VPP real-time scheduling method based on cloud edge collaborative architecture and ADMM consistency optimization

Abstract

The invention belongs to the technical field of power system optimization scheduling, and particularly discloses a VPP real-time scheduling method based on cloud edge collaborative architecture and ADMM consistency optimization, a cloud operator uses a particle swarm algorithm to carry out global cost optimal calculation on real-time power deviation, and after an edge side transformer area group receives a scheduling instruction through a main node, the scheduling instruction is sent to the main node; the ADMM algorithm is adopted to realize the consistency iterative optimization of the cost increment rate of each transformer area, and dynamic coupling is carried out through default penalty constraint and an electric power spot market transaction strategy. According to the VPP real-time scheduling method based on the cloud edge collaborative architecture and the ADMM consistency optimization, the problem of real-time power mismatch caused by distributed resource prediction deviation and load abrupt change is effectively solved through a double-level collaborative mechanism, so that a virtual power plant meets the day-ahead contract response requirement, and the real-time scheduling efficiency of the virtual power plant is improved. Multiple deviation correction modes such as energy storage charging and discharging, resource regulation and control or market transaction can be flexibly selected according to real-time electricity price fluctuation, and finally the real-time scheduling target with the optimal operation cost is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system optimization and scheduling, and in particular to a VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization. Background Art

[0002] With the large-scale access of renewable energy and the widespread participation of demand-side resources, virtual power plants (VPPs), as the core technology for aggregating distributed resources, need to address the power deviation problem caused by prediction errors and load fluctuations in real-time scheduling. Traditional VPP scheduling methods mostly use centralized optimization, which has significant shortcomings in handling the real-time response of distributed resources (such as wind power, photovoltaics, energy storage, etc.): On the one hand, centralized scheduling relies on global computing in the cloud, which is difficult to meet the rapid requirements of real-time control; on the other hand, the randomness of distributed resources and sudden load increases can easily lead to large deviations between actual power and day-ahead plans, resulting in high penalty costs for breach of contract and even affecting grid stability. In addition, existing methods lack an efficient distributed optimization mechanism when coordinating resources in multiple substations, and cannot achieve economically optimal allocation with consistent cost increment rates.

[0003] In existing technologies, VPPs mainly correct deviations in real time through electricity spot market transactions or local resource regulation, but fail to effectively integrate cloud-edge collaborative architecture and distributed optimization algorithms, leading to the following problems: 1) The global optimization on the cloud side is disconnected from the local response on the edge side, making it difficult to adapt to sudden power fluctuations; 2) There is a lack of coordination mechanism between stations, and resource scheduling relies on cloud communication, which poses a risk of delay; 3) Traditional algorithms (such as linear programming) have difficulty handling non-convex optimization problems of large-scale distributed resources and cannot ensure that the slight increase rate of cost at each node is consistent.

[0004] To address the above issues, a real-time scheduling method that integrates a cloud-edge collaborative architecture with a distributed optimization algorithm is urgently needed. Cloud computing can provide global optimization capabilities, while edge computing can improve local response speed and real-time data processing efficiency. At the same time, the alternating direction multiplier method (ADMM), as a distributed optimization algorithm, can decompose complex problems into multiple subproblems for parallel solution and achieve global consistency by coordinating shared variables. It is suitable for the collaborative optimization of VPP multi-station resources. However, existing research has not yet combined the cloud-edge collaborative architecture with the ADMM algorithm to achieve the dual goals of global cost optimization and rapid edge response in VPP real-time scheduling, nor has it fully considered the dynamic game relationship between electricity spot market transactions and default penalties.

[0005] Therefore, the present invention proposes a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization, aiming to solve the problems of insufficient real-time performance, low resource coordination efficiency, and high operating costs of traditional centralized scheduling, and provide VPP with an efficient, economical, and reliable real-time control solution. Summary of the Invention

[0006] The purpose of the present invention is to provide a VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization. It aims to achieve consistency optimization of the slight increase rate of cost of substation groups by using the alternating direction multiplier method through a combination of cloud-side global optimization and edge-side distributed computing, continuously correct real-time power deviations and achieve optimal operating costs of virtual power plants. It is particularly suitable for dealing with scenarios where the day-ahead scheduling plan of a virtual power plant (VPP) deviates from the actual operating power.

[0007] To achieve the above objectives, the present invention provides a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization, comprising the following steps:

[0008] S1. Input the training data of the VPP particle swarm optimization algorithm, including the initial real-time market transaction power, regulation power, and default power;

[0009] S2. Set the initial state of the VPP particle swarm optimization algorithm, including the initial position of the particles, the initial velocity of the particles, and the number of iterations u;

[0010] S3, particle swarm cross mutation, obtain the area control power after the uth iteration Default Power and market trading power

[0011] S4. Combine the day-ahead deviation with the minimum VPP real-time operating cost as the optimal goal, make decisions on the control power, default power and real-time market transaction rate, and calculate the real-time deviation correction cost C. VPP , and the constraints are power balance constraints, substation group control power constraints, market transaction constraints and breach of contract constraints;

[0012] S5. Confirm the individual and global optimal solutions in the iterative process, denoted as and If C' is not satisfied VPP >C VPP , and As a new iteration value to participate in the next cycle calculation, if C' VPP >C VPP , and Participate in the next cycle calculation as a new iteration value;

[0013] S6. Let u=u+1, if u≥u max , then exit the loop and output the result of the optimal target, otherwise return to S3 to continue the loop operation;

[0014] S7, when u≥umax When , the loop is exited and the result of the optimal target is output. The substation-master receives the optimized total control power of the substation group from the cloud, and the substation-master exchanges information with adjacent substations based on the total control power.

[0015] S8. Use the ADMM consensus algorithm to make each substation achieve a unified target incremental rate λ, and model it based on the optimal target and constraints of local control costs;

[0016] S9. Introducing the Lagrange multiplier λ j,t Construct augmented Lagrangian function;

[0017] S10, iteratively updating the substation control power variables, shared variables and Lagrange multipliers respectively;

[0018] S11. Determine whether the convergence conditions are met. If so, output the actual regulated power of each substation. If not, return to step S9, continue updating the augmented Lagrangian function, and iteratively update the substation regulated power variables, shared variables, and Lagrangian multipliers again until convergence.

[0019] Preferably, in S1, the VPP particle swarm optimization algorithm is specifically:

[0020] The cloud sends the control power of each substation group to the substation-master of the edge server. The substation-master completes the distributed optimization calculation of the control power of each substation in the area and performs secondary optimization allocation. The expression is as follows:

[0021]

[0022] Where, M is the number of stations in the ith station group; C j.t is the control cost of the jth station at time t, P j.t is the regulated power of the j-th substation at time t.

[0023] Preferably, S4 is specifically:

[0024] Taking the day-ahead deviation into consideration and minimizing the VPP real-time operating cost as the optimal goal, decisions are made on the regulated power, default power, and real-time market transaction rate. The objective function is as follows:

[0025] minC VPP =C DK +C PC -C MA (2);

[0026] Where C VPP is the operating cost of VPP in the real-time phase, C DK is the total cost of VPP resource control, C PCis the cost of liquidated damages, C MA It is the real-time profit from transactions between VPP and the electricity spot market;

[0027] Calculate the real-time deviation correction cost C' according to formula (2) VPP , and with power balance constraints, substation group control power constraints, market transaction constraints and default constraints as constraints, the expression is as follows:

[0028] a. Power balance constraints are:

[0029] ΔP VPP.t =P am.t -P VPP.t (3);

[0030] Where, P am.t is the planned output of VPP predicted by the day before in period t, P VPP.t is the actual output during the actual operation phase, ΔP VPP.t is the difference between the two;

[0031]

[0032] Where ΔP VPP1.t P is the transaction power deviation of VPP participating in the real-time electricity spot market. D1.i.t is the resource regulation power actually used by the grid group i to participate in the electricity market transaction during period t, ΔP VPP2.t is the power deviation to be corrected for VPP, P D2.i.t The resource control power of the station group i in the real-time deviation correction in time period t;

[0033] The total control power of the station group i is the sum of the control power participating in market transactions and real-time deviation correction. The specific formula is as follows:

[0034]

[0035] b. The power constraint of the substation group control is:

[0036] 0≤P D.i.t ≤P Dmax.i.t (6);

[0037] Where, P Dmax.i.t is the maximum control power of the station group i during period t;

[0038] c. Market transaction constraints:

[0039]

[0040] Where, and It is the maximum power that the VPP can purchase and sell in the real-time electricity spot market;

[0041] d. The default constraints are:

[0042] εP am.t ≤P A.t ≤P am.t (8);

[0043] Where ε is the default coefficient.

[0044] Preferably, S8 is specifically:

[0045] Each substation in the VPP substation group is considered as a node. The ADMM consensus algorithm is used to make each substation reach a unified cost increment rate λ. By adjusting the control variables, the cost increment rates of each substation are made consistent, that is:

[0046]

[0047] By introducing a shared variable z to express the consistency of the cost increment rate of each area, that is, each area has an expression for this constraint, as follows:

[0048]

[0049] Based on the optimal target and constraints of local control cost, model:

[0050]

[0051] Preferably, S9 is specifically:

[0052] Augmented Lagrangian function:

[0053]

[0054] Preferably, in S10, first update the control power variable P of each station area j,t , we need to fix the shared variable z and the Lagrange multiplier λ j,t , its specific expression is as follows:

[0055]

[0056] Where, is the control power variable of station j after k+1 iterations, z k is the shared variable after k iterations, is the Lagrange multiplier variable after the last iteration, and are the Lagrange multiplier variables for the upper and lower power limits after the last iteration.

[0057] Preferably, in S10, after updating the control power variable P of each substationj,t Then, the current cost increment rate is calculated according to formula (1) and formula (9), and the new shared variable z is obtained. k+1 , that is, z after k+1 iterations k+1 , its specific expression is as follows:

[0058]

[0059] Preferably, in S10, the Lagrange multiplier λ of each station is updated j,t , its specific expression is as follows:

[0060]

[0061] Then update the Lagrange multipliers of the upper and lower power limits, and their specific expressions are as follows:

[0062]

[0063] Preferably, in S10, the judgment condition for termination of iteration is set:

[0064]

[0065] Where, if after k+1 iterations, the change in power scheduling is very small and the change in cost increment rate is also very small, then the iteration ends and the result converges, and θ is the corresponding convergence residual.

[0066] Therefore, the present invention adopts the above-mentioned VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization, and the beneficial effects are as follows:

[0067] (1) Due to the participation in the day-ahead demand-side response market, when it comes to the execution time and the real-time response process, the inaccurate prediction of distributed resources and the uncontrollability of load will inevitably have a great impact on the actual response results, and even the situation of severe shortage of photovoltaic output or sudden increase in load due to sudden weather changes. At this time, it is necessary to quickly dispatch the distributed resources under the substation site to deal with such emergencies. The present invention constructs a cloud-edge collaborative architecture that combines cloud-side global optimization with edge-side distributed computing, and adopts a substation resource optimization strategy based on the ADMM consistency algorithm, so that the virtual power plant can take into account both global economy and edge response agility in real-time scheduling.

[0068] (2) The cloud operator of the present invention uses the particle swarm algorithm to perform global cost optimization calculations on real-time power deviations. After the edge-side substation group receives the dispatch instruction through the master node, the ADMM algorithm is used to achieve iterative optimization of the consistent cost increment rate of each substation. This two-level collaborative mechanism effectively solves the real-time power mismatch problem caused by the distributed resource prediction deviation and load mutation. Through the dynamic coupling of the breach penalty constraint and the power spot market transaction strategy, the virtual power plant can flexibly select multiple deviation correction methods such as energy storage charging and discharging, resource regulation or market transaction according to the real-time electricity price fluctuations while meeting the day-ahead contract response requirements, and ultimately achieve the real-time dispatch target with the optimal operating cost. It is particularly suitable for dealing with scenarios where the day-ahead dispatch plan of a virtual power plant (VPP) deviates from the actual operating power.

[0069] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a "master-slave" distributed substation group structure diagram of an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention;

[0071] Figure 2 This is a flowchart of an embodiment of a VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization of the present invention;

[0072] Figure 3 This is a schematic diagram of a substation group according to an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention;

[0073] Figure 4 This is a diagram of the real-time response execution status of each substation participating in the day-ahead plan in scenario 1 of an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention, where (a) is the day-ahead and intra-day real-time response power deviation in scenario 1, (b) is the optimized scheduling strategy of substation 1 in scenario 1, (c) is the optimized scheduling strategy of substation 2 in scenario 1, and (d) is the optimized scheduling strategy of substation 3 in scenario 1;

[0074] Figure 5 This is a diagram showing the real-time response execution of each substation participating in the day-ahead plan under scenario 2 of an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention, wherein (a) is the day-ahead and intra-day real-time response power deviation under scenario 2, (b) the optimized scheduling strategy of substation 1 under scenario 2 is, (c) is the optimized scheduling strategy of substation 2 under scenario 2, and (d) is the optimized scheduling strategy of substation 3 under scenario 2;

[0075] Figure 6 This is a diagram of the real-time response execution status of each substation participating in the day-ahead plan under scenario three of an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention, wherein (a) is the day-ahead and intra-day real-time response power deviation under scenario three, (b) is the optimized scheduling strategy of substation 1 under scenario three, (c) is the optimized scheduling strategy of substation 2 under scenario three, and (d) is the optimized scheduling strategy of substation 3 under scenario three;

[0076] Figure 7 This is a diagram of the real-time response execution status of each substation participating in the day-ahead plan under scenario 4 of an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention, wherein (a) is the day-ahead and intra-day real-time response power deviation under scenario 4, (b) is the optimized scheduling strategy of substation 1 under scenario 4, (c) is the optimized scheduling strategy of substation 2 under scenario 4, and (d) is the optimized scheduling strategy of substation 3 under scenario 4;

[0077] Figure 8 This is a cost iteration diagram for each substation in scenario 4 of an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention;

[0078] Figure 9 This is a diagram of the residual iteration of the ADMM algorithm in scenario four of an embodiment of a VPP real-time scheduling method based on a cloud-edge collaborative architecture and ADMM consistency optimization of the present invention. DETAILED DESCRIPTION

[0079] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0080] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0081] In this invention, the cloud layer consists of the VPP operator's central server, which is responsible for receiving edge-side data, performing global optimization and overall management, and distributing control tasks to each edge node. The edge layer is composed of VPP site substations, equipped with intelligent distribution transformers (i.e., converged terminals), which are responsible for integrating, monitoring, and sensing substation resources. Compared with data transmission between the cloud and the edge, direct interaction between edge nodes is more efficient and real-time.

[0082] Example 1

[0083] A VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization, such as Figure 1As shown in the figure, based on the cloud-edge collaboration, a distributed control architecture on the edge side is further designed, and multiple substations with adjacent communication capabilities are combined into substation groups. Each substation is equipped with an independent local controller, which can exchange information with adjacent substations to form an edge node network. Within each substation group, a "master-slave" distributed control strategy is adopted, and edge computing technology is used to optimize the resource scheduling of the VPP real-time stage. In the real-time operation stage, the cloud (the central server of the VPP operator) takes the minimum cost of operating deviation correction as the global optimization goal. After completing the control calculation, the cloud will send the control power of each substation group to the substation "master" of the edge server. The substation "master" will complete the distributed optimization calculation of the control power of each substation in its own area, and perform secondary optimization allocation. The expression is as follows:

[0084]

[0085] In formula (1), M is the number of stations in the ith station group; C j.t is the control cost of the jth station at time t, P j.t is the regulated power of the j-th substation at time t.

[0086] The ADMM algorithm is a distributed optimization algorithm particularly well-suited for optimization problems with local constraints and shared variables. Its core concept is to decompose the original optimization problem into multiple local subproblems and then optimize them alternately, ensuring that their shared variables gradually converge to a consistent solution. The basic idea of ​​the ADMM algorithm is to use the augmented Lagrange multiplier method combined with a splitting technique to gradually optimize each subproblem, while maintaining consistency through shared variables and constraints. Shared variables are then coordinated, allowing each local subproblem to gradually approach the global optimal solution.

[0087] For local substation groups, an optimization algorithm that combines the particle swarm algorithm (PSO) and the ADMM algorithm can be used to achieve the goal of minimizing the control cost of the local substation group under the condition of a consistent overall cost increase rate. In this method, the particle swarm algorithm is used to guide the global search for solutions, so that the overall cost of controlling all substations to handle day-ahead power deviations is optimal, while the ADMM algorithm is used to handle local optimization. Each time the ADMM algorithm optimizes a subproblem, the particle swarm algorithm can search for a suitable spatial region for the solution and guide the ADMM algorithm to optimize the update direction. The particle swarm algorithm performs global optimization, and the ADMM algorithm is used as a local optimization algorithm for the substation group to solve the VPP real-time optimization scheduling model, such as Figure 2 The specific steps are as follows:

[0088] 1) Input the training data of the VPP particle swarm optimization algorithm, including the initial real-time market trading power, regulation power, and default power;

[0089] 2) Set the initial state of the VPP particle swarm optimization algorithm, including the initial position of the particles (the solution in the search space), and set the initial velocity of the particles. The number of iterations u is 0 at this time.

[0090] 3) Particle swarm cross-mutation can obtain the control power of the substation after the uth iteration Default Power and market trading power

[0091] 4) Taking into account day-ahead deviations and minimizing VPP real-time operating costs as the optimal goal, decisions are made on resource control power, default power, and real-time market transaction power. The specific expressions are as follows:

[0092] minC VPP =C DK +C PC -C MA (2)

[0093] Where C VPP is the operating cost of VPP in the real-time phase; C DK Regulate the total cost of VPP resources; C PC C is the cost of compensation for breach of contract; MA It is the real-time profit from transactions between VPP and the electricity spot market.

[0094] Calculate the real-time deviation correction cost C' using formula (2) VPP , and the constraints are power balance constraints, substation group control power constraints, market transaction constraints, and default constraints. The specific expression is as follows:

[0095] a. Power balance constraints are:

[0096] ΔP VPP.t =P am.t -P VPP.t (3)

[0097] In formula (3), P am.t P is the planned output predicted by VPP in period t, which is also the contracted power. VPP.t is the actual output during the actual operation phase, ΔP VPP.t The difference between the two.

[0098]

[0099] In formula (4), ΔP VPP1.t P is the transaction power deviation of VPP participating in the real-time electricity spot market. D1.i.tΔP is the resource control power actually used by the station group i to participate in the power market transaction during period t. The market transaction power deviation during period t is the difference between the real-time transaction power and the resource control power participating in the power market. VPP2.t is the power deviation to be corrected for VPP, P D2.i.t is the resource control power of the substation group i in time period t after real-time deviation correction. The power deviation to be corrected is the sum of the resource control power after real-time deviation correction and the default power.

[0100] The total deviation of the day-ahead power forecast in real time is the sum of the deviation of the VPP participating in market transactions and the power deviation to be corrected. The total regulated power of substation group i is the sum of the regulated power participating in market transactions and the regulated power of real-time deviation correction. The specific formula is as follows:

[0101]

[0102] b. The power constraint of the substation group control is:

[0103] 0≤P D.i.t ≤P Dmax.i.t (6)

[0104] In formula (6), P Dmax.i.t is the maximum control power of the station group i during period t.

[0105] c. Market transaction constraints:

[0106]

[0107] In formula (7), and It is the maximum power that a VPP can purchase or sell electricity in the real-time electricity spot market.

[0108] d. The default constraints are:

[0109] εP am.t ≤P A.t ≤P am.t (8)

[0110] In formula (8), ε is the default coefficient, which ranges from 0.6 to 0.9.

[0111] 5) Confirm the individual and global optimal solutions in the iterative process, denoted as and According to formula (1), the current optimal real-time deviation correction cost is calculated and recorded as C VPP ; and perform selection operation, if C′ is not satisfied VPP >C VPP , then the particle swarm updated in the uth iteration is and This reduces the cost of real-time deviation correction. and Participate in the new loop calculation as the value of the next iteration; if C' VPP >C VPP , then the updated and The goal of reducing real-time operating costs has not been achieved. and Participate in the next cycle calculation as a new iteration value;

[0112] 6) Let u=u+1, if u≥u max , then exit the loop and output the optimal result, otherwise return to step 3) to continue the loop operation;

[0113] 7) When u≥u max When , the loop is exited and the optimal result is output. Then the master substation will receive the optimized total control power of the substation group from the cloud, and the master substation group will exchange information with the adjacent follower substations according to the total control power.

[0114] 8) The ADMM algorithm is used to solve the distributed optimization model for the site cluster under the VPP. Each station in the VPP station cluster is regarded as a node. The ADMM consistency algorithm is used to make each station achieve a unified cost increment rate λ. By adjusting these control variables, the cost increment rate of each station is consistent, that is:

[0115]

[0116] By introducing a shared variable z, that is, each area has this constraint expression, the consistency of the cost increment rate of each area is expressed as follows:

[0117]

[0118] According to the above optimal objectives and constraints of local control costs, the following model can be built:

[0119]

[0120] 9) Introducing the Lagrange multiplier λ j,t Constructing the augmented Lagrangian function can be expressed as:

[0121]

[0122] 10) Iteratively update the area control power variables, shared variables and Lagrange multipliers respectively, specifically:

[0123] 10-1. Update the control power variables of each substation:

[0124] In the update area, the control variable P j,t When , it is necessary to fix the shared variable z and the Lagrange multiplier λ j,t , this process is equivalent to optimizing the local optimization problem, and its specific expression is as follows:

[0125]

[0126] Where, is the control power variable of station j after k+1 iterations, z k is the shared variable after iteration k, that is, the last iteration, is the Lagrange multiplier variable after the last iteration. and are the Lagrange multiplier variables for the upper and lower power limits after the last iteration.

[0127] 10-2. Update shared variables:

[0128] Next, we need to update the shared variable z so that the incremental cost rate is consistent across all substations. The shared variable z is the average incremental cost rate across all substations, and is updated using the following formula:

[0129]

[0130] After updating the control power variable P of each substation j,t After that, the current cost increment rate can be calculated according to equations (1) and (9), and the new shared variable is obtained, which is z after k+1 iterations. k+1 .

[0131] 10-3. Update the Lagrange multiplier:

[0132] Next, update the Lagrange multiplier λ of each station j,t In order to ensure the consistency of the cost increment rate of all substations, the iterative formula is as follows:

[0133]

[0134] Next, the Lagrange multipliers of the upper and lower power limits are updated to ensure that the power regulation of the substation meets the upper and lower limit constraints. The specific expression is as follows:

[0135]

[0136] For the ADMM algorithm, you can set the judgment condition for iteration termination:

[0137]

[0138] Where, if after k+1 iterations, the change in power scheduling and the change in cost increment rate are small, the iteration ends and the result converges, and θ is the corresponding convergence residual. Ultimately, by using cost increment rate consistency as the objective function and employing the ADMM consistency algorithm, the goal is to achieve a globally optimal solution by ensuring a consistent cost increment rate for each substation by alternately optimizing the substation's scheduling variables, shared variables, and Lagrange multipliers.

[0139] 11) Determine whether the convergence conditions are met. If so, output the actual regulated power of each substation. If not, continue updating the augmented Lagrangian function and iteratively update the variables in step 10) again until convergence.

[0140] Example 2

[0141] In order to verify the combined algorithm of PSO and ADMM proposed in Example 1, the PSO algorithm is used as the global cost optimization algorithm for real-time operation of VPP, and the ADMM algorithm is used as the local optimization algorithm for the substation group, so that the slight increase rate of cost of each substation tends to be consistent. Taking a VPP demonstration project of the Jibei Power Grid as an example, each substation in the VPP is divided into multiple substation groups. The VPP is divided into two substation groups, namely Substation 1 and Substation 2, and each substation has three stations on the way, namely Substation 1-1, Substation 1-2, Substation 1-3 and Substation 2-1, Substation 2-2, Substation 2-3, and each station uses Station 1 as the "master" of the substation group, that is, Substation 1-1 and Substation 2-1, and the voltage level of each substation is 10kV / 380V. As Figure 3 shown.

[0142] In this embodiment, four scenarios are considered for the global optimization goal established by the VPP operation platform, and the four scenarios are as follows.

[0143] Scenario 1: Prioritize reducing VPP resource control costs;

[0144] Scenario 2: Prioritize reducing penalty compensation costs;

[0145] Scenario 3: Prioritize increasing revenue from real-time electricity spot market transactions;

[0146] Scenario 4: Taking into account the costs and benefits in scenarios 1, 2, and 3, the goal is to minimize the total operating cost of the VPP, which is the VPP overall operating cost minimization strategy shown in formula (1).

[0147] Simulations were conducted for each of the four optimized scheduling scenarios, including the real-time response execution of each substation participating in the day-ahead plan under these four scenarios. The specific scenario strategy analysis is shown below:

[0148] First, let’s analyze scenario 1. Figure 4 As shown. In this scenario, the power deviation will be optimized to a certain extent relative to the power deviation of centralized optimization control, and the penalty cost will be reduced accordingly relative to centralized control, but there will still be a certain deviation overall, because when considering the minimum control cost, almost no controllable resources are controlled to complete the response, and only the electricity spot market is used to purchase electricity to meet the real-time demand. In the actual process, there will be a certain delay in the execution of electricity purchase and sale in the electricity market, which will lead to real-time deviation. The following is a specific strategy for optimizing the scheduling of adjustable resources in Area 1 to Area 3 in the scenario. Then, in the scenario, the real-time internal resources of Areas 1 to 3 are regulated, and the VPP operation platform sends the real-time strategy to each edge layer. When considering the minimum control cost, all wind power and photovoltaic power generation resources participate in the real-time scheduling plan, and energy storage and gas turbines are hardly scheduled. Therefore, the real-time control cost is minimized, as shown Figure 4 As shown in (b), (c), and (d), during the response process, the wind power resources in Area 3 are sufficient. When the wind power resources are sufficient at night, the excess power will be sold to the other two areas. Therefore, even if Area 1 has no wind power resources and Area 2 has less wind power resources, it is still possible to meet the day-ahead planned power requirements by purchasing a small amount of electricity from the market and more electricity from Area 3. When the load demand increases during the day, the distributed resources can no longer meet the load demand. To meet the real-time power requirements, all areas purchase electricity from the electricity spot market.

[0149] like Figure 5 As shown in the figure, the real-time response execution status of each substation participating in the day-ahead plan under scenario 2. The second scenario gives priority to reducing the cost of default. In this scenario, the real-time power within the day is generally greater than or equal to the target power, so the cost of default will be relatively small. However, due to the delay in the purchase of electricity in the electricity spot market or the real-time regulation of resources to respond, when the power demand rises sharply during the peak power consumption, the real-time operating power within the day is less than the target power at certain moments. Figure 5As shown in (a), the real-time power was lower than the target power in both the time periods of 10:45-11:30 and 18:30-20:00, but the deviation was relatively small. Therefore, the penalty cost of this scenario is generally the smallest among the four scenarios. The real-time power in the figure is very consistent with the target power. At 19:30, the real-time response power did not keep up with the changes in the target power curve, but at 19:45, the real-time power reached the day-ahead target power. This fully demonstrates that in the process of cloud-edge collaborative optimization and control, the control rate of the real-time response adjustment control strategy is very fast, and the power supply and demand can be balanced in a very short time under the condition of a sudden load increase. When minimizing the default cost is prioritized, each substation will, on the basis of distributed resource power supply, utilize the adjustable resources within the substation, including energy storage discharge and gas turbines, to supplement the real-time power shortage of market demand. During the 0:00-5:00 and 22:00-0:00 periods, Substation 2 will purchase electricity from the market, and distributed resources and gas turbines will jointly output power to meet the load demand while charging the energy storage. The energy storage resources in Substation 2 are relatively sufficient. The energy storage is charged during the period of 3:00-5:00. The purpose is to allow the substation to have more energy storage resources to call upon when electricity is scarce during the day. However, Substation 1 only has photovoltaic resources and no photovoltaic resources are output at night, so it can only purchase electricity from Substation 3 and the power market to meet the load demand. During the 9:00 AM to 10:00 PM period, all three substations meet their day-ahead planned power requirements through the output of distributed energy storage and gas turbines, as well as electricity purchased from the power market. By leveraging multiple resources, the VPP ensures that the day-ahead planned requirements are met. Therefore, the penalty cost in this scenario is relatively small, but the regulation cost and electricity purchase cost from the spot market are relatively high.

[0150] like Figure 6 Figure 2 shows the real-time response execution of each grid area participating in the day-ahead plan under Scenario 3. Scenario 3 prioritizes maximizing real-time spot market revenue. Therefore, between 00:00 and 5:00, grid areas 2 and 3, when distributed resources are sufficient, regulate their internal resources to sell electricity to the spot market. Grid area 1, lacking wind power resources, supplies power to its internal loads through energy storage and gas turbines. Grid area 2, due to insufficient nighttime energy storage charging, has a relatively low energy storage output compared to Scenario 2, so gas turbine output is increased to ensure power balance. From 9:00 to 15:00, grid areas 1 and 3 regulate their internal gas turbines and energy storage to meet response requirements while selling all excess electricity to the spot market. This regulation maximizes real-time market revenue in Scenario 3, but the regulation and penalty costs are relatively high.

[0151] like Figure 7As shown in the diagram of the real-time response execution of each substation participating in the day-ahead plan under scenario four, taking into account resource regulation costs, market returns, and default costs, the overall operational deviation will be larger than when the penalty cost is lowest. This is because, considering the electricity spot market, if the real-time electricity price is higher than the penalty cost, within the default power range, the VPP can choose to accept the day-ahead real-time power deviation, selling more electricity to the electricity market when the electricity price is high and charging more electricity to distributed energy storage when the electricity price is low. If the regulation cost is lower than the penalty cost, the VPP can choose to accept the power deviation penalty cost. Therefore, during the 0:00-5:00 and 9:00-10:00 periods, the penalty cost per unit power is lower, and the intraday real-time power will be lower than the day-ahead target power, so the VPP actively accepts the penalty cost. When the deviation aligns with real-time market demand, and if there's enough electricity available within the VPP to participate in the real-time market, the VPP will profit from the excess electricity from the real-time deviation. During the early morning hours, when electricity demand is low, grid stations 2 and 3 will use some of their wind power resources for energy storage charging and market trading. As shown in the diagrams for scenarios 1 to 3 above, energy storage charging is prioritized during the 12:00 AM to 4:00 AM period, when electricity prices are low. During the 5:00 AM to 6:00 AM period, when electricity prices are high, spot market trading is prioritized. When the real-time power is lower than the target power, the goal is to minimize the real-time operating cost of the VPP. When the resource regulation cost is low, the real-time power deviation is corrected by dispatching the substation resources first. During the 17:30-20:00 period, the resource regulation cost is lower than the penalty cost, and the three substations actively choose to increase the output of gas turbines and energy storage to meet the load demand; if the default penalty cost is low, a certain response deviation is allowed, and the default penalty cost is borne first. During the 15:00-19:30 period, when the penalty cost is lower than the resource regulation cost, the three substations actively reduce the output of energy storage and gas turbines; when the electricity spot market income is higher than the resource dispatch cost, within the constraint range of the default power , giving priority to participating in electricity spot market transactions. When electricity prices are higher during the 7:00-8:00 and 9:00-10:00 periods, substations 2 and 3 sell electricity in the electricity market. When electricity prices are high during the 16:00-17:00 period, substation 1 adjusts the distributed resources in the substation to produce more power and sell electricity to the electricity market. Overall, VPP achieves the goal of minimizing real-time operating costs by considering multiple cost factors in scenario 4. In addition, when electricity prices fluctuate, the VPP operation platform, as the cloud, sends real-time dispatch instructions to the edge side of the substation, and the substation executes the cloud dispatch instructions very quickly, which fully demonstrates the real-time and accuracy of cloud-edge collaborative optimization scheduling.

[0152] like Figure 8 As shown in the figure, after about 60 iterations, the cost of each area reaches a constant value, which meets the result of the consistent convergence of the slight increase rate of the cost of each area required by the ADMM algorithm.

[0153] like Figure 9 As shown, after 60 iterations, the residual is basically 0, which is consistent with the above Figure 8 The optimal convergence diagram of the substation cost corresponds to this, proving that the algorithm converges to the lowest comprehensive cost of VPP.

[0154] In summary, to address the significant power discrepancies between the day-ahead and real-time target power during the real-time response phase of VPP day-ahead market dispatch plans and actual target power when wind and PV output forecasts are inaccurate or when load fluctuates significantly, a cloud-edge collaborative real-time optimization scheduling framework has been constructed, leveraging the speed and accuracy of cloud and edge computing. The VPP operator, acting as the cloud, is responsible for implementing the global cost-optimizing operation strategy. The edge-side agent utilizes the ADMM algorithm to perform distributed optimization calculations on the VPP's substation resources. Overall, cloud-edge collaborative optimization scheduling achieves rapid and accurate real-time response and global cost optimization. This cloud-edge collaborative scheduling approach effectively implements local substation control, effectively resolving communication issues between the substation and the VPP control platform. It also enables rapid adjustments, addressing the drawback of slow global control scheduling. When there is a significant discrepancy between actual and predicted distributed resources or when load fluctuates, internal resources can be locally absorbed to avoid further losses caused by load fluctuations. The ADMM algorithm ensures consistency in substation cost increments, ensuring optimal allocation of dispatched power across substations and minimizing the VPP's global operating costs. The PSO algorithm effectively improves computational speed and accuracy. Combining the two algorithms achieves the ultimate optimal scheduling. Simulations simulated four scenarios for optimizing VPP operating costs, taking into account regulation costs, penalty costs, real-time spot market electricity purchase and sales revenue, and a comprehensive consideration of all three costs. The simulations presented specific operational strategies for three substations within a substation cluster. The cloud-edge collaborative strategy reduced operating costs by 34% through real-time deviation correction, demonstrating the comprehensive economic and practical benefits of scenario four, considering all aspects of costs. The simulations also verified the convergence of the final iteration and the consistency of the incremental cost rate across substations.

[0155] Therefore, the present invention adopts the above-mentioned VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization, cloud-side global optimization and edge-side distributed collaborative mechanism, which effectively solves the problem of difficulty in balancing global economy and real-time response and insufficient dynamic adaptability in the construction of traditional scheduling models, and can enable the scheduling strategy to continuously adapt to the wind and solar output fluctuations and load mutation conditions in the virtual power plant. This method enables the virtual power plant to more accurately dynamically and collaboratively optimize the resources of each substation during operation, providing a new control paradigm for multi-time scale scheduling. This method can provide an efficient working method for the collaborative optimization of the cloud and edge sides in existing virtual power plants, thereby solving the problems of delayed response and economic imbalance between substations caused by traditional centralized optimization, so that virtual power plants can better meet the higher requirements of modern power grids for real-time frequency regulation speed, economy and operational stability.

[0156] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization, characterized by: The following steps are involved: S1. Input the training data of the VPP particle swarm optimization algorithm, including the initial real-time market transaction power, regulation power, and default power; S2. Set the initial state of the VPP particle swarm optimization algorithm, including the initial position of the particles, the initial velocity of the particles, and the number of iterations u; S3, particle swarm cross mutation, obtain the area control power after the uth iteration Default Power and market trading power S4. Combine the day-ahead deviation with the minimum VPP real-time operating cost as the optimal goal, make decisions on the control power, default power and real-time market transaction rate, and calculate the real-time deviation correction cost C. VPP , and the constraints are power balance constraints, substation group control power constraints, market transaction constraints and breach of contract constraints; S5. Confirm the individual and global optimal solutions in the iterative process, denoted as and If C' is not satisfied VPP >C VPP , and As a new iteration value to participate in the next cycle calculation, if C' VPP >C VPP , and Participate in the next cycle calculation as a new iteration value; S6. Let u=u+1, if u≥u max , then exit the loop and output the result of the optimal target, otherwise return to S3 to continue the loop operation; S7, when u≥u max When , the loop is exited and the result of the optimal target is output. The substation-master receives the optimized total control power of the substation group from the cloud, and the substation-master exchanges information with adjacent substations based on the total control power. S8. Use the ADMM consensus algorithm to make each substation achieve a unified target incremental rate λ, and model it based on the optimal target and constraints of local control costs; S9. Introducing the Lagrange multiplier λ j,t Construct augmented Lagrangian function; S10, iteratively updating the substation control power variables, shared variables and Lagrange multipliers respectively; S11. Determine whether the convergence conditions are met. If so, output the actual regulated power of each substation. If not, return to step S9, continue updating the augmented Lagrangian function, and iteratively update the substation regulated power variables, shared variables, and Lagrangian multipliers again until convergence.

2. A VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 1, characterized in that: In S1, the VPP particle swarm optimization algorithm is specifically: The cloud sends the control power of each substation group to the substation-master of the edge server. The substation-master completes the distributed optimization calculation of the control power of each substation in the area and performs secondary optimization allocation. The expression is as follows: Where, M is the number of stations in the ith station group; C j.t is the control cost of the jth station at time t, P j.t is the regulated power of the j-th substation at time t.

3. The VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 2 is characterized in that: S4 is specifically: Taking the day-ahead deviation into consideration and minimizing the VPP real-time operating cost as the optimal goal, decisions are made on the regulated power, default power, and real-time market transaction rate. The objective function is as follows: minC VPP =C DK +C PC -C MA (2); Where C VPP is the operating cost of VPP in the real-time phase, C DK is the total cost of VPP resource control, C PC is the cost of liquidated damages, C MA It is the real-time profit from transactions between VPP and the electricity spot market; Calculate the real-time deviation correction cost C' according to formula (2) VPP , and with power balance constraints, substation group control power constraints, market transaction constraints and default constraints as constraints, the expression is as follows: a. Power balance constraints are: ΔP VPP.t =P am.t -P VPP.t (3); Where, P am.t is the planned output of VPP predicted by the day before in period t, P VPP.t is the actual output during the actual operation phase, ΔP VPP.t is the difference between the two; Where ΔP VPP1.t P is the transaction power deviation of VPP participating in the real-time electricity spot market. D1.i.t is the resource regulation power actually used by the grid group i to participate in the electricity market transaction during period t, ΔP VPP2.t is the power deviation to be corrected for VPP, P D2.i.t The resource control power of the station group i in the real-time deviation correction in time period t; The total control power of the station group i is the sum of the control power participating in market transactions and real-time deviation correction. The specific formula is as follows: b. The power constraint of the substation group control is: 0≤P D.i.t ≤P Dmax.i.t (6); Where, P Dmax.i.t is the maximum control power of the station group i during period t; c. Market transaction constraints: Where, and It is the maximum power that the VPP can purchase and sell in the real-time electricity spot market; d. The default constraints are: εP am.t ≤P A.t ≤P am.t (8); Where ε is the default coefficient.

4. The VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 3 is characterized in that: S8 specifically: Each substation in the VPP substation group is considered as a node. The ADMM consensus algorithm is used to make each substation reach a unified cost increment rate λ. By adjusting the control variables, the cost increment rates of each substation are made consistent, that is: By introducing a shared variable z to express the consistency of the cost increment rate of each area, that is, each area has an expression for this constraint, as follows: Based on the optimal target and constraints of local control cost, model:

5. The VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 4 is characterized in that: S9 specifically: Augmented Lagrangian function:

6. A VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 5, characterized in that: In S10, the control power variable P of each substation is first updated. j,t , we need to fix the shared variable z and the Lagrange multiplier λ j,t , its specific expression is as follows: Where, is the control power variable of station j after k+1 iterations, z k is the shared variable after k iterations, is the Lagrange multiplier variable after the last iteration, and are the Lagrange multiplier variables for the upper and lower power limits after the last iteration.

7. The VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 6 is characterized in that: In S10, the control power variable P of each substation is updated. j,t Then, the current cost increment rate is calculated according to formula (1) and formula (9), and the new shared variable z is obtained. k+1 , that is, z after k+1 iterations k+1 , its specific expression is as follows:

8. The VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 7 is characterized in that: In S10, the Lagrange multiplier λ of each station is updated j,t , its specific expression is as follows: Then update the Lagrange multipliers of the upper and lower power limits, and their specific expressions are as follows:

9. The VPP real-time scheduling method based on cloud-edge collaborative architecture and ADMM consistency optimization according to claim 8 is characterized in that: In S10, set the judgment condition for iteration termination: Where, if after k+1 iterations, the change in power scheduling is very small and the change in cost increment rate is also very small, then the iteration ends and the result converges, and θ is the corresponding convergence residual.

Images