CN112448392B - Regional power grid harmonic treatment method and device based on complex-valued neural network - Google Patents

Regional power grid harmonic treatment method and device based on complex-valued neural network Download PDF

Info

Publication number
CN112448392B
CN112448392B CN202011238149.5A CN202011238149A CN112448392B CN 112448392 B CN112448392 B CN 112448392B CN 202011238149 A CN202011238149 A CN 202011238149A CN 112448392 B CN112448392 B CN 112448392B
Authority
CN
China
Prior art keywords
harmonic
current
neural network
power grid
complex
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202011238149.5A
Other languages
Chinese (zh)
Other versions
CN112448392A (en
Inventor
金庆忍
姚知洋
郭敏
陈卫东
韩帅
周柯
阮诗雅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Electric Power Research Institute of Guangxi Power Grid Co Ltd
Original Assignee
Electric Power Research Institute of Guangxi Power Grid Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Electric Power Research Institute of Guangxi Power Grid Co Ltd filed Critical Electric Power Research Institute of Guangxi Power Grid Co Ltd
Priority to CN202011238149.5A priority Critical patent/CN112448392B/en
Publication of CN112448392A publication Critical patent/CN112448392A/en
Application granted granted Critical
Publication of CN112448392B publication Critical patent/CN112448392B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/01—Arrangements for reducing harmonics or ripples
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J3/00—Circuit arrangements for ac mains or ac distribution networks
    • H02J3/18—Arrangements for adjusting, eliminating or compensating reactive power in networks
    • H02J3/1821—Arrangements for adjusting, eliminating or compensating reactive power in networks using shunt compensators
    • H02J3/1835—Arrangements for adjusting, eliminating or compensating reactive power in networks using shunt compensators with stepless control
    • H02J3/1842—Arrangements for adjusting, eliminating or compensating reactive power in networks using shunt compensators with stepless control wherein at least one reactive element is actively controlled by a bridge converter, e.g. active filters
    • H—ELECTRICITY
    • H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
    • H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
    • H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
    • H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
    • Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
    • Y02E40/20—Active power filtering [APF]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Power Engineering (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention provides a regional power grid harmonic wave treatment method and device based on a complex value neural network, wherein a real-time monitoring device is used for collecting voltage information and current information of each node to form a data pair; training a complex-valued neural network using the data pairs; solving the optimal treatment current by using a complex value neural network; the APF is controlled to output the optimal treatment current to the power grid to complete treatment, the optimal treatment current is adjusted in real time, and the overall distribution condition of harmonic waves in the power grid is scientifically considered, so that a better treatment effect on the overall harmonic wave problem of the power grid is achieved.

Description

Regional power grid harmonic treatment method and device based on complex-valued neural network
Technical Field
The invention belongs to the field of power quality optimization of a power grid, and particularly relates to a regional power grid harmonic treatment method and device based on a complex-valued neural network.
Background
With the development of power electronics technology in recent years, more and more power electronics devices such as rectifiers, inverters and DC-DC converters are widely used in power grids due to their efficient and convenient characteristics of converting electric energy. However, due to the non-linear characteristics of the power electronics itself, the investment in power electronics inevitably creates a large number of non-negligible harmonic problems in the grid, degrading the power quality of the grid.
To solve such problems, various power quality control devices have been developed. The Active Power Filter (APF) has good development prospect due to the advantages of flexible compensation, high response speed and the like. The active power filter can be divided into a parallel type and a series type according to the installation mode. The parallel active power filter (SAPF) generally works in a manner of compensating a problem load in situ, and compensating the problem load by connecting the active power filter and a load to be treated in parallel. The harmonic current injected into the power grid by the load is detected, the harmonic component of the harmonic current is extracted, the harmonic current is inverted and serves as an output instruction of the main circuit, the main circuit is controlled by the harmonic current opposite to the load to be injected into the power grid through the instruction, and therefore the total current flowing into the power grid from the connection point does not contain the harmonic component, and the effect of harmonic suppression is achieved. From the filtering point of view, the harmonic components equivalent to the load flow out through the active filter without entering the grid, and are therefore referred to as active power filters.
When the active filter is used for treating concentrated high-power nonlinear loads, a good treatment effect can be obtained. However, for a local power grid with a plurality of distributed nonlinear loads, in order to optimize the overall power quality of the power grid, a local compensation method requires a plurality of active power filters to perform treatment at the same time. On the one hand, this leads to increased costs. On the other hand, if only part of the problem loads are treated, or when the installation position and the compensation load are not selected reasonably, the whole power quality of the power grid cannot be guaranteed to be optimized, and in a serious case, the power quality is even worsened compared with the power quality before treatment. Therefore, how to perform system-level comprehensive optimization treatment on the voltage harmonics of the local power grid containing a plurality of distributed nonlinear loads by using a small number of active power filters becomes a problem worthy of research.
Disclosure of Invention
The invention aims to provide a regional power grid harmonic treatment method and device based on a complex value neural network, and solves the problems that in the prior art, the treatment cost of a local power grid with a plurality of distributed nonlinear loads is high, and the overall power quality cannot be guaranteed.
The purpose of the invention is realized by the following technical scheme:
in a first aspect, the invention provides a regional power grid harmonic treatment method based on a complex-valued neural network, which comprises the following steps:
s1, collecting voltage information and current information of each node to form a data pair;
s2, training a complex-valued neural network by using data pairs;
s3, solving the optimal treatment current by using a complex value neural network;
and S4, controlling the APF to output the optimal treatment current to the power grid to finish treatment.
Further, the method for regional power grid harmonic treatment based on the complex-valued neural network further comprises the step of monitoring whether the harmonic current phasor meets the precision requirement in real time:
acquiring harmonic current phasor output by each harmonic source of a power grid in real time and recording the harmonic current phasor as [ I'1,I'2,...,I'n]If, if
Figure GDA0003728140600000021
If the value is less than the set threshold value, the precision requirement is met; otherwise, the step S3 is circulated until
Figure GDA0003728140600000022
Is smaller than the set threshold.
Furthermore, the voltage information and the current information of each node are collected by installing a real-time monitoring device on each node of the power grid and collecting the voltage information and the current information of each node through the real-time monitoring device.
Further, the construction method of the complex-valued neural network comprises the following steps: and taking the harmonic current phasor in each data pair as input and the harmonic voltage phasor as output, and establishing a fitting relation between the harmonic current and the harmonic voltage.
Further, the expression of the fitting relationship between the harmonic current and the harmonic voltage is as follows:
Figure GDA0003728140600000023
wherein:
Figure GDA0003728140600000024
representing the h-harmonic voltage phasor at each node,
Figure GDA0003728140600000025
representing the h-harmonic current phasors output by the n distributed harmonic sources,
Figure GDA0003728140600000026
represents h times of treatment current phasor output by the APF,
Figure GDA0003728140600000027
representing the h-order harmonic propagation characteristics of the power grid.
Further, the method for solving the optimal treatment current by using the complex value neural network comprises the following steps: setting the output current phasor of the current harmonic source acquired in real time as the first n inputs, adopting a traversal method for the rest p inputs, and obtaining the amplitude range of the governing current in traversal
Figure GDA0003728140600000028
The phase range is [0,2 π]And taking a group of inputs which enable the output phasor amplitude sum of the neural network to be the minimum as the optimal governing current.
In a second aspect, the invention provides a regional power grid harmonic treatment device based on a complex-valued neural network, which comprises a real-time monitoring device, an active power filter and a training complex-valued neural network; the real-time monitoring device is arranged on each node of the power grid and is used for acquiring voltage information and current information of each node; the training complex value neural network is used for solving the optimal treatment current; the active power filter is used for outputting the optimal treatment current to a power grid as the compensation current of the power grid.
The method for solving the optimal treatment current comprises the following steps: setting the output current phasor of the current harmonic source acquired in real time as the first n inputs, adopting a traversal method for the rest p inputs, and obtaining the amplitude range of the governing current in traversal
Figure GDA0003728140600000031
The phase range is [0,2 π]And taking a group of inputs which enable the output phasor amplitude sum of the neural network to be the minimum as the optimal governing current.
Furthermore, the real-time monitoring device is also used for monitoring whether the harmonic current phasor meets the precision requirement in real time.
Furthermore, the neurons of the complex-valued neural network adopt real-imaginary part neurons, the network structure adopts a full-connection structure, and the training algorithm is a complex-valued error back propagation algorithm.
Further, the current information is harmonic current phasor, and the voltage information is harmonic voltage phasor.
The regional power grid harmonic wave treatment method and device based on the complex value neural network can comprehensively treat the power grid distributed with a plurality of harmonic wave sources, so that the overall voltage harmonic wave condition of the power grid is improved. The monitoring equipment is utilized to obtain current information output by a plurality of groups of treatment equipment and harmonic sources in the power grid to be treated and voltage information of each node corresponding to the current information, and complex value neural network fitting modeling is utilized to obtain the relationship between the treatment current and the node voltage, so that the APF compensation instruction of the active power filter is generated not only based on a single node, but also scientifically considers the integral distribution condition of the harmonic in the power grid, and the better treatment effect on the integral harmonic problem of the power grid is achieved.
Drawings
FIG. 1 is a schematic diagram of an algorithm flow of a regional power grid harmonic treatment method based on a complex-valued neural network according to the present invention;
FIG. 2 is a schematic diagram of a 18-node grid system structure;
FIG. 3 (a) is a line graph of fifth harmonic neural network error as a function of iteration;
FIG. 3 (b) is a line graph of the seventh harmonic neural network error as a function of iteration;
FIG. 4 (a) shows the voltage amplitude of node 2 fifth harmonic;
FIG. 4 (b) shows the voltage amplitude of the fifth harmonic at node 3;
FIG. 4 (c) is the node 4 fifth harmonic voltage amplitude;
FIG. 4 (d) shows the voltage amplitude of the fifth harmonic at node 5;
FIG. 4 (e) is the node 6 fifth harmonic voltage amplitude;
FIG. 4 (f) is the node 7 fifth harmonic voltage amplitude;
FIG. 4 (g) shows the voltage amplitude of node 8 fifth harmonic;
FIG. 4 (h) shows the voltage amplitude of node 9 for the fifth harmonic;
FIG. 4 (i) is the sum of the squares of the voltage amplitudes of the fifth harmonic from node 2 to node 9;
FIG. 5 (a) is the node 2 seventh harmonic voltage amplitude;
FIG. 5 (b) is the voltage amplitude of the seventh harmonic of node 3;
FIG. 5 (c) is the voltage amplitude of the seventh harmonic of node 4;
FIG. 5 (d) is the voltage amplitude of the seventh harmonic of node 5;
FIG. 5 (e) is the voltage amplitude of the seventh harmonic at node 6;
FIG. 5 (f) is the voltage amplitude of the seventh harmonic at node 7;
FIG. 5 (g) is the node 8 seventh harmonic voltage amplitude;
FIG. 5 (h) is the voltage amplitude of the seventh harmonic at node 9;
FIG. 5 (i) is the sum of the squares of the voltage amplitudes of the seventh harmonic from node 2 to node 9;
fig. 6 (a) shows the output fifth harmonic current amplitude of the nonlinear load 1;
fig. 6 (b) shows the phase of the fifth harmonic current output by the nonlinear load 1;
FIG. 6 (c) shows the magnitude of the seventh harmonic current output by the nonlinear load 1;
FIG. 6 (d) shows the output seventh harmonic current phase of the nonlinear load 1;
fig. 7 (a) shows the output fifth harmonic current amplitude of the nonlinear load 2;
fig. 7 (b) shows the phase of the fifth harmonic current output by the nonlinear load 2;
FIG. 7 (c) shows the magnitude of the current of the seventh harmonic output by the nonlinear load 2;
FIG. 7 (d) shows the output seventh harmonic current phase of the nonlinear load 2;
FIG. 8 (a) shows the APF output fifth harmonic current amplitude;
FIG. 8 (b) shows APF output fifth harmonic current phase;
FIG. 8 (c) shows the APF output seventh harmonic current amplitude;
fig. 8 (d) shows the phase of the seventh harmonic current output by the APF.
Detailed Description
The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
The embodiments of the present disclosure are described below with specific examples, and other advantages and effects of the present disclosure will be readily apparent to those skilled in the art from the disclosure of the present disclosure. It is to be understood that the described embodiments are merely illustrative of some, and not restrictive, of the embodiments of the disclosure. The disclosure may be carried into practice or applied to various other specific embodiments, and various modifications and changes may be made in the details within the description and the drawings without departing from the spirit of the disclosure. It should be noted that the features in the following embodiments and examples may be combined with each other without conflict. All other embodiments, which can be derived by a person skilled in the art from the embodiments disclosed herein without making any creative effort, shall fall within the protection scope of the present disclosure.
The invention discloses a regional power grid harmonic treatment method based on a complex-valued neural network, which comprises the following steps of:
s1, collecting voltage information and current information of each node to form a data pair.
Further, in a preferred embodiment of the present application, the collecting voltage information and current information of each node is to install a real-time monitoring device on each node of the power grid, and collect the voltage information and current information of each node through the real-time monitoring device.
The invention does not limit the concrete form of the real-time monitoring device.
The current information comprises harmonic current phasors of the harmonic source and the APF, and the voltage information comprises harmonic voltage phasors. A harmonic current phasor and a harmonic voltage phasor of the same harmonic source form a set of data pairs. The data pair format is shown in table 1, and the number of data pairs is related to the number of grid nodes and the number of harmonic sources and APFs. Taking a power grid with m nodes containing n harmonic sources and p APFs as an example, it is suggested to collect at least m-x (n + p) group data pairs.
Figure GDA0003728140600000051
TABLE 1
And S2, training a complex-valued neural network by using the data pairs.
Further, in a preferred embodiment of the present application, the neurons of the complex-valued neural network are real-imaginary neurons, the network structure is a fully connected structure, and the training algorithm is a complex-valued error back-propagation algorithm. The neuron number of the input layer is n + p, and the neuron number of the output layer is m.
Further, in a preferred embodiment of the present application, the method for constructing the complex-valued neural network includes: and the harmonic current phasor in each data pair is used as input, the harmonic voltage phasor is used as output, and the fitting relation between the harmonic current and the harmonic voltage of the treatment node is established.
Taking an m-node three-phase power grid containing n distributed harmonic sources as an example, an APF is required to output a treatment current to reduce the integral harmonic content of the power grid, and then any h-order harmonic voltage phasor of each node of the power grid can be regarded as a function of the harmonic source and the h-order harmonic current phasor output by the APF, that is, the fitting relationship between the harmonic current and the harmonic voltage is as follows:
Figure GDA0003728140600000052
in the above formula
Figure GDA0003728140600000053
Representing the h-harmonic voltage phasor at each node,
Figure GDA0003728140600000054
representing the h-harmonic current phasors output by the n distributed harmonic sources,
Figure GDA0003728140600000055
represents h times of treatment current phasor output by the APF,
Figure GDA0003728140600000056
representing h harmonic propagation characteristics of the power grid, and obtained by complex value neural network modeling
Figure GDA0003728140600000057
The training process of the complex-valued neural network does not belong to the disclosure of the present invention, and is not described here again.
And S3, solving the optimal treatment current by using a complex value neural network.
Further, in a preferred embodiment of the present application, the method for obtaining the optimal treatment current by using the complex-valued neural network includes: after the neural network training is finished, the current harmonic source output current phasor (marked as [ I ] acquired by the real-time monitoring device1,I2,...,In]) Setting the first n inputs, adopting a traversal method for the rest p inputs, and obtaining the amplitude range of the treatment current in the traversal
Figure GDA0003728140600000061
The phase range is [0,2 π]And taking a group of inputs which enable the output phasor amplitude sum of the neural network to be the minimum as the optimal governing current.
Solving the optimal treatment current problem can be regarded as a constrained optimization problem, and the expression is as follows:
Figure GDA0003728140600000062
in the formula I1，…，InRepresenting the harmonic current phasor output by each harmonic source under the current power grid operation state, and Amp (·) represents a function of phasor amplitude.And (3) by using the complex-valued neural network obtained by training, taking the current of each harmonic source as the first n inputs, and solving the governing current phasor which minimizes the sum of squares of the harmonic voltage amplitudes of each node by adopting a traversal method.
And S4, controlling the APF to output the optimal treatment current to finish treatment.
Preferably, in an embodiment of the present application, the method for harmonic control of a regional power grid based on a complex-valued neural network further includes a step of monitoring in real time whether the harmonic current phasor satisfies the precision, and specifically includes:
the monitoring device collects harmonic current phasor output by each harmonic source in real time and records the harmonic current phasor as [ I'1,I'2,...,I'n]If, if
Figure GDA0003728140600000063
And if the value is smaller than the set threshold value, the accuracy requirement is considered to be met. Otherwise, the step S3 is circulated until
Figure GDA0003728140600000064
Is smaller than the set threshold.
The set threshold value can be set according to actual requirements, and the size of the threshold value should not be taken as a limitation to the present invention.
The beneficial effects of monitoring whether the harmonic current phasor meets the precision in real time are as follows: when a harmonic source in a power grid changes along with the change of an external voltage, iterative optimization needs to be performed on the harmonic source in order to ensure the accuracy of an optimal solution. The current of each harmonic source is collected in real time according to monitoring equipment, and the optimal treatment current is solved and updated in real time by using a complex value neural network, so that the harmonic content of the whole power grid is always minimum.
When the complex-valued neural network is constructed, each harmonic source is regarded as a current source, so that the establishment of a model reflecting the complex external characteristics of the harmonic source is avoided, the complex-valued neural network only reflects the harmonic propagation characteristics of a power grid, and the difficulty in training the neural network is reduced. However, when the optimal treatment current is solved, the output of the harmonic source generates certain offset due to the change of the working state of the power grid after treatment, and the treatment current deviates from the optimization. Therefore, an iterative process needs to be introduced into the treatment to avoid the influence of the phenomenon.
The invention also provides a regional power grid harmonic treatment device based on the complex-valued neural network, which comprises a real-time monitoring device, an active power filter and a training complex-valued neural network. The real-time monitoring device is installed on each node of the power grid and used for collecting voltage information and current information of each node. And training a complex-valued neural network for solving the optimal treatment current. The active power filter is used for outputting the optimal treatment current to the power grid as the compensation current of the power grid. The current information comprises harmonic current phasors of all harmonic sources and the APF, and the voltage information comprises harmonic voltage phasors of all harmonic sources and the APF.
Further, in a preferred embodiment of the present application, the real-time monitoring device is further configured to monitor whether the harmonic current phasor meets the precision requirement in real time. When a harmonic source in a power grid changes along with the change of an external voltage, iterative optimization needs to be performed on the harmonic source in order to ensure the accuracy of an optimal solution. The current of each harmonic source is collected in real time according to monitoring equipment, the optimal treatment current is solved and updated in real time by using a complex value neural network, and the updated optimal treatment current is compensated to the power grid through the APF, so that the integral harmonic content of the power grid is always the minimum.
Further, in a preferred embodiment of the present application, the neurons of the complex-valued neural network are real-imaginary neurons, the network structure is a fully connected structure, and the training algorithm is a complex-valued error back-propagation algorithm. The device comprises an input layer and an output layer, wherein the neuron number of the input layer is n + p, and the neuron number of the output layer is m.
To better illustrate the benefits of the present invention, the following description is presented in conjunction with specific test procedures and test data.
And (3) constructing a power grid system of IEEE typical 18 nodes in MATLAB/Simulink, and testing by adopting a MATLAB-Simulink joint simulation method. The structure of the 18-node system is shown in fig. 2.
In order to better verify the effectiveness of the method, a nonlinear load is taken as a harmonic source, and due to the harmonic coupling characteristic of the nonlinear load, the h-order harmonic current output of the harmonic source is influenced by the additional h-order harmonic voltage and other-order harmonic voltages, so that the treatment difficulty is increased.
Treatment objective and parameter design
A typical Non-linear load (NLL) three-phase six-pulse uncontrolled rectifier bridge is used as a harmonic source and connected with nodes 7 and 9, an APF is arranged at a node 4, 5 and 7 harmonic currents of the APF are collected as input, 5 and 7 harmonic voltages of the nodes 2-9 are output, and two complex-value neural networks are trained. And then the APF is utilized to treat the nodes 2 to 9, so that the sum of squares of 5 and 7 harmonic voltage amplitudes of the 8 nodes is minimum. The simulation parameter information is shown in tables 2, 3 and 4 below.
Figure GDA0003728140600000071
Figure GDA0003728140600000081
TABLE 2 load per node, parallel capacitor parameters
Figure GDA0003728140600000082
TABLE 3 line impedance parameters
Figure GDA0003728140600000083
TABLE 4 Power supply, non-Linear load parameters
(II) training neural networks
After 100 groups of 5 harmonic data pairs and 100 groups of 7 harmonic data pairs are acquired, two complex value neural networks (real part type and imaginary part type) with the structures of 3-10-10-8 are trained, 100 groups of data are divided into 75 groups of training sets, 15 groups of verification sets and 15 groups of test sets, and the three-set error is shown as a broken line graph along with the change of iteration times in fig. 3 (a) and 3 (b).
As can be seen from fig. 3, the errors of the fifth harmonic neural network and the seventh harmonic neural network reach about 0.3 after 30000 iterations, and the neural network predicted values are the fifth and seventh harmonic voltage phasors of the nodes 2 to 9, so that the square average of the difference phasor amplitude of the neural network to the predicted value phasor (normalization) of a single node voltage and the actual value phasor (normalization) of the node voltage is 0.0375, and the accuracy meets the governing requirement.
(III) analysis of treatment waveforms
After modeling is completed, the harmonic voltages of the nodes 2-9 can be comprehensively treated through the fifth harmonic neural network and the seventh harmonic neural network. In the treatment, five and seven harmonic currents of two nonlinear loads are sampled every 0.06s, and the five and seven harmonic currents obtained by sampling are used as constraint conditions of an optimization problem through a model, so that treatment currents which enable the square sum of the amplitudes of five and seven harmonic voltages of nodes 2-9 to be minimum under the current harmonic current are calculated and treated. Fig. 4 (a) to 4 (i) and fig. 5 (a) to 5 (i) show the voltage amplitudes of the fifth and seventh harmonics of each node and the square sum of the voltage amplitudes along with the waveform of the governance, the voltage is per unit value, and the reference value is 12.5kV.
As can be seen from fig. 4 (a) to 4 (i), as the harnessing progresses, the sum of the square of the amplitudes of the 8 nodes decreases gradually, and the algorithm achieves the effect of comprehensively harnessing the fifth harmonic. The voltage amplitude of the fifth harmonic of each node is also reduced to a certain extent, but the nodes 2, 3 and 4 generate bounce at 0.3-0.5 s. The harmonic voltage amplitudes of a few nodes are slightly rebounded, the harmonic voltage amplitudes of a large number of nodes are monotonically reduced along with the treatment, the sum of the square amplitudes is free from rebounding, the electric energy quality of the nodes is sacrificed by the algorithm to ensure the lowest integral harmonic content, and in the actual engineering, certain important nodes can be protected in an optimized mode in a weighting mode.
As can be seen from fig. 5 (a) to 5 (i), in the governing process, only the voltage amplitude of the node 8 monotonically decreases as the governing proceeds, most nodes have a bounce phenomenon, and a certain bounce is generated by the sum of squares of the amplitudes, and it is known that, due to the existence of the harmonic coupling characteristic, the governing on the fifth harmonic voltage each time changes the seventh harmonic current output by the nonlinear load before the fifth harmonic voltage stabilizes, resulting in a change in the propagation characteristic of the seventh harmonic network. After the fifth harmonic treatment is stabilized for 0.5s, the seventh harmonic current is not influenced by the fifth harmonic treatment any more, and the seventh harmonic voltage amplitude of each node is reduced accordingly and tends to be stable after 0.7 s. This can also be derived from the harmonic current outputs of the two nonlinear loads, which are shown in fig. 6 (a) to 6 (d) and fig. 7 (a) to 7 (d).
As can be seen from fig. 6 (a) to 6 (d), the fifth harmonic current output by the nonlinear load 1 is gradually stabilized as the treatment progresses, and the output seventh harmonic current fluctuates greatly before 0.5s, the amplitude change of the fifth harmonic current is firstly reduced and then increased again, and the fifth harmonic current tends to be stable until 0.5s later, which verifies the characteristics of the six-pulse wave uncontrolled rectifier bridge, that is, the low harmonic has a large influence on the high harmonic, and the high harmonic has a relatively small influence on the low harmonic.
The same conclusions can be drawn from the amplitudes and phases of the five and seven harmonic currents output by the nonlinear load 2 in fig. 7 (a) to 7 (d). The abatement current output by the APF is therefore affected similarly, as shown in fig. 8 (a) to 8 (d).
In conclusion, the complex value neural network is utilized to model the relationship between harmonic source current and node harmonic voltage in the distributed harmonic source system, and the obtained model can be used for calculating the optimal treatment current which enables the sum of the squares of the harmonic voltage amplitudes of all nodes to be minimum when the harmonic source is in a current working state, so that the effect of comprehensive harmonic treatment is achieved. And because the model does not contain the harmonic source, when the harmonic source changes, the model does not need to be modeled again, and the optimal treatment current under different states can be calculated based on the completed model in an iterative manner, so that the adverse effect caused by the output change of the harmonic source is avoided, the treatment effect of each harmonic is not counteracted mutually, and the content of each harmonic is minimized.
In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "connected," "secured," and the like are to be construed broadly and can, for example, be fixedly connected, detachably connected, or integrally formed; may be mechanically coupled, may be electrically coupled or may be in communication with each other; they may be directly connected or indirectly connected through intervening media, or they may be interconnected within two elements or in a relationship where two elements interact with each other unless otherwise specifically limited. The specific meanings of the above terms in the present invention can be understood according to specific situations by those of ordinary skill in the art.
The above description is for illustrative purposes only and is not intended to limit the present invention, and any modifications, equivalents, improvements, etc. within the spirit and scope of the present invention should be included within the scope of the present invention as defined by the appended claims.

Claims (9)

1. A regional power grid harmonic treatment method based on a complex value neural network is characterized by comprising the following steps:
s1, collecting voltage information and current information of each node to form a data pair;
s2, training a complex-valued neural network by using data pairs;
s3, solving the optimal treatment current by using a complex value neural network;
s4, controlling the APF to output optimal treatment current to the power grid to finish treatment;
the method for solving the optimal treatment current by using the complex value neural network comprises the following steps: setting the output current phasor of the current harmonic source acquired in real time as the first n inputs, and recording as [ I ]1,I2,...,In]Adopting a traversal method for inputting the other p APF output treatment current vectors, wherein the amplitude range of the treatment current is obtained in the traversal method
Figure FDA0003764189470000011
The phase range is [0,2 π]Taking a group of inputs which enable the harmonic voltage phasor amplitude square sum of each node of the neural network output to be minimum as the maximumThe current is preferably controlled.
2. The regional power grid harmonic treatment method based on the complex-valued neural network as claimed in claim 1, further comprising the step of monitoring whether harmonic current phasor meets the precision requirement in real time:
acquiring harmonic current phasor output by each harmonic source of a power grid in real time and recording the harmonic current phasor as [ I'1,I'2,...,I'n]If, if
Figure FDA0003764189470000012
If the value is less than the set threshold value, the precision requirement is met; otherwise, the step S3 is circulated until
Figure FDA0003764189470000013
Is smaller than the set threshold.
3. The regional power grid harmonic treatment method based on the complex-valued neural network as claimed in claim 1, wherein the collecting of the voltage information and the current information of each node is performed by installing a real-time monitoring device on each node of the power grid and collecting the voltage information and the current information of each node through the real-time monitoring device.
4. The complex-valued neural network-based regional power grid harmonic treatment method according to claim 1, wherein the complex-valued neural network is constructed by the method comprising the following steps: and taking the harmonic current phasor in each data pair as input and the harmonic voltage phasor as output, and establishing a fitting relation between the harmonic current and the harmonic voltage.
5. The complex-valued neural network-based regional power grid harmonic treatment method according to claim 4, wherein the expression of the fitting relationship between the harmonic current and the harmonic voltage is as follows:
Figure FDA0003764189470000014
wherein:
Figure FDA0003764189470000015
representing the h-harmonic voltage phasor at each node,
Figure FDA0003764189470000016
representing the h-harmonic current phasor output by the n distributed harmonic sources,
Figure FDA0003764189470000017
represents the h times of treatment current phasor output by the APF,
Figure FDA0003764189470000018
representing the h-harmonic propagation characteristics of the grid.
6. The device for treating the regional power grid harmonic wave based on the complex-valued neural network comprises a real-time monitoring device and an active power filter, and is characterized by also comprising a training complex-valued neural network; the real-time monitoring device is arranged on each node of the power grid and is used for acquiring voltage information and current information of each node; the training complex value neural network is used for solving the optimal treatment current; the active power filter is used for outputting optimal treatment current to a power grid as compensation current of the power grid;
the method for solving the optimal treatment current comprises the following steps: setting the output current phasor of the current harmonic source acquired in real time as the first n inputs, and recording as [ I ]1,I2,...,In]Adopting a traversal method for inputting the other p APF output treatment current vectors, wherein the amplitude range of the treatment current is obtained in the traversal method
Figure FDA0003764189470000021
The phase range is [0,2 π]And taking a group of inputs which enable the harmonic voltage phasor amplitude sum of each node output by the neural network to be minimum as the optimal governing current.
7. The complex-valued neural network-based regional power grid harmonic treatment device of claim 6, wherein the real-time monitoring device is further configured to monitor whether harmonic current phasors meet precision requirements in real time.
8. The device for regional power grid harmonic treatment based on a complex-valued neural network as claimed in claim 6, wherein the neurons of the complex-valued neural network are real-imaginary neurons, the network structure is a fully-connected structure, and the training algorithm is a complex-valued error back-propagation algorithm.
9. The device as claimed in claim 6, wherein the current information is harmonic current phasor, and the voltage information is harmonic voltage phasor.
CN202011238149.5A 2020-11-09 2020-11-09 Regional power grid harmonic treatment method and device based on complex-valued neural network Active CN112448392B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011238149.5A CN112448392B (en) 2020-11-09 2020-11-09 Regional power grid harmonic treatment method and device based on complex-valued neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011238149.5A CN112448392B (en) 2020-11-09 2020-11-09 Regional power grid harmonic treatment method and device based on complex-valued neural network

Publications (2)

Publication Number Publication Date
CN112448392A CN112448392A (en) 2021-03-05
CN112448392B true CN112448392B (en) 2022-11-01

Family

ID=74736548

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011238149.5A Active CN112448392B (en) 2020-11-09 2020-11-09 Regional power grid harmonic treatment method and device based on complex-valued neural network

Country Status (1)

Country Link
CN (1) CN112448392B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114285038B (en) * 2021-11-24 2023-10-31 上海希形科技有限公司 APF resonance prediction method based on voting perceptron

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103278686A (en) * 2013-05-10 2013-09-04 国家电网公司 Harmonic analysis filtering system and intelligently selected harmonic detection method

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3818777B2 (en) * 1998-06-22 2006-09-06 関西電力株式会社 Active filter control method
CN107147120B (en) * 2017-06-29 2020-01-07 河海大学常州校区 RBF dual neural network self-adaptive sliding mode control method of active power filter
CN109755940A (en) * 2019-03-19 2019-05-14 河南理工大学 A kind of control method of the virtual Active Power Filter-APF of alternating current-direct current mixing micro-capacitance sensor
CN111431184B (en) * 2020-04-10 2021-07-16 华南理工大学 Optimal virtual harmonic resistance control method based on BP neural network algorithm

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103278686A (en) * 2013-05-10 2013-09-04 国家电网公司 Harmonic analysis filtering system and intelligently selected harmonic detection method

Also Published As

Publication number Publication date
CN112448392A (en) 2021-03-05

Similar Documents

Publication Publication Date Title
Kwon et al. Harmonic instability analysis of a single-phase grid-connected converter using a harmonic state-space modeling method
Agarwal et al. Application of LMS-based NN structure for power quality enhancement in a distribution network under abnormal conditions
Çelik Lyapunov based harmonic compensation and charging with three phase shunt active power filter in electrical vehicle applications
CN109995034B (en) Regional power grid harmonic comprehensive treatment method considering harmonic coupling characteristics
Costa-Castelló et al. High-performance control of a single-phase shunt active filter
Osorio et al. A practical design procedure for robust H2 controllers applied to grid-connected inverters
Kazemtarghi et al. Active compensation-based harmonic reduction technique to mitigate power quality impacts of EV charging systems
Sujith et al. Optimization of harmonics with active power filter based on ADALINE neural network
CN113193560B (en) Method for equally dividing harmonic power in island micro-grid
CN104333002A (en) Mixed active power filter based on ip-iq detection method and hysteresis control
Prince et al. Power quality enhancement in a distribution network using PSO assisted Kalman filter—Based shunt active power filter
CN108462213B (en) Multifunctional grid-connected inverter control method and system based on conservation power theory
CN112448392B (en) Regional power grid harmonic treatment method and device based on complex-valued neural network
Chen et al. Fractional-order multiperiodic odd-harmonic repetitive control of programmable AC power sources
Theodoro et al. A flexible co-simulation framework for penetration studies of power electronics based renewable sources: A new algorithm for phasor extraction
CN110266044B (en) Microgrid grid-connected control system and method based on energy storage converter
Abou Houran et al. Active power filter module function to improve power quality conditions using GWO and PSO techniques for solar photovoltaic arrays and battery energy storage systems
Cheng et al. State-and-disturbance-observer-based current control scheme for LCL-filtered single-phase grid-tied inverters under nonideal conditions
CN108964013B (en) UPQC optimal output tracking control method and device based on state observer
CN113839396A (en) Dynamic reactive power compensation system and method for repairing line loss of power grid
CN113991675A (en) Micro-grid harmonic compensation method based on distributed event trigger control
Karamta et al. A review of power system state estimation: Techniques, state-of-the-art and inclusion of FACTS controllers
Li et al. High-frequency oscillation mechanism analysis of wind farm-side MMC station considering converter transformer stray capacitance
Pan et al. LCL APF based on fractional-order fast repetitive control strategy
Božiček et al. Performance evaluation of the DSP-based improved time-optimal current controller for STATCOM

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant