JP2022179294A - Distributed type architecture for self-adapting virtual power station and economical distribution method therefor - Google Patents

Abstract

To provide a distributed type architecture and economical distribution method therefor, suitable for a virtual power station with a large amount of aggregated distributed type resource.SOLUTION: A self-adapting virtual power station distributed type architecture includes an agent layer, a gateway layer, a database, a message queue server, an adjustment layer and a power network. The agent layer includes a plurality of groups of DERU. The gateway layer is a middle layer and includes an edge computing function, and transmits an adjustment signal issued by the adjustment layer to the DERU. The adjustment layer includes a repetition management service for promoting repetition. To relieve a pressure load of communication with high simultaneous execution, the issuance or subscription of all messages are asynchronous in the time frame of each repetition and to read it as necessary, the message queue server writes important information in the queue into the database.SELECTED DRAWING: Figure 1

Description

TECHNICAL FIELD The present invention relates to the technical field of electrical engineering and its automation, and more particularly to the distributed architecture of adaptive virtual power plants and their economic allocation methods.

With the expansion of power grid scale and the daily increase in the penetration rate of Distributed Energy Resources (DER), the safe and reliable coordination and economic operation of the power system are becoming more and more important. And efficient management and participation in the economic operation of power systems has become one of the research focuses and challenges in the power sector. A Virtual Power Plant (VPP) is an ideal configuration form of an aggregated DER with the help of advanced control, instrumentation, communication, and other technologies and management mechanisms.

The existing economic allocation methods have the following main drawbacks. (1) In the centralized dispatch method, the lowest layer DER unit (DERU) needs to upload their respective parameters and constraints, the central controller solves the problem collectively, and after obtaining the optimal solution, each unit Send dispatch instructions to . Due to the decentralized nature of the large amount of DERU, the centralized method places high demands on the communication bandwidth and the performance of the central storage equipment, and even a single point of failure in the dispatch center can paralyze the system. And it is necessary to upload all the information of the lowest layer, which is disadvantageous to the resource property owner's privacy protection. (2) With the development of agent technology, some scholars have proposed a decentralized economic allocation method that treats DERUs with certain computing and communication capabilities as agents. Each agent's status is equal without the need for a dispatch center, and each unit, after exchanging necessary information with neighboring units, integrates the information obtained to make independent decisions, allowing each unit to exercise independent decision-making ability and individual intelligence. For example, the current typical method--the decentralized economic allocation method based on multi-agent coherence--chooses the marginal cost of each agent's output as the coherence variable, and the coherence of each unit's incremental cost is satisfied. and achieve optimum economic efficiency for the entire region. In the prior art, a leader-follower power system decentralized economic allocation strategy is proposed, but the leader needs to collect the active power shortage ΔP of the entire region in each iteration, which causes the concentration problem. , and the iteration step size has a great impact on the convergence performance. Also, the iteration termination condition may be that if the condition ΔP=0 is met and the iteration terminates, the marginal costs of each agent may not be exactly equal. Prior literature also provided a fully decentralized economic allocation method that does not require a leader, but the initial total output of DERU iterations needs to be equal to the target amount. All of the above distributed dispatch algorithms require the formation of a connected graph in which each agent has a strong communication connection. Therefore, in practical applications, the information silo increases the extra workload of the system, and it is recommended that agents operate during initialization and in order to avoid penalizing the plug-and-play of distributed resources and the online fault tolerance of the system. You should check your system connectivity when exiting . In addition, the convergence speed of the distributed Sdispatch algorithm is highly dependent on the connectivity of the communication topology, and the number of iterations to achieve convergence increases significantly with increasing DER scale, resulting in a large amount of distributed resources. application in the context of (3) Regarding a method for constructing a strongly connected graph that is independent of all DERUs, the conventional literature proposes a distributed computation method based on the Alternate Directional Multiplier Method (ADMM), in which information is distributed among agents. can effectively avoid information silos without the need to exchange Its application is limited because it is obtained by trial and error. Also, all the above methods rely on the DERU cost function as a quadratic function of the output power, which limits the generality of the methods.

The purpose of the present invention is to establish a decentralized economic allocation method and architecture suitable for a virtual power plant that aggregates a large amount of distributed resources. The method has the characteristics of stable convergence condition and self-adaptive adjustment, can realize DERU plug-and-play, the method can deal with the coordinator's single point of failure problem, and improves the efficiency of computation and communication. , while protecting privacy and providing strong fault tolerance. The method has universal applicability as it can be applied to situations where the incremental cost function of DERU is a general monotonically increasing function.

In order to achieve the above object, the present invention as a technical solution,
Including agent layer, gateway layer, database, message queue server, coordination layer and power grid;
The agent layer includes multiple groups of distributed power generation units (DERUs), the units of the DERU being the lowest-level power units or adjustable loads in the virtual power plant and equipped with sensing and computing functions. , can respond to a regulation signal according to its own incremental cost function and execute a final output command, and can continuously regulate the output within an output limit value; Connected to the bus and established a communication connection with the local energy information gateway, one gateway can connect to dozens of DERUs;
The gateway layer is an intermediate layer, with edge computing function, collectively uploads the information of the lowest DERU in each iteration, and conveys the coordination signal issued by the coordination layer to the DERU;
The coordination layer has an iteration management service to facilitate iteration, and all message publications or subscriptions are asynchronous in each iteration timeframe to reduce the load of highly concurrency communication. ;
The message queue server provides a self-adapting virtual power plant distributed architecture that can write important information in the queue to a database so that it can be read as needed.

Step S1 in which the power grid issues an output command, i.e., an output target quantity X _D to the virtual power plant;
step S2 in which the coordination layer issues an initialization coordination signal, ie an incremental cost value λ, to the gateway, and the gateway sends the coordination signal to the online DERU;
DERU calculates its own output size x ^l _i (k) and sensitivity 1/λ′ _i (x _i (k)) and compares with the output limit value, if the output limit value is not exceeded, DERU's output size and its sensitivity to the conditioned signal to the gateway; if the output limit is exceeded, limit the DERU output to the output limit, reset its sensitivity 1/λ' _i (x _i (k)) to zero, and from step S3 of reporting the output size and sensitivity of the DERU to the gateway;
a step S4 of aggregating the total output value and the total sensitivity value of the DERUs to which the gateway is connected and issuing them to the coordination layer via a message queue;
a step S5 in which the adjustment layer calculates the total output value of all DERUs and the upper limit of the step size, and calculates the output error ΔX(k), that is, the difference value between the output target amount X _D and the total output value;
When the output error is small enough, the iteration ends, the adjustment layer issues a unified iteration end signal, each DERU outputs according to the final iteration calculation result, the virtual power plant completes the economic allocation; if so, step S6 of changing the magnitude of the adjustment signal and returning to step S2 for the next iteration until the error is sufficiently small;
Also provided is a decentralized economic allocation method for a self-adapting virtual power plant comprising:

In step S1, if the virtual coordinator of the coordination layer fails, the computing device is re-designated as the new virtual coordinator, and the new virtual coordinator subscribes to the output size x _l and step size h _l statistics by all gateways. , l may be the gateway number.

In step S4, if the coordination layer virtual coordinator fails, the computing device is redesignated as the new virtual coordinator, and the new virtual coordinator subscribes to the output size x _l and step size h _l of all gateway statistics, and the new The virtual coordinator downloads from the database or message queue the output target quantity X _D and the adjustment signal λ(k) of the current iteration and the total output and total sensitivity values of all connected DERUs issued by each gateway. may

In step S6, the rule for changing the adjustment signal size is: if the total output value is lower than the output target amount X _D , increase the adjustment signal in the next iteration; if the total output value is higher than the output target amount X _D , The iteration may be terminated after reducing the adjustment signal in the next iteration until the difference between the total output value and the output target quantity _XD is sufficiently small.

その中で、ｋが反復回数であり；ｈが正のスカラー、反復ステップサイズであり；Ｘ_DがＶＰＰの出力目標量であり、サイズが、

であり、ｘ_iがｉ番目のＤＥＲＵの出力であり、軽減できる負荷については、ｘ_ｉがその負荷軽減量であり；ｘ_ｉの出力制約式がｘ^min _i≦ｘ_i≦ｘ^max _i ｉ＝１、２、…、Ｎであり、その中でｘ^min _iとｘ^max _iはそれぞれが、ｉ番目のＤＥＲＵ出力の上限と下限であり；

をｉ番目ＤＥＲＵの増分コストとし、出力制約式を考慮しない場合、等増分率法則（Ｅｑｕａｌ ｉｎｃｒｅｍｅｎｔａｌ ｒａｔｅ ｃｒｉｔｅｒｉｏｎ ）により、λ₁＝λ₂＝…λ_N＝λ’ 且つ、

を満たす場合、仮想発電所の出力コストが最も低く、経済配分を達成し；出力制約式を考慮する場合、ソルビングプロセスで出力制限を越えるＤＥＲＵ出力を対応する制限値に固定し、制限値に達しないＤＥＲＵの増分コストが一致する場合、仮想発電所は経済配分を達成し；
毎回反復における目標量に対する仮想発電所の総出力の誤差を以下に定義し、

ＤＥＲＵ出力の更新ルールが、

即ち、調整信号がＤＥＲＵの出力範囲を超えると、出力は制限値に固定され、この時点で制限値に達したＤＥＲＵは、調整信号の変化に応答し続けず；その中で、λ^-1 _i（・）がλ_i（・）の逆函数であってもよい。 In step S6, the rule for updating the adjustment signal is as follows:

where k is the number of iterations; h is a positive scalar, the iteration step size; X _D is the output target amount of VPP, and the size is

^, where _xi is the output _of the i _- th DERU, and for the load _that can be mitigated _{, xi} is the ^amount of the load that can be mitigated; 1, 2, . . . , N, where x ^min _i and x ^max _i are respectively the upper and lower bounds of the i-th DERU output;

be the incremental cost of the i-th DERU, and the output constraint is not considered, then by the Equal incremental rate criterion, λ ₁ =λ ₂ = . . . λ _N =λ' and

If the output constraint equation is considered, the DERU output exceeding the output limit is fixed to the corresponding limit value in the solving process, and the limit value is reached. If the incremental costs of non-DERUs are matched, the virtual power plant achieves economic allocation;
Define the error of the total output of the virtual power plant to the target quantity at each iteration as

The update rule for the DERU output is

That is, when the ^regulated signal exceeds the DERU's output range, the output is clamped to the limit value, and the DERU that has reached the limit value at this point will not continue to respond to changes in the regulated signal _; (·) may be the inverse function of λ _i (·).

ここで、ｘ_iとｆ_iのそれぞれがｉ番目のＤＥＲＵの出力及び発電コストであり、削減できる負荷については、ｘ_iとｆ_iはそれぞれがその負荷削減の量と調整コストであり、電力システムでのＤＥＲＵ発電又は調整コストは、限界費用の増加の法則を満たしているので、コスト関数ｆ_iが単調増加関数でありａ_i、ｂ_i、ｃ_iはそれぞれが二次項、一次項係数及び定数項であり、いずれも正数であり、ｈが毎回の反復で変化せずに、ＤＥＲＵの出力制限が考慮されない場合、前記自己適応化仮想発電所の分散型経済配分方法が効果的に収束するための必要且つ十分な条件が以下の通りであり、

この場合、調整ステップサイズｈが任意の値を取ることができず、正数を取る必要があり、且つすべてのＤＥＲＵのコスト関数の二次項係数にのみ関連する上限があり、一次項係数及び定数項とは関係ないので、ステップサイズの上限を以下に定義でき、

したがって、前記の収束条件が以下の通りであり、
ｈ＝ｒｈ_max ０＜ｒ＜１
この式は、即ちすべてのＤＥＲＵのコスト関数が二次関数である場合の反復収束条件であり；
（２）ＤＥＲＵのコスト関数が一般的な増加関数である場合：
ＤＥＲＵのコスト関数の導関数、即ち増分コスト関数λ_i（ｘ_i）が増加関数であり、この時点でのステップ上限が以下の通りであり、

式中、λ’_i（ｘ_i）はｘ_iに対するλ_i（ｘ_i）の一次導関数であり、即ちｆ_iがｘ_iに対する二次導関数であり；
（３）ＤＥＲＵが出力限界に達し、又は通信障害が発生した場合：
ＤＥＲＵが出力限界に達すると、ＤＥＲＵは調整信号の変化に応答できなくなり、したがってｈ_maxを計算するとき、それを無視するべきであり、したがって反復プロセスでは、ステップサイズの集中型更新ルールが、ＤＥＲＵの出力限界を考慮し、ＤＥＲＵの増分コスト関数が一般的な増加関数である場合、ｋ番目の反復では、ステップサイズｈの集中型更新ルールのアルゴリズムが以下の通りであり、
ａ１：ξ←０
ａ２：ｆｏｒ ｉ＝１：Ｎ
ａ３：ｘ_ｉ（ｋ）←ｍａｘ（ｍｉｎ（λ‐１ｉ（λ（ｋ），ｘmax i），ｘmin i）
ａ４：ｉｆ ｘmin i＜ｘ_ｉ（k）＜ｘmax i ｔｈｅｎ ξ←ξ+１／λ' ｉ（ｘ_ｉ（ｋ））
ａ５：ｈ_ｍａｘ（ｋ）←２／ξ
ａ６：ｈ（ｋ）＝ｒｈ_ｍａｘ（ｋ）
特殊な場合の処理：調整信号が対応する出力値が一部のＤＥＲＵの上限より大きく、残りのＤＥＲＵの下限より小さい場合、反復でデッドゾーンが発生し、このとき、ステップサイズの更新は、分母が０の状況が発生し、即ちアルゴリズムａ５行のξ＝０、この状況が発生した場合、アルゴリズムａ４行の制限値の判定は片側のみを取ってもよく、即ち、誤差ΔＸ（ｋ）がプラスである場合、上限のみを判断し、誤差ΔＸ（ｋ）がマイナスである場合、下限のみを判断する；即ち、分母ξは出力が誤差方向に調整できるＤＥＲＵによって総合的に決定できる。 There are three convergence conditions for the iteration step size h:
(1) If all DERU cost functions are quadratic:
If the DERU cost function is a conventional quadratic function:

where x _i and f _i are the output and generation costs of the i-th DERU, respectively, and for the load that can be reduced, x _i and f _i are the amount of load reduction and the adjustment cost, respectively, and the power system Since the DERU generation or adjustment cost at , satisfies the law of increasing marginal cost, the cost function f _i is a monotonically increasing function and a _i , b _i , c _i are the quadratic term, the linear term coefficient and the constant terms, both of which are positive numbers, h does not change at every iteration, and the DERU output limit is not considered, the decentralized economic allocation method of the self-adapting virtual power plant converges effectively. The necessary and sufficient conditions for

In this case, the adjustment step size h cannot take any value, it must take a positive number, and there is an upper bound associated only with the quadratic coefficient of the cost function of all DERUs, the linear term coefficient and the constant term, we can define an upper bound on the step size as

Thus, the convergence conditions for the above are:
h=rh _max 0<r<1
This expression is the iterative convergence condition when the cost function of all DERUs is quadratic;
(2) If the DERU cost function is a general increasing function:
The derivative of the DERU cost function, the incremental cost function λ _i (x _i ), is the increasing function, and the upper step limit at this point is:

where λ' _i (x _i ) is the first derivative of λ _i (x _i ) with respect to x _i , i.e. f _i is the second derivative with respect to x _i ;
(3) When the DERU reaches its output limit or a communication failure occurs:
When the DERU reaches its output limit, it can no longer respond to changes in the regulation signal, so it should be ignored when calculating h _max , so in the iterative process, the centralized update rule of the step size is applied to the DERU Considering the output bound of , and the incremental cost function of DERU is a general increasing function, at the kth iteration, the algorithm for the centralized update rule with step size h is:
a1: ξ←0
a2: for i=1:N
a3: x _i (k)←max(min(λ-1i(λ(k), xmax i), xmin i)
a4: if xmin i< _xi (k)<xmax i then ξ←ξ+1/λ'i( _xi (k))
a5: h _max (k)←2/ξ
a6:h(k)= _rhmax (k)
Special Case Handling: If the output value to which the adjustment signal corresponds is greater than the upper bound of some DERUs and less than the lower bound of the remaining DERUs, an iteration will experience a dead zone, when updating the step size is 0, i.e., ξ = 0 in line Algorithm a5, if this situation occurs, the determination of the limit value in line Algorithm a4 may take only one side, i.e. the error ΔX(k) is positive If , then only the upper bound is judged, and if the error ΔX(k) is negative, only the lower bound;

The present invention has the following beneficial effects compared with the prior art. A decentralized economic allocation method according to an embodiment of the present invention is self-adaptive and can adapt to the high volume and variability characteristics of distributed resources within a virtual power plant. Step-size algorithm decoupling and hierarchical distributed architecture can adapt to the distributed characteristics of distributed resources and realize the plug-and-play of distributed resources.

1) The existence of a theoretical upper bound on the step size ensures that the method has a high convergence, and at the same time the method's self-adaptiveness can effectively solve the economic allocation problem of massive and distributed resources.
2) The lightweight design distributes the computational and communication pressure of the system, greatly improving the operational efficiency and adjustment speed of the virtual power plant.
3) Proposing a coordinator hosting handling method in the event of a single point of failure, combined with plug-and-play functionality to support distributed resources, the whole method has high fault tolerance.
4) Parameter update function and layered architecture protects parameter information of distributed resources and protects privacy.

It is a distributed architecture of virtual power plant based on message queue transmission mechanism. Fig. 3 is a flow chart of a decentralized economic allocation method for virtual power plants;

The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only one of the embodiments of the present invention. Some but not all embodiments. All other embodiments obtained by persons skilled in the art without creative work based on the embodiments of the present invention fall within the protection scope of the present invention.

(1. Virtual power plant economic allocation model)
The economic allocation of a virtual power plant is the rational allocation of the output of each generator group to minimize the total power generation cost under the condition that the total output target of the virtual power plant and the output range of each distributed resource are met. It means to place in

x ^min _i ≤ x _i ≤ x ^max _i i = 1, 2, ..., N (equation 1c)

をｉ番目ＤＥＲＵの増分コスト（Ｉｎｃｒｅｍｅｎｔａｌ Ｃｏｓｔ， ＩＣ）とする。出力制約（１ｃ）を考慮しない場合、等増分率法則（Ｅｑｕａｌ ｉｎｃｒｅｍｅｎｔａｌ ｒａｔｅ ｃｒｉｔｅｒｉｏｎ ）により、λ₁＝λ₂＝…λ_N＝λ’ 且つ（１ｂ）を満たす場合、ＶＰＰの出力コストが最も低く、経済配分を達成する。式（１ｃ）を考慮する場合の制約と最適化問題については、ソルビングプロセスで出力制限値を越えたＤＥＲＵ出力を対応する制限値に固定し、制限値に達しないＤＥＲＵの増分コストが一致する場合、ＶＰＰ（仮想発電所）は経済配分を達成した。 where x _i and f _i are the generation output and generation cost of the i-th DERU, respectively, and in the case of a reducible load, x _i and f _i are its load curtailment and adjustment cost, respectively. Since the DERU generation or regulation cost in a power system satisfies the marginal cost law, f _i (the cost function) is a monotonically increasing function. N is the total number of DERUs, F is the total power generation cost of the virtual power plant, X _D is the output target amount of the VPP, and x ^min _i and x ^max _i are the upper and lower limits of the output power of the i-th DERU, respectively.

Let be the Incremental Cost (IC) of the i-th DERU. Without considering the output constraint (1c), the VPP _has the lowest output cost if λ ₁ =λ ₂ = . Achieve economic allocation. For constraints and optimization problems when considering equation (1c), if the solving process fixes the DERU output that exceeds the output limit to the corresponding limit, and the incremental cost of the DERU that does not reach the limit matches , VPP (virtual power plant) achieved economic allocation.

(2.Description of decentralized economic distribution method for virtual power plants)
For the distributed dispatch method, since there is no direct communication between each unit, an upper-layer entity is required to coordinate the iteration direction of each unit. VC). In addition, the distributed architecture design distributes the computational load, so that the coordinator has significantly reduced performance requirements compared to traditional central controllers, which can also be burdened by local energy information gateways. . Therefore, this method weakens the demands on the functionality of the VC through proper architecture and process design, so that middle-tier entities with all (or most) edge intelligence can play the role of coordinator. , causes all units as a whole to optimally complete one total task. Also, since the system is lightweight and does not place high demands on VC performance, it can more conveniently host reconciliation functions in the event of a VC failure, providing some degree of fault tolerance.

In order to satisfy the equal rate law of economic allocation, at each iteration, the VCs issue a unified incremental cost value λ as the adjustment signal, and each DERU calculates and reports its output accordingly. In order to satisfy the power balance constraint in equation (2), based on the feedback idea, the VC compares the aggregated DERU total output with the target value _XD , and if it is lower than _XD , the adjusted signal at the next iteration , and if higher than X _D , reduce the adjustment signal until the error between the total DERU output at VPP and the target value X _D is sufficiently small, then the iteration ends and VC returns to the unified iteration end A signal is issued, each DERU outputs according to the final iterative calculation result, and the VPP completes the economic allocation. In order to ensure fault tolerance, in each iteration, the system self-adaptively adjusts the step size from bottom to top according to the DERU online situation and running status to ensure the speed of convergence.

(3. Iterative update rules for parameters)
The update rule of the adjustment signal is as follows.

where k is the number of iterations; h is a positive scalar, the iteration step size, which has a significant effect on convergence. The error of the VPP total output to target amount at each iteration is defined below.

ＤＥＲＵ出力の更新ルールは次の通りである。

The update rule for the DERU output is as follows.

That is, when the regulation signal exceeds the output range of the DERU, the output is clamped to the limit value, and the DERU, which has reached the limit value at this point, will not continue to respond to changes in the regulation signal. Among them, λ ^-1 _i (·) is the inverse function of λ _i (·).

(4.Convergence condition)
((1) All DERU cost functions are quadratic functions)
The DERU cost function is a conventional quadratic function:

, where a _i , b _i , and c _i are the quadratic term, the linear term coefficient, and the constant term, respectively, and are positive numbers), then h remains unchanged at each iteration, and the output of DERU is If no constraints are considered, the sufficient and necessary conditions for the above process to effectively converge are as follows.

where the adjustment step size h cannot take an arbitrary value, it must take a positive number, and there is an upper bound associated only with the quadratic term coefficient of the cost function of all DERUs, the linear term coefficient and the constant item is irrelevant. Therefore, an upper bound on the step size can be defined as follows.

Therefore, the above convergence condition is as follows.

h=rh _max 0<r<1

This expression is the iterative convergence condition, ie when the cost function of all DERUs is quadratic.

((2) DERU's cost function is a general increasing function)
More generally, the derivative of the DERU cost function, the incremental cost function λ _i (x _i ), is the incremental function, and the upper step limit at this point is:

where λ' _i (x _i ) is the first derivative of λ _i (x _i ) with respect to x _i ie the second derivative of f _i versus x _i .

((3) When DERU reaches its output limit or communication failure occurs)
When the DERU reaches its output limit, it can no longer respond to changes in the adjustment signal and should therefore be ignored when calculating h _max , so in the iterative process, the step-size centralized update rule is: Street.

(Algorithm 1: If DERU's incremental cost function is a general increasing function, considering the output limit of DERU, then at the kth iteration, centralized update rule with step size h)
Input: Incremental cost function λ _i (x) _i of all DERUs, adjusted signal λ(k);
Output: Changed step size h(k) of the training signal in this iteration

If a DERU encounters a communication failure, it will not upload its own parameter values during the iteration, and finally will not receive the iteration end signal issued by the VC, and will output according to the original working point, and other DERUs will The step-size update rule of Algorithm 1 can also guarantee convergence because it does not affect the iterative process of . In a physical sense, 1/λ' _i (x _i (k)) is the sensitivity of the i th DERU's output to the conditioning signal at the k th iteration. The higher the sensitivity, the stronger the ability of the DERU to respond to the adjustment signal within the output limit range, ie, the greater the ratio of the DERU output change micro-increments to the adjustment signal change micro-increments. When viewed in combination with Algorithm 1, the step size has a self-adaptive feature, the higher the number of DERUs with response function, the higher the output sensitivity of each DERU, which in turn reduces the step size and the possibility of over-tuning. becomes smaller. Conversely, the smaller the number of responsive DERUs and the smaller the output sensitivity of each DERU, the larger the step size and the smaller the possibility of under-adjustment.

Special Case Handling: If the output value corresponding to the adjustment signal is greater than the upper bound of some DERUs and less than the lower bound of the remaining DERUs, a dead zone occurs in the iteration, when the step size update is is 0 (.xi.=0 on line 5 of Algorithm 1). If this situation occurs, the limit determination in line 4 of Algorithm 1 may be one-sided, i.e., if the error ΔX(k) is positive, only the upper limit is determined, and the error ΔX(k ) is negative, only the lower bound is determined. That is, the denominator ξ can be comprehensively determined by the DERU that can adjust the output in the error direction.

Algorithm 1 describes the step-size update rule in a centralized situation, and experiments show that r=0.5 gives the fastest convergence speed. Moreover, by computing the internal architecture and decoupling of the distributed virtual power plant, we can design it as a distributed step-size update scheme. At each iteration, each DERU computes its own output x _i (k) and 1/λ' _i (x _i (k)), so that each unit's independent decision-making and individual intelligence is exercised, Computational load is distributed. In addition, due to the layered distributed architecture of "DERU-gateway-VC", the gateway aggregates the parameter information of dozens of connected DERUs as an intermediate-layer entity and uploads the aggregated intermediate data. The computational burden of VC is further reduced, the parameter information of a single DERU is protected from VC, and privacy is protected. In terms of fault tolerance, computing the step size with Algorithm 1 guarantees reliability and speed of convergence in the event of a communication failure in the DERU. In the event of a communication failure on the VC, hosting allows the coordination function to be hosted on other computational devices, preserving the necessary data during the iteration process so that all gateways establish communication connections with the new VC and iterate. can continue. Therefore, the VC hosting mechanism combined with step-size self-adaptiveness can achieve high fault tolerance and high availability of the overall system.

(V. Virtual Power Plant Distributed Architecture)
From the description of the decentralized economic allocation method and the convergence conditions, the upper layer VC of the virtual power plant is required to count the total DERU output and the necessary parameter information at each iteration, update the step size, and adjust the uniformity. Emitted after computing the signal. All DERUs feed back their output according to the adjustment signal and in this way feed back until the error between the DERU's total output and the output command target is within tolerance. As shown in Figure 1, a distributed architecture of virtual power plant based on message queue transmission mechanism was designed to decouple different parts and enhance the scalability and fault tolerance of the system. Inside the virtual power plant, the DERU is the lowest layer (agent layer) power supply or adjustable load, equipped with sensing and computing capabilities, responding to adjustment signals according to its own incremental cost function, and executing the final output command. , the power output can be continuously adjusted within the output limits. The DERUs are connected to the power bus of the virtual power plant through the energy interface and establish a communication connection with the local energy information gateway, and one gateway can connect dozens of DERUs. The gateway layer in the middle has the edge computing function, and in each iteration, it collects and uploads the information of the lowest layer DERU, and conveys the coordination signal issued by the VC to the DERU. Coordination layer VCs logically establish communication connections with all gateways, update step sizes and coordination signals to adjust the total output of the virtual power plant, and finally meet the power grid's output command. . The coordination layer has an iteration management service to facilitate iteration, and every iteration timeframe, all message publications or subscriptions are asynchronous, alleviating the pressure of highly concurrency communication. A message queue server writes important information in a queue to a database, making it ready for reading when needed.

The message queue sending mechanism only requires that all devices in the gateway layer and coordination layer establish a connection with the message queue server, and each device subscribes to the messages it needs through listeners. A gateway in the gateway layer needs to subscribe to the incentive signaling information issued by the VC, and the VC needs to subscribe to the summarized information of all the gateway's parameters. Every DERU computes its own output separately to respond to the coordination signal, and after gateway layer aggregation, the communication and computation load of the VC is reduced. VC uses Algorithm 1 and Equation (1) to update the step size and excitation signal, and the layered and distributed architecture and communication decoupling design makes VC very light in communication and computation. , VC can also be borne by the gateway device. In the event of a single point of failure, the functionality of the VC can be hosted on other gateways and all gateways can resume iteration by communicating only with the new VC. Due to its lightweight design, VCs can be directly borne by any gateway in the gateway layer. If a VC fails, a PROBATE message is issued to trigger the coordinator hosting mechanism. After the reconciliation function is hosted on another gateway or computation server, the new VC downloads only the iteration-related parameter information from the message queue or database and then continues configuring the iteration. Therefore, the database or message queue only needs to hold the total output instructions and some information in the current iteration in order to restart the iteration, so the storage load is relatively light.

である。ｘ^l _iは、ｌ番目のゲートウェイに接続されたｉ番目のＤＥＲＵの出力である。Ｘ_lとｈ_lは、それぞれがｌ番目のゲートウェイに接続されているすべてのＤＥＲＵの総出力と、増分コスト関数の導関数の逆数の合計である。ＶＣは、すべてのゲートウェイによって発行されたＸ_lとｈ_l情報をサブスクライブする必要がある。方法は、ステップサイズの更新及び計算プロセスをデカップリングし、且つＶＣホスティングメカニズムを介してＶＣ単一障害点を解決し、方法のフォールトトレランスと可用性を向上させる。ＶＣに障害が発生した場合、新しいＶＣは今回反復でのＸ_l、ｈ_l、λ及びＸ_Dをダウンロードする必要があり、そして同じフローを通じてＶＣの機能を実行して反復を復元できるため、コーディネーターの単一障害点の問題が解決される。ＤＥＲＵ又はゲートウェイに障害が発生し、又はオフラインになった場合、ステップサイズの計算はそれを無視する。反復プロセスでＤＥＲＵを追加する場合、収束条件を満たすようにステップサイズも下から上に更新されるため、収束パフォーマンスに影響を与えることなく、分散型リソースのプラグアンドプレイを実現できる。 (6. Method flow design)
The basic flow of the decentralized economic allocation method is shown in Figure 2. where l is the gateway number and L is the number of gateways in the gateway layer. The number of DERUs connected to the lth gateway is _n1 . Namely

is. x ^l _i is the output of the i-th DERU connected to the l-th gateway. X _l and h _l are the sum of the total output of all DERUs each connected to the lth gateway and the reciprocal of the derivative of the incremental cost function. A VC needs to subscribe to the X _l and h _l information published by all gateways. The method decouples the step size update and computation process, and solves the VC single point of failure through the VC hosting mechanism, improving the fault tolerance and availability of the method. If a VC fails, the new VC needs to download X _l , h _l , λ and X _D in this iteration, and can perform the VC's function through the same flow to restore the iteration, so the coordinator single point of failure problem is solved. If a DERU or gateway fails or goes offline, the step size calculation ignores it. When adding DERUs in the iterative process, the step size is also updated from bottom to top to satisfy the convergence condition, thus realizing plug-and-play of distributed resources without affecting the convergence performance.

Both the distributed architecture and the method flow of the virtual power plant are characterized by a high degree of decoupling, the gateway summarizes the information of the lowest layers and uploads the intermediate results to the coordination layer after the summation calculation, so that each iteration And, since the coordinator does not know the specific parameters and status information of individual DERU units, it has a privacy-preserving feature. In addition, the architecture can be improved, such as multi-layer middle layer design, decoupling each body in the same layer, upper layer uploading information of lower layer collectively, making each device's communication and calculation lighter. can be done.

In the description herein, descriptions that refer to terms such as "one embodiment," "example," "particular example," and the like refer to particular features, structures, materials, or properties being described in combination with the embodiment or example. is included in at least one embodiment or example of the invention. In this specification, the generalized descriptions of the terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or illustrations.

The preferred embodiments of the invention disclosed above are used only to help explain the invention. The preferred embodiments have not been described in every detail, nor are the inventions limited to only the particular embodiments described. Of course, many modifications and variations can be made in light of the content of this specification. This specification has chosen and specifically described these embodiments in order to better explain the principles and practical applications of the present invention so that those skilled in the art can fully understand and use the invention. . The invention is limited only by the claims and their full scope and equivalents.

Claims (7)

Including agent layer, gateway layer, database, message queue server, coordination layer and power grid;
the agent layer includes multiple groups of distributed power supply units (DERUs), the DERU units being the lowest power supply units or adjustable loads in a virtual power plant, with sensing and computing capabilities; The DERU is capable of responding to a regulation signal and executing a final output command according to its incremental cost function, continuously regulating the output within a range of output limits, and said DERU is connected to the virtual power plant power bus through an energy interface. and establish a communication connection with the local energy information gateway, one gateway can connect to dozens of DERUs;
The gateway layer is an intermediate layer, with edge computing function, collectively uploads the information of the lowest DERU in each iteration, and conveys the coordination signal issued by the coordination layer to the DERU;
The coordination layer has an iteration management service to facilitate iteration, and all message publications or subscriptions are asynchronous in each iteration timeframe to reduce the load of highly concurrency communication. ;
A self-adaptive virtual power plant distributed architecture characterized in that said message queue server can write important information in the queue to a database so that it can be read as needed. Step S1 in which the power grid issues an output command, i.e., an output target quantity X _D to the virtual power plant;
step S2 in which the coordination layer issues an initialization coordination signal, ie an incremental cost value λ, to the gateway, and the gateway sends the coordination signal to the online DERUs;
DERU calculates its own output size x ^l _i (k) and sensitivity 1/λ′ _i (x _i (k)) and compares with the output limit value, if the output limit value is not exceeded, DERU's output size and its sensitivity to the conditioned signal to the gateway, and if the output limit is exceeded, limit the DERU output to the output limit, reset its sensitivity 1/λ' _i (x _i (k)) to zero, and from step S3 of reporting the output size and sensitivity of the DERU to the gateway;
a step S4 of summarizing the total output value and the total sensitivity value of the DERUs to which the gateway is connected and issuing them to the coordination layer via a message queue;
a step S5 in which the adjustment layer calculates the total output value of all DERUs and the upper limit of the step size, and calculates the output error ΔX(k), that is, the difference value between the output target amount X _D and the total output value;
When the output error is small enough, the iteration ends, the adjustment layer issues a unified iteration end signal, each DERU outputs according to the final iteration calculation result, the virtual power plant completes the economic allocation, and the output error is if so, step S6 of changing the magnitude of the adjustment signal and returning to step S2 for the next iteration until the error is sufficiently small;
A decentralized economic allocation method for a self-adapting virtual power plant, comprising: In step S1, if the virtual coordinator of the coordination layer fails, the computing equipment is redesignated as the new virtual coordinator, and the new virtual coordinator subscribes to the output size _xl and step size _hl statisticed by all gateways. 3. The decentralized economic allocation method of self-adapting virtual power plants according to claim 2, wherein l is a gateway number. In step S4, if the coordination layer virtual coordinator fails, the computing device is re-designated as a new virtual coordinator, and the new virtual coordinator subscribes to the output size x _l and step size h _l of all gateway statistics. , the new virtual coordinator obtains from the database or message queue the output target quantity X _D and the adjustment signal λ(k) of the current iteration and the total output and total sensitivity values of all connected DERUs issued by each gateway. The decentralized economic allocation method of self-adaptive virtual power plant according to claim 3, characterized by downloading. In step S6, the rule for changing the adjustment signal size is: if the total output value is lower than the output target amount X _D , increase the adjustment signal in the next iteration; if the total output value is higher than the output target amount X _D , Self-adaptive virtual power generation according to claim 4, characterized in that the iteration ends when the adjustment signal in the next iteration is reduced until the difference between the total output value and the output target value X _D is sufficiently small. decentralized economic allocation method. The update rule for the adjustment signal is as follows,

where k is the number of iterations, h is a positive scalar, the iteration step size; X _D is the VPP output target, and the size is

, where _xi is the output of the i-th DERU, and for the load that can be relieved, _xi is its load reduction amount, and the output constraint equation for _xi is x ^min _i ≤ xi ≤ x ^max _i i=1 , 2, . . . , N, where x ^min _i and x ^max _i are respectively the upper and lower bounds of the i-th DERU output;

be the incremental cost of the i-th DERU, and the output constraint is not considered, then by the Equal incremental rate criterion, λ ₁ =λ ₂ = . . . λ _N =λ' and

If the virtual power plant has the lowest output cost, achieves economic allocation, and considers the output constraint equation, the solving process fixes the DERU output exceeding the output limit to the corresponding limit value, and the limit value is reached. If the incremental costs of non-DERUs are matched, the virtual power plant achieves economic allocation;
Define the error of the total output of the virtual power plant to the target quantity at each iteration as

The update rules for the DERU output are as follows:

That is, when the regulation signal exceeds the output range of the DERU, the output is clamped to the limit value, and the DERU which reaches the limit value at this point will not continue to respond to the regulation signal, in which λ ^-1 _i ( ) is an inverse function of λ _i (·). There are three convergence conditions for the iteration step size h:
(1) All DERU cost functions are quadratic:
The DERU cost function is a conventional quadratic function:

where x _i and f _i are the output and generation costs of the i-th DERU, respectively, and for the load that can be reduced, x _i and f _i are the amount of load reduction and the adjustment cost, respectively, and the power system Since the DERU generation or adjustment cost at , satisfies the law of increasing marginal cost, the cost function f _i is a monotonically increasing function and a _i , b _i , c _i are the quadratic term, the linear term coefficient and the constant terms, both of which are positive numbers, h does not change at every iteration, and the DERU output limit is not considered, the decentralized economic allocation method of the self-adapting virtual power plant converges effectively. The necessary and sufficient conditions for

This is because the adjustment step size h cannot take any value, it must be considered positive, and there is an upper bound associated only with the quadratic coefficient of the cost function of all DERUs, the linear term coefficient and the constant term, we can define an upper bound on the step size as

Thus, the convergence conditions for the above are:
h=rh _max 0<r<1
This expression is the iterative convergence condition when the cost function of all DERUs is quadratic;
(2) DERU's cost function is a general increasing function:
The derivative of the DERU cost function, the incremental cost function λ _i (x _i ), is the increasing function, and the upper step limit at this point is:

where λ' _i (x _i ) is the first derivative of λ _i (x _i ) with respect to x _i , i.e. the second derivative of f _i versus x _i ;
(3) DERU reaches output limit or communication failure occurs:
When the DERU reaches its power limit, it can no longer respond to changes in the regulation signal, so it should be ignored when calculating h _max , so in the iterative process, the centralized update rule of the step size is applied to the DERU Considering the output limit of , and the incremental cost function of DERU is a general increasing function, at the kth iteration, the algorithm for the centralized update rule with step size h is:

Special Case Handling: If the output value to which the adjustment signal corresponds is greater than the upper bound of some DERUs and less than the lower bound of the remaining DERUs, an iteration will experience a dead zone, when updating the step size is 0, i.e., ξ=0 in line Algorithm a5, and if this situation occurs, the limit determination in line Algorithm a4 may be one-sided, i.e. the error ΔX(k) is positive, only the upper limit is judged, and if the error ΔX(k) is negative, only the lower limit is judged, that is, the denominator ξ can be determined synthetically by DERU, the output of which can be adjusted in the direction of the error.
The decentralized economic allocation method of self-adaptive virtual power plant according to claim 6, characterized in that:

Images