CN103679292B - Electricity collaborative optimization method for double batteries of intelligent micro power grid - Google Patents
Electricity collaborative optimization method for double batteries of intelligent micro power grid Download PDFInfo
- Publication number
- CN103679292B CN103679292B CN201310695793.9A CN201310695793A CN103679292B CN 103679292 B CN103679292 B CN 103679292B CN 201310695793 A CN201310695793 A CN 201310695793A CN 103679292 B CN103679292 B CN 103679292B
- Authority
- CN
- China
- Prior art keywords
- network
- control strategy
- circulation
- battery control
- judgment
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 230000005611 electricity Effects 0.000 title claims abstract description 31
- 238000005457 optimization Methods 0.000 title claims abstract description 22
- 238000011217 control strategy Methods 0.000 claims abstract description 58
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 230000006870 function Effects 0.000 claims description 34
- 241001269238 Data Species 0.000 claims description 5
- 238000011156 evaluation Methods 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000001105 regulatory effect Effects 0.000 abstract 2
- 210000004027 cell Anatomy 0.000 description 38
- 238000004146 energy storage Methods 0.000 description 26
- 239000011159 matrix material Substances 0.000 description 8
- 230000006978 adaptation Effects 0.000 description 7
- 238000012549 training Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 230000007935 neutral effect Effects 0.000 description 3
- 238000010606 normalization Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000009825 accumulation Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000000875 corresponding effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000004899 motility Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 239000002253 acid Substances 0.000 description 1
- 239000013543 active substance Substances 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007717 exclusion Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 238000012905 input function Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000008929 regeneration Effects 0.000 description 1
- 238000011069 regeneration method Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses an electricity collaborative optimization method for double batteries of an intelligent micro power grid. The electricity collaborative optimization method comprises the steps that related parameters are initialized; overall circulation is started, and a network judgment weight is initialized; outer circulation is started, a battery control strategy is initialized, and the battery control strategy is regulated according to the actual electricity use conditions; internal circulation is started, the neural network weight is circularly corrected and judged through the collected system state and the regulated battery control strategy, and the performance of the current battery control strategy is evaluated by utilizing the network judgment weight; whether the internal circulation is finished or not is judged, if not, the internal circulation is repeated, if not, whether the outer circulation is finished or not is judged, if the outer circulation is not finished, the outer circulation is repeated, otherwise whether the overall circulation is finished or not is judged, and if the overall circulation is not finished, the overall circulation is repeated; after a searching program completely operates, the optimum battery control strategy is stored according to the performance obtained from judgment, and the electricity use cost is calculated.
Description
Technical field
The invention belongs to intelligent grid electric energy optimizing technical field, be specifically related to a kind of intelligent micro-grid double cell electric energy association
Same optimal control method.
Background technology
Accumulator, also known as rechargeable battery, be the conventional energy storage device of intelligent micro-grid user's residential terminal, storage chemistry
Energy, is directly translated into a kind of electrochemical equipment of electric energy if desired by chemical energy.Accumulator is to be standard by rechargeable
Design, realize recharging by reversible chemical reaction, each side such as integrated cost considers, energy storage accumulator mainly uses
Lead-acid accumulator.Its work process: the electric energy outside utilizing during charging makes internal active substance regeneration, is chemistry electrical power storage
Can, need again chemical energy to be converted into electric energy output during electric discharge, it is achieved the storing process of electric energy.
Batteries to store energy is an important ring of residential electric power energy storage section in intelligent grid, and has use temperature range
Extensively, the feature such as charge acceptance strong, life-span length, easy care, wherein energy storage battery is to ensure that smart grid security is stably transported
The visual plant of row, it is possible to emergency power supply is provided for residential customer, cuts down the power consumption of peak of power consumption period, reduction electrical network
Peak load difference, is the extremely important ingredient of intelligent grid.In actual application, on the one hand by energy storage electric power storage
The working mechanism in pond is analyzed double-energy storage battery and is filled (putting) electrical characteristics, on the other hand double negative according to intelligent micro-grid residential customer end
Double-energy storage battery is filled the parameters such as (putting) electricity order, time and carries out optimum control by the practical operation situation carried, and Intelligent Optimal is micro-
The double cell electric energy of electrical network residential customer end, reduces the peak load difference of electrical network, improves the efficiency of operation of power networks and reduces use
The actual electric cost at family.But in actual double cell collaborative optimization running, the load change of residential customer end relates to
Artificial subjective factor, it is difficult to accurately predict, time span is long simultaneously, controls effect the most fairly obvious and non-linear in the short time
The feature such as serious makes double cell electric energy work in coordination with optimization to be difficult to set up accurate mechanism model, assist to the battery power of intelligent grid
The biggest difficulty is brought with optimal control.Therefore, based on operation of power networks relevant historical data, design a set of effective double cell
Electric energy works in coordination with optimization control scheme, to realize reducing electrical network maximum load, promoting network load balance, the motility of raising electrical network
It is that intelligent grid develops the problem needing solution badly with electric cost compatible, reduction user.
Summary of the invention
(1) to solve the technical problem that
The technical problem to be solved is to utilize intelligent grid dual user load histories service data, Spot Price
Information and neutral net, build intelligent micro-grid double cell electric energy and work in coordination with optimal control network method, use and move based on self adaptation
The self-learning method of state planning, it is achieved the collaborative optimal control of double cell electric energy.
(2) technical scheme
The present invention proposes a kind of intelligent micro-grid double cell electric energy cooperative optimization method, comprising:
S1, double cell electric energy cooperative optimization method is carried out relevant parameter initialization, including Spot Price, a time period
Two interior terminal use's load datas and judge modular neural network parameter;
S2, unlatching global loops, initialize and pass judgment on network weight, uses random method to realize within the specific limits;
S3, unlatching outer circulation, initialize battery control strategy, and according to actual electricity consumption situation to described battery control strategy
It is adjusted;
Circulation in S4, unlatching, uses and performs dependence heuristic dynamic programming method, utilizes the system mode and tune collected
Battery control strategy circulation after whole is revised and is passed judgment on neural network weight, and utilizes this judge network weight to evaluate and test present battery control
The performance of system strategy;Judge whether current interior circulation completes, if it is not yet done, return circulation in step S4 continues, otherwise
Judge whether current outer circulation completes, if it is not yet done, return step S3 to continue outer circulation, otherwise judge that the current overall situation is followed
Whether ring completes, if it is not yet done, return step S2 to continue global loops, otherwise continues next step;
After S5, search utility run completely, preserved optimum battery control strategy also according to passing judgment on the performance obtained
Calculate electric cost.
(3) beneficial effect
The present invention, according to two user's history load operation data of load end and Spot Price information, builds double cell
Electric energy works in coordination with optimal control network method, the optimal control plan in using self adaptation dynamic programming method to predict the corresponding time period
Slightly, reduce the peak valley load difference of electrical network, improve motility and the compatibility of electrical network, reduce the electric cost of user simultaneously.
Accompanying drawing explanation
Fig. 1 is intelligent micro-grid dual-battery structure schematic diagram in the present invention;
Fig. 2 is the flow chart that in the present invention, intelligent micro-grid double cell electric energy works in coordination with optimal control method;
Fig. 3 be in the present invention ADHDP method realize block diagram.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with specific embodiment, and reference
Accompanying drawing, the present invention is described in further detail.
Fig. 1 is intelligent micro-grid dual-battery structure schematic diagram in the present invention.As it is shown in figure 1, intelligent micro-grid is by external public affairs
Common-battery net 1 provides external public electric wire net power supply, and pricing mode is Spot Price, is only house end and provides electric energy, it is impossible to reversely
Conveying electric energy;The electric energy that external utility network 1 provides can charge to energy-storage battery 2 and energy-storage battery 4, simultaneously can also be for living
Residence 3 and house 5 provide electric energy;Energy-storage battery 2 and energy-storage battery 4 can also be respectively house 3 and house 5 electric terminal provides electricity
Can, but intelligent dwelling 3 or house 5 synchronization can only by external public electric wire net 1, energy-storage battery 2 and energy-storage battery 4 in one
Powering for it, same residential terminal can not discharge by energy-storage battery 2 and energy-storage battery 4 simultaneously, simultaneously two energy-storage batteries it
Between can not charge mutually.Controller can determine at a time by external public electric wire net 1, energy storage according to actual operation result
In battery 2 and energy-storage battery 4, one or two are that the electric terminal of house 3 and house 5 is powered, it is also possible to by external public
Electrical network 1 is charged for energy-storage battery.It should be noted that this Fig. 1 is schematic sketch, actual intelligent micro-grid with
And user's residential terminal also includes other each workpieces, such as straight AC converter, transformator, controller and laundry
The residential electric power terminals such as machine, but it is all those skilled in the art and knows, and do not affect the control method of the present invention, therefore
It is not added with at this repeating.
