Distributed energy base optimization method and terminal based on adaptive ADMM

Abstract

The application discloses a kind of distributed energy base optimization method and terminal based on adaptive ADMM, based on the uncertainty scenario set established to establish the distributed energy base model with minimum operating cost as objective function, and decoupling is carried out to it, and subproblem is obtained, the risk of distributed energy base model is quantified using CVaR theory, and risk is added to objective function, based on updated objective function using adaptive ADMM to solve subproblem, obtain solving result, to this using CVaR theory quantifies risk Ensure that the benefits and risks of each subject can be balanced, combined with adaptive ADMM to solve the problem, compared with traditional centralized optimization algorithm, only local communication between nodes is needed, global optimal solution can be solved by parallel computing, without global coordination or hierarchical division, can improve optimization speed, in the case of less error, the problem can be solved faster, to improve solving efficiency, and ensure data privacy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of distributed energy base, and particularly relates to a distributed energy base optimization method based on adaptive ADMM and a terminal. BACKGROUND

[0002] At present, some problems about uncertain parameters of the distributed energy base are solved, such as new energy output and the uncertainty of the decision maker to the risk. The traditional view is to regard the uncertain coefficient as a stable number. However, due to the imperfect prediction method, additional rotating equipment capacity and supplementary standby resources need to be provided, which increases the power cost. The traditional method models the uncertainty and fluctuation as a non-constant value optimization method, such as the solution of robust optimization and uncertain optimization problem. The realization of any possible parameter must be feasible, and is based on the optimization in the worst case. The random programming algorithm needs to know the probability distribution of the uncertain set.

[0003] In terms of solving the traditional model, existing commercial non-linear programming solvers, such as IPOPT (non-linear optimization solver), can usually solve non-linear optimization with continuous variables and are easy to implement. However, they can only find local solutions and cannot find global optimal solutions. When the model becomes larger, it may lead to slow convergence speed or even convergence failure. In terms of non-linear term processing, piecewise linearization (MILP) based on mixed integer linear programming can be used, in which the approximation accuracy can be well controlled by adjusting the number of linear segments, but the computational efficiency is still a major challenge, and this method belongs to NP-hard. And for the engineering practice and operation data privacy problem, the centralized processing of the increasingly large data is difficult, the hardware requirements of the central data processor are increasing, and the centralized processing of a large amount of private data is easy to cause data leakage.

[0004] The main problem of the above method is that the existing uncertain parameters are too conservative for the traditional uncertain set, and the traditional solver is difficult to find the optimal solution or the solving efficiency is slow, and the privacy problem is increasingly valued. The transmission of private information to the central processor may cause privacy leakage. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a distributed energy base optimization method based on adaptive ADMM and a terminal, which can improve the solving efficiency and ensure data privacy.

[0006] To solve the above technical problems, the technical scheme adopted by the present application is:

[0007] A distributed energy base optimization method based on adaptive ADMM, comprising the steps of:

[0008] constructing an uncertainty scenario set of the distributed energy base;

[0009] establishing a distributed energy base model with a target function of minimum operation cost based on the uncertainty scenario set, and decoupling the distributed energy base model to obtain a sub-problem;

[0010] quantifying a risk of the distributed energy base model using CVaR theory, and adding the risk to the target function to obtain an updated target function;

[0011] solving the sub-problem based on the updated target function using adaptive ADMM to obtain a solution result.

[0012] In order to solve the above technical problems, another technical solution adopted by the present application is:

[0013] A distributed energy base optimization terminal based on adaptive ADMM comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the following steps when executing the computer program:

[0014] constructing an uncertainty scenario set of the distributed energy base;

[0015] establishing a distributed energy base model with a target function of minimum operation cost based on the uncertainty scenario set, and decoupling the distributed energy base model to obtain a sub-problem;

[0016] quantifying a risk of the distributed energy base model using CVaR theory, and adding the risk to the target function to obtain an updated target function;

[0017] solving the sub-problem based on the updated target function using adaptive ADMM to obtain a solution result.

[0018] The beneficial effects of the present application are that: based on the constructed uncertainty scenario set, a distributed energy base model with a target function of minimum operation cost is established, the distributed energy base model is decoupled to obtain a sub-problem, the risk of the distributed energy base model is quantified using the CVaR theory, and the risk is added to the target function, the sub-problem is solved based on the updated target function using the adaptive ADMM, and a solving result is obtained, so as to quantify the risk by using the CVaR theory to ensure that the benefits and risks of each subject can be balanced, the problem is solved by using the adaptive ADMM, compared with the traditional centralized optimization algorithm, only local communication between nodes is needed, the global optimal solution can be solved by parallel computing, without global coordination or hierarchical division, the optimization speed can be improved, in the case of small error, the problem can be solved faster, thereby improving the solving efficiency, and the data privacy is ensured. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A step flow chart of a distributed energy base optimization method based on adaptive ADMM for an embodiment of the present application is shown in the figure.

[0020] Figure 2 A structure schematic diagram of a distributed energy base optimization terminal based on adaptive ADMM for an embodiment of the present application is shown in the figure.

[0021] Figure 3 A large scenario set generated in the distributed energy base optimization method based on adaptive ADMM for an embodiment of the present application is shown in the figure.

[0022] Figure 4 An uncertainty scenario set after scenario reduction in the distributed energy base optimization method based on adaptive ADMM for an embodiment of the present application is shown in the figure.

[0023] Figure 5 A distributed energy base model decoupling schematic diagram in the distributed energy base optimization method based on adaptive ADMM for an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0024] To explain the technical content, the achieved purposes and effects of the present application in detail, the following will be explained in combination with the embodiments and the accompanying drawings.

[0025] Please refer to Figure 1 A distributed energy base optimization method based on adaptive ADMM, including the steps of:

[0026] An uncertainty scenario set of the distributed energy base is constructed;

[0027] Based on the aforementioned set of uncertain scenarios, a distributed energy base model is established with the objective function of minimizing operating costs. The distributed energy base model is then decoupled to obtain sub-problems.

[0028] The risk of the distributed energy base model is quantified using CVaR theory, and the risk is added to the objective function to obtain the updated objective function.

[0029] The subproblem is solved using the updated objective function using Adaptive ADMM to obtain the solution result.

[0030] As can be seen from the above description, the beneficial effects of this invention are as follows: A distributed energy base model with minimum operating cost as the objective function is established based on a set of constructed uncertainty scenarios. The distributed energy base model is decoupled to obtain sub-problems. The risk of the distributed energy base model is quantified using CVaR theory, and the risk is added to the objective function. Based on the updated objective function, the sub-problems are solved using adaptive ADMM to obtain the solution results. This utilizes CVaR theory to quantify risk, ensuring a balance between benefits and risks for each entity. Combined with adaptive ADMM to solve the problem, compared to traditional centralized optimization algorithms, only local communication between nodes is needed to solve the global optimal solution through parallel computation, without requiring global coordination or hierarchical partitioning. This improves optimization speed, and with smaller errors, the problem can be solved faster, thereby improving solution efficiency and ensuring data privacy.

[0031] Furthermore, the set of uncertainties in constructing the distributed energy base includes:

[0032] Generate a sampling scenario of wind and solar power output for a distributed energy base;

[0033] A large scene set is generated using time series analysis based on the aforementioned wind and solar power output sampling scenarios.

[0034] The large set of scenes is reduced based on the principle of probabilistic distance to obtain a set of uncertain scenes.

[0035] As described above, when faced with the growth in the scale of the scene set and the corresponding increase in computing demand, the use of time series analysis to generate a large scene set based on the wind and solar power output sampling scene can effectively control the scene scale while maintaining computing accuracy. The large scene set can be reduced based on the principle of probabilistic distance, thereby reducing energy uncertainty and improving model robustness.

[0036] Furthermore, the reduction of the large scene set based on the probabilistic distance principle to obtain an uncertain scene set includes:

[0037] Calculate the Euclidean distance between each pair of scenes in the large scene set;

[0038] A central scene is determined from the large scene set, such that the sum of the probabilistic distances between the central scene and other scenes in the large scene set is minimized;

[0039] Determine the target scene with the smallest Euclidean distance to the central scene, and add the probability of the central scene to the probability of the target scene;

[0040] The central scene is removed from the large scene set to obtain a new scene set;

[0041] Determine whether the number of scenes in the new scene set has reached the preset number of scenes. If yes, then the new scene set is used as an uncertain scene set. If no, then return to perform the calculation of the Euclidean distance between each pair of scenes in the large scene set.

[0042] As described above, by continuously reducing the number of scenes in a large scene set to reach the preset number of scenes, a small number of representative scenes are used to approximate the generated large scene, thereby reducing the complexity of subsequent calculations and maintaining a certain level of reliability.

[0043] Furthermore, the objective function is:

[0044]

[0045]

[0046]

[0047]

[0048] In the formula, W represents the operating cost, and C Inv C represents the total investment cost of the energy base. Run C represents the operating and maintenance costs of energy base equipment. Dep C represents the total depreciation cost of energy base equipment. pumped This represents the total cost of a pumped storage power station. C represents the penalty cost for wind curtailment at energy bases. res C represents the cost of generating electricity. buy This indicates the cost of purchasing and selling electricity from the higher-level power grid. This indicates the unit cost of generating electricity from renewable energy sources. c represents the power output of new energy sources at time t, where T represents the dispatch period, and c represents the dispatch cycle. run This indicates the unit power generation maintenance cost. This represents the total power generation at time t. This indicates the price at which electricity is purchased and sold from the higher-level power grid. This indicates the purchase and sale of electricity from the superior power grid.

[0049] As can be seen from the above description, the objective function is to minimize the operating cost, thereby ensuring the economic efficiency of the distributed energy base operation.

[0050] Furthermore, the step of incorporating the risk into the objective function to obtain the updated objective function includes:

[0051]

[0052] In the formula, β represents the decision-maker's risk preference, W represents operating costs, δ represents the threshold, and N... Ω Let α represent the total number of samples, α represent the confidence level, and η represent the total number of samples. ω Let ω represent the auxiliary function, and let ω represent the sampling point in the discrete operation.

[0053] As described above, incorporating the quantified risks into the objective function allows power distribution system operators to reasonably weigh risks and costs, making various resource allocation and combination schemes more robust and economical.

[0054] Furthermore, the subproblem is solved using Adaptive ADMM based on the updated objective function, and the solution results include:

[0055] Determine the initial values ​​for the number of iterations, the penalty factor, the Lagrange multipliers, and the convergence criterion, and initialize each agent decision variable;

[0056] Iterate through the subproblems to obtain the optimal decision solution, and then pass the coupling variables to the adjacent agents;

[0057] The scheduling scheme is obtained by solving the subproblem of each adjacent agent based on the coupling variables;

[0058] Determine whether the original residual and the dual residual satisfy the convergence criterion. If yes, the iteration ends and the scheduling scheme is used as the solution result. If no, update the penalty factor, increment the iteration count by one, and return to execute the subproblem of the iteration.

[0059] As described above, in traditional ADMM, the penalty variable is usually stable, which may lead to convergence problems. Adaptive ADMM mainly introduces an adaptive step size element for the penalty factor during the iteration process. This allows the penalty factor to be adaptively adjusted according to the relationship between the original residual and the dual residual as the optimization objective value changes, thereby effectively improving the efficiency of model solving.

[0060] Furthermore, the sub-problem is:

[0061]

[0062]

[0063]

[0064]

[0065] In the formula, x represents each agent's decision variable, x k+1 Let x represent the agent decision variables in the (k+1)th iteration. k Let y represent the agent decision variables in k iterations, and let y represent the global variable of the distributed energy base. k y represents the global variable for the distributed energy base in k iterations. k+1 Let λ be the global variable representing the distributed energy base in the (k+1)th iteration. i,t Represents the first Lagrange multiplier. Denotes the first Lagrange multiplier in k iterations. Let λ represent the first Lagrange multiplier in the (k+1)th iteration. j,t Indicates the second Lagrange multiplier. Denotes the second Lagrange multiplier in k iterations. Let ρ represent the second Lagrange multiplier in the (k+1)th iteration. i-j ρ represents the first penalty factor. i-j,k Let ρ represent the first penalty factor in the kth iteration. j-i ρ represents the second penalty factor. j-i,k Let T represent the second penalty factor for k iterations, T represent the scheduling period, and I represent the number of sub-agents in the energy base. Indicates the first coupling variable. Indicates the first auxiliary variable. Represents the first auxiliary variable in the kth iteration. Indicates the second coupling variable. Indicates the second auxiliary variable. This represents the second auxiliary variable for k iterations.

[0066] As described above, by decoupling and decomposing the distributed energy base model into sub-problems for distributed optimization, the entire system can be optimized through only a small amount of information exchange between adjacent subsystems during the solution process. This greatly improves network communication efficiency and ensures the privacy of information of each entity.

[0067] Furthermore, the convergence criterion is:

[0068]

[0069] In the formula, Represents the first auxiliary variable in the (k+1)th iteration. This represents the second auxiliary variable for iteration k+1. This represents the convergence accuracy of the original residual in the k-th iteration. ρ represents the convergence accuracy of the dual residual in the k-th iteration. i-j ρ represents the first penalty factor. j-i This represents the second penalty factor.

[0070] As described above, by determining the convergence criterion to determine whether the current solution is the optimal operating strategy, the effectiveness of the solution can be guaranteed.

[0071] Furthermore, the update of the penalty factor is as follows:

[0072]

[0073]

[0074] In the formula, ρ i-j,k+1 Let ρ represent the first penalty factor in the (k+1)th iteration. j-i,k+1 ν represents the second penalty factor for the (k+1)th iteration. j-i,k ν represents the deceleration factor of the second penalty factor in k iterations. i-j,k κ represents the deceleration factor of the first penalty factor in k iterations. j-i,k κ represents the acceleration factor of the second penalty factor in k iterations. i-j,k The acceleration factor represents the first penalty factor in k iterations. Represents the original residual. This represents the dual residual.

[0075] As described above, by updating the penalty factor to adaptively adjust with changes in the optimization objective value, convergence problems can be avoided, thus finding the global optimum and ensuring the reliability of the solution results.

[0076] Please refer to Figure 2 Another embodiment of the present invention provides a distributed energy base optimization terminal based on adaptive ADMM, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements each step of the above-described distributed energy base optimization method based on adaptive ADMM.

[0077] The distributed energy base optimization method and terminal based on adaptive ADMM described above are applicable to distributed energy base optimization and scheduling scenarios. The following detailed implementation methods illustrate this:

[0078] Please refer to Figure 1 , Figures 3-5 Embodiment 1 of the present invention is as follows:

[0079] An optimization method for distributed energy bases based on adaptive ADMM includes the following steps:

[0080] S1. A set of uncertain scenarios for constructing distributed energy bases, specifically including S11-S13:

[0081] S11. Generate a sampling scenario of wind and solar power output for a distributed energy base.

[0082] Specifically, autoregressive parameters and moving average parameters are set to generate sampling scenarios for wind and solar power output in distributed energy bases.

[0083] S12. Based on the aforementioned wind and solar power output sampling scenarios, a large scene set is generated using time series analysis, such as... Figure 3 As shown.

[0084] S13. Based on the principle of probabilistic distance, the large scene set is reduced to obtain an uncertain scene set, such as... Figure 4 As shown, specifically including S131-S135:

[0085] S131. Calculate the Euclidean distance between each pair of scenes in the large scene set to quantify the differences between scenes.

[0086] S132. Determine a central scene from the large scene set, wherein the sum of the probability distances between the central scene and other scenes in the large scene set is minimized.

[0087] S133. Determine the target scene with the smallest Euclidean distance to the central scene, and add the probability of the central scene to the probability of the target scene.

[0088] S134. Remove the central scene from the large scene set to obtain a new scene set.

[0089] S135. Determine whether the number of scenes in the new scene set has reached the preset number of scenes. If yes, then the new scene set is used as an uncertain scene set. If no, then return to execute S131-S135.

[0090] The number of preset scenarios can be flexibly set according to the actual situation.

[0091] S2. Based on the set of uncertain scenarios, establish a distributed energy base model with the objective function of minimizing operating costs, and decouple the distributed energy base model to obtain sub-problems, specifically including S21-S22:

[0092] S21. Based on the set of uncertain scenarios, establish a distributed energy base model with the objective function of minimizing operating costs.

[0093] The objective function is:

[0094]

[0095]

[0096]

[0097]

[0098] In the formula, W represents the operating cost, and C Inv C represents the total investment cost of the energy base. Run C represents the operating and maintenance costs of energy base equipment. Dep C represents the total depreciation cost of energy base equipment. pumped This represents the total cost of a pumped storage power station. C represents the penalty cost for wind curtailment at energy bases. res C represents the cost of generating electricity. buy This indicates the cost of purchasing and selling electricity from the higher-level power grid. This indicates the unit cost of generating electricity from renewable energy sources. c represents the power output of new energy sources at time t, where T represents the dispatch period, and c represents the dispatch cycle. run This indicates the unit power generation maintenance cost. This represents the total power generation at time t. This indicates the price at which electricity is purchased and sold from the higher-level power grid. This indicates the purchase and sale of electricity from the superior power grid.

[0099] The distributed energy base model consists of distributed resources such as wind power, fuel cells, energy storage, and loads. The model also includes constraints on active power balance, transmission system, battery operation, pumped storage, electrolyzer, fuel cell, and external characteristics.

[0100] The active power balance constraint is as follows:

[0101] P Eb,i,t +P fc,i,t +P h,i,t =P ele,i,t +P bat,i,t +P p,i,t +P wind,i,t +P load,i,t ;

[0102] In the formula, P Eb,i,t P represents the wind power output of the energy base. fc,i,t P represents the power generation capacity of a fuel cell. h,i,t P represents the power output of a pumped-storage hydroelectric power station when releasing water. ele,i,t P represents the power of the electrolytic cell.bat,i,t P represents battery power. p,i,t P represents the pumping power of a pumped storage power station during pumping. wind,i,t P represents the power of wind curtailment. load,i,t This indicates the load power.

[0103] Among them, the wind curtailment power in energy bases should be less than the wind power output: 0≤P wind,i,t ≤P Eb,i,t ;

[0104] The constraints of the power transmission system are:

[0105] 0≤P net,i,t ≤P trans ;

[0106] 0.9P net,i,t-1 ≤P net,i,t ≤1.1P net,i,t-1 ;

[0107] In the formula, P net,i,t P represents the current power transmission system capacity during the current time period. trans P represents the on-grid power of the power transmission system. net,i,t-1 This indicates the capacity of the power transmission system in the previous period.

[0108] Traditional battery models assume that the battery is always charged and discharged at its rated power. However, these models cannot accurately estimate the relationship between battery power changes and SOC (State of Charge), and cannot meet the expected response of the optimal capacity configuration of the ESS (Energy Storage System). Therefore, this invention considers the dynamic power characteristics of the battery during charging and discharging. The battery operating constraints are as follows:

[0109]

[0110] In the formula, P bat,max S indicates the upper limit of battery power. dth S represents the threshold at which the battery's rated power limit switches to a variable power limit SOC. cth S represents the threshold value at which the battery's rated lower power limit switches to a variable lower power limit. bat,i,t This indicates the real-time SOC state of the battery; when 0 ≤ P bat,i,t When P indicates that the battery is charging; when P bat,i,t When the value is ≤0, it indicates that the battery is in a discharged state;

[0111] Changes in battery SOC and battery power P bat,i,t The relationships between them, as well as the upper and lower limits of SOC, are as follows:

[0112]

[0113] In the formula, S bat,i,t+1 This indicates the battery's real-time SOC state for the next time period, ΔT represents the continuous operating time, and E... bat S represents the total battery capacity. bat,min S represents the lower bound of SOC. bat,max This indicates the upper limit of SOC.

[0114] The operation of a hydroelectric power plant primarily depends on the pumped storage capacity and the installed capacity of the hydraulic turbines. The operational model constraints mainly consist of three parts: inflow and outflow rates, storage capacity, and power constraints. The pumped storage constraints are as follows:

[0115]

[0116]

[0117] In the formula, γ H P represents the binary variable representing the water released by a pumped storage power station. psmax Indicates the installed capacity of the turbine unit, γ P c represents the binary variable representing the pumping action of a pumped-storage power station. psmin E represents the minimum pumped storage capacity ratio. psmax E represents the maximum capacity of pumped storage. ps,i,t E represents the capacity of pumped storage at time t. ps,i,t-1 c represents the capacity of pumped storage at time t-1. psmax α represents the maximum pumped storage capacity ratio. H α represents pumping efficiency. P This indicates the water discharge efficiency.

[0118] The constraints of the electrolytic cell are:

[0119]

[0120] P ele,i,t =η ele V he,i,t ;

[0121] In the formula, n represents the minimum operating power of the electrolytic cell. ele,i,t Represents a binary variable, n ele,i,t =1 indicates the start-up state of the electrolytic cell, n ele,i,t =0 indicates that the electrolytic cell is in a stopped state. η represents the maximum operating power of the electrolytic cell. ele V represents the electrical energy consumed to produce one unit volume of hydrogen gas. he,i,t This indicates the volume of hydrogen produced during the operation of the electrolyzer.

[0122] The constraints of the fuel cell are:

[0123]

[0124] In the formula, n fc i ,t This indicates the start / stop status of the fuel cell. η represents the maximum power of the fuel cell. fc V represents the electrical energy produced by a fuel cell consuming a unit volume of hydrogen. hf,i,t This indicates the output of the fuel cell.

[0125] The external characteristic constraints include wind power constraints, wind speed constraints, and pitch angle constraints. The wind power constraints are as follows:

[0126] 0≤P wind,i ≤P wind,max ;

[0127] In the formula, P wind,i P represents the wind power output of the wind farm at time i. wind,max This indicates the rated maximum wind power output of the wind farm.

[0128] The wind speed constraint is:

[0129] v cutin ≤v work ≤v uto t ;

[0130] In the formula, v cutin Indicates the cut-in wind speed, v work Indicates the operating wind speed of the fan, v cutout This indicates the cut-out wind speed.

[0131] The pitch angle constraint is:

[0132] β min ≤β≤β max ;

[0133] In the formula, β represents the wind turbine pitch angle, β min β represents the lower limit of the wind turbine pitch angle. max This indicates the upper limit of the wind turbine pitch angle.

[0134] S22. Decouple the distributed energy base model to obtain a sub-problem, specifically:

[0135]

[0136]

[0137]

[0138]

[0139] In the formula, x represents each agent's decision variable, x k+1 Let x represent the agent decision variables in the (k+1)th iteration. k Let y represent the agent decision variables in k iterations, and let y represent the global variable of the distributed energy base. k y represents the global variable for the distributed energy base in k iterations. k+1 Let λ be the global variable representing the distributed energy base in the (k+1)th iteration. i,t Represents the first Lagrange multiplier. Denotes the first Lagrange multiplier in k iterations. Let λ represent the first Lagrange multiplier in the (k+1)th iteration. j,t Indicates the second Lagrange multiplier. Denotes the second Lagrange multiplier in k iterations. Let ρ represent the second Lagrange multiplier in the (k+1)th iteration. i-j ρ represents the first penalty factor. i-j,k Let ρ represent the first penalty factor in the kth iteration. j-i ρ represents the second penalty factor. j-i,k Let I represent the second penalty factor for k iterations, and let I represent the number of sub-agents in the energy base. Indicates the first coupling variable. Indicates the first auxiliary variable. Represents the first auxiliary variable in the kth iteration. Indicates the second coupling variable. Indicates the second auxiliary variable. This represents the second auxiliary variable in the k iterations. Meanwhile, each agent within the county-level network element aims to minimize its own operating cost.

[0140] Compared to traditional centralized optimization methods, distributed algorithms only require local communication between nodes to solve for the global optimum through parallel computation, without the need for global coordination or hierarchical partitioning. In terms of communication, this algorithm only requires finding a consistent, strongly connected graph, resulting in low communication requirements. It also offers advantages such as strong adaptability, data privacy protection, and distributed computation. To use the adaptive ADMM (Alternating Direction Multiplier Method) algorithm for distributed optimization, the distributed energy base model needs to be decoupled into regions. During decoupling, node 1 and node 2 are respectively treated as virtual power injection nodes into sub-regions. During the solution process, sub-regions A and B only need to exchange a small amount of information with their adjacent subsystems to achieve distributed optimization of the entire system. Figure 5 As shown, sub-agents Agent1 and Agent2 only need to interact with variables X1 = [P1, Q1, U1] and X2 = [P2, Q2, U2], which greatly improves network communication efficiency and ensures the privacy of information of each entity.

[0141] The reason for introducing auxiliary variables is to allow all agents to consider a globally optimal solution, thus decoupling them as individual entities. After decoupling, all coupled variables must satisfy the consistency condition:

[0142]

[0143] In situations where renewable energy generation output is uncertain, the CVaR method is introduced. This method allows distribution system operators to reasonably weigh risks and costs, making various resource allocation and combination schemes more robust and economical.

[0144] In CvaR, the loss function f(x,y) is first established, where x and y are the decision variable and random variable, respectively, and x∈R. n y∈R m R n and R m Let n and m be real number spaces, respectively. The probability density function of the random variable y is set to p(y). If the decision variable x and the threshold δ are determined, then the cumulative distribution function of the loss function is as follows:

[0145]

[0146] Given a confidence level α∈(0,1), the VaR function can be expressed as follows for a given decision variable x:

[0147]

[0148] The CVaR function corresponding to this VaR function is as follows:

[0149]

[0150] To simplify the CVaR solution function to a certain extent, an auxiliary function was first constructed:

[0151]

[0152] Where, [f(x,y)-δ] + =max{f(x,y)-δ,0}, then the CVaR function can be transformed into:

[0153] V CVaR-α (x)=min α∈R F α (x,δ);

[0154] Secondly, since integral operations are continuous and difficult to solve, sampling points can be used to replace integral calculations, transforming it into discrete operations, which can further simplify the calculation process of CVaR.

[0155]

[0156] By introducing the auxiliary function η ω N Ω Given the total number of samples, the final formula for calculating risk cost is:

[0157]

[0158] The above risk cost calculation is applied to the distributed energy base model as follows:

[0159] S3. Quantify the risk of the distributed energy base model using CVaR theory, and add the risk to the objective function to obtain the updated objective function, specifically:

[0160]

[0161] In the formula, β represents the decision-maker's risk preference, β∈[0,1], and the smaller the value, the more risk-averse the decision-maker is. W represents the operating cost, δ represents the threshold, and N... Ω Let α represent the total number of samples, α represent the confidence level, and η represent the total number of samples. ω Let ω represent the auxiliary function, and let ω represent the sampling point in the discrete operation.

[0162] S4. Based on the updated objective function, the sub-problem is solved using Adaptive ADMM to obtain the solution result, specifically including S41-S44:

[0163] S41. Determine the initial values ​​for the number of iterations, the penalty factor, the Lagrange multipliers, and the convergence criterion, and initialize each agent decision variable.

[0164] The initial values ​​of the iteration number and the Lagrange multipliers are both 0.

[0165] S42. Iterate over the subproblem to obtain the optimal decision solution, and pass the coupling variable to the adjacent agent.

[0166] S43. The scheduling scheme is obtained by solving the subproblems of each adjacent agent according to the coupling variables.

[0167] S44. Determine whether the original residual and the dual residual satisfy the convergence criterion. If yes, the iteration ends and the scheduling scheme is used as the solution result. If no, update the penalty factor, increment the iteration number by one, and return to execute S42-S44.

[0168] The convergence criterion is as follows:

[0169]

[0170] In the formula, Represents the first auxiliary variable in the (k+1)th iteration. This represents the second auxiliary variable for iteration k+1. This represents the convergence accuracy of the original residual in the k-th iteration. ρ represents the convergence accuracy of the dual residual in the k-th iteration. i-j ρ represents the first penalty factor. j-i This represents the second penalty factor.

[0171] The updated penalty factor is:

[0172]

[0173]

[0174] In the formula, ρ i-j,k+1 Let ρ represent the first penalty factor in the (k+1)th iteration. j-i,k+1 ν represents the second penalty factor for the (k+1)th iteration. j-i,k ν represents the deceleration factor of the second penalty factor in k iterations. i-j,k κ represents the deceleration factor of the first penalty factor in k iterations. j-i,k κ represents the acceleration factor of the second penalty factor in k iterations. i-j,k The acceleration factor represents the first penalty factor in k iterations. Represents the original residual. This represents the dual residual.

[0175] Please refer to Figure 2 Embodiment two of the present invention is as follows:

[0176] A distributed energy base optimization terminal based on adaptive ADMM includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the distributed energy base optimization method based on adaptive ADMM in Embodiment 1.

[0177] In summary, this invention provides a distributed energy base optimization method and terminal based on adaptive ADMM. It establishes a distributed energy base model with the objective function of minimizing operating costs based on a constructed set of uncertain scenarios. The distributed energy base model is decoupled to obtain sub-problems. The risk of the distributed energy base model is quantified using CVaR theory and incorporated into the objective function. Based on the updated objective function, adaptive ADMM is used to solve the sub-problems, yielding the solution results. This method utilizes CVaR theory to quantify risk, ensuring a balance between benefits and risks for each entity. Compared to traditional centralized optimization algorithms, this approach only requires local communication between nodes and can solve the entire problem through parallel computing. The local optimum solution is obtained without the need for global coordination or hierarchical partitioning, which improves the optimization speed. Under conditions of small error, the problem can be solved more quickly, thereby improving the solution efficiency and ensuring data privacy. In addition, when faced with the growth of scene set size and the corresponding increase in computational demand, a large scene set is generated by using time series analysis based on wind and solar power output sampling scenes. This can effectively control the scene size while maintaining computational accuracy. The large scene set is reduced based on the principle of probabilistic distance, reducing energy uncertainty and improving model robustness. Furthermore, the large scene set is continuously reduced to reach the preset number of scenes, so that a small number of representative scenes can be used to approximate the generated large scene, reducing the complexity of subsequent calculations and maintaining a certain degree of reliability.

[0178] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A distributed energy base optimization method based on adaptive ADMM, characterized in that, Including the following steps: Uncertainty scenarios for constructing distributed energy bases; Based on the aforementioned set of uncertain scenarios, a distributed energy base model is established with the objective function of minimizing operating costs. The distributed energy base model is then decoupled to obtain sub-problems. The risk of the distributed energy base model is quantified using CVaR theory, and the risk is added to the objective function to obtain the updated objective function. The subproblem is solved using the updated objective function to obtain the solution result; The set of uncertainties in constructing a distributed energy base includes: Generate a sampling scenario of wind and solar power output for a distributed energy base; A large scene set is generated using time series analysis based on the aforementioned wind and solar power output sampling scenarios. The large scene set is reduced based on the principle of probabilistic distance to obtain an uncertain scene set; The reduction of the large scene set based on the probabilistic distance principle to obtain an uncertain scene set includes: Calculate the Euclidean distance between each pair of scenes in the large scene set; A central scene is determined from the large scene set, such that the sum of the probabilistic distances between the central scene and other scenes in the large scene set is minimized; Determine the target scene with the smallest Euclidean distance to the central scene, and add the probability of the central scene to the probability of the target scene; The central scene is removed from the large scene set to obtain a new scene set; Determine whether the number of scenes in the new scene set has reached the preset number of scenes. If yes, then the new scene set is regarded as an uncertain scene set. If no, then return to perform the calculation of the Euclidean distance between each pair of scenes in the large scene set. The step of adding the risk to the objective function to obtain the updated objective function includes: ； In the formula, β represents the decision-maker's risk preference, and W represents the operating cost. Indicates the threshold. Let α represent the total number of samples and α represent the confidence level. Describes an auxiliary function. Represents the sampling point in discrete computation; The sub-problem is solved using the updated objective function using Adaptive ADMM, and the solution results include: Determine the initial values ​​for the number of iterations, the penalty factor, the Lagrange multipliers, and the convergence criterion, and initialize each agent decision variable; Iterate through the subproblems to obtain the optimal decision solution, and then pass the coupling variables to the adjacent agents; The scheduling scheme is obtained by solving the subproblem of each adjacent agent based on the coupling variables; Determine whether the original residual and the dual residual satisfy the convergence criterion. If yes, the iteration ends and the scheduling scheme is used as the solution result. If no, update the penalty factor, increment the iteration count by one, and return to execute the subproblem of the iteration.

2. The distributed energy base optimization method based on adaptive ADMM according to claim 1, characterized in that, The objective function is: ； ； ； ； In the formula, W represents the operating cost. This indicates the total investment cost of the energy base. This indicates the operating and maintenance costs of energy base equipment. This represents the total depreciation cost of equipment at the energy base. This represents the total cost of a pumped storage power station. This indicates the penalty cost for wind curtailment at energy bases. Indicates the cost of energy generation. This indicates the cost of purchasing and selling electricity from the higher-level power grid. This indicates the unit cost of generating electricity from renewable energy sources. This represents the power output of new energy sources at time t, where T represents the dispatch period. This indicates the unit power generation maintenance cost. This represents the total power generation at time t. This indicates the price at which electricity is purchased and sold from the higher-level power grid. This indicates the purchase and sale of electricity from the superior power grid.

3. The distributed energy base optimization method based on adaptive ADMM according to claim 1, characterized in that, The sub-problem is: ； ； ； ； In the formula, x represents each agent's decision variable. Let each agent's decision variable represent the (k+1)th iteration. Let represent the agent decision variables in k iterations, and y represent the global variable of the distributed energy base. This represents a global variable representing the distributed energy base in k iterations. This represents the global variable for the distributed energy base in the (k+1)th iteration. Represents the first Lagrange multiplier. Denotes the first Lagrange multiplier in k iterations. Denotes the first Lagrange multiplier in the (k+1)th iteration. Indicates the second Lagrange multiplier. Denotes the second Lagrange multiplier in k iterations. Denotes the second Lagrange multiplier in the (k+1)th iteration. Indicates the first penalty factor. This represents the first penalty factor in the kth iteration. Indicates the second penalty factor. Let T represent the second penalty factor for k iterations, T represent the scheduling period, and I represent the number of sub-agents in the energy base. Indicates the first coupling variable. Indicates the first auxiliary variable. Represents the first auxiliary variable in the kth iteration. Indicates the second coupling variable. Indicates the second auxiliary variable. This represents the second auxiliary variable for k iterations.

4. The distributed energy base optimization method based on adaptive ADMM according to claim 3, characterized in that, The convergence criterion is as follows: ； In the formula, Represents the first auxiliary variable in the (k+1)th iteration. This represents the second auxiliary variable for iteration k+1. This represents the convergence accuracy of the original residual in the k-th iteration. This represents the convergence accuracy of the dual residual in the k-th iteration. Indicates the first penalty factor. This represents the second penalty factor.

5. The distributed energy base optimization method based on adaptive ADMM according to claim 3, characterized in that, The updated penalty factor is: ； ； In the formula, This represents the first penalty factor in the (k+1)th iteration. This represents the second penalty factor in the k+1th iteration. The deceleration factor represents the second penalty factor in the kth iteration. The deceleration factor represents the first penalty factor in the kth iteration. The acceleration factor represents the second penalty factor in k iterations. The acceleration factor represents the first penalty factor in k iterations. Represents the original residual. This represents the dual residual.

6. A distributed energy base optimization terminal based on adaptive ADMM, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the distributed energy base optimization method based on adaptive ADMM as described in any one of claims 1 to 5.

Images