CN112036553A

CN112036553A - Non-signal injection type user-phase topological relation identification method based on genetic algorithm

Info

Publication number: CN112036553A
Application number: CN202011126494.XA
Authority: CN
Inventors: 徐文; 孙大璟; 唐明群; 葛善虎; 高尚源
Original assignee: Jiangsu Denang Electric Power Design Consulting Co ltd; Jiangsu Qihou Intelligent Electrical Equipment Co ltd
Current assignee: Jiangsu Denang Electric Power Design Consulting Co ltd; Jiangsu Qihou Intelligent Electrical Equipment Co ltd
Priority date: 2020-10-20
Filing date: 2020-10-20
Publication date: 2020-12-04
Anticipated expiration: 2040-10-20
Also published as: CN112036553B

Abstract

The invention discloses a non-signal injection type user-phase topological relation identification method based on a genetic algorithm, which comprises the following steps: obtaining effective data after data cleaning according to the electric quantity data information; randomly generating an initialized chromosome population data matrix Pop _ data according to the valid data; entering a circulating process: calling a fitness function; calling a selection function, a cross function and a variation function of the genetic algorithm to obtain updated NewPopdata; entering a reset function call of a genetic algorithm; and obtaining the final NewPopdata, and obtaining a judgment result Best _ Pop of each row corresponding to each user table through the mapping relation. The invention can solve the problem of low-voltage distribution area topology identification by utilizing the existing adoption system data, namely, other equipment is not required to be installed.

Description

Non-signal injection type user-phase topological relation identification method based on genetic algorithm

Technical Field

The invention relates to the technical field of power distribution network topology identification, in particular to a non-signal injection type user-phase topological relation identification method based on a genetic algorithm.

Background

At present, the identification of the home-phase topological relation of a low-voltage station area is mainly a line checking method: for power supply of users in a transformer area, opening each user meter box, checking the specific wiring mode of each user meter, determining the specific use, and then recording and marking, the method is characterized in that: the workload is large, the efficiency is low, and once a user changes or increases the capacity, the mark is needed again.

The existing distribution area low-voltage distribution topology is basically blank or needs manual participation, is complex in work and long in time, cannot be automatically updated and identified, and a self-adaptive distribution area low-voltage distribution network topology identification method is urgently needed to solve the problems.

Meanwhile, the utilization and acquisition system of the low-voltage transformer area is widely applied nationwide, and the system acquires the electric energy data of the low-voltage side outgoing line multifunctional meter and the household meter of the jurisdiction of the distribution transformer area, and comprises the following steps: voltage, current, active, reactive, and wattage measurements, etc.; although the data acquisition mode, the communication mode, the management architecture and the like are slightly different, the massive data are accumulated day by day, and the possibility is undoubtedly provided for the intelligent analysis of the big data.

Disclosure of Invention

The purpose of the invention is as follows: the invention aims to provide a non-signal injection type user-phase topological relation recognition method based on a genetic algorithm, which is based on a large amount of data accumulated by a system, automatically deduces a low-voltage topological connection relation of a distribution area by adopting an artificial intelligent Genetic Algorithm (GA), automatically generates a user-phase topological relation, can achieve an accurate recognition rate not lower than 97% according to two months of data accumulated by the existing user acquisition system, and basically meets the commercial requirement.

The technical scheme is as follows: the invention discloses a low-voltage distribution network topology identification method based on the cooperation of a unilateral optimization algorithm and a genetic algorithm, which comprises the following steps:

(1) obtaining effective data after data cleaning according to the electric quantity data information;

(2) randomly generating an initialized chromosome population data matrix Pop _ data according to the valid data;

(3) entering a circulating process:

(4) calling a Fitness function, returning the proximity degree delta E between the distribution side and the corresponding user table in each chromosome, and converting the delta E into Fitness for measuring the proximity degree in each chromosome;

(5) entering selection function, cross function and mutation function call of a genetic algorithm to obtain updated NewPopdata, which contains Pop _ num chromosomes with preferred selection results;

(6) setting a premature condition, entering a reset function call of a genetic algorithm, entering the step (2) if the premature condition is established, and otherwise, continuing the cycle process;

(7) and obtaining the final NewPopdata, and obtaining a judgment result Best _ Pop of each row corresponding to each user table through the mapping relation.

Further, the electric quantity data is any one of current, voltage, reactive power, electric energy and active power, and the data cleaning method comprises the steps of eliminating a full 0 household meter set, eliminating a weak value household meter set and eliminating an invalid time sample space set in an effective household meter set; screening the cleaned data comprises acquiring that the electrical quantity data of the outgoing line side of the transformer in the transformer area has P_a，P_b，P_c，P_Total-changeThe data for acquiring the electric meter sample at the user side has P_a，P_b，P_c，P_{General household}Obtaining the P of the outgoing line of the distribution transformer_Total-changeAnd accumulating and calculating an error delta P corresponding to the user meter P at the same sampling time, setting parameters, keeping the distribution transformer and the user meter sampling data at the same time when the delta P is less than or equal to the sampling time, adding the data into an effective data sample set, otherwise giving up the data, and meeting the number K of the effective time:

≤10％，K≥300；

≤15％，K≥400；

≤20％，K≥600；

≤25％，K≥1200；

≤35％，K≥2000。

further, in the step (1), effective data after data cleaning is obtained according to the electrical quantity data information, a storage list ValidDataSet is returned, each qualified sample includes a distribution transformer of a distribution area and a combined set sample set of three-phase and single-phase electrical quantities of a user, m _ tnum in the storage list ValidDataSet represents the number of qualified user tables, Validcount _ num represents the number of qualified data samples, and a total electrical quantity matrix E of a distribution transformer side is obtained according to the data of the storage list ValidDataSet_{Distribution transformer assembly}[j]Distribution side A, B, C single-phase electric quantity matrix E_1-a[j]、E_1-b[j]、E_1-b[j]And meter-level single-phase electric quantity matrix

