Rapid construction method of power data analysis model based on big data

By monitoring and processing distribution network data, building power data feature vectors and using self-organized competitive neural network training, the problem of slow construction of power data analysis models in the existing technology is solved, and fast and accurate intelligent power data analysis is achieved. .

CN118916646AInactive Publication Date: 2024-11-08JIANGSU ELECTRIC POWER INFORMATION TECH
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202411413307.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-11
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing power data analysis model cannot quickly build intelligent models while ensuring accuracy and efficiency, especially in special emergency scenarios.

Method used

By monitoring and collecting the current data and voltage data of phase A, phase B and phase C during the distribution network operation, a power data feature vector is constructed, and a self-organized competition neural network is used for training to build a power data analysis model.

Benefits of technology

It realizes the rapid construction of intelligent power data analysis models while ensuring the accuracy and efficiency of power failure analysis, simplifying the model training process, reducing the computing resource requirements and manual labeling costs, and improving the efficiency of model construction and analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916646A_ABST
    Figure CN118916646A_ABST
Patent Text Reader

Abstract

The invention discloses a rapid construction method of an electric power data analysis model based on big data, and relates to the technical field of electric power data analysis models, and the method comprises the following steps: monitoring and collecting current data and voltage data of an A phase, a B phase and a C phase during the operation of a power distribution network; performing data calculation processing by using the current data and the voltage data, and constructing a first power data set; performing feature extraction processing by using the first power data set, constructing a second power data set, and establishing a power data feature vector; training a self-organizing competitive neural network based on the second power data set and the power data feature vector, and constructing a power data analysis model; the method is used for solving the problem that an intelligent electric power data analysis model cannot be quickly constructed by using big data while the accuracy and the high efficiency of electric power fault analysis based on electric power data cannot be ensured by an existing electric power data analysis model technology.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of power data analysis models, and in particular to a method for quickly constructing a power data analysis model based on big data. Background Art

[0002] Power data analysis models are a series of mathematical and statistical tools used to extract valuable information from power data, predict future trends and optimize power grid performance. These models help power systems improve performance, safety and efficiency by conducting in-depth analysis of large-scale power data. They cover multiple power fields, including energy management, fault detection, and improvement of the stability and safety of power grid systems.

[0003] Traditional power data analysis models are built based on simple mathematical processing or fixed algorithms, and power data analysis is performed through some simple characteristic values ​​of power data, such as mean, variance, and rising and falling trends. The accuracy of power data analysis is not high enough, and it cannot repeatedly mine the hidden characteristics of power data. Moreover, it relies on manual judgment, which is not efficient and smart enough. The current mainstream power data analysis models are often based on deep learning technology, which uses big data to train deep learning models and then perform power data analysis. However, deep learning models are often complex, and the training process is indeed complex and time-consuming, requiring a lot of computing resources. In addition, relevant parameters must be adjusted and optimized during the training process, and it is also time-consuming to mark the training data before training begins.

[0004] For example, a patent application with publication number CN112508363A discloses a method and device for analyzing the state of an electric power information system based on deep learning. The solution is to first label the training data, then train the deep learning model with the labeled data, and finally obtain the corresponding analysis model. The training process is complicated and time-consuming, and it is impossible to quickly build a model. In some special emergency scenarios, building an efficient and accurate electric power data analysis model based on deep learning technology cannot meet the time efficiency requirements. Therefore, the existing electric power data analysis model technology cannot ensure the accuracy and efficiency of power fault analysis based on power data, while using big data to quickly build an intelligent electric power data analysis model. Summary of the invention

[0005] The present invention aims to solve at least one of the technical problems in the prior art to a certain extent, by monitoring and collecting the current data and voltage data of phase A, phase B and phase C when the distribution network is running; using the current data and voltage data for processing to establish a power data feature vector; and then training the self-organizing competitive neural network to construct a power data analysis model; to solve the problem that the existing power data analysis model technology cannot ensure the accuracy and efficiency of power fault analysis based on power data while using big data to quickly build an intelligent power data analysis model.

[0006] To achieve the above objectives, the present application provides a method for quickly constructing a power data analysis model based on big data, comprising the following steps: Monitor and collect current and voltage data of phase A, phase B and phase C during distribution network operation; Performing data calculation and processing using the current data and the voltage data, and constructing a first power data set; Performing feature extraction processing using the first power data set, constructing a second power data set, and establishing a power data feature vector; The self-organizing competitive neural network is trained based on the second power data set and the power data feature vector to construct a power data analysis model.

[0007] Furthermore, monitoring and collecting the current data and voltage data of phase A, phase B and phase C during the operation of the distribution network includes the following sub-steps: Collecting voltage data: collecting phase A voltage information, phase B voltage information, and phase C voltage information of the distribution network at first time intervals within a first time period, and marking the phase A voltage information, phase B voltage information, and phase C voltage information as voltage data; Collect current data: collect the A-phase current information, B-phase voltage and current, and C-phase current information of the distribution network at a first time interval within a first time period, and mark the A-phase current information, B-phase voltage and current, and C-phase current information as current data; the first time period is T, and the first time interval is d.

[0008] Furthermore, the data calculation process using the current data and the voltage data includes the following sub-steps: Calculate active power: Based on the current data and voltage data, use the active power calculation formula to calculate the active power of phase A, phase B, and phase C respectively. The active power calculation formula is as follows: , where P represents the active power of any phase circuit among phase A, phase B and phase C, U 相 and I 相 represents the voltage value and current value of the corresponding phase circuit, φ represents the phase difference between the voltage and current of the corresponding phase circuit; the active power of phase A, the active power of phase B and the active power of phase C are marked as active power data; Calculate reactive power: Based on the current data and voltage data, use the reactive power calculation formula to calculate the reactive power of phase A, phase B, and phase C respectively. The reactive power calculation formula is as follows: , where Q represents the active power of any phase circuit among phase A, phase B and phase C, U 相 and I 相 represents the voltage value and current value of the corresponding phase circuit, φ represents the phase difference between the voltage and current of the corresponding phase circuit; the reactive power of phase A, the reactive power of phase B and the reactive power of phase C are marked as reactive power data.

[0009] Furthermore, constructing the first power data set includes the following sub-steps: The current data, voltage data, active power data and reactive power data are classified and stored according to the associated phase A, phase B or phase C. Each phase in phase A, phase B and phase C contains the current data, voltage data, active power data and reactive power data of the phase. After completion, the first power data set is obtained.

[0010] Furthermore, performing feature extraction processing using the first power data set includes the following sub-steps: Current feature extraction processing: Based on the current data of phase A, phase B, and phase C, the maximum current and the minimum current of phase A, phase B, and phase C at the same time are screened and extracted, and the maximum current is marked as I max , the minimum current is marked as I min ; The current imbalance degree is calculated based on the maximum current and the minimum current using the current imbalance degree formula. The current imbalance degree formula is as follows: , where I un Represents the current unbalance, I max Represents the maximum current value among the current data of phase A, phase B, and phase C, I min Representing the minimum current value in the current data of phase A, phase B and phase C; obtaining the current imbalance at each moment at the first time interval within the first time period, marked as current imbalance data; The current imbalance threshold is set based on the current imbalance data, marked as I th , based on the current imbalance data, the number of current imbalances greater than the current imbalance threshold is counted, marked as X0; Based on the current data, set the current over-threshold and the current under-threshold, mark the current over-threshold as Bi1, and the current under-threshold as Bi2; based on the current data, count the number of currents greater than the current over-threshold, mark them as X1; count the number of currents less than the current under-threshold, mark them as X2; Calculate the Euclidean distance between phase A current and phase B current based on the current data, marked as L AB , calculate the Euclidean distance between phase A current and phase C current, marked as L AC , calculate the Euclidean distance between the current of phase B and the current of phase C, marked as L BC ; Then, voltage feature extraction is performed based on the voltage data.

[0011] Furthermore, the voltage feature extraction process based on the voltage data includes the following sub-steps: Set the voltage over-limit threshold and voltage under-limit threshold, mark the voltage over-limit threshold as Bv1, and the voltage under-limit threshold as Bv2; count the number of voltages greater than the voltage over-limit threshold based on voltage data, mark it as X3; count the number of voltages less than the voltage under-limit threshold, mark it as X4; Based on the voltage data of phase A, phase B and phase C, respectively calculate the effective values ​​of phase A voltage, phase B voltage and phase C voltage in the first time period; mark the effective value of phase A voltage in the first time period as EA, the effective value of phase B voltage in the first time period as EB, and the effective value of phase C voltage in the first time period as EC; The voltage data of phase A, phase B and phase C are respectively composed into vector sequences with dimension m: m 1,U m 2,....,U m N-m+1 , U m i ={V(i), V(i+1), ..., V(i+m-1)}, where N is the number of voltage data of phase A, phase B and phase C, i is a positive integer and 1≤i≤N-m+1, V(i) represents the voltage value of the i-th sampling point of a phase, U m i Represents the i-th vector sequence; calculate U m i with U m j The Chebyshev distance of m ij , where j is a positive integer and 1≤j≤N-m+1, statistics d m ij The number of values ​​less than or equal to r is marked as c i , calculate ci / N-m+1, marked as c i m,r , define Φ m,r c i m,r The average value of the logarithmic representation, Φ m,r for: ; The voltage data of phase A, phase B and phase C are respectively formed into vector sequences with dimension m+1, and then Φ is calculated. m+1,r , calculate the ApEn of phase A A相 , ApEn of phase B B相 and ApEn of phase C C相 ,ApEn=Φ m,r -Φ m+1,r ;Define F=max(ApEn A相 , ApEn B相 , ApEn C相 ); Load feature extraction and processing are performed based on active power data and reactive power data.

[0012] Furthermore, the load feature extraction process based on the active power data and the reactive power data includes the following sub-steps: Based on the active power data and the reactive power data, the total load power of each phase at each time instant in the first time period is calculated, which is marked as total power data; the average total load power of the distribution network in the first time period is calculated, which is marked as S; Based on the total power data, the number of loads that exceed 80% of the normal load upper limit is counted and marked as X5.

[0013] Furthermore, constructing a second power data set and establishing a power data feature vector includes the following sub-steps: Mark the power imbalance data, the total power data, X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, F and S as a second power data set; Based on the second power data set, the power data feature vector is defined as Y, where Y={X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, S, F}.

[0014] Furthermore, training the self-organizing competitive neural network based on the second power data set and the power data feature vector to construct a power data analysis model includes the following sub-steps: Normalize the power data feature vector Y to obtain Y0; Training the current analysis sub-model: setting four current anomaly classifications, including slight current anomaly, relatively serious fault, severe fault, and very serious fault; setting the first learning rate and the first number of training rounds, setting the current weight vector, setting the output layer neurons to 4, and training Y0 as the input of the self-organizing competitive neural network based on the second power data. After completion, the current analysis sub-model is obtained; Training the voltage analysis sub-model: set 4 voltage anomaly classifications, including slight voltage anomaly, serious voltage fault, low voltage and high voltage; set the second learning rate and the second number of training rounds, set the voltage weight vector, set the output layer neurons to 4, and train Y0 as the input of the self-organizing competitive neural network based on the second power data. After completion, the voltage analysis sub-model is obtained.

[0015] Furthermore, training the self-organizing competitive neural network based on the second power data set and the power data feature vector, and constructing the power data analysis model further includes the following sub-steps: Training the load analysis sub-model: set three load anomaly classifications, including slight load anomaly, line overload, and line super overload, set the third learning rate and the third number of training rounds, set the load weight vector, set the output layer neurons to 3, and train Y0 as the input of the self-organizing competitive neural network based on the second power data. After completion, the load analysis sub-model is obtained; The current analysis sub-model, the voltage analysis sub-model and the load analysis sub-model are connected in parallel to obtain a power data analysis model.

[0016] Beneficial effects of the present invention: The present invention collects the current data and voltage data of phase A, phase B and phase C during the operation of the distribution network through monitoring; then uses the current data and voltage data for data calculation and processing, and constructs a first power data set; then uses the first power data set for feature extraction and processing, constructs a second power data set, and establishes a power data feature vector; finally, based on the second power data set and the power data feature vector, the self-organizing competitive neural network is trained to construct a power data analysis model; while ensuring the accuracy and efficiency of power fault analysis based on power data, an intelligent power data analysis model can be quickly constructed using big data, and the model training is simple and efficient, and the data analysis is accurate and practical; The present invention establishes a power data feature vector through feature extraction processing, which can reduce the amount of data input during model training and analysis, reduce the interference of useless data on the model, improve the efficiency and reliability of model construction, and improve the accuracy of power data analysis; construct a power data analysis model based on a self-organizing competitive neural network, which is relatively simple in structure, usually requires less computing resources and a shorter training time, and does not require label data during the training process. It can automatically classify input data, greatly reducing the cost and time of manual labeling, and greatly improving the efficiency of model construction; constructing a current analysis sub-model, a voltage analysis sub-model, and a load analysis sub-model respectively can improve the accuracy of model analysis, and decomposing a complex model into different sub-models can greatly reduce the amount of calculation for each analysis; improving the efficiency of model construction and analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of the steps of the method of the present invention; Figure 2 It is a schematic diagram of the structure of the self-organizing competitive neural network of the present invention; Figure 3 It is a schematic diagram of the structure of the power data analysis model of the present invention; Figure 4 This is a flow chart of power data analysis of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Example 1, please refer to Figure 1 As shown, the present application provides a method for quickly constructing a power data analysis model based on big data; the method comprises the following steps: Step S1, monitoring and collecting current data and voltage data of phase A, phase B and phase C when the distribution network is running; step S1 includes the following sub-steps: Step S101, collecting A-phase voltage information, B-phase voltage information and C-phase voltage information of the distribution network at first time intervals within a first time period; Step S102, marking the A-phase voltage information, the B-phase voltage information and the C-phase voltage information as voltage data; Step S103, collecting A-phase current information, B-phase voltage and current, and C-phase current information of the distribution network at first time intervals within a first time period; Step S104, marking the A-phase current information, the B-phase voltage and current, and the C-phase current information as current data, the first time period is T, the first time interval is d, in this embodiment, the first time period is one day, that is, 24 hours, and the first time interval is 15 minutes; In the specific implementation process, three-phase electricity is a form of transmission of electric energy, also known as; three-phase alternating current, which is a power source composed of three alternating potentials with the same frequency, equal amplitude, and phase difference of 120°; it contains three phases A, B, and C, the voltage between phases is 380V, and the voltage between phases and the neutral line is 220V; usually the alternating current produced, transmitted and distributed by power plants and power grids is three-phase electricity; three-phase alternating current has many uses, and most of the AC power equipment in industry, such as motors, use three-phase alternating current; for the collection of current data and voltage data, it is necessary to collect current data and voltage data within multiple first time periods. For example, in this embodiment, the first time period is 1 day, that is, it is necessary to collect current data and voltage data for multiple days.

[0020] Step S2, using the current data and the voltage data to perform data calculation and processing, and construct a first power data set; Step S2 includes the following sub-steps: Step S201, calculate active power: Based on the current data and voltage data, use the active power calculation formula to calculate the active power of phase A, phase B and phase C respectively. Active power refers to the power part that is actually consumed in the AC circuit and converted into other forms of energy, such as heat energy, light energy or mechanical energy; it is a measure of the actual role of electrical energy in practical applications; the active power calculation formula is as follows: , where P represents the active power of any phase circuit among phase A, phase B and phase C, U 相 and I 相 represents the voltage and current values ​​of the corresponding phase circuit, and φ represents the phase difference between the voltage and current of the corresponding phase circuit; for example, at a certain moment, U 相 =220v, I 相 =12A, φ=60°, then the active power P of the phase A circuit at this moment is 220*12*cos60°=1320w; the active power of phase A, the active power of phase B and the active power of phase C are marked as active power data.

[0021] Step S202, calculate reactive power: Based on the current data and voltage data, the reactive power of phase A, phase B and phase C are calculated respectively using the reactive power calculation formula. Reactive power exists in an AC circuit with inductance or capacitance. It is generated by the inductor coil and transformer, or the capacitor in the process of storing and releasing energy in the AC circuit. It does not directly do work, but is a part of the power required for doing work. The reactive power calculation formula is as follows: , where Q represents the active power of any phase circuit among phase A, phase B and phase C, U 相 and I 相represents the voltage and current value of the corresponding phase circuit, and φ represents the phase difference between the voltage and current of the corresponding phase circuit; for example, at a certain moment, U 相 =218v,I 相 =5A, φ=30°, then the reactive power P of the phase B circuit at this moment is P=218*5*sin30°=545w; the reactive power of phase A, phase B and phase C are marked as reactive power data.

[0022] Step S203, classify and store the current data, voltage data, active power data and reactive power data according to the associated phase A, phase B or phase C, and each phase of phase A, phase B and phase C contains the current data, voltage data, active power data and reactive power data of the phase; for example, phase A contains the collected current data, voltage data and calculated active power data and reactive power data of phase A; Step S204, after completion, a first power data set is obtained; In the specific implementation process, in the AC circuit, all inductive or capacitive elements will establish an electric field between the magnetic poles of the inductor coil after being energized; therefore, in the first half of each cycle of the AC power when the instantaneous power is positive, they will absorb energy from the power supply to establish a magnetic field or an electric field; and in the second half when the instantaneous power is negative, the magnetic field or electric field energy established returns to the power supply; therefore, the average value of this power in the entire cycle is equal to zero; that is, the energy of the power supply and the magnetic field energy or the electric field energy are undergoing energy conversion without consuming power; reactive power is the part of electric power that is reversibly converted due to the presence of reactive elements in the AC circuit, referring to pure inductance or pure capacitance, and it expresses the maximum rate of AC power supply energy and magnetic field or electricity.

[0023] In actual work, for any inductive load with coils and iron cores, the power consumed by them to establish a magnetic field during operation is reactive power. Although reactive power does not do actual work, it is responsible for establishing the necessary alternating magnetic field in the power system, which is essential for the normal operation of equipment such as motors and transformers. Without sufficient reactive power, these devices will not be able to work properly. Active power and reactive power are two different types of power in an AC circuit. In a three-phase AC system, the existence and circulation of these two types of power are essential for the stable operation of the entire power system. Active power and reactive power not only affect the efficiency of the power system, but also have a profound impact on the stability and security of the power network.

[0024] Step S3, using the first power data set to perform feature extraction processing, construct a second power data set, and establish a power data feature vector; Step S3 includes the following sub-steps: Step S301, performing current feature extraction processing; Step S301 includes the following sub-steps: Step S3011, based on the current data of phase A, phase B and phase C, filter and extract the maximum current and the minimum current among the currents of phase A, phase B and phase C at the same time, and mark the maximum current as I max , the minimum current is marked as I min For example, at a certain moment, the current of phase A is 13A, the current of phase B is 10A, and the current of phase C is 17A. The current of phase C>the current of phase A>the current of phase B. Therefore, at this moment, I max =17A,I min =10A; Step S3012, calculating the current imbalance degree using a current imbalance degree formula based on the current maximum value and the current minimum value. The current imbalance degree formula is as follows: , where I un Represents the current unbalance, I max Represents the maximum current value among the current data of phase A, phase B, and phase C, I min The minimum current value in the current data representing phase A, phase B and phase C is obtained; the current imbalance at each moment in the first time interval within the first time period is obtained, and marked as current imbalance data; for example, in this embodiment, the first time period is 1 day, that is, 24 hours, and the first time interval is 15 minutes, so there are 96 data sampling points in one day, and the current imbalance of 96 data sampling points needs to be calculated in one day.

[0025] Step S3013, setting a current imbalance threshold based on the current imbalance data. The setting of the current imbalance threshold should be specifically set according to the actual application scenario. In this embodiment, the current imbalance threshold is set according to the current imbalance data, all current imbalances are arranged in order of magnitude, and the median is taken as the current imbalance threshold; marked as I th , based on the current imbalance data, the number of current imbalances greater than the current imbalance threshold is counted, marked as X0; Step S3014, based on the current data, set the current too large threshold and the current too small threshold. The current too large threshold and the current too small threshold should be based on the current data and the actual application scenario. In this embodiment, based on the current data statistics of the maximum current value and the non-zero current minimum value, based on the current maximum value*0.8 as the current too large threshold, based on the non-zero current minimum value, if the non-zero current minimum value is less than 18A, then set 18A as the current too small threshold, if the non-zero current minimum value is greater than 18A, then set the non-zero current minimum value*1.2 as the current too small threshold; mark the current too large threshold as Bi1, and the current too small threshold as Bi2; Step S3015, based on the current data, the number of currents greater than the current too large threshold is counted, marked as X1; the number of currents less than the current too small threshold is counted, marked as X2; Step S3016, calculating the Euclidean distance between the A-phase current and the B-phase current based on the current data, marked as L AB , calculate the Euclidean distance between phase A current and phase C current, marked as L AC , calculate the Euclidean distance between the current of phase B and the current of phase C, marked as L BC ; L AB , L AC and L BC The calculation formula is as follows: , where i represents the sampling point of the current data at each moment, M represents the number of the current data sampling points. In this embodiment, within a first time period, M=96, IAi represents the A-phase current size at the i sampling point, IBi represents the B-phase current size at the i sampling point, and ICi represents the C-phase current size at the i sampling point.

[0026] Step S302, performing voltage feature extraction processing based on the voltage data; Step S302 includes the following sub-steps: Step S3021, setting a voltage over-threshold and a voltage under-threshold, marking the voltage over-threshold as Bv1, and the voltage under-threshold as Bv2; the voltage over-threshold and the voltage under-threshold can be based on my country's regulations on low-voltage distribution networks and actual application scenarios. my country's standard voltage for low-voltage distribution networks is 220V, and the normal deviation range is -10% to +7%, which is approximately [198V, 235V]. In this embodiment, the voltage over-threshold is 230V, and the voltage under-threshold is 205V; Step S3022, based on the voltage data, counting the number of voltages greater than the voltage-too-large threshold, marked as X3; counting the number of voltages less than the voltage-too-small threshold, marked as X4; Step S3023, based on the voltage data of phase A, phase B and phase C, respectively calculate the effective values ​​of phase A voltage, phase B voltage and phase C voltage in the first time period; the voltages of phase A, phase B and phase C are all AC voltages, and the effective value of AC voltage refers to the value of this DC voltage when the heat generated by AC power through the same resistor in one cycle is the same as the heat generated by a certain DC voltage in the same time; the effective value of AC voltage is also called root mean square value, and its calculation method is to take the square root of the average value of the sum of the squares of the voltages in one cycle; for example, in this embodiment, the voltage of each phase is an AC voltage with a sinusoidal waveform, and its effective value is the peak value divided by √2, that is, Veffective=Umax / √2, where Umax is the maximum value or amplitude of the voltage.

[0027] Step S3024, marking the effective value of the A phase voltage in the first time period as EA, marking the effective value of the B phase voltage in the first time period as EB, and marking the effective value of the C phase voltage in the first time period as EC; Step S3025, the voltage data of phase A, phase B and phase C are respectively formed into vector sequences with a dimension of m: m 1,U m 2,....,U m N-m+1 , U m i ={V(i), V(i+1), ..., V(i+m-1)}, m can be called the pattern dimension, which is the window length of the time series used to calculate ApEn; m is generally a small positive integer. In this embodiment, m=2, where N is the number of voltage data of phase A, phase B and phase C, i is a positive integer and 1≤i≤N-m+1, V(i) represents the voltage value of the i-th sampling point of a certain phase, U m i Represents the i-th vector sequence; calculate U m i with U m j The Chebyshev distance of m ij , where j is a positive integer and 1≤j≤N-m+1, statistics d m ij The number of similarity tolerance thresholds r is less than or equal to r. When the value of r is large, more information will be lost, and when the value is small, the statistical characteristics cannot be estimated ideally. The general value range is [0.1*std(U), 0.25std(U)], where std is the standard deviation; marked as c i , calculate ci / N-m+1, marked as c i m,r ;d m ij The calculation formula is as follows ; where d m ij For U m i with U m j Chebyshev distance; Step S3026, define Φ m,r c i m,r The average value of the logarithmic representation, Φ m,r for: ; The voltage data of phase A, phase B and phase C are respectively formed into vector sequences with dimension m+1, and then Φ is calculated. m+1,r ; Step S3027, calculate the ApEn of phase A A相 , ApEn of phase B B相 and ApEn of phase C C相 ,ApEn=Φ m,r -Φ m+1,r ;Define F=max(ApEn A相 , ApEn B相 , ApEn C相 ); ApEn is used to measure the irregularity and complexity of the signal sequence. The more complex the time series, the larger the corresponding approximate entropy, and the more regular the time series, the smaller the corresponding approximate entropy; so F represents the complexity of the voltage curve in the three-phase voltage.

[0028] Step S303, performing load feature extraction processing based on active power data and reactive power data; Step S303 includes the following sub-steps: Step S3031, based on the active power data and the reactive power data, calculate the total load power of each phase at each time in the first time period and at the first time interval. The total load power is also called apparent power, which represents the total power that the AC power source can provide, including active power and reactive power. The calculation formula is: , where S 相i represents the total power of phase A, phase B or phase C at sampling point i, P 相i Represents the active power of the corresponding phase, Q 相i Represents the reactive power of the corresponding phase; marked as total power data; Step S3032, calculating the average total load power of the distribution network in the first time period; that is, summing the total load power of each phase at each time and dividing it by the number of acquisition times; marked as S; Step S3032, counting the number of loads exceeding 80% of the normal load upper limit based on the total power data, marked as X5; Step S304, marking the power imbalance data, the total power data, X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, F and S as a second power data set; Step S305, defining a power data feature vector Y based on the second power data set, where Y={X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, S, F}; In the specific implementation process, because the power data is associated with time, the effective information it contains exists not only in the numerical value of the data, but also in the time corresponding to the numerical value. It has time-related characteristics such as trend and periodicity, and belongs to time series data. Time series data is often obtained by multiple sampling within a certain period of time, and its data volume is often very large. For most neural network models, direct input is not feasible, so feature extraction is required to establish a power data feature vector; in the distribution network, three-phase current imbalance, voltage abnormality and distribution network overload are three types of faults that endanger the distribution network. In this embodiment, the power data feature vector is Y={X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, S, F}, where X0, X1, X2, L AB , L AC and L BC It is used to describe the cause of three-phase current unbalance, X3, X4, EA, EB, EC and F are used to describe the cause of voltage abnormality, and X5 and S are used to describe the severity of distribution network overload.

[0029] Step S4, training the self-organizing competitive neural network based on the second power data set and the power data feature vector to construct a power data analysis model; Step S4 includes the following sub-steps: Step S401, normalize the power data feature vector Y to obtain Y0; because some items in the power data feature vector are numerically larger than other items, they will affect the network training and the accuracy of the final model. In this embodiment, Y={X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, S, F}; wherein X0, X1, X2, X3, X4 and X5 are normalized by dividing by the number of sampling moments M, i.e. X0 / M, X1 / M, X2 / M, X3 / M, X4 / M and X5 / M; LAB, LAC and LBC are normalized by dividing by the maximum value among LAB, LAC and LBC; EA, EB and EC are normalized by dividing by the maximum value among EA, EB and EC, and S and F are set to 1.

[0030] Step S402, training the current analysis sub-model: setting 4 current anomaly classifications, including slight current anomaly, relatively serious fault, serious fault, and very serious fault; setting a first learning rate and a first number of training rounds, which can be set according to the actual scenario, and the learning rate value range is (0, 1]. In this embodiment, the first learning rate is 0.01, and the first number of training rounds is 200 rounds; The current weight vector is set. Because the main classification basis of current anomaly classification is the number of points with excessive current unbalance, the number of points with excessive and insufficient current, the Euclidean distance between the three-phase currents, and the waveform complexity of the three-phase voltage, and irrelevant quantities need to be set to zero, so in this embodiment, the current weight vector is H1=(3, 3, 3, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0.5); the number of output layer neurons is set to 4, that is, the current-related data is clustered into 4 categories, namely, slight current anomaly, more serious fault, serious fault and very serious fault; Based on the second power data, Y0 is used as the input of the self-organizing competitive neural network for training. See also Figure 2 As shown in the figure, the self-organizing neural network is an unsupervised learning method with a simple structure, consisting of an input layer and an output layer. The self-organizing competitive neural network detects certain regularities of clustering objects, so the resulting clusters can be regarded as a mapping from basic features to advanced features. For example, current anomalies are mapped through current-related data, and for different types of faults and anomalies, the importance of each element in the input vector is different, so the input neurons are weighted, for example, the current weight vector is H1, and the network structure also changes. After completion, the current analysis sub-model is obtained; the first learning rate and the first number of training rounds can be set according to the actual scenario, and the value range of the first learning rate is (0, 1]. In this embodiment, the first learning rate is 0.01 and the first number of training rounds is 200.

[0031] Step S403, training the voltage analysis sub-model: setting 4 voltage anomaly classifications, including slight voltage anomaly, serious voltage fault, too low voltage and too high voltage; setting the second learning rate and the second number of training rounds, setting the voltage weight vector, setting the output layer neurons to 4, based on the second power data, using Y0 as the input of the self-organizing competitive neural network for training, and obtaining the voltage analysis sub-model after completion; in this embodiment, the second learning rate is 0.01, and the second number of training rounds is 200 rounds; the main classification basis for voltage anomaly classification is the waveform complexity of the three-phase voltage, the number of points where the voltage is too large and too small, and the average value of the three-phase voltage, so in this embodiment, the voltage weight vector is H2=(0, 0, 0, 3, 3, 0, 0, 0, 1, 1, 1, 0, 0, 3); Step S404, training load analysis submodel: setting 3 load abnormality classifications, including slight load abnormality, line overload and line super overload, setting the third learning rate and the third training round number, setting the load weight vector, setting the output layer neurons to 3, based on the second power data, using Y0 as the input of the self-organizing competitive neural network for training, and obtaining the load analysis submodel after completion; the number of points where the voltage is too large and too small and the average value of the three-phase voltage; in this embodiment, the third learning rate is 0.01, and the third training round number is 200 rounds; the main classification of load abnormality classification is based on the load rate average value and the number of points where the load rate exceeds 80%. Since a certain degree of flow imbalance and low voltage may also occur when a load abnormality occurs, the load weight vector in this embodiment is H3=(1, 0, 0, 0, 1, 3, 0, 0, 0, 0, 0, 0, 3, 0); Step S405, see Figure 3 As shown; the current analysis sub-model, the voltage analysis sub-model and the load analysis sub-model are connected in parallel to obtain a power data analysis model; In the specific implementation process, please refer to Figure 4 As shown in the figure, the power data is analyzed by using the power data analysis model. First, the current data and voltage data are collected, and then data calculation and feature extraction are performed, and normalization is performed to obtain the power feature vector Y0. Y0 is input into the power data analysis model, that is, the current analysis sub-model, voltage analysis sub-model and load analysis sub-model are respectively input to obtain the current analysis results, voltage analysis results and load analysis results. According to the classification to which the results belong, the fault situation is obtained.

[0032] Embodiment 2, the present application provides an electronic device, which may include: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus. The memory stores computer-readable instructions, and the processor can call the instructions in the memory. When the computer-readable instructions are executed by the processor, the steps in a method for quickly building a power data analysis model based on big data are executed to achieve the following functions: monitor and collect the current data and voltage data of phase A, phase B and phase C when the distribution network is running; use the current data and voltage data to perform data calculation and processing, and construct a first power data set; use the first power data set to perform feature extraction processing, construct a second power data set, and establish a power data feature vector; train the self-organizing competitive neural network based on the second power data set and the power data feature vector to construct a power data analysis model.

[0033] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0034] Embodiment 3. The present application also provides a computer-readable storage medium. The present application provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps in the method for quickly constructing a power data analysis model based on big data are executed to achieve the following functions: monitor and collect current data and voltage data of phase A, phase B, and phase C when the distribution network is running; use current data and voltage data to perform data calculation and processing, and construct a first power data set; use the first power data set to perform feature extraction processing, construct a second power data set, and establish a power data feature vector; train a self-organizing competitive neural network based on the second power data set and the power data feature vector to construct a power data analysis model.

[0035] Through the description of the above implementation methods, the embodiments of the present invention can be provided as methods, systems or computer program products. Based on such an understanding, the above technical solutions can be essentially or partly contributed to the prior art in the form of software products, which can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and include several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0036] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or units can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of systems, modules and units can be electrical, mechanical or other forms.

[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for quickly constructing a power data analysis model based on big data, characterized in that: The steps include: Monitor and collect current and voltage data of phase A, phase B and phase C during distribution network operation; Performing data calculation and processing using the current data and the voltage data, and constructing a first power data set; Performing feature extraction processing using the first power data set, constructing a second power data set, and establishing a power data feature vector; The self-organizing competitive neural network is trained based on the second power data set and the power data feature vector to construct a power data analysis model.

2. According to a method for rapidly constructing a power data analysis model based on big data according to claim 1, it is characterized in that: Monitoring and collecting the current and voltage data of phase A, phase B, and phase C during the operation of the distribution network includes the following sub-steps: Collecting voltage data: collecting phase A voltage information, phase B voltage information, and phase C voltage information of the distribution network at first time intervals within a first time period, and marking the phase A voltage information, phase B voltage information, and phase C voltage information as voltage data; Collect current data: collect the A-phase current information, B-phase voltage and current, and C-phase current information of the distribution network at a first time interval within a first time period, and mark the A-phase current information, B-phase voltage and current, and C-phase current information as current data; the first time period is T, and the first time interval is d.

3. According to a method for rapidly constructing a power data analysis model based on big data according to claim 2, it is characterized in that: The data calculation process using the current data and the voltage data includes the following sub-steps: Calculate active power: Based on the current data and voltage data, use the active power calculation formula to calculate the active power of phase A, phase B, and phase C respectively. The active power calculation formula is as follows: , where P represents the active power of any phase circuit among phase A, phase B and phase C, U 相 and I 相 represents the voltage and current values ​​of the corresponding phase circuit, and φ represents the phase difference between the voltage and current of the corresponding phase circuit; The active power of phase A, the active power of phase B and the active power of phase C are marked as active power data; Calculate reactive power: Based on the current data and voltage data, use the reactive power calculation formula to calculate the reactive power of phase A, phase B, and phase C respectively. The reactive power calculation formula is as follows: , where Q represents the active power of any phase circuit among phase A, phase B and phase C, U 相 and I 相 represents the voltage and current values ​​of the corresponding phase circuit, and φ represents the phase difference between the voltage and current of the corresponding phase circuit; The A-phase reactive power, the B-phase reactive power, and the C-phase reactive power are marked as reactive power data.

4. According to claim 3, a method for rapidly constructing a power data analysis model based on big data is characterized in that: Constructing the first power data set includes the following sub-steps: The current data, voltage data, active power data and reactive power data are classified and stored according to the associated phase A, phase B or phase C. Each phase in phase A, phase B and phase C contains the current data, voltage data, active power data and reactive power data of the phase. After completion, the first power data set is obtained.

5. According to the method for rapidly constructing a power data analysis model based on big data according to claim 4, it is characterized in that: The feature extraction process using the first power data set includes the following sub-steps: Current feature extraction processing: Based on the current data of phase A, phase B, and phase C, the maximum current and the minimum current of phase A, phase B, and phase C at the same time are screened and extracted, and the maximum current is marked as I max , the minimum current is marked as I min ; The current imbalance degree is calculated based on the maximum current and the minimum current using the current imbalance degree formula. The current imbalance degree formula is as follows: , where I un Represents the current unbalance, I max Represents the maximum current value among the current data of phase A, phase B, and phase C, I min represents the minimum current value among the current data of phase A, phase B and phase C; Obtaining the current imbalance at each moment at a first time interval within a first time period, and marking it as current imbalance data; The current imbalance threshold is set based on the current imbalance data, marked as I th , based on the current imbalance data, the number of current imbalances greater than the current imbalance threshold is counted, marked as X0; Based on the current data, set the current over-threshold and the current under-threshold, mark the current over-threshold as Bi1, and the current under-threshold as Bi2; The number of currents greater than the current excessive threshold value based on the current data is marked as X1; Count the number of currents that are less than the current too small threshold, marked as X2; Calculate the Euclidean distance between phase A current and phase B current based on the current data, marked as L AB , calculate the Euclidean distance between phase A current and phase C current, marked as L AC , calculate the Euclidean distance between the current of phase B and the current of phase C, marked as L BC ; Then, voltage feature extraction is performed based on the voltage data.

6. The method for rapidly constructing a power data analysis model based on big data according to claim 5, characterized in that: The voltage feature extraction process based on voltage data includes the following sub-steps: Set a voltage over-limit threshold and a voltage under-limit threshold, mark the voltage over-limit threshold as Bv1, and mark the voltage under-limit threshold as Bv2; count the number of voltages greater than the voltage over-limit threshold based on voltage data, and mark them as X3; Count the number of voltages that are less than the voltage threshold, marked as X4; Based on the voltage data of phase A, phase B and phase C, respectively calculate the effective values ​​of phase A voltage, phase B voltage and phase C voltage in the first time period; mark the effective value of phase A voltage in the first time period as EA, the effective value of phase B voltage in the first time period as EB, and the effective value of phase C voltage in the first time period as EC; The voltage data of phase A, phase B and phase C are respectively composed into vector sequences with dimension m: m 1,U m 2,....,U m N-m+1 , U m i ={V(i), V(i+1), ..., V(i+m-1)}, where N is the number of voltage data of phase A, phase B and phase C, i is a positive integer and 1≤i≤N-m+1, V(i) represents the voltage value of the i-th sampling point of a phase, U m i Represents the i-th vector sequence; calculate U m i with U m j The Chebyshev distance of m ij , where j is a positive integer and 1≤j≤N-m+1, statistics d m ij The number of values ​​less than or equal to r is marked as c i , calculate ci / N-m+1, marked as c i m,r , define Φ m,r c i m,r The average value of the logarithmic representation, Φ m,r for: ; The voltage data of phase A, phase B and phase C are respectively formed into vector sequences with dimension m+1, and then Φ is calculated. m+1,r , calculate the ApEn of phase A A相 , ApEn of phase B B相 and ApEn of phase C C相 ,ApEn=Φ m,r -Φ m+1,r ; Define F = max (ApEn A相 , ApEn B相 , ApEn C相 ); Load feature extraction and processing are performed based on active power data and reactive power data.

7. The method for rapidly constructing a power data analysis model based on big data according to claim 6, characterized in that: Load feature extraction based on active power data and reactive power data includes the following sub-steps: Based on the active power data and the reactive power data, the total load power of each phase at each time instant in the first time period is calculated, which is marked as total power data; the average total load power of the distribution network in the first time period is calculated, which is marked as S; Based on the total power data, the number of loads that exceed 80% of the normal load upper limit is counted and marked as X5.

8. The method for rapidly constructing a power data analysis model based on big data according to claim 7, characterized in that: Constructing the second power data set and establishing the power data feature vector includes the following sub-steps: Mark the power imbalance data, the total power data, X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, F and S as a second power data set; Based on the second power data set, the power data feature vector is defined as Y, where Y={X0, X1, X2, X3, X4, X5, LAB, LAC, LBC, EA, EB, EC, S, F}.

9. The method for rapidly constructing a power data analysis model based on big data according to claim 8, characterized in that: Training the self-organizing competitive neural network based on the second power data set and the power data feature vector to construct a power data analysis model includes the following sub-steps: Normalize the power data feature vector Y to obtain Y0; Training the current analysis sub-model: Set four current anomaly classifications, including slight current anomaly, relatively serious fault, serious fault, and very serious fault; Set the first learning rate and the first number of training rounds, set the current weight vector, set the output layer neurons to 4, and train Y0 as the input of the self-organizing competitive neural network based on the second power data. After completion, the current analysis sub-model is obtained; Training the voltage analysis sub-model: set 4 voltage anomaly classifications, including slight voltage anomaly, serious voltage fault, low voltage and high voltage; set the second learning rate and the second number of training rounds, set the voltage weight vector, set the output layer neurons to 4, and train Y0 as the input of the self-organizing competitive neural network based on the second power data. After completion, the voltage analysis sub-model is obtained.

10. The method for rapidly constructing a power data analysis model based on big data according to claim 9, characterized in that: The self-organizing competitive neural network is trained based on the second power data set and the power data feature vector, and the construction of the power data analysis model also includes the following sub-steps: Training the load analysis sub-model: set three load anomaly classifications, including slight load anomaly, line overload, and line super overload, set the third learning rate and the third number of training rounds, set the load weight vector, set the output layer neurons to 3, and train Y0 as the input of the self-organizing competitive neural network based on the second power data. After completion, the load analysis sub-model is obtained; The current analysis sub-model, the voltage analysis sub-model and the load analysis sub-model are connected in parallel to obtain a power data analysis model.

Citation Information

Patent Citations

  • Power distribution network fault clustering analysis method based on self-organizing competitive network

    CN115169441A