Power grid operation data cleaning method and system based on machine learning

Abstract

The invention relates to a machine learning-based power grid operation data cleaning method and system. The method comprises the following steps of obtaining real-time operation data of a power grid structure; sequentially performing low-dimensional representation and data reconstruction on the real-time operation data by adopting a pre-trained auto-encoder model, and performing screening based on the reconstructed data to obtain abnormal data and normal data; based on the abnormal data, adopting an Apriori algorithm to classify the abnormal data into abnormal state data and dirty data, and removing the dirty data; and adopting a pre-trained GRU recurrent neural network model to complement the abnormal data from which the dirty data is removed to obtain the complemented abnormal data, and completing the data cleaning process. Compared with the prior art, the method has the advantages of realizing efficient and low-threshold power grid operation data cleaning and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid data processing, and in particular to a method and system for cleaning power grid operation data based on machine learning. Background Art

[0002] The application of power grid intelligence and big data has promoted the research and development of power system operation data cleaning technology. With the improvement of the level of power grid intelligence, the power system is increasingly dependent on real-time perception and precise dispatching decisions, all of which are based on high-quality data. Therefore, the research on power grid operation data cleaning technology has emerged, aiming to improve the accuracy and efficiency of power system dispatching decisions through data cleaning technology, optimize the operating cost of the power system, and ensure the reliable power supply and sustainable development of the power system. This has important practical significance for power system operators and decision makers, and is also in line with the development direction of power system transformation and upgrading.

[0003] The development of power system and power equipment status perception technology relies on a large amount of data provided by IoT sensors and power system detection equipment. These data support the power grid production command decision-making system and are an important guarantee for the reliable operation of the power grid. With the application of big data and artificial intelligence technology in equipment status analysis, the level of intelligent equipment management has been effectively improved. The development of smart grids has promoted the exploration and application of big data in power systems, and at the same time has put forward higher requirements for the cleaning of power grid operation data. The construction of an integrated, high-speed, two-way communication network for power systems provides a good foundation for real-time perception of the operating status of power systems and power equipment.

[0004] However, working in complex and changeable natural and industrial conditions can easily lead to various problems with detection data, such as data redundancy, data loss, data errors, etc. These detection data problems will further affect the effectiveness of power system and power equipment status perception, and further affect the effectiveness of power system dispatching decisions. Therefore, the research on power grid operation data cleaning came into being, and power grid operation data cleaning has also become an important means to improve the quality of power grid data. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for cleaning power grid operation data based on machine learning with high efficiency and low threshold.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for cleaning power grid operation data based on machine learning, comprising the following steps:

[0008] Obtain real-time operating data of the power grid structure;

[0009] The pre-trained autoencoder model is used to perform low-dimensional representation and data reconstruction on the real-time operation data in turn, and the reconstructed data is screened to obtain abnormal data and normal data;

[0010] Based on the abnormal data, the Apriori algorithm is used to classify the abnormal data into abnormal state data and dirty data, and the dirty data is removed;

[0011] A pre-trained GRU recursive neural network model is used to complete the abnormal data after removing the dirty data, so as to obtain the completed abnormal data and complete the data cleaning process.

[0012] Furthermore, the real-time operating data of the power grid structure includes the real-time operating temperature, voltage, current, power, and frequency of each device in the power grid structure.

[0013] Furthermore, the method further includes preprocessing the real-time operation data before sequentially performing low-dimensional representation and data reconstruction, wherein the preprocessing step includes:

[0014] The real-time operation data is standardized, wherein the expression of data standardization is:

[0015]

[0016] Where, X std is the real-time operation data after data standardization, X is the original real-time operation data obtained, μ is the mean, and σ is the standard deviation;

[0017] The real-time operation data after data standardization is normalized to obtain the normalized real-time operation data, completing the preprocessing process, wherein the normalized expression is:

[0018]

[0019] Where, X norm is the normalized real-time running data, X min is the minimum value in the data, X max is the maximum value in the data.

[0020] Furthermore, the step of obtaining abnormal data and normal data includes:

[0021] Calculate the reconstruction error of each data point in the real-time operation data based on the reconstruction data and the real-time operation data;

[0022] Set reconstruction error threshold;

[0023] It is determined whether the reconstruction error of each data point is greater than the reconstruction error threshold. If so, the data point is regarded as an abnormal data point. If not, the data point is regarded as a normal data point, thus completing the screening process of abnormal data and normal data.

[0024] Furthermore, the training process of the autoencoder model includes:

[0025] Obtain historical operation data of the power grid structure;

[0026] Construct an autoencoder model, which includes an encoder and a decoder. The encoder is used to compress the input data into a low-dimensional representation, and the decoder is used to reconstruct the low-dimensional representation. The encoder expression formula is:

[0027] z=φ(W e ·x+b e )

[0028] The decoder formula is:

[0029]

[0030] In the formula, z is the low-dimensional representation of the data, φ is the activation function, and W e and b e are the weight and bias of the encoder respectively, x is the input data, To reconstruct the data, W d and b d are the weight and bias of the decoder respectively;

[0031] The historical operation data is input into the autoencoder model for training to minimize the reconstruction error between the input data and the reconstructed data to complete the training process, wherein the autoencoder model uses the mean square error as the loss function during the training process, and the expression of the loss function is:

[0032]

[0033] In the formula, n is the number of data samples, x i and are the input and reconstructed data of the i-th sample respectively.

[0034] Furthermore, before classifying the abnormal data, the abnormal data is binarized, and the step of binarizing the data includes:

[0035] The abnormal data is intercepted by using a sliding window with a length of L, and a linear fitting is performed on the intercepted abnormal data;

[0036] According to the linear fitting results, the data is normalized and the data binarization process is completed.

[0037] Furthermore, the step of dividing the data into abnormal state data and dirty data includes:

[0038] Based on the abnormal data, an Apriori algorithm is used to generate a candidate k-item set;

[0039] Check whether all non-empty subsets of the candidate k-item set are frequent. If so, proceed to the next step. If not, remove the infrequent non-empty subsets and proceed to the next step.

[0040] Calculate the support of each frequent candidate k-item set, and take the candidate item sets whose support is greater than or equal to the minimum support threshold as frequent item sets;

[0041] For each frequent item set and its frequent non-empty subset, generate candidate rules and calculate the confidence of the candidate rules. The candidate rules whose confidence is greater than or equal to the minimum confidence threshold are taken as association rules.

[0042] According to the generated association rules, the abnormal data is classified into abnormal state data and dirty data.

[0043] Furthermore, the calculation expressions of the support and confidence are respectively:

[0044]

[0045] Where P support (A→B) is the support, P confidence (A→B) is the confidence level, A and B are the antecedent and consequent of the rule, respectively, n count (A∪B) is the number of occurrences of A∪B in database D.

[0046] Furthermore, the training steps of the GRU recursive neural network model include:

[0047] Obtain the historical operation time series data of the power grid structure containing missing data, expressed as:

[0048] X=[x1,x2,…,x t ,…,x T ]

[0049] Where X is the historical operation time series data, x1, x2, x T is the running time series data point, x t are missing data points;

[0050] Initialize the values ​​of missing data:

[0051]

[0052] Build and train a GRU recurrent neural network model to complete missing data values, where the missing data values ​​after completion are expressed as:

[0053]

[0054] The present invention also provides a cleaning system for power grid operation data based on machine learning, comprising:

[0055] Data acquisition module: obtains real-time operation data of the power grid structure;

[0056] Reconstruction and screening module: using a pre-trained autoencoder model to perform low-dimensional representation and data reconstruction on the real-time operation data in turn, and screening based on the reconstructed data to obtain abnormal data and normal data;

[0057] Abnormal classification and elimination module: Based on the abnormal data, the Apriori algorithm is used to classify the abnormal data into abnormal state data and dirty data, and the dirty data is eliminated;

[0058] Completion module: A pre-trained GRU recursive neural network model is used to complete the abnormal data after removing the dirty data, to obtain the completed abnormal data, and to complete the data cleaning process.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] (1) The present invention uses an autoencoder model to filter out abnormal data in the power grid operation data, and further uses the Apriori algorithm to remove dirty data in the abnormal data and retain abnormal state data. Finally, the GRU recursive neural network model is used to capture the long-term dependencies in the data sequence, effectively predict and fill in the missing data, and achieve efficient and low-threshold power grid operation data cleaning.

[0061] (2) The present invention uses an autoencoder, where the encoder part compresses high-dimensional data into a latent space of lower dimension, thereby reducing the storage requirements of the data and extracting key features from the data. The decoder part reconstructs the compressed features back to the dimensions of the original data, so that important information is retained, while noise and redundancy are effectively removed, thus solving the problem of data redundancy.

[0062] (3) The present invention mines data association rules for abnormal data through the Apriori algorithm, classifies abnormal data by discovering frequent item sets and association rules in the data, and eliminates dirty data, including data errors. It then further captures long-term dependencies on the retained abnormal state data, predicts and fills in missing data, and solves problems such as missing data and data errors in operating data.

[0063] (4) The present invention significantly improves the quality of power grid data by effectively cleaning the power grid operation data, thereby providing a basis for improving the effectiveness of power system and power equipment status perception. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0065] Figure 2 It is the principle diagram of the AutoEncoder algorithm of the present invention;

[0066] Figure 3 Apriori algorithm schematic diagram of the present invention;

[0067] Figure 4 This is a schematic diagram of the GRU recursive neural network algorithm of the present invention. DETAILED DESCRIPTION

[0068] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0069] The present embodiment provides a method for cleaning power grid operation data based on machine learning. The method comprehensively considers various different power grid structure operation data pollution situations, introduces the AutoEncoder algorithm to perform low-dimensional representation of power grid data and reconstruct the data, and screens out abnormal data in the power grid structure operation data by comparing the data reconstruction error. On the basis of screening out abnormal data, the Apriori algorithm is further used to mine data association rules, and the abnormal data is classified and the "dirty data" in the abnormal data is eliminated by discovering frequent item sets and association rules in the data. The data in abnormal state is retained, and the data gaps generated by eliminating "dirty data" are effectively predicted and filled in by the present invention using the GRU recursive neural network algorithm. By capturing the long-term dependencies in the time series data, the present invention can effectively predict and fill in the missing data. Through the above machine learning method, the present invention assists users in achieving efficient and low-threshold cleaning of power grid structure operation data. Specifically, such as Figure 1 As shown, the method comprises the following steps:

[0070] S1. Collect the original data of the grid structure operation.

[0071] These data include the real-time operating temperature, voltage, current, power, frequency, etc. of transmission lines, transformers and other equipment. The data can be obtained from various sensors, smart meters and SCADA systems. By collecting the above operating data and combining it with the topological structure of the power grid, a power grid structure operation database is built.

[0072] S2. Introduce the AutoEncoder algorithm to filter abnormal data.

[0073] Normalize the collected data to be cleaned; compress the data to be cleaned to a low dimension by building an autoencoder model; train the autoencoder model to minimize the reconstruction error; calculate the reconstruction error, and filter the data with a reconstruction error greater than the specified value as abnormal data, otherwise it is normal data. The corresponding principles are as follows Figure 2 As shown, the specific implementation of this step includes:

[0074] (1) Realize data standardization and normalization of the grid structure operation data. Standardization is to convert the data into a distribution with a mean of 0 and a standard deviation of 1, that is, to distribute the data in [-1,1]. The specific formula for achieving standardization is as follows:

[0075]

[0076] In the formula: X is the original data, μ is the mean, σ is the standard deviation, X std The data after standardization.

[0077] Normalization is to scale the data to a specific range, usually [0,1]. The specific formula for normalization is as follows:

[0078]

[0079] Where: X mim is the minimum value of the data, X max is the maximum value of the data, X norm The data after normalization

[0080] (2) Construct an autoencoder model, which consists of an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, and the decoder reconstructs the low-dimensional representation into the original data.

[0081] The main task of the encoder is to compress the input data into a low-dimensional representation. Its structure is a multi-layer neural network. The specific formula for implementing the encoder is as follows:

[0082] z=φ(W e ·x+b e )

[0083] Where: x is the input data, W e and b e are the weights and biases of the encoder, φ is the activation function such as ReLU or sigmoid, and z is the low-dimensional representation of the data.

[0084] The task of the decoder is to reconstruct the original data from the low-dimensional representation. Its structure is also a multi-layer neural network. The formula for implementing the decoder is as follows:

[0085]

[0086] Where: is the reconstructed data, W d and b d are the weights and biases of the decoder respectively.

[0087] (3) Train the autoencoder model and use normal data to train the autoencoder so that it can effectively reconstruct the input data. During the training process, the input data x and the reconstructed data The reconstruction error between , the loss function used is the mean square error (MSE), and the calculation formula is as follows:

[0088]

[0089] Where n is the number of data samples, x i and are the input and reconstructed data of the i-th sample respectively.

[0090] (4) Calculate the reconstruction error. After training, use the autoencoder to reconstruct all data and calculate the reconstruction error of each data point. Data points with large reconstruction errors may be abnormal data. The calculation formula for reconstruction error is as follows:

[0091]

[0092] Where: e i is the reconstruction error of the ith data point, and m is the number of features for each data point.

[0093] (5) Determine the abnormal data and set a reconstruction error threshold. According to the set threshold, the data points with reconstruction errors greater than the threshold are screened out. These data points are the detected abnormal data.

[0094] S3. Use Apriori algorithm to classify abnormal data.

[0095] Binarize the collected continuous data into discrete data; use the Apriori algorithm to generate frequent item sets and association rules, and classify abnormal data based on association rules; retain abnormal state data and remove "dirty data". The corresponding principles are as follows Figure 3 As shown, the specific implementation of this step includes:

[0096] (1) Data preprocessing, data binarization. The Apriori algorithm is usually applied to discrete data, so continuous data needs to be binarized. First, the data to be processed is processed using a sliding window of length L. data The original data is intercepted, a linear fit is performed on the intercepted data, and then the data is normalized to the range of [0,1]. Finally, the [0,1] interval is divided into 8 segments and the data is represented by binary value.

[0097] (2) Use the Apriori algorithm to generate frequent item sets, starting from the frequent 1-item set, and gradually generate candidate k-item sets until no new frequent item sets can be generated. The process of generating candidate k-item sets can be completed by the connection operation. For each candidate k-item set, check whether all its (k-1) item subsets are frequent, where these subsets are used as the antecedents of the rule. If not, prune the candidate item set. The pruning process uses the "Apriori property", that is, all non-empty subsets of a frequent item set must also be frequent, so that each frequent item set and its non-empty subsets are combined into a candidate rule. Calculate the support of each candidate k-item set, and take the candidate item sets with support greater than or equal to the minimum support threshold as frequent item sets; for each frequent item set and its frequent non-empty subset, generate candidate rules and calculate the confidence of the candidate rules. Take the candidate rules with confidence greater than or equal to the minimum confidence threshold as association rules, that is, filter out strong rules that meet the conditions as the final association rule output. The support formula is as follows:

[0098]

[0099] Where: n count (A∪B) is the number of occurrences of A∪B in database D

[0100] The confidence formula is as follows:

[0101]

[0102] (3) Classify abnormal data. According to the generated association rules, the abnormal data is classified into different types, namely abnormal state data and "dirty data". The abnormal state data is retained and the "dirty data" (data quality issues) is eliminated.

[0103] S4. Use the GRU recurrent neural network algorithm to complete the missing data.

[0104] Collect the time series data that needs to be supplemented with missing data; normalize the time series data; build a GRU training model to output prediction results; use the trained GRU model to predict and supplement the missing data. The corresponding principles are as follows Figure 4 As shown, the specific implementation of this step includes:

[0105] (1) Data preparation: Collect and prepare time series data containing missing data. The collected time series data set is as follows:

[0106] X=[x1,x2,…,x t ,…,x T ]

[0107] Among them, some data points x t is missing.

[0108] (2) Handle missing values ​​and initialize them to a certain value, such as 0 or mean filling:

[0109]

[0110] (3) Build and train the GRU model. The core formula of the GRU unit is as follows:

[0111] Reset Gate t and update gate z t :

[0112] r t =σ(W r x t +U r h t-1 +b r )

[0113] z t =σ(W z x t +U z h t-1 +b z )

[0114] Candidate hidden state

[0115]

[0116] Current hidden state h t :

[0117]

[0118] Where: σ represents the sigmoid function, ⊙ represents element-wise multiplication.

[0119] (4) Use the GRU model for data completion. After the model training is completed, the GRU model can be used to predict and fill in missing data. Pass the input data containing missing values ​​to the model to obtain the prediction results, and use the prediction results output by the model to fill in the missing values. The formula is as follows:

[0120]

[0121] In summary, this embodiment introduces the AutoEncoder model to filter out abnormal data in the power grid operation data, further uses the Apriori algorithm to eliminate the "dirty data" in the abnormal data, retains the abnormal state data, and finally adopts the GRU recursive neural network model to effectively predict and fill in the missing data by capturing the long-term dependencies in the time series data.

[0122] Example 2

[0123] This embodiment provides a cleaning system for power grid operation data based on machine learning, including:

[0124] Data acquisition module: obtains real-time operation data of the power grid structure;

[0125] Reconstruction and screening module: using a pre-trained autoencoder model to perform low-dimensional representation and data reconstruction on the real-time operation data in turn, and screening based on the reconstructed data to obtain abnormal data and normal data;

[0126] Abnormal classification and elimination module: Based on the abnormal data, the Apriori algorithm is used to classify the abnormal data into abnormal state data and dirty data, and the dirty data is eliminated;

[0127] Completion module: A pre-trained GRU recursive neural network model is used to complete the abnormal data after removing the dirty data, to obtain the completed abnormal data, and to complete the data cleaning process.

[0128] The rest is the same as in Example 1.

[0129] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method 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), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0130] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0131] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0132] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0133] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0134] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0135] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for cleaning power grid operation data based on machine learning, characterized in that: The following steps are involved: Obtain real-time operating data of the power grid structure; The pre-trained autoencoder model is used to perform low-dimensional representation and data reconstruction on the real-time operation data in turn, and the reconstructed data is screened to obtain abnormal data and normal data; Based on the abnormal data, the Apriori algorithm is used to classify the abnormal data into abnormal state data and dirty data, and the dirty data is removed; A pre-trained GRU recursive neural network model is used to complete the abnormal data after removing the dirty data, so as to obtain the completed abnormal data and complete the data cleaning process.

2. The method for cleaning power grid operation data based on machine learning according to claim 1, characterized in that: The real-time operation data of the power grid structure includes the real-time operation temperature, voltage, current, power, and frequency of each device in the power grid structure.

3. The method for cleaning power grid operation data based on machine learning according to claim 1, characterized in that: The method further includes preprocessing the real-time operation data before sequentially performing low-dimensional representation and data reconstruction, wherein the preprocessing step includes: The real-time operation data is standardized, wherein the expression of data standardization is: In the formula, X std is the real-time operation data after data standardization, X is the original real-time operation data obtained, μ is the mean, and σ is the standard deviation; The real-time operation data after data standardization is normalized to obtain the normalized real-time operation data, completing the preprocessing process, wherein the normalized expression is: In the formula, X norm is the normalized real-time running data, X min is the minimum value in the data, X max is the maximum value in the data.

4. The method for cleaning power grid operation data based on machine learning according to claim 1, characterized in that: The steps of obtaining abnormal data and normal data include: Calculate the reconstruction error of each data point in the real-time operation data based on the reconstruction data and the real-time operation data; Set reconstruction error threshold; It is determined whether the reconstruction error of each data point is greater than the reconstruction error threshold. If so, the data point is regarded as an abnormal data point. If not, the data point is regarded as a normal data point to complete the screening process of abnormal data and normal data.

5. The method for cleaning power grid operation data based on machine learning according to claim 1, characterized in that: The training process of the autoencoder model includes: Obtain historical operation data of the power grid structure; Construct an autoencoder model, which includes an encoder and a decoder. The encoder is used to compress the input data into a low-dimensional representation, and the decoder is used to reconstruct the low-dimensional representation. The encoder expression formula is: z=φ(W e ·x+b e ) The decoder formula is: In the formula, z is the low-dimensional representation of the data, φ is the activation function, and W e and b e are the weight and bias of the encoder respectively, x is the input data, To reconstruct the data, W d and b d are the weight and bias of the decoder respectively; The historical operation data is input into the autoencoder model for training to minimize the reconstruction error between the input data and the reconstructed data to complete the training process, wherein the autoencoder model uses the mean square error as the loss function during the training process, and the expression of the loss function is: In the formula, n is the number of data samples, x i and are the input and reconstructed data of the i-th sample respectively.

6. The method for cleaning power grid operation data based on machine learning according to claim 1, characterized in that: Before classifying the abnormal data, the method further includes binarizing the abnormal data. The binarization step includes: The abnormal data is intercepted by using a sliding window with a length of L, and a linear fitting is performed on the intercepted abnormal data; According to the linear fitting results, the data is normalized and the data binarization process is completed.

7. The method for cleaning power grid operation data based on machine learning according to claim 1, characterized in that: The step of dividing the data into abnormal state data and dirty data includes: Based on the abnormal data, an Apriori algorithm is used to generate a candidate k-item set; Check whether all non-empty subsets of the candidate k-item set are frequent. If so, proceed to the next step. If not, remove the infrequent non-empty subsets and proceed to the next step. Calculate the support of each frequent candidate k-item set, and take the candidate item sets whose support is greater than or equal to the minimum support threshold as frequent item sets; For each frequent item set and its frequent non-empty subset, generate candidate rules and calculate the confidence of the candidate rules. The candidate rules whose confidence is greater than or equal to the minimum confidence threshold are taken as association rules. According to the generated association rules, the abnormal data is classified into abnormal state data and dirty data.

8. The method for cleaning power grid operation data based on machine learning according to claim 7, characterized in that: The calculation expressions of the support and confidence are respectively: Where P support (A→B) is the support, P confidence (A→B) is the confidence level, A and B are the antecedent and consequent of the rule, respectively, n count (A∪B) is the number of occurrences of A∪B in database D.

9. The method for cleaning power grid operation data based on machine learning according to claim 1, characterized in that: The training steps of the GRU recursive neural network model include: Obtain the historical operation time series data of the power grid structure containing missing data, expressed as: X=[x1,x2,…,x t ,…,x T ] Where X is the historical operation time series data, x1, x2, x T is the running time series data point, x t are missing data points; Initialize the values ​​of missing data: Build and train a GRU recurrent neural network model to complete missing data values, where the missing data values ​​after completion are expressed as:

10. A cleaning system for power grid operation data based on machine learning, characterized in that: include: Data acquisition module: obtains real-time operation data of the power grid structure; Reconstruction and screening module: using a pre-trained autoencoder model to perform low-dimensional representation and data reconstruction on the real-time operation data in turn, and screening based on the reconstructed data to obtain abnormal data and normal data; Abnormal classification and elimination module: Based on the abnormal data, the Apriori algorithm is used to classify the abnormal data into abnormal state data and dirty data, and the dirty data is eliminated; Completion module: A pre-trained GRU recursive neural network model is used to complete the abnormal data after removing the dirty data, to obtain the completed abnormal data, and to complete the data cleaning process.

Images