CN112865116B - Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network - Google Patents

Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network Download PDF

Info

Publication number
CN112865116B
CN112865116B CN202110029005.7A CN202110029005A CN112865116B CN 112865116 B CN112865116 B CN 112865116B CN 202110029005 A CN202110029005 A CN 202110029005A CN 112865116 B CN112865116 B CN 112865116B
Authority
CN
China
Prior art keywords
reactive power
region
voltage
parallel
capacitor
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
CN202110029005.7A
Other languages
Chinese (zh)
Other versions
CN112865116A (en
Inventor
殷林飞
王涛
陆悦江
高放
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.)
Guangxi University
Original Assignee
Guangxi University
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 Guangxi University filed Critical Guangxi University
Priority to CN202110029005.7A priority Critical patent/CN112865116B/en
Publication of CN112865116A publication Critical patent/CN112865116A/en
Application granted granted Critical
Publication of CN112865116B publication Critical patent/CN112865116B/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
    • H02J3/12—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load
    • H02J3/16—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load by adjustment of reactive power
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0637—Strategic management or analysis, e.g. setting a goal or target of an organisation; Planning actions based on goals; Analysis or evaluation of effectiveness of goals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • 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
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H02J3/46—Controlling of the sharing of output between the generators, converters, or transformers
    • H02J3/50—Controlling the sharing of the out-of-phase component
    • 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
    • Y02E40/30—Reactive power compensation
    • 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
    • Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
    • 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
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Abstract

The invention provides a thirteen-region diagram reactive power optimization method of a parallel federal diagram neural network. The method completes reactive power optimization by changing the position of a tap joint of the on-load tap changing transformer and switching the compensation parallel capacitor bank, divides the regional operation mode of the power system into thirteen real-time operation state regions according to the upper limit and the lower limit of reactive power and the upper limit and the lower limit of voltage, and is used for determining the implementation mode of the reactive power optimization adjustment measure of the system at the next moment. Secondly, a parallel federal graph neural network method is provided. The method carries out prediction training on historical data, judges the running state of the current system, determines the region of the system according to the running condition of the current system, and determines reactive power optimization measures. The method can ensure the operation privacy of the interconnected power system and ensure the parallel operation reactive power optimization scheme of each area of the interconnected power system.

Description

Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network
Technical Field
The invention belongs to the field of reactive power optimization of an electric power system, relates to a reactive power optimization method based on an artificial intelligence technology, and is suitable for the reactive power optimization problem of the electric power system.
Background
Under the development trend of an electric power system, the requirement of a power grid user on the electric energy quality is higher and higher, the voltage is an important index related to the electric energy quality of the whole electric power system, and the reactive power optimization problem of the electric power system is not only to enable the voltage of the electric power system to reach a normal level, so that the active loss of the system is reduced, but also to enable the electric power system to operate safely and stably. Most of the current reactive power optimization of the power system is to adjust the reactive power output of a generator set, change the transformation ratio of a transformer, use reactive power compensation devices such as a compensation reactor and a capacitor, and the like. However, at present, the power system is moving to the operation mode of a large-scale interconnected power system, and more distributed energy sources are connected to the power system, so that the overall power quality of the power system is affected, and the voltage is ensured to be in a normal operation state. The reactive power optimization process of the power system is very complex, and at present, many reactive power optimization problems are only system optimization aiming at different optimization targets, such as a compensation equipment capacity minimum optimization model, a network loss minimum optimization model, a voltage deviation specified value minimum optimization model, an optimization model considering voltage safety and the like. Solving such optimization problems typically uses some analytical methods such as weight methods, fuzzy optimization, penalty function methods, etc.; the methods need to construct corresponding mathematical models firstly, but when a big data complex solving process is faced by an analytic method, the decoupling process is complex, and a result without solution is obtained in the solving operation process. Currently, for solving the optimization problem of multiple objective functions, an intelligent method is a relatively common optimization method, such as an annealing method, a particle swarm method, a wolf method, and the like. At present, a neural network is often used in a series of processes such as prediction and solution of a power system, and the neural network is trained through big data to better fit the requirements of the optimized power system, so that a good optimization effect can be obtained.
At present, the reactive power regulation of the voltage of a transformer substation usually adopts a regulation mode of an on-load tap changer and a compensation parallel capacitor bank, but under the condition of completing double-parameter regulation of the voltage and the reactive power, manual regulation is difficult to accurately complete the regulation of work content, so that a transformer substation voltage regulation and control and reactive power compensation automatic control system is particularly critical. The artificial neural network can analyze big data, has the capabilities of collective operation and self-adaptive learning, and has very obvious effects on the aspects of finishing the predictive analysis of the power system, flexibly guiding the operation and the like. In the reactive power optimization process, if the transformer transformation ratio is frequently switched, unnecessary damage to equipment is caused; meanwhile, in the voltage regulation process of a certain node of the power system, the voltage quality of other nodes can be influenced, and the working times of the reactive power optimization regulation are reduced as far as possible, so that the normal stable operation state of the power system is kept. Although the conventional nine-zone graph method can correspondingly reduce the reactive power regulation operation frequency of the power system at present, the invention can more accurately divide the operation state of the power system into zones by utilizing a thirteen-zone graph method, and can more accurately complete corresponding regulation measures in different operation states so as to achieve the purpose of reducing the work frequency. Meanwhile, currently, under the trend of large-scale interconnected power systems, information privacy before each region is particularly critical, and a plurality of methods for solving the distributed optimization problem are provided, such as a series of distributed optimization methods such as an auxiliary problem principle, a lagrange multiplier method and the like. However, the federated learning approach not only can accomplish the distributed optimization problem, but also can ensure the security and privacy protection of the data provided by all participants.
Disclosure of Invention
The invention provides a reactive power optimization method for a thirteen-region diagram of a parallel federal diagram neural network. The method completes reactive power optimization by changing the position of a tap joint of the on-load tap changing transformer and switching the compensation parallel capacitor bank, divides the regional operation mode of the power system into thirteen real-time operation state regions according to the upper limit and the lower limit of reactive power and the upper limit and the lower limit of voltage, and is used for determining the implementation mode of the reactive power optimization adjustment measure of the system at the next moment. Secondly, a parallel federal graph neural network method is provided, historical data are subjected to predictive analysis by the method and are used for judging the operation state of the current system, the region to which the system belongs is determined according to the operation condition of the current system, and reactive power optimization measures are determined. And finally, the parallel federal diagram neural network method is utilized to ensure the operation privacy of the interconnected electric power system, simultaneously ensure that all areas of the interconnected electric power system can simultaneously run in parallel to calculate to obtain an optimal reactive power compensation mode, simultaneously complete adjustment measures and control all the areas of the electric power system to cooperatively perform reactive power optimization.
The method divides the regional operation mode of the power system into thirteen real-time operation state regions according to the upper limit and the lower limit of the reactive power and the upper limit and the lower limit of the voltage, and is used for determining the implementation mode of the reactive power optimization regulation measure of the system at the next moment. Compared with the traditional nine-region method, the thirteen-region method can more accurately distinguish the adjustment measures required by different operation modes of the power system.
According to the control measure of voltage reactive power regulation, thirteen regions in different operation states are divided on a two-dimensional numerical axis by constructing upper and lower voltage limits and upper and lower reactive power limits, and the operation states and the control strategy of each region are as follows:
region 0, operating state: the voltage is normal and the reactive power is normal; and (3) control strategy: not acting;
region 1, operating state: the voltage exceeds the upper limit and simultaneously the reactive power exceeds the lower limit; and (3) control strategy: lowering the tap, and if the tap is at the lowest level, putting the capacitor into use;
zone 2, operating state: when the voltage exceeds the upper limit, the reactive power is normal; and (3) control strategy: a tap is reduced; if the voltage is at the lowest level, the capacitor is used;
zone 3, operating state: when the voltage exceeds the upper limit, the reactive power is normal; and (3) control strategy: lowering the tap;
zone 4, operating state: the voltage exceeds the upper limit, and the reactive power exceeds the upper limit; and (3) control strategy: a capacitor is put into use; if the capacitor is not available, the tap is reduced;
zone 5, operating state: the voltage is qualified, and the reactive power exceeds the upper limit; and (3) control strategy: a capacitor is put into use; if the capacitor is not available, the tap is reduced;
zone 6, operating state: the voltage is qualified, and the reactive power exceeds the upper limit; and (3) control strategy: a capacitor is put into use;
zone 7, operating state: the voltage exceeds the lower limit, and the reactive power exceeds the upper limit; and (3) control strategy: a lift-off joint; if the maximum gear is reached, the capacitor is put into use;
region 8, operating state: the voltage exceeds the lower limit, and the reactive power is normal; and (3) control strategy: a lift-off joint; if the voltage is at the highest level, the capacitor is used;
region 9, operating state: the voltage exceeds the lower limit, and the reactive power is normal; and (3) control strategy: a lift-off joint;
zone 10, operating state: voltage crosses the lower limit, and reactive power crosses the lower limit; and (3) control strategy: putting the capacitor into use, and raising the branch connector if the capacitor is not available;
region 11, operating state: the voltage is qualified, and the reactive power exceeds the lower limit; and (3) control strategy: putting the capacitor into use, and raising the branch connector if the capacitor is not available;
region 12, operating state: the voltage is qualified, and the reactive power exceeds the lower limit; and (3) control strategy: a capacitor is put into use;
through the partition mode, the operation state of the whole power system is divided into thirteen area modes, and meanwhile, corresponding control strategies are selected according to different operation states.
After the operation state area is divided, the parallel federal graph neural network is constructed by using the thirteen-area graph reactive power optimization method based on the parallel federal graph neural network method, the operation data of the electric power system is analyzed and processed, the graph neural network is trained, and after the current operation state is analyzed by using the federal graph neural network, the operation state area where the operation state area is located is obtained and corresponding control and adjustment are carried out.
The parallel federal graph neural network provided by the invention has a very remarkable effect on processing graph data, a large amount of historical data of a control mode corresponding to a thirteen-region graph law are used for analyzing data characteristic values among all region graphs through the learning and training of the parallel federal graph neural network, and a parallel federal graph neural network training model is obtained and is used for predicting and analyzing the current operation condition of the system to obtain reactive power optimization regulation measures which should be taken by the system at present.
The network has four layers, namely an input layer, two hidden layers and an output layer. And each hidden layer comprises a plurality of neurons, the operation state of each node of the analysis system through analyzing the thirteen-region graph method is defined by the characteristic value and the relevant node, and the information and the state of each node should comprise the information of the adjacent node.
The model aims to learn the historical data of the running conditions of the relative nodes in each region graphThe operating state is set to learn a state embedding Hu∈RmIt contains neighborhood information for each node u. State embedding HuIs an m-dimensional vector of node u, which can be used to generate output data OuThrough OuThen, the reactive power optimization scheduling mode required in the current operation mode may be determined, the output function is assigned to the weight coefficient wpi (i ═ 1,2, L, n) of each region, and the global parameter coefficient wgi (i ═ 1,2, L, n) is assigned to the input function, where H is HuAnd OuThe specific expression of (a) is as shown in the following formulas (1) and (2):
Hu＝f(xu,xc[u],hn[u],xn[u]) (1)
Ou＝g(hu,xu) (2)
f is a parameter function, called a local transfer function, shares data among all nodes, and updates the node state according to input neighborhood data; g is an output function for describing an output process; x is the number ofuA vector of eigenvalues that are u; x is the number ofc[u]A vector of eigenvalues of its neighboring nodes; h isn[u]Neighborhood state function of u, xn[u]Is a function of the eigenvalues of the nodes.
H, O, X and XNRespectively representing all state functions, all output functions, all node characteristic values and vectors formed by superposing all node characteristics, and obtaining the following formulas (3) and (4):
H＝F(H,X) (3)
O＝G(H,XN) (4)
the global transfer function F and the global output function G are respectively accumulation formulas of F and G of all nodes in the graph; the value of H is the motionless point of equation (3) and is uniquely defined in the case where F is the shrink mapping; equation (3) sets any initial value H (0); the computational processes described in F and G can be interpreted as a feed-forward graph neural network.
The model is iterated as shown in equation (5):
Ht+1＝F(Ht,X) (5)
wherein HtRepresents the t-th iteration of H; the exponential operation performed by equation (5) quickly converges to the solution of the equation.
Data of data holders are generally subjected to centralized encryption and integration, and a graph neural network can encrypt data and gradient of each party in a homomorphic encryption mode in a training process and exchange edge information to complete parallel cooperative computing of a system.
Firstly, an initialized parameter model is sent to each region training neural network through centralized processing parameter server equipment, then a trainer trains the neural network through a local historical data set, each training process obtains each region parameter value wpi (i is 1,2, L, n), then information is encrypted and sent to a parameter server for processing, the latest global parameter wgi (i is 1,2, L, n) is obtained, and the updated global parameter is sent to each region for retraining until a training end condition is reached. In order to achieve higher operation speed, the model adopts a multi-parameter server calculation mode to perform parallel calculation on different weight parameters of the system, compares the operation results of updating the parameters each time, selects the best operation result, and sends the best operation result to each region to perform the next iterative calculation, so that the calculation speed is increased, and the calculation time is reduced.
Information privacy between regions is important for distributed interconnected power systems. Therefore, the model provided by the invention can meet independent parallel operation of each region, processes information in a centralized manner, adopts a multi-parameter server parallel weight parameter operation mode for calculation, and improves the cooperative work capacity among the regions, namely, the model sets a corresponding graph neural network for each region, and constructs a calculation model for cooperative calculation by using a parallel federal learning method, thereby reducing the calculation time required by optimization and improving the information privacy of each region.
Drawings
FIG. 1 is a thirteen-zone diagram reactive power optimization regulation control diagram of the method of the present invention.
FIG. 2 is a parallel federal graph neural network learning training diagram of the method of the present invention.
Detailed Description
The invention provides a reactive power optimization method for a thirteen-region diagram of a parallel federal diagram neural network, which is explained in detail by combining the attached drawings as follows:
FIG. 1 is a thirteen-zone diagram reactive power optimization regulation control diagram of the method of the present invention. The thirteen-region diagram method provided by the invention is mainly used for comprehensively controlling the voltage and the reactive power of the power system, and ensures that the power system can keep a normal and reasonable running state in the running process. Aiming at the running state of the power system, according to the control measures of voltage reactive power regulation, thirteen regions under different running states are divided on a two-dimensional numerical axis by constructing upper and lower voltage limits and upper and lower reactive power limits, the thirteen regions are divided into thirteen regions with the number of 0-12, when the running state of the power system is in the region of 1-12, corresponding regulation and control measures are needed, so that the final running mode of the power system is in the region of No. 0, and the regulation and control measures are divided into the adjustment of a tap joint of an on-load tap-changing transformer or the switching of a capacitor bank. Delta Q+、ΔQ-、ΔV+And Δ V-After the tap of the on-load tap changing transformer or the switched capacitor bank is adjusted, the reactive power changes to a relatively large value or a relatively small value, and the voltage changes to a relatively large value or a relatively small value. After the partition, the implementation of the corresponding voltage scheduling instruction is more accurate, redundant or unnecessary operations can be reduced as much as possible, the operation purpose is more accurate, the voltage fluctuation of the power system is greatly reduced, and a better regulation and control scheme is obtained.
FIG. 2 is a parallel federal graph neural network learning training diagram of the method of the present invention. Firstly, the invention provides a parallel federal graph neural network to solve the distributed computing problem of the interconnected power system and ensure the information privacy of each region. Dividing the region into 1-n regions, training a graph neural network by using the operation condition historical data set of each region in the region 1, wherein the network has four layers, namely an input layer, two hidden layers and an output layer, inputting the initial information of each region into the trained graph neural network to obtain the weight coefficient wpi (i is 1,2, L, n) of each region, leading the encrypted information of each region into a parallel operation server by encrypting the weight coefficient, obtaining corresponding parameters in a plurality of parameter servers by different parameter calculation modes, comparing and analyzing by the parallel operation server, selecting the encrypted global parameter wgi (i is 1,2, L, n) which best meets the requirement, returning to each region for decryption, continuously substituting into the neural network, and stopping iteration after meeting the requirement, and outputting the region information of the respective regions, and then comparing according to the corresponding thirteen-region graphs to obtain specific regulation control measures.

Claims (1)

1. A thirteen-region diagram reactive power optimization method of a parallel federal diagram neural network is characterized in that reactive power optimization is completed by changing the position of a tap joint of an on-load tap-changing transformer and switching a compensation parallel capacitor bank, the regional operation mode of a power system is divided into thirteen real-time operation state regions according to the upper limit and the lower limit of reactive power and the upper limit and the lower limit of voltage, and the thirteen real-time operation state regions are used for determining the implementation mode of reactive power optimization regulation measures of the system at the next moment;
the thirteen-zone diagram method divides thirteen zones under different operating states according to control measures of voltage reactive power regulation by constructing upper and lower voltage limits and upper and lower reactive power limits, and the operating states and control strategies of each zone are as follows:
region 0, operating state: the voltage is normal and the reactive power is normal; and (3) control strategy: not acting;
region 1, operating state: the voltage exceeds the upper limit and simultaneously the reactive power exceeds the lower limit; and (3) control strategy: lowering the tap, and if the tap is at the lowest level, putting the capacitor into use;
zone 2, operating state: when the voltage exceeds the upper limit, the reactive power is normal; and (3) control strategy: a tap is reduced; if the voltage is at the lowest level, the capacitor is used;
zone 3, operating state: when the voltage exceeds the upper limit, the reactive power is normal; and (3) control strategy: lowering the tap;
zone 4, operating state: the voltage exceeds the upper limit, and the reactive power exceeds the upper limit; and (3) control strategy: a capacitor is put into use; if the capacitor is not available, the tap is reduced;
zone 5, operating state: the voltage is qualified, and the reactive power exceeds the upper limit; and (3) control strategy: a capacitor is put into use; if the capacitor is not available, the tap is reduced;
zone 6, operating state: the voltage is qualified, and the reactive power exceeds the upper limit; and (3) control strategy: a capacitor is put into use;
zone 7, operating state: the voltage exceeds the lower limit, and the reactive power exceeds the upper limit; and (3) control strategy: a lift-off joint; if the maximum gear is reached, the capacitor is put into use;
region 8, operating state: the voltage exceeds the lower limit, and the reactive power is normal; and (3) control strategy: a lift-off joint; if the voltage is at the highest level, the capacitor is used;
region 9, operating state: the voltage exceeds the lower limit, and the reactive power is normal; and (3) control strategy: a lift-off joint;
zone 10, operating state: voltage crosses the lower limit, and reactive power crosses the lower limit; and (3) control strategy: putting the capacitor into use, and raising the branch connector if the capacitor is not available;
region 11, operating state: the voltage is qualified, and the reactive power exceeds the lower limit; and (3) control strategy: putting the capacitor into use, and raising the branch connector if the capacitor is not available;
region 12, operating state: the voltage is qualified, and the reactive power exceeds the lower limit; and (3) control strategy: a capacitor is put into use;
dividing the running state of the whole power system into thirteen regional modes, and selecting corresponding control strategies according to different running states;
setting a corresponding graph neural network for each region, and constructing a calculation model of cooperative calculation by using a parallel federal learning method; a parallel federal diagram neural network is provided for judging the current running state of each region of the interconnected power system and obtaining a corresponding regulation and control mode;
the parallel federal graph neural network trains the neural network through historical data sets of all regions, analyzes data characteristic values among graphs of all regions, obtains a parallel federal graph neural network training model, has four layers, namely an input layer, two hidden layers and an output layer, and inputs initial information of all regions into the trained parallel federal graph neural networkIn the network, the weight coefficient wp of each region is obtainediAnd i is 1,2, …, n, the encryption information of each area is led into the parallel operation server through the encryption processing weight coefficient, corresponding parameters are obtained in a plurality of parameter servers through different parameter calculation modes, then the parallel operation server carries out contrastive analysis, and the encryption global parameter coefficient wg which best meets the requirement is selectediReturning to each region, decrypting, continuously substituting into the neural network for iterative computation, stopping iteration after the requirement is met, and outputting region information of each region; the model aims to set the running state as a learning state embedded H by learning the historical data of the running condition of the relative nodes in each region graphu∈RmIt contains neighborhood information for each node u; state embedding HuIs an m-dimensional vector of node u for generating output data OuThrough OuDetermining a reactive power optimization scheduling mode required to be carried out in the current operation mode, and assigning an output function to the weight coefficient wp of each regioniThe global parameter coefficient wgiIs assigned to an input function, where HuAnd OuThe specific expression of (a) is as follows:
Hu＝f(xu,xc[u],hn[u],xn[u]) (1)
Ou＝g(Hu,xu) (2)
f is a parameter function, called a local transfer function, shares data among all nodes, and updates the node state according to input neighborhood data; g is an output function for describing an output process; x is the number ofuA vector of eigenvalues that are u; x is the number ofc[u]A vector of eigenvalues of its neighboring nodes; h isn[u]Neighborhood state function of u, xn[u]Is a function of the eigenvalues of the nodes;
h, O, X and XNRespectively representing all state functions, all output functions, all node characteristic values and vectors formed by superposing all node characteristics, and obtaining the following formula:
H＝F(H,X) (3)
O＝G(H,XN) (4)
the global transfer function F and the global output function G are respectively accumulation formulas of transfer functions and output functions of all nodes; the value of H is the fixed point of equation (3) and is uniquely defined in the case of F-shrink mapping; equation (3) sets any initial value H (0); the calculation processes described in F and G are feed-forward graph neural networks;
the model iterates as shown in the following equation:
Ht+1＝F(Ht,X) (5)
wherein HtRepresents the t-th iteration of H; performing exponential function operation through a formula (5) to quickly converge to a solution of the equation;
carrying out centralized encryption and integration on data of a data holder, carrying out encryption processing on data and gradient of each part by a graph neural network in a homomorphic encryption mode in a training process, and carrying out edge information exchange to complete parallel cooperative calculation of a system;
and setting a corresponding graph neural network for each region, and constructing a collaborative computing calculation model by using a parallel federal learning method, thereby reducing the calculation time required by optimization and improving the information privacy of each region.
CN202110029005.7A 2021-01-11 2021-01-11 Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network Active CN112865116B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110029005.7A CN112865116B (en) 2021-01-11 2021-01-11 Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110029005.7A CN112865116B (en) 2021-01-11 2021-01-11 Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network

Publications (2)

Publication Number Publication Date
CN112865116A CN112865116A (en) 2021-05-28
CN112865116B true CN112865116B (en) 2022-04-12

Family

ID=76002170

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110029005.7A Active CN112865116B (en) 2021-01-11 2021-01-11 Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network

Country Status (1)

Country Link
CN (1) CN112865116B (en)

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102354995B (en) * 2011-09-28 2013-12-11 广东省电力调度中心 Control method and control system for VQC (voltage quality control) of transformer substation
CN103166235A (en) * 2013-04-10 2013-06-19 镇江泰源电器科技有限公司 System for controlling power station reactive power automatic compensation based on third generation (3G) technology
CN103972905B (en) * 2014-04-28 2016-08-17 广州供电局有限公司 400V distributing net platform region reactive power compensation ability method
US20170124464A1 (en) * 2015-10-28 2017-05-04 Fractal Industries, Inc. Rapid predictive analysis of very large data sets using the distributed computational graph
AU2017292924B2 (en) * 2016-07-07 2022-08-18 The Regents Of The University Of California Implants using ultrasonic backscatter for sensing physiological conditions
CN108039726B (en) * 2017-12-12 2021-02-09 国网山东省电力公司德州供电公司 Energy local area network distributed cooperative control method based on multi-agent system
CN108242811B (en) * 2018-03-09 2019-10-29 广东电网有限责任公司汕头供电局 A kind of alternating current-direct current paired running receiving-end system VQC control method
CN109638823B (en) * 2018-12-19 2022-08-02 国网河南省电力公司电力科学研究院 Automatic partitioning method and device for state estimation of radiation type power grid
CN110311372A (en) * 2019-06-30 2019-10-08 南京理工大学 Sub-area division method based on spectral clustering
CN110503157B (en) * 2019-08-27 2021-11-19 安徽大学 Image steganalysis method of multitask convolution neural network based on fine-grained image
CN111222628B (en) * 2019-11-20 2023-09-26 深圳前海微众银行股份有限公司 Method, device, system and readable storage medium for optimizing training of recurrent neural network
CN111797974B (en) * 2020-06-01 2022-10-11 武汉大学 Power system state estimation method combining particle filtering and convolutional neural network
CN111553470B (en) * 2020-07-10 2020-10-27 成都数联铭品科技有限公司 Information interaction system and method suitable for federal learning
CN111967506A (en) * 2020-07-31 2020-11-20 西安工程大学 Electroencephalogram signal classification method for optimizing BP neural network by artificial bee colony
CN112200321B (en) * 2020-12-04 2021-04-06 同盾控股有限公司 Inference method, system, device and medium based on knowledge federation and graph network

Also Published As

Publication number Publication date
CN112865116A (en) 2021-05-28

Similar Documents

Publication Publication Date Title
Sun et al. Two-stage volt/var control in active distribution networks with multi-agent deep reinforcement learning method
Dobbe et al. Toward distributed energy services: Decentralizing optimal power flow with machine learning
Fukuyama et al. A particle swarm optimization for reactive power and voltage control in electric power systems
Hu et al. Multi-agent deep reinforcement learning for voltage control with coordinated active and reactive power optimization
Abiyev et al. A type-2 fuzzy wavelet neural network for system identification and control
Van der Lee et al. A tuning algorithm for model predictive controllers based on genetic algorithms and fuzzy decision making
Wang et al. Distributed extremum seeking for optimal resource allocation and its application to economic dispatch in smart grids
Lian Reactive power optimization based on adaptive multi-objective optimization artificial immune algorithm
CN115313403A (en) Real-time voltage regulation and control method based on deep reinforcement learning algorithm
Li et al. Multi-objective decomposition optimization algorithm based on adaptive weight vector and matching strategy
Yin et al. Sequential reconfiguration of unbalanced distribution network with soft open points based on deep reinforcement learning
CN112865116B (en) Thirteen-area diagram reactive power optimization method of parallel federal diagram neural network
Menhas et al. Real/binary co-operative and co-evolving swarms based multivariable PID controller design of ball mill pulverizing system
Zeng et al. Solving multiobjective optimal reactive power dispatch using improved multiobjective particle swarm optimization
CN113422371A (en) Distributed power supply local voltage control method based on graph convolution neural network
Chen et al. A Novel Approach Based on Modified and Hybrid Flower Pollination Algorithm to Solve Multi-objective Optimal Power Flow.
Hoseini et al. A new multi objective optimization approach in distribution systems
CN114707613B (en) Layered depth strategy gradient network-based power grid regulation and control method
Xue et al. Multi-model modelling and predictive control based on local model networks
Wang et al. Cloud model-based artificial immune network for complex optimization problem
Li et al. Research on an improved coordinating method based on genetic algorithms and particle swarm optimization
Oukati Sadegh et al. Decentralised coordination of facts devices for power system stability enhancement using intelligent programming
Mu et al. Graph multi-agent reinforcement learning for inverter-based active voltage control
Matsumoto et al. Voltage and Reactive Power Control by Parallel Modified Brain Storm Optimization
Obert et al. Efficient Distributed Energy Resource Voltage Control Using Ensemble Deep Reinforcement Learning

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