CN113452026A - Intelligent training method, evaluation method and system for weak evaluation of power system - Google Patents

Intelligent training method, evaluation method and system for weak evaluation of power system Download PDF

Info

Publication number
CN113452026A
CN113452026A CN202110731653.7A CN202110731653A CN113452026A CN 113452026 A CN113452026 A CN 113452026A CN 202110731653 A CN202110731653 A CN 202110731653A CN 113452026 A CN113452026 A CN 113452026A
Authority
CN
China
Prior art keywords
power
power system
network
generator
intelligent agent
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.)
Granted
Application number
CN202110731653.7A
Other languages
Chinese (zh)
Other versions
CN113452026B (en
Inventor
姚伟
曾令康
文劲宇
黄彦浩
汤涌
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.)
Huazhong University of Science and Technology
China Electric Power Research Institute Co Ltd CEPRI
Original Assignee
Huazhong University of Science and Technology
China Electric Power Research Institute Co Ltd CEPRI
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 Huazhong University of Science and Technology, China Electric Power Research Institute Co Ltd CEPRI filed Critical Huazhong University of Science and Technology
Priority to CN202110731653.7A priority Critical patent/CN113452026B/en
Publication of CN113452026A publication Critical patent/CN113452026A/en
Application granted granted Critical
Publication of CN113452026B publication Critical patent/CN113452026B/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
    • H02J3/04—Circuit arrangements for ac mains or ac distribution networks for connecting networks of the same frequency but supplied from different sources
    • H02J3/06—Controlling transfer of power between connected networks; Controlling sharing of load between connected networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • 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
    • 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
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • 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
    • H02J3/14—Circuit arrangements for ac mains or ac distribution networks for adjusting voltage in ac networks by changing a characteristic of the network load by switching loads on to, or off from, network, e.g. progressively balanced loading
    • H02J3/144—Demand-response operation of the power transmission or distribution network
    • 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
    • H02J3/48—Controlling the sharing of the in-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
    • H02J2203/10—Power transmission or distribution systems management focussing at grid-level, e.g. load flow analysis, node profile computation, meshed network optimisation, active network management or spinning reserve management
    • 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]
    • 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
    • H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
    • H02J2300/40—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation wherein a plurality of decentralised, dispersed or local energy generation technologies are operated simultaneously
    • 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
    • Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
    • Y02B70/00—Technologies for an efficient end-user side electric power management and consumption
    • Y02B70/30—Systems integrating technologies related to power network operation and communication or information technologies for improving the carbon footprint of the management of residential or tertiary loads, i.e. smart grids as climate change mitigation technology in the buildings sector, including also the last stages of power distribution and the control, monitoring or operating management systems at local level
    • Y02B70/3225—Demand response systems, e.g. load shedding, peak shaving
    • 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
    • Y04S20/00—Management or operation of end-user stationary applications or the last stages of power distribution; Controlling, monitoring or operating thereof
    • Y04S20/20—End-user application control systems
    • Y04S20/222—Demand response systems, e.g. load shedding, peak shaving

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Power Engineering (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Economics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Game Theory and Decision Science (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

The invention discloses a training method, an evaluation method and a system for a power system weakness evaluation intelligent agent, and belongs to the field of power system weakness evaluation. The method is based on a deep reinforcement learning algorithm and a power system cascading failure model, an agent based on a deep Q network decides an attack line which most easily causes the power system to crash, a power flow transfer process after the attacked line exits from running is simulated based on the power system cascading failure model, and the power transmission line with the most serious power flow out-of-limit is automatically cut. And continuously utilizing the intelligent agent to decide the attack line until the outage line or the loss load reaches a certain threshold value, judging that the power system is crashed, and outputting an attack sequence decided by the intelligent agent. In the process, experience samples required by reinforcement learning are stored and the updating agent is trained. According to the method, the intelligent agent obtained by deep reinforcement learning algorithm training can be used for effectively deciding the attack sequence which most easily causes the breakdown of the power system under the current tide working condition, so that the weakness degree of the power system is evaluated.

Description

Intelligent training method, evaluation method and system for weak evaluation of power system
Technical Field
The invention belongs to the technical field of weak assessment of power systems, and particularly relates to a training method, an assessment method and a system for an intelligent agent for weak assessment of a power system.
Background
In modern large interconnected power systems, as loads continue to increase, the power delivered by the lines continues to approach transmission capacity. At this time, if the transmission line is out of operation due to fault tangent, the power flow of the power system is transferred, which may cause the power flow of other transmission lines to exceed the limit, and further cause cascading faults, resulting in shutdown of a large number of transmission lines, and finally may develop into accidents such as grid collapse of the power system, large-scale power failure, and the like. The weak evaluation of the power system can find out the weak power transmission line under the current power system structure and tide operation mode, and the dispatching work is carried out for guiding the power system operation technicians, so that the safe and reliable operation of the power system is guaranteed.
Most of existing weak line identification work of the power system is performed aiming at power supply sufficiency, and the condition that after the weak line exits from operation, a load of the power system is lost and a grid frame is broken down possibly due to cascading failure accidents caused by power flow transfer is mainly considered, so that the power supply sufficiency cannot be guaranteed. However, such evaluation work of the weak line of the power system is performed by calculating indexes such as a power flow entropy and an information entropy in a certain power flow operation mode of the power system, and the weak line of the power system corresponding to the operation mode is obtained. In the actual operation process, the operation modes of the power system are varied, and after the trend mode is changed, the existing power system weak evaluation method needs to execute the calculation process again and simulate the cascading failure evolution process to obtain the power system weak line.
Disclosure of Invention
Aiming at the defects of the related art, the invention aims to provide an intelligent agent training method, an evaluation method and an intelligent agent training system for evaluating the weakness of a power system, and aims to solve the problems that in the prior art, the weak evaluation of the power system is carried out through an algorithm, a calculation process needs to be executed again after a load flow operation mode is changed, and a cascading failure evolution process is simulated to obtain a weak line of the power system.
To achieve the above object, an aspect of the present invention provides a power system weakness assessment agent training method, including the following steps:
s1, reading the load flow calculation data and obtaining the load flow state of the power system;
s2, the intelligent agent decides the number of the attack line according to the current power flow state;
s3, executing line attack, performing sub-network detection and pretreatment of the power system after disconnection, adjusting power flow calculation data of the power system, and calculating new power flow of the power system;
s4, judging whether a power flow out-of-limit line exists or not, if yes, cutting off any power flow out-of-limit line, returning to the step S3, and if not, continuing to execute the step S5;
s5, adding the current training sample data into an intelligent agent model training sample library, training and updating intelligent agent parameters;
s6, judging whether the power flow state reaches a first termination condition, if not, returning to the step S2, and if so, outputting an attack sequence obtained by the decision of the intelligent agent;
and S7, judging whether the intelligent agent training process reaches a second termination condition, if not, returning to the step S1, and if so, outputting the trained intelligent agent model.
Further, the power flow calculation data includes:
information of the power transmission line: the serial numbers of the left and right buses, the reactance resistance capacitance, the transformer transformation ratio, the rated transmission capacity and the line running state;
bus information: the active power output, the load active power and the bus networking state of the generator;
generator information: and (5) a generator speed regulation coefficient.
Furthermore, the intelligent agent is represented by a deep Q network, the dimensionality of an output layer is the number of the transmission lines which can be attacked by the power system, and the deep Q network calculates the Q value of the output neuron corresponding to each transmission line by using the input state data.
Further, in step S3, the sub-network detection and pre-processing includes:
if the power system is divided into more than 2 sub-networks after the disconnection, respectively bringing the power transmission line, the generator and the load into the corresponding sub-networks, keeping the power balance of each sub-network, and having no power interaction with each other;
if no generator is arranged in the sub-network, when the sub-network is processed, the transmission line state of the sub-network is changed from running to shutdown, and the bus networking state is changed from network access to network disconnection.
Further, in step S3, the adjusting the power system power flow calculation data includes:
if the active power in the current sub-network is unbalanced, proportionally distributing the amount of the active power unbalance according to the speed regulation coefficient of the generator with the frequency regulation capability, if the active power is surplus, reducing the active power output of the generator, and if the active power is insufficient, increasing the active power output of the generator;
if the adjusted active output of the generator exceeds the upper limit of the active output of the generator, setting the active output of the generator as the upper limit of the active output of the generator, and bearing the residual active output up-regulation quantity by other generators; if the active output power up-regulation quantity still remains, reducing the load to achieve power balance;
if the adjusted active output of the generator is lower than the lower limit of the active output of the generator, setting the active output of the generator as the lower limit of the active output of the generator, and bearing the remaining active output down-regulation quantity by other generators; if the active power output down-regulation quantity still remains, the generator is cut off to achieve power balance;
if all the generators in the sub-network are cut off after the adjustment, the sub-network correspondingly adjusts the states of the power transmission line and the bus according to the output of the passive network.
Further, the step S5 includes:
taking the power flow state before the decision of the intelligent agent is cut, the decision of the cut, the power flow state after the cut, an immediate return function and a Markov decision process ending mark as a reinforced learning sample and storing the reinforced learning sample into an intelligent agent training sample library;
and extracting sample data from the intelligent agent training sample library by adopting a prior experience sampling method.
Further, the determining whether the power flow state reaches the first termination condition includes:
and judging whether the total number of the current outage lines reaches a preset scale or whether the loss load reaches a preset proportion.
Further, whether the intelligent agent training process reaches a second termination condition comprises:
and judging whether the update times of the intelligent agent parameters or the initial power flow data read in an accumulated mode reach a preset threshold value or not.
The invention also provides a method for evaluating the weakness of the power system, which comprises the following steps:
(1) reading initial load flow calculation data, and acquiring the current initial load flow state of the power system;
(2) the intelligent agent obtained by training by adopting the training method decides the number of the attack line according to the current tide state;
(3) performing line attack, performing sub-network detection and pretreatment of the power system after disconnection, adjusting power system load flow calculation data, and calculating new power system load flow;
(4) judging whether a power flow out-of-limit line exists or not, if so, cutting off any power flow out-of-limit line, returning to the step (3), and if not, continuing to execute the step (5);
(5) judging whether the power flow state reaches a first termination condition, if not, returning to the step (2), and if so, outputting an attack sequence obtained by the decision of the intelligent agent;
(6) and evaluating the weakness degree of the power system in the initial power flow state according to the length of the attack sequence.
Yet another aspect of the present invention provides a power system weakness evaluation system, including: a computer-readable storage medium and a processor;
the computer-readable storage medium is used for storing executable instructions;
the processor is used for reading executable instructions stored in the computer readable storage medium and executing the power system weakness assessment method.
Through the technical scheme, compared with the prior art, the weak line evaluation method for the power system can provide the weak line of the power system without simulating the cascading failure process and calculating related indexes again after the power flow operation mode of the power system is changed. Specifically, the evaluation method is based on a deep Q network algorithm, an intelligent agent decision attack line formed by a deep neural network is adopted, in an intelligent agent training stage, after the attack line is determined, cascading failure simulation calculation of a power system is executed, firstly, the active output and the load of a generator in each sub-network are adjusted, and active power balance is realized; secondly, according to the direct current power flow calculation result of the power system, cutting off the line of the power flow out of limit in the power system; and then storing a reinforcement learning sample for training the intelligent agent model, training and updating network parameters of the intelligent agent, and finally outputting a line attack sequence after the intelligent agent executes a plurality of times of line attack decisions to cause the grid structure of the power system to collapse. When the method is applied to the online weakness evaluation of the actual operation of the power system after the training is finished, the weakest route can be decided directly according to the current tide state. In addition, the evaluation method can provide a line attack sequence which causes the grid frame of the power system to be crashed, judge the weakness degree of the power system in the current tide operation mode according to the length of the attack sequence and guide the operation technicians of the power system to carry out related scheduling operation.
Drawings
Fig. 1 is a flowchart of a method for evaluating a power system weakness under a sequence attack according to an embodiment of the present invention;
fig. 2 is a flow chart of a neutron network power flow adjustment in the cascading failure model.
Fig. 3 is a single line diagram of a bus system 39 of the IEEE standard new england 10 machine.
FIG. 4 is a diagram of the decision effect of the agent in the initial operation condition sample of the test set trend during the training process.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
A method 100 for evaluating a power system weakness under a sequence attack, as shown in fig. 1, includes:
step 110, reading initial load flow calculation data of the power system, and calculating direct current load flow of the power system;
step 120, according to the current power flow state, the intelligent agent decides the corresponding attack tangent line number;
step 130, detecting and processing sub-networks generated by the power system after the circuit is disconnected, adjusting load flow calculation data such as output and load of the generator in the sub-networks one by one, and calculating the adjusted direct current load flow of the power system;
step 140, judging whether a line with the power flow out of limit exists or not based on the calculated direct current power flow, if so, cutting off the line with the most serious power flow out of limit, and executing step 130, otherwise, executing step 150;
step 150, storing current training sample data, training a model, and correcting and updating intelligent agent parameters;
step 160, judging whether the number of the outage lines exceeds a first threshold value, if not, executing the step 2, and if so, outputting an attack sequence obtained by the decision of the intelligent agent;
and 170, judging whether the iteration times exceed a second threshold, if not, executing the step 1, and if so, outputting the trained intelligent agent model.
The weak line identification method provided by the embodiment considers the influence of power flow transfer to cause cascading failure of the power system after the failure of the power transmission line is removed in different operation modes. Specifically, the identification method is based on a double-depth Q network (DBQN) algorithm, an epsilon-greedy strategy is adopted to determine an attacked target line under different tide operation modes, and a power system cascading failure caused by tide transfer after the attacked line is withdrawn from operation is simulated through a power system direct current tide cascading failure model. And then generating a sign of whether the net rack is crashed or not and an immediate return function according to the simulation calculation result to form a sample library of the training intelligent agent. And finally, extracting quantitative samples from the sample library to train and update parameters of the Q estimation network, and obtaining the power transmission line attack which can make a decision under different tide operation conditions and most easily causes serious cascading failure of the power system.
According to the embodiment, the DBQN algorithm can be used for effectively identifying the weak degree of the power system under the sequence attack of different power flow operation modes, and the identification accuracy is high. Secondly, when the DBQN trains and updates Q estimation network parameters, the action a ' of the next state s ' of the Q estimation network decision is firstly adopted, then the Q value of the a ' in the next state is estimated by adopting the Q target network, so that the problem of over-estimation of the Q value caused by the action of the next state of the Q target network decision and the estimation of the Q value can be effectively avoided, and the training process of the DBQN intelligent agent is more stable. In addition, when the intelligent agent Q estimation network is trained, a priority experience sampling method is adopted, more valuable samples can be utilized more fully, and the convergence rate of the intelligent agent training process is improved. Finally, the power system weakness evaluation method of the embodiment can be simultaneously applied to off-line operation mode evaluation and on-line operation weakness line identification, for off-line operation mode calculation, an attack sequence causing grid collapse of the power system can be obtained by combining with a cascading failure model of the power system, and the weakness degree of the operation mode is evaluated according to the length of the attack sequence (including the number of power transmission lines); for on-line operation, after the operation mode is changed, the trained intelligent body can directly give out the transmission line fault which most easily causes serious cascading failure of the power system.
Preferably, the initial power flow calculation data of the power system read in step 110 is composed of load active power and generator active power output setting;
when the power system actually operates, the load active power of the bus can be changed constantly, and in order to ensure power supply margin, the active power of the generator in the power grid can be changed along with the change of the load active power of the bus, so that the active power balance of the power grid is maintained. Other load flow calculation data generally do not change much. Before the intelligent agent training process is executed, different load active power levels are set, and a large amount of initial load flow operation condition calculation data are generated.
A large number of initial trend operating conditions can provide enough sample data for the training of the intelligent agent, and the trained intelligent agent can not only aim at the trend operating conditions for the training, but also provide an attack sequence which most easily causes the grid structure of the power system to crash for a new operating condition. Therefore, in the actual online operation process of the power grid, after the tidal current operation mode changes, the trained intelligent body can quickly decide the attack line which is most prone to cause the large-range cascading failure, evaluate the weakest line in the current state and guide the power grid operation personnel to execute the relevant scheduling operation.
Preferably, the intelligent agent is composed of two deep neural networks with the same structure;
the Q estimation network is used for deciding an attack line in the current power flow state, and the Q target network is used for evaluating the accuracy of the Q estimation network on Q value estimation in the training process.
The Q estimation network and the Q target network have the same structure, the Q estimation network is randomly initialized, then the whole set of network parameters are given to the Q target network, and then the whole set of network parameters are given to the Q target network again after the Q estimation network is updated for a certain number of times (generally, 50 times). The delay updating of the Q target network can relieve the over-estimation of the Q value of the action in the next state in the process of training the intelligent agent. Therefore, the method is beneficial to the stability of the training process of the Q estimation network.
Preferably, step 120 includes:
the transmission active power of the power transmission line, the load active power and the generator active power obtained by the direct current power flow calculation form a one-dimensional vector which is used as state input information of the intelligent agent; compared with the table lookup method Q learning, the input state of the reinforcement learning algorithm agent can only be discrete, and the input state of the agent adopting the deep neural network to estimate the action Q value can be continuous. Because the operation mode change of the power system is continuous, the dimension disaster problem can be generated by discretization processing of the input state, and the dimension disaster problem can not be generated because the intelligent agent based on the deep Q network supports the input state as continuous quantity.
Q estimation network calculates Q value corresponding to attack each transmission line, corrects Q value calculated by output layer according to current transmission line state of power system, if the line quits operation, Q value corresponding to the line is negative infinite; if the line is in normal operation, the Q value corresponding to the line maintains the value calculated by the estimation network. By the method, the attack line obtained by the decision of the intelligent agent can be effectively prevented from being a line which is already stopped, namely, an invalid decision is avoided.
And deciding the attack line in the current state according to the probability that the candidate normal operation line corresponding to the maximum Q value is selected as 1-epsilon and the probability that the candidate normal operation line corresponding to the non-maximum Q value is selected as epsilon. By adopting an epsilon-greedy strategy, the intelligent Q network can balance the utilization of learning knowledge and the exploration solution space during decision, so that the Q learning training process can be prevented from falling into local optimization, and the accuracy of the identification result of the weak line is guaranteed.
Preferably, in steps 130 and 140, a direct current power flow cascading failure model is used, and a power flow state of the power system without the power flow out-of-limit of the power transmission line, which is finally stabilized after the power flow transfer caused by the quitting operation of the attacked line causes the cascading failure of the power system, is obtained through simulation calculation.
Preferably, step 130 includes: in the power system, after an attacked line exits from operation, detecting each formed sub-network; adjusting the active output and the active load of the generator of each sub-network; and calculating the direct current power flow of the power system based on the power flow calculation data adjusted by each sub-power system.
When the sub-network is detected, if the sub-network is a passive network, the states of all buses and power transmission lines in the network are modified to be off. Fig. 2 shows a flow of adjusting the active power output and the active load of the generator in each sub-network, which specifically includes:
(1) counting the total active load P within a subnetwork1Active power output P of each generatorgAnd upper limit PmaxAnd a lower limit Pmin；
(2) Calculating the active power unbalance amount delta P of the power system as Pl-∑PgIf Δ P > 0 and Δ P > ∑ (P)max-Pg) If the power is seriously deficient, increasing the active power of the generator to be insufficient to meet all active load requirements, cutting off the load, recalculating the active power unbalance delta P after load cutting, and turning to the step (4), otherwise, turning to the step (3) without serious power deficiency;
(3) if Δ P < 0 and | Δ P | > ∑ (P)g-Pmin) If the power is seriously surplus, the reduction of the active power output of the generator is difficult to meet the active power balance constraint, and the capacity (P) needs to be adjusted according to the power of the generatorg-Pmin) Cutting off the generator in an ascending order until the requirement of | < delta | is less than or equal to sigma (P)g-Pmin) Recalculating the active power unbalance amount delta P after the generator tripping, and turning to the step (4), otherwise, directly turning to the step (4) without serious surplus of power;
(4) and adjusting the active output of the generator according to the delta P and the speed regulation coefficient of the generator in proportion, and transferring the residual quantity of the active adjustment quantity originally distributed to the generator to other generators if the active output of the generator reaches the upper limit or the lower limit in the adjustment process.
Preferably, step 140 comprises: judging whether a line with a power flow out-of-limit exists or not based on the direct current power flow calculated in the step 130, if so, cutting off the line with the most serious power flow out-of-limit, and executing the step 130 again, otherwise, executing the step 150;
preferably, step 150 comprises:
(1) the intelligent agent decides a power flow state before tangent, a tangent decision, a power flow state after tangent, an immediate return function and a Markov Decision (MDP) end mark as a reinforced learning sample and stores the reinforced learning sample in an intelligent agent model training sample library;
(2) and extracting a certain amount of sample data from the model training sample library regularly, training the intelligent agent, and updating the network parameters of the intelligent agent according to the loss function optimization equation.
Wherein, the total number N of the lines exiting from the operation is determined according to the power flow state after the tangent linetotalWhether the first threshold is reached (50% of the total number of lines can be taken in a small system) determines the MDP end flag TdoneIf the operation is finished, the value is 1, otherwise, the value is 0, and the number N of newly-added and quitted-operation lines is combinednewDetermining the immediate return function r to be 0.5 (N)new-1)+5Tdone。
And calculating the time difference error of the sample, and storing the time difference error as the attribute mark of the sample into the reinforcement learning sample library together.
Further, the time difference error calculation formula when the intelligent agent model is constructed by adopting a double-depth Q network (DBQN) algorithm is as follows:
Figure BDA0003138441530000101
wherein e is the time difference error of the sample, s is the current state of the sample, a is the current decision action of the sample, and thetaiEstimating network parameters for Q at the ith iterative training of the agent, Q (s, a; theta)i) Is the Q-estimated value of the sample, r is the immediate reward function of the sample, s 'is the next state of the sample, a' is the next decision action of the sample,
Figure BDA0003138441530000103
for the Q target network parameters at the ith iterative training of the agent,
Figure BDA0003138441530000102
the next decision action a 'to obtain the maximum Q value in the s' state evaluated for the sample with the Q estimation network,
Figure BDA0003138441530000104
the Q value obtained by executing the next decision action a 'in the s' state evaluated by the sample Q target network, gamma is a discount factor, and the value range is [0, 1 ]]。
And when the intelligent agent model is trained, extracting a certain amount of samples by adopting a preferential empirical sampling method according to the time difference errors of the samples. The greater the time difference error, the higher the probability of being selected. And a preferential experience sampling method is adopted, so that a sample with higher utilization value is extracted during training, and the training convergence process of the intelligent agent is accelerated.
Updating network parameters of a Q estimation network of the intelligent agent according to a loss function optimization equation by using the extracted samples;
when the Q estimation network parameter updating interval is a certain number of times (50 times can be taken), the Q estimation network parameter is given to the Q target network, and when the other Q estimation network updating times are less than the interval value, the Q target network parameter is unchanged;
recalculating and updating the time difference error of the extracted training sample by using the updated Q estimation network;
further, the loss function optimization equation is:
Figure BDA0003138441530000111
Figure BDA0003138441530000112
wherein L isi(θi) As a loss function of the extracted training samples, e is the time difference error of the samples,
Figure BDA0003138441530000113
is Li(θi) For thetaiPartial differential of (a) for correcting the agent network parameter thetai，
Figure BDA0003138441530000114
Is Q (s, a; theta)i) For thetaiPartial differential of (a). When the intelligent agent is constructed by adopting the DBQN algorithm, the Q estimation network is adopted for the action decision estimation of the next state, and the Q value which can be generated by the action in the next state is evaluated by using the Q target network, so that the phenomenon that the parameter fluctuation of the training process is large due to overhigh Q value estimation caused by adopting the Q target network simultaneously for the action estimation and the Q value evaluation of the next state can be avoided, and the stability of the training process of the intelligent agent is improved.
Preferably, step 160 comprises: judging whether the number of outage lines in the current state of the power system reaches a first threshold value, if not, executing a step 120, continuing to utilize an intelligent agent to decide a tangent line, if so, outputting an attack sequence of the power system under the initial tide, finishing weak assessment according to the number of power transmission lines contained in the sequence, and guiding the grid structure reinforcement planning and tide operation mode optimization formulation of the power system.
It should be noted that, as the number of training iterations increases, the Q value of the action decision is estimated more and more accurately by the Q estimation network of the agent, and the evaluation of the weakness of the power system by the attack sequence which leads to the grid collapse of the power system under a certain initial power flow operation, which is obtained by the decision, is more accurate through interaction with the cascading failure simulation model in the offline operation mode calculation.
Preferably, step 170 comprises: judging whether the number of times of updated iteration of the intelligent agent model parameters reaches a second threshold value or not, if not, executing the step 110, and continuously reading a new initial power flow; and if the number of training iterations of the intelligent agent reaches a second threshold value, storing and outputting the trained intelligent agent model parameters.
The offline trained intelligent agent can rapidly give a line fault which most easily causes serious cascading faults in the current power flow state after the power flow operation mode is changed in the online operation process of the power system, so that power system operation personnel are guided to pay attention to the power flow of the line, and the influence caused by power flow transfer after the line fault is reduced through power flow scheduling measures in advance.
For example, as shown in fig. 3, the IEEE standard new england 10 machine 39 bus system, in combination with a cascading failure simulation calculation model of the dc power flow thereof, evaluates the weakness degree of the system through a sequence attack. The system comprises 10 generators, and each generator is provided with a speed regulator and a power system stabilizer. The system comprises 46 transmission lines, and the number of neurons of the network output layer is estimated corresponding to the intelligent agent Q. The system contains 21 loads. Under the attack of the sequence, when one transmission line is attacked, the line exits from operation, and the caused power flow transfer and the influence thereof are obtained by the simulation of a cascading failure model.
The evaluation flow of the power system weakness evaluation method under the sequence attack provided by the embodiment is shown in fig. 1, and specifically includes the following steps:
(1) setting the structure parameter, greedy coefficient epsilon and learning rate alpha of the Q network of the intelligent agent as 0.001, discount factor gamma as 0.9, training extracted sample number as 32, Q target network updating frequency as 20, intelligent agent evaluation effect testing frequency as 50 and iteration number upper limit as 30000;
(11) the structural parameters of the Q network of the intelligent agent comprise: the network I has 5 layers, wherein the number of neurons in an input layer is consistent with the dimension of a one-dimensional vector of an input state, and is 46+10+21, 77, wherein 46 is the dimension representing active power transmitted on a power transmission line in the input vector, 10 is the dimension representing active power output of a generator in the input vector, 21 is the dimension representing active power load in the input vector, the number of neurons in an output layer is consistent with the number of the power transmission line, 46, and the number of neurons in a hidden layer in the other three layers is 256;
(12) the greedy coefficient epsilon is also given in a decreasing manner, with an initial value epsilon0Every 1 iteration, the decrement of the greedy coefficient is delta epsilon 0.0001, and finally, the greedy coefficient is stabilized at epsilonf＝0.1；
(2) The method comprises the steps that different initial operation working condition structures are constructed, a system is divided into 3 regions, loads in each region are set according to 4 proportionality coefficients of 0.8, 0.9, 1.0, 1.1 and the like, 64 load levels are obtained in total, active power output of each generator in the system is obtained by utilizing an optimal power flow algorithm under each load level, and power flow calculation data of different initial operation working conditions in 64 are obtained by means of parameters of a power transmission line and adjustment coefficients of the generators in the system; wherein, 10 working conditions are randomly extracted as a test working condition set, and the rest 54 working conditions are taken as a training working condition set.
(3) Randomly reading an initial power flow calculation data, calculating the direct current power flow of the power system, and simultaneously sorting the active power on the power transmission line, the active power output of the generator and the active power of the load in the settlement result to obtain an input state vector;
(4) making an attack line decision to generate a random number, and if the random number is smaller than epsilon, randomly selecting a power transmission line which does not quit operation in the current state as the current attack line; if the random number is larger than epsilon, calculating Q values for attacking the power transmission lines according to the input state vector, modifying the Q values corresponding to the power transmission lines which quit operation into negative infinity according to the current input state, and selecting the power transmission line with the maximum Q value as the current attack line;
(5) performing cascading failure simulation on the power system, calling a power system cascading failure simulation calculation model of the direct current power flow, and simulating a power flow transfer process after an attacked line exits from running to obtain power system power flow calculation data after power flow adjustment;
(51) in the power system, after an attacked line exits from running, each formed sub-network is detected, and if the sub-network is a passive network, the states of all buses and power transmission lines in the network are modified to be off-line;
(52) adjusting the active output and the active load of the generator of each sub-network;
(53) and calculating the direct current power flow of the power system based on the power flow calculation data adjusted by each sub-network.
(6) Judging whether the power transmission line with the out-of-limit power flow exists or not according to the adjusted power flow state, if so, cutting off the power transmission line with the most serious out-of-limit power flow, executing the step (5), and if not, directly executing the step (7);
(7) storing training samples, extracting a certain amount of samples to train the intelligent agent, and updating parameters of the intelligent agent;
(71) updating an MDP process ending mark, judging whether the number of the power transmission lines which are out of operation in the power system without the load flow out-of-limit lines and obtained by chain fault model simulation calculation after the attacked lines are cut off reaches 23 (a first threshold value is set to be 50% of a bus line), if so, the ending mark is 1, otherwise, the ending mark is 0;
(72) counting the number of newly-added and quitted-running power transmission lines under the current attack, and formulating an immediate return function under the current attack line according to an immediate return function formulation rule by combining an MDP process ending mark;
(73) calculating the time difference error of the current sample to be stored, and storing the time difference error as a sample attribute into a training sample library;
(74) extracting 32 samples from a sample library by adopting a prior empirical sampling method, calculating the correction quantity of the Q estimation network parameters according to a loss function optimization equation, and updating the Q estimation network parameters;
(75) judging whether the Q estimation network parameter updating times reach multiples of 20 times, if so, giving a complete set of parameters of the Q estimation network to the Q target network;
(76) whether the Q-cut estimation network parameter updating times reach multiples of 50 times or not is judged, if yes, the Q estimation network is combined with a direct current power flow cascading failure model to estimate the sequence attack length required by the collapse of the network frame under the working condition of 10 test sets, and the model with the optimal estimation effect is stored;
(8) judging whether the MDP process is finished or not, if not, continuing to execute the step (4), and deciding the next attack line by the agent; if the decision is finished, outputting an attack sequence which causes the grid frame breakdown of the power system and is decided by the agent;
(9) judging whether the iteration times of the updated parameters of the training model are 30000 times or not, if not, continuing to execute the step (3), and randomly extracting an initial operation working condition in the training working condition set to continue the training process; and if the iteration times reach 30000 times, outputting the intelligent agent model with the optimal evaluation effect in the training process.
(10) For each working condition in the test set, referring to an evaluation flow shown in fig. 1, obtaining attack lines through random decision until the net rack collapses, repeating each working condition for 5000 times, obtaining 33 total of the minimum number of attack lines under each test working condition, namely 4, 3, 4, 2 and 3, and taking the minimum number of attack lines as an approximate shortest attack sequence reference value for evaluating the decision effect of the intelligent agent;
(11) the length and the drawing of an attack sequence obtained by decision under the initial operation condition of a test set under a DBQN intelligent body sampled by prior experience in the training process are shown in a figure 4;
as can be seen from fig. 4, the sum of the lengths of the shortest attack sequences required for causing the rack crash in 10 initial operating conditions of the test set is 33, the number of times of iterative update of the DBQN agent adopting the prior empirical sampling increases with the training process, the sum of the lengths of the attack sequences obtained by the decision in the initial operating conditions of the test set gradually decreases, and gradually approaches the minimum value 33, wherein the sum of the lengths of the attack sequences determined by the optimal parameter model in the training process of the agent can reach 36, that is, the agent makes 0.3 more decisions in the average initial operating condition of the test set. The embodiment illustrates that the optimal intelligent agent model obtained by training the power system weakness evaluation method under the sequence attack can provide an attack line which most easily causes the grid frame breakdown of the power system, the shortest attack sequence which causes the grid frame breakdown can be obtained by matching with the cascading failure model, and the accuracy of evaluating the weakness degree of the power system is high.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (10)

1. An intelligent training method for evaluating weakness of a power system is characterized by comprising the following steps:
s1, reading the load flow calculation data and obtaining the load flow state of the power system;
s2, the intelligent agent decides the number of the attack line according to the current power flow state;
s3, executing line attack, performing sub-network detection and pretreatment of the power system after disconnection, adjusting power flow calculation data of the power system, and calculating new power flow of the power system;
s4, judging whether a power flow out-of-limit line exists or not, if yes, cutting off any power flow out-of-limit line, returning to the step S3, and if not, continuing to execute the step S5;
s5, adding the current training sample data into an intelligent agent model training sample library, training and updating intelligent agent parameters;
s6, judging whether the power flow state reaches a first termination condition, if not, returning to the step S2, and if so, outputting an attack sequence obtained by the decision of the intelligent agent;
and S7, judging whether the intelligent agent training process reaches a second termination condition, if not, returning to the step S1, and if so, outputting the trained intelligent agent model.
2. The power system weakness assessment agent training method according to claim 1, wherein the power flow calculation data comprises:
information of the power transmission line: the serial numbers of the left and right buses, the reactance resistance capacitance, the transformer transformation ratio, the rated transmission capacity and the line running state;
bus information: the active power output, the load active power and the bus networking state of the generator;
generator information: and (5) a generator speed regulation coefficient.
3. The training method for the intelligent agent for evaluating the weakness of the power system as claimed in claim 1, wherein the intelligent agent is represented by a deep Q network, the dimension of an output layer is the number of the transmission lines which can be attacked by the power system, and the deep Q network calculates the Q value of the output neuron corresponding to each transmission line by using the input state data.
4. The power system weakness assessment agent training method according to claim 1, wherein in the step S3, the sub-network detection and preprocessing comprises:
if the power system is divided into more than 2 sub-networks after the disconnection, respectively bringing the power transmission line, the generator and the load into the corresponding sub-networks, keeping the power balance of each sub-network, and having no power interaction with each other;
if no generator is arranged in the sub-network, when the sub-network is processed, the transmission line state of the sub-network is changed from running to shutdown, and the bus networking state is changed from network access to network disconnection.
5. The training method for power system weakness assessment agent according to claim 4, wherein in the step S3, the adjusting power system load flow calculation data includes:
if the active power in the current sub-network is unbalanced, proportionally distributing the amount of the active power unbalance according to the speed regulation coefficient of the generator with the frequency regulation capability, if the active power is surplus, reducing the active power output of the generator, and if the active power is insufficient, increasing the active power output of the generator;
if the adjusted active output of the generator exceeds the upper limit of the active output of the generator, setting the active output of the generator as the upper limit of the active output of the generator, and bearing the residual active output up-regulation quantity by other generators; if the active output power up-regulation quantity still remains, reducing the load to achieve power balance;
if the adjusted active output of the generator is lower than the lower limit of the active output of the generator, setting the active output of the generator as the lower limit of the active output of the generator, and bearing the remaining active output down-regulation quantity by other generators; if the active power output down-regulation quantity still remains, the generator is cut off to achieve power balance;
if all the generators in the sub-network are cut off after the adjustment, the sub-network correspondingly adjusts the states of the power transmission line and the bus according to the output of the passive network.
6. The power system weakness assessment agent training method according to claim 1, wherein the step S5 includes:
taking the power flow state before the decision of the intelligent agent is cut, the decision of the cut, the power flow state after the cut, an immediate return function and a Markov decision process ending mark as a reinforced learning sample and storing the reinforced learning sample into an intelligent agent training sample library;
and extracting sample data from the intelligent agent training sample library by adopting a prior experience sampling method.
7. The power system weakness assessment agent training method according to claim 1, wherein the determining whether the power flow state reaches a first termination condition comprises:
and judging whether the total number of the current outage lines reaches a preset scale or whether the loss load reaches a preset proportion.
8. The power system vulnerability assessment agent training method of claim 1, wherein whether the outage agent training process reaches a second termination condition comprises:
and judging whether the update times of the intelligent agent parameters or the initial power flow data read in an accumulated mode reach a preset threshold value or not.
9. A method for evaluating the weakness of an electric power system is characterized by comprising the following steps:
(1) reading initial load flow calculation data, and acquiring the current initial load flow state of the power system;
(2) an agent obtained by training by adopting the method of any one of claims 1-8, and determining the number of an attack line according to the current power flow state;
(3) performing line attack, performing sub-network detection and pretreatment of the power system after disconnection, adjusting power system load flow calculation data, and calculating new power system load flow;
(4) judging whether a power flow out-of-limit line exists or not, if so, cutting off any power flow out-of-limit line, returning to the step (3), and if not, continuing to execute the step (5);
(5) judging whether the power flow state reaches a first termination condition, if not, returning to the step (2), and if so, outputting an attack sequence obtained by the decision of the intelligent agent;
(6) and evaluating the weakness degree of the power system in the initial power flow state according to the length of the attack sequence.
10. An electrical power system weakness assessment system, comprising: a computer-readable storage medium and a processor;
the computer-readable storage medium is used for storing executable instructions;
the processor is configured to read executable instructions stored in the computer-readable storage medium and execute the power system weakness assessment method of claim 9.
CN202110731653.7A 2021-06-29 2021-06-29 Intelligent training method, evaluation method and system for weak evaluation of power system Active CN113452026B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110731653.7A CN113452026B (en) 2021-06-29 2021-06-29 Intelligent training method, evaluation method and system for weak evaluation of power system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110731653.7A CN113452026B (en) 2021-06-29 2021-06-29 Intelligent training method, evaluation method and system for weak evaluation of power system

Publications (2)

Publication Number Publication Date
CN113452026A true CN113452026A (en) 2021-09-28
CN113452026B CN113452026B (en) 2022-09-20

Family

ID=77814232

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110731653.7A Active CN113452026B (en) 2021-06-29 2021-06-29 Intelligent training method, evaluation method and system for weak evaluation of power system

Country Status (1)

Country Link
CN (1) CN113452026B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114123178A (en) * 2021-11-17 2022-03-01 哈尔滨工程大学 Intelligent power grid partition network reconstruction method based on multi-agent reinforcement learning
CN114266487A (en) * 2021-12-24 2022-04-01 国网湖北省电力有限公司经济技术研究院 Transformer substation fault handling method suitable for digital handover scene
CN114386330A (en) * 2022-01-14 2022-04-22 合肥工业大学 Power grid cascading failure prediction method based on Q learning network attack

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017004626A1 (en) * 2015-07-01 2017-01-05 The Board Of Trustees Of The Leland Stanford Junior University Systems and methods for providing reinforcement learning in a deep learning system
WO2017028631A1 (en) * 2015-08-19 2017-02-23 中国电力科学研究院 Multi-fault coordinated integrative decision-making assistance method for static security and storage medium

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2017004626A1 (en) * 2015-07-01 2017-01-05 The Board Of Trustees Of The Leland Stanford Junior University Systems and methods for providing reinforcement learning in a deep learning system
WO2017028631A1 (en) * 2015-08-19 2017-02-23 中国电力科学研究院 Multi-fault coordinated integrative decision-making assistance method for static security and storage medium

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
JUN YAN ET AL: "Q-Learning-Based Vulnerability Analysis of Smart", 《IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY》 *
曾令康等: "基于双Q学习的考虑暂态稳定约束的电网薄弱线路辨识", 《中国电机工程学报》 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114123178A (en) * 2021-11-17 2022-03-01 哈尔滨工程大学 Intelligent power grid partition network reconstruction method based on multi-agent reinforcement learning
CN114123178B (en) * 2021-11-17 2023-12-19 哈尔滨工程大学 Multi-agent reinforcement learning-based intelligent power grid partition network reconstruction method
CN114266487A (en) * 2021-12-24 2022-04-01 国网湖北省电力有限公司经济技术研究院 Transformer substation fault handling method suitable for digital handover scene
CN114386330A (en) * 2022-01-14 2022-04-22 合肥工业大学 Power grid cascading failure prediction method based on Q learning network attack
CN114386330B (en) * 2022-01-14 2024-02-20 合肥工业大学 Power grid cascading failure prediction method based on Q learning network attack

Also Published As

Publication number Publication date
CN113452026B (en) 2022-09-20

Similar Documents

Publication Publication Date Title
CN113452026B (en) Intelligent training method, evaluation method and system for weak evaluation of power system
CN111628501B (en) AC/DC large power grid transient voltage stability assessment method and system
CN111191918A (en) Service route planning method and device for smart power grid communication network
CN111159922B (en) Key line identification method and device for cascading failure of power system
CN110460091B (en) Method for acquiring optimal planning of power transmission network under new energy access
CN105610171B (en) Minimum startup sequence optimization method based on unit load dynamic reactive response
CN112630659A (en) Lithium battery SOC estimation method based on improved BP-EKF algorithm
CN109657913B (en) Transmission and distribution network joint risk assessment method considering distributed power supply
CN110165667B (en) Transmission and distribution cooperative reactive power optimization method and system considering static voltage safety constraint
JP2007240521A (en) Method of calculating state quantity of secondary battery
CN112288326A (en) Fault scene set reduction method suitable for toughness evaluation of power transmission system
CN115588983B (en) Transient voltage stability assessment method
CN110783964A (en) Risk assessment method and device for static security of power grid
CN112310980A (en) Safety and stability evaluation method and system for direct-current blocking frequency of alternating-current and direct-current series-parallel power grid
CN110163540A (en) Electric power system transient stability prevention and control method and system
CN109873406B (en) A kind of electric system weakness route discrimination method
CN112821424A (en) Power system frequency response analysis method based on data-model fusion drive
CN116316611B (en) Power supply method and system based on low-voltage transformer area
CN113625118B (en) Single-phase earth fault line selection method based on optimized pulse neural membrane system
CN114565058A (en) Training method, device, equipment and medium for island detection model
CN102231169A (en) Agent map-based brittleness self-organization criticality analysis method of marine electric power system
CN112165104B (en) Multi-station fast switching capacitor control method
CN107767003B (en) Receiving-end power grid stability assessment method
CN112653129B (en) Transient power angle stability margin estimation method, device and system
CN110957742B (en) Electric power system small-interference stable auxiliary decision-making method and system based on deep 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