Randomly generating an M multiplied by N initialization chromosome population data matrix Pop _ data, wherein M is the number of constructed chromosome initial populations, N is 3 multiplied by M _ tnum, 3 represents the discrimination result of only A, B, C phases of the user table, and randomly generating the chromosome population according to M _ tnum and Pop _ num

Further, the fitness function constructing process is as follows:

calculating the change rate Delta E of the A/B/C three phases of the distribution transformer of the transformer area_1-a[i]、ΔE_1-b[i]、ΔE_1-c[i]Calculating the change rate Delta E of the user table P_1-a1、ΔE_1-b1、ΔE_1-c1：

Obtaining Pop _ data _ A [ i, j ], Pop _ data _ B [ i, j ] and Pop _ data _ C [ i, j ] matrixes;

to obtain E_1-a1Rate of change matrix, E_1-b1Rate of change matrix, E_1-c1A rate of change matrix;

calculating the cumulative sum value delta E of the distribution transformation change rate and the absolute value of the corresponding row user table change rate_a，ΔE_b，ΔE_c；

Proximity Δ E between the distribution and the corresponding user table in each chromosome₁＝ΔE_a+ΔE_b+ΔE_c；

Calculate Fit [ p ]]＝ΔE₁[p]，p＝1，2，…Pop_num-1，Fit_max＝max(Fit[p])，

Fitness[P]＝Fit_max-Fit[P]，P＝1，2，…，Pop_num-1。

Further, the selection function construction method comprises

Fitness normalization processing:

P_fitness[j]＝Fitness[j]/Sumfit_ness，j＝0，1，…，Pop_num-1；

storing Fit _ max5[ w ] arrays on the chromosomes of the top five in the ranking according to the 5 corresponding larger chromosomes of the P _ fitness [ j ] from big to small;

randomly selecting ms which belongs to a value of [0, 1] and meets the condition: ms is less than or equal to max (P _ fixness [ j ]), j is 0,1,2, …, Pop _ num-1.

Further, the cross algorithm is as follows: selecting an operation random number c _ rand E (0, 1) and cross probability factors pc and pc E (0.8, 1), if c _ rand is more than or equal to pc, performing no cross operation on the two groups of chromosomes, and keeping the original gene of the chromosome group unchanged; if c _ rand < pc is satisfied, the chromosome set is subject to crossover operation.

Further, the mutation algorithm is:

1) judging whether mutation operation is needed: when the random number count _ rand is larger than or equal to pm, the count _ rand belongs to (0, 1) and does not operate the mutation, and when the count _ rand is smaller than pm, the next operation is continued;

2) selecting a random integer t₁Rand is not more than Poplenge-1, and

rounding down when t ₁0 denotes the gene to which phase A belongs, when t is₁1-1 represents a gene to which phase B belongs, when t is₁1-2 represents a gene belonging to the C-phase, and t is₁The value of the _randlocus gene is put in a variable t _ sample _ 1;

3) taking the second random number t₂_rand≤Poplenge-1，t₂_rand≠t₁A ran d, and

rounding down when t ₂0 denotes the gene to which phase A belongs, when t is₂1-1 represents a gene to which phase B belongs, when t is₂1-2 represents a gene belonging to the C-phase, and t is₂The value of the _randlocus gene is put in a variable t _ sample _ 2;

4) the distinguishing process comprises the following steps:

4.1) if t _ sample _1+ t _ sample _2 is 0, the gene bit to be exchanged belongs to an invalid bit, no mutation operation is carried out, and the step 3) is returned to continue random selection;

4.2) if t _ sample _1+ t _ sample _2 ≠ 0 and t₁_1≠t₂And (1) exchanging the positions of the gene locus values to be exchanged in the different phase regions, storing the positions into corresponding arrays, and finishing the chromosome mutation operation.

Further, when the loop body is executed for no more than half times, the accumulated value exceeds the preset reselection rate, the calculation of the genetic algorithm is carried out again through a reselection mechanism: sorting the values of the Pop _ data genes from large to small; the maximum number of the same gene values was counted and recorded as num.

Furthermore, each user table corresponds to the discrimination result of each row of Best _ Pop [ m _ num ] [6], and the discrimination result comprises a 0 th bit representing the user table number; 1/2/3, which respectively represent phase A, phase B and phase C; the 2 nd bit indicates the reliability of discrimination as 0/1, which indicates unreliability/credibility, respectively; the 3 rd, 4 th and 5 th positions respectively represent the number of the A phase, the B phase and the C phase in the group

Has the advantages that: compared with the prior art, the invention has the advantages that: the artificial intelligence algorithm based on the electrical characteristics has certain universality, the principle and the method can be popularized to the fields of transmission, transformation and power distribution, lay a foundation for the analysis of large data of a power grid and the deep application of artificial intelligence, can be widely applied to the field of electric power ubiquitous Internet of things, and bring considerable social and economic benefits. As follows:

1. the problem of low-voltage distribution area topology identification can be solved by utilizing the existing acquisition system data, namely, other equipment does not need to be installed, and people do not need to be sent to a site for actual measurement and investigation, so that only a little is cost saved. Meanwhile, the efficiency is greatly improved, and the operation management level is improved.

2. The operation condition of the platform area can be effectively monitored, and more reliable basis and means are provided for monitoring, analyzing and managing low-voltage three-phase unbalance.

3. According to the data collected by the user table, deep analysis of the user load characteristics is performed, and auxiliary reference is provided for tracking the load dynamic operation characteristics.

Drawings

FIG. 1 is a schematic diagram of a mining system;

FIG. 2 is a functional block diagram of the present invention;

FIG. 3 is a flow chart of a method implementation of the present invention;

FIG. 4 is a flow chart of the general control function in the present invention;

FIG. 5 is a flow chart of the cumulative number of times a user belongs to an A/B/C user;

FIG. 6 is a flow chart of statistical analysis for judging whether 10 subscriber lists belong to a distribution area;

FIG. 7 is a flow chart of confidence determination;

FIG. 8 is a schematic diagram of a chromosome set to be crossed;

FIG. 9 is a schematic diagram of the crossing rule of the i chromosome to the i +1 chromosome;

FIG. 10 is a schematic diagram of the crossing rule of i +1 chromosome to i chromosome.

Detailed Description

The technical solution of the present invention is described in detail below with reference to the accompanying drawings, but the scope of the present invention is not limited to the embodiments.

The invention is based on a large amount of data accumulated by the acquisition system, adopts artificial intelligent Genetic Algorithm (GA), automatically deduces the low-voltage topological connection relation of the distribution area, automatically generates the user-phase topological relation, and can achieve the accurate recognition rate not lower than 97 percent according to the two months of data accumulated by the existing user acquisition system, thereby basically meeting the commercial requirement.

The genetic algorithm is a self-adaptive global optimization probability search algorithm formed by simulating the genetic and evolutionary processes of organisms in natural environment, and mainly comprises 3 processes of selection, intersection and variation.

The mining system as shown in fig. 1 comprises:

(1) low-pressure side: refers to a low voltage user intelligent device comprising: household meters, leakage protection devices, low-voltage capacitors, SVG/SVC and other automatic devices of residential users; the sampled information includes: voltage, current, active, reactive, and wattage measurements, etc.; a variety of communication means may be employed to communicate with the cell concentrator, such as: LORA \ broadband carrier \ RS485\ network \ narrowband carrier, etc.; the sampling time interval can be different from 1 point/5 min, 1 point/15 min, 1 point/60 min and 1 point/day according to different communication modes and requirements.

(2) A platform area side: the system refers to automation equipment such as a platform area acquisition concentrator, a distribution transformer low-voltage outlet multifunctional ammeter and the like; the sampled information includes: voltage, current, active, reactive, and wattage measurements, etc.; the master station communicates with the mobile station by means of a public wireless network.

(3) The main station side: the system is a master station system of the acquisition system and is used for acquiring and managing uploaded data for analysis and management.

The method comprises the steps of installing a set of low-voltage distribution area topological intelligent analysis software on a master station system side, reading historical data stored by a sampling system through forward physical isolation equipment, analyzing by adopting an artificial intelligence algorithm, judging and reasoning the family-to-variation and family-to-phase topological relation of related distribution areas, and carrying out graph-model display according to a unified distribution area information access model defined based on the IEC 61850 standard. The functional structure is shown in fig. 2.

The invention installs a set of low-voltage distribution area topological intelligent analysis software, analyzes by adopting an artificial intelligence algorithm through the acquired data information, judges and infers the family-phase topological relation of the relevant distribution area, and performs graph-model integrated display.

Data selection, data cleaning and data screening method based on electric quantity trend self-adaptive adjustment

In order to improve the convergence stability and the discrimination accuracy of the algorithm, special processing needs to be performed on the acquired data of the existing active power, and the uniqueness processing is mainly embodied in the following aspects.

1. The data selection method comprises the following steps:

in electrical quantity multidimensional data (current, voltage, reactive power and electric energy), active power is taken as an example of a research object of a data sample, on one hand, the change rate characteristic of the data is more suitable for identification of a genetic algorithm than other data, and the precision is 5% -10% better than that of other data.

2. The data cleaning method comprises the following steps:

three pretreatment modes are mainly adopted:

firstly, removing all 0 user table sets: for a sample with a sampling value of 0, indicating that the user is not powered up in the period of the homologous data, the user is rejected.

Secondly, rejecting weak value user table sets: the user is rejected for a sampled value exceeding 1/3 of 0 or a data value relatively small during the whole period.

Thirdly, eliminating sample space sets at invalid time in the valid user table set: in a plurality of time sample spaces, if no index exists at a certain time of a certain user table, all the time data corresponding to all the corresponding valid user tables are removed.

3. The data screening method comprises the following steps:

the specific screening steps are as follows (taking electric power P as research content):

1) the electrical quantity data of the outgoing line side of the transformer in the transformer area is P_a,P_b,P_c,P_Total-changeEtc., generally 1 point/15 minutes, etc；

2) The data sampled by the user side ammeter has P_a,P_b,P_c,P_{General household}Etc. since the user-side meter is generally a single-phase meter and does not know the specific acquisition phase, only P is generally available_{General household}Waiting for data, representing single-phase data;

3) p for taking outgoing line of distribution transformer_Total-changeAnd accumulating with a user table P corresponding to the same sampling time:

calculated error Δ P ═ Abs ((P)_Total-change-P_{General household})/P_Total-change)×100；

Setting parameters, and when the delta P is less than or equal to the delta P, reserving the distribution transformer and the user electric meter sampling data at the moment and adding the data into an effective data sample set; otherwise, the data is regarded as bad data, and the time sampling data is discarded.

4) Based on the last step, once the value of the data starting time period sum is determined, the amount of the screened time point data is determined accordingly. Because the values of the subsequent intelligent reasoning algorithm for participating in calculation are related, the following steps are mainly adopted: the data volume of the effective time point is required, so an adaptive selection mode needs to be adopted here, that is, the data volume K of the effective time point needs to satisfy: the value of K is related to the following steps:

≤10％,K≥300；

≤15％,K≥400；

≤20％,K≥600；

≤25％,K≥1200；

≤35％,K≥2000。

with the control accuracy of automatic adjustment, the conditions that the data amount K satisfies are different, and once the data preparation conditions are satisfied, the intelligent recognition system can be started. The processed data samples eliminate data mutation burrs, eliminate bad data, effectively eliminate the influence of sampling abnormal points on algorithm convergence, and the processed sample data can be applied to intelligent reasoning calculation.

Random population construction method based on electrical node sample mapping

A combined set sample set of distribution transformation and three-phase and user single-phase P, wherein the user single-phase sample set is put into P_{General assembly}Returning to a memory list for ValidDataSet:

decomposing to obtain a single-phase active one-dimensional matrix on the distribution transformer side:

the distribution transformer A/B/C phase total active one-dimensional matrix is as follows:

for the total active power of the distribution transformer side,

wherein j is 0,1, 2., Validcount _ num-1; j, j + 3.., 4 Validcount _ num-3;

distribution transformation A phase active one-dimensional matrix:

the phase A of the distribution transformer side has active power,

where j is 0,1, 2., Validcount _ num-1; j +1, j + 4.., 4 Validcount _ num-2;

and (3) distribution transformation B-phase active one-dimensional matrix:

the active power of the B phase at the side of the distribution transformer,

where j is 0,1, 2., Validcount _ num-1; j +2, j +5, 4 Validcount _ num-1;

the distribution transformation C phase active one-dimensional matrix:

the active power is provided for the C phase at the distribution transformer side,

where j is 0,1,2, …, Validcount _ num-1; j +3, j +6, … 4, Validcount _ num;

single-phase active matrix of family table level:

wherein i is 0,1,2,. m _ num-1; j ═ 0,1, 2., Validcount _ num-1; m ═ i; n j, j + 3.., 4 Validcount _ num-3;

randomly generating an initial chromosome string structure data matrix, each string structure data is called a chromosome, and M chromosomes form a population:

wherein M is the number of chromosomes in the structure, and N is the length of each chromosome

For a user-phase topology discrimination algorithm, the specific construction mode is as follows: m is the number of the initial population of the constructed chromosomes, and generally takes the value M as 100, N is the length of each chromosome, N as 3 × M _ num, 3 represents that the user table has only A, B, C phase discrimination possibility, the station areas are all performed under a specific single station area, and M _ num is the total number of the user tables.

The uniqueness operation process comprises the following steps:

1) initial chromosome placement population

2) i is 0; i + +; each chromosome in the i.ltoreq. (M-1)//. clan population is to be cycled to;

3) j is 0; j + +; j is less than or equal to (N-1)///' each gene of each chromosome is required to be counted;

4) taking a random integer 1 not less than t _ num not more than N-1, when the ith gene of t _ num in the Pop _ data [ i, j ] is unique (non-zero values of each chromosome are not repeated), then: pop _ data [ i, j ] ═ i, t _ num ];

5) return 3);

6) return 4);

7) random chromosome populations were produced.

Finally, each single-phase user meter can be in

The unique corresponding home position can be found in each chromosome, the specific position is randomly generated at first, and finally after GA algorithm iteration, the position is corresponding to a certain phase position in the A/B/C three phases.

And skillfully determining a certain phase in the user table and the corresponding A/B/C phase uniquely through a random number generation mode.

Construction method based on electric quantity trend deviation fitness function

The fitness function construction of the chromosome is the most important ring in the encoding process, and the fitness indicates the superiority and inferiority of an individual or a solution. Different problems, the fitness function is defined differently.

For the user-phase, according to the coding mode, each m _ num cis-position gene position represents an A/B/C user access condition, so that the whole chromosome is sequentially divided into 3 partitions (if the relation is a user-phase change relation, the number of the partitions is determined by the number of the station areas), the power of the users of each partition is calculated according to the user numbers on the gene positions in the partitions, and then the power error of each partition is obtained by subtracting the total power of the collected known partitions, and is recorded as err. Since the present invention seeks a chromosome when err is 0 and the value of fitness is proportional to the quality level of the chromosome, the fitness function fitness max (err) -err is constructed so that the value of the fitness function coincides with the direction of evolution.

The fitness function is constructed as follows:

1. the change rate of the three phases of A/B/C of the distribution transformer of the transformer area is as follows:

distribution transformation A phase active one-dimensional matrix:

the phase A of the distribution transformer side has active power,

where j is 0,1, 2., Validcount _ num-1; j +1, j + 4.., 4 Validcount _ num-2;

rate of change of distribution phase a:

for the active change rate of the phase A at the distribution side,

where i ═ 0,1, 2., Validcount _ num-2;

and (3) distribution transformation B-phase active one-dimensional matrix:

the active power of the B phase at the side of the distribution transformer,

where j is 0,1, 2., Validcount _ num-1; j +2, j +5, 4 Validcount _ num-1;

rate of change of distribution phase B:

for the active change rate of the B phase at the distribution side,

where i ═ 0,1, 2., Validcount _ num-2;

the distribution transformation C phase active one-dimensional matrix:

where j is 0,1, 2., Validcount _ num-1; j +3, j +6, 4 Validcount _ num;

rate of change of distribution phase C:

for the active change rate of the phase C at the distribution side,

where i ═ 0,1, 2., Validcount _ num-2;

2. subscriber table P rate of change Δ E_1-a1、ΔE_1-b1、ΔE_1-c1：

Single-phase active matrix of family table level:

obtaining: pop _ data _ A [ i, j ] matrix:

because the length of the Pop _ data is 3 times of the length of A/B/C, the front corresponds to the A phase, the middle corresponds to the B phase and the final corresponds to the C phase. The corresponding A phase is processed according to the effective gene code m _ num, and the A phase is set to zero when exceeding the gene code, and the A phase is reserved.

if b_ij≤m_num，c_ij＝b_ij，else c_ij＝0，

i∈[1，Pop_num]，j∈[1，validcount_num]

Obtaining: pop _ data _ B [ i, j ] matrix:

because the length of the Pop _ data is 3 times of the length of A/B/C, the front corresponds to the A phase, the middle corresponds to the B phase and the final corresponds to the C phase. The corresponding B phase is processed according to the effective gene code m _ num, and the position exceeding the gene code is set to zero, and the remained position is reserved.

if b_ij≤m_num，d_ij＝b_ij，else d_ij＝0，

i∈[1，Pop_num]，j∈[validcount num，2*validcount_num-1]

3. Obtaining: pop _ data _ C [ i, j ] matrix:

because the length of the Pop _ data is 3 times of the length of A/B/C, the front corresponds to the A phase, the middle corresponds to the B phase and the final corresponds to the C phase. The corresponding C phase is processed according to the effective gene code (m _ num), and the excess gene code is set to zero, and the remained C phase is reserved.

if b_ij≤m_num，e_ij＝b_ij，else e_ij＝0，

i∈[1，Pop_num]，j∈[2*validcount_num，3*validcount_num-1]

4. Obtaining:

the transformation matrix process is as follows:

initialization:

taking each corresponding non-zero t _ num in the Pop _ data _ A as an effective user table, and taking t _ num as E_φ[]And accumulating the sampling points of the corresponding columns of the corresponding index rows.

if c_ij≠0，i∈[0，m_num-1]，j∈[0，validcount_num-1]

Taking each corresponding non-zero line t _ num in the Pop _ data _ B as an effective user table, and taking t _ num as E_φ[]And accumulating the sampling points of the corresponding columns of the corresponding index rows.

Taking each corresponding non-zero line t _ num in the Pop _ data _ C as an effective user table, and taking t _ num as E_φ[]And accumulating the sampling points of the corresponding columns of the corresponding index rows.

5. Obtaining: e_1-a1The rate of change matrix (one-dimensional reduction compared to the pre-transform matrix) procedure is as follows:

subtracting the previous term from the next term of the matrix and dividing by the next term

6. Obtaining: e_1-b1The rate of change matrix (one-dimensional reduction compared to the pre-transform matrix) procedure is as follows:

7. Obtaining: e_1-c1The rate of change matrix (one-dimensional reduction compared to the pre-transform matrix) procedure is as follows: subtracting the previous term from the next term of the matrix and dividing by the next term

8. Calculating Delta E_a，ΔE_b，ΔE_c：

For Delta E_aIn other words, the method is a one-dimensional array composed of Pop _ num numerical values, each numerical value is an accumulated sum value of the distribution change rate and the absolute value of the change rate of the corresponding row user table, and the calculation formula is as follows:

a is a multiplication factor, generally, A is 1, m is equal to [0, Pop _ num-1], ABS represents an absolute value

For Delta E_bIn other words, the method is a one-dimensional array composed of Pop _ num numerical values, each numerical value is an accumulated sum value of the distribution change rate and the absolute value of the change rate of the corresponding row user table, and the calculation formula is as follows:

a is a multiplication factor, typically A is 1, m ∈ [0, Pop _ num-1]

For Delta E_cIn other words, the method is a one-dimensional array composed of Pop _ num numerical values, each numerical value is an accumulated sum value of the distribution change rate and the absolute value of the change rate of the corresponding row user table, and the calculation formula is as follows:

a is a multiplication factor, typically A is 1, m ∈ [0, Pop _ num-1]

9. Calculating Delta E₁：

ΔE₁The closeness degree between the distribution transformation and the corresponding user table in each chromosome is reflected, the smaller the value is, the closer the value is to the true value, and the calculation formula is as follows:

ΔE₁＝ΔE_a+ΔE_b+ΔE_c

10. will be Delta E₁Conversion to Fitness in each chromosome, which measures how close together:

take Fit [ p ]]＝ΔE₁[p]，p＝1，2，…Pop_num-1

Calculate the Fit _ max maximum: fit _ max ═ max (Fit [ p ])

Fitness[P]＝Fit_max-Fit[P]，P＝1，2，…，Pop_num-1

Fitness and Δ E₁The relationship is transformed such that a greater Fitness value in each subsequent chromosome indicates that the corresponding chromosome set is closer to the actual true value of the distribution.

Selection function construction method based on 'elite' preferred mechanism

And (4) carrying out primary 'elite' strategy screening on chromosomes in the population according to the calculation result of the Fitness Fitness. The elite strategy is to construct a probability interval based on fitness, wherein the higher fitness is selected for multiple times as a member of the next generation population, and chromosomes with lower fitness are eliminated, and the elite strategy comprises the following specific steps:

4.1) normalization processing:

fitness normalization processing procedure:

P_fitness[j]＝Fitness[j]/Sumfit_ness，j＝0，1，…Pop_num-1

4.2) chromosome preservation of top five of the ranking

Placing the corresponding 5 larger chromosomes in the order of P _ fitness [ j ] from large to small: in the array of Fit _ max5[ w ], w is 0,1,2, 3, 4, j is 0,1,2, …, Pop _ num-1;

4.3) "Elite" strategy screening construction method:

sorting the Pop _ data according to P _ fixness [ j ], j-0, 1,2, … and Pop _ num-1 from small to large, and correspondingly obtaining D _ Popdata [ m ] [ n ], m-0, 1,2 … and Pop _ num-1; n-0, 1,2, …, poplene-1;

and randomly selecting ms which belongs to the value of [0, 1], and in order to meet the requirement that each selection is successful and effective, reducing the range of random selection and meeting the condition: and the ms is less than or equal to max (P _ fixness [ j ]), j is 0,1,2, … and Pop _ num-1, and the Pop _ num is selected for a total number of times.

If when the probability ms is between j and j +1 chromosomes, the j +1 chromosome corresponding to D _ Popdata is selected, i.e.:

P_fitness[j]≤ms≤P_fitness[j+1]，j＝0，1，2，…，Pop_num-1；

and storing the j +1 chromosome corresponding to the selected D _ Popdata into NewPopdata, and generating a new array NewPopdata after Pop _ num times of ms selection, wherein the new array NewPopdata contains Pop _ num chromosomes with more preferable selection results.

Searching whether New Popdata contains Fit _ max5[ w ], w is 0,1,2, 3, 4, if not, P _ fitness [ j ], j is 0,1,2, … and Pop _ num-1 of the New Popdata, and replacing the chromosome sample with the minimum P _ fitness corresponding to the chromosome sample by the chromosome sample which does not appear in Fit _ max5[ w ], w is 0,1,2, 3, 4, thereby effectively preserving the previous 5 superior chromosomes and eliminating the chromosome with the smaller P _ fitness value.

Fifth, chromosome group crossover operator construction method based on electric quantity data

Generating a new array NewPopdata which contains Pop _ num chromosomes, taking two adjacent chromosomes as a group of chromosome groups to be crossed, combining two chromosome groups in turn to form a total group

The complete genome is taken down.

Taking the i, i-0, 1,2, …, Pop _ num-2 chromosomes to perform pairwise intersection operation with the i +1, i-0, 1,2, …, Pop _ num-2 chromosomes:

(1) before the two groups of chromosomes to be crossed are subjected to cross operation, a randomly selected operation random number c _ rand belongs to (0, 1), if the condition that c _ rand is larger than or equal to pc (pc is a cross probability factor and generally takes the value pc belongs to (0.8, 1) so that most of chromosomes can meet the condition of cross operation) is met, the two groups of chromosomes are not subjected to cross operation, and the chromosome group keeps the original gene unchanged.

(2) If c _ rand < pc is satisfied, the chromosome set is subjected to crossover operation, and the specific steps are as follows:

two random integers are taken:

t₁，t₂and t is₁<t₂，t₁∈(0，Poplenge-1)，t₂∈(0，Poplenge-1)

Poplenge＝3*m_num

Will [ t ] of the ith strip₁，t₂]The gene of the fragment is related to [ t ] of item i +1₁，t₂]The genes were cross-manipulated according to the following rules:

suppose the parent i and i +1 chromosomes are, with the color index corresponding to [ t [₁，t₂]The portion of the segments to be intersected is shown in figure 8,

the rule for the generation of the next generation chromosome by the intersection of the i chromosome to the i +1 chromosome is shown in fig. 9:

1) mapping i chromosome to [ t ]₁，t₂]The position gene completely inherits the sub-generation of the corresponding position;

2) find out i +1 chromosome and i chromosome [ t ]₁，t₂]The same gene position and located in the frame selection part;

3) the i +1 chromosome is linked to the i chromosome [ t ]₁，t₂]The gene positions with different segments are inserted into vacant positions in sequence.

In the same way, another next-generation stain generated by the i +1 chromosome crossing the i chromosome can be obtained, and the result is shown in FIG. 10, and the chromosome band boundary mutation operator construction method based on the electrical quantity data

The mutation of the research is that a certain gene position of a chromosome is randomly changed by meeting a mutation probability factor pm (pm is less than or equal to 10 percent), a user is randomly selected, the gene corresponding to the user is exchanged to another randomly selected position, namely the access position of a certain user is randomly changed, and the position of the gene position change has a precondition: because the areas are in the same region (for example, the A/B/C has three corresponding regions), the effect of variation can not be achieved, and therefore, the design requirement of the basic precondition is still existed when the corresponding position is selected.

The Pop _ num chromosome is operated according to the same rule as follows.

The mutation operation for a single chromosome is as follows:

5) judging whether the receiving needs variation operation: when the random number count _ rand is larger than or equal to pm, the count _ rand belongs to (0, 1) and the mutation is not operated; when count _ rand < pm, the next operation is continued.

6) Selecting a random integer t₁Rand is not more than Poplenge-1, and

rounding down when t ₁0 denotes the gene to which phase a belongs; when t is₁1-1 represents a gene to which phase B belongs; when t is₁The term "1-2" denotes a gene to which the C-phase belongs. Will t₁The value of the _randlocus gene is put in a variable t _ sample _ 1;

similarly, the above steps are taken to obtain the second random number t₂_rand≤Poplenge-1，t₂_rand≠t₁A ran d, and

rounding down when t ₂0 denotes the gene to which phase a belongs; when t is₂1-1 denotes

7) The gene to which phase B belongs; when t is₂_1＝2Represents the gene to which phase C belongs. Will t₂The value of the _randlocus gene is put in a variable t _ sample _ 2;

8) the distinguishing process comprises the following steps:

Seven, early-maturing prevention reselection mechanism construction method

Considering the possibility of the degradation of the judgment accuracy due to the "premature" condition, i.e. the termination of the loop condition by entering the local convergence point in a special case, the calculation of the genetic algorithm will be resumed by the reselection mechanism when the accumulated value exceeds the pre-established reselection rate pw within a short time (less than 2500 times) of the loop body execution.

The specific construction method comprises the following steps:

1. finding out the maximum number of chromosome composed by the genes with the same number in the Pop _ data _ A, and the main operation comprises two steps:

1.1) sorting the gene values in the Pop _ data _ A from large to small;

1.2) counting the maximum number of the same gene values, and recording as num _ a;

2. finding out the maximum number of chromosome elements consisting of genes with the same number in the Pop _ data _ B, and mainly operating in two steps:

1.3) sorting the Pop _ data _ B from large to small;

1.4) counting the maximum number of the same gene values, and recording as num _ b;

3. finding out the maximum number of chromosomes formed by genes with the same number in the Pop _ data _ C, and mainly operating in two steps:

1.5) sorting the Pop _ data _ C from large to small;

1.6) counting the maximum number of the same gene values, and recording as num _ c.

The judgment basis is as follows:

setting the number of times of total circulation as T (generally, T is more than or equal to 5000), and under the condition that the execution of a circulation body is not more than 2500 times, satisfying (num _ a + num _ b + num _ c)/3 × Pop _ num is more than or equal to pw, wherein pw belongs to [0.8, 1] value, restarting a reselection mechanism, namely, starting from random function random selection again, and redoing a genetic algorithm.

Eighthly, a non-signal injection type user-phase topological relation flow based on a genetic algorithm:

designing a master control flow:

1) the preset parameter T is 10000, pc is 0.8, pm is 0.1, pw is 0.8, and Pop _ num is 100

2) According to the electric quantity data information, the related data content is obtained after data cleaning, and the method comprises the following steps:

3) initializing Pop _ data, and randomly generating Pop _ data [ ] according to m _ num and Pop _ num;

4) entering a T10000 circulation process

5) Calling a fitness function module, and returning Fit _ ness, delta E₁

6) Entering into the selective function call of the genetic algorithm to obtain

NewPopdata，Fit_max5，MBest_pop[]＝max(Fit_max5)

7) Entering cross function call of a genetic algorithm to obtain updated NewPopdata;

8) entering cross function call of a genetic algorithm to obtain updated NewPopdata;

9) when T is 500, calling a reset function of the genetic algorithm, if the early-maturing condition is met, entering the step 3), and if not, continuing to execute the next step;

10) return 4)

11) And obtaining final NewPopdata, and obtaining Best _ Pop according to the following determined mapping relation, wherein the concrete judgment is as follows:

12) each user table corresponds to the discrimination result of each row of Best _ Pop [ m _ num ] [6], and the discrimination result comprises a 0 th bit representing user table number; 1/2/3, which respectively represent phase A, phase B and phase C; the 2 nd bit indicates the reliability of discrimination as 0/1, which indicates unreliability/credibility, respectively; the 3 rd, 4 th and 5 th positions represent the numbers of the A phase, the B phase and the C phase in the group respectively.

Specific running test example:

1. a platform area: 173 platform area

2. Data type: the partial screenshots of the data source of the power data of the actual 2019

year

4 and 5 months of the power are shown as follows

3. User number: 80

4. And (3) total circulation: 10000 times

5. 2026, the number of cycles to obtain the minimum value of Δ E1; minimum value of Delta E1 17552.573203380536

6. The 0 th bit represents the user table number (has a corresponding unique relationship with the actual user table number); 1/2/3, which respectively represent phase A, phase B and phase C; the 2 nd bit indicates the reliability of discrimination as 0/1, which indicates unreliability/credibility, respectively; the 3 rd, 4 th and 5 th positions represent the numbers of the A phase, the B phase and the C phase in the group respectively. The running test results are as follows (the phases A, B and C are 23, 25 and 22 respectively, wherein 11 user tables are accurately judged when waiting for subsequent power utilization due to sporadic power utilization or no power utilization):

[4,3,1,0,0,100]

[5,3,1,0,0,100]

[7,2,1,0,100,0]

[8,2,1,0,100,0]

[9,3,1,0,0,100]

[11,3,1,0,0,100]

[13,2,1,1,99,0]

[14,3,1,1,0,99]

[15,1,1,100,0,0]

[16,1,1,100,0,0]

[17,1,1,99,0,1]

[19,2,1,1,99,0]

[20,1,1,92,8,0]

[21,2,1,0,99,1]

[22,1,1,100,0,0]

[23,2,1,0,100,0]

[24,2,1,0,100,0]

[26,3,1,0,0,100]

[27,2,1,0,100,0]

[28,1,1,100,0,0]

[29,1,1,99,1,0]

[30,2,1,0,100,0]

[31,3,1,0,0,100]

[32,1,1,100,0,0]

[33,3,1,0,34,66]

[34,1,1,99,1,0]

[35,3,1,0,0,100]

[36,2,1,0,100,0]

[37,2,1,0,99,1]

[38,1,1,100,0,0]

[39,1,1,100,0,0]

[41,2,1,0,100,0]

[42,2,1,0,100,0]

[43,3,1,0,0,100]

[44,3,1,1,0,99]

[45,3,1,0,0,100]

[46,2,1,0,100,0]

[47,2,1,1,99,0]

[48,2,1,0,100,0]

[49,1,1,100,0,0]

[50,1,1,100,0,0]

[51,1,1,100,0,0]

[52,1,1,100,0,0]

[53,3,1,0,0,100]

[54,2,1,0,100,0]

[55,2,1,0,100,0]

[56,3,1,0,0,100]

[57,2,1,0,100,0]

[58,1,1,100,0,0]

[59,1,1,100,0,0]

[61,2,1,0,100,0]

[62,3,1,0,0,100]

[63,2,1,1,99,0]

[64,3,1,0,1,99]

[65,3,1,0,1,99]

[66,3,1,0,1,99]

[67,3,1,0,0,100]

[68,1,1,100,0,0]

[69,1,1,100,0,0]

[70,1,1,99,0,1]

[72,2,1,0,100,0]

[73,3,1,0,0,100]

[74,2,1,0,100,0]

[75,2,1,0,100,0]

[76,1,1,99,1,0]

[77,3,1,0,0,100]

[78,1,1,99,1,0]

[79,2,1,0,100,0]

[80,3,1,0,0,100]

[81,1,1,100,0,0]。

the artificial intelligence algorithm based on the electrical characteristics has certain universality, the principle and the method can be popularized to the fields of transmission, transformation and power distribution, lay a foundation for the analysis of large data of a power grid and the deep application of artificial intelligence, can be widely applied to the field of electric power ubiquitous Internet of things, and bring considerable social and economic benefits. As follows:

As noted above, while the present invention has been shown and described with reference to certain preferred embodiments, it is not to be construed as limited thereto. Various changes in form and detail may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. A non-signal injection type user-phase topological relation identification method based on a genetic algorithm is characterized by comprising the following steps:

(3) entering a circulating process:

2. The method for identifying the non-signal-injection-type user-phase topological relation based on the genetic algorithm as claimed in claim 1, wherein: the electric quantityThe data is any one of current, voltage, reactive power, electric energy and active power, and the data cleaning method comprises the steps of eliminating all 0 household meter sets, eliminating weak value household meter sets and eliminating invalid time sample space sets in valid household meter sets; screening the cleaned data comprises acquiring that the electrical quantity data of the outgoing line side of the transformer in the transformer area has P_a，P_b，P_c，P_Total-changeThe data for acquiring the electric meter sample at the user side has P_a，P_b，P_c，P_{General household}Obtaining the P of the outgoing line of the distribution transformer_Total-changeAnd accumulating and calculating an error delta P corresponding to the user meter P at the same sampling time, setting parameters, keeping the distribution transformer and the user meter sampling data at the same time when the delta P is less than or equal to the sampling time, adding the data into an effective data sample set, otherwise giving up the data, and meeting the number K of the effective time:

≤10％，K≥300；

≤15％，K≥400；

≤20％，K≥600；

≤25％，K≥1200；

≤35％，K≥2000。

3. the method for identifying the non-signal-injection-type user-phase topological relation based on the genetic algorithm as claimed in claim 2, wherein: in the step (1), effective data after data cleaning is obtained according to the electrical quantity data information, a storage list ValidDataSet is returned, each qualified sample comprises a distribution transformer, a combined set sample set of three-phase and single-phase electrical quantities of a user, m _ tnum in the storage list ValidDataSet represents the number of qualified user tables, Validcount _ num represents the number of qualified data samples, and a total electrical quantity matrix E at the distribution transformer side is obtained according to the data of the storage list ValidDataSet_{Distribution transformer assembly}[j]Distribution side A, B, C single-phase electric quantity matrix E_1-a[j]、E_1-b[j]、E_1-b[j]And meter-level single-phase electric quantity matrix

b_ij≠b_ik，j∈[1，N]，k∈[1，N]，i∈[1，M]。

4. The method for identifying the non-signal-injection type user-phase topological relation based on the genetic algorithm as claimed in claim 3, wherein: the fitness function construction process is as follows:

calculating the change rate Delta E of the A/B/C three phases of the distribution transformer of the transformer area_1-a[i]、ΔE_1-b[i]、ΔE_1-c[i]Calculating the change rate Delta E of the user table P_1-a1、ΔE_1-b1、ΔE_1-c1；

Calculate Fit [ p ]]＝AE₁[p]，p＝1，2，…Pop_num-1，Fit_max＝max(Fit[p])，

Fitness[P]＝Fit_max-Fit[P]，P＝1，2，…，Pop_num-1。

5. The method for identifying the non-signal-injection type user-phase topological relation based on the genetic algorithm as claimed in claim 4, wherein: the selection function construction method comprises the following steps

Fitness normalization processing:

P_fitness[j]＝Fitness[j]/Sumfit_ness，j＝0，1，…，Pop_num-1；

6. The method for identifying the non-signal-injection type user-phase topological relation based on the genetic algorithm as claimed in claim 5, wherein: the cross algorithm is as follows: selecting an operation random number c _ rand E (0, 1) and cross probability factors pc and pc E (0.8, 1), if c _ rand is more than or equal to pc, performing no cross operation on the two groups of chromosomes, and keeping the original gene of the chromosome group unchanged; if c _ rand < pc is satisfied, the chromosome set is subject to crossover operation.

7. The method for identifying the non-signal-injection type user-phase topological relation based on the genetic algorithm as claimed in claim 6, wherein: the mutation algorithm is as follows:

1) judging whether mutation operation is needed: when the random number count _ rand is larger than or equal to pm, the count _ rand belongs to (0, 1) and the mutation is not operated, and when the count _ rand is smaller than pm, the next operation is continued;

2) selecting a random integer t₁Rand is not more than Poplenge-1, and

rounding down when t₁0 denotes the gene to which phase A belongs, when t is₁1-1 represents a gene to which phase B belongs, when t is₁1-2 represents a gene belonging to the C-phase, and t is₁The value of the _randlocus gene is put in a variable t _ sample _ 1;

3) the same as the above steps, getSecond random number t₂_rand≤Poplenge-1，t₂_rand≠t₁A ran d, and

rounding down when t₂0 denotes the gene to which phase A belongs, when t is₂1-1 represents a gene to which phase B belongs, when t is₂1-2 represents a gene belonging to the C-phase, and t is₂The value of the _randlocus gene is put in a variable t _ sample _ 2;

4) the distinguishing process comprises the following steps:

8. The method for identifying the non-signal-injection-type user-phase topological relation based on the genetic algorithm as claimed in claim 1, wherein: when the loop body is executed for no more than half times, the accumulated value exceeds the preset reselection rate, the calculation of the genetic algorithm is carried out again through a reselection mechanism: sorting the values of the Pop _ data genes from large to small; the maximum number of the same gene values was counted and recorded as num.

9. The method for identifying the non-signal-injection-type user-phase topological relation based on the genetic algorithm as claimed in claim 1, wherein: each user table corresponds to the judgment result of each line of Best _ Pop and comprises a 0 th bit user table number; 1/2/3, which respectively represent phase A, phase B and phase C; the 2 nd bit indicates the reliability of discrimination as 0/1, which indicates unreliability/credibility, respectively; the 3 rd, 4 th and 5 th positions represent the numbers of the A phase, the B phase and the C phase in the group respectively.