Method and system for allocating distributed generation instructions for virtual power plants in distribution networks

Abstract

The present invention discloses a method and system for allocating distributed generation instructions for a virtual power plant in a distribution network. The method first linearizes the AC power flow model of the distribution network, models the line active power and node voltage in the distribution network, and represents the node voltage and line active power as a mapping of the node power variation. Then, constraints including the power balance constraint, node voltage constraint, line capacity constraint, upper and lower limit constraints of the adjustable unit power regulation, and ramp rate constraint of each adjustable unit are set to construct a generation instruction scheduling optimization model with the highest frequency regulation performance and the lowest carbon emission as the dual optimization goals. Finally, a distributed algorithm based on Nesterov momentum acceleration is adopted to solve the generation instruction scheduling optimization model within each control cycle, thereby achieving fast and accurate solution of instructions, and setting communication conditions to reduce the communication pressure during distributed information exchange, thereby improving the convergence speed of the model. The method of the present invention has fast convergence speed and high precision.

Description

Technical Field

[0001] The present invention belongs to the field of internal power instruction distribution of virtual power plants, and in particular relates to a method and system for distributing distributed power generation instructions of virtual power plants oriented to a distribution network. Background Art

[0002] As the world transitions to zero-carbon emissions, the number of distributed energy resources (DERs) is increasing. However, a significant portion of these resources is not directly managed by the power system. Virtual power plant (VPP) technology offers an effective solution for coordinated control and optimized management of internal DERs. By aggregating and precisely controlling various DERs, VPPs can provide auxiliary frequency regulation services. From the perspective of an externally controlled power source, a VPP tracks power regulation commands (PRCs) sent by the grid dispatch center and adjusts its power in real time to meet the power system's frequency regulation needs. From the perspective of internal regulation and control allocation, it requires efficient allocation of internal DERs with diverse characteristics through generation command dispatch (GCD). Traditional generation command decomposition is mostly centralized, with limited ability to process massive amounts of data, prone to single points of failure during communication, and difficulty meeting unit privacy requirements. Therefore, GCD for virtual power plants is gradually evolving towards decentralized, flexible, and open solutions, with distributed algorithms becoming an effective approach to addressing GCD.

[0003] There are three major pain points in the distributed GCD of virtual power plants: First, when designing the objective function, most of them assume that all DERs in the VPP are connected to an equivalent node, ignoring power flow changes in the network, voltage node over-limit, and other situations. Since traditional load changes are slow and highly predictable, traditional GCD based on slower time scale economic dispatch will not cause significant losses. However, in modern VPPs with high renewable energy penetration, short-term power fluctuations will increase significantly and be difficult to predict, and the internal topology of the VPP is complex. Therefore, the traditional distributed GCD method cannot quickly respond to frequency regulation instructions while reducing carbon emissions within the virtual power plant, achieving a cleaner and low-carbon virtual power plant operation effect, and the distributed GCD model becomes complex and time-sensitive. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention provides a method and system for allocating distributed generation instructions for virtual power plants (VPPs) in distribution networks. This method introduces a linearized AC power flow model to model the active power and node voltage in the VPP. The power regulation limits, line capacity constraints, and node voltage constraints of each frequency regulation resource are considered as key constraints. An optimal GCD model is designed with the dual optimization goals of maximizing frequency regulation performance and reducing carbon emissions. The algorithm solution utilizes a distributed algorithm based on Nesterov accelerated gradients to achieve rapid and accurate instruction resolution. Communication conditions are set to reduce communication pressure during distributed information exchange and improve the convergence speed of distributed GCD iterations.

[0005] The specific technical solutions of the present invention are as follows:

[0006] A method for allocating distributed power generation instructions for a virtual power plant in a distribution network, the method comprising the following steps:

[0007] Step 1: Linearize the AC power flow model of the distribution network, model the line active power and node voltage in the distribution network, and express the node voltage and line active power as a mapping of node power changes;

[0008] Step 2: Set constraints, including power balance constraints for each adjustable unit, node voltage constraints, line capacity constraints, upper and lower limits for adjustable unit power regulation, and ramp rate constraints, to build a generation command scheduling optimization model with the dual optimization objectives of maximizing frequency regulation performance and minimizing carbon emissions.

[0009] Step 3: Using a distributed algorithm based on Nesterov momentum acceleration, the power generation instruction scheduling optimization model is solved in each control cycle to achieve fast and accurate solution of the instructions, and communication conditions are set to reduce the communication pressure during distributed information exchange and improve the convergence speed of the power generation instruction scheduling optimization model.

[0010] Furthermore, the step three specifically includes the following sub-steps:

[0011] S3.1: Decouple the Lagrange multiplier in the power generation command scheduling optimization model into n subproblems, so that each adjustable unit can locally solve the Lagrange multiplier corresponding to itself;

[0012] S3.2: Initialize the Lagrange multiplier, iteration step size, and number of iterations, and introduce the performance deficiency index of each adjustable unit and the power adjustment value at the previous moment;

[0013] S3.3: Update the Lagrange multiplier using a distributed algorithm based on Nesterov momentum acceleration, specifically including: calculating the actual power instruction of the i-th adjustable unit at time t corresponding to iteration round k+1 according to the Lagrange multiplier update result Intermediate variables ; Combined with the power regulation upper and lower limit constraints and climbing rate constraints of each adjustable unit, Correction is performed to obtain the optimal actual power instruction of the i-th adjustable unit at time t corresponding to iteration round k+1;

[0014] S3.4: Repeat the iteration according to the updated result of the Lagrange multiplier until the iteration termination condition is reached, and obtain the optimal actual power instruction of the i-th adjustable unit at time t.

[0015] Furthermore, the step S3.3 is implemented by the following sub-steps:

[0016] S3.3.1: Use the Nesterov momentum term instead of the Lagrange multiplier and perform gradient updates on the Nesterov momentum term. Use the exponentially weighted moving average method to calculate the weighted sum v of the Nesterov momentum gradient. i ;

[0017] S3.3.2: Introduce a correction step to eliminate the local incremental cost ψ obtained by Nesterov accelerated gradient descent i and the weighted sum v of the Nesterov momentum gradient i The difference between i Ability to stabilize at the optimal point during iterative engineering;

[0018] S3.3.3: Combine the correction information of neighboring nodes to achieve global consistency.

[0019] Furthermore, in step 3, communication conditions are set to reduce the communication pressure during distributed information exchange, specifically:

[0020] During the communication process of the adjustable unit, if the change in the Lagrange multiplier between two adjacent iterations is less than the set termination condition, the information exchange will be stopped.

[0021] Furthermore, in the step 2,

[0022] The power balance constraint is to meet the balance between the total frequency regulation mileage and the frequency regulation demand when responding to the total instruction issued by the dispatching center;

[0023] The node voltage constraint is that the node voltage of the important node is within the upper and lower limits of the voltage safety range; the important node is a node to which the virtual power plant adjustable unit is connected;

[0024] The line capacity constraint is to ensure that the power of all lines is within the upper and lower limits of the power of the current line during the power generation instruction scheduling process;

[0025] The upper and lower limits of the power adjustment of the adjustable units are constrained to ensure that the adjustment power of each adjustable unit is within the upper and lower limits of the power adjustment;

[0026] The ramp rate constraint is that the power regulation rate of each adjustable unit is within an upper and lower limit range of the power regulation rate of the adjustable unit.

[0027] Furthermore, the power generation command scheduling optimization model is a linear weighted sum of the frequency regulation contribution function and the carbon emission function of all adjustable units;

[0028] The frequency modulation contribution function of the adjustable unit is specifically the square of the difference between the power adjustment value of the adjustable unit at the current moment and the power adjustment value at the previous moment, multiplied by the performance deficiency index σ of the corresponding adjustable unit. i ;

[0029] The carbon emission function of the adjustable unit is specifically a quadratic function of the actual power instruction of the adjustable unit at the current moment, and the coefficient of the quadratic function is the carbon emission coefficient.

[0030] Furthermore, in the step 1,

[0031] Linearizing the AC power flow model of the distribution network, specifically: decoupling the voltage amplitude and phase angle of the AC power flow model of the distribution network;

[0032] The node voltage is expressed as a mapping of the node power change. Specifically, the active-voltage sensitivity matrix is ​​introduced to express the node voltage as the relationship between the active power changes of each node;

[0033] The line active power is expressed as a mapping of node power changes. Specifically, based on the linearized AC power flow model of the distribution network, an active-voltage sensitivity matrix is ​​introduced to express the node voltage as the relationship between the active power changes of each node; and a power transmission allocation matrix is ​​introduced to express the node phase angle difference as the relationship between the total output power of each node.

[0034] Furthermore, it includes one or more processors for implementing a method for allocating distributed power generation instructions of a virtual power plant for a distribution network.

[0035] An electronic device, comprising:

[0036] one or more processors;

[0037] A storage device is used to store one or more programs. When the one or more programs are executed by the electronic device, the electronic device implements a method for allocating distributed power generation instructions of a virtual power plant for a distribution network.

[0038] A computer-readable storage medium stores a program, which, when executed by a processor, implements a method for allocating distributed power generation instructions for a virtual power plant in a distribution network.

[0039] The beneficial effects of the present invention are as follows:

[0040] (1) The present invention introduces a linearized AC power flow model to model the distribution network where the virtual power plant is located, and takes the node voltage constraint and line capacity constraint as one of the constraints of the multi-objective optimization problem to prevent the voltage and power of the adjustable unit from exceeding the limit during power adjustment.

[0041] (2) The present invention uses a distributed algorithm based on Nesterov momentum acceleration to achieve fast and accurate solution of instructions, accelerate convergence speed, and ensure convergence accuracy. The method of the present invention is not prone to single point failure, has high scalability and flexibility, and protects the privacy of adjustable units in the virtual power plant;

[0042] (3) The present invention sets communication conditions during the iteration process to reduce the communication pressure during distributed information exchange and improve the convergence speed of distributed GCD. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, without paying any creative work, they can also obtain drawings of other embodiments based on these drawings.

[0044] Figure 1 This is a flow chart of a method for rapidly allocating distributed power generation instructions of a virtual power plant for a low-carbon distribution network according to an embodiment of the present invention.

[0045] Figure 2 This is a power residual comparison chart of the improved distributed algorithm based on Nesterov momentum acceleration, the precise diffusion algorithm, the standard diffusion algorithm, and the consensus algorithm.

[0046] Figure 3 This is the power adjustment value result diagram of each adjustable unit, and the horizontal axis is time (s).

[0047] Figure 4 Tracking P for Virtual Power Plant RRC The power curve, the horizontal axis is time (s).

[0048] Figure 5 This is the carbon emission curve of the virtual power plant participating in the secondary frequency regulation process, and the horizontal axis is time (s). DETAILED DESCRIPTION

[0049] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments, and the purpose and effects of the present invention will become more apparent. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0050] The present invention provides a method for quickly allocating distributed power generation instructions for a virtual power plant in a low-carbon distribution network. First, a linearized AC power flow model is introduced. The voltage sensitivity matrix and the power transmission allocation matrix are used to express the active power flow of the line as a mapping relationship of the node power change. The node voltage constraint and the line capacity constraint are used as important constraints of the multi-objective optimization problem to prevent the adjustable unit from exceeding the limit during power adjustment. When constructing the optimization problem, a GCD model with the dual optimization goals of improving frequency regulation and reducing carbon emissions is designed to achieve improved frequency regulation performance and reduced carbon emissions. For this optimization model, the Lagrange multiplier in the optimization model is decoupled into n sub-problems so that the problem can be solved locally in each adjustable unit; a distributed algorithm based on Nesterov momentum acceleration is adopted, and the Nesterov momentum gradient is introduced instead of the traditional gradient term to accelerate the convergence speed and ensure the convergence accuracy. Communication conditions are set during the iteration process to reduce the communication pressure during distributed information exchange and improve the convergence speed of distributed GCD.

[0051] This embodiment is introduced by taking a virtual power plant composed of n distributed resources as an example. Figure 1 As shown in FIG, the method for rapidly allocating distributed generation instructions of a virtual power plant for a low-carbon distribution network specifically includes the following steps:

[0052] S1: Linearize the AC power flow model of the distribution network, model the line active power and node voltage in the distribution network, and express the node voltage and line active power as a mapping of node power changes.

[0053] The element Y in the power system admittance matrix Y pq It can be expressed as:

[0054]

[0055] Where G pq is the element Y in the power system admittance matrix pq The real part, B pq is the element Y in the power system admittance matrix pq The imaginary part of y pp is the parallel admittance of node p, y pq is the admittance of the line l between nodes p and q, ypq =g pq +j*b pq , g pq represents the conductance of the line l between nodes p and q, b pq represents the susceptance of the line l between nodes p and q.

[0056] The present invention does not consider reactive power, and the AC power flow equation of the distribution network is expressed as follows:

[0057]

[0058] Where, P p represents the AC power flow of node p, num represents the total number of nodes, v p is the voltage amplitude at node p, v q is the voltage amplitude at node q, θ pq is the voltage phase angle difference between nodes p and q.

[0059] The power flow of the distribution network has two characteristics:

[0060] 1) The node voltage amplitude is close to 1p.u.;

[0061] 2) The phase angle difference between the two ends is very small.

[0062] Therefore, by decoupling the voltage amplitude and phase angle of the AC power flow equation, we can obtain:

[0063]

[0064] Therefore, the AC power flow equation in branch power form can be linearized as:

[0065]

[0066] Where, P pq Represents the AC power flow from node p to node q.

[0067] According to the voltage sensitivity formula:

[0068]

[0069] Among them, v p,0 is the voltage of node p in steady state, S pf is the element in the pth row and fth column of the active-voltage sensitivity matrix S, ΔP f is the change in active power output at node f. Therefore, g pq (v p -v q ) can be expressed as:

[0070]

[0071] According to the power transmission allocation matrix, -b pq θ pq It can be expressed as:

[0072]

[0073]

[0074] Where, P f is the total output power of node f, P f,0 is the initial power of node f, ρ lf is the element in the power transfer allocation matrix ρ, ρ=[ρ lf ]∈R m×num , m is the total number of lines. The power transmission allocation matrix ρ can be expressed as:

[0075]

[0076] Where B line is the line-node admittance matrix, B x It is the node admittance matrix calculated based on Kirchhoff's law. The specific calculation formula is as follows:

[0077]

[0078]

[0079] where X i,j represents the line reactance from node i to j.

[0080] The active power flow formula of the distribution network after linearization is as follows:

[0081]

[0082] In order to make the expression of optimization function constraints more concise, the active power flow formula of line l after linearization of distribution network is l and the active power output change ΔP at node f f The relationship can be written in the form of a mapping,

[0083]

[0084] in is ΔP f Since the present invention only considers nodes with virtual power plant resources in the distribution network, the loads of other nodes are known constants, so the above mapping formula can be simplified to:

[0085]

[0086] Where n is the number of adjustable resources of the virtual power plant in the distribution network, Indicates the power adjustment value of the i-th adjustable resource.

[0087] S2: Set constraints including power balance constraints of each adjustable unit, node voltage constraints, line capacity constraints, upper and lower limits of adjustable unit power regulation, and ramp rate constraints, and build a GCD optimization model with the dual optimization goals of maximizing frequency regulation performance and minimizing carbon emissions.

[0088] The GCD optimization model is expressed as the linear weighted sum of the frequency modulation contribution function and the carbon emission function of the adjustable unit, as follows:

[0089]

[0090] Where, represents the frequency modulation contribution function of the i-th adjustable resource, represents the carbon emission function of the i-th adjustable resource, ϵ is the linear weighted weight coefficient;

[0091]

[0092]

[0093]

[0094] Where ΔT is the implementation period of each control interval, t represents the current time point, t=ΔT, 2*ΔT,…, N*ΔT, N is the total number of control periods; a Co2,i 、b Co2,i 、c Co2,i are the three carbon emission coefficients of each adjustable unit, represents the actual power instruction of the i-th adjustable unit at time t, represents the initial power of the i-th adjustable unit at time t; is the power adjustment value of the i-th adjustable unit at time t; σ i Represents the performance deficiency indicator of the i-th adjustable unit.

[0095] The constraints of the GCD optimization model specifically include:

[0096] Power command balance constraint, in response to the total command P issued by the dispatch center PRC At (t), the balance between the total frequency regulation mileage and frequency regulation demand is met:

[0097]

[0098] Node voltage constraints. To ensure the safety of node voltages within the virtual power plant during power regulation, important node voltage constraints are set as follows:

[0099]

[0100] Where, v j is the node voltage of the important node j. In the present invention, the nodes to which the virtual power plant adjustable unit is connected are regarded as important nodes. and are the upper and lower limits of the voltage safety range of important node j respectively.

[0101] Line capacity constraints are used to prevent line capacity from being exceeded during the GCD process, ensuring safe and stable operation of the power grid:

[0102]

[0103]

[0104]

[0105] Where, and Indicates the upper and lower limits of the power of line l, L max and L min Represents the set of upper and lower limits of line capacity;

[0106] The power adjustment lower limit constraint is used to ensure that the power adjustment of each adjustable unit does not exceed the power adjustment limit.

[0107]

[0108] Where, and They represent the upper and lower limits of the power adjustment of the i-th adjustable unit respectively.

[0109] The ramp rate constraint specifically states that the power regulation of each adjustable unit should not exceed the regulation rate limit:

[0110]

[0111] Where, is the power adjustment output of the i-th adjustable unit in the previous control period t-ΔT, V i is the ramp rate of the i-th adjustable unit in the virtual power plant.

[0112] S3: A distributed algorithm based on Nesterov momentum acceleration is used to solve the GCD optimization model in each control cycle to achieve fast and accurate solution of instructions. Communication conditions are set to reduce the communication pressure during distributed information exchange and improve the convergence speed of distributed GCD iteration.

[0113] Solving the GCD optimization model in each control cycle using an improved distributed algorithm based on Nesterov momentum acceleration includes the following sub-steps:

[0114] S3.1: Decouple the Lagrange multiplier in the GCD optimization model into n subproblems, so that each adjustable unit can locally solve the Lagrange multiplier corresponding to it.

[0115] The objective function of the GCD model, which includes power command constraints, node voltage constraints, line capacity constraints, upper and lower power adjustment constraints of each adjustable unit, and ramp rate constraints, is rewritten in the form of a Lagrangian dual function, as follows:

[0116]

[0117] where λ is the Lagrange multiplier for the equality constraint, and is the Lagrange multiplier of the voltage constraint of important node j, and is the Lagrange multiplier of the capacity constraint of line l.

[0118] Decouple the coupled Lagrangian functions into n subproblems. Each adjustable unit can locally solve its corresponding Lagrangian multiplier, as shown below:

[0119]

[0120]

[0121]

[0122]

[0123] Where λ i is the Lagrange multiplier of the equational constraint in the decoupled adjustable unit i, and They represent the Lagrange multipliers of the voltage constraints of the important nodes j in the adjustable unit i after decoupling, and are the Lagrange multipliers of the line capacity constraint in the decoupled adjustable unit i, OF i represents the objective function of the i-th adjustable unit.

[0124] S3.2: Initialize the Lagrange multiplier, iteration step size, and number of iterations, and introduce the performance deficiency index of each adjustable unit and the power adjustment value at the previous moment;

[0125] S3.3: Update the Lagrange multiplier using a distributed algorithm based on Nesterov momentum acceleration, specifically including: calculating the actual power command of the i-th adjustable unit at time t corresponding to iteration round k+1 based on the Lagrange multiplier update result Intermediate variables ; Combined with the power regulation upper and lower limit constraints and climbing rate constraints of each adjustable unit, Correction is performed to obtain the optimal actual power instruction of the i-th adjustable unit at time t corresponding to iteration round k+1.

[0126] Due to λ i , , , and Updates are independent of each other and do not affect each other, so λ i Taking the update of as an example, the iterative process of the distributed algorithm is as follows:

[0127] Step 1: Use the Nesterov momentum term Instead of the Lagrange multiplier, the Nesterov momentum term is updated with a gradient, and the weighted sum v of the Nesterov momentum gradient is calculated using the exponentially weighted moving average method. i , as follows:

[0128]

[0129] Where k is the number of iterations of the distributed algorithm, v i is the Nesterov momentum term, β is the Nesterov momentum delay coefficient, is the Nesterov momentum coefficient, is the local incremental cost obtained by Nesterov accelerated gradient descent, is the iteration step size. .

[0130] Step 2: To make the local incremental cost During the iteration process, it can be stabilized at the optimal point, and a correction step is introduced to eliminate and v i The differences:

[0131]

[0132] Where, is the local incremental cost taking into account bias correction.

[0133] Step 3, information diffusion step, combines the correction information of neighbor nodes to achieve Global consistency:

[0134]

[0135] Where, Φ W =[ω i,j ]∈R n×n represents the line weight set, if , then ω i,j >0, it means that the i-th adjustable unit and the j-th adjustable unit are neighbors; if , then ω i,j =0,Φ E Represents a set of communication lines; the specific weights are defined as follows:

[0136]

[0137] Among them, n i 、n j denote the number of neighboring nodes of adjustable unit i and adjustable unit j respectively; and Update and λ i Updates are similar and can be performed in parallel.

[0138] Calculate the optimal power adjustment instruction of the i-th adjustable unit at time t corresponding to the iteration round k+1 according to the Lagrange multiplier update result Intermediate variables ; Based on intermediate variables Combined with the upper and lower limits of the power adjustment constraints of each distributed resource, the optimal actual power instruction of the i-th adjustable unit at time t corresponding to the iteration round k+1 is obtained as follows:

[0139]

[0140]

[0141] S3.4: Repeat the iteration according to the updated result of the Lagrange multiplier until the iteration termination condition is reached, and obtain the optimal actual power command of the i-th adjustable unit at time t.

[0142] The communication conditions are set during the iteration process to reduce the communication pressure during the distributed information exchange and improve the convergence speed of the distributed GCD process, as follows:

[0143] During the communication process of the adjustable unit, if the change of the Lagrange multiplier between two adjacent iterations is less than the set termination condition, the information exchange will be stopped.

[0144]

[0145] Among them, ε is the termination condition set by the system. When the above conditions are met, the neighbor node will use the last received to update local information.

[0146] The effects of the present invention will be further described below with reference to a specific embodiment.

[0147] In this embodiment, a virtual power plant consisting of five distributed frequency regulation units (1-5, 1: photovoltaic, 2: micro gas turbine, 3: micro gas turbine, 4: energy storage, 5: small hydropower) verifies the effectiveness of this method. The weight coefficient This article sets it to 0.9, PRC =10MW, the power residuals of the improved distributed algorithm based on Nesterov momentum acceleration, the precise diffusion algorithm, the standard diffusion algorithm, and the consensus algorithm are compared. Figure 2 The calculation formula of its power residual is as follows:

[0148]

[0149] in, is the set of frequency-adjusted mileage commands for each adjustable unit in the kth iteration, The optimal value set obtained by the solver. The power residual curve is used to compare the convergence speed of the four consistency algorithms. Compared with the remaining three methods, the method of the present invention has a faster convergence speed and higher convergence accuracy.

[0150] The model was built on the MATLAB / Simulink platform. The virtual power plant system consisting of 5 distributed frequency regulation units (1-5, 1: photovoltaic, 2: micro gas turbine, 3: micro gas turbine, 4: energy storage, 5: small hydropower) was verified to have a P of 200s. PRC =10MW virtual power plant response performance, the total input time in the system is 600s, Figure 3 This is the instruction distribution of the five adjustable units in the GCD real-time control process. It can be seen that the frequency regulation performance and carbon emissions of different adjustable units are different, which will affect the power output. The frequency regulation response speed of the gas turbine is slow and the carbon emissions are large, so the frequency regulation power is low; the carbon emissions of distributed resources dominated by renewable energy such as photovoltaics are low. Its carbon emission coefficient quantifies the carbon emissions of the entire operation cycle (including transportation, installation, operation and maintenance), which is generally low, so it bears more power adjustment. In addition, there are the influences of node voltage constraints and line capacity constraints. The power adjustment values ​​of the adjustable units are all within the constraints. In addition, four methods (Method 1: using a centralized solver, regarded as a reference value for other algorithms; Method 2: fixed ratio allocation method; Method 3: consensus algorithm; the present invention) are compared. Figure 4 The virtual power plant response P calculated by four methods is PRCThe command response curve shows that the fixed-ratio allocation method has poor tracking response capabilities due to the lack of dynamic optimization, resulting in poor frequency modulation performance. The consensus algorithm has low accuracy, slow convergence, and no way to calculate the optimal frequency modulation solution. In comparison, the present invention has the best frequency modulation performance, demonstrating its improvement in frequency modulation performance. Figure 5 After comparing the carbon emissions of the four methods, it was found that the method proposed in the present invention also has the smallest cumulative carbon emissions.

[0151] Those skilled in the art will understand that the foregoing descriptions are merely preferred embodiments of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art will still be able to modify the technical solutions described in the foregoing examples or substitute equivalents for some of the technical features therein. Any modifications, equivalent substitutions, etc. made within the spirit and principles of the invention shall be included within the scope of protection of the invention.

Claims

1. A method for allocating distributed power generation instructions for a virtual power plant in a distribution network, characterized in that: The method comprises the following steps: Step 1: Linearize the AC power flow model of the distribution network, model the line active power and node voltage in the distribution network, and express the node voltage and line active power as a mapping of node power changes; Step 2: Set constraints, including power balance constraints for each adjustable unit, node voltage constraints, line capacity constraints, upper and lower limits for adjustable unit power regulation, and ramp rate constraints, to build a generation command scheduling optimization model with the dual optimization objectives of maximizing frequency regulation performance and minimizing carbon emissions. Step 3: Using a distributed algorithm based on Nesterov momentum acceleration, the power generation instruction scheduling optimization model is solved in each control cycle to achieve fast and accurate solution of instructions, and communication conditions are set to reduce the communication pressure during distributed information exchange, thereby improving the convergence speed of the power generation instruction scheduling optimization model; The step three specifically includes the following sub-steps: S3.1: Decouple the Lagrange multiplier in the power generation command scheduling optimization model into n subproblems, so that each adjustable unit can locally solve the Lagrange multiplier corresponding to itself; S3.2: Initialize the Lagrange multiplier, iteration step size, and number of iterations, and introduce the performance deficiency index of each adjustable unit and the power adjustment value at the previous moment; S3.3: Update the Lagrange multiplier using a distributed algorithm based on Nesterov momentum acceleration, specifically including: calculating the actual power instruction of the i-th adjustable unit at time t corresponding to iteration round k+1 according to the Lagrange multiplier update result Intermediate variables ; Combined with the power regulation upper and lower limit constraints and climbing rate constraints of each adjustable unit, Correction is performed to obtain the optimal actual power instruction of the i-th adjustable unit at time t corresponding to iteration round k+1; The S3.3 is implemented by the following sub-steps: S3.3.1: Use the Nesterov momentum term instead of the Lagrange multiplier and perform gradient updates on the Nesterov momentum term. Use the exponentially weighted moving average method to calculate the weighted sum v of the Nesterov momentum gradient. i ; S3.3.2: Introduce a correction step to eliminate the local incremental cost ψ obtained by Nesterov accelerated gradient descent i and the weighted sum v of the Nesterov momentum gradient i The difference between i Ability to stabilize at the optimal point during iterative engineering; S3.3.3: Combine the correction information of neighboring nodes to achieve global consistency; S3.4: Repeat the iteration according to the updated result of the Lagrange multiplier until the iteration termination condition is reached, and obtain the optimal actual power instruction of the i-th adjustable unit at time t.

2. The method for allocating distributed power generation instructions of a virtual power plant for a distribution network according to claim 1, characterized in that: In step 3, communication conditions are set to reduce the communication pressure during distributed information exchange, specifically: During the communication process of the adjustable unit, if the change in the Lagrange multiplier between two adjacent iterations is less than the set termination condition, the information exchange will be stopped.

3. The method for allocating distributed power generation instructions of a virtual power plant for a distribution network according to claim 1, characterized in that: In the step 2, The power balance constraint is to meet the balance between the total frequency regulation mileage and the frequency regulation demand when responding to the total instruction issued by the dispatching center; The node voltage constraint is that the node voltage of the important node is within the upper and lower limits of the voltage safety range; the important node is a node to which the virtual power plant adjustable unit is connected; The line capacity constraint is to ensure that the power of all lines is within the upper and lower limits of the power of the current line during the power generation instruction scheduling process; The upper and lower limits of the power adjustment of the adjustable units are constrained to ensure that the adjustment power of each adjustable unit is within the upper and lower limits of the power adjustment; The ramp rate constraint is that the power regulation rate of each adjustable unit is within an upper and lower limit range of the power regulation rate of the adjustable unit.

4. The method for allocating distributed power generation instructions of a virtual power plant for a distribution network according to claim 1, characterized in that: The power generation command scheduling optimization model is the linear weighted sum of the frequency regulation contribution function and the carbon emission function of all adjustable units; The frequency modulation contribution function of the adjustable unit is specifically the square of the difference between the power adjustment value of the adjustable unit at the current moment and the power adjustment value at the previous moment, multiplied by the performance deficiency index σ of the corresponding adjustable unit. i ; The carbon emission function of the adjustable unit is specifically a quadratic function of the actual power instruction of the adjustable unit at the current moment, and the coefficient of the quadratic function is the carbon emission coefficient.

5. The method for allocating distributed power generation instructions of a virtual power plant for distribution network according to claim 1, characterized in that: In the step 1, Linearizing the AC power flow model of the distribution network, specifically: decoupling the voltage amplitude and phase angle of the AC power flow model of the distribution network; The node voltage is expressed as a mapping of the node power change. Specifically, the active-voltage sensitivity matrix is ​​introduced to express the node voltage as the relationship between the active power changes of each node; The line active power is expressed as a mapping of node power changes. Specifically, based on the linearized AC power flow model of the distribution network, an active-voltage sensitivity matrix is ​​introduced to express the node voltage as the relationship between the active power changes of each node; and a power transmission allocation matrix is ​​introduced to express the node phase angle difference as the relationship between the total output power of each node.

6. A virtual power plant distributed power generation instruction distribution system for distribution network, characterized in that: The method comprises one or more processors for implementing the method for allocating distributed power generation instructions of a virtual power plant for a distribution network as described in any one of claims 1 to 5.

7. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, enables the electronic device to implement the method for allocating distributed power generation instructions of a virtual power plant for a distribution network as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that A program is stored thereon, and when the program is executed by a processor, the method for allocating distributed power generation instructions of a virtual power plant for a distribution network as described in any one of claims 1 to 5 is implemented.

Images