Fig. 2 shows that the intelligent micro-grid double cell electric energy that the present invention proposes works in coordination with the flow chart of optimal control method.As
Shown in Fig. 2, the method includes the following steps performed successively:
S1, double cell electric energy cooperative optimization method is carried out relevant parameter initialization, including Spot Price, a time period
Two interior terminal use's load datas and judge modular neural network parameter;
S2, unlatching global loops, initialize and pass judgment on network weight, uses random method to realize within the specific limits;
S3, unlatching outer circulation, initialize battery and control parameter, enter according to the electric energy that double cell is stored by its initialization data
Row is optimized and revised;
Circulation in S4, unlatching, uses and performs dependence heuristic dynamic programming method (ADHDP), utilizes the system collected
State and battery control strategy, as passing judgment on the input data selecting module, are revised and are passed judgment on network weight, evaluate and test present battery control
System strategy performance, it is judged that in current, whether circulation completes, if it is not yet done, return circulation in step S4 continues, otherwise sentences
Whether disconnected current outer circulation completes, if it is not yet done, return step S3 to continue outer circulation, otherwise judges current global loops
Whether complete, if it is not yet done, return step S2 to continue global loops, otherwise continue next step;
S5, searching procedure run the best battery control strategy occurred afterwards completely and calculate electric cost.
Introduce each step above-mentioned separately below.
S1, double cell electric energy cooperative optimization method is carried out relevant parameter initialization, including Spot Price, a time period
Two interior terminal use's load datas and judge modular neural network parameter;
The double cell electric energy cooperative optimization method of the present invention is that self adaptation thoery of dynamic programming method based on data is carried out
Build, it is not necessary to concrete mathematical model, i.e. utilize self adaptation dynamic programming algorithm processing and basis during problem analysis
Process the spy that data characteristics adjusts processing method automatically, processing sequence makes itself and problem characteristic match with processing parameter of data
Point, filling when adjusting and optimizing residential customer end exists double cell (putting) electricity order and the method for amount.According to intelligent micro-grid
Middle electrical network and residential customer part relevant historical data, self adaptation dynamic programming can learn automatically, updates current
The structure of system optimizing control, it is achieved the electric energy of double cell is collaborative to be optimized and revised.
According to performing dependence heuristic dynamic programming algorithm and the demand of intelligent micro-grid, at intelligent micro-grid double cell
The starting stage of electric energy Cooperative Optimization Algorithm needs to carry out the initial work of parameter, runs for algorithm and carries out basic data deposit
Work.
The relevant parameter initial work of algorithm includes Spot Price, two terminal residential customer bearing powers and adaptive
Answer the parameter passing judgment on modular neural network in dynamic programming algorithm.
Spot Price: using simplest peak valley pricing strategy, i.e. peak of power consumption period electricity rates high, electricity consumption is low herein
Paddy period electricity rates are low, to encourage user to avoid the Peak power use period.Spot Price information data is with reference to China some areas
Spot Price data obtain, change an electricity consumption electricity price per hour, and electricity price data be normalized.
Two residential customer loads: with reference to the residential customer history electricity consumption data of power supply department collection accumulation, be analyzed
Process and obtain community building and determine the user load data of two ordinary residences in the time period, and the two load data is carried out
Normalized, it is simple to self adaptation dynamic programming method uses.
Algorithm parameter initializes: need to carry out the self adaptation dynamic programming method of intelligent micro-grid in the algorithm incipient stage
Double cell reserve of electricity initializes with initial quantity of electricity, and global iterative initializes with interior circulation, outer circulation number of times, discount factor and god
Through network structure initial work, wherein discount factor refers to the variable used in ADHDP algorithm, and this variable embodies
The performance of the previous moment system size on the impact of current time system, its span be (0,1].
Wherein, the structure arranging judge network is 7-20-1, and wherein 7 for passing judgment on network input layer number of nodes, and 20 for commenting
Sentencing network hidden node quantity, 1 for passing judgment on network output layer number of nodes, and its hidden node quantity can rule of thumb be adjusted
Whole to obtain optimal Approximation effect.
In actual application, Spot Price information can be obtained by the real time data that power department is issued, user
The historical data accumulation of load can be used to again transport the intelligent micro-grid electric energy Cooperative Optimization Algorithm of each time period
Calculating planning and adjusting, be allowed to more press close to current practice, algorithm can carry out learning adjusting and optimizing automatically.Therefore, reality herein
Time electricity price and two residential customer load datas close to practical situation.
S2, unlatching global loops, initialize and pass judgment on network weight, uses random method to realize within the specific limits;
(i.e. perform to rely on heuristic dynamic programming method according to intelligent micro-grid double cell electric energy Cooperative Optimization Algorithm
ADHDP), need that this is passed judgment on network and carry out network weight initial work, considering weight convergence speed with stable
In the case of property, within global iterative, determine weights W1、W2In the range of (-0.01,0.01), random initializtion obtains.Wherein
W1For passing judgment on the weights between network input layer node and hidden node, W2For pass judgment on network hidden node and output layer node it
Between weights.The input value passing judgment on network includes normalized Spot Price Cnorm(t), normalized user load Pload1_norm
(t) and Pload2_norm(t), the real-time electric energy E of normalized double cellb1_norm(t) and Eb2_normFilling of (t), battery 1 and battery 2
(putting) electric control strategy u1(t) and u2T (), output valve is performance index function.
Pass judgment on network weight to initialize within global iterative, judge can be re-started when each global loops starts
The initialization of network weight, thus on the basis of ensureing network stabilization and convergence rate, better ensure that the receipts of neutral net
Hold back, in order to find double cell optimal control policy as early as possible.
S3, unlatching outer circulation, initialize battery and control parameter u (t), the electricity stored double cell according to its initialization data
Adjustment can be optimized;
Intelligent micro-grid single battery control strategy is { 1: battery charges;0: battery is hung up;-1: battery is to this lateral load
Power supply (such as battery 1, load 1 is this lateral load);-2: battery is (such as battery 1, negative to heteropleural load supplying
Carry 2 and be heteropleural load;);-3: battery is simultaneously to this heteropleural load supplying (i.e. battery is simultaneously to load 1 and load 2 power supply) }.
Consider battery 1 and battery 2 and the practical situation of two loads, all control strategies of intelligent dwelling double-battery charge (putting) electricity
As shown in table 1, but in initialization procedure, there are 7 { (u1(t), u2(t)) }=(-1 ,-2), (-1 ,-3), (-2 ,-1), (-2 ,-
3), (-3 ,-1), (-3 ,-2), (-3 ,-3) } due to two pieces of accumulator, to fill the strategy mutual exclusion of (putting) electricity untenable, need again to adjust
Whole.
The feasible control scheme list of table 1 double cell
The initial work method of the control strategy of user's house end double cell is that, within outer circulation, control strategy exists
{ 1,0 ,-1, in-2 ,-3}, random initializtion obtains its control strategy.Outer circulation ensure that, when circulation starts every time, intelligence is micro-
The control strategy of the double cell of electrical network re-starts random initializtion so that the control strategy of double cell relates to as much as possible
All of control strategy.Table 2 and table 3 are with u1T ()={ 1,0 ,-1, illustrate energy-storage battery as a example by-2 ,-3} fills (putting) electricity
Control strategy adjusts, and works as u2(t) and u1When () clashes t, default priority considers u1(t)。
Table 2 u1(t)=and 1,0, control Developing Tactics table when-1 ,-2}
Table 3 u1(t)={ during-3}, control Developing Tactics table
After battery control strategy initializes, according to the practical situation of user power utilization, the battery in set time section is controlled
Strategy correspondingly adjusts, and the practical situation of wherein said user power utilization includes the electricity consumption number of residential customer in set time section
According to this and the electric energy memory state etc. of double cell, simultaneously the filling of double cell (putting) electrically with fill (putting) power consumption and also can calculate
Arrive.Adjustment mode after battery control strategy initializes illustrates as a example by energy-storage battery 1, and concrete adjustment mode is shown in Table 2 Hes
Table 3.
In the process, guaranteed output conservation is wanted between load and battery, electrical network
Pload1(t)+Pload2(t)=Pbattery1(t)+Pbattery2(t)+Pgrid(t)。
Wherein, Pload1T () represents the realtime power of load 1 (load 1 is all electrical appliances in house 1), Pload2(t)
Represent the realtime power of load 2 (load 2 is all electrical appliances in house 2), Pbattery1T () represents filling in real time of battery 1
(putting) electrical power, Pbattery2(t) represent battery 2 fill (putting) electrical power, P in real timegridT () represents the real-time merit obtained from electrical network
Rate.
Circulation in S4, unlatching, acquisition system state (including: normalized Spot Price Cnorm(t), normalized user
Load Pload1_norm(t) and Pload2_norm(t), the real-time electric energy E of normalized double cellb1_norm(t) and Eb2_norm(t)), and
Use and perform to rely on heuristic dynamic programming method (ADHDP), utilize the system mode and battery control strategy conduct collected
Pass judgment on the input data of network, revise the weights passing judgment on network, evaluate and test present battery control strategy performance, it is judged that circulation in current
Whether complete, if it is not yet done, return circulation in step S4 continues, otherwise judge whether current outer circulation completes, if
Not yet complete, then return step S3 and continue outer circulation, otherwise judge whether current global loops completes, if it is not yet done,
Return step S2 and continue global loops, otherwise continue next step;
Outer circulation start the S3 stage be calculated adjustment after double cell control strategy and accumulator fill (putting) electricity
Mode and quota data and utility function and system mode and battery control strategy matrix, wherein, utility function is according to battery
Control strategy obtains.
Fig. 3 show in the present invention perform rely on heuristic dynamic programming method (ADHDP) realize block diagram.Such as Fig. 3 institute
Showing, this stage utilizes and performs to rely on heuristic dynamic programming method according to network the input data, i.e. system mode being previously obtained
With battery control strategy matrix, training is passed judgment on mixed-media network modules mixed-media accordingly, is revised its network weight, and utilize judge network evaluation to work as
Front battery control strategy performance.
According to the Bellman principle of optimization, dynamic programming method is used for finding optimum control for complex nonlinear optimization problem
The behavior of system strategy.Working in coordination with optimization problem for intelligent micro-grid double cell electric energy, current system is represented by:
X (t+1)=F [x (t), u (t)].
X (t) represents that the state vector of t system (includes normalized Spot Price Cnorm(t), normalized user
Load Pload1_norm(t) and Pload2_norm(t), the real-time electric energy E of normalized double cellb1_norm(t) and Eb2_norm(t)), u
T () represents the control strategy of two accumulator of t, F is when t battery control strategy u (t), current system conditions x
T () is to the transfer function of next system mode x (t+1).
For evaluating the performance index function of the system of system quality it is:
Wherein U represents the utility function of system, and it is for representing expense produced by system power consumption at a time,
γ is discount factor, span (0,1].The target of dynamic programming is exactly to find optimum battery a making property of control strategy
Can minimize by target function.According to Bellman principle, optimal performance index function is:
Therefore the optimal control policy of t is
For in concrete intelligent micro-grid double cell electric energy cooperative optimization method, concrete parameter is expressed as follows:
System mode vector is
x(t)=[Cnorm(t), Pload1_norm(t), Pload2_norm(t), Eb1_norm(t), Eb2_norm(t)],
Wherein CnormT () is the Spot Price after normalization, Pload1_norm(t) and Pload2_normT () is the most normalized
User load 1 and user load 2, Eb1_norm(t) and Eb2_normT () is respectively battery 1 and the real-time energy storage of battery 2 after normalization
Electricity.
The control strategy of energy-storage battery is
u(t)=(u1(t), u2(t)),
Wherein u1(t) and u2What t () was respectively battery 1 and battery 2 fills (putting) electric control strategy.
Utility function computing formula is
U(t)=c(t)(Pload1(t)+Pload2(t)+Ebattery1_change(t)+Ebattery2_change(t))/Umax,
Wherein c (t) represents current time Spot Price, Pload1(t) and Pload2T () is respectively user load 1 and user bears
Carry 2, Ebattery1_change(t) and Ebattery2_changeT () fills (putting) in being respectively the current slot of energy-storage battery 1 and battery 2
Electricity electric energy, be negative value during electric discharge, during charging be on the occasion of.UmaxMaximum for utility function set in advance.
Performance index function formula according to system can deduce
J [x (t)]=U [x (t), u (t)]+γ J [x (t+1)].
Therefore, intelligent micro-grid double cell electric energy cooperative optimization method performs rely on heuristic dynamic programming and pass judgment on network
Target be by training minimize judge network output error.Owing to J [x (t+1)] is the performance indications letter of subsequent time
Number, is therefore difficult to obtain.For the ease of solving, we define new performance index function Q (t)=J [x (t+1)], make effectiveness letter
Number U [x (t), u (t)] is abbreviated as U (t), then we both can obtain:
Q(t-1)=U(t)+γQ(t)
It can be seen that performance index function Q (t) in above formula only performance with the utility function of t and a upper moment refers to
Scalar functions is correlated with, and unrelated with the performance index function in t+1 moment.We use judge network approximate solution Q (t) below.
Neural network forecast error is passed judgment in definition
ec(t)=γQ(t)+U(t)-Q(t-1)。
Q (t) is for passing judgment on the t output function of network, and Q (t-1) is the t-1 moment output function passing judgment on network, and γ is
Discount factor.
Judge network minimizes error function and is
The right value update rule then passing judgment on network obtains based on gradient descent method
wc(t+1)=wc(t)+Δwc(t),
Wherein wcT () represents the judge network weight of t, Δ wcT () represents its weights variable quantity, lcWhen () represents t t
Carve the learning rate passing judgment on network, lc(t)>0。
Pass judgment on network along with the increase of frequency of training, its forecast error ecT () gradually levels off to zero.Therefore, performance indications
Relation between function Q (t-1) and Q (t) can be approximately:
Q(t-1)≈U(t)+γQ(t)。
The structure passing judgment on network includes input layer, hidden layer and output layer three-decker, the activation between input layer and hidden layer
Function is bipolar S type function, the linear function of the activation primitive between hidden layer and output layer.Making hidden layer neuron number is L,
Output layer neuron number is M.System mode in known one section of set time section and battery control strategy matrix
Statecontrol and utility function matrix U, can obtain output according to first network calculations
Q=purline(W2*σ(W1* statecontrol)),
Wherein
σ(W1*statecontrol)∈RL,
purline(W2*σ(W1*statecontrol))∈RM, purline (z)=z.
And the performance index function matrix that Q is all moment Q (t) in this section of set time section, statecontrol=
[x, u], x represents all moment x (t) state vectors in this section of set time section, and it is all that u represents in this section of set time section
The control strategy in moment u (t), U is the utility function matrix of all moment U (t).
From front, the performance index function of previous moment t-1 can be calculated by following formula
Q(t-1)=U(t)+γQ(t)。
The target output value now passing judgment on network is Q (t-1), and input function matrix is statecontrol, can have
In the case of the inputoutput data determined, this is passed judgment on network and carry out a Weight Training, obtain the weights W trained1And W2, its
Middle W1For passing judgment on the weights between network input layer node and hidden node, W2For passing judgment on network hidden node and output layer node
Between weights.
The two weights W subsequently1And W2It is delivered to select module, selects module according to these weights and system mode matrix x
T () (includes normalized Spot Price, the realtime power of normalized user load 1 and the realtime power of load 2, normalization
The real-time electric energy of double cell) all possible battery control strategy combination u is calculated, it is thus achieved that corresponding select module letter
Number output valve, judges the best performance of current any control strategy according to the size of this output valve;Such as, if, with currently
System mode and pass judgment on the weights of network all battery control strategies calculated, the selection module that obtains output result is
Little control strategy is current optimum battery control strategy.Now select the hidden layer neuron number of module also for L, output
Layer neuron number is also M.
FS=purline(W2*σ(W1* statecontrol)),
Wherein
σ(W1*statecontrol)∈RL,
purline(W2*σ(W1*statecontrol))∈RM, purline (z)=z, and FSFor selecting modularity function.
According to all selection modularity functions output result obtained, find the selection modularity function output knot that functional value is minimum
Fruit also finds the control strategy u of its correspondence.According to control strategy u and two actual user load data, to intelligent micro-grid
Two pieces of batteries fill (putting) electricity order carry out planning and adjusting, according to formula with amount
Cost=(Pload1+Pload2+Ebattery1_change+Ebattery2_change)*C
Calculate the cost of user.Wherein C is to Spot Price matrixes all in this section of set time section, Ebattery1_change
And Ebattery2_changeBe respectively energy-storage battery 1 and battery 2 fills (putting) electricity electric energy, is negative value during electric discharge, during charging be on the occasion of.
Setting cycle-index according to interior circulation, if circulation not yet terminates in now, continues to return S4 step section start and continues
Continuous network weight of passing judgment on is trained, if interior circulation is over, now the inventive method may proceed to whether judge current outer circulation
Terminate.If outer circulation not yet terminates, then return to again the double cell control strategy of intelligent micro-grid be initialized at S3
Assignment, is optimized adjustment according to initialization result to the electric energy of two pieces of battery storage subsequently;If outer circulation terminates the most, then originally
Inventive method may proceed to judge whether current global loops terminates, if global loops not yet terminates, can return at S2 commenting
Sentence two weights W of modular neural network1、W2Carry out random initializtion, start to calculate next time;If global loops is the most
Terminate, then the method may proceed to carry out next step S5.
S5, searching procedure run the best battery control strategy occurred afterwards completely and calculate electric cost.
Pass judgment on after network trains every time and terminate in interior circulation, recalculate intelligent micro-grid according to the weights of this neutral net
The control strategy of double cell, is then adjusted correspondingly according to actual user's load and battery energy storage data, and according to phase
The result answered calculates the cost of double cell user.During the continuous circuit training of network, constantly calculate each training result
User cost value Cost, then with minima Cost of current system recordminCompare, it may be judged whether be better than currently known
Minimum user cost, if less than record value, then will record minima with currency update, do not update.Side of the present invention
Method continually searches for Neural Network Data when best battery control strategy and optimal user cost in a program, until network is instructed
Practice and terminate.
The advantage of control algolithms based on data maximum is Model free control.Real system exists serious non-linear, no
The factor such as definitiveness, time variation, when obtaining accurate mathematical model, model-free adaption optimal control just can
Play the advantage of self, overcome the contradiction between theoretical model and actual application, reduce the requirement to model and have preferably
Comprehensive Control effect.
Particular embodiments described above, has been carried out the purpose of the present invention, technical scheme and beneficial effect the most in detail
Describe in detail bright it should be understood that the foregoing is only the specific embodiment of the present invention, be not limited to the present invention, all
Within the spirit and principles in the present invention, any modification, equivalent substitution and improvement etc. done, should be included in the protection of the present invention
Within the scope of.
Claims (9)
1. an intelligent micro-grid double cell electric energy cooperative optimization method, comprising:
S1, double cell electric energy cooperative optimization method is carried out relevant parameter initialization, including Spot Price, in the time period
Two terminal use's load datas and judge modular neural network parameter;
S2, unlatching global loops, initialize and pass judgment on network weight, uses random method to realize in preset range;
S3, unlatching outer circulation, initialize battery control strategy, and carry out described battery control strategy according to actual electricity consumption situation
Adjust;
Circulation in S4, unlatching, uses and performs dependence heuristic dynamic programming method, after utilizing the system mode collected and adjusting
Battery control strategy circulation revise pass judgment on neural network weight, and utilize this judge network weight evaluation and test present battery control plan
Performance slightly;Judge whether current interior circulation completes, if it is not yet done, return circulation in step S4 continues, otherwise judge
Whether current outer circulation completes, if it is not yet done, return step S3 to continue outer circulation, otherwise judges that current global loops is
No complete, if it is not yet done, return step S2 to continue global loops, otherwise continue next step;
After S5, search utility run completely, preserved optimum battery control strategy according to passing judgment on the performance obtained and calculate
Electric cost.
2. the method for claim 1, it is characterised in that in step sl, Spot Price data are with reference to China partly
The Spot Price information in district and the Spot Price that obtains, two user load data messages are according to residential building electricity consumption historical data
Obtain.
3. the method for claim 1, it is characterised in that in step sl, the structure arranging judge network is 7-20-1,
Wherein 7 for passing judgment on network input layer number of nodes, and 20 for passing judgment on network hidden node quantity, and 1 for passing judgment on network output layer node
Quantity, and its hidden node quantity can rule of thumb be adjusted obtaining optimal Approximation effect.
4. the method for claim 1, it is characterised in that in step s 2, with random in the range of (-0.01,0.01)
Initial method obtains passes judgment on network weight.
5. the method for claim 1, it is characterised in that in step s 4, the input data of described judge network and defeated
Go out data to include:
Input data: normalized Spot Price Cnorm(t), normalized user load Pload1_norm(t) and Pload2_norm(t),
The real-time electric energy E of normalized double cellb1_norm(t) and Eb2_norm(t), the battery control strategy u of double cell1(t) and u2(t);
Output data: according to Bellman equation, utilize and pass judgment on output Q (t) in a moment and utility function U (t) and folding on network
Button factor gamma is calculated data Q (t-1) of current time, and computing formula is as follows:
Q (t-1)=U (t)+γ Q (t);
Wherein, utility function U (t) represents the cost that current time t institute power consumption produces.
6. the method for claim 1, it is characterised in that in step S4, utility function computing formula is:
U (t)=c (t) (Pload1(t)+Pload2(t)+Ebattery1_change(t)+Ebattery2_change(t))/Umax;
Wherein, c (t) represents current time Spot Price, Pload1(t) and Pload2T () is respectively user 1 bearing power and user 2
Bearing power, Ebattery1_change(t) and Ebattery2_changeT () is respectively rechargeable electrical energy in double cell current slot, electric discharge
Time be negative value, during charging be on the occasion of;UmaxMaximum for utility function set in advance.
7. the method for claim 1, it is characterised in that in step S4, the right value update rule passing judgment on network is as follows:
wc(t+1)=wc(t)+Δwc(t)
Wherein, wcT () represents the judge network weight of t, Δ wcT () represents its weights variable quantity, lcT () represents that t is commented
Sentence the learning rate of network, lc(t) > 0;Q (t) represents the output passing judgment on network, ecT () represents the forecast error passing judgment on network, Ec
T () minimizes error function for passing judgment on network.
8. the method for claim 1, it is characterised in that pass judgment on present battery control strategy performance in step S4 concrete
For: according to described judge network, utilize the neural network weight obtained that all of battery control strategy is evaluated, it is thus achieved that
Optimum battery control strategy under Current Situation of Neural Network weights and the evaluation of estimate of correspondence thereof.
9. the method for claim 1, it is characterised in that step S4 also includes, the optimum that the most current interior circulation obtains
Evaluation of estimate corresponding to battery control strategy and current optimal value, if the optimum battery control strategy that in current, circulation obtains is corresponding
Evaluation of estimate be better than optimal value, then update described optimal value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310695793.9A CN103679292B (en) | 2013-12-17 | 2013-12-17 | Electricity collaborative optimization method for double batteries of intelligent micro power grid |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310695793.9A CN103679292B (en) | 2013-12-17 | 2013-12-17 | Electricity collaborative optimization method for double batteries of intelligent micro power grid |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103679292A CN103679292A (en) | 2014-03-26 |
CN103679292B true CN103679292B (en) | 2017-01-11 |
Family
ID=50316776
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310695793.9A Active CN103679292B (en) | 2013-12-17 | 2013-12-17 | Electricity collaborative optimization method for double batteries of intelligent micro power grid |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103679292B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104022503B (en) * | 2014-06-18 | 2017-01-04 | 中国科学院自动化研究所 | A kind of intelligent micro-grid electric energy optimizing control method with energy storage device |
CN105071421B (en) * | 2015-08-17 | 2019-03-15 | 中国科学院自动化研究所 | Office Building Energy Consumption management method |
CN111200285B (en) * | 2020-02-12 | 2023-12-19 | 燕山大学 | Micro-grid hybrid coordination control method based on reinforcement learning and multi-agent theory |
CN111532150A (en) * | 2020-05-15 | 2020-08-14 | 国网辽宁省电力有限公司电力科学研究院 | Self-learning-based electric vehicle charging control strategy optimization method and system |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103138636A (en) * | 2011-11-28 | 2013-06-05 | 北汽福田汽车股份有限公司 | Hybrid power supply system for photovoltaic off network, photovoltaic combined network and energy storage |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8203445B2 (en) * | 2006-03-28 | 2012-06-19 | Wireless Environment, Llc | Wireless lighting |
-
2013
- 2013-12-17 CN CN201310695793.9A patent/CN103679292B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103138636A (en) * | 2011-11-28 | 2013-06-05 | 北汽福田汽车股份有限公司 | Hybrid power supply system for photovoltaic off network, photovoltaic combined network and energy storage |
Also Published As
Publication number | Publication date |
---|---|
CN103679292A (en) | 2014-03-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104022503B (en) | A kind of intelligent micro-grid electric energy optimizing control method with energy storage device | |
CN109103912B (en) | Industrial park active power distribution system scheduling optimization method considering power grid peak regulation requirements | |
CN105846461A (en) | Self-adaptive dynamic planning control method and system for large-scale energy storage power station | |
CN104538999B (en) | Photovoltaic generating energy storage system and energy scheduling method thereof | |
CN104200297B (en) | The energy-optimised dispatching method of household mixed power supply system under Spot Price environment | |
CN104485727B (en) | A kind of off-network photovoltaic generation energy-storage system and its energy dispatching method | |
CN105870950B (en) | A kind of multi-objective coordinated system of the adjustable resource of intelligent building and optimal control method | |
CN108347062A (en) | Microgrid energy based on gesture game manages distributed multiple target Cooperative Optimization Algorithm | |
Kavousi-Fard et al. | Optimal distribution feeder reconfiguration for increasing the penetration of plug-in electric vehicles and minimizing network costs | |
CN106058855A (en) | Active power distribution network multi-target optimization scheduling method of coordinating stored energy and flexible load | |
CN110070292B (en) | Micro-grid economic dispatching method based on cross variation whale optimization algorithm | |
CN105119312B (en) | A kind of photovoltaic energy storage dispatching method based on mixed integer nonlinear programming | |
CN103679292B (en) | Electricity collaborative optimization method for double batteries of intelligent micro power grid | |
CN107067190A (en) | The micro-capacitance sensor power trade method learnt based on deeply | |
CN110247411A (en) | A kind of user side instrumentation intelligence light storage plant capacity dynamic allocation method | |
CN109390949A (en) | A kind of household photovoltaic, energy storage and with can control method | |
CN112821470A (en) | Micro-grid group optimization scheduling strategy based on niche chaos particle swarm algorithm | |
CN110165666A (en) | A kind of active distribution network dispatching method based on IGDT | |
CN109494813A (en) | A kind of power dispatching method, electronic equipment and storage medium | |
CN114123256B (en) | Distributed energy storage configuration method and system adapting to random optimization decision | |
CN110021947B (en) | Distributed energy storage power system operation optimization method based on reinforcement learning | |
CN116093466A (en) | Ordered charging method, ordered charging device, computer readable storage medium and computer equipment | |
Harrold et al. | Battery control in a smart energy network using double dueling deep q-networks | |
CN115660346A (en) | Large-scale vehicle network interactive load regulation and control method based on edge control | |
CN116542137A (en) | Multi-agent reinforcement learning method for distributed resource cooperative scheduling |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |