CN109934470A

CN109934470A - It polymerize the information physical modeling and control method of extensive air conditioner load

Info

Publication number: CN109934470A
Application number: CN201910147401.2A
Authority: CN
Inventors: 张沛超; 王永权
Original assignee: Shanghai Jiaotong University
Current assignee: Shanghai Jiaotong University
Priority date: 2019-02-27
Filing date: 2019-02-27
Publication date: 2019-06-25
Anticipated expiration: 2039-02-27
Also published as: CN109934470B

Abstract

The present invention relates to a kind of information physical modelings and control method for polymerizeing extensive air conditioner load, this method comprises: in load layer, air conditioner load is considered as information physical system CPS, the air conditioner load Probability Control based on half markoff process is proposed, can guarantee the blocking time of air-conditioning and keeps the diversity of cluster state；And the coupled relation of air conditioner load physical process and control process is established using hybrid model, realize the autonomous control of air conditioner load；In polymerization quotient sheaf, the control method for coordinating that air conditioner load is polymerized to virtual regulating units is proposed based on market equilibrium mechanism, this method energy effective protection user data privacy simultaneously significantly reduces communication cost；Meanwhile virtual regulating units can provide the offer curve and pondage for being similar to conventional power unit to scheduling, to participate in the Optimized Operation of power grid layer.Compared with prior art, the present invention has many advantages, such as to significantly reduce communication cost, is convenient for Optimized Operation.

Description

It polymerize the information physical modeling and control method of extensive air conditioner load

Technical field

The present invention relates to the temperature control loads in electric system to participate in peak regulation technique, extensive empty more particularly, to a kind of polymerization Adjust the information physical modeling and control method of load.

Background technique

With the large-scale grid connection of wind-powered electricity generation and being continuously increased for load peak-valley difference, peak regulation service is maintaining electric system supply and demand Increasingly important role is played in balance.Traditional way is to increase generating set power output in the load peak period.But The peak load period is often shorter, in order to meet this portion requirements and it is increased power generation and power transmission and distribution investment utilization rate it is very low.In addition, adjusting The fired power generating unit [being detailed in document 1] that peak unit is mostly at high cost and the feature of environmental protection is poor.At the same time, with air conditioner load (air Conditioning load, ACL) based on accounting of the temperature control load in total load persistently increase.It is pre- according to International Energy Agency Meter will be equal to the current total electricity demand [being detailed in document 2] of China to the year two thousand fifty whole world air conditioning energy consumption.If air-conditioning can be made Peak regulation service is participated in load scale, is the effective means realizing peak of power consumption period network load and cutting down.

Has a possibility that more documents temperature control load participates in peak regulation at present.For example, document [3] is by Jiangsu The estimation for saving summer air conditioning load proposes the several method for cutting down peak load, and has estimated corresponding peak clipping effect；Document [4] frame and implementation that public building air conditioner load participates in peak regulation are proposed；Document [5] considers the public of certain scale Building air-conditioner load proposes the control strategy for stopping to cut load by floor wheel；Document [6] from the peak regulation cost of Generation Side, Propose the optimisation strategy that peak regulation is participated in comprising the flexible load including air conditioner load.

Temperature control load substantial amounts and position dispersion, generally pass through Load aggregation quotient (load aggregator, LA) and tune The interaction of degree center.In order to implement effectively to control, polymerization quotient needs to establish the cluster models of temperature control load.There are two classes representative at present Cluster models.The first kind is the deterministic models with state sequence model [being detailed in document 7-9] for representative.But such methods pair Temperature control load is controlled using direct load, needs to obtain equipment control.Second class is stochastic model.For example, document [10,11] By the relationship of Fokker-Planck establishing equation cluster state probability distribution and migration probability, to realize to air conditioner load The control of cluster；The Temperature Distribution of load in cluster is divided into several sections by document [12,13] proposition, then can using Ma Er Husband's chain describes the dynamic process of the temperature control load quantity transition in each section.But the above method such as needs to collect there are several limitations The model parameter and user preference [being detailed in document 13,14] of load [can only be detailed in text for the identical homotype temperature control load of parameter Offer 10,13] etc..In addition, there is also following common problems for model above and control method: 1) polymerizeing between quotient and temperature control load needs It frequently to communicate to obtain load condition and send control command, control cost is larger；2) no matter switch is taken to temperature control load [being detailed in document 7], desired temperature control [being detailed in document 8,9] or probability control [being detailed in document 12,13] are controlled, is all existed Information security issue is such as subject to implement load the malice control [being detailed in document 15] " with opening with pass "；3) it needs to collect The model parameter and user preference of temperature control load, it is difficult to guarantee the data-privacy of user；4) air-conditioning blocking time is not accounted for about Beam is easy aggravation equipment attrition [being detailed in document 12,13].In the above problem, polymerization quotient is primarily upon the first two problem, and user Then pay close attention to latter two problems.Solving the above problems for system is that the flexibility of temperature control load is able to the premise of large-scale application.

Summary of the invention

It is extensive empty that it is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide a kind of polymerizations Adjust the information physical modeling and control method of load.

The purpose of the present invention can be achieved through the following technical solutions:

A kind of information physical modeling and control method polymerizeing extensive air conditioner load, this method comprises:

In load layer, air conditioner load is considered as information physical system CPS, proposes the air-conditioning based on half markoff process Load Probability control method can guarantee the blocking time of air-conditioning and keep the diversity of cluster state；And utilize hybrid system The model foundation coupled relation of air conditioner load physical process and control process, realizes the autonomous control of air conditioner load；

In polymerization quotient sheaf, the coordination control that air conditioner load is polymerized to virtual regulating units is proposed based on market equilibrium mechanism Method processed, this method energy effective protection user data privacy simultaneously significantly reduce communication cost；Meanwhile virtual regulating units can be to Scheduling provides the offer curve and pondage for being similar to conventional power unit, to participate in the Optimized Operation of power grid layer.

Preferably, the hybrid model by continuous physical process in CPS and discrete control process into Row simultaneous models to obtain, specifically:

Airconditioning control process is departure process, is described using finite state machine, proposes a kind of Probability Control, is introduced u₀And u₁Two migration probability state of a control transfers, and blocking time t is set_lock, wherein t_lockInclude blocking time after closing t_offlockWith blocking time t after opening_onlock；

Indoor temperature change generated in case is continuous physical process, can be indicated with equivalent thermal parameter ETP model, second order expression It is as follows:

In formula, T_a、T_mAnd T₀Respectively indoor air temperature, indoor solid temperature and outdoor temperature；U_aFor indoor and outdoor etc. Imitate impedance；H_mEquivalent impedance between room air and solid；Q_aAnd Q_mRespectively indoor gas and solid heat transfer amount；Q_ac、Q_iWith Q_sRespectively air-conditioning heat output, indoor airflow heat output and solar radiation heat；C_aAnd C_mThe respectively heat of room air and solid Hold；M (t) is air-conditioner switch state, and m (t)=1 represents air-conditioning unlatching, and m (t)=0 represents air-conditioning closing；f₁,f₂For heat transfer coefficient.

Preferably, the Semi-Markov Process of the control process specifically:

1) in state spaceOn, as system leave state m, the probability into state n is p_mn, full Foot:

2) under the premise of the state of next entrance is n, there is distribution F until the time for occurring to shift from m to n_mn (t), which can be Arbitrary distribution；

Then it is in the distribution F of state m_m(t) meet:

The average waiting time of state m are as follows:

M=1 is enabled, 2,3,4 respectively indicate ON, and the state transfer of OFF, ONLOCK and OFFLOCK state, air conditioner load have One-way can then be obtained by formula (2):

p₁₄=p₄₂=p₂₃=p₃₁=1 (5)

If α_mnFor from state m to the mobility of state n, when the execution period Δ t of half markoff process is more than setting threshold When value, then have: u₀≈α₁₄Δt,u₁≈α₂₃Δ t, if the residence time of state 1,2 meets exponential distribution, that is, have it is without memory, Then F₁₄(t) and F₂₃(t) be respectively parameter be α₁₄And α₂₃Exponential distribution, it may be assumed that

The average waiting time of state 1,2 can be then obtained by formula (4) are as follows:

And state 3,4 residence time disobey exponential distribution, but following determining is worth:

T₃=t_onlock, T₄=t_offlock (8)

If semi-Markov chain be it is irreducible, after finite time, can be restrained in certain shape probability of state To the definite value unrelated with original state, the referred to as probability of stability；Opposite, the probability before converging to the probability of stability is referred to as temporary State probability, according to the property of semi-Markov chain, the probability of stability p of each state_mAre as follows:

In formula, π_mThe Stationary Distribution of expression state m, Stationary Distribution meet:

The result of formula (5), which is substituted into formula (10), to be obtained:

π₁=π₂=π₃=π₄=0.25 (11)

Substitution formula (9) obtains the probability of stability of each state are as follows:

Obviously, the probability of stability that air conditioner load is in each state is migration probability u₀And u₁Function.

Preferably, the feature power of the physical process includes:

1) rated power P_rate,ij

P_rate,ijPower after being opened for air-conditioning j in air conditioner load cluster i, the air-conditioning heat output Q in it and formula (1)_a,iIt is full The following relationship of foot:

Q_a,ij=COP_ij×P_rate,ij (13)

In formula, COP_ijFor air-conditioning Energy Efficiency Ratio；I=1 ... M, j=1 ... N_i, M is number of clusters, N_iFor the air-conditioning number of cluster i Amount；

2) stable state it is expected power P_exp,ij

For the air conditioner load using ON/OFF control, the power at certain momentFor discrete random variable, according to discrete The stable state of expectation of a random variable formula, the air conditioner load it is expected power are as follows:

It is apparent from P_exp,ijIt is also migration probability u₀And u₁Function, above formula establishes physical process in air-conditioning hybrid model With the connection of control process, wherein p₁And p₃The probability of stability of the ON (m=1) and ONLOCK (m=3) that obtained by formula (12)；

The then desired value of the general power of air conditioner load cluster i are as follows:

In this way, can control the power of entire cluster, by P by the migration probability for controlling each air conditioner load_exp,ijReferred to as For the expectation power of air conditioner load, by P_exp,iReferred to as cluster general power；

3) reference power P_base,ij

P_base,ijIt is defined as air-conditioning j in cluster i and maintains set temperature T under no-control state applies_set,ijRequired expectation function Rate, can be in the hope of reference power by formula (1) are as follows:

Wherein COP_ijFor air-conditioning Energy Efficiency Ratio；U_a,ijFor the equivalent thermal resistance of indoor and outdoor；Q_m,ijFor indoor solid heat transfer amount；T_oFor Outdoor temperature；

As it can be seen that reference power is influenced by outdoor temperature, but since outdoor temperature variation is slower, within a control period, It is believed that reference power is basically unchanged；

The then reference power of cluster i are as follows:

The flexibility of air conditioner load cluster is by making cluster general power P_exp,iDeviate reference power P_base,ijCome what is reached；

4) dynamic power P_d,ij

P_d,ijIt is defined as within a control period, expectation power needed for making current room temperature change to assigned temperature, if Air-conditioning is k-th of period with P_d,ijIn one control period of continuous service, room temperature is made to change to temperature T_a,ijIt (k), then can by formula (1) In the hope of dynamic power are as follows:

A_1,ij、B_1,ij、C_1,ij、D_1,ij(k) indicate that coefficient related with outdoor temperature and air-conditioning parameter, physical relationship are seen below Literary appendix A；T_a,ij(k) indoor gas temperature, the T in kth all end of term are indicated_m,ij(k-1) the indoor solid in all end of term of kth -1 is indicated Temperature；

If the room temperature range that air-conditioning j allows is [T_min,ij,T_max,ij], optimum temperature T_set,ij, then dynamic power has as follows Particular value:

1) as assigned temperature T_a,ij(k)=T_min,ij, remember that required dynamic power is P_max0,ij；

2) as assigned temperature T_a,ij(k)=T_max,ij, remember that required dynamic power is P_min0,ij；

3) as assigned temperature T_a,ij(k)=T_set,ij, dynamic power is power needed for returning to optimum temperature, is denoted as P_set0,ij；

It is arranged temperature " soft-sided circle ", i.e., temperature range is reduced into [T_min,ij+δ,T_max,ij- δ], wherein δ is taken as:

δ=0.1 × (T_max,ij-T_min,ij) (19)。

Preferably, it polymerize to implement polymerization to air conditioner load with counter, the market equilibrium mechanism is applied to cluster Coordinated control, enable each air-conditioning construct demand curve d according to respective electricity consumption urgency and flexibility_ij(π), demand curve tool There is monotone decreasing characteristic, includes three key points: A (1, P_min,ij), B (0, P_set,ij) and C (- 1, P_max,ij)；Wherein, in abscissa Price π only represent control signal, referred to as virtual price, and be [- 1,1] by its scope limitation；Ordinate is dynamic power； The case where being likely to occur less than 0 due to the dynamic power estimated by formula (18) or be greater than rated power, so in demand curve Three critical power ratios are as follows:

P_min,ijIt indicates to consider the minimum dynamic power after soft-sided circle；P_set,ijIt indicates to return to room temperature most after considering soft-sided circle The dynamic power of thermophilic degree；P_max,ijIt indicates to consider the maximum dynamic power after soft-sided circle；P_max0,ijExpression does not consider soft-sided circle Maximum dynamic power, value are determined by formula (18)；P_min0,ijIndicate the minimum dynamic power for not considering soft-sided circle, value is by formula (18) it determines；P_set0,ijExpression does not consider the dynamic power that soft-sided circle makes room temperature return to optimum temperature, and value is determined by formula (18)； P_rate,ijIndicate the rated power of air conditioner load.

Air-conditioning can upgrade demand curve in each control period according to the present situation, in this way, both can accurately report its spirit Activity also can guarantee users'comfort when responding arbitrary virtual price signal；

Before preparing to release control to air-conditioning cluster, the signal that virtual price is 0 can be issued in advance, it is bent according to demand Line, the responding power of each air conditioner load are P_set,ij, that is, allow room temperature to return to the expectation power of set temperature, in this manner, respectively Air conditioner load will keep the probability control mode of half Markov to run a period of time, release room temperature again close to after set temperature Control, can effectively keep the diversity of air-conditioning state in cluster and alleviate load oscillatory occurences.

Preferably, air conditioner load locally controls specifically:

When air conditioner load is connected to virtual priceAfterwards, curve determines that its responding power is according to demandThen Air conditioner load need to determine migration probability u₀And u₁Value makes it is expected that power meets:

But u only can not be uniquely determined by above formula₀And u₁, require supplementation with condition, when air-conditionings a large amount of in cluster be in ON or When the average waiting time of OFF state is more than setting high level or is less than setting low value, it will lead to and be trapped in some state in cluster Air-conditioning quantity it is excessive, to destroy diversity, it is contemplated that the average waiting time of ON or OFF state depend on u₀And u₁, therefore be This phenomenon is avoided, forms following supplementary condition, it is assumed that t_onlock=t_offlock=t_lock:

In formula, function Rand indicates the execution interval of half markoff process for generating uniform random number, Δ t；

If), probability u is opened at this time₁It is larger and close probability u₀It is smaller, it should prevent because of u₁It is excessive and lead Cause the average waiting time T of OFF state₂It is too small, bring formula (7) into formula (22), it is known that T₂It is restricted to [t_lock/1.5,t_lock/ 0.5] section, with blocking time t_lockIn the same order of magnitude, u₁Why it is randomly generated, is to further keep air conditioner load The diversity of cluster is obtaining u by formula (22)₁Afterwards, then according to formula (21) u is solved₀, acquire u₀And u₁Afterwards, air conditioner load can be by State machine model carries out local control.

Preferably, coordinated control is carried out to extensive air conditioner load using the market equilibrium mechanism, detailed process is such as Under:

1) polymerization process, when each air conditioner load forms demand curve d_ijAfter (π), polymerization quotient i is formed always need according to the following formula Seek curve D_i(π):

2) anti-polymerization process, polymerization quotient i obtain responding powerVirtual price is acquired according to the following formula:

The virtual price is broadcast in cluster by polymerization quotient, and each air conditioner load obtains respective responding powerDue to For market clearing price, thus met in cluster level:

Due to implementing to control to extensive air-conditioning using unified price signal, therefore it is lower to control cost.

Preferably, the external behavior of virtual regulating units includes peak regulation power, peak capacity and peak regulation expense, wherein peak regulation Power definition is the value that virtual robot arm responding power deviates reference power, only considers that air conditioner load is cut down in summer peak of power consumption The scene of load is equivalent to conventional power unit and increases power output, then the peak regulation power of virtual robot arm i and the relationship of responding power are as follows:

Peak capacityIt is defined as the maximum peak regulation power of virtual robot arm, is calculated in combination with formula (23) by following formula:

In formula, D_i(1) the minimum response power of virtual robot arm i is indicated；P_base,iFor the reference power of air conditioner load i, value It is determined by formula (16).

Air-conditioning state SOA index is defined, when only considering summer reduction plans scene, SOA is defined as:

Then the average SOA of virtual robot arm i is indicated are as follows:

And in order to measure the peak regulation degree of virtual robot arm, define peak regulation depth delta P_ratio,iAre as follows:

It enables each virtual robot arm participate in peak regulation scheduling by way of quotation herein, assumes that each virtual robot arm is reported as the following formula herein Valence:

In formula, λ_rFor Spot Price；a_iIndicate peak regulation depth compensation coefficient；b_iComfort level penalty coefficient is indicated, because of Δ P_ratio,iWithChange is marked, the upper limit for easily estimating above-mentioned quotation is (a_i+b_i)λ_r；

Above formula show virtual robot arm it is next control the period peak regulation depth it is bigger, currently averagely SOA it is higher, then It offers higher；

Then the peak regulation expense of virtual robot arm i may be expressed as:

F_i(ΔP_a,i)=λ_c,iΔP_a,i (32)

Obviously, peak regulation expense is the quadratic function of peak regulation power.

Compared with prior art, the present invention proposes a kind of based on half markoff process first against air-conditioner control system Probability Control, can guarantee the blocking time of air-conditioning and keep the diversity of cluster state, and be easy to establish and control Journey and physical process contact；The physical model for being then based on air conditioner load proposes a kind of building method of demand curve, energy Enough while expressing electricity consumption urgency and flexibility, the data-privacy and information security of effective protection user.

On the basis of above- mentioned information physical model, proposed based on market equilibrium mechanism a kind of by extensive air conditioner load It is polymerized to the control method for coordinating of virtual regulating units, this method uses single control signal and very low control frequency, can show Writing reduces communication cost.Meanwhile virtual regulating units can provide to be similar to the offer curve of conventional power unit and adjust to scheduling and hold Amount is convenient for Optimized Operation.

Detailed description of the invention

Fig. 1 is the block schematic illustration of virtual peak regulation control；

Fig. 2 is the hybrid model schematic diagram of air conditioner load；

Fig. 3 is demand curve schematic diagram；

Fig. 4 is the information physical coupling model schematic diagram of air conditioner load；

Fig. 5 is that different conditions transient state probability changes schematic diagram；

Fig. 6 is cluster reference power and actual power curve graph；

Fig. 7 is load peak and oscillation of power phenomenon comparison schematic diagram；

Fig. 8 is air conditioner load cluster track situation schematic diagram；

Fig. 9 is tracking error mean value schematic diagram；

Figure 10 is virtual robot arm Δ P_ratio,iWithVariation schematic diagram；

Figure 11 is clear electricity price and Spot Price schematic diagram.

Specific embodiment

Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete Site preparation description, it is clear that described embodiment is a part of the embodiments of the present invention, rather than whole embodiments.Based on this hair Embodiment in bright, those of ordinary skill in the art's every other reality obtained without making creative work Example is applied, all should belong to the scope of protection of the invention.

It is recognized herein that the autonomous control for realizing temperature control load is the key that solve the above problems.For this purpose, this paper presents one Kind polymerize the information physical modeling and control method of extensive air conditioner load.Compared with existing method, the main contributions of this paper exist In:

1) in load level, air conditioner load is considered as information physical system (cyber-physical system, CPS), is mentioned Go out based on half markoff process (semi-markov process)^[16]Air conditioner load Probability Control, and using mixed The coupled relation of miscellaneous system (hybrid system) model foundation air conditioner load physical process and control process, realizes load Autonomous control.

2) in polymerization quotient sheaf face, a kind of coordinated control side of extensive air conditioner load based on market equilibrium mechanism is proposed Method, can be equivalent at the virtual regulating units with the similar external characteristics of conventional power unit by a large amount of randomness air conditioner loads, and participates in Optimized Operation.

1, virtual peak regulation system overall structure

Virtual peak regulation, which refers to, is polymerized to virtual regulating units for extensive air conditioner load by polymerization quotient, participates in peak load regulation network tune Degree^[4,5].Fig. 1 is the overall structure of virtual peak regulation system, from top to bottom includes control centre, virtual regulating units and air conditioner load Three layers.It is herein the control period with 0.5h, each control period includes such as the next stage:

1) air-conditioning is submitted a tender.In a upper control all end of term, each air-conditioning is independently calculated according to information such as respective state, preferences Reference power simultaneously forms demand curve, is reported to corresponding polymerization quotient.

2) it polymerize.The demand curve of air conditioner load in cluster is polymerized to total demand curve by each polymerization quotient, meanwhile, with similar In the mode of conventional power unit, peak capacity and quotation strategy are provided to scheduling.

3) Optimized Operation.The peak capacity according to each virtual regulating units and quotation strategy are dispatched, each virtual peak regulation is calculated The optimal peak regulation power of unit is simultaneously assigned to corresponding polymerization quotient.

4) anti-polymerization.Polymerization quotient determines virtual price signal according to market equilibrium mechanism and is broadcast to the air-conditioning in cluster and bears Lotus, to achieve the purpose that peak regulation power distribution to each air conditioner load.

5) local control.The responding power that each air conditioner load undertakes needed for being determined according to virtual price, and then determine half horse The migration probability of Er Kefu model realizes the autonomous operation within this control period.

The above-mentioned stage 1,5 belongs to the autonomous control layer of each air conditioner load；Stage 2,4 belongs to the coordinated control of each polymerization quotient Layer；Stage 3 belongs to the optimum control layer of control centre.The control method of above three level is introduced separately below.

2, the air conditioner load autonomous control based on hybrid model

CPS is the integrated system of calculating process and physical process^[17].Air conditioner load is considered as CPS herein, be in order to by its The joint dynamic modeling and reciprocation of physical process and control process are preposition to be realized to user side, is avoided all via in control The conventional control mode in heart composition " acquisition-processing-control " circuit^[7-9], to realize the autonomous control of air conditioner load, reach drop The purpose of low control cost, protection privacy of user.Simultaneous is carried out to physical process continuous in CPS and discrete control process to build The important method of mould is hybrid system modeling^[17,18], therefore the model of air conditioner load CPS is established first with this method herein.

2.1 hybrid model

It is illustrated in figure 2 the hybrid model of air conditioner load.Airconditioning control process is departure process, using finite state Machine description.This paper presents a kind of Probability Controls, introduce u₀And u₁Two migration probability state of a control transfers.In addition, in order to It prevents from aggravating equipment attrition because of frequent start-stop or even damages compressor, air-conditioning needs to be arranged certain blocking time t_lock(packet Containing blocking time t after closing_offlockWith blocking time t after opening_onlock), value is generally 3~5min^[14].The state machine of Fig. 2 Comprising four states, wherein ON and ONLOCK is the open state of air-conditioning, and difference is that the former has met blocking time and wants It asks, can close at any time；And OFF and OFFLOCK are in off state, but the former can open at any time.

Indoor temperature change generated in case is continuous physical process, can use equivalent thermal parameter (equivalent thermal Parameters, ETP) model expression^[14,19], second order expression is as follows:

In formula, T_a、T_mAnd T₀Respectively indoor air temperature, indoor solid temperature and outdoor temperature；U_aFor indoor and outdoor etc. Imitate impedance；H_mEquivalent impedance between room air and solid；Q_aAnd Q_mRespectively indoor gas and solid heat transfer amount；Q_ac、Q_iWith Q_sRespectively air-conditioning heat output, indoor airflow heat output and solar radiation heat；C_aAnd C_mThe respectively heat of room air and solid Hold；M (t) is air-conditioner switch state, and m (t)=1 represents air-conditioning unlatching, and m (t)=0 represents air-conditioning closing；f₁,f₂For heat transfer coefficient, F is taken herein₁=f₂=0.5.

In Fig. 2, the autonomy of air conditioner load, which had both been shown, to express electricity consumption urgency and flexibility by bid information, Virtual price signal can be responded and complete local control by showing again.It will elaborate below to this.

The Semi-Markov Process of 2.2 control processes

It is without memory that markoff process requires the Annual distribution of systematic state transfer to have, i.e. obedience exponential distribution.But In Fig. 2, the blocking time of air conditioner load is the constant of setting, so that the transfer time of blocking and disobeying index point Cloth.For this purpose, introducing the Controlling model that half markoff process establishes air conditioner load herein.The process is defined as follows^[16]:

2) under the premise of the state of next entrance is n, there is distribution F until the time for occurring to shift from m to n_mn (t), which can be Arbitrary distribution.

Then it is in the distribution F of state m_m(t) meet:

The average waiting time of state m are as follows:

M=1 is enabled, 2,3,4 respectively indicate ON, OFF, ONLOCK and OFFLOCK state.As shown in Figure 2, the shape of air conditioner load State transfer has one-way, then can be obtained by formula (2):

p₁₄=p₄₂=p₂₃=p₃₁=1 (37)

If α_mnFor from state m to the mobility of state n.(this when the execution period Δ t of half markoff process is sufficiently small Text is 2s) have: u₀≈α₁₄Δt,u₁≈α₂₃Δt^[20].If the residence time of state 1,2 meets exponential distribution, that is, have memoryless Property, then F₁₄(t) and F₂₃(t) be respectively parameter be α₁₄And α₂₃Exponential distribution, it may be assumed that

T₃=t_onlock, T₄=t_offlock (40)

If semi-Markov chain be it is irreducible, after finite time, can be restrained in certain shape probability of state To the definite value unrelated with original state, the referred to as probability of stability^[16].Opposite, the probability before converging to the probability of stability is claimed For transient state probability.According to the property of semi-Markov chain, the probability of stability p of each state_mAre as follows:

In formula, π_mThe Stationary Distribution of expression state m.Stationary Distribution meets:

π₁=π₂=π₃=π₄=0.25 (43)

The feature power of 2.3 physical processes

2.3.1 rated power P_rate,ij

Q_a,ij=COP_ij×P_rate,ij (45)

In formula, COP_ijFor air-conditioning Energy Efficiency Ratio；I=1 ... M, j=1 ... N_i, M is number of clusters, N_iFor the air-conditioning number of cluster i Amount.

2.3.2 stable state it is expected power P_exp,ij

For the air conditioner load using ON/OFF control, the power at certain momentFor discrete random variable (P_rate,ijOr 0).According to the desired value formula of discrete random variable, the stable state of the air conditioner load it is expected power are as follows:

It is apparent from P_exp,ijIt is also migration probability u₀And u₁Function.Above formula establishes physical process in air-conditioning hybrid model With the connection of control process.

In this way, can control the power of entire cluster by the migration probability for controlling each air conditioner load.Hereinafter by P_exp,ij The referred to as expectation power of air conditioner load, by P_exp,iReferred to as cluster general power.

2.3.3 reference power P_base,ij

P_base,ijIt is defined as air-conditioning j in cluster i and maintains set temperature T under no-control state applies_set,ijRequired expectation function Rate.It can be in the hope of reference power by formula (1) are as follows:

As it can be seen that reference power is influenced by outdoor temperature.But due to outdoor temperature change it is slower, a control period (such as In 0.5h), it is believed that reference power is basically unchanged.

The then reference power of cluster i are as follows:

The flexibility of air conditioner load cluster is by making cluster general power P_exp,iDeviate reference power P_base,iCome what is reached.

2.3.4 dynamic power P_d,ij

P_d,ijIt is defined as within a control period, expectation power needed for making current room temperature change to assigned temperature.If Air-conditioning is k-th of period with P_d,ijIn one control period of continuous service, room temperature is made to change to temperature T_a,ij(k).It then can by formula (1) In the hope of dynamic power are as follows:

The meaning and derivation process of each parameter of above formula are shown in appendix A.

If the room temperature range that air-conditioning j allows is [T_min,ij,T_max,ij], optimum temperature T_set,ij.Then dynamic power has as follows Particular value:

3) as assigned temperature T_a,ij(k)=T_set,ij, dynamic power is power needed for returning to optimum temperature, is denoted as P_set0,ij。

It should be noted that room temperature control can be made to exist due to using probability to control herein and forcing to consider blocking time Error.For this purpose, being similar to document [10] herein is provided with temperature " soft-sided circle ", i.e., temperature range is reduced into [T_min,ij+δ, T_max,ij- δ], wherein δ is taken as:

δ=0.1 × (T_max,ij-T_min,ij) (51)

2.4 demand curve

It polymerize to implement polymerization to air conditioner load with counter, market equilibrium mechanism is applied to the coordination control of cluster herein System^[19].For this purpose, each air-conditioning is enabled to construct the demand curve as shown in Fig. 3 (a) according to respective electricity consumption urgency and flexibility herein d_ij(π).The demand curve has monotone decreasing characteristic, includes three key points: A (1, P_min,ij), B (0, P_set,ij) and C (- 1, P_max,ij).Wherein, the price π in abscissa only represents control signal, does not have monetary unit, therefore herein referred to as virtual valence Lattice, and be [- 1,1] by its scope limitation；Ordinate is dynamic power.Since the dynamic power estimated by formula (18) is likely to occur The case where less than 0 or being greater than rated power, so three critical power ratios in demand curve are as follows:

Air-conditioning can upgrade demand curve in each control period according to the present situation.In this way, both can accurately report its spirit Activity also can guarantee users'comfort when responding arbitrary virtual price signal.

Before preparing to release control to air-conditioning cluster, the signal that virtual price is 0 can be issued in advance.It is bent according to demand Line, the responding power of each air conditioner load are P_set,ij, that is, room temperature is allowed to return to the expectation power of set temperature.In this manner, respectively Air conditioner load will keep the probability control mode of half Markov to run a period of time, release room temperature again close to after set temperature Control, can effectively keep the diversity of air-conditioning state in cluster and alleviate load oscillatory occurences.

2.5 air conditioner load local control methods

But u only can not be uniquely determined by above formula₀And u₁, require supplementation with condition.Existing research discovery, when air-conditioning in cluster After the diversity of state is destroyed, serious load peak or oscillatory occurences will lead to^[14].And at air-conditionings a large amount of in cluster When the average waiting time of ON or OFF state excessive or too small (compared to blocking time), it will lead in cluster and be trapped in certain The air-conditioning quantity of a state is excessive, to destroy diversity.In view of the average waiting time of ON or OFF state depends on u₀With u₁, therefore to avoid this phenomenon, following supplementary condition formed herein is (assuming that t_onlock=t_offlock=t_lock):

In formula, function Rand is for generating uniform random number.

It is now larger with responding powerThe case where explain formula (22).Probability u is opened at this time₁It is larger and close Close probability u₀It is smaller, it should prevent because of u₁Average waiting time T that is excessive and leading to OFF state₂It is too small.Bring formula (7) into formula (22), it is known that T₂It is restricted to [t_lock/1.5,t_lock/ 0.5] section, with blocking time t_lockIn the same order of magnitude.u₁Why It is randomly generated, is the diversity in order to further keep air conditioner load cluster.U is being obtained by formula (22)₁Afterwards, then according to formula (21) it solves U out₀, the existence proof of solution is shown in Appendix B.Acquire u₀And u₁Afterwards, air conditioner load can carry out this by the state machine model of Fig. 2 Ground control.

Fig. 4 illustrates the coupled relation of air conditioner load information and physical model, as seen from the figure: 1) Semi-Markov Process energy The relationship for enough establishing air conditioner load expectation power and control system migration probability, to link up continuous physical process and discrete control Process processed.Probability control preferably ensure that the diversity of cluster state；2) using demand curve as air conditioner load CPS and external Contact interface, can preferably protect the data-privacy of user, improve the safety of control.

3, the coordinated control of virtual regulating units

Herein, polymerization quotient need to have dual function: carry out coordinated control to extensive air conditioner load；Be polymerized for Virtual regulating units (being hereinafter virtual robot arm), and receive dispatching of power netwoks.The realization of the two functions is discussed separately below Method.

3.1 polymerization based on market equilibrium mechanism and anti-polymerization

As previously mentioned, carrying out coordinated control to extensive air conditioner load using market equilibrium mechanism herein, process is as follows:

1) polymerization process.When each air conditioner load forms the demand curve d as shown in Fig. 3 (a)_ijAfter (π), polymerization quotient i is pressed Total demand curve D is formed according to following formula_i(π):

2) anti-polymerization process.Polymerization quotient i obtains responding powerVirtual price is acquired according to the following formula:

The virtual price is broadcast in cluster by polymerization quotient, and each air conditioner load obtains respective sound according to the method for 2.5 sections Answer powerDue toFor market clearing price, thus met in cluster level:

Fig. 3 (b) illustrates the above process.Due to implementing to control to extensive air-conditioning using unified price signal, therefore control Cost processed is lower.

The external behavior of 3.2 virtual regulating units

Similar to conventional power unit, when virtual robot arm participates in peak regulation, need to provide peak capacity and peak regulation cost to scheduling Etc. information.

3.2.1 peak regulation power and peak capacity

Peak regulation power definition is the value that virtual robot arm responding power deviates reference power.Only consider air conditioner load in the summer herein The scene (be equivalent to conventional power unit and increase power output) of reduction plans when season peak of power consumption.The then peak regulation power and sound of virtual robot arm i Answer the relationship of power are as follows:

In formula, D_i(1) the minimum response power of virtual robot arm i is indicated.

3.2.2 peak regulation expense

In order to measure the comfort level of user, air-conditioning state (state of air-conditioner, SOA) is defined herein and is referred to Mark, when only considering summer reduction plans scene, SOA is defined as:

Then the average SOA of virtual robot arm i can be indicated are as follows:

And in order to measure the peak regulation degree of virtual robot arm, peak regulation depth delta P is defined herein_ratio,iAre as follows:

Each virtual robot arm is enabled to participate in peak regulation scheduling by way of quotation herein^[21,22].Difference polymerization quotient is according to its technology The factors such as management level, air-conditioning quantity and the requirement for comfort level, can voluntarily formulate quotation strategy.Each void is assumed herein Quasi- unit is offered as the following formula:

In formula, λ_rFor Spot Price；a_iIndicate peak regulation depth compensation coefficient；b_iIndicate comfort level penalty coefficient.Because of Δ P_ratio,iWithChange is marked, the upper limit for easily estimating above-mentioned quotation is (a_i+b_i)λ_r。

Above formula show virtual robot arm it is next control the period peak regulation depth it is bigger, currently averagely SOA it is higher, then It offers higher.

Then the peak regulation expense of virtual robot arm i can indicate are as follows:

F_i(ΔP_a,i)=λ_c,iΔP_a,i (64)

4, level optimum control is dispatched

Control centre solves following Optimal Scheduling Problem after the quotation and peak capacity for collecting each virtual robot arm:

In formula, P_targetFor total peak regulation power；M is number of clusters.

The calculated result of above formula is handed down to each virtual robot arm by control centre, and the latter is calculated by formula (26) Responding power, and then distribute by the anti-polymerization of 3.1 sections to each air-conditioning.

Formula (33) is typical quadratic programming problem, can be converted into dual problem solution^[23], the Lagrange of dual problem Multiplier result is clear electricity price.It should be noted that the clear electricity price that goes out herein is settled accounts for practical, it is only used for different from Section 2 The virtual price of coordination.

5 simulation examples

5.1 parameter setting

This paper example investigates somewhere air conditioner load and participates in peak regulation, it is assumed that this area includes that 4 Load aggregation quotient (are denoted as void Quasi- unit 1~4), virtual robot arm parameter is as shown in table 1, and different penalty coefficients are used to reflect different cost levels and relax The relative weighting moderately compensated.The rated power and physochlaina infudibularis exponential model of air conditioner load are exported all in accordance with the information of table 3^[19].At this In table, U (a, b) indicates being uniformly distributed between [a, b], and N (avg, std) indicates normal distribution.4 virtual robot arms it is specified General power is 143.7MW.

The control period of system is set as 0.5h.The initial temperature and switch state of each air conditioner load are randomly generated, empty Blocking time after adjusting load to open/close is disposed as 3min.Wherein table 1 is virtual robot arm parameter, and table 2 is air conditioner load heat Parameter calculation basis.

Table 1

Table 2

The verifying of 5.2 control methods

Before the virtual peak regulation effect of verifying air conditioner load cluster, the feasibility of this paper control method is first verified that.

5.2.1 the probability of stability is verified

Emulation is compared respectively with 1 000 and 10 000 parameters and the identical air-conditioning of original state.By formula (22) Supplementary condition, this emulation migration probability be taken as u₀=0.007 5 and u₁=0.001 2.It can be calculated at this time by formula (7) The average waiting time of ON state is 4.4min, close with blocking time.It is by the theoretical value that formula (12) can obtain each probability of stability p₁=0.119, p₂=0.719, p₃=0.081 and p₄=0.081.Actual emulation result is as shown in figure 5, transient state probability is in figure Refer to the practical accounting of each air-conditioning state in cluster.It can be seen that the distribution of air-conditioning cluster state is passed through converges on theoretical value after a certain period of time, Convergence time is less than 0.5h.Simultaneously it could be observed that air-conditioning quantity is more, the randomness of cluster polymerization property is lower.

5.2.2 reference power is estimated

By taking virtual robot arm 1 as an example, enable in cluster all air conditioner loads by set temperature free-running operation.Outdoor temperature variation, The estimated value and cluster actual power of cluster air-conditioning reference power are as shown in Figure 6.As seen from the figure, in addition to changing in outdoor temperature The faster period is outer (period near such as 12:00 and 16:00), and formula (16) can estimate reference power well.

5.2.3 it is compared with temperature control method

Still by taking virtual robot arm 1 as an example, the implementation process of temperature control method are as follows: starting control season all air-conditioning set temperature tune It is 1 DEG C high, set temperature is recalled into initial value when releasing control.It compares with temperature control method, is controlled in context of methods for convenience The expectation power setting of state processed is the steady state power of temperature control method state of a control, and the mode for releasing control is shown in 2.4 sections.In order to exclude The influence of outdoor temperature fluctuation, this emulation assume that outdoor temperature is constant.As seen from Figure 7, temperature control method is starting and is releasing control When aggregate power very big spike can all occur, moment is close to 0 or reaches nominal total power, and with oscillatory occurences.And it is our Method can allow cluster power smoothly to reach target response power, apparent spike and oscillatory occurences do not occur.

5.3 virtual peak regulation effects

5.3.1 overall tracking effect

Two periods of noon peak (11:00~14:00) and evening peak (18:00~21:00) are located at, peak load regulation network ability is not Foot needs air conditioner load to provide virtual peak regulation.Total reference power, general objective responding power and the real response of 4 virtual robot arms Power is as shown in Figure 8.According to formula (26), the difference of reference power and responding power is peak regulation power, and peak regulation power is in 4 void Distribution condition in quasi- unit is also shown in Fig. 8.As seen from the figure, when total peak regulation power changes, air conditioner load cluster power is deposited In convergence process, it is similarly to the climbing rate of conventional power unit.

The calculation formula of tracking error are as follows:

Similar to conventional power unit, the climbing time of 5min is reserved for each virtual robot arm herein, therefore calculating tracking error This period is not counted in when mean value.By taking noon peak as an example, each tracking error mean value for controlling the period is as shown in Figure 9.It can be seen that when vacation If air conditioner load ETP parameter is accurate, error mean is less than 0.5%, illustrate context of methods can be very good realize power with Track.

5.3.2 quotation strategy and cost analysis

According to formula (31) it is found that the quotation strategy of each virtual robot arm is determined by penalty coefficient, and actual price will also depend on In respective Δ P_ratio,iWithBy the available following rule of Figure 10:

1)ΔP_ratio,iWithThere are certain positive correlations: the peak regulation power Δ P that virtual robot arm undertakes_ratio,i It is smaller, thenSmaller, the influence to users'comfort is smaller.

2) respectively the control period existsRule, this makes comfort level penalty coefficient b_iTo the shadow of quotation It rings than peak regulation depth compensation coefficient a_iGreatly.Due in 4 virtual robot arms, the b of unit 2_iCoefficient is maximum, makes its highest of offering, leads The peak regulation depth for causing it to be undertaken is always minimum.

3) penalty coefficient of virtual robot arm 1 and 4 is identical, and only capacity is different.It is by result as it can be seen that identical in quotation strategy When, unit capacity is to Δ P_ratio,iWithInfluence it is very small.This can make the global comfort of air-conditioning cluster substantially not by The influence of its scale.

The clear electricity price out in each control period is as shown in figure 11.The Spot Price data of peak regulation period come from document^[24].In conjunction with Figure 10 with peak regulation process it is found that continue, the Δ P of each virtual robot arm_ratio,iWithIt is gradually increased, leads to clear electricity price out It is in rising trend.This is also illustrated in fact, and compared with generator, air conditioner load belongs to finite energy type resource, cannot be for a long time Participate in peak regulation.

Total peak regulation expense (being peak regulation income for polymerizeing quotient) of each virtual robot arm is as shown in table 3.Compared to virtual robot arm 2,3, virtual robot arm 1 offer it is lower, so income is higher；The scale of virtual robot arm 4 is about twice of virtual robot arm 1, and other Parameter is similar, therefore income is also about twice.Wherein table 3 is the peak regulation expense of virtual robot arm.

The operational efficiency of conventional power unit can be improved although needing to pay reimbursement for expenses in angle from control centre, It reduces the consumption of fossil fuel and reduces carbon emission, and electric generation investment can be delayed, therefore have great importance.

Table 3

5.3.3 to the influence of users'comfort

As previously mentioned, being reduced using the method for temperature soft-sided circle herein because probability controls and considers air-conditioning blocking time institute Bring room temperature controls error.Equally by taking noon peak as an example, the room temperature threshold crossing time accounting of all air conditioner loads is shown in Table 4.As it can be seen that Users'comfort is significantly improved after introducing temperature soft-sided circle.Can further it be improved by increasing the soft-sided circle δ in formula (19) Comfort level, but this can reduce the flexibility of air conditioner load.

Table 4

The inaccurate bring of 5.4 parameters influences

Since there is employed herein information physical Coupling methods, therefore air conditioner load needs more accurate ETP model parameter.But In practice there is certain evaluated error in these parameters^[25,26].Assume herein the evaluated error of each parameter be limited to ± 15% range.The universal situation less than normal of estimates of parameters is considered in emulation and takes extreme condition, i.e., the estimated value of each parameter is equal It is the 85% of actual value.

5.4.1 to the influence of tracking error

Tracking error when parameter is less than normal has been shown on Fig. 9.As seen from the figure, parameter lax pair tracking error has no obvious shadow It rings.

5.4.2 to the influence of users'comfort

Temperature beyond limit ratio when parameter is less than normal has been listed in table 4.Reference power estimated value is less than normal at this time known to formula (16), Convolution (26) can make that target response power is less than normal, actual load is cut down it is found that virtual robot arm lowers load in this reference value Amount is more than that peak regulation instructs, so that temperature beyond limit ratio increases.But due to using temperature soft-sided circle, control method is to user The influence of comfort level is still limited.

5.4.3 to the influence of peak regulation cost

Influence to cost is as shown in table 3.Convolution (16), (18), (26) are it is found that when parameter is less than normal, peak capacity Estimated value it is less than normal so that the Δ P under identical peak regulation power_ratio,iBigger than normal, then all virtual robot arms all improve quotation, thus Increase income.

It sums up, the inaccurate problem of air conditioner load parameter has substantially no effect on the tracking accuracy of virtual robot arm；For polymerization It, can be along with the reduction of comfort level when thus income being caused to increase for quotient.

In order to reduce the control cost for extensive air conditioner load and improve control effect, this paper presents be based on mixing The air conditioner load information physical modeling method of system.First against air-conditioner control system, propose a kind of based on half Markov mistake The Probability Control of journey can guarantee the blocking time of air-conditioning and keep the diversity of cluster state, and be easy to establish control Process and physical process contact；The physical model for being then based on air conditioner load proposes a kind of building method of demand curve, It can be while expressing electricity consumption urgency and flexibility, the data-privacy and information security of effective protection user.

Appendix A

The derivation process of dynamic power:

T is eliminated by formula (1)_m:

In formula,

Solving (A.1) can obtain:

In formula,

Then extract Q in formula (A.3)_a, the expression formula of dynamic power can be obtained:

P_d(k)=A₁T_a(k)+B₁T_a(k-1)+C₁T_m(k-1)+D₁(k)(A.5)

In formula,

Appendix B

It is larger with air conditioner load responding powerThe case where prove solution existence:

By formula (7) it is found that working as u₁∈(0.5Δt/t_lock,1.5Δt/t_lock) when, T₂∈(2t_lock/3,2t_lock).At this time:

IfThen T₁It can be expressed as T₂Function:

Convolution (7) is it is found that u₀<1.So for any u₁∈(0.5Δt/t_lock,1.5Δt/t_lock) can determine completely The u of sufficient condition₀。

The document of reference is specific as follows:

[1] the photo-thermal power station heat storage capacity configuration method of Cui Yang, Yang Zhiwen, Yan Gangui reduction fired power generating unit peak regulation cost [J] Proceedings of the CSEE, 2018,38 (6): 1605-1611.

Cui Yang,Yang Zhiwen,Yan Gangui.Capacity configuration of thermal energy storage within CSP to reduce the cost of peak load regulation[J] .Proceedings of the CSEE,2018,38(6):1605-1611(in Chinese).

[2]International Energy Agency:Air conditioning use emerges as one of the key drivers of global electricity-demand growth[EB/OL].2018[2018-08-13

16:47:00].http://suo.im/5qFQbP.

[3] Song Hongkun, Tang Guoqing, Lu Yi wait the peak clipping of the analysis of summer air conditioning loads of Jiangsu and demand side management measure Effect calculates [J] electric power network technique, 2006,30 (17): 88-91.

Song Hongkun,Tang Guoqing,Lu Yi,et al.Analysis on summer air- conditioning loads composition Jiangsu Province and estimation of peak load shifting effect by DSM measures[J].Power System Technology,2006,30(17):88-91 (in Chinese).

[4] Yang Yongbiao, Yan Qingguo, Xu Shiming wait public building air conditioner load to participate in thinking [J] of the virtual peak regulation of power grid Automation of Electric Systems, 2015,39 (17): 103-107.

Yang Yongbiao,Yan Qingguo,Xu Shiming,et al.Thinking of public building air-conditioning load participating in grid with virtual peak clipping[J].Automation of Electric Power Systems,2015,39(17):103-107(in Chinese).

[5] the electric load peak clipping plan a few days ago of Xu Qingshan, Yang Chenxing, Yan Qingguo meter and scale air conditioner heat balancing inertia Slightly [J] electric power network technique, 2016,40 (1): 156-163.

Xu Qingshan,Yang Chenxing,Yan Qingguo.Strategy of day-ahead power peak load shedding considering thermal equilibriuminertia of large-scale air conditioning loads[J].Power System Technology,2016,40(1):156-163(in Chinese).

[6] Yang Nan, Wang Bo, Liu Dichen wait to consider the large-scale wind power random optimization dispatching method of flexible load peak regulation [J] electrotechnics journal, 2013,28 (11): 231-238.

Yang Nan,Wang Bo,Liu Dichen,et al.Large-scale wind power stochastic optimization scheduling method considering flexible load peaking[J] .Transactions of China Electrotechnical Society,2013,28(11):231-238(in Chinese).

[7]Lu N,Chassin D.A State-Queueing Model of Thermostatically Controlled Appliances[J].IEEE Transactions on Power Systems,2004,19(3):1666- 1673.

[8] Wang Chengshan, Liu Mengxuan, Lu Ning are smoothly square using the microgrid interconnection tie power fluctuation of resident's temperature control load control system Method [J] Proceedings of the CSEE, 2012,32 (25): 36-43.

Wang Chengshan,Liu Mengxuan,Lu Ning.A tie-line Power smoothing method for microgrid using residential thermostatically-controlled loads[J] .Proceedings of the CSEE,2012,32(25):36-43(in Chinese).

[9] wangdan, Fan Menghua, Jia Hongjie consider the response of household temperature control workload demand and efficiency electricity that user comfortably constrains Factory models [J] Proceedings of the CSEE, 2014,34 (13): 2071-2077.

Wang Dan,Fan Menghua,Jia Hongjie.User comfort constraint demand response for residential thermostatically-controlled loads and efficient power plant modeling[J].Proceedings of the CSEE,2014,34(13):2071-2077(in Chinese).

[10]Paccagnan,Dario,Kamgarpour M,et al.On the Range of Feasible Power Trajectories for a Population of Thermostatically Controlled Loads[C].IEEE 54th Annual Conference,2015.

[11] Liu Meng, Liang Wen, Zhang Ye wait the numerical value La Pula of temperature control load group Fokker-Planck equation polymerization model This inverse transformation method for solving [J] electric power system protection and control, 2017,45 (23): 17-23.

Liu Meng,Liang Wen,Zhang Ye,et al.Numerical inverse laplace transform solving method of thermostatically controlled load group's Fokker-Planck equation aggregation model[J].Power System Protection and Control,2017,45 (23):17-23(in Chinese).

[12]Esmaeil S,Abate A.Aggregation and Control of Populations of Thermostatically Controlled Loads by Formal Abstractions[J].IEEE Transactions on Control Systems Technology,2015,23(3):975-990.

[13]Mathieu J,Koch S,Callaway D.State Estimation and Control of Electric Loads to Manage Real-Time Energy Imbalance[J].IEEE Transactions on Power Systems,2013,28(1):430-440.

[14]Zhang W,Lian J,Chang C,et al.Aggregated Modeling and Control of Air Conditioning Loads for Demand Response[J].IEEE Transactions on Power Systems,2013,28(4):4655-4664.

[15] Wu Yibei, Li June, it is old burbulent, wait the extensive controllable burden of by power distribution network risk point under malice control scene Analyse [J] Automation of Electric Systems, 2018,42 (10): 30-37.

Wu Yibei,Li Jun'e,Chen Xiong,et al.Risk analysis of distribution network with large-scale controllable loads with attacks[J].Automation of Electric Power System,2018,42(10):30-37(in Chinese).

[16]Harlamov B.Continuous Semi-Markov Processes[M].Great Britain:ISTE Ltd,2008.

[17]Derler P,Lee Ea,Vincentelli As.Modeling Cyber–Physical Systems [J].Proceedings of the IEEE,2012,100(1):13-28.

[18] mixed color reagent of flexible load and control [J] China in Wang Yun, Liu Dong, Li Qingsheng active distribution network Electrical engineering journal, 2016,36 (8): 2142-2150.

Wang Yun,Liu Dong,Li Qingsheng.Modeling and control of flexible load based on hybrid systemin active distribution network[J].Proceedings of the CSEE,2016,36(8):2142-2150(in Chinese).

[19] Yao's Yao, Zhang Peichao participate in stabilizing the side of microgrid interconnection tie power fluctuation based on the air conditioner load of market control Method [J] Proceedings of the CSEE, 2018,38 (3): 782-791.

Yao Yao,Zhang Peichao.A market-based control method for air conditioner loads to smooth microgrid tie-line power fluctuation[J] .Proceedings of the CSEE,2018,38(3):782-791(in Chinese).

[20]Privault N.Understanding Markov Chains[M].Springer Singapore, 2013.

[21] Zhou Yizhou, Sun Guoqiang, Huang Wenjin wait multizone virtual plant comprehensive energy coordinated scheduling Optimized model [J] Proceedings of the CSEE, 2017,37 (23): 6780-6791.

Zhou Yizhou,Sun Guoqiang,Huang Wenjin,et al.Optimized multi-regional integrated energy coordinated scheduling of a virtual power plant[J] .Proceedings of the CSEE,2017,37(23):6780-6791(in Chinese).

[22] virtual plant schedule risk analysis [J] China electricity of Fan Songli, Ai Qian, the He Xing based on chance constrained programming Machine engineering journal, 2015,35 (16): 4025-4034.

Fan Songli,Ai Qian,He Xing.Risk analysis on dispatch of virtual power plant based on chance constrained programming[J].Proceedings of the CSEE, 2015,35(16):4025-4034(in Chinese).

[23] Chen Baolin Optimum Theory and the Beijing algorithm [M]: publishing house, Tsinghua University, 2005.

[24]Song M,Gao C,Yan H,et al.Thermal Battery Modeling of Inverter Air Conditioning for demand response[J].IEEE Transactions on Smart Grid,2017:1- 13.

[25]Song M,Gao C,Shahidehpour M,et al.Impact of Uncertain Parameters on TCL Power Capacity Calculation via HDMR for Generating Power Pulses[J] .IEEE Transactions on Smart Grid,2018:1-12.

[26]Sun Y,Elizondo M,Lu S,et al.The Impact of Uncertain Physical Parameters on HVAC Demand Response[J].IEEE Transactions on Smart Grid,2014,5 (2):916-923.

The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any Those familiar with the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or replace It changes, these modifications or substitutions should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with right It is required that protection scope subject to.

Claims

1. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load, which is characterized in that this method comprises:

In load layer, air conditioner load is considered as information physical system CPS, proposes the air conditioner load based on half markoff process Probability Control can guarantee the blocking time of air-conditioning and keep the diversity of cluster state；And utilize hybrid model The coupled relation for establishing air conditioner load physical process and control process realizes the autonomous control of air conditioner load；

In polymerization quotient sheaf, the coordinated control side that air conditioner load is polymerized to virtual regulating units is proposed based on market equilibrium mechanism Method, this method energy effective protection user data privacy simultaneously significantly reduce communication cost；Meanwhile virtual regulating units can be to scheduling The offer curve and pondage for being similar to conventional power unit are provided, to participate in the Optimized Operation of power grid layer.

2. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load according to claim 1, special Sign is that the hybrid model is built by carrying out simultaneous to physical process continuous in CPS and discrete control process Mould obtains, specifically:

Airconditioning control process is departure process, is described using finite state machine, and a kind of Probability Control is proposed, and introduces u₀With u₁Two migration probability state of a control transfers, and blocking time t is set_lock, wherein t_lockInclude blocking time t after closing_offlock With blocking time t after opening_onlock；

Indoor temperature change generated in case is continuous physical process, can indicate that second order expression is such as with equivalent thermal parameter ETP model Under:

In formula, T_a、T_mAnd T_oRespectively indoor air temperature, indoor solid temperature and outdoor temperature；U_aFor the equivalent resistance of indoor and outdoor It is anti-；H_mEquivalent impedance between room air and solid；Q_aAnd Q_mRespectively indoor gas and solid heat transfer amount；Q_ac、Q_iAnd Q_sPoint It Wei not air-conditioning heat output, indoor airflow heat output and solar radiation heat；C_aAnd C_mThe respectively thermal capacitance of room air and solid；m It (t) is air-conditioner switch state, m (t)=1 represents air-conditioning unlatching, and m (t)=0 represents air-conditioning closing；f₁,f₂For heat transfer coefficient.

3. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load according to claim 2, special Sign is, the Semi-Markov Process of the control process specifically:

1) in state spaceOn, as system leave state m, the probability into state n is p_mn, meet:

2) under the premise of the state of next entrance is n, there is distribution F until the time for occurring to shift from m to n_mn(t), should Distribution can be Arbitrary distribution；

Then it is in the distribution F of state m_m(t) meet:

The average waiting time of state m are as follows:

M=1 is enabled, 2,3,4 respectively indicate ON, and OFF, ONLOCK and OFFLOCK state, the state transfer of air conditioner load has unidirectional Property, then it can be obtained by formula (2):

p₁₄=p₄₂=p₂₃=p₃₁=1 (5)

If α_mnFor from state m to the mobility of state n, when the execution period Δ t of half markoff process is more than given threshold, Then have: u₀≈α₁₄Δt,u₁≈α₂₃Δ t has without memory, then F if the residence time of state 1,2 meets exponential distribution₁₄ (t) and F₂₃(t) be respectively parameter be α₁₄And α₂₃Exponential distribution, it may be assumed that

T₃=t_onlock, T₄=t_offlock (8)

If semi-Markov chain be it is irreducible, after finite time, in certain shape probability of state can converge to The unrelated definite value of original state, the referred to as probability of stability；Opposite, it is general that the probability before converging to the probability of stability is referred to as transient state Rate, according to the property of semi-Markov chain, the probability of stability p of each state_mAre as follows:

π₁=π₂=π₃=π₄=0.25 (11)

4. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load according to claim 2, special Sign is that the feature power of the physical process includes:

1) rated power P_rate,ij

P_rate,ijPower after being opened for air-conditioning j in air conditioner load cluster i, the air-conditioning heat output Q in it and formula (1)_a,iMeet such as Lower relationship:

Q_a,ij=COP_ij×P_rate,ij (13)

In formula, COP_ijFor air-conditioning Energy Efficiency Ratio；I=1 ... M, j=1 ... N_i, M is number of clusters, N_iFor the air-conditioning quantity of cluster i；

2) stable state it is expected power P_exp,ij

For the air conditioner load using ON/OFF control, the power at certain momentFor discrete random variable, according to Discrete Stochastic The stable state of the desired value formula of variable, the air conditioner load it is expected power are as follows:

It is apparent from P_exp,ijIt is also migration probability u₀And u₁Function, above formula establishes physical process and control in air-conditioning hybrid model The connection of process processed, wherein p₁And p₃It is the probability of stability of the ON and ONLOCK that are obtained by formula (12)；

In this way, can control the power of entire cluster, by P by the migration probability for controlling each air conditioner load_exp,ijIt is referred to as empty The expectation power for adjusting load, by P_exp,iReferred to as cluster general power；

3) reference power P_base,ij

P_base,ijIt is defined as air-conditioning j in cluster i and maintains set temperature T under no-control state applies_set,ijRequired expectation power, by formula It (1) can be in the hope of reference power are as follows:

As it can be seen that reference power is influenced by outdoor temperature, but since outdoor temperature variation is slower, within a control period, can recognize It is basically unchanged for benchmark power；

The then reference power of cluster i are as follows:

4) dynamic power P_d,ij

P_d,ijIt is defined as within a control period, expectation power needed for making current room temperature change to assigned temperature, if air-conditioning exists K-th of period is with P_d,ijIn one control period of continuous service, room temperature is made to change to temperature T_a,ijIt (k), then can be in the hope of by formula (1) Dynamic power are as follows:

P_d,ij(k)=A_1,ijT_a,ij(k)+B_1,ijT_a,ij(k-1)+C_1,ijT_m,ij(k-1)+D_1,ij(k) (18)

A_1,ij、B_1,ij、C_1,ij、D_1,ij(k) indicate that coefficient related with outdoor temperature and air-conditioning parameter, physical relationship see below literary annex A；T_a,ij(k) indoor gas temperature, the T in kth all end of term are indicated_m,ij(k-1) the indoor solid temperature in all end of term of kth -1 is indicated；

If the room temperature range that air-conditioning j allows is [T_min,ij,T_max,ij], optimum temperature T_set,ij, then dynamic power has following special Value:

δ=0.1 × (T_max,ij-T_min,ij) (19)

5. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load according to claim 4, special Sign is, polymerize to implement polymerization to air conditioner load with counter, the market equilibrium mechanism is applied to the coordination control of cluster System enables each air-conditioning construct demand curve d according to respective electricity consumption urgency and flexibility_ij(π), the demand curve have monotone decreasing Characteristic includes three key points: A (1, P_min,ij), B (0, P_set,ij) and C (- 1, P_max,ij)；Wherein, the price π in abscissa is only Control signal, referred to as virtual price are represented, and is [- 1,1] by its scope limitation；Ordinate is dynamic power；Due to pressing formula (18) the case where dynamic power estimated is likely to occur less than 0 or is greater than rated power, so three in demand curve close Key power are as follows:

P_min,ijIt indicates to consider the minimum dynamic power after soft-sided circle；P_set,ijIt indicates to make room temperature return to most thermophilic after considering soft-sided circle The dynamic power of degree；P_max,ijIt indicates to consider the maximum dynamic power after soft-sided circle；P_max0,ijIndicate the maximum for not considering soft-sided circle Dynamic power, value are determined by formula (18)；P_min0,ijIndicate the minimum dynamic power for not considering soft-sided circle, value is true by formula (18) It is fixed；P_set0,ijExpression does not consider the dynamic power that soft-sided circle makes room temperature return to optimum temperature, and value is determined by formula (18)；P_rate,ij Indicate the rated power of air conditioner load.

Air-conditioning can upgrade demand curve in each control period according to the present situation, in this way, both can accurately report its flexibility, Also it can guarantee users'comfort when responding arbitrary virtual price signal；

Before preparing to release control to air-conditioning cluster, the signal that virtual price is 0 can be issued in advance, according to demand curve, respectively The responding power of air conditioner load is P_set,ij, that is, room temperature is allowed to return to the expectation power of set temperature, in this manner, each air-conditioning Load will keep the probability control mode of half Markov to run a period of time, and room temperature is made to release control again close to after set temperature System, can effectively keep the diversity of air-conditioning state in cluster and alleviate load oscillatory occurences.

6. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load according to claim 4, special Sign is that air conditioner load locally controls specifically:

When air conditioner load is connected to virtual priceAfterwards, curve determines that its responding power is according to demandThen air-conditioning Load need to determine migration probability u₀And u₁Value makes it is expected that power meets:

But u only can not be uniquely determined by above formula₀And u₁, condition is required supplementation with, when air-conditionings a large amount of in cluster are in ON or OFF shape When the average waiting time of state is more than setting high level or is less than setting low value, the air-conditioning that some state is trapped in cluster will lead to Quantity is excessive, to destroy diversity, it is contemplated that the average waiting time of ON or OFF state depends on u₀And u₁, therefore to avoid this Kind phenomenon, forms following supplementary condition, it is assumed that t_onlock=t_offlock=t_lock:

IfProbability u is opened at this time₁It is larger and close probability u₀It is smaller, it should prevent because of u₁It is excessive and lead to OFF The average waiting time T of state₂It is too small, bring formula (7) into formula (22), it is known that T₂It is restricted to [t_lock/1.5,t_lock/ 0.5] area Between, with blocking time t_lockIn the same order of magnitude, u₁Why it is randomly generated, is to further keep air conditioner load cluster Diversity is obtaining u by formula (22)₁Afterwards, then according to formula (21) u is solved_0,Acquire u₀And u₁Afterwards, air conditioner load can be by state machine Model carries out local control.

7. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load according to claim 4, special Sign is, carries out coordinated control to extensive air conditioner load using the market equilibrium mechanism, detailed process is as follows:

1) polymerization process, when each air conditioner load forms demand curve d_ijAfter (π), polymerization quotient i forms aggregate demand song according to the following formula Line D_i(π):

The virtual price is broadcast in cluster by polymerization quotient, and each air conditioner load obtains respective responding powerDue toFor city Field equilibrium price, thus met in cluster level:

8. a kind of information physical modeling and control method for polymerizeing extensive air conditioner load according to claim 4, special Sign is, the external behavior of virtual regulating units includes peak regulation power, peak capacity and peak regulation expense, wherein peak regulation power definition Deviate the value of reference power for virtual robot arm responding power, only considers the field of air conditioner load reduction plans in summer peak of power consumption Scape is equivalent to conventional power unit and increases power output, then the peak regulation power of virtual robot arm i and the relationship of responding power are as follows:

In formula, D_i(1) the minimum response power of virtual robot arm i is indicated；P_base,iFor the reference power of air conditioner load i, value is by formula (16) it determines.

Then the average SOA of virtual robot arm i is indicated are as follows:

It enables each virtual robot arm participate in peak regulation scheduling by way of quotation herein, assumes that each virtual robot arm is offered as the following formula herein:

Above formula show virtual robot arm it is next control the period peak regulation depth it is bigger, currently averagely SOA it is higher, then offer It is higher；

Then the peak regulation expense of virtual robot arm i may be expressed as:

F_i(ΔP_a,i)=λ_c,iΔP_a,i (32)