CN113890098A - Micro-grid unplanned island switching method based on artificial emotion SARSA learning - Google Patents

Micro-grid unplanned island switching method based on artificial emotion SARSA learning Download PDF

Info

Publication number
CN113890098A
CN113890098A CN202111235755.6A CN202111235755A CN113890098A CN 113890098 A CN113890098 A CN 113890098A CN 202111235755 A CN202111235755 A CN 202111235755A CN 113890098 A CN113890098 A CN 113890098A
Authority
CN
China
Prior art keywords
load
function
emotion
learning
power
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.)
Pending
Application number
CN202111235755.6A
Other languages
Chinese (zh)
Inventor
王灿
田福银
应宇辰
李欣然
褚四虎
徐恒山
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.)
China Three Gorges University CTGU
Original Assignee
China Three Gorges University CTGU
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 China Three Gorges University CTGU filed Critical China Three Gorges University CTGU
Priority to CN202111235755.6A priority Critical patent/CN113890098A/en
Publication of CN113890098A publication Critical patent/CN113890098A/en
Pending legal-status Critical Current

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
    • H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
    • H02J3/388—Islanding, i.e. disconnection of local power supply from the 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
    • H02J3/12—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
    • H02J3/24—Arrangements for preventing or reducing oscillations of power in networks
    • H02J3/241—The oscillation concerning frequency
    • 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
    • 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)
  • Power Engineering (AREA)
  • Supply And Distribution Of Alternating Current (AREA)

Abstract

A micro-grid unplanned island switching method based on artificial emotion SARSA learning comprises the following steps: calculating the power shortage generated on the connecting line; constructing a micro-grid state information set and an action decision set; constructing a reward value function based on the load priority; establishing a desired reward function based on load priority; constructing an artificial emotion quantization function; constructing an artificial emotion coefficient output function; introducing a quadratic function to convert the emotion coefficient into the actual influence of SARSA learning, applying the actual influence to the learning rate of the SARSA learning agent, and outputting a time-varying learning rate; an update function of the state-action value function is constructed. By performing a load shedding action that maximizes the accumulation of the value function, the tie line power deficit is eliminated. The method can ensure the power supply reliability of important loads and maintain the stability of system frequency when the micro-grid generates an unplanned island due to the distribution network fault.

Description

Micro-grid unplanned island switching method based on artificial emotion SARSA learning
Technical Field
The invention belongs to the technical field of grid-connected and off-grid switching of a microgrid, and particularly relates to a microgrid unplanned island switching method based on artificial emotion SARSA learning.
Background
The micro-grid is used as an effective way for the distributed power supply to be connected into the power distribution network, and not only can the increasing power demand of users be met, but also the on-site consumption of distributed energy resources can be promoted. The key for ensuring the power supply reliability of the micro-grid is to stably switch the micro-grid among different operation modes. When the power distribution network fails to cause an unplanned isolated island of the microgrid, if the output of the distributed power supply in the microgrid cannot make up for the power shortage on the connecting line, the system frequency in the microgrid can vibrate, and the safe and stable operation of the microgrid is influenced. In order to ensure safe and stable operation of the microgrid, it is necessary to deeply research an unplanned island switching method of the microgrid.
In the prior art documents:
an improved storage battery PQ and V/f control method is provided in the literature [1] study on smooth grid-connected and off-grid switching control strategy of light storage microgrid (Liumeng super, king pig iron, warm element, study on smooth grid-connected and off-grid switching control strategy of light storage microgrid [ J ] study on renewable energy sources 2020,38(12):1633 and 1639), and transient current impact in the switching process of grid-connected and off-grid of the microgrid is effectively reduced. The document does not take into account the power shortage of the system.
Document [2] coded controlled from grid-connected to isolated operation for the purpose of this/single-phase hybrid controlled and smooth (C.Wang, X.Li, T.Tian, et al. coded controlled from isolated operation for the purpose of this/single-phase hybrid controlled and smooth [ J ]. IEEE Transactions on Industrial Electronics 2020,67(3):1921 and 1931.) proposes an island strategy that coordinates the switching process with load shedding and inhibits the power oscillation after switching of the smart grid. However, the switching strategy only considers planned island switching, and does not consider the emergency situation that the power distribution network fails to cause unplanned island switching.
A micro-grid load shedding rate based on a hidden lift method is provided in a document [3] a stable control optimization generator tripping method based on a hidden lift method (Luzaki, Anzhiyi, Xuchou, Zhang Fangyuan, white poplar, Chenhongliang. [ J ] a stable control optimization generator tripping method based on a hidden lift method [ power system automation, 2016,40(05):139- & 144. ]), and the power shortage in an island micro-grid is rapidly eliminated. But this document does not take into account the priority of the load during offloading.
Disclosure of Invention
In order to solve the technical problems, the invention provides a microgrid unplanned island switching method based on artificial emotion SARSA learning, which can effectively ensure continuous power supply of important loads and maintain the stability of system frequency when dealing with microgrid unplanned island switching.
The technical scheme adopted by the invention is as follows:
a micro-grid unplanned island switching method based on artificial emotion SARSA learning comprises the following steps:
step 1: when the micro-grid is subjected to unplanned islanding due to the distribution network fault, the relationship between the output of the distributed power supply in the micro-grid and the power shortage on the connecting line is judged, and if the output of the micro-grid can balance the power shortage of the connecting line, the unplanned islanding process is finished. If the output of the microgrid cannot balance the power shortage on the tie line, entering the step 2;
step 2: the method comprises the following steps that power shortage of a microgrid connecting line, total active power output of photovoltaic, total active power output of stored energy and total active power demand of a load are used as state input of an emotional body in artificial emotion SARSA learning, a logic body of the SARSA learning divides a discrete state information set according to the power shortage of the connecting line, and the load is used as an action decision set;
and step 3: dividing the load into three priorities according to the influence degree of the load interruption on economy and society, and constructing a reward value function based on the load priorities;
and 4, step 4: the emotion body perception system for SARSA learning is internally provided with the source load storage power information so as to output an artificial emotion coefficient, and the artificial emotion coefficient acts on the learning rate so as to generate a time-varying learning rate;
and 5: and constructing an updating function of the state-action value function, selecting the corresponding load shedding action when the value of the state-action value function is maximum, and eliminating the power shortage of the tie line by executing the load shedding action.
In step 1, the power shortage generated on the interconnection line is calculated, and the power shortage can be represented as:
PCL＝∑PES+∑PPV-∑PL
wherein: pCLIs the power deficit produced on the tie line; sigma PESThe sum of all energy storage powers in the micro-grid system; sigma PPVThe sum of all photovoltaic powers in the microgrid system; sigma PLIs the sum of all load powers in the microgrid system.
In the step 1, according to a formula PCL＝∑PES+∑PPV-∑PLAnd judging the relationship between the distributed power supply output in the microgrid and the power shortage on the connecting line. If PCLWhen the output of the distributed power supply in the microgrid is greater than the power shortage on the connecting line, the power balance in the microgrid can be realized by adjusting the output of the distributed power supply, and the problem of the power shortage does not exist in the microgrid; otherwise PCL<0, the output of the distributed power supply in the microgrid is smaller than the power shortage on the connecting line, and the problem of power shortage exists in the microgrid.
