An intelligent microgrid multi-area collaborative scheduling method and system based on heavy ball accelerated distributed optimization

Abstract

This invention provides a multi-zone collaborative scheduling method and system for intelligent microgrids based on heavy-ball acceleration distributed optimization, belonging to the field of intelligent microgrid technology. In the intelligent microgrid model, a minimum network loss objective function is defined to determine the generator output power by summing the active power losses. Power flow constraints, active and reactive power constraints of distributed sources, active and reactive power output constraints of generator sets, reactive power compensation constraints, and per-unit voltage and current constraints are defined. Second-order cone relaxation is used to process non-convex functions to obtain the constraints required for power flow solution. The intelligent microgrid model is optimized and controlled based on the ADMM distributed algorithm and an improved algorithm using heavy-ball acceleration technology. This invention can reduce network losses and improve the system's solution accuracy and convergence speed while ensuring that node voltage stability remains basically unchanged.

Description

Technical Field

[0001] This invention belongs to the field of intelligent microgrid technology, and proposes a multi-zone collaborative scheduling method and system for intelligent microgrids based on heavy ball acceleration distributed optimization. Background Technology

[0002] The ability to generate electricity using inexhaustible energy sources like solar and wind power is of great practical significance for promoting ecological civilization and high-quality development, and for addressing climate change issues. Therefore, distributed generation based on solar photovoltaic and wind power, including renewable power generation and distributed energy storage systems, has been vigorously developed. However, the utilization of new energy sources such as distributed photovoltaic and wind power is inherently uncertain and volatile, and the power generation and control methods of various energy sources differ, posing significant challenges to the reliability and stability of the power grid. In severe cases, this can even lead to the collapse of the power grid system, giving rise to smart microgrids.

[0003] Smart microgrids primarily consist of energy storage devices, distributed power sources, energy conversion, and protection devices. They are self-protecting, controllable, and manageable systems with advantages such as high reliability, autonomous operation, and environmental friendliness. They can achieve functions such as fault detection and protection, power balance control, system operation optimization, and power quality management through their own control and management of energy supply. Smart microgrids are mainly divided into stand-alone smart microgrids and grid-connected smart microgrids. Stand-alone smart microgrids are not connected to the external power grid and rely on their own internal coordination and optimization scheduling. Grid-connected smart microgrids are connected to the power grid and can exchange power. Unlike traditional power grid dispatching systems, smart microgrid optimization scheduling technology needs to coordinate multiple energy sources, fully utilize and explore the different characteristics of different energy sources, and requires comprehensive optimization algorithms to rationally allocate energy resources while ensuring the safe and stable operation of the smart microgrid.

[0004] With the increasing integration of renewable energy into smart microgrids and their growing penetration rate, the shortcomings of traditional centralized optimization methods are becoming increasingly apparent. Centralized optimization methods involve concentrating all decision-making and power in a single central entity. Under this method, the central entity manages and controls all aspects of the system or process. However, it has three main drawbacks. First, centralized optimization lacks robustness, requiring the central entity to communicate with controllable units to obtain global information before it can control those units, making it unsuitable for flexible topologies. Second, the need for a central entity to access global information places high demands on communication bandwidth and the processing of large amounts of data, increasing microgrid operating costs and communication burden. Finally, the central entity's access to global information under centralized optimization increases the risk of information leakage.

[0005] Distributed optimization methods decompose large tasks into subtasks and assign them to multiple agents, using these agents to achieve parallel and rapid solutions to large problems. Utilizing local information for coordination and control, they offer lower computational and communication overhead compared to centralized optimization methods, providing greater flexibility and reliability, better privacy protection, and enhanced network scalability due to their independence from a central node. Currently, mainstream distributed optimization methods include Consensus-based Optimization, Alternating Direction Method of Multipliers (ADMM), and Distributed Gradient Descent, as well as MapReduce and asynchronous optimization algorithms.

[0006] Distributed optimization methods in smart microgrids are mainly applied to unit combination, reactive power optimization, and economic dispatch, and secondarily to relay protection, electricity pricing, and user data storage. The primary goals are to reduce losses, lower costs, and make the system more stable and flexible. Among numerous distributed optimization algorithms, the Alternating Direction Multiplier Method (ADMM) has attracted considerable attention due to its excellent parallel processing capabilities and high solution efficiency. However, the standard ADMM algorithm still has some shortcomings in terms of convergence speed and number of iterations, which to some extent limits its practical application in smart microgrids. To overcome these limitations, researchers have proposed several improved ADMM algorithms. These algorithms mainly focus on improving the convergence speed and reducing the number of iterations, but in practical applications, they often neglect the handling of branch current flow equality constraints, which may lead to a decrease in solution accuracy. Furthermore, current research mainly focuses on convex optimization problems, while the handling of non-convex optimization problems remains insufficient. Summary of the Invention

[0007] This invention provides a multi-zone collaborative scheduling method and system for intelligent microgrids based on heavy ball acceleration distributed optimization. It constructs a model that considers branch power flow constraints, uses the second-order cone (SOC) relaxation technique to make the model convex, and applies the ADMM algorithm model with heavy ball acceleration technology to perform real-time power allocation for the intelligent microgrid. The voltage stability and network loss of each node and the convergence of the model are analyzed.

[0008] The methods include:

[0009] S101: Define the minimum network loss objective function for the sum of active power losses in the smart microgrid model, and use it to determine the generator output power;

[0010] S102: Define power flow constraints, active and reactive power constraints of distributed power sources, active and reactive power output constraints of generator sets, reactive power compensation constraints, and per-unit voltage and current constraints.

[0011] S103: Use second-order cone relaxation to process non-convex functions to obtain the constraints required for power flow solution;

[0012] S104: An improved algorithm based on the ADMM distributed algorithm and the heavy ball acceleration technique is used to optimize the control of the smart microgrid model.

[0013] The present invention also provides a smart microgrid multi-zone collaborative scheduling system based on heavy ball acceleration distributed optimization, the system comprising: a microgrid output power determination module, a constraint definition module, a power flow solution constraint definition module, and an optimization control module;

[0014] The microgrid output power determination module is used to define the minimum network loss objective function of the sum of active power losses in the smart microgrid model, thereby determining the generator output power;

[0015] The constraint definition module is used to define power flow constraints, active and reactive power constraints of distributed power sources, active and reactive power output constraints of generator sets, reactive power compensation constraints, and per-unit voltage and current constraints.

[0016] The power flow solution constraint definition module is used to process non-convex functions using second-order cone relaxation to obtain the constraints required for power flow solution.

[0017] The optimization control module improves the algorithm based on the ADMM distributed algorithm and the heavy ball acceleration technology to optimize the control of the smart microgrid model.

[0018] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the steps of the intelligent microgrid multi-zone cooperative scheduling method based on heavy ball acceleration distributed optimization.

[0019] The present invention also provides a readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a multi-zone collaborative scheduling method for intelligent microgrids based on heavy ball acceleration distributed optimization.

[0020] As can be seen from the above technical solutions, the present invention has the following advantages:

[0021] The method provided by this invention uses a distributed optimization algorithm to optimize the control of a smart microgrid. The distributed optimization method utilizes local information to achieve coordinated control, which has a smaller computational load and communication burden compared to the centralized optimization method. It also has higher flexibility and reliability, better privacy protection, and enhances the scalability of the network because it does not rely on a central node.

[0022] This invention constructs a model that considers branch power flow constraints and uses the second-order cone (SOC) relaxation technique to convexify the model, thereby improving the solution accuracy and enabling the ADMM algorithm to solve it directly.

[0023] This invention utilizes the cplex solver on the MATLAB platform to simulate an improved IEEE 33-node intelligent microgrid system, analyzing the voltage stability and network loss of each node, as well as the model convergence. Simulation results show that the improved algorithm can reduce network loss and improve the solution accuracy and convergence speed of the system while maintaining the basic stability of node voltage. Attached Figure Description

[0024] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 The flowchart shows a multi-zone collaborative scheduling method for intelligent microgrids based on heavy ball acceleration distributed optimization.

[0026] Figure 2 A power flow model diagram for a smart microgrid branch;

[0027] Figure 3 Schematic diagram of distributed optimization solution principle;

[0028] Figure 4 Flowchart for solving the ADMM algorithm with the heavy ball term;

[0029] Figure 5 This is a schematic diagram of the specific partitioning of the calculation example;

[0030] Figure 6 To optimize the time-varying network loss diagram;

[0031] Figure 7 To optimize the time-varying network loss diagram;

[0032] Figure 8 To optimize the per-unit value diagram of time-varying node voltages;

[0033] Figure 9 The per-unit value diagram of the optimized time-varying node voltage. Detailed Implementation

[0034] The intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization provided by this invention uses a distributed optimization algorithm to optimize and control the intelligent microgrid, and achieves coordinated control through local information. Compared with centralized optimization methods, the method of this invention has less computation and communication burden, higher flexibility and reliability, better privacy protection, and enhances network scalability because it does not rely on a central node.

[0035] This invention proposes an ADMM algorithm model using heavy ball acceleration technology for real-time optimization scheduling and power allocation of smart microgrids. The SOCP (second-order cone) relaxation technique is used to convexify the optimization problem of the smart microgrid model, improving the solution accuracy and enabling the ADMM algorithm to solve it directly.

[0036] like Figure 1 A flowchart illustrating a preferred embodiment of the intelligent microgrid multi-zone cooperative scheduling method based on heavy-sphere accelerated distributed optimization of the present invention is shown. The method can be applied to one or more electronic devices, which are devices capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Their hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0037] Electronic devices can be any electronic product that allows human-computer interaction, such as personal computers, tablets, smartphones, personal digital assistants (PDAs), and interactive network television (IPTV).

[0038] The networks in which electronic devices are located include, but are not limited to, the Internet, wide area networks, metropolitan area networks, local area networks, and virtual private networks (VPNs).

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] The methods involved in this embodiment include:

[0041] S101: Define the minimum network loss objective function for the sum of active power losses in the smart microgrid model, and use it to determine the generator output power.

[0042] In an exemplary embodiment, the objective function in the smart microgrid model is to minimize network losses, thereby determining the generator's output power. The expression for the minimum network loss objective function is as follows:

[0043] (2-1)

[0044] Indicates the branch resistance. The formula above represents the sum of active power losses in each line, where the branch current squared is represented by the formula.

[0045] S102: Defines power flow constraints, active and reactive power constraints of distributed power sources, active and reactive power output constraints of generator sets, reactive power compensation constraints, and per-unit voltage and current constraints.

[0046] In an exemplary embodiment, the construction of an intelligent microgrid model is a prerequisite for optimizing scheduling using algorithms. The construction of the intelligent microgrid model mainly includes setting the objective function, introducing and processing constraints. The intelligent microgrid model constructed in this embodiment considers multiple constraints, including power flow constraints, active and reactive power output constraints of wind and photovoltaic power generation, active and reactive power output constraints of generator units, upper and lower limit constraints of static var compensators, and upper and lower limit constraints of voltage and current.

[0047] The power flow constraint in this embodiment is:

[0048] (2-2)

[0049] (2-3)

[0050] The above formula indicates that the active power flowing into and out of each node is equal to the reactive power. Inject active power into the nodes. Active power is consumed by node loads; Inject reactive power into the node. The reactive power consumed by the node load These are the active power and reactive power of the branch circuit, respectively. , Let J represent the total active power and reactive power of the set of branches starting from node J. Indicates the active power and reactive power of the branch. This represents the active power and reactive power of a node.

[0051] Optionally, this implementation uses the intelligent microgrid branch power flow model such as Figure 2 As shown.

[0052] If Defined as apparent power, then the branch With node pair The current relationship is as follows:

[0053] (2-4)

[0054] (2-5)

[0055] (2-6)

[0056] Represents the square of the branch voltage These represent the branch voltage and branch current, respectively. Let i and j represent the set of nodes and the set of branches, respectively, where i and j represent nodes and (i, j) represent branches. This indicates the branch circuit reactance.

[0057] in, yes The branch impedance. From 2-4 and 2-6, we can obtain the following formula:

[0058] (2-7)

[0059] (2-8)

[0060] Further simplification yields the following formula:

[0061] (2-9)

[0062] Current constraints

[0063]

[0064] In this embodiment, the distributed power source can be constrained by the active and reactive power of photovoltaic and wind power as follows:

[0065] Photovoltaic active and reactive power constraints

[0066] (2-14)

[0067] (2-15)

[0068] Wind turbine active and reactive power constraints

[0069] (2-16)

[0070] (2-17)

[0071] The active and reactive power output values ​​of distributed power sources should both be between the maximum and minimum values ​​generated by photovoltaic and wind power.

[0072] i This indicates the node where distributed power sources such as photovoltaics and wind turbines are located. , These represent the upper and lower limits of the output active power and the upper and lower limits of the output reactive power of the photovoltaic distributed power source, respectively. These represent the upper and lower limits of the output active power and the upper and lower limits of the output reactive power of wind power distributed generation, respectively.

[0073] The active and reactive power output constraints of the generator set in this embodiment are as follows:

[0074] (2-18)

[0075] (2-19)

[0076] Each generator set has its own generating capacity. To ensure the normal operation of the smart microgrid, the active and reactive power output of the generator set should be within its capacity range. 'i' represents the node where the generator set is located. , These represent the upper and lower limits of the generator set's output active power and output reactive power, respectively.

[0077] Reactive power compensation (SVC) constraints are

[0078] (2-20)

[0079] This embodiment uses a continuous static var compensator (SVC) to ensure that the compensated reactive power is within the SVC compensation capacity range. The reactive power compensation device is located at node i. These represent the upper and lower limits of the static var compensator, respectively.

[0080] The per-unit constraints for voltage and current are:

[0081] (2-21)

[0082] (2-22)

[0083] These represent the upper and lower limits of the square of the voltage, respectively. This represents the upper limit of the square of the current. Excessive current can cause many problems for the circuit. First, it can cause the circuit conductors to overheat and burn. Second, it can damage the insulation performance of the circuit. The current density must be maintained below the standard value. To ensure the safe operation of the smart microgrid, this example uses a constraint that the square of the branch current is less than 10 (the square of the rated line current).

[0084] To ensure the normal operation of the smart microgrid, the voltage at the grid nodes is generally maintained within 0.05 per unit value, i.e., 0.95-1.05. In this embodiment, the square of the voltage is used as the variable, so the voltage magnitude should be between 0.9025 and 1.1025.

[0085] S103: Use second-order cone relaxation to process non-convex functions to obtain the constraints required for power flow solution.

[0086] According to embodiments of this application, the cplex mathematical optimization technique or gurobi can solve the minimum optimization problem of a convex quadratic objective function and the maximum optimization problem of a concave quadratic objective function.

[0087] In this embodiment, the squares of voltage and current are used as variables, which can linearize the distributed optimization problem (objective function). The quadratic equality constraint shown in equation (2-13) will cause the distributed optimization problem to be non-convex, so the SOCP (second-order cone) relaxation technique is used to make it convex.

[0088] It should be noted that second-order cone programming (SOCP) relaxation has been proven to work under the following conditions: (1) the upper voltage bounds of all nodes in the smart microgrid are the same, and (2) the lower power bounds of each node in the smart microgrid are negative. The second-order cone programming relaxation technique can replace the original constraints.

[0089] In this embodiment, second-order cone programming refers to convex optimization with constraints of second-order cones. The general form of second-order cone programming is as follows:

[0090]

[0091]

[0092]

[0093] Optimization parameters include: The optimization variables are: .

[0094] Alternatively, second-order cone programming, as a very special type of nonlinear programming, has many efficient solution algorithms, such as the interior-point method. Therefore, many programming problems can be transformed into second-order cone programming problems for solution.

[0095] In this embodiment, the second-order cone can first be understood as a cone (cone) for a vector space. With a subset C of it, for any point x in subset C and any positive number If the product of all elements in a set C still belongs to a subset C, then C is called a cone. The mathematical definition of a cone is that it is a subset of a set C. The set formed by x is called a cone. It should be noted that a cone is not necessarily continuous; it can be a set of several rays passing through the origin; a cone is always unbounded.

[0096] Secondly, for a convex cone... If C is both a convex set and a cone, then C is a convex cone. In short: a set that is both a convex set and a cone is a convex cone.

[0097] Next, understand the standard cone, specifically the definition of an n-dimensional standard cone:

[0098]

[0099] A standard cone is a convex cone, and the variables include x and t.

[0100] The cone uses the 2-norm, so it is called a second-order cone. Below is an expression for a second-order cone.

[0101]

[0102] A second-order cone can be viewed as a standard cone. Affine transformation was performed:

[0103]

[0104] Below is an example of transforming a quadratic programming problem into a second-order cone programming problem. For the quadratic programming constraints: This can be transformed into a second-order cone constraint:

[0105] (2-23)

[0106]

[0107]

[0108]

[0109]

[0110]

[0111] (2-24)

[0112] Quadratic Equality Constraints This will result in a non-convex smart microgrid model, requiring further expansion of the feasible region to solve the model. The "=" in the equation constraint will be relaxed to "...". "back, It can be rewritten in the form of Second Order Cone Programming (SOCP):

[0113] (2-25)

[0114]

[0115]

[0116] (2-26)

[0117] The cplex or gurobi solvers can directly solve norm-constrained problems; simply add norm constraints to the constraint set.

[0118] Based on the above method, the parameters and variables required for the optimization problem are introduced, and power flow constraints, active and reactive power constraints of distributed sources, active and reactive power output constraints of generator sets, reactive power compensation (SVC) constraints, and voltage and current per-unit constraints are defined. By combining the processing method for the quadratic terms of voltage and current and using second-order cone relaxation to handle non-convex functions, the optimization problem of the smart microgrid model is transformed into a convex optimization problem. Finally, all the constraints required for power flow solution that can be directly solved are obtained.

[0119] S104: An improved algorithm based on the ADMM distributed algorithm and the heavy ball acceleration technique is used to optimize the control of the smart microgrid model.

[0120] In one exemplary embodiment, distributed optimization methods are increasingly being used in smart microgrids due to their superior performance compared to centralized optimization methods. Their superior performance is primarily reflected in avoiding the leakage of user privacy, accelerating computation through distributed multi-machine processing, and enhancing network scalability by not relying on a central node.

[0121] Before performing distributed optimization scheduling, the smart microgrid topology is partitioned. The smart microgrid is divided into m sub-regions. For any region p, its adjacent region q is connected by a line, called a connecting line. The distributed optimization algorithm, as the name suggests, performs optimization scheduling calculations for each region separately. Then, it iterates continuously, taking into account conditions such as consistent current flow at the connecting line, consistent active and reactive power flow, and consistent voltage at the same node, until the error is controlled within a certain range, thus achieving global optimized scheduling of the smart microgrid. A detailed partitioning and connecting line diagram is shown below. Figure 3 As shown.

[0122] In subregion p, the variable of the branch at the junction is the line active power P. p and reactive power Q p and voltage U p Current I p In subregion q, the variables at the junction include active power P. q and reactive power Q q and voltage U q Current I q .

[0123] In this embodiment, the constraints that the independent sub-region p needs to satisfy are:

[0124] (3-1)

[0125] Let be the objective function for subregion p, i.e., the minimum network loss for subregion p. It refers to the equality constraint among all constraints. It refers to inequality constraints among all constraints. This indicates that the combination conditions (boundary conditions, such as active power and reactive power, voltage square and current square) in subregions p and q are equal.

[0126] The Alternating Direction Method of Multipliers (ADMM) in this embodiment is widely used in various fields such as signal processing, image processing, machine learning, and engineering computing due to its advantages of good convergence performance and fast convergence speed.

[0127] This embodiment uses the standard ADMM distributed optimization algorithm, and the Lagrangian function of sub-region p is as follows:

[0128] (3-2)

[0129] Let G be the square of the current in subregion q, and G be the set of all branches connected to subregion p. For the Lagrange multiplier vectors in the iteration, The penalty parameter is used. The solution is performed separately for each independent sub-region. After obtaining the results, iterative calculations are performed using the following steps:

[0130] (3-3)

[0131] In the formula, k represents the iteration number. The Lagrange multiplier vectors of independent subregions are used. and boundary conditions The convergence is determined by the approximation of zero, and the data error is considered. The calculation formula is determined by the Lagrange multiplier vector and the boundary conditions, and is as follows:

[0132] (3-4)

[0133] , These represent the residuals of the Lagrange multipliers and the boundary conditions after the k-th iteration, respectively. The algorithm stops iterating and outputs the optimal solution when the value is less than a certain minimum value 'a'.

[0134] The traditional ADMM algorithm for optimizing the scheduling of smart microgrids consists of the following steps: First, the smart microgrid is partitioned, and the combined branches and parameters are determined. Second, initial values ​​are given for algorithm iteration. Third, each independent sub-region is solved independently using a cplex solver, and the values ​​of intermediate variables during iteration are obtained according to the algorithm formula. Fourth, the system error is calculated using the error formula, and it is determined whether the error is less than the set error. If it is less, the optimized scheduling result is output, and the iteration ends. If it is greater than the set error, the iteration returns to step three and continues until the output error is within the set value.

[0135] This embodiment introduces a heavy-ball term. Heavy-ball acceleration is a technique that can accelerate algorithm convergence. As discussed in the introduction, the application of heavy-ball acceleration in various algorithms has significantly improved convergence performance and greatly reduced the number of iterations. The principle behind heavy-ball acceleration is the introduction of a heavy-ball term during the iteration process.

[0136] This embodiment introduces the heavy ball acceleration technique into the ADMM algorithm. A heavy ball term is introduced into the penalty parameter in the iteration process of the traditional ADMM algorithm to realize a variable step size ADMM algorithm to improve the problems of poor convergence performance, large number of iterations, and long iteration time of the traditional ADMM algorithm under the same accuracy.

[0137] For a multi-node smart microgrid system, since the wind turbine and photovoltaic power generation are different in different time periods, and the size and type of load are also different, the smart microgrid system needs to consider the impact of time variation on itself. At the same time, it can learn the penalty parameters of previous time periods and apply them to new time periods to make the changes of penalty parameters more reasonable, which is beneficial to accelerating the convergence speed of the algorithm.

[0138] Classic heavy-ball acceleration techniques are often used to accelerate the convergence of multidimensional networks. In this embodiment, heavy-ball acceleration is applied to the change of one-dimensional parameters, namely penalty parameters.

[0139] The heavy ball acceleration technique can be seen as an acceleration technique that introduces the momentum term of a heavy ball into the traditional gradient descent method, resulting in a faster convergence speed.

[0140] Soviet mathematician BT Polyak first proposed the heavy ball momