clusters according to the density of the data points and their mutual accessibility, and identifies the noise points (i.e., points that do not belong to any cluster).

[0056] Step S250: training a plurality of back propagation models based on the fifth wind power data set, and repairing the abnormal power data and the abnormal wind speed data based on the plurality of back propagation models to obtain a cleaned wind power data set.

[0057] In some embodiments, a back propagation (BP) neural network is a commonly used supervised learning algorithm that predicts unknown data by learning the mapping relationship between input and output. In the fifth wind power data set, multiple back propagation models are trained and used to predict and repair abnormal power data and abnormal wind speed data. The repaired data is merged with the non-abnormal values ​​in the original data to obtain a cleaned wind power data set.

[0058] The wind power data cleaning method, device, equipment and storage medium provided by the embodiment of the present invention first pre-process the original wind power data set, and then remove the missing wind speed data and power data in the original wind power data set to obtain a first wind power data set; use the isolation forest algorithm to detect and remove the global outliers of the wind speed data and power data in the first wind power data set to obtain a second wind power data set; use the outlier detection algorithm based on boundary-aware local density to detect and remove the local outliers of the wind speed data and power data in the second wind power data set to obtain a third wind power data set; based on the second wind power data set and the third wind power data set, determine a fourth wind power data set; use the density-based noise spatial clustering algorithm to detect and remove the fourth wind power data set to obtain a fifth wind power data set; train multiple back propagation models based on the fifth wind power data set, and repair the abnormal power data and abnormal wind speed data based on the multiple back propagation models to obtain a cleaned wind power data set. Thus, the present invention provides a multi-mode machine learning data preprocessing method consisting of multiple outlier detection algorithms and BPNN that can realize abnormal data repair. By systematically combining them, their respective advantages are fully utilized to perform more accurate outlier detection and data repair; thereby meeting the requirements for real-time, accuracy, integrity, consistency, scalability and automatic alarm of wind power data cleaning algorithms in the intelligent upgrade of the functions of the centralized control station monitoring system.

[0059] In some embodiments, the above step S220 can be implemented by the following steps S221 to S224:

[0060] Step S221, randomly select β sample points from the first wind power data set as sub-samples and put them into the root node of an isolated tree; randomly specify a dimension q, and randomly generate a cutting point p within the current node data range, where p is generated between the maximum value and the minimum value of the specified dimension.

[0061] Step S222: The cutting point p generates a hyperplane, and the sample points with a dimension less than p are placed in the left child node, and the sample points with a dimension greater than or equal to p are placed in the right child node.

[0062] Step S223, determine whether only one sample point of all leaf nodes reaches the specified height. If so, the construction of a single isolated tree is completed. If not, return to the previous step.

[0063] Step S224, repeat the above steps t times, so as to obtain t isolated trees and generate an isolation forest model; then, substitute each sample point x into the isolation forest model and calculate the anomaly score, namely:

[0064] ;

[0065] In the formula, h(x) is the height of sample x in the isolated tree, and c(β) is the average path length; if the c(β) anomaly score is close to 1, it is an abnormal sample; if the c(β) anomaly score is much less than 0.5, it is a normal sample; if the c(β) anomaly score is about 0.5, it means that there is no abnormal sample.

[0066] In some embodiments, the above step S230 can be implemented by the following steps S231 to S234:

[0067] Step S231: determining the first wind power data set as data to be detected.

[0068] Step S232: Calculate the current data point x in the first wind power data set. i The sum of the squared distances of the k nearest neighbors gives the local density.

[0069] Step S233, calculate the distance between the current data point and all other data points in the first wind power data set to obtain a distance matrix; based on the local density, calculate the sum of the number of elements in the distance matrix that meet the preset conditions; the preset condition indicates that the element value in the distance matrix is ​​less than the local density value to determine whether it is an abnormal point.

[0070] Step S234, based on the neighborhood density and deviation parameter of each data point in the first wind power data set, determine the neighborhood boundary of each data point; compare each data point with the corresponding neighborhood boundary, and determine the data points located outside the neighborhood boundary and with lower neighborhood density as outliers; and screen the abnormal factors based on a preset threshold to obtain the third wind power data set.

[0071] In some embodiments, the above step S240 can be implemented by the following steps S241 to S243:

[0072] Step S241: Integrate the common data in the second wind power data set and the third wind power data set to obtain the fourth wind power data set; and set the radius and the minimum number of sample points.

[0073] Step S242, traverse the fourth wind power data set to determine whether each data point in the fourth wind power data set is clustered and marked; if not, classify the core points within the radius neighborhood of the core object and mark the core points; traverse all core pairs of the current class and mark them.

[0074] Step S243: output all clusters, core points, boundary points and outliers in the fourth wind power data set, and remove abnormal data to obtain the fifth wind power data set.

[0075] Based on the above embodiment, the above step S250 can be implemented by the following steps S251 to S252:

[0076] Step S251, using the fifth wind power data set and the back propagation neural network to train the multiple back propagation models; wherein the first back propagation model is used to represent the relationship between wind speed and power at time t; the second back propagation model is used to represent the relationship between power and wind speed at time t; the third back propagation model is used to represent the relationship between power at time tk, ..., t-1 and power at time t; the fourth back propagation model is used to represent the relationship between wind speed at time tk, ..., t-1 and wind speed at time t.

[0077] Step S252, using the first back propagation model, repairing the abnormal power data at time t based on the wind speed data at time t; using the second back propagation model, repairing the abnormal wind speed data at time t based on the power data at time t; if the wind speed data and power data at time t are both eliminated, using the third back propagation model, according to the power data at time tk, ..., t-1, repairing the power data at time t, and using the fourth back propagation model according to the wind speed data at time tk, ..., t-1, repairing the wind speed data at time t, to obtain the cleaned wind power data set.

[0078] The following describes an exemplary application of an embodiment of the present invention in a practical application scenario.

[0079] The technical solution adopted by the present invention is to use random forest to detect and remove global outliers of data{1} to obtain data{2}; then use BLDOD to detect and remove local outliers of data{1} to obtain data{3}; integrate the common data of data{2} and data{3} to obtain data{4}; then use DBSCAN algorithm to detect and remove abnormal data to obtain data{5}; use data{5} to train BP model; and then use BP model to repair data. In this way, missing values, local outliers and global outliers in wind speed and power data can be identified and data repaired.

[0080] The present invention provides a multi-mode machine learning data preprocessing method for wind power data, a detection process, such as Figure 3 As shown, the algorithm includes the following steps:

[0081] Step 1: preprocess the wind power data set Original-data, where the wind power data set includes wind speed and power data. Eliminate the missing data in the data to be processed Original-data and its wind speed or power data together to obtain data{1}, and record the search address where the eliminated data is located.

[0082] Step 2: Use random forest to detect and remove global outliers in the wind speed and power data of data{1} to obtain data{2}, and record the retrieval address of the removed data.

[0083] Step 3: Use BLDOD to detect and remove local outliers in the wind speed and power data of data{1} to obtain data{3}, and record the retrieval address of the removed data.

[0084] Step 4: Integrate the common data in data{2} and data{3} to obtain data{4}; then, detect and remove abnormal data from data{4} using the DBSCAN density detection algorithm to obtain data{5}, and record the retrieval address of the removed data.

[0085] Step 5, use data{5} and BPNN to train different BPmodels. Among them, BPmodel{1} represents the relationship between wind speed WindSpeed(t) and power Power(t) at time t; BPmodel{2} represents the relationship between power Power(t) and wind speed WindSpeed(t) at time t; BPmodel{3} represents the relationship between power Power(tk, …, t-1) and power Power(t) at time tk, …, t-1; BPmodel{4} represents the relationship between wind speed WindSpeed(tk, …, t-1) and wind speed WindSpeed(t) at time tk, …, t-1. Use BPmodel to fill in the deleted data. The abnormal power data at time t is repaired by BPmodel{1} based on the wind speed data at time t; the abnormal wind speed data at time t is repaired by BPmodel{2} based on the power data at time t. If the wind speed and power data at time t are both removed, the power data at time t can be repaired using BPmodel{3} based on the power data at time tk, ..., t-1, and the wind speed data at time t can be repaired using BPmodel{4} based on the wind speed data at time tk, ..., t-1.

[0086] It should be noted that in step 1, the wind power data set is first normalized to avoid adverse effects caused by excessive differences between data.

[0087] It should be noted that the isolation forest algorithm in step 2 uses a binary search tree structure to randomly split the data set in a recursive manner to isolate abnormal samples. Since abnormal samples are small in number and isolated from most samples, they will be isolated earlier, and have better robustness than traditional clustering algorithms.

[0088] It should be noted that the Boundary-aware Local Density Outlier Detection (BLDOD) algorithm in step 3 is an unsupervised local outlier detection method. Specifically, the BLDOD algorithm first calculates the neighborhood density of each data point, which can be done by evaluating the distance between the data point and its neighbors. Then, the boundary of each point is determined based on the neighborhood density and deviation parameter of the data point. The boundary setting is based on the Chebyshev inequality, which can be used to estimate the fluctuation range of the data.

[0089] Step 2 above can be achieved by:

[0090] Step 2.1, Data preparation and feature selection: Ensure that data{1} is prepared, that is, all missing wind speed or power data has been removed from the original data. Determine the features to be used to build the isolated random forest model. In this scenario, the main features are wind speed and power data.

[0091] Step 2.2, randomly select β sample points as subsamples and put them into the root node of an isolated tree. Randomly specify a dimension q, and randomly generate a cutting point p within the current node data range, where p is generated between the maximum and minimum values ​​of the specified dimension.

[0092] Step 2.3, the cutting point p generates a hyperplane, and the sample points with specified dimensions less than p are placed in the left child node, and those greater than or equal to p are placed in the right child node.

[0093] Step 2.4, determine whether there is only one sample point in all leaf nodes that reaches the specified height. If so, the construction of a single isolated tree is completed. If not, return to step 2.2.

[0094] Step 2.5, repeat the above operation t times to obtain t isolated trees and generate an isolation forest. Then, each sample point x is brought into the isolation forest and the anomaly score is calculated, that is:

[0095] (1);

[0096] In the formula, h(x) is the height of sample x in the isolated tree, c(β) is the average path length; if the anomaly score of c(β) is close to 1, it is an abnormal sample; if the anomaly score of c(β) is much less than 0.5, it is a normal sample; if the anomaly score of c(β) is about 0.5, it means that there is no abnormal sample. In this way, the global outliers of wind speed and power data of data{1} are detected and removed to obtain data{2}, and the retrieval address of the removed data is recorded.

[0097] Step 3 above can be achieved by:

[0098] Step 3.1: Use the data{1} obtained in step 1 as the data X to be detected.

[0099] Step 3.2, calculate the current data point x by formula 2 i The sum of the squared distances of the k-nearest neighbors gives the local density ,

[0100] (2);

[0101] In the formula, is the local density, λ is the weight factor set, is the variance vector of the current data point x.

[0102] Step 3.3, calculate the distance between all other data points and the current data point Xi by using formula 3, and calculate the distance matrix D by using formula 4, satisfying the condition diag(DDT) ½ The sum of the number of elements ≤ε.

[0103] (3);

[0104] In the formula, D is the distance matrix, X is the data to be detected, Indicates the current data point.

[0105] (4);

[0106] In the formula, Nj represents the sum of the number of conditions that satisfy formula (4), and the diag() function returns a diagonal matrix, where the elements on the diagonal come from the diagonal elements of the input matrix; DD T It refers to the transpose of the D matrix multiplied by itself, and 1 is the local density obtained by formula (1).

[0107] Step 3.4, determine the boundary of each point based on the neighborhood density and deviation parameter of the data point. The boundary setting is based on the Chebyshev inequality, which is used to estimate the fluctuation range of the data. By comparing the data point with its neighborhood boundary, the BLDOD algorithm can identify points that are outside the boundary and have a low neighborhood density as outliers. Finally, the anomaly factors are sorted by size, and the abnormal data is filtered according to the threshold. The local outliers of data{1} are detected and eliminated to obtain data{3}, and the retrieval address of the eliminated data is recorded;

[0108] (5);

[0109] Where, μ=E(X), E(X) is the mathematical expectation of the random variable X; is the variance of the random variable X; is a variable, and .

[0110] Step 4 above can be achieved by:

[0111] Step 4 is as follows:

[0112] Step 4.1: Integrate the data {2} obtained in step 2 and data {3} obtained in step 3 to obtain data {4}. At the same time, set the following two main parameters: radius eps and minimum number of sample points min_samples.

[0113] The radius eps refers to the maximum distance from one observation to another, beyond which the two are no longer considered neighbors. The minimum number of sample points min_samples refers to the minimum number of neighbors.

[0114] Step 4.2, traverse data{4} and determine whether each data point is marked by clustering. If not, the core points in the neighborhood of a core object eps are classified into one category through the following set of formulas, and the core points are marked. Then traverse all core objects of the current class and mark them until all data points are traversed;

[0115] (6);

[0116] in, is a dataset, dist(p,q) is the distance metric between objects p and q, is the critical variable to be set.

[0117] Step 4.3, output all clusters, core points, boundary points and outliers in data{4}, remove abnormal data to obtain data{5}, and record the retrieval address of the removed data.

[0118] Step 5 above can be achieved by:

[0119] Step 5.1, use data{5} and BPNN to train different BPmodels. Among them, BPmodel. Among them, BPmodel{1} represents the relationship between wind speed WindSpeed(t) and power Power(t) at time t; BPmodel{2} represents the relationship between power Power(t) and wind speed WindSpeed(t) at time t; BPmodel{3} represents the relationship between power Power(tk, …, t-1) and power Power(t) at time tk, …, t-1; BPmodel{4} represents the relationship between wind speed WindSpeed(tk, …, t-1) and wind speed WindSpeed(t) at time tk, …, t-1.

[0120] Step 5.2, use BPmodel to fill in the deleted data. The abnormal power data at time t is repaired using BPmodel{1} based on the wind speed data at time t; the abnormal wind speed data at time t is repaired using BPmodel{2} based on the power data at time t. If both the wind speed and power data at time t are deleted, the power data at time t can be repaired using BPmodel{3} based on the power data at time tk, ..., t-1, and the wind speed data at time t can be repaired using BPmodel{4} based on the wind speed data at time tk, ..., t-1.

[0121] The data cleaning performance of the present invention was tested, and the results are as follows:

[0122] Figure 4 The abnormal data distribution diagram of data set #1-#5 is given. The data set used is Data-1. Five test sets (#1-#5) are selected from it. Each test set contains 10,000 continuous time points. In order to facilitate the performance of IFBD, the normal data in the test set is replaced with abnormal data. Figure 4 , 100 abnormal data are randomly distributed in each data set and the data of power and wind speed are independent. In this test, the abnormal data are divided into global outliers, local outliers, and missing values. Among them, the specific settings of abnormal data are: global outliers account for 20% of the total number of outliers, local outliers account for 70% of the total number of outliers, and missing values ​​account for 10% of the total number of outliers.

[0123] Figure 5 (a) is the distribution diagram of data set #5 containing abnormal data; Figure 5 (b) is the distribution diagram of the data set after iForest removes global outliers and BLDOD algorithm identifies local outliers; Figure 5 (c) is the distribution diagram of the data set after removing distribution abnormal data; Figure 5 (d) is the distribution diagram of the data set after cleaning. In Figure 5 (b), there is a phenomenon of normal data being judged as abnormal data near the power values ​​of 0 MW and 16 MW. For this phenomenon, IFBD will retain the information of misjudged data in this section and restore these data in the data repair link of BPNN. Although the repaired data is different from the original data, the repaired data conforms to the distribution characteristics of the data in the data set and has little impact on the subsequent network training. Figure 5 (c) uses the DBSCAN algorithm to identify the distribution abnormal data in the wind speed-wind power scatter plot, such as data points in sparse density areas, which can significantly reduce misjudgments. Figure 5 (d) shows the good data that is finally retained.

[0124] Figure 6 (a) is the result of repairing the wind speed value using the wind speed sequence change; Figure 6 (b) is the result of repairing the power value using the wind power sequence change. As can be seen from Figure 6, the repair result of the IFBD data cleaning system is very close to the normal data, which proves the effectiveness of the BPNN model in data repair, especially its robustness in the multi-source information collaborative repair scenario.

[0125] Table 1 is the performance evaluation table of IFBD. The repair results of data sets #1-#5 are evaluated from four dimensions: determination coefficient (R2), root mean square error (RMSE), true positive rate (TPR), and false positive rate (FPR). It can be seen from the results that the R2 of the repair results of abnormal data is above 0.95, and the RMSE is around 0.4. The repair results can well restore the relationship between wind power data. The TPR of the IFBD data cleaning system is above 90% and the FPR does not exceed 5%. Compared with the three methods of iForest, BLDOD, and DBSCAN, the performance of the IFBD data cleaning system is stronger than that of a single method. iForest, BLDOD, and DBSCAN are all abnormal scores of the obtained data. The higher the value of the abnormal score, the higher the possibility that the data is abnormal data. However, in Table 1, when the TPR of the three methods is not high, the value of their FPR is very high. This situation occurs mainly because a part of the wind power data misleads the single method. Therefore, this influence makes iForest, BLDOD, and DBSCAN unable to give full play to their own advantages. The IFBD data cleaning system gradually removes abnormal data based on the advantages of various methods. It achieves a high TPR and a very low FPR. The test results of abnormal data recognition of the IFBD data cleaning system in 5 data sets are above 90%, but all of them have a recognition rate of more than 95%. This phenomenon is because in the setting of abnormal data, some local outliers have little change compared to normal data, and these local outliers have little effect on the distribution and change characteristics of the original data. Therefore, this small part of local outliers is difficult to identify.

[0126] Table 1 Performance evaluation of IFBD

[0127]

[0128] Figure 7 Schematic diagram of the structure of the wind power data cleaning device provided by the embodiment of the present invention. Figure 7As shown, the wind power data cleaning device 700 includes: a first elimination module 701, which is used to pre-process the original wind power data set, and then eliminate the missing wind speed data and power data in the original wind power data set to obtain a first wind power data set; a second elimination module 702, which is used to detect and eliminate the global outliers of the wind speed data and power data in the first wind power data set by using an isolation forest algorithm to obtain a second wind power data set; a third elimination module 703, which is used to detect and eliminate the global outliers of the wind speed data and power data in the second wind power data set by using an outlier detection algorithm based on boundary-aware local density. The method comprises the following steps: detecting and removing local outliers of the wind power data and the power data to obtain a third wind power data set; a fourth elimination module 704 is used to determine a fourth wind power data set based on the second wind power data set and the third wind power data set; detecting and removing the fourth wind power data set by applying a density-based noise spatial clustering algorithm to obtain a fifth wind power data set; and a data repair module 705 is used to train multiple back propagation models based on the fifth wind power data set, and repair the power abnormality data and the wind speed abnormality data based on the multiple back propagation models to obtain a cleaned wind power data set.

[0129] In some embodiments, the second elimination module is further used to randomly select β sample points from the first wind power data set as subsamples and put them into the root node of an isolated tree; randomly specify a dimension q, and randomly generate a cutting point p within the current node data range, where p is generated between the maximum value and the minimum value of the specified dimension; the cutting point p generates a hyperplane, and the sample points with the specified dimension less than p are placed in the left child node, and those greater than or equal to p are placed in the right child node; determine whether all leaf nodes have only one sample point reaching the specified height, if so, the construction of a single isolated tree is completed, if not, return to the previous step; repeat the above steps t times, so as to obtain t isolated trees and generate an isolation forest model; then, substitute each sample point x into the isolation forest model, and calculate the abnormality score, that is:

[0130] ;

[0131] In the formula, h(x) is the height of sample x in the isolated tree, and c(β) is the average path length; if the c(β) anomaly score is close to 1, it is an abnormal sample; if the c(β) anomaly score is much less than 0.5, it is a normal sample; if the c(β) anomaly score is about 0.5, it means that there is no abnormal sample.

[0132] In some embodiments, the third elimination module is further used to determine the first wind power data set as the data to be detected; calculate the current data point x in the first wind power data set iThe sum of the squares of the distances of the k nearest neighbors is used to obtain the local density; the distance between the current data point and all other data points in the first wind power data set is calculated to obtain a distance matrix; based on the local density, the sum of the number of elements in the distance matrix that meet the preset conditions is calculated; the preset condition indicates that the element value in the distance matrix is ​​less than the local density value, so as to determine whether it is an abnormal point; based on the neighborhood density and deviation parameter of each data point in the first wind power data set, the neighborhood boundary of each data point is determined; each data point is compared with the corresponding neighborhood boundary, and the data points located outside the neighborhood boundary and with lower neighborhood density are determined as outliers; the abnormal factors are screened based on the preset threshold to obtain the third wind power data set.

[0133] In some embodiments, the fourth elimination module is also used to integrate the common data in the second wind power data set and the third wind power data set to obtain the fourth wind power data set; and set the radius and the minimum number of sample points; traverse the fourth wind power data set to determine whether each data point in the fourth wind power data set is clustered; if not, the core points within the radius neighborhood of the core object are grouped and marked; traverse all core points of the current class and mark them; output all clusters, core points, boundary points and outliers in the fourth wind power data set, and eliminate abnormal data to obtain the fifth wind power data set.

[0134] In some embodiments, the data repair module is also used to train the multiple back propagation models using the fifth wind power data set and the back propagation neural network; wherein the first back propagation model is used to represent the relationship between wind speed and power at time t; the second back propagation model is used to represent the relationship between power and wind speed at time t; the third back propagation model is used to represent the relationship between power at time tk, ..., t-1 and power at time t; the fourth back propagation model is used to represent the relationship between wind speed at time tk, ..., t-1 and wind speed at time t; using the first back propagation model, the abnormal power data at time t is repaired based on the wind speed data at time t; using the second back propagation model, the abnormal wind speed data at time t is repaired based on the power data at time t; if the wind speed data and power data at time t are both eliminated, the third back propagation model is used to repair the power data at time t according to the power data at time tk, ..., t-1, and the fourth back propagation model is used to repair the wind speed data at time t according to the wind speed data at time tk, ..., t-1, to obtain the cleaned wind power data set.

[0135] It should be noted that the description of the device of the embodiment of the present invention is similar to the description of the above method embodiment, and has similar beneficial effects as the same method embodiment, so it will not be repeated. For technical details not disclosed in the embodiment of the device, please refer to the description of the method embodiment of the present invention for understanding.

[0136] It should be noted that in the embodiment of the present invention, if the above-mentioned wind power data cleaning method is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a terminal to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), disk or optical disk and other media that can store program code. In this way, the embodiment of the present invention is not limited to any specific combination of hardware and software.

[0137] Correspondingly, an embodiment of the present invention provides a wind power data cleaning device, Figure 8 Schematic diagram of the structure of the wind power data cleaning device provided by the embodiment of the present invention. Figure 8 As shown, the wind power data cleaning device 800 at least includes: a processor 801 and a computer-readable storage medium 802 configured to store executable instructions, wherein the processor 801 generally controls the overall operation of the wind power data cleaning device. The computer-readable storage medium 802 is configured to store instructions and applications executable by the processor 801, and can also cache data to be processed or processed by each module in the processor 801 and the wind power data cleaning device 800, which can be implemented through flash memory (FLASH) or random access memory (RAM, Random Access Memory).

[0138] An embodiment of the present invention provides a storage medium storing executable instructions, wherein the executable instructions are stored. When the executable instructions are executed by a processor, the processor will be caused to execute the method provided by the embodiment of the present invention, for example, Figure 2 The method shown.

[0139] In some embodiments, the storage medium can be a computer-readable storage medium, for example, a ferroelectric random access memory (FRAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory, a magnetic surface memory, an optical disk, or a compact disk read-only memory (CD-ROM); it can also be various devices including one or any combination of the above memories.

[0140] In some embodiments, executable instructions may be in the form of a program, software, software module, script or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a stand-alone program or as a module, component, subroutine or other unit suitable for use in a computing environment.

[0141] As an example, executable instructions may, but need not correspond to, a file in a file system, may be stored as part of a file storing other programs or data, such as one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files storing one or more modules, subroutines, or code portions). As an example, executable instructions may be deployed to be executed on one electronic device, or on multiple electronic devices located at one location, or on multiple electronic devices distributed at multiple locations and interconnected by a communication network.

[0142] The above description is only an embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement and improvement made within the spirit and scope of the present invention are included in the protection scope of the present invention.

[0143] It should be understood that "one embodiment" or "an embodiment" mentioned throughout the specification means that specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present invention, the size of the serial number of the above-mentioned processes 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 embodiment of the present invention. The serial numbers of the above-mentioned embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0144] It should be noted that, in this article, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of further restrictions, an element defined by the statement "comprises one..." does not exclude the presence of other identical elements in the process, method, article or device including the element. In several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored, or not executed.

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

Claims

1. A wind power data cleaning method, characterized in that: The method comprises: After preprocessing the obtained original wind power data set, the missing wind speed data and power data in the original wind power data set are removed to obtain a first wind power data set; Using an isolation forest algorithm to detect and remove global outliers in the wind speed data and the power data in the first wind power data set, to obtain a second wind power data set; Using an outlier detection algorithm based on boundary-aware local density, local outliers of wind speed data and power data in the second wind power data set are detected and removed to obtain a third wind power data set; Based on the second wind power data set and the third wind power data set, a fourth wind power data set is determined; by applying a density-based noise spatial clustering algorithm, the fourth wind power data set is detected and eliminated to obtain a fifth wind power data set; A plurality of back propagation models are trained based on the fifth wind power data set, and the abnormal power data and the abnormal wind speed data are repaired based on the plurality of back propagation models to obtain a cleaned wind power data set.

2. The method according to claim 1, characterized in that The method of using the isolation forest algorithm to detect and remove global outliers of the wind speed data and the power data in the first wind power data set to obtain a second wind power data set includes: Randomly select β sample points from the first wind power data set as subsamples and put them into the root node of an isolated tree; randomly specify a dimension q, and randomly generate a cutting point p within the current node data range, where p is generated between the maximum value and the minimum value of the specified dimension; The cutting point p generates a hyperplane, and the sample points with a specified dimension less than p are placed in the left child node, and those greater than or equal to p are placed in the right child node; Determine whether there is only one sample point among all leaf nodes that reaches the specified height. If so, the construction of a single isolated tree is completed. If not, return to the previous step. Repeat the above steps t times to obtain t isolated trees and generate an isolation forest model; then, substitute each sample point x into the isolation forest model and calculate the anomaly score, that is: ； In the formula, h(x) is the height of sample x in the isolated tree, and c(β) is the average path length; if the c(β) anomaly score is close to 1, it is an abnormal sample; if the c(β) anomaly score is much less than 0.5, it is a normal sample; if the c(β) anomaly score is about 0.5, it means that there is no abnormal sample.

3. The method according to claim 1, characterized in that The outlier detection algorithm based on boundary-aware local density is used to detect and remove local outliers of wind speed data and power data in the second wind power data set to obtain a third wind power data set, including: Determining the first wind power data set as data to be detected; Calculate the current data point x in the first wind power data set i The sum of the squared distances of the k nearest neighbors gives the local density; Calculate the distance between the current data point and all other data points in the first wind power data set to obtain a distance matrix; based on the local density, calculate the sum of the number of elements in the distance matrix that meet a preset condition; the preset condition indicates that the element value in the distance matrix is ​​less than the local density value, so as to determine whether it is an abnormal point; Based on the neighborhood density and deviation parameters of each data point in the first wind power data set, the neighborhood boundary of each data point is determined; each data point is compared with the corresponding neighborhood boundary, and the data points located outside the neighborhood boundary and with low neighborhood density are determined as outliers; and the abnormal factors are screened based on a preset threshold to obtain the third wind power data set.

4. The method according to claim 1, characterized in that: The method further comprises: determining a fourth wind power data set based on the second wind power data set and the third wind power data set; and detecting and eliminating the fourth wind power data set by applying a density-based noise spatial clustering algorithm to obtain a fifth wind power data set, including: Integrate the common data in the second wind power data set and the third wind power data set to obtain the fourth wind power data set; and set the radius and the minimum number of sample points; Traversing the fourth wind power data set, determining whether each data point in the fourth wind power data set is clustered and marked; if not, classifying the core points within the radius neighborhood of the core object and marking the core points; traversing all core pairs of the current class and marking them; All clusters, core points, boundary points and outliers in the fourth wind power data set are output, and abnormal data are removed to obtain the fifth wind power data set.

5. The method according to claim 1, characterized in that The step of training a plurality of back propagation models based on the fifth wind power data set, and repairing the abnormal power data and the abnormal wind speed data based on the plurality of back propagation models to obtain a cleaned wind power data set includes: The plurality of back propagation models are trained using the fifth wind power data set and the back propagation neural network; wherein the first back propagation model is used to represent the relationship between wind speed and power at time t; the second back propagation model is used to represent the relationship between power and wind speed at time t; the third back propagation model is used to represent the relationship between power at time tk, ..., t-1 and power at time t; the fourth back propagation model is used to represent the relationship between wind speed at time tk, ..., t-1 and wind speed at time t; Using the first back propagation model, the abnormal power data at time t is repaired based on the wind speed data at time t; using the second back propagation model, the abnormal wind speed data at time t is repaired based on the power data at time t; if the wind speed data and the power data at time t are both eliminated, using the third back propagation model, according to the power data at time tk, ..., t-1, the power data at time t is repaired, and according to the wind speed data at time tk, ..., t-1, the wind speed data at time t is repaired using the fourth back propagation model to obtain the cleaned wind power data set.

6. A wind power data cleaning device, characterized in that: The device comprises: A first elimination module is used to eliminate missing wind speed data and power data from the original wind power data set after preprocessing the obtained original wind power data set, so as to obtain a first wind power data set; A second elimination module is used to detect and eliminate global outliers of wind speed data and power data in the first wind power data set by using an isolation forest algorithm to obtain a second wind power data set; a third elimination module, configured to detect and eliminate local outliers of the wind speed data and the power data in the second wind power data set by using an outlier detection algorithm based on boundary-aware local density, so as to obtain a third wind power data set; a fourth elimination module, configured to determine a fourth wind power data set based on the second wind power data set and the third wind power data set; and detect and eliminate the fourth wind power data set by applying a density-based noise spatial clustering algorithm to obtain a fifth wind power data set; The data repair module is used to train multiple back propagation models based on the fifth wind power data set, and repair the abnormal power data and abnormal wind speed data based on the multiple back propagation models to obtain a cleaned wind power data set.

7. An electronic device, characterized in that: include: A memory for storing executable instructions; The processor is used to implement the wind power data cleaning method according to any one of claims 1 to 5 when executing the executable instructions stored in the memory.

8. A computer-readable storage medium, characterized in that: Executable instructions are stored, which are used to cause a processor to execute the executable instructions to implement the wind power data cleaning method according to any one of claims 1 to 5.

Images