In the step 2, the power shortage of the microgrid connecting line, the total active power output of the photovoltaic, the total active power output of the stored energy and the total active power demand of the load are used as the state input of the emotion body in the artificial emotion SARSA learning in the step 4; the constructed microgrid state information set and action decision set,
the set of state information may be represented as:
S＝[S1,S2,…,Sn]
wherein: s is a state information set; snA discrete set of states that are power shortfalls for the tie line; s1Is the target state of the system.
The action decision set may be expressed as:
A＝[A1,A2,…An]
wherein: a is an action decision set; a. thenCorresponding load power removed in the network.
And in the step 3, the load is classified into G grade according to the influence degree of the load interruption on economy and society. G represents the maximum load level, and generally, G is 3. The primary load is an important load for uninterrupted power supply, and once the primary load is interrupted, huge economic loss is caused; the secondary load is an interruptible load, and certain economic loss is caused by interruption of the interruptible load; three levels of loading are adjustable non-critical loads. Where the larger G, the lower the priority of the load.
In step 3, based on the reward value function of the load priority, the reward value function may be represented as:
Figure BDA0003317444380000031
wherein: r istAs a function of the reward value; stIs in the current state; a istThe current action of the agent; paLoad power cut for action a; v*(a) Is the desired reward under action a.
Constructing a load priority based desired reward function, which can be expressed as:
Figure BDA0003317444380000032
wherein: g is the load priority; g ═ 1 represents a primary load, and is an important load for uninterrupted power supply; g ═ 2 denotes secondary load, which is interruptible load; g-3 represents a three-level load, which is an adjustable non-critical load.
In the step 4, the emotion body sensing system learned by the SARSA learns the stored power information, that is, the power shortage of the microgrid connecting line, the total active power output of the photovoltaic, the total active power output of the stored energy, and the total active power demand of the load.
In the step 4, an artificial emotion quantization function is constructed, and the artificial emotion quantization function can be expressed as:
Figure BDA0003317444380000033
wherein: f. ofnOutputting synthesis for the emotion function; thetaiThe method comprises the steps of storing power information for artificial emotion perceived source load in a microgrid; omegaiIs the weight of each information; lambda [ alpha ]iIs thetaiAnd ωiThe product of (a).
Constructing an artificial emotion coefficient output function, wherein the artificial emotion coefficient output function can be expressed as:
Figure BDA0003317444380000041
wherein: k is a radical ofnA range for maximizing artificial emotion; eta is an emotion coefficient; etamaxIs the maximum value of the affective coefficient.
In the step 4, a quadratic function is introduced, the emotion coefficient is converted into the actual influence of SARSA learning, the actual influence is acted on the learning rate of the SARSA learning intelligent agent, and the time-varying learning rate is output;
introducing a quadratic function to convert the emotion coefficients into the actual influence of SARSA learning:
Cf(η)＝Kaη2+Kbη
wherein: cfThe actual influence of the emotional coefficient; kaAnd KbIs the conversion factor.
The learning rate of the SARSA learning agent is acted on by actual influence, and the learning rate of time variation is output:
αη←αCf(η)
wherein: alpha is alphaηThe learning rate of the time variation after the artificial emotion acts; α is a learning rate.
In the step 5, the value of the state-action value function is selected to be the mostAnd when the load is large, the corresponding load shedding action is executed to eliminate the power shortage of the connecting line. In the state-action value function Q(s)t,at) When the value is maximum, a corresponding microgrid state information set and a load shedding action strategy set exist, and the load shedding action is executed to eliminate the power shortage of the tie line.
In the step 5, the step of processing the image,
an update function of the state-action value function is constructed, which can be expressed as:
Q(st,at)←Q(st,at)+αη(rt(st,at)+γQ(st,at)-Q(st,at))
wherein: q(s)t,at) As a function of state-action value; st is the current state; alpha is alphatThe current action of the agent; alpha is alphaηIs a time-varying learning rate based on artificial emotion; gamma is a discount factor; r ist(st,at) As a function of the prize value.
The invention discloses a micro-grid unplanned island switching method based on artificial emotion SARSA learning, which has the following technical effects:
1) aiming at the problem of power shortage during the switching of the micro-grid unplanned island, the method considers the load priority in the reward value function of the artificial emotion SARSA learning, and reasonably cuts off part of the load during the period of the unplanned island to ensure the power supply reliability of important load.
2) According to the method, the micro-grid source load storage power information is sensed by using the emotion body, and the time-varying learning rate based on artificial emotion is constructed so as to accelerate the exploration speed of the intelligent agent.
3) The invention provides a micro-grid unplanned island switching method based on artificial emotion SARSA learning. The method takes the load as a decision object, and can quickly eliminate the power shortage on the connecting line by executing the load shedding method based on the artificial emotion SARSA learning. Furthermore, load priority is taken into account in the reward value function to ensure the power reliability of important loads. Compared with the traditional switching method, the method disclosed by the invention can effectively ensure continuous power supply of important loads and maintain the stability of system frequency when the power shortage of the tie line occurs in the case of the unplanned island switching of the micro-grid.
Drawings
Fig. 1 is a flowchart of a microgrid unplanned island switching method.
Fig. 2 is a diagram of a microgrid topology with improved IEEE-13 nodes.
Fig. 3(a) is an example of a load shedding object configuration ratio (a load shedding method based on a hidden enumeration method) in the case of an unplanned island switching of a microgrid;
fig. 3(b) is a load shedding object composition ratio diagram (the method of the present invention) in the case of the micro grid unplanned islanding switching.
Fig. 4 is a system frequency waveform diagram during unplanned islanding switching of the microgrid.
Fig. 5(a) is a load shedding object composition ratio map (load shedding method based on the implicit method) when the photovoltaic power supply PV2 exits from the unplanned islanding switching of the microgrid;
fig. 5(b) is a load shedding object composition ratio graph (method of the present invention) when photovoltaic power source PV2 is considered to be out of operation for a microgrid unplanned island switch.
Fig. 6 is a waveform diagram of system frequency when photovoltaic power source PV2 is considered to exit the unplanned islanded switching of the microgrid in operation.
Fig. 7(a) is a decision result diagram of the method of the present invention in an unplanned islanding switching operation state of the microgrid;
FIG. 7(b) is a diagram of the decision results of the method of the present invention in the operating state where source storage is incorporated into the microgrid;
fig. 7(c) is a decision result diagram of the method of the present invention in a microgrid node branch disturbance operating state.
Detailed Description
A micro-grid unplanned island switching method based on artificial emotion SARSA learning comprises the following steps:
s1, calculating the power shortage generated on the connecting line, wherein the power shortage can be represented as:
PCL＝∑PES+∑PPV-∑PL
wherein: pCLIs the power deficit produced on the tie line; sigma PESThe sum of all energy storage powers in the micro-grid system; sigma PPVThe sum of all photovoltaic powers in the microgrid system; sigma PLIs the sum of all load powers in the microgrid system.
S2, constructing a state information set and an action decision set of the microgrid:
the microgrid state information set may be represented as:
S＝[S1,S2,…,Sn]
wherein: s is a state information set; snA discrete set of states that are power shortfalls for the tie line; s1Is the target state of the system.
The action decision set may be expressed as:
A＝[A1,A2,…An]
wherein: a is an action decision set; a. thenCorresponding load power removed in the network.
S3, constructing a reward value function based on the load priority, wherein the reward value function can be expressed as:
Figure BDA0003317444380000061
wherein: r istAs a function of the reward value; stIs in the current state; a istThe current action of the agent; paLoad power cut for action a; v*(a) Is the desired reward under action a.
S4, establishing an expected reward function based on the load priority, wherein the expected reward function can be expressed as:
Figure BDA0003317444380000062
wherein: g is the load priority; g ═ 1 represents a primary load, and is an important load for uninterrupted power supply; g ═ 2 denotes secondary load, which is interruptible load; g-3 represents a three-level load, which is an adjustable non-critical load.
S5, constructing an artificial emotion quantization function, wherein the artificial emotion quantization function can be expressed as follows:
Figure BDA0003317444380000064
wherein: f. ofnOutputting synthesis for the emotion function; thetaiThe method comprises the steps of storing power information for artificial emotion perceived source load in a microgrid; omegaiIs the weight of each information; lambda [ alpha ]iIs thetaiAnd ωiThe product of (a).
S6, constructing an artificial emotion coefficient output function, wherein the emotion output function can be expressed as:
Figure BDA0003317444380000063
wherein k isnA range for maximizing artificial emotion; eta is an emotion coefficient; etamaxIs the maximum value of the affective coefficient.
S7, introducing a quadratic function to convert the emotion coefficient into the actual influence of SARSA learning, applying the actual influence to the learning rate of the SARSA learning intelligent agent, and outputting the time-varying learning rate:
introducing a quadratic function to convert the emotion coefficients into the actual influence of SARSA learning:
Cf(η)＝Kaη2+Kbη
wherein: cfThe actual influence of the emotional coefficient; kaAnd KbIs the conversion factor.
The learning rate of the SARSA learning agent is acted on by actual influence, and the learning rate of time variation is output:
αη←αCf(η)
wherein alpha isηThe learning rate of the time variation after the artificial emotion acts; α is a learning rate.
S8, constructing an updating function of the state-action value function, wherein the updating function can be expressed as:
Q(st,at)←Q(st,at)+αη(rt(st,at)+γQ(st,at)-Q(st,at))
wherein: q(s)t,at) As a function of state-action value; stIs in the current state; alpha is alphatThe current action sum of the intelligent agent; alpha is alphaηIs a time-varying learning rate based on artificial emotion; gamma is a discount factor; r ist(st,at) As a function of the prize value.
Example (b):
fig. 1 is a flowchart of a microgrid unplanned island switching method. The micro-grid central controller detects the occurrence of an unplanned island by receiving terminal measurement information, sends out a control signal, and collects the source load and storage parameters in the micro-grid. Logic in the artificial emotion SARSA learning divides a discrete state information set according to the power shortage of a connecting line, removes loads as an action decision set, and constructs a reward function based on load priority. The emotion body in the artificial emotion SARSA learning generates an emotion coefficient by sensing source load storage power information of the micro-grid, acts the artificial emotion on the learning rate, and then generates the time-varying learning rate to accelerate the exploration speed of the intelligent body. The logic in the SARSA learning starts from an initial state, searches for and updates the Q value function by the time-varying learning rate, and eliminates the power shortage on the tie line by executing the load shedding operation that maximizes the Q value accumulation.
Wherein: the system parameters are as follows: rated power of photovoltaic power sources PV1 and PV2 in the microgrid is 70kW and 60kW respectively; the rated power of the energy storage devices BS1 and BS2 is 50kW and 130kW respectively; the rated power of the Load1-9 is respectively 35kW, 30kW, 40kW, 45kW, 60kW, 20kW, 25kW and 20 kW; loads Load1, Load4 and Load8 are primary loads, loads Load2, Load6 and Load9 are secondary loads, and loads Load3, Load5 and Load7 are tertiary loads; the tie line power is 130 kW.
Fig. 2 is a diagram of a microgrid topology with improved IEEE-13 nodes. This little electric wire netting contains 2 photovoltaic power, 2 energy memory, 9 loads. When the power distribution network breaks down to cause the switch to be disconnected, the micro-grid is passively switched into an unplanned island operation mode.
Fig. 3(a) and 3(b) are diagrams of load shedding object configuration ratios at the time of microgrid unintended islanding switching. Wherein: the load shedding result of the load shedding method based on the copy-out method is shown in fig. 3(a), and the load shedding result of the method of the present invention is shown in fig. 3 (b). After the Load shedding method based on the hidden enumeration method is executed, the cut-off proportions of the primary Load, the secondary Load and the tertiary Load are respectively 19.2%, 46.2% and 34.6%, and the Load shedding objects comprise Load5, Load6 and Load8, but the priority of the Load is not considered in the method, and the primary Load8 is selected as the Load shedding object, so that the important Load is powered off. Under the same condition, in the Load shedding objects of the method, the secondary Load and the tertiary Load respectively account for 26.9% and 73.1%, and the Load shedding objects comprise Load2, Load3, Load5 and Load 7.
Fig. 4 is a system frequency waveform diagram during unplanned islanding switching of the microgrid. After the load shedding method based on the implicit enumeration method is executed, the maximum value and the minimum value of frequency fluctuation are respectively 50.01Hz and 49.71 Hz; in the same case, after the method of the invention is carried out, the maximum value and the minimum value of the frequency fluctuation are 49.98Hz and 49.77Hz respectively, and the frequency fluctuation amplitude generated by the method of the invention is smaller. After the load shedding method based on the hidden enumeration method is executed, the frequency is recovered to the rated value of 50Hz after 0.29s, but the frequency can be recovered to the rated value of 50Hz only by 0.23s when the method is executed, and the method can shorten the frequency recovery time by 20.69%.
Fig. 5(a) and 5(b) are proportion diagrams of load shedding objects when the photovoltaic power source PV2 is taken out of operation in the micro grid unplanned island switching. The photovoltaic power source PV2 being out of service causes a greater power deficit, which can reach 190 kW. Among them, the load shedding result of the load shedding method based on the copy-out method is shown in fig. 5(a), and the load shedding result of the method of the present invention is shown in fig. 5 (b). After the load shedding method based on the implicit enumeration method is executed, the cut-off proportions of the primary load, the secondary load and the tertiary load are respectively 34.2%, 42.1% and 23.7%; after the method of the invention is executed, the cutting proportions of the secondary load and the tertiary load are respectively 50% and 50%. In the method, loads of Load2, Load3, Load5, Load6 and Load7 are selected for cutting, and primary loads are not included; and loads Load4, Load5, Load6, Load8 and Load9 are selected to be cut in the Load shedding method based on the hidden enumeration method, and loads Load4 and Load8 are all primary loads, so that the important loads are powered off.
Fig. 6 is a waveform diagram of system frequency when photovoltaic power source PV2 is considered to exit the unplanned islanded switching of the microgrid in operation. After the load shedding method based on the implicit enumeration method is executed, the maximum value and the minimum value of frequency fluctuation are respectively 50.02Hz and 49.59 Hz; in the same case, after the method of the present invention is performed, the maximum value and the minimum value of the frequency fluctuation are 49.99Hz and 49.65Hz respectively, and it can be seen that the frequency fluctuation amplitude caused by the method of the present invention is smaller. After the load shedding method based on the hidden enumeration method is executed, the frequency is recovered to the rated value of 50Hz after 0.42s, but the frequency can be recovered to the rated value of 50Hz only by 0.37s when the method is executed, and the method can shorten the frequency recovery time by 11.9 percent. The conclusion is consistent with the result of FIG. 4, and the superiority of the method of the invention is further verified.
Fig. 7(a), fig. 7(b) and fig. 7(c) are decision result diagrams of the method of the present invention under different operation states of the microgrid. When the microgrid is in an unplanned island, the decision result of the method is shown in fig. 7 (a); when t is 5s, the rated power of the photovoltaic modules is 50kW, 70kW and 90kW respectively, the energy storage elements and the load are merged into the microgrid at the nodes 3, 4 and 5, and the decision result of the method in the simulation case is shown in fig. 7 (b); when t is 9s, the branch of node 11 or 12 in the microgrid is disturbed, so that the loads Load7 and Load8 exit the system, the decision result of the method in the simulation case is shown in fig. 7(c), and the action number on the abscissa in fig. 7(c) represents the Load number. When the running state of the micro-grid changes along with the change of time, the Q values are converged near the reward value, and the distribution of the Q values is consistent with the change trend of the reward value, so that the method disclosed by the invention can obtain a better load shedding decision when the running state of the micro-grid changes.
In the method, an intelligent agent for artificial emotion SARSA learning carries out interactive learning through information of the operating environment of the microgrid, and a training data set of a load shedding decision is continuously enriched, so that the optimal load shedding decision can adapt to the change of the structure and the operating state of the microgrid. The test result fully shows that the method has stronger adaptability to the change of the micro-grid structure and the operation state.

Claims (8)

1. A micro-grid unplanned island switching method based on artificial emotion SARSA learning is characterized by comprising the following steps:
step 1: when the micro-grid is subjected to unplanned islanding due to the distribution network fault, judging the magnitude relation between the output of a distributed power supply in the micro-grid and the power shortage on a connecting line, and if the output of the micro-grid can balance the power shortage of the connecting line, ending the unplanned islanding process; if the output of the microgrid cannot balance the power shortage on the tie line, entering the step 2;
step 2: taking the power shortage of the microgrid connecting line, the total active power output of the photovoltaic, the total active power output of the stored energy and the total active power demand of the load as the state input of the emotional body in the artificial emotion SARSA learning, dividing a discrete state information set according to the power shortage of the connecting line, and taking the load as an action decision set;
and step 3: dividing the load into three priorities according to the influence degree of the load interruption on economy and society, and constructing a reward value function based on the load priorities;
and 4, step 4: the emotion body perception system for SARSA learning is internally provided with the source load storage power information so as to output an artificial emotion coefficient, and the artificial emotion coefficient acts on the learning rate so as to generate a time-varying learning rate;
and 5: and constructing an updating function of the state-action value function, and eliminating the power shortage of the connecting line by executing the load shedding action which enables the value function to be accumulated to be maximum.
2. The microgrid unplanned island switching method based on artificial emotion SARSA learning of claim 1, characterized in that: in step 1, the power shortage generated on the interconnection line is calculated, and the power shortage can be represented as:
PCL＝∑PES+∑PPV-∑PL
wherein: pCLIs the power deficit produced on the tie line; sigma PESThe sum of all energy storage powers in the micro-grid system; sigma PPVThe sum of all photovoltaic powers in the microgrid system; sigma PLIs the sum of all load powers in the microgrid system.
3. The microgrid unplanned island switching method based on artificial emotion SARSA learning of claim 1, characterized in that: in the step 2, the micro-grid state information set and the action decision set are constructed,
the set of state information may be represented as:
S＝[S1,S2,…,Sn]
wherein: s is a state information set; snA discrete set of states that are power shortfalls for the tie line; s1Is the target state of the system;
the action decision set may be expressed as:
A＝[A1,A2,…An]
wherein: a is an action decision set; a. thenCorresponding load power removed in the network.
4. The microgrid unplanned island switching method based on artificial emotion SARSA learning of claim 1, characterized in that: in step 3, based on the reward value function of the load priority, the reward value function may be represented as:
Figure FDA0003317444370000021
wherein: r istAs a function of the reward value; stIs in the current state; a istFor making an intelligenceThe current action of the energy body; paLoad power cut for action a; v*(a) Is the desired reward under action a;
constructing a load priority based desired reward function, which can be expressed as:
Figure FDA0003317444370000022
wherein: g is the load priority; g ═ 1 represents a primary load, and is an important load for uninterrupted power supply; g ═ 2 denotes secondary load, which is interruptible load; g-3 represents a three-level load, which is an adjustable non-critical load.
5. The microgrid unplanned island switching method based on artificial emotion SARSA learning of claim 1, characterized in that: in the step 4, an artificial emotion quantization function is constructed, and the artificial emotion quantization function can be expressed as:
Figure FDA0003317444370000023
wherein: f. ofnOutputting synthesis for the emotion function; thetaiThe method comprises the steps of storing power information for artificial emotion perceived source load in a microgrid; omegaiIs the weight of each information; lambda [ alpha ]iIs thetaiAnd ωiThe product of (a);
constructing an artificial emotion coefficient output function, wherein the artificial emotion coefficient output function can be expressed as:
Figure FDA0003317444370000024
wherein: k is a radical ofnA range for maximizing artificial emotion; eta is an emotion coefficient; etamaxIs the maximum value of the affective coefficient.
6. The microgrid unplanned island switching method based on artificial emotion SARSA learning of claim 1, characterized in that: in the step 4, a quadratic function is introduced, the emotion coefficient is converted into the actual influence of SARSA learning, the actual influence is acted on the learning rate of the SARSA learning intelligent agent, and the time-varying learning rate is output;
introducing a quadratic function to convert the emotion coefficients into the actual influence of SARSA learning:
Cf(η)＝Kaη2+Kbη
wherein: cfThe actual influence of the emotional coefficient; kaAnd KbIs the conversion coefficient;
the learning rate of the SARSA learning agent is acted on by actual influence, and the learning rate of time variation is output:
αη←αCf(η)
wherein: alpha is alphaηThe learning rate of the time variation after the artificial emotion acts; α is a learning rate.
7. The microgrid unplanned island switching method based on artificial emotion SARSA learning of claim 1, characterized in that: in step 5, an update function of the state-action value function is constructed, and the update function can be represented as:
Q(st,at)←Q(st,at)+αη(rt(st,at)+γQ(st,at)-Q(st,at))
wherein: q(s)t,at) As a function of state-action value; stIs in the current state; alpha is alphatThe current action of the agent; alpha is alphaηIs a time-varying learning rate based on artificial emotion; gamma is a discount factor; r ist(st,at) As a function of the prize value.
8. A micro-grid unplanned island switching method based on artificial emotion SARSA learning is characterized by comprising the following steps:
s1, calculating the power shortage generated on the connecting line, wherein the power shortage can be represented as:
PCL＝∑PES+∑PPV-∑PL
wherein: pCLIs the power deficit produced on the tie line; sigma PESThe sum of all energy storage powers in the micro-grid system; sigma PPVThe sum of all photovoltaic powers in the microgrid system; sigma PLThe sum of all load powers in the micro-grid system;
s2, constructing a state information set and an action decision set of the microgrid:
the microgrid state information set may be represented as:
S＝[S1,S2,…,Sn]
wherein: s is a state information set; snA discrete set of states that are power shortfalls for the tie line; s1Is the target state of the system; the action decision set may be expressed as:
A＝[A1,A2,…An]
wherein: a is an action decision set; a. thenLoad power correspondingly cut off in the network;
s3, constructing a reward value function based on the load priority, wherein the reward value function can be expressed as:
Figure FDA0003317444370000031
wherein: r istAs a function of the reward value; stIs in the current state; a istThe current action of the agent; paLoad power cut for action a; v*(a) Is the desired reward under action a;
s4, establishing an expected reward function based on the load priority, wherein the expected reward function can be expressed as:
Figure FDA0003317444370000041
wherein: g is the load priority; g ═ 1 represents a primary load, and is an important load for uninterrupted power supply; g ═ 2 denotes secondary load, which is interruptible load; g-3 represents a tertiary load, which is an adjustable non-essential load;
s5, constructing an artificial emotion quantization function, wherein the artificial emotion quantization function can be expressed as follows:
Figure FDA0003317444370000042
wherein: f. ofnOutputting synthesis for the emotion function; thetaiThe method comprises the steps of storing power information for artificial emotion perceived source load in a microgrid; omegaiIs the weight of each information; lambda [ alpha ]iIs thetaiAnd ωiThe product of (a);
s6, constructing an artificial emotion coefficient output function, wherein the emotion output function can be expressed as:
Figure FDA0003317444370000043
wherein k isnA range for maximizing artificial emotion; eta is an emotion coefficient; etamaxThe maximum value of the emotion coefficient;
s7, introducing a quadratic function to convert the emotion coefficient into the actual influence of SARSA learning, applying the actual influence to the learning rate of the SARSA learning intelligent agent, and outputting the time-varying learning rate:
introducing a quadratic function to convert the emotion coefficients into the actual influence of SARSA learning:
Cf(η)＝Kaη2+Kbη
wherein: cfThe actual influence of the emotional coefficient; kaAnd KbIs the conversion coefficient;
the learning rate of the SARSA learning agent is acted on by actual influence, and the learning rate of time variation is output:
αη←αCf(η)
wherein alpha isηThe learning rate of the time variation after the artificial emotion acts; alpha is the learning rate;
s8, constructing an updating function of the state-action value function, wherein the updating function can be expressed as:
Q(st,at)←Q(st,at)+αη(rt(st,at)+γQ(st,at)-Q(st,at))
wherein: q(s)t,at) As a function of state-action value; stIs in the current state; alpha is alphatThe current action sum of the intelligent agent; alpha is alphaηIs a time-varying learning rate based on artificial emotion; gamma is a discount factor; r ist(st,at) As a function of the prize value.
CN202111235755.6A 2021-10-22 2021-10-22 Micro-grid unplanned island switching method based on artificial emotion SARSA learning Pending CN113890098A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111235755.6A CN113890098A (en) 2021-10-22 2021-10-22 Micro-grid unplanned island switching method based on artificial emotion SARSA learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111235755.6A CN113890098A (en) 2021-10-22 2021-10-22 Micro-grid unplanned island switching method based on artificial emotion SARSA learning

Publications (1)

Publication Number Publication Date
CN113890098A true CN113890098A (en) 2022-01-04

Family

ID=79013449

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111235755.6A Pending CN113890098A (en) 2021-10-22 2021-10-22 Micro-grid unplanned island switching method based on artificial emotion SARSA learning

Country Status (1)

Country Link
CN (1) CN113890098A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104052078A (en) * 2013-03-12 2014-09-17 珠海优特电力科技股份有限公司 Regulation and control method and system for switching grid-connected operation mode to island operation mode of microgrid
CN109842148A (en) * 2018-02-01 2019-06-04 大全集团有限公司 A kind of micro-grid connection turns the stable control method of isolated operation mode
CN111817349A (en) * 2020-07-31 2020-10-23 三峡大学 Multi-microgrid passive grid-connected and off-grid switching control method based on deep Q learning
CN112186796A (en) * 2020-09-04 2021-01-05 三峡大学 Multi-microgrid non-planned island control method based on artificial emotion reinforcement learning

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104052078A (en) * 2013-03-12 2014-09-17 珠海优特电力科技股份有限公司 Regulation and control method and system for switching grid-connected operation mode to island operation mode of microgrid
CN109842148A (en) * 2018-02-01 2019-06-04 大全集团有限公司 A kind of micro-grid connection turns the stable control method of isolated operation mode
CN111817349A (en) * 2020-07-31 2020-10-23 三峡大学 Multi-microgrid passive grid-connected and off-grid switching control method based on deep Q learning
CN112186796A (en) * 2020-09-04 2021-01-05 三峡大学 Multi-microgrid non-planned island control method based on artificial emotion reinforcement learning

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
龙军;罗群;覃龙;: "低压微网孤岛模式下改进的下垂控制策略", 广西大学学报(自然科学版), no. 03, pages 754 - 763 *

Similar Documents

Publication Publication Date Title
Rehman et al. An advanced virtual synchronous generator control technique for frequency regulation of grid-connected PV system
Hong et al. Multiscenario underfrequency load shedding in a microgrid consisting of intermittent renewables
Zhou et al. Compartmentalization strategy for the optimal economic operation of a hybrid AC/DC microgrid
CN104037765B (en) The method of active power distribution network service restoration scheme is chosen based on improved adaptive GA-IAGA
CN112271727B (en) Fault recovery method for flexible power distribution network containing flexible soft switch
Nemati et al. Optimization of microgrids short term operation based on an enhanced genetic algorithm
CN115207935B (en) Reactive power coordination optimization method for improving transient voltage stability of voltage weak area
Amirtharaj et al. Optimal utilization of renewable energy sources in MG connected system with integrated converters: an AGONN approach
CN115719979A (en) Source load storage coordination control method and system for off-grid operation of new energy microgrid
Liu et al. Grid-friendly energy prosumers based on the energy router with load switching functionality
Santurino et al. Optimal coordinated design of under-frequency load shedding and energy storage systems
Kottas et al. New operation scheme and control of smart grids using fuzzy cognitive networks
CN110311386A (en) A kind of capacity configuration optimizing method based on the spontaneous idle and SVG reactive compensation of PSO new energy power station
Alanazi et al. Designing a new controller in the operation of the hybrid PV-BESS system to improve the transient stability
Al Jajeh et al. Islanding of a microgrid using a distributed multi-agent control system
CN112186796B (en) Multi-microgrid non-planned island control method based on artificial emotion reinforcement learning
Ashourian et al. Controlling and modeling power-electronic interface DERs in islanding mode operation micro grid
CN112039126A (en) Multi-time scale coordinated scheduling method and system for power distribution network containing distributed power supply
CN113890098A (en) Micro-grid unplanned island switching method based on artificial emotion SARSA learning
CN109412192A (en) From the back-to-back soft straight device operation method of energy storage multiterminal
Bao et al. Cooperative control strategy of multiple VSGs in microgrid based on adjacent information
Du et al. Two-step solution to optimal load shedding in a micro-grid
Joshi et al. Control uncertainty based LQR for frequency regulation using distributed energy resources
Nair et al. Satin bower bird algorithm for controller parameter optimization in an autonomous AC microgrid
Behera et al. Battery energy management and power control in microgrids using PI and fuzzy logic controller based SMA

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