District behavior portraying method and device based on multi-source heterogeneous data and medium

Abstract

The invention discloses a transformer area behavior portraying method and device based on multi-source heterogeneous data and a medium. The method comprises the steps that a multi-source data set of a low-voltage transformer area is acquired; generating a portrait label for the data which can be directly monitored in the multi-source data set to obtain a first label data set; through deep learning and statistical calculation modes, portrait labels are generated for data which cannot be directly monitored in the multi-source data set, and a second label data set is obtained; and performing clustering processing on the multi-source data set, the first label data set and the second label data set to obtain a zone area behavior portrait. According to the transformer area behavior portrait method and device based on the multi-source heterogeneous data and the medium, the labels of the multi-source data set are constructed in a flexible mode according to the source of the data, features and modes of transformer area behaviors can be depicted more finely through clustering processing, and the problem that it is difficult to depict transformer area behavior portraits accurately can be solved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method, device and medium for profiling substation behavior based on multi-source heterogeneous data. Background Art

[0002] The continuous deepening of the construction of new power systems has made accurate profiling of low-voltage substation behaviors of great significance; currently, behavioral profiling is mainly focused on electric vehicles, sales websites, and data decision-making, and there is little research on substation profiling and accurate analysis of distribution network substations; as the penetration rate of "high proportion of new energy" and "high proportion of power electronic equipment" continues to increase, detailed distribution network substation profiling can enable operation and maintenance personnel to better understand the status quo of the substation and provide effective support for substation transformation and future decision-making. The traditional method is to use big data analysis and data mining technology to conduct in-depth analysis of substation data to portray substation behavior.

[0003] However, accurate portraits of distribution station areas currently face three major challenges: first, it is difficult to clean multi-source data in the area, making it difficult to screen out useful data; second, it is difficult to extract feature labels and characterize behaviors in the area, making it difficult to effectively identify feature areas; third, it is difficult to access data and process multi-source heterogeneous data, making it difficult to effectively analyze large amounts of text data and non-standardized data. Summary of the invention

[0004] The present invention provides a method, device and medium for profiling substation behavior based on multi-source heterogeneous data to solve the problem that it is difficult to accurately portray substation behavior.

[0005] Obtain multi-source datasets of low-pressure areas;

[0006] Generate a portrait label for the data that can be directly monitored in the multi-source data set to obtain a first label data set;

[0007] Generate a portrait label for the data that cannot be directly monitored in the multi-source data set by means of deep learning and statistical calculation to obtain a second label data set;

[0008] Clustering is performed on the multi-source data set, the first label data set, and the second label data set to obtain a substation behavior portrait.

[0009] For data that can be directly monitored, the present invention directly and quickly generates portrait labels, thereby improving the efficiency of data processing and label generation. For data that cannot be directly monitored, portrait labels are generated through deep learning and statistical calculations; among them, deep learning can capture complex data relationships and improve the accuracy of estimation and correction, while statistical calculations provide a solid theoretical basis and enhance the reliability of labels; this method of processing data by category is conducive to the flexible and accurate generation of labels. Clustering processing divides the substation behavior into different categories or groups, thereby more finely portraying the characteristics and patterns of the substation behavior. This refined portrait helps to more deeply understand the operating conditions and needs of the substation, thereby accurately portraying the substation behavior portrait.

[0010] Compared with the prior art, the present invention constructs labels for multi-source data sets in a flexible manner based on the source of the data, overcoming the limitation that traditional methods are difficult to process multiple categories of data. Through clustering processing, the intrinsic connections and potential laws between the data can be discovered, thereby more finely characterizing the characteristics and patterns of substation behavior. Therefore, it can solve the problem of difficulty in accurately portraying substation behavior portraits.

[0011] As a preferred solution, by means of deep learning and statistical calculation, a portrait label is generated for the data that cannot be directly monitored in the multi-source data set to obtain a second label data set, specifically:

[0012] If the intermediate parameters of the first data set in the multi-source data set are obtained by direct monitoring, generating a portrait label for the first data set by statistical calculation to obtain a statistical label data set;

[0013] If the intermediate parameters of the second data set in the multi-source data set are not obtained by direct monitoring, generating a portrait label for the second data set by deep learning to obtain a deep label data set;

[0014] The second label dataset is composed of the statistical label dataset and the deep label dataset.

[0015] For data sets obtained through direct monitoring, this preferred solution generates portrait labels through statistical calculations, which can fully utilize the original characteristics and distribution patterns of the data and reduce errors caused by human intervention or model complexity. For data sets obtained through indirect monitoring, labels are generated through deep learning models, which can capture complex patterns and potential relationships in the data and improve the accuracy and generalization ability of labels.

[0016] As a preferred solution, a portrait label is generated for the second data set by deep learning to obtain a deep label data set, specifically:

[0017] Normalizing the second data set to obtain a third data set;

[0018] If the degree of dispersion of the third data set is within a preset standard range, correcting the error of the third data set to obtain a fourth data set;

[0019] If the degree of discreteness of the third data set is not within the standard range, correcting the estimated value of the third data set by discretization classification to obtain a fourth data set;

[0020] The fourth data set is converted into a label output to obtain the deep label data set.

[0021] In this preferred solution, when the degree of dispersion of the third data set is within the preset standard range, it means that the overall volatility and variability of the data are within an acceptable range, but there may still be local errors or noise. Therefore, directly correcting the errors of the third data set can further reduce the impact of these errors on subsequent analysis and improve the accuracy and reliability of the data. When the degree of dispersion of the data set exceeds the standard range, it usually means that there are significant outliers or large variability in the data; directly processing such a data set may introduce unnecessary noise or errors, affecting the results of subsequent analysis, so by calculating and correcting the estimated value, this abnormal situation can be effectively dealt with, making the data more in line with the actual distribution and the needs of subsequent analysis.

[0022] As a preferred solution, the estimated value of the third data set is corrected by discretization classification to obtain a fourth data set, specifically:

[0023] According to the average difference and heterogeneity ratio of the data in the third data set, the third data set is divided into data to obtain a discretized classification data set;

[0024] Learning the long-term dependencies of the discretized classification data set through a long short-term memory network to generate an estimated value about a preset target;

[0025] Training the estimated value through a radial basis function network to obtain specific data information;

[0026] The estimated value is corrected based on the specific data information to obtain the fourth data set.

[0027] The combined use of long short-term memory network and radial basis function network in this preferred solution not only utilizes the advantages of long short-term memory network in capturing long-term dependencies, but also gives full play to the strengths of radial basis function network in nonlinear fitting and rapid learning. This combination makes the entire processing flow more robust and reliable.

[0028] As a preferred solution, the calculation formula of the heterogeneity ratio is specifically:

[0029]

[0030] Among them, Σf i is the total frequency of the values ​​in the third data set, f m is the frequency of data in the third data set that meets the preset normal standard.

[0031] As a preferred solution, after obtaining the multi-source data set of the low-voltage station area, the method further includes:

[0032] Classifying the multi-source data sets according to structural categories and update forms to obtain transition data sets;

[0033] Invalid data in the transition data set is deleted, and missing data in the transition data set is completed to obtain a multi-source data set after refinement processing.

[0034] This preferred solution classifies multi-source data sets according to structural categories and update forms, making the organizational structure of the data set clearer. Since invalid data may mislead the analysis results or reduce the analysis accuracy, deleting invalid data can reduce noise and redundancy in the data set and improve the overall quality of the data; and completing these missing data can make the refined multi-source data set more complete and reliable.

[0035] The present invention also provides a station area behavior profiling device based on multi-source heterogeneous data, including a data module, a first label module, a second label module and a profiling module;

[0036] Wherein, the data module is used to obtain multi-source data sets of low-voltage areas;

[0037] The first label module is used to generate a portrait label for the data that can be directly monitored in the multi-source data set to obtain a first label data set;

[0038] The second label module is used to generate a portrait label for the data that cannot be directly monitored in the multi-source data set by means of deep learning and statistical calculation, so as to obtain a second label data set;

[0039] The portrait module is used to perform clustering processing on the multi-source data set, the first label data set and the second label data set to obtain a station area behavior portrait.

[0040] As a preferred solution, the second label module includes a statistical unit, a depth unit and a construction unit;

[0041] The statistical unit is used to generate a portrait label for the first data set in the multi-source data set by statistical calculation to obtain a statistical label data set if the intermediate parameter of the first data set in the multi-source data set is obtained by direct monitoring;

[0042] The depth unit is used to generate a portrait label for the second data set in the multi-source data set by deep learning to obtain a deep label data set if the intermediate parameters of the second data set in the multi-source data set are not obtained by direct monitoring;

[0043] The construction unit is used to construct the second label data set from the statistical label data set and the deep label data set.

[0044] As a preferred solution, the depth unit includes a normalization subunit, a correction subunit, an estimation subunit and a conversion subunit;

[0045] The normalization subunit is used to perform normalization processing on the second data set to obtain a third data set;

[0046] The correction subunit is used to correct the error of the third data set to obtain a fourth data set if the discrete degree of the third data set is within a preset standard range;

[0047] The estimation subunit is configured to correct the estimated value of the third data set by discretization classification to obtain a fourth data set if the discreteness of the third data set is not within the standard range;

[0048] The conversion subunit is used to convert the fourth data set into a label output to obtain the deep label data set.

[0049] As a preferred solution, the estimated value of the third data set is corrected by discretization classification to obtain a fourth data set, specifically:

[0050] According to the average difference and heterogeneity ratio of the data in the third data set, the third data set is divided into data to obtain a discretized classification data set;

[0051] Learning the long-term dependencies of the discretized classification data set through a long short-term memory network to generate an estimated value about a preset target;

[0052] Training the estimated value through a radial basis function network to obtain specific data information;

[0053] The estimated value is corrected based on the specific data information to obtain the fourth data set.

[0054] As a preferred solution, the calculation formula of the heterogeneity ratio is specifically:

[0055]

[0056] Among them, ∑f i is the total frequency of the values ​​in the third data set, f mis the frequency of data in the third data set that meets the preset normal standard.

[0057] As a preferred solution, after the data module, a classification unit and a processing unit are also included:

[0058] The classification unit is used to classify the multi-source data set according to the structural category and the update form to obtain a transition data set;

[0059] The processing unit is used to delete invalid data in the transition data set and refine the missing data in the transition data set to obtain a refined multi-source data set.

[0060] The present application also provides a storage medium having a computer program stored thereon. The computer program is called and executed by a computer to implement the above-mentioned method for profiling substation behavior based on multi-source heterogeneous data. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of a method for profiling substation behavior based on multi-source heterogeneous data provided in an embodiment of the present application;

[0062] Figure 2 It is a schematic diagram of the data cleaning step method provided in an embodiment of the present application;

[0063] Figure 3 It is a block diagram of the statistical analysis method provided in the embodiment of the present application;

[0064] Figure 4 This is a schematic diagram of the label extraction process provided by the embodiment of the present application;

[0065] Figure 5 It is a block diagram of a deep learning label extraction method combining LSTM and RBF provided in an embodiment of the present application;

[0066] Figure 6 This is a model framework diagram of a deep learning label extraction method combining LSTM and RBF methods provided in an embodiment of the present application;

[0067] Figure 7 It is a flow chart of a deep learning extraction method combining LSTM and RBF methods provided in an embodiment of the present application;

[0068] Figure 8 It is a schematic diagram of a method for profiling substation behavior based on multi-source heterogeneous data provided in an embodiment of the present application;

[0069] Fig. 9 It is a structural schematic diagram of a station area behavior profiling device based on multi-source heterogeneous data provided in an embodiment of the present application. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0071] In the description of the present application, it should be understood that the terms "first", "second", "third" and "fourth" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first", "second", "third" and "fourth" may explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, "several" means two or more.

[0072] The embodiment of the present application provides a method for profiling substation behavior based on multi-source heterogeneous data, which is mainly used to overcome the problems of large amounts of useless data, inaccurate profiling, and difficulty in processing heterogeneous data in existing solutions, so as to accurately portray substation behavior.

[0073] Embodiment 1:

[0074] See also Figure 1 The embodiment of the present application provides a method for generating a behavior profile of a substation based on multi-source heterogeneous data, including S1 to S4. The specific implementation steps are as follows:

[0075] S1. Obtain multi-source datasets of low-pressure areas.

[0076] Step S1 of the embodiment of the present application includes S1.1 to S1.2; wherein S1.1 is a process of acquiring a multi-source data set, and S1.2 is a process of performing data cleaning, specifically:

[0077] S1.1. Obtain a multi-source data set of the low-voltage area; wherein the multi-source data set includes dispatching data, marketing data, metering data, production data and geographic data of the low-voltage area.

[0078] S1.2. Unify multi-source data sets into IoT data or collect them into data warehouses through MQTT (Message Queuing Telemetry Transport) protocol;

[0079] According to the data form, the multi-source data sets are classified to obtain transitional data sets; wherein the data forms include structured data, unstructured data, semi-structured data, real-time data and offline data;

[0080] Based on statistics, machine learning and other methods, invalid data, abnormal data and non-standard data in the transitional data set are screened and deleted, and missing data in the transitional data set are completed by difference completion, model completion, fitting completion and mode completion to obtain a multi-source data set after refined processing;

[0081] Multi-source data sets are buffered and stored in the form of archival data, historical data, and real-time data to establish a heterogeneous database.

[0082] To apply the embodiments of this application, please refer to Figure 2 , Figure 2 It is a schematic diagram of the data cleaning step method provided in the embodiment of the present application, which shows the process of performing data cleaning on a multi-source data set in embodiment S1.2.

[0083] In this embodiment S1.2, the multi-source data set is classified according to the structural category and the update form, so that the organizational structure of the data set is clearer. Since invalid data may mislead the analysis results or reduce the analysis accuracy, deleting invalid data can reduce the noise and redundancy in the data set and improve the overall quality of the data; and completing these missing data can make the multi-source data set after fine processing more complete and reliable.

[0084] S2. Generate portrait labels for data that can be directly monitored in the multi-source data set to obtain a first label data set.

[0085] Step S2 of the embodiment of the present application is specifically as follows:

[0086] Generate portrait labels for the data that can be directly monitored in the multi-source data set to obtain a first label data set; wherein the data that can be directly monitored in the multi-source data set includes data that can be directly obtained, such as power supply radius, transformer capacity, power supply line diameter, and number of power supply households.

[0087] S3. Generate portrait labels for data that cannot be directly monitored in multi-source data sets through deep learning and statistical calculations to obtain a second label data set.

[0088] Step S3 of the embodiment of the present application includes S3.1 to S3.3; wherein S3.1 is a process of generating a statistical label data set based on statistical analysis data, S3.2 is a process of processing deep learning data to generate a third data set, and S3.3 is a process of generating a deep label data set based on the third data set and obtaining a final second label data set, specifically:

[0089] S3.1. For data in a multi-source data set that cannot be directly monitored, if the intermediate parameters of the first data set in the multi-source data set are directly monitored, the first data set is standardized and normalized to obtain a unified data set; wherein the first data set is statistical analysis data, including average load rate, power stability, power fluctuation rate and average load rate, etc.;

[0090] According to the frequency of data deviation from the estimated value in unit time, the unified data set is analyzed for data frequency to obtain the data distribution, including the data's central tendency, degree of dispersion, and whether there are outliers.

[0091] Combined with the data distribution, mathematical formulas, modeling and other methods are used to perform correlation analysis on the unified data set to identify potential patterns in the data, and based on this, establish correlation relationships between data labels to obtain a statistical label data set;

[0092] The results of data frequency analysis and correlation analysis are displayed in an intuitive way such as visualization.

[0093] To apply the embodiments of this application, please refer to Figure 3 , Figure 3 It is a block diagram of the statistical analysis method provided in an embodiment of the present application, which represents the general process of generating a statistical label data set based on a first data set (statistical analysis data).

[0094] S3.2. For data in a multi-source data set that cannot be directly monitored, if the intermediate parameters of the second data set in the multi-source data set are not obtained by direct monitoring, the data in the second data set that seriously deviates from the preset actual value is eliminated; and the second data set is normalized to convert the second data set into a numerical value within [-1, 1] to obtain a third data set; wherein the second data set is deep learning data, including temperature and humidity sensitivity values, GIS data sensitivity, weather data sensitivity, and new energy index of the substation area.

[0095] The third data set is randomly divided into a training set and a test set at a ratio of 75% and 25%, and RBF (Radial Basis Function Network), LSTM (Long Short-Term Memory Network) and BP Neural Network (Backpropagation Neural Network) are trained;

[0096] Among them, the selection of the number of hidden layer nodes in the BP neural network is subject to the joint constraints of the input layer and the output layer. It is necessary to make a preliminary selection of the number of nodes according to the preset formula, and then through multiple experimental verifications, a reasonable value is taken around the number of nodes initially selected, so that the estimation accuracy of the BP neural network reaches the optimal level.

[0097] The normalization formula is:

[0098]

[0099] The preset formula is:

[0100]

[0101] Where, l is any sample data in the second data set; l min is the minimum value of the data in the second data set; l max is the maximum value of the data in the second data set; l i is the data in the third data set after normalization;

[0102] P represents the number of hidden layer nodes that need to be selected in the BP neural network; m and n represent the number of nodes in the output layer and the number of nodes in the input layer of the BP neural network respectively; a is an integer selected as needed, ranging from 1 to 10.

[0103] To apply the embodiments of the present application, please refer to Table 1, which is a neural network data training table provided in the embodiments of the present application, indicating the process of training RBF and LSTM.

[0104] Table 1 Neural network data training table

[0105]

[0106]

[0107] To apply the embodiment of the present application, please refer to Table 2, which is a BP neural network data training table provided in the embodiment of the present application, indicating the process of training the BP neural network.

[0108] Table 2 BP neural network data training table

[0109]

[0110] S3.3, if the discrete degree and data measurement error of the third data set are within the preset standard range, the error of the third data set is directly corrected by the RBF method to obtain the fourth data set;

[0111] If the degree of dispersion and data measurement error of the third data set are not within the standard range, the third data set is divided according to the mean difference and heterogeneity ratio of the data in the third data set with weights of 80% and 20% to obtain a discretized classification data set;

[0112] The LSTM is used to learn the long-term dependencies of discretized classification data sets and generate estimates about preset targets. The preset targets are usually classification labels, which indicate the categories to which the data points belong and can be set according to actual usage.

[0113] Through RBF training estimation, the error can be further reduced to obtain specific data information;

[0114] The estimated value is corrected based on the specific data information to obtain a fourth data set.

[0115] The fourth data set is converted into a label output using a BP neural network to obtain a deep label data set.

[0116] The second label dataset is composed of a statistical label dataset and a deep label dataset.

[0117] The calculation formula of the mean difference is:

[0118]

[0119] The calculation formula of the heterogeneity ratio is:

[0120]

[0121] The third data set is divided according to the weights of 80% and 20%, which can be expressed as:

[0122]

[0123] l d is the mean difference, l n is any data in the third data set, is the average value of all data in the third data set;

[0124] V r is the heterogeneity ratio, ∑f i is the total frequency of values ​​in the third data set, f m is the frequency of data in the third data set that meets the preset normal standard.

[0125] To apply the embodiments of this application, please refer to Figures 4 to 7 ;

[0126] Figure 4 The present invention provides a flowchart of label extraction, including "(a) overall flowchart" and "(b) label extraction method determination flowchart". (a) Overall flowchart represents the process of extracting labels from corresponding data according to the categories of directly extracted data, deep learning data, and statistical analysis data, thereby establishing a behavior profile of the substation area on this basis. (b) Label extraction method determination flowchart represents the process of extracting labels from corresponding data according to the categories of directly detectable and non-directly detectable, thereby establishing a behavior profile of the substation area on this basis.

[0127] Figure 5It is a block diagram of a deep learning label extraction method combining LSTM and RBF provided in an embodiment of the present application, which represents a deep learning label extraction method combining LSTM and RBF;

[0128] Figure 6 This is a model framework diagram of a deep learning label extraction method combining LSTM and RBF methods provided in an embodiment of the present application, indicating Figure 5 The model framework of deep learning label extraction method combining LSTM and RBF;

[0129] Figure 7 It is a flow chart of the deep learning extraction method combining LSTM and RBF methods provided in an embodiment of the present application, which represents the general process of generating a deep label data set based on a third data set by the deep learning extraction method combining LSTM and RBF methods; it also includes improving the deep label data set by selecting operations, crossover operations and updating fitness values.

[0130] In this embodiment S3.3, when the degree of dispersion of the third data set is within the preset standard range, it means that the overall volatility and variability of the data are within an acceptable range, but there may still be local errors or noise. Therefore, directly correcting the errors of the third data set can further reduce the impact of these errors on subsequent analysis and improve the accuracy and reliability of the data. When the degree of dispersion of the data set exceeds the standard range, it usually means that there are significant outliers or large variability in the data; directly processing such a data set may introduce unnecessary noise or errors, affecting the results of subsequent analysis, so by calculating and correcting the estimated value, this abnormal situation can be effectively dealt with, so that the data is more in line with the actual distribution and the needs of subsequent analysis;

[0131] In addition, the combination of LSTM and RBF not only takes advantage of the LSTM in capturing long-term dependencies, but also takes advantage of the RBF in nonlinear fitting and fast learning. This combination makes the entire processing flow more robust and reliable.

[0132] In addition, since neural networks can process data sets with low discretization levels faster and with accurate results (using only the RBF method), and process data sets with high discretization levels more slowly and with larger output errors (using the RBF+LSTM method), the division of discretization levels is crucial.

[0133] Overall, for data sets obtained through direct monitoring, this embodiment S3 generates portrait labels through statistical calculations, which can make full use of the original characteristics and distribution patterns of the data and reduce errors caused by human intervention or model complexity. For data sets obtained through indirect monitoring, generating labels through deep learning models can capture complex patterns and potential relationships in the data and improve the accuracy and generalization ability of labels.

[0134] S4. Perform clustering processing on the multi-source data set, the first label data set, and the second label data set to obtain a behavior portrait of the substation area.

[0135] Step S4 of the embodiment of the present application is specifically as follows:

[0136] Clustering is performed on the multi-source data set, the first label data set and the second label data set to obtain the behavior portrait of the substation area.

[0137] To apply the embodiments of this application, please refer to Figure 8 ; Figure 8 : is a schematic diagram of a method for profiling a substation behavior based on multi-source heterogeneous data provided by an embodiment of the present application, which shows the general process of performing data cleaning, label extraction and data analysis on a multi-source data set in this embodiment to establish a substation behavior profile, specifically:

[0138] Data cleaning: A complete data cleaning solution consists of all or part of the steps of data collection and processing, data classification, data screening, data completion and buffer storage. The cleaned data is passed to the label extraction and data analysis steps.

[0139] Label extraction: Generate image labels through deep learning, direct extraction, and statistical analysis, and pass them to the data analysis stage;

[0140] Data analysis: Analyze the data and labels, and form a behavioral portrait of the substation through clustering methods.

[0141] Overall, this embodiment has the following beneficial effects:

[0142] For data that can be directly monitored, this application directly and quickly generates portrait labels, which improves the efficiency of data processing and label generation. For data that cannot be directly monitored, portrait labels are generated through deep learning and statistical calculations; among them, deep learning can capture complex data relationships and improve the accuracy of estimation and correction, while statistical calculations provide a solid theoretical basis and enhance the reliability of labels; this method of processing data by category is conducive to flexible and accurate generation of labels. Clustering processing divides the behavior of the substation into different categories or groups, thereby more finely portraying the characteristics and patterns of the substation behavior. This refined portrait helps to more deeply understand the operating conditions and needs of the substation, thereby accurately portraying the behavior of the substation.

[0143] In addition, this application uses a multi-source data cleaning method to achieve automatic collection, completion, and screening of multi-source data. Based on the area behavior label generation method, the extraction method of different labels is determined. According to the normalized multi-source data processing method and the data classification method based on the average difference and heterogeneous ratio, the calculation speed and calculation accuracy of the neural network are improved. Therefore, this application solves the problem that traditional solutions are difficult to carry out area profiling.

[0144] Embodiment 2:

[0145] See also Fig. 9 , the embodiment of the present application provides a station area behavior profiling device based on multi-source heterogeneous data, including a data module 10, a first label module 20, a second label module 30 and a profiling module 40;

[0146] Wherein, the data module 10 is used to obtain a multi-source data set of a low-pressure area;

[0147] A first label module 20, used to generate a portrait label for the data that can be directly monitored in the multi-source data set to obtain a first label data set;

[0148] The second label module 30 is used to generate a portrait label for the data that cannot be directly monitored in the multi-source data set by means of deep learning and statistical calculation, so as to obtain a second label data set;

[0149] The portrait module 40 is used to perform clustering processing on the multi-source data set, the first label data set and the second label data set to obtain a station area behavior portrait.

[0150] In one embodiment, the data module 10 includes a data unit, a conversion unit, a classification unit, and a processing unit; wherein the data module is a process of acquiring a multi-source data set, and the conversion unit, the classification unit, and the processing unit are processes of performing data cleaning, specifically:

[0151] The data unit is used to obtain a multi-source data set of a low-voltage area; wherein the multi-source data set includes dispatching data, marketing data, metering data, production data and geographic data of the low-voltage area.

[0152] A conversion unit is used to unify multi-source data sets into IoT data or collect them into a data warehouse through the MQTT protocol (Message Queuing Telemetry Transport) and other methods;

[0153] A classification unit is used to classify multi-source data sets according to data forms to obtain transitional data sets; wherein the data forms include structured data, unstructured data, semi-structured data, real-time data, and offline data;

[0154] A processing unit is used to screen and delete invalid data, abnormal data and non-standard data in the transition data set based on statistics, machine learning and other methods, and to complete the missing data in the transition data set by means of difference completion, model completion, fitting completion and mode completion, so as to obtain a multi-source data set after fine processing;

[0155] The processing unit is also used to buffer and store multi-source data sets in the form of archival data, historical data and real-time data, thereby establishing a heterogeneous database.

[0156] To apply the embodiments of this application, please refer to Figure 2 , Figure 2 It is a schematic diagram of the data cleaning step method provided in an embodiment of the present application, which shows the process of data cleaning of a multi-source data set by a conversion unit, a classification unit and a processing unit in the embodiment.

[0157] In this embodiment, the conversion unit, classification unit and processing unit classify the multi-source data set according to the structural category and update form, so that the organizational structure of the data set is clearer. Since invalid data may mislead the analysis results or reduce the analysis accuracy, deleting invalid data can reduce the noise and redundancy in the data set and improve the overall quality of the data; and completing these missing data can make the multi-source data set after fine processing more complete and reliable.

[0158] In one embodiment, the first label module 20 is specifically:

[0159] Generate portrait labels for the data that can be directly monitored in the multi-source data set to obtain a first label data set; wherein the data that can be directly monitored in the multi-source data set includes data that can be directly obtained, such as power supply radius, transformer capacity, power supply line diameter, and number of power supply households.

[0160] In one embodiment, the second label module 30 includes a statistical unit, a normalization subunit, a correction subunit, an estimation subunit, a conversion subunit and a construction unit; wherein the statistical unit is a process of generating a statistical label data set according to statistical analysis data, the normalization subunit is a process of processing deep learning data to generate a third data set, and the correction subunit, the estimation subunit, the conversion subunit and the construction unit are processes of generating a deep label data set according to the third data set and obtaining the final second label data set, specifically:

[0161] A statistical unit is used for standardizing and per-unitizing the data in the multi-source data set that cannot be directly monitored, if the intermediate parameters of the first data set in the multi-source data set are directly monitored, to obtain a unified data set; wherein the first data set is statistical analysis data, including average load rate, power stability, power fluctuation rate and average load rate, etc.;

[0162] The statistical unit is also used to perform data frequency analysis on the unified data set according to the frequency of data deviation from the estimated value in unit time, and obtain the distribution of the data, including the central tendency of the data, the degree of dispersion, and whether there are outliers.

[0163] The statistical unit is also used to combine the distribution of data and use mathematical formulas, modeling and other methods to perform correlation analysis on the unified data set, so as to identify potential patterns in the data and establish correlation relationships between data labels to obtain a statistical label data set;

[0164] The statistical unit is also used to display the results of data frequency analysis and correlation analysis in an intuitive way such as visualization.

[0165] To apply the embodiments of this application, please refer to Figure 3 , Figure 3 It is a block diagram of the statistical analysis method provided in an embodiment of the present application, which represents the general process of generating a statistical label data set based on a first data set (statistical analysis data).

[0166] The normalization subunit is used for the data that cannot be directly monitored in the multi-source data set. If the intermediate parameters of the second data set in the multi-source data set are not obtained by direct monitoring, the data in the second data set that seriously deviates from the preset actual value will be eliminated; and the second data set is normalized to convert the second data set into a numerical value within [-1, 1] to obtain a third data set; wherein the second data set is deep learning data, including temperature and humidity sensitivity values, GIS data sensitivity, weather data sensitivity and new energy index of the substation area.

[0167] The normalization subunit is also used to randomly divide the third data set into a training set and a test set at a ratio of 75% and 25%, and train RBF (Radial Basis Function Network), LSTM (Long Short-Term Memory Network) and BP Neural Network (Backpropagation Neural Network);

[0168] Among them, the selection of the number of hidden layer nodes in the BP neural network is subject to the joint constraints of the input layer and the output layer. It is necessary to make a preliminary selection of the number of nodes according to the preset formula, and then through multiple experimental verifications, a reasonable value is taken around the number of nodes initially selected, so that the estimation accuracy of the BP neural network reaches the optimal level.

[0169] The normalization formula is:

[0170]

[0171] The preset formula is:

[0172]

[0173] Where, l is any sample data in the second data set; l min is the minimum value of the data in the second data set; l max is the maximum value of the data in the second data set; l i is the data in the third data set after normalization;

[0174] P represents the number of hidden layer nodes that need to be selected in the BP neural network; m and n represent the number of nodes in the output layer and the number of nodes in the input layer of the BP neural network respectively; a is an integer selected as needed, ranging from 1 to 10.

[0175] To apply the embodiments of the present application, please refer to Table 1, which is a neural network data training table provided in the embodiments of the present application, indicating the process of training RBF and LSTM.

[0176] Table 1 Neural network data training table

[0177]

[0178]

[0179] To apply the embodiment of the present application, please refer to Table 2, which is a BP neural network data training table provided in the embodiment of the present application, indicating the process of training the BP neural network.

[0180] Table 2 BP neural network data training table

[0181]

[0182] A correction subunit, configured to directly correct the error of the third data set by using an RBF method to obtain a fourth data set if the discrete degree and the data measurement error of the third data set are within a preset standard range;

[0183] an estimation subunit, for dividing the third data set into data according to the weights of 80% and 20% according to the mean difference and heterogeneity ratio of the data in the third data set to obtain a discretized classification data set if the degree of dispersion and the data measurement error of the third data set are not within the standard range;

[0184] The estimation subunit is also used to learn the long-term dependencies of the discretized classification data set through LSTM to generate an estimated value about a preset target; wherein the preset target is usually a classification label, indicating the category to which the data point belongs, and can be set according to the actual usage;

[0185] The estimation subunit is also used to further reduce the error and obtain specific data information through RBF training estimation value;

[0186] The estimation subunit is further used to correct the estimation value based on the specific data information to obtain a fourth data set.

[0187] The conversion subunit is used to convert the fourth data set into a label output by using a BP neural network to obtain a deep label data set.

[0188] A construction unit is used to construct a second label dataset from a statistical label dataset and a deep label dataset.

[0189] The calculation formula of the mean difference is:

[0190]

[0191] The calculation formula of the heterogeneity ratio is:

[0192]

[0193] The third data set is divided according to the weights of 80% and 20%, which can be expressed as:

[0194]

[0195] l d is the mean difference, l n is any data in the third data set, is the average value of all data in the third data set;

[0196] V r is the heterogeneity ratio, ∑f i is the total frequency of values ​​in the third data set, f mis the frequency of data in the third data set that meets the preset normal standard.

[0197] To apply the embodiments of this application, please refer to Figures 4 to 7 ;

[0198] Figure 4 The present invention provides a flowchart of label extraction, including "(a) overall flowchart" and "(b) label extraction method determination flowchart". (a) Overall flowchart represents the process of extracting labels from corresponding data according to the categories of directly extracted data, deep learning data, and statistical analysis data, thereby establishing a behavior profile of the substation area on this basis. (b) Label extraction method determination flowchart represents the process of extracting labels from corresponding data according to the categories of directly detectable and non-directly detectable, thereby establishing a behavior profile of the substation area on this basis.

[0199] Figure 5 It is a block diagram of a deep learning label extraction method combining LSTM and RBF provided in an embodiment of the present application, which represents a deep learning label extraction method combining LSTM and RBF;

[0200] Figure 6 This is a model framework diagram of a deep learning label extraction method combining LSTM and RBF methods provided in an embodiment of the present application, indicating Figure 5 The model framework of deep learning label extraction method combining LSTM and RBF;

[0201] Figure 7 It is a flow chart of the deep learning extraction method combining LSTM and RBF methods provided in an embodiment of the present application, which represents the general process of generating a deep label data set based on a third data set by the deep learning extraction method combining LSTM and RBF methods; it also includes improving the deep label data set by selecting operations, crossover operations and updating fitness values.

[0202] In the correction subunit, estimation subunit, conversion subunit and construction unit of this embodiment, when the degree of discreteness of the third data set is within the preset standard range, it means that the overall volatility and variability of the data are within an acceptable range, but there may still be local errors or noise. Therefore, directly correcting the errors of the third data set can further reduce the impact of these errors on subsequent analysis and improve the accuracy and reliability of the data. When the degree of discreteness of the data set exceeds the standard range, it usually means that there are significant outliers or large variability in the data; directly processing such a data set may introduce unnecessary noise or errors, affecting the results of subsequent analysis. Therefore, by calculating and correcting the estimated values, this abnormal situation can be effectively dealt with, making the data more in line with the actual distribution and the needs of subsequent analysis;

[0203] In addition, the combination of LSTM and RBF not only takes advantage of the LSTM in capturing long-term dependencies, but also takes advantage of the RBF in nonlinear fitting and fast learning. This combination makes the entire processing flow more robust and reliable.

[0204] In addition, since neural networks can process data sets with low discretization levels faster and with accurate results (using only the RBF method), and process data sets with high discretization levels more slowly and with larger output errors (using the RBF+LSTM method), the division of discretization levels is crucial.

[0205] Overall, the second label module 30 of this embodiment generates portrait labels through statistical calculation for the data set obtained by direct monitoring, which can make full use of the original characteristics and distribution rules of the data and reduce errors caused by human intervention or model complexity. For the data set obtained by indirect monitoring, the deep learning model is used to generate labels, which can capture the complex patterns and potential relationships in the data and improve the accuracy and generalization ability of the labels.

[0206] In one embodiment, the portrait module 40 is specifically:

[0207] Clustering is performed on the multi-source data set, the first label data set and the second label data set to obtain the behavior portrait of the substation area.

[0208] To apply the embodiments of this application, please refer to Figure 8 ; Figure 8 : is a schematic diagram of a method for profiling a substation behavior based on multi-source heterogeneous data provided in an embodiment of the present application, which shows the general process of performing data cleaning, label extraction and data analysis on a multi-source data set in the second embodiment to establish a substation behavior profile, specifically:

[0209] Data cleaning: A complete data cleaning solution consists of all or part of the steps of data collection and processing, data classification, data screening, data completion and buffer storage. The cleaned data is passed to the label extraction and data analysis steps.

[0210] Label extraction: Generate image labels through deep learning, direct extraction, and statistical analysis, and pass them to the data analysis stage;

[0211] Data analysis: Analyze the data and labels, and form a behavioral portrait of the substation through clustering methods.

[0212] Overall, this embodiment has the following beneficial effects:

[0213] For data that can be directly monitored, this application directly and quickly generates portrait labels, which improves the efficiency of data processing and label generation. For data that cannot be directly monitored, portrait labels are generated through deep learning and statistical calculations; among them, deep learning can capture complex data relationships and improve the accuracy of estimation and correction, while statistical calculations provide a solid theoretical basis and enhance the reliability of labels; this method of processing data by category is conducive to flexible and accurate generation of labels. Clustering processing divides the behavior of the substation into different categories or groups, thereby more finely portraying the characteristics and patterns of the substation behavior. This refined portrait helps to more deeply understand the operating conditions and needs of the substation, thereby accurately portraying the behavior of the substation.

[0214] In addition, this application uses a multi-source data cleaning method to achieve automatic collection, completion, and screening of multi-source data. Based on the area behavior label generation method, the extraction method of different labels is determined. According to the normalized multi-source data processing method and the data classification method based on the average difference and heterogeneous ratio, the calculation speed and calculation accuracy of the neural network are improved. Therefore, this application solves the problem that traditional solutions are difficult to carry out area profiling.

[0215] Embodiment three:

[0216] The embodiment of the present application provides a computer-readable storage medium, the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the method for profiling the behavior of a substation based on multi-source heterogeneous data;

[0217] Wherein, the method for profiling the behavior of a substation based on multi-source heterogeneous data, if implemented in the form of a software functional unit and used as an independent product, can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0218] The above are preferred embodiments of the present invention. It should be noted that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for profiling substation behavior based on multi-source heterogeneous data, characterized in that: include: Obtain multi-source datasets of low-pressure areas; Generate a portrait label for the data that can be directly monitored in the multi-source data set to obtain a first label data set; Generate a portrait label for the data that cannot be directly monitored in the multi-source data set by means of deep learning and statistical calculation to obtain a second label data set; Clustering is performed on the multi-source data set, the first label data set, and the second label data set to obtain a substation behavior portrait.

2. A method for profiling substation behavior based on multi-source heterogeneous data as claimed in claim 1, characterized in that: By means of deep learning and statistical calculation, a portrait label is generated for the data that cannot be directly monitored in the multi-source data set, and a second label data set is obtained, which is specifically: If the intermediate parameters of the first data set in the multi-source data set are obtained by direct monitoring, generating a portrait label for the first data set by statistical calculation to obtain a statistical label data set; If the intermediate parameters of the second data set in the multi-source data set are not obtained by direct monitoring, generating a portrait label for the second data set by deep learning to obtain a deep label data set; The second label dataset is composed of the statistical label dataset and the deep label dataset.

3. A method for profiling substation behavior based on multi-source heterogeneous data as claimed in claim 2, characterized in that: Generate portrait labels for the second data set by deep learning to obtain a deep label data set, specifically: Normalizing the second data set to obtain a third data set; If the degree of dispersion of the third data set is within a preset standard range, correcting the error of the third data set to obtain a fourth data set; If the degree of discreteness of the third data set is not within the standard range, correcting the estimated value of the third data set by discretization classification to obtain a fourth data set; The fourth data set is converted into a label output to obtain the deep label data set.

4. A method for profiling substation behavior based on multi-source heterogeneous data as claimed in claim 3, characterized in that: The estimated value of the third data set is corrected by discretization classification to obtain a fourth data set, which is specifically: According to the average difference and heterogeneity ratio of the data in the third data set, the third data set is divided into data to obtain a discretized classification data set; Learning the long-term dependencies of the discretized classification data set through a long short-term memory network to generate an estimated value about a preset target; Training the estimated value through a radial basis function network to obtain specific data information; The estimated value is corrected based on the specific data information to obtain the fourth data set.

5. A method for profiling substation behavior based on multi-source heterogeneous data as claimed in claim 4, characterized in that: The calculation formula of the heterogeneity ratio is specifically: Among them, Σf i is the total frequency of the values ​​in the third data set, f m is the frequency of data in the third data set that meets the preset normal standard.

6. The method for profiling substation behavior based on multi-source heterogeneous data according to claim 1, characterized in that: After obtaining the multi-source data set of the low-voltage area, the method further includes: Classifying the multi-source data sets according to structural categories and update forms to obtain transition data sets; Invalid data in the transition data set is deleted, and missing data in the transition data set is completed to obtain a multi-source data set after refinement processing.

7. A station area behavior profiling device based on multi-source heterogeneous data, characterized in that: It includes a data module, a first label module, a second label module and a portrait module; Wherein, the data module is used to obtain multi-source data sets of low-voltage areas; The first label module is used to generate a portrait label for the data that can be directly monitored in the multi-source data set to obtain a first label data set; The second label module is used to generate a portrait label for the data that cannot be directly monitored in the multi-source data set by means of deep learning and statistical calculation, so as to obtain a second label data set; The portrait module is used to perform clustering processing on the multi-source data set, the first label data set and the second label data set to obtain a station area behavior portrait.

8. The device for profiling substation behavior based on multi-source heterogeneous data as claimed in claim 7, characterized in that: The second label module includes a statistical unit, a depth unit and a construction unit; The statistical unit is used to generate a portrait label for the first data set in the multi-source data set by statistical calculation to obtain a statistical label data set if the intermediate parameter of the first data set in the multi-source data set is obtained by direct monitoring; The depth unit is used to generate a portrait label for the second data set in the multi-source data set by deep learning to obtain a deep label data set if the intermediate parameters of the second data set in the multi-source data set are not obtained by direct monitoring; The construction unit is used to construct the second label data set from the statistical label data set and the deep label data set.

9. The device for profiling substation behavior based on multi-source heterogeneous data as claimed in claim 8, characterized in that: The depth unit includes a normalization subunit, a correction subunit, an estimation subunit and a conversion subunit; The normalization subunit is used to perform normalization processing on the second data set to obtain a third data set; The correction subunit is used to correct the error of the third data set to obtain a fourth data set if the discrete degree of the third data set is within a preset standard range; The estimation subunit is configured to correct the estimated value of the third data set by discretization classification to obtain a fourth data set if the discreteness of the third data set is not within the standard range; The conversion subunit is used to convert the fourth data set into a label output to obtain the deep label data set.

10. A storage medium, characterized in that: The storage medium stores a computer program, which is called and executed by a computer to implement a method for profiling substation behavior based on multi-source heterogeneous data as described in any one of claims 1 to 6 above.

Images