Waveform data labeling method under typical scene of user demand side

Abstract

The invention relates to a waveform data labeling method in a user demand side typical scene, which comprises the following steps of: 1, acquiring original data: acquiring power consumption data of three types of typical users to obtain power consumption information of the users; 2, data preprocessing; 3, data generated by the users are mined and analyzed, operation rules of the users during electricity utilization are found out through data mining and analysis, and user behavior models of the electricity household users are obtained; 4, constructing a model label: performing labeling processing on the features of the user according to the user model obtained through analysis; 5, performing correlation analysis according to the harmonic data: performing correlation analysis by using the harmonic data of the user and the current data, and perfecting the portrait of the household power user; step 6, constructing a power consumer portrait based on the user power consumption characteristics and the harmonic characteristics; the method has the advantages of fusing the power utilization and harmonic characteristics of the user, constructing the user portrait, improving the fault judgment precision, and improving the stability and reliability of the power grid.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grids, and in particular relates to a waveform data labeling method in a typical scenario on a user demand side. Background Art

[0002] As the State Grid Corporation of China continues to promote the construction of smart grids, the emergence of unconventional grid operating environments such as special power sources or distributed power sources, as well as the application of a large number of power electronic equipment, have brought various impacts on the metering performance of electric energy metering devices, which in turn affects the stability of safe production; at the same time, with the continuous growth of modern society's demand for electricity and the increasing complexity of the power system, how to effectively manage and optimize power resources has become a problem that continues to be solved; the diversity and uncertainty of users' electricity consumption behaviors make it more difficult to rationally allocate power resources and improve energy utilization efficiency; therefore, accurately describing the electricity consumption behaviors and characteristics of electricity users, and then realizing personalized power management and optimization, is the key to improving power system efficiency and user satisfaction; traditional The construction method of power user portrait usually focuses on the basic power consumption characteristics of users, such as load rate, peak-to-valley difference rate, etc., but when facing harmonic problems, the impact of harmonic characteristics on power consumption behavior is often ignored. Harmonics, as an important signal component in the power system, have a profound impact on the stability and energy efficiency of the power system. Harmonic characteristics include harmonic content, harmonic distribution, and the impact of each harmonic on power. These characteristics can provide additional insights into user power consumption behavior and help to have a more comprehensive understanding of user power usage. Therefore, it is very necessary to provide a waveform data annotation method for typical scenarios on the user demand side that integrates user power consumption and harmonic characteristics, constructs user portraits, improves fault judgment accuracy, and improves power grid stability and reliability. Summary of the invention

[0003] 1. Technical issues

[0004] In view of the above-mentioned existing technical status, this application mainly addresses the following technical problems:

[0005] 1. The diversity and uncertainty of users’ electricity consumption behavior make it more difficult to rationally allocate power resources and improve energy efficiency;

[0006] 2. How to accurately describe the electricity consumption behavior and characteristics of power users, realize personalized power management and optimization, and improve power system efficiency and user satisfaction;

[0007] 3. The current complex operating conditions caused by unconventional power grid operating environments such as special power sources or distributed power sources access and the application of a large number of power electronic equipment have various impacts on the metering performance of electric energy metering devices, resulting in unstable production safety. Traditional power user portraits are limited to focusing on the basic power consumption characteristics of users and will ignore the impact of harmonic characteristics on power quality.

[0008] (II) Technical solution

[0009] The purpose of the present invention is to overcome the shortcomings of the prior art and to provide a waveform data labeling method in typical scenarios on the user demand side that integrates user electricity consumption and harmonic characteristics, builds user portraits, improves fault judgment accuracy, and improves grid stability and reliability.

[0010] The object of the present invention is achieved by providing a waveform data annotation method in a typical scenario on the user demand side, the method comprising the following steps:

[0011] Step 1: Acquisition of original data: Collect electricity consumption data of three typical users to obtain electricity consumption information of users;

[0012] Step 2: Data preprocessing: Filter and clean the original complex data information;

[0013] Step 3: Mining and analyzing the data generated by users: Through data mining and analysis, find out the operating rules of users when using electricity and obtain the user behavior model of household electricity users;

[0014] Step 4: Build model labels: Label the user's features based on the user model obtained through analysis;

[0015] Step 5: Correlation analysis based on harmonic data: Use the user's harmonic data and current data for correlation analysis to improve the portrait of household electricity users;

[0016] Step 6: Build a power user profile based on the user's power consumption characteristics and harmonic characteristics.

[0017] Furthermore, the data acquisition in step 1 is based on the data middle platform to obtain archival data, operating data, event data, and auxiliary analysis data in the marketing business system and the electricity consumption information collection system; wherein ① the archival data includes: electricity meter information, user information, and metering point information; ② the operating data includes: daily forward and reverse electricity indication curve of the measuring point, daily active power curve, daily reactive power curve, daily power factor curve, daily voltage curve, and daily current curve; ③ the event data includes: electricity meter flow reverse event, electricity meter power reverse event; ④ the auxiliary analysis data includes: phase-by-phase forward and reverse electricity indication data, and phase angle data.

[0018] Furthermore, the data preprocessing in step 2 has the following specific processing rules: ① filter user types; ② eliminate null values; ③ eliminate reverse settlement users; ④ eliminate photovoltaic users; ⑤ eliminate no-load; ⑥ eliminate abnormal data such as voltage loss and phase failure; ⑦ eliminate incomplete daily data of voltage, current, power and power factor.

[0019] Furthermore, the step 6 of constructing a power user portrait based on the user's power consumption characteristics and harmonic characteristics includes the following steps:

[0020] Step 6.1: First, use the random forest and Kendall correlation coefficient methods to analyze the influence of user electricity consumption characteristics on electricity consumption behavior and the redundancy between characteristics;

[0021] Step 6.2: Construct a feature fitness evaluation index and use the sparrow search algorithm for iterative optimization to screen out the most representative user electricity consumption features;

[0022] Step 6.3: Secondly, through the correlation analysis of load data and harmonic data, the relationship between harmonic content and load data in each scenario is explored to reveal the performance of harmonic characteristics in different power consumption scenarios;

[0023] Step 6.4: Finally, combine the user's electricity consumption characteristics with the harmonic characteristics to build a comprehensive power user portrait;

[0024] Step 6.5: Verify the effectiveness through case analysis, which provides new perspectives and methods for the optimization and management of power systems.

[0025] Furthermore, the random forest algorithm in step 6.1 is specifically as follows: by constructing multiple decision trees and integrating their results, the mathematical description of the decision tree is as follows: for a sample set D containing n categories Hi (i=1, 2, ..., n), in, Where I is the expected entropy of the sample; d i is the number of samples belonging to category Hi; d is the total number of samples; E(A) is the expected entropy of a single feature A of the sample; k is the total number of features in the sample set; d ij is the jth feature of the sample belonging to category i.

[0026] Furthermore, the Kendall correlation coefficient method in step 6.1 is specifically as follows: for any features A1, A2, the calculation method of the Kendall correlation coefficient is as follows: Among them, B and C are sample pairs with consistent and inconsistent rankings respectively; T1 and T2 are the number of samples that appear the same in the sample sets of features A1 and A2 respectively.

[0027] Furthermore, the feature fitness evaluation index in step 6.2 is constructed as follows: for any feature set a∈A, its evaluation function P(a) is: P(a)=R(a)-K(a)(9), where: Where R(a) and K(a) are the importance value and redundancy of the feature set a respectively; N a is the number of features contained in feature set a; m1 and m2 are two different features in feature set a.

[0028] Furthermore, the sparrow search algorithm in step 6.2 is iteratively optimized, specifically: the sparrow population is divided into finders, followers and scouts, and the position of the finder is updated during the predation process as follows: in, is the position of the i-th sparrow in the t-th iteration and the j-th dimension; α is a random number between (0, 1]; R2∈[0,1] and ST∈[0.5,1] are the warning value and the safety value respectively; Q is a random number that follows a normal distribution; L is a 1×d matrix.

[0029] Furthermore, the position of the follower is updated as follows: Among them, x p With x worst are the best position and the global worst position of the discoverer respectively; n is the population size; A is a 1×d matrix;

[0030] The positions of the scout positions are updated as follows: Among them, x best is the global optimal position; β is a normally distributed random number with a mean of 0 and a variance of 1; k is a random number between [-1,1]; f i is the fitness value of the current sparrow; f g With f w are the current global optimal and worst fitness values ​​respectively; ε is the minimum constant.

[0031] A recorder with clock synchronization constructs a user portrait combining power consumption characteristics and harmonic characteristics based on the waveform data annotation method in the typical scenario on the user demand side as described above. The recorder includes a signal processing module, a power consumption characteristic screening module and a recording data processing and analysis module; wherein the signal processing module is used to analyze the recording data for testing, generate appropriate debugging data, and feed back to the recorder debugging personnel, so as to adjust the function parameters of the recorder; the power consumption characteristic screening module is used to analyze the user-side power consumption behavior data using the load data of the residential area, and screen the user's power consumption characteristics through machine learning and statistical methods; the recording data processing and analysis module is used to process the collected recording data, perform correlation analysis in combination with the power consumption behavior data, and extract harmonic characteristics related to the user's power consumption behavior.

[0032] (III) Beneficial effects

[0033] 1. The present invention processes and analyzes the recording data to realize hardware joint debugging of new recording equipment, which helps to timely discover and correct problems of the recording equipment during the research and development and debugging stage, and ensure the normal operation of the subsequent recording process;

[0034] 2. The present invention constructs a user portrait method that integrates user power consumption characteristics and harmonic characteristics by performing feature analysis on the recorded data of users in typical scenarios collected by the recorder and combining it with the user-side power load data of the communication data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Flow chart of the random forest algorithm of the present invention.

[0036] Figure 2 It is a schematic diagram of the power curve of the newspaper building of the present invention.

[0037] Figure 3 It is a schematic diagram of the power curve of the university science and technology park of the present invention.

[0038] Figure 4 It is a schematic diagram of the power curve of Yujin Food City of the present invention.

[0039] Figure 5 It is a schematic diagram of the power curve of the Du Kang Hotel of the present invention.

[0040] Figure 6 It is a schematic diagram of the power curve of the Zhengzhou City Urban Lighting Management Office on Zhongyuan Road of the present invention.

[0041] Figure 7 It is a schematic diagram of the power curve of Zhengzhou City Lighting Management Office of the present invention.

[0042] Figure 8 It is a schematic diagram of the importance of various electricity consumption characteristics of the present invention.

[0043] Fig. 9 It is a process diagram of the SSA iterative solution of the optimal feature set of the present invention.

[0044] Fig.10 A user portrait schematic diagram of the present invention. DETAILED DESCRIPTION

[0045] The present invention is further described below in conjunction with embodiments and / or drawings.

[0046] Example 1

[0047] like Figure 1-10As shown, a waveform data annotation method in a typical scenario on the user demand side includes the following steps:

[0048] Step 1: Acquisition of original data: Collect electricity consumption data of three typical users to obtain electricity consumption information of users;

[0049] In the present invention, specifically: based on the data middle platform, the archival data, operation data, event data and auxiliary analysis data in the marketing business system and the electricity consumption information collection system are obtained; wherein ① the archival data includes the electric energy meter information, the user information and the metering point information; ② the operation data includes the daily forward and reverse electric energy indication curve of the measuring point, the daily active power curve, the daily reactive power curve, the daily power factor curve, the daily voltage curve and the daily current curve; ③ the event data includes the electric energy meter flow reverse event and the electric energy meter power reverse event; ④ the auxiliary analysis data includes the phase-by-phase forward and reverse electric energy indication data and the phase angle data.

[0050] The data sources are shown in the following table.

[0051]

[0052] Data archiving: Because the data in the data center can only be retained for one month, when accessing long-term data, the center must aggregate and seal it into a temporary table, mainly including the user load curve data table, voltage curve data table, and current curve data table.

[0053] The archiving rules are as follows:

[0054] 1. The data of the previous month shall be sealed at the beginning of each month;

[0055] 2. The data is only stored for electricity users in typical scenarios;

[0056] 3. The archived data only contains the user's load curve data;

[0057] 4. Aggregate the sealed data in each table using the metering point number as the primary key.

[0058] The aggregated data facilitates subsequent testing and adjustment of models in other analytical work.

[0059] Step 2: Data preprocessing: Filter and clean the original complex data information, remove useless information, and lay the foundation for subsequent data mining;

[0060] In the present invention, the main cleaning rules are as follows:

[0061] 1. Filter user types: During the data analysis period, only users whose electricity metering points are used for settlement on the electricity sales side and whose metering points are classified as electricity customers are retained for incorrect wiring metering anomalies.

[0062] 2. Eliminate null values: remove the time points where the collected result data is null.

[0063] 3. Eliminate reverse settlement users: By screening whether the indication type in the user's meter reading power information is active reverse total users, users whose settlement method is reverse settlement during the data analysis period are removed, thereby reducing interference with the calculation of user electricity consumption characteristics.

[0064] 4. Eliminate photovoltaic users (gc_flag is 1, there is no gc_flag field in the archive table c_cons, and there is an fc_gc power generation user table).

[0065] 5. Eliminate users with no load (current value and voltage value are zero at 96 points), low power (less than 10% of rated current), and low collection success rate (less than 80%).

[0066] 6. Eliminate abnormal data such as pressure loss and phase failure.

[0067] 7. Eliminate incomplete daily data of voltage, current, power and power factor.

[0068] Step 3: Mining and analyzing the data generated by users: Through data mining and analysis, find out the operating rules of users when using electricity and obtain the user behavior model of household electricity users;

[0069] Step 4: Build model labels: Label the user's features based on the user model obtained through analysis;

[0070] Step 5: Correlation analysis based on harmonic data: Use the user's harmonic data and current data for correlation analysis to improve the portrait of household electricity users;

[0071] In the present invention, if the original data contains a large amount of text information, a text mining algorithm is used to process the data to obtain valuable keywords and then label the user appropriately; if in the e-commerce field, it is necessary to mine the information based on the user's search for goods, use the corresponding recommendation algorithm, mine and analyze the information related to the user, and recommend the information to the user in a timely manner to ensure the company's precise marketing.

[0072] Step 6: Build a power user profile based on the user's power consumption characteristics and harmonic characteristics.

[0073] The present invention is a waveform data annotation method under typical scenarios on the user demand side. During use, the present invention processes and analyzes the recorded data to realize hardware joint debugging of new recording equipment, which helps the recording equipment to promptly discover and correct problems during the research and development and debugging stages, and ensure normal operation during the subsequent recording process; the present invention constructs a user portrait method that integrates user power consumption characteristics and harmonic characteristics by performing feature analysis on the recorded data of users in typical scenarios collected by the recorder, combined with the power load data on the user side of the communication data center; the present invention has the advantages of integrating user power consumption and harmonic characteristics, constructing user portraits, improving fault judgment accuracy, and improving power grid stability and reliability.

[0074] Example 2

[0075] like Figure 1-10 As shown, a waveform data annotation method in a typical scenario on the user demand side includes the following steps:

[0076] Step 1: Acquisition of original data: Collect electricity consumption data of three typical users to obtain electricity consumption information of users;

[0077] Step 2: Data preprocessing: Filter and clean the original complex data information, remove useless information, and lay the foundation for subsequent data mining;

[0078] Step 3: Mining and analyzing the data generated by users: Through data mining and analysis, find out the operating rules of users when using electricity and obtain the user behavior model of household electricity users;

[0079] Step 4: Build model labels: Label the user's features based on the user model obtained through analysis;

[0080] Step 5: Correlation analysis based on harmonic data: Use the user's harmonic data and current data for correlation analysis to improve the portrait of household electricity users;

[0081] Step 6: Build a power user profile based on the user's power consumption characteristics and harmonic characteristics.

[0082] In the present invention, the method used in the present invention to construct a power user portrait based on the user's power consumption characteristics and harmonic characteristics is as follows:

[0083] First, we use the random forest and Kendall correlation coefficient methods to analyze the influence of user electricity consumption characteristics on electricity consumption behavior and the redundancy between characteristics, construct a characteristic fitness evaluation index, and use the sparrow search algorithm for iterative optimization to screen out the most representative user electricity consumption characteristics;

[0084] Specifically: ① Random forest is an algorithm based on ensemble learning. By constructing multiple decision trees and integrating their results, the algorithm can effectively improve the prediction accuracy and stability of the model. The flowchart of the random forest algorithm is as follows: Figure 1 As shown:

[0085] The mathematical description of the decision tree is as follows: For a sample set D containing n categories Hi (i = 1, 2, ..., n), in, Where I is the expected entropy of the sample; d i is the number of samples belonging to category Hi; d is the total number of samples; E(A) is the expected entropy of a single feature A of the sample; k is the total number of features in the sample set; d ij is the jth feature of the sample belonging to category i.

[0086] The entropy gain and entropy gain rate of feature A are calculated as: Gain(A)=I(d1,d2,...,d n )—E(A)(4), in, Wherein, Gain(A) and Gain′(A) are the entropy gain and entropy gain rate of feature A respectively.

[0087] Random forest is not only suitable for classification and prediction tasks, but also can effectively evaluate the influence of each feature on the target variable. The importance of feature A P(A) can be calculated by the following formula: Where T is the number of decision trees in the random forest; is the contribution of feature A to reducing information uncertainty in the tth decision tree.

[0088] ② In the power user feature set, there is often a certain correlation between feature individuals. Using only the random forest algorithm to calculate the influence of features will ignore the redundancy problem caused by the correlation. Therefore, the present invention uses the Kendall correlation coefficient to calculate the redundancy between features. The Kendall correlation coefficient mainly measures the correlation between variables by comparing the sorting consistency between sample pairs, that is, how many pairs of possible sample pairs have the same sorting; the value range of the correlation coefficient is between -1 and 1, where 1 represents a completely consistent sorting, -1 represents a completely inconsistent sorting, and 0 represents no correlation; the advantage of the Kendall correlation coefficient is its robustness to outliers and its relatively loose requirements on data distribution. It is not as strict as the Pearson correlation coefficient method and is suitable for variables of various data types and distribution forms;

[0089] For any feature A1, A2, the calculation method of Kendall correlation coefficient is as follows: Among them, B and C are sample pairs with consistent and inconsistent rankings respectively; T1 and T2 are the number of samples that appear the same in the sample sets of features A1 and A2 respectively.

[0090] ③ Taking into account the influence of features calculated by random forest and the redundancy between features, the present invention constructs an adaptive evaluation function. For any feature set a∈A, its evaluation function P(a) is: P(a)=R(a)-K(a)(9), where Where R(a) and K(a) are the importance value and redundancy of the feature set a respectively; N a is the number of features contained in feature set a; m1 and m2 are two different features in feature set a.

[0091] ④ The SSA algorithm uses the sparrow's predation and anti-predation behavior as inspiration to move the population's position to achieve the purpose of searching for the optimal solution. The sparrow population is divided into discoverers, followers, and scouts. The discoverer's position is updated during the predation process as follows: in, is the position of the i-th sparrow in the t-th iteration and the j-th dimension; α is a random number between (0, 1]; R2∈[0,1] and ST∈[0.5,1] are the warning value and the safety value respectively; Q is a random number that follows a normal distribution; L is a 1×d matrix, and each element is 1.

[0092] The follower's position is updated as follows: Among them, x p With x worst are the best position and the global worst position of the discoverer respectively; n is the population size; A is a 1×d matrix, each element is 1 or -1, and A + =A T (AA T ) -1 .

[0093] The positions of the scout positions are updated as follows: Among them, x best is the global optimal position; β is a normally distributed random number with a mean of 0 and a variance of 1, which is used to control the step size; k is a random number between [-1,1]; f i is the fitness value of the current sparrow; f g With f w are the current global optimal and worst fitness values ​​respectively; ε is the minimum constant used to avoid the denominator being 0.

[0094] Second, through the correlation analysis of load data and harmonic data, the relationship between harmonic content and load data in each scenario is explored to reveal the performance of harmonic characteristics in different power consumption scenarios; finally, the user's power consumption characteristics are combined with harmonic characteristics to build a comprehensive power user portrait;

[0095] 3. The effectiveness of this method is verified through case analysis, which provides a new perspective and method for the optimization and management of power systems.

[0096] Specifically: the data required for the power user portrait comes from the recorded data of the data center and the new recorder. Therefore, it is necessary to first perform hardware coordination on the new recorder to ensure the accuracy of the recorded data, then perform harmonic data correlation analysis, construct harmonic related features, and finally construct the user portrait in combination with the power user's electricity consumption data.

[0097] ① Hardware joint debugging: The recording device with clock synchronization and real-time recording of the present invention is based on high-frequency data acquisition technology, with an acquisition frequency of 50 acquisition cycles per second, 200 points are collected in each cycle, that is, 10,000 points are collected per second, which meets the power quality analysis requirements of the power grid; the collected data is AC data, and it is necessary to obtain harmonics, current, voltage, power and other data for analysis through transformation. The transformation method used in the present invention is Fourier transform, and the calculation method of Fourier transform is as follows: Among them, X(k) is the frequency domain signal; x(n) is the time domain signal; N is the number of sample points of the signal.

[0098] In order to test the performance of the recorder, the recorder needs to consider three dimensions for testing, namely voltage (57.7V, 100V, 220V, 380V); current (0.1A, 0.5A, 1A, 2.5A, 5A); temperature (0 degrees, 30 degrees, 60 degrees, 300 degrees, 330 degrees); according to the requirements of the recorder tester, the original recorded data collected in the above scenarios needs to be decomposed into the following fields:

[0099] u1_dc (DC component)_maximum value, u1_dc (DC component)_minimum value, u1_dc (DC component)_difference ratio;

[0100] u2_dc (DC component)_maximum value, u2_dc (DC component)_minimum value, u2_dc (DC component)_difference ratio;

[0101] u3_dc (DC component)_maximum value, u3_dc (DC component)_minimum value, u3_dc (DC component)_difference ratio;

[0102] i1_dc (DC component)_maximum value, i1_dc (DC component)_minimum value, i1_dc (DC component)_difference ratio;

[0103] i2_dc (DC component)_maximum value, i2_dc (DC component)_minimum value, i2_dc (DC component)_difference ratio;

[0104] i 3_dc (DC component)_maximum value, i 3_dc (DC component)_minimum value, i 3_dc (DC component)_difference ratio;

[0105] u1 fundamental wave amplitude_maximum value, u1 fundamental wave amplitude_minimum value, u1 fundamental wave amplitude_difference ratio;

[0106] u2 fundamental wave amplitude_maximum value, u2 fundamental wave amplitude_minimum value, u2 fundamental wave amplitude_difference ratio;

[0107] u3 fundamental wave amplitude_maximum value, u3 fundamental wave amplitude_minimum value, u3 fundamental wave amplitude_difference ratio;

[0108] i 1 fundamental wave amplitude_maximum value, i 1 fundamental wave amplitude_minimum value, i 1 fundamental wave amplitude_difference ratio;

[0109] i 2 fundamental wave amplitude_maximum value, i 2 fundamental wave amplitude_minimum value, i 2 fundamental wave amplitude_difference ratio;

[0110] i 3 fundamental wave amplitude_maximum value, i 3 fundamental wave amplitude_minimum value, i 3 fundamental wave amplitude_difference ratio;

[0111] u2 fundamental wave phase difference_maximum value, u2 fundamental wave phase difference_minimum value, u2 fundamental wave phase difference_difference value;

[0112] u3 fundamental wave phase difference_maximum value, u3 fundamental wave phase difference_minimum value, u3 fundamental wave phase difference_difference value;

[0113] i 1 fundamental wave phase difference_maximum value, i 1 fundamental wave phase difference_minimum value, i 1 fundamental wave phase difference_difference value;

[0114] i 2 fundamental wave phase difference_maximum value, i 2 fundamental wave phase difference_minimum value, i 2 fundamental wave phase difference_difference value;

[0115] i 3 fundamental wave phase difference_maximum value, i 3 fundamental wave phase difference_minimum value, i 3 fundamental wave phase difference_difference value;

[0116] u1_3_am_maximum, u1_3_am_minimum, u1_3_am_difference ratio;

[0117] u2_3_am_maximum, u2_3_am_minimum, u2_3_am_difference ratio;

[0118] u3_3_am_maximum, u3_3_am_minimum, u3_3_am_difference ratio;

[0119] i 1_3_am_maximum, i 1_3_am_minimum, i 1_3_am_difference ratio;

[0120] i 2_3_am_maximum, i 2_3_am_minimum, i 2_3_am_difference ratio;

[0121] i 3_3_am_maximum, i 3_3_am_minimum, i 3_3_am_difference ratio;

[0122] u1_5_am_maximum, u1_5_am_minimum, u1_5_am_difference_ratio;

[0123] u2_5_am_maximum, u2_5_am_minimum, u2_5_am_difference_ratio;

[0124] u3_5_am_maximum, u3_5_am_minimum, u3_5_am_difference ratio;

[0125] i 1_5_am_maximum, i 1_5_am_minimum, i 1_5_am_difference ratio;

[0126] i 2_5_am_maximum, i 2_5_am_minimum, i 2_5_am_difference_ratio;

[0127] i 3_5_am_maximum, i 3_5_am_minimum, i 3_5_am_difference ratio;

[0128] u1_7_am_maximum, u1_7_am_minimum, u1_7_am_difference ratio;

[0129] u2_7_am_maximum, u2_7_am_minimum, u2_7_am_difference ratio;

[0130] u3_7_am_maximum, u3_7_am_minimum, u3_7_am_difference ratio;

[0131] i 1_7_am_maximum, i 1_7_am_minimum, i 1_7_am_difference ratio;

[0132] i 2_7_am_maximum, i 2_7_am_minimum, i 2_7_am_difference ratio;

[0133] i 3_7_am_maximum, i 3_7_am_minimum, i 3_7_am_difference ratio;

[0134] u1_3_ph_maximum value, u1_3_ph_minimum value, u1_3_ph_difference ratio;

[0135] u2_3_ph_maximum value, u2_3_ph_minimum value, u2_3_ph_difference ratio;

[0136] u3_3_ph_maximum value, u3_3_ph_minimum value, u3_3_ph_difference ratio;

[0137] i 1_3_ph_maximum value, i 1_3_ph_minimum value, i 1_3_ph_difference ratio;

[0138] i 2_3_ph_maximum, i 2_3_ph_minimum, i 2_3_ph_differenceratio;

[0139] i 3_3_ph_maximum value, i 3_3_ph_minimum value, i 3_3_ph_difference ratio;

[0140] u1_5_ph_maximum value, u1_5_ph_minimum value, u1_5_ph_difference ratio;

[0141] u2_5_ph_maximum value, u2_5_ph_minimum value, u2_5_ph_difference ratio;

[0142] u3_5_ph_maximum value, u3_5_ph_minimum value, u3_5_ph_difference ratio;

[0143] i 1_5_ph_maximum value, i 1_5_ph_minimum value, i 1_5_ph_difference ratio;

[0144] i 2_5_ph_maximum, i 2_5_ph_minimum, i 2_5_ph_differenceratio;

[0145] i 3_5_ph_maximum, i 3_5_ph_minimum, i 3_5_ph_difference ratio;

[0146] u1_7_ph_maximum value, u1_7_ph_minimum value, u1_7_ph_difference ratio;

[0147] u2_7_ph_maximum value, u2_7_ph_minimum value, u2_7_ph_difference ratio;

[0148] u3_7_ph_maximum value, u3_7_ph_minimum value, u3_7_ph_difference ratio;

[0149] i 1_7_ph_maximum value, i 1_7_ph_minimum value, i 1_7_ph_difference ratio;

[0150] i2_7_ph_maximum, i2_7_ph_minimum, i2_7_ph_differenceratio;

[0151] i 3_7_ph_maximum value, i 3_7_ph_minimum value, i 3_7_ph_difference ratio;

[0152] P1_maximum value, P1_minimum value, P1_difference ratio;

[0153] P2_maximum value, P2_minimum value, P2_difference ratio;

[0154] P3_maximum value, P3_minimum value, P3_difference ratio;

[0155] Q1_maximum value, Q1_minimum value, Q1_difference ratio;

[0156] Q2_maximum value, Q2_minimum value, Q2_difference ratio;

[0157] Q3_maximum value, Q3_minimum value, Q3_difference ratio;

[0158] P_maximum value, P_minimum value, P_difference ratio;

[0159] Q_max, Q_min, Q_difference ratio.

[0160] Based on the above analysis results, the testers compared the normal data range with the results, debugged the equipment according to the abnormal data, re-collected and analyzed the data, and finally obtained the recording equipment that met the needs.

[0161] ②Data preparation: 1) Middle station data: Curve data: Electric energy indication table, 96 collection points every day, data after January 1 of a certain year, using three categories and six typical users, respectively involving office buildings, catering, and urban lighting; Daily label data: Daily electricity data currently has 2024 data; Monthly label data: Monthly electricity data for 2024; 2) Recording data: Waveform data: A week-long recording data collection was conducted on three categories and six typical users, respectively involving office buildings, catering, and urban lighting, and a total of A, B, and C three-phase current and voltage data were collected. Among the above six categories of users, only one user in the catering industry is three-phase four-wire, and the rest are three-phase three-wire; Date data: The newly developed recorder has added a time recording function, which automatically marks the time during collection.

[0162] ③Data processing: 1) The recorded data is transformed by Fourier transform. After the useless fields are removed, the remaining fields are:

[0163] Time, UA DC component, UA fundamental amplitude, UA effective value, UA3rd harmonic amplitude, UA5th harmonic amplitude, UA7th harmonic amplitude, UA9th harmonic amplitude, UA11th harmonic amplitude, UA13th harmonic amplitude, UA15th harmonic amplitude, UA17th harmonic amplitude, UA19th harmonic amplitude, UA21st harmonic amplitude, UA23rd ​​harmonic amplitude, UA25th harmonic amplitude, UA27th harmonic amplitude, UA29th harmonic amplitude, UA31st harmonic amplitude, UA33rd harmonic amplitude, UA35th harmonic amplitude, UA37th harmonic amplitude;

[0164] UC DC component, UC fundamental wave amplitude, UC effective value, UC3rd harmonic amplitude, UC5th harmonic amplitude, UC7th harmonic amplitude, UC9th harmonic amplitude, UC11th harmonic amplitude, UC13th harmonic amplitude, UC15th harmonic amplitude, UC17th harmonic amplitude, UC19th harmonic amplitude, UC21st harmonic amplitude, UC23rd harmonic amplitude, UC25th harmonic amplitude, UC27th harmonic amplitude, UC29th harmonic amplitude, UC31st harmonic amplitude, UC33rd harmonic amplitude, UC35th harmonic amplitude, UC37th harmonic amplitude;

[0165] IA DC component, IA fundamental wave amplitude, IA effective value, IA3rd harmonic amplitude, IA5th harmonic amplitude, IA7th harmonic amplitude, IA9th harmonic amplitude, IA11th harmonic amplitude, IA13th harmonic amplitude, IA15th harmonic amplitude, IA17th harmonic amplitude, IA19th harmonic amplitude, IA21st harmonic amplitude, IA23rd ​​harmonic amplitude, IA25th harmonic amplitude, IA27th harmonic amplitude, IA29th harmonic amplitude, IA31st harmonic amplitude, IA33rd harmonic amplitude, IA35th harmonic amplitude, IA37th harmonic amplitude;

[0166] IC DC component, IC fundamental wave amplitude, IC effective value, IC3rd harmonic amplitude, IC5th harmonic amplitude, IC7th harmonic amplitude, IC9th harmonic amplitude, IC11th harmonic amplitude, IC13th harmonic amplitude, IC15th harmonic amplitude, IC17th harmonic amplitude, IC19th harmonic amplitude, IC21st harmonic amplitude, IC23rd harmonic amplitude, IC25th harmonic amplitude, IC27th harmonic amplitude, IC29th harmonic amplitude, IC31st harmonic amplitude, IC33rd harmonic amplitude, IC35th harmonic amplitude, IC37th harmonic amplitude;

[0167] Power factor 1, P1, Q1, Power factor 3, P3, Q3, P, Q, IA harmonic content, IC harmonic content.

[0168] 2) Missing values ​​are filled by linear interpolation method.

[0169] 3) Outliers are directly eliminated.

[0170] After data screening, there are approximately 181.44 million valid samples.

[0171] ④ Power consumption feature screening: The power curves of 6 scenarios are as follows: 1) Office building: Figure 2-3 ; 2) Catering: such as Figure 4-5 ; 3) Urban lighting: such as Figure 6-7 shown.

[0172] From the power curves of six typical users in three categories, we can see that the main electricity consumption time for office buildings is from 9:00 to 12:00 in the morning and from 14:00 to 18:00 in the afternoon, which is a typical double peak type at noon; the electricity consumption of catering is mainly concentrated in the peak dining hours at noon and in the evening, which is a double peak type during the day; the characteristics of urban lighting are more prominent, with stable power from 17:00 in the evening to nearly 6:00 in the morning of the next day, and the power is close to 0 at the rest of the time.

[0173] The electricity consumption characteristics of power users are reflected in the characteristics of user load curves. The main analysis indicators for load characteristics are: maximum load t1, minimum load t2, average load t3, load rate t4, peak-to-valley difference rate t5, peak power consumption rate t6, normal power consumption rate t7, and valley power consumption rate t8. The above characteristic indicators constitute the original feature set of power users, that is, the original feature set is T = {t1, t2, ..., t8}.

[0174] By sorting out the data of 6 users, the above features were selected and their importance was analyzed using random forest. The analysis results are as follows Figure 8 As shown by Figure 8 It can be seen that the average power value and the maximum power value are of the highest importance, while the importance of the minimum power value, load factor, peak-to-valley difference rate, peak load factor, valley load factor, and normal load factor are roughly the same and relatively low. The calculation results of the Kendall correlation coefficient between different features are shown in the table below.

[0175]

[0176] It can be seen from the above table that there is a high positive correlation between the maximum power and the average power, and between the peak-to-valley difference rate and the normal load rate, all of which are not less than 0.5; there is a high negative correlation between the peak-to-valley difference rate and the peak load rate, the peak load rate and the normal load rate, and the normal load rate and the valley load rate, all of which are lower than -0.5; it can be seen that the analysis method based on the Kendall correlation coefficient can effectively identify the correlation between different features.

[0177] First, a certain number of initial populations are generated randomly (the population size is 100), and then SSA is used for iterative solution. During the iteration process, the better individuals in the population are selected through the fitness function. The larger the fitness value, the greater the probability of being selected. Then, according to the sparrow's predation rules, a new generation of population is finally formed. A certain stopping criterion is set (the present invention is set to an upper limit of 200 iterations), and the algorithm repeats the iterative update of the population according to the above rules until the stopping criterion is met.

[0178] The individual with the highest fitness value obtained based on the above algorithm is T = {0, 0, 0, 1, 1, 1, 0, 1}, that is, the optimal feature set solution result is T = {t4, t5, t6, t8}. The process of SSA iterative solution of the optimal feature set is shown in Fig. 9 As can be seen from the figure, the number of iterations to obtain the optimal individual is 12, and the fitness value of the corresponding individual is 1.2405, which shows the high efficiency of SSA in solving the optimal feature set. However, the average fitness of the population is about 0.9 at this time, while the average fitness of the population finally obtained by SSA is about 1.2. This result shows that when obtaining the optimal individual, the differences between individuals in the population are still large, and the algorithm still has a large search space in the global range, that is, the algorithm has a strong ability to search for the global optimal solution.

[0179] ⑤ Harmonic characteristic analysis: In daily electricity use, the voltage is relatively stable, while the current fluctuates greatly. Therefore, the current and the corresponding harmonic proportion are mainly analyzed. For the three-phase three-wire scenario, only phases A and C are considered. For the three-phase four-wire scenario, phases A, B, and C are considered. The Kendall correlation coefficient method is used to determine the correlation between the current and the corresponding harmonic content in the above scenarios. The results are shown in the following table.

[0180] Science Park Newspaper Building Food City Du Kang Zhengzhou City Lighting Zhongyuan Road Lighting Phase A -0.66 -0.54 -0.37 -0.14 0.37 -0.36 Phase C -0.68 -0.47 -0.4 -0.18 0.41 -0.41

[0181] It can be seen from the above table that the harmonic content of the office building type scene has a more obvious negative correlation with the corresponding current value; the harmonic content of the catering type has a smaller negative correlation with the corresponding current value; and the harmonics of the urban lighting type will only undergo irregular sudden distortion, and since there is no fixed upward or downward trend, there is no fixed positive or negative correlation between the harmonic content and the corresponding current.

[0182] In addition to analyzing the relationship between harmonic content and corresponding current, the relationship between harmonics and user scenarios can also be observed by analyzing the proportion of each harmonic in the total harmonics. The present invention calculates the 3rd to 37th harmonics for each scenario. Among office users, the 5th and 7th harmonics are the main components. Among catering users, the 3rd, 5th and 7th harmonics account for a relatively large proportion. In urban lighting, except for the 5th and 7th harmonics which are slightly higher, the proportions of the other multiple harmonics are relatively even.

[0183] In summary, we can summarize the characteristics of harmonics in each scenario: 1. In office building scenarios, the harmonic content is highly negatively correlated with the current, and the harmonics are mainly composed of the 5th and 7th harmonics; 2. In the catering industry scenario, the harmonic content is relatively low negatively correlated with the current, and the harmonics are mainly composed of the 3rd, 5th and 7th harmonics; 3. In urban lighting scenarios, the harmonic content has no fixed positive or negative correlation with the current, and the harmonics are mainly composed of the 5th and 7th harmonics, with multiple other harmonics evenly distributed.

[0184] ⑥ User portrait construction: Based on the summarized power consumption characteristics and harmonic characteristics, a user portrait can be jointly constructed. Taking office buildings as an example, the user portrait is as follows: Fig.10 As shown in the figure, it can be seen that user portraits are divided into two types of label libraries, namely electricity consumption behavior characteristics and harmonic characteristics. The electricity consumption behavior label library can formulate time-of-use electricity prices for them in a targeted manner, which is conducive to guiding users to shift peak loads and assisting the power grid in peak shaving; the harmonic characteristic label library can evaluate the impact of harmonics on power quality and find the impact of each harmonic on power, which is helpful to optimize users' energy efficiency, improve overall energy utilization efficiency, predict and maintain equipment failures, and formulate targeted demand response strategies.

[0185] ⑦Result analysis: Through comprehensive analysis of the wave recording data provided by the data center and the newly developed wave recorder, the data normalization acquisition and analysis tasks are summarized as follows:

[0186] 1. Through machine learning and statistical methods, we analyzed users’ electricity consumption data and found the electricity consumption feature set with the best combination of influence and redundancy.

[0187] 2. Using the waveform decomposition method and hardware joint debugging, we assisted the testers in completing a new recording device with clock synchronization function.

[0188] 3. By converting the recording data provided by the new recorder and analyzing the harmonic data, we found the harmonic composition ratios of three categories of six typical users, and correlated the harmonic data with the electricity consumption data, and obtained the correlation between the harmonic data and electricity consumption data of each type of user.

[0189] 4. The user's electricity consumption characteristics and harmonic characteristics are integrated to jointly construct a user portrait method. The portrait library can be supplemented later by adding recording data of other types of scenarios.

[0190] The present invention is a waveform data labeling method under a typical scenario on the user demand side. During use, the present invention implements hardware joint debugging of a new recording device for processing and analyzing the recording data, which helps the recording device to timely discover and correct problems in the research and development and debugging stage, and ensure normal operation during the subsequent recording process; the present invention constructs a user portrait method that integrates user power consumption characteristics and harmonic characteristics by performing feature analysis on the recording data of users in typical scenarios collected by the recorder, and combining the power load data on the user side of the communication data center; the present invention can achieve synchronization of real-time data and collection time through in-depth research on the clock synchronization and real-time recording technology of the electric energy recorder, and ensure the reliability of data monitoring in multiple user scenarios, which not only provides strong data support for subsequent data extraction, analysis and fault diagnosis, but also effectively reduces the occurrence of power grid faults, improves the stability and reliability of the power grid, and thus reduces the economic losses caused by power outages and other faults; and it helps to improve the fault analysis and judgment accuracy of the power grid in accidents, thereby ensuring the safe operation of the power grid; the present invention has the advantages of integrating user power consumption and harmonic characteristics, constructing user portraits, improving fault judgment accuracy, and improving power grid stability and reliability.

[0191] Example 3

[0192] like Figure 1-10 As shown, a recorder with clock synchronization constructs a user portrait combining power consumption characteristics and harmonic characteristics based on the waveform data annotation method in the typical scenario on the user demand side as described above. The recorder includes a signal processing module, a power consumption characteristic screening module and a recording data processing and analysis module; wherein the signal processing module is used to analyze the recording data for testing, generate appropriate debugging data, and feed back to the recorder debugging personnel, so as to adjust the function parameters of the recorder; the power consumption characteristic screening module is used to analyze the user-side power consumption behavior data using the load data of the residential area, and screen the user's power consumption characteristics through machine learning and statistical methods; the recording data processing and analysis module is used to process the collected recording data, perform correlation analysis in combination with the power consumption behavior data, and extract harmonic characteristics related to the user's power consumption behavior.

Claims

1. A waveform data annotation method in a typical scenario on the user demand side, characterized by: The method comprises the following steps: Step 1: Acquisition of original data: Collect electricity consumption data of three typical users to obtain electricity consumption information of users; Step 2: Data preprocessing: Filter and clean the original complex data information; Step 3: Mining and analyzing the data generated by users: Through data mining and analysis, find out the operating rules of users when using electricity and obtain the user behavior model of household electricity users; Step 4: Build model labels: Label the user's features based on the user model obtained through analysis; Step 5: Correlation analysis based on harmonic data: Use the user's harmonic data and current data for correlation analysis to improve the portrait of household electricity users; Step 6: Build a power user profile based on the user's power consumption characteristics and harmonic characteristics.

2. The waveform data labeling method in a typical scenario on the user demand side according to claim 1, characterized in that: The data acquisition in step 1 is based on the data center to obtain the archive data, operation data, event data, and auxiliary analysis data in the marketing business system and the electricity consumption information collection system; Among them, ① archive data includes: electricity meter information, user information, metering point information; ② operation data includes: daily forward and reverse electricity indication curve of measurement point, daily active power curve, daily reactive power curve, daily power factor curve, daily voltage curve, daily current curve; ③ event data includes: electricity meter flow reverse event, electricity meter power reverse event; ④ auxiliary analysis data includes: phase-by-phase forward and reverse electricity indication data, phase angle data.

3. The waveform data labeling method in a typical scenario on the user demand side according to claim 1, characterized in that: The data preprocessing in step 2 has the following specific processing rules: ① filter user types; ② eliminate null values; ③ eliminate reverse settlement users; ④ eliminate photovoltaic users; ⑤ eliminate no-load; ⑥ eliminate abnormal data such as loss of voltage and phase failure; ⑦ eliminate incomplete daily data of voltage, current, power and power factor.

4. The waveform data annotation method in a typical scenario on the user demand side according to claim 1, characterized in that: The step 6 of constructing a power user profile based on the user's power consumption characteristics and harmonic characteristics includes the following steps: Step 6.1: First, use the random forest and Kendall correlation coefficient methods to analyze the influence of user electricity consumption characteristics on electricity consumption behavior and the redundancy between characteristics; Step 6.2: Construct a feature fitness evaluation index and use the sparrow search algorithm for iterative optimization to screen out the most representative user electricity consumption features; Step 6.3: Secondly, through the correlation analysis of load data and harmonic data, the relationship between harmonic content and load data in each scenario is explored to reveal the performance of harmonic characteristics in different power consumption scenarios; Step 6.4: Finally, combine the user's electricity consumption characteristics with the harmonic characteristics to build a comprehensive power user portrait; Step 6.5: Verify the effectiveness through case analysis, which provides new perspectives and methods for the optimization and management of power systems.

5. The waveform data labeling method in a typical scenario on the user demand side according to claim 4, characterized in that: The random forest algorithm in step 6.1 is specifically: by constructing multiple decision trees and integrating their results, the mathematical description of the decision tree is as follows: for a sample set D containing n categories Hi (i = 1, 2, ..., n), in, Where I is the expected entropy of the sample; d i is the number of samples belonging to category Hi; d is the total number of samples; E(A) is the expected entropy of a single feature A of the sample; k is the total number of features in the sample set; d ij is the jth feature of the sample belonging to category i.

6. The waveform data labeling method in a typical scenario on the user demand side according to claim 5, characterized in that: The Kendal 1 correlation coefficient method in step 6.1 is specifically as follows: for any features A1, A2, the calculation method of the Kendal 1 correlation coefficient is as follows: Among them, B and C are sample pairs with consistent and inconsistent rankings respectively; T1 and T2 are the number of samples that appear the same in the sample sets of features A1 and A2 respectively.

7. The waveform data labeling method in a typical scenario on the user demand side according to claim 4, characterized in that: The feature fitness evaluation index in step 6.2 is constructed as follows: for any feature set a∈A, its evaluation function P(a) is: P(a)=R(a)-K(a)(9), where: Where R(a) and K(a) are the importance value and redundancy of the feature set a respectively; N a is the number of features contained in feature set a; m1 and m2 are two different features in feature set a.

8. The waveform data labeling method in a typical scenario on the user demand side according to claim 7, characterized in that: The sparrow search algorithm in step 6.2 is iteratively optimized, specifically: the sparrow population is divided into finders, followers and scouts, and the position of the finder is updated during the predation process as follows: in, is the position of the i-th sparrow in the t-th iteration and the j-th dimension; α is a random number between (0, 1]; R2∈[0,1] and ST∈[0.5,1] are the warning value and the safety value respectively; Q is a random number that follows a normal distribution; L is a 1×d matrix.

9. The waveform data labeling method in a typical scenario on the user demand side according to claim 8, characterized in that: The follower's position is updated as follows: Among them, x p With x worst are the best position and the global worst position of the discoverer respectively; n is the population size; A is a 1×d matrix; The positions of the scout positions are updated as follows: Among them, x best is the global optimal position; β is a normally distributed random number with a mean of 0 and a variance of 1; k is a random number between [-1,1]; f i is the fitness value of the current sparrow; f g With f w are the current global optimal and worst fitness values ​​respectively; ε is the minimum constant.

10. A oscilloscope with clock synchronization, based on the waveform data annotation method in a typical scenario on the user demand side as described in any one of claims 1 to 9, constructs a user portrait combining power consumption characteristics and harmonic characteristics, characterized in that: The recorder includes a signal processing module, an electricity consumption feature screening module and a recording data processing and analysis module; wherein the signal processing module is used to analyze the recording data for testing, generate appropriate debugging data, and feed back to the recorder debugging personnel, so as to adjust the function parameters of the recorder; the electricity consumption feature screening module is used to analyze the user-side electricity consumption behavior data using the load data of the residential area, and screen the user's electricity consumption characteristics through machine learning and statistical methods; the recording data processing and analysis module is used to process the collected recording data, perform correlation analysis in combination with the electricity consumption behavior data, and extract harmonic characteristics related to the user's electricity consumption behavior.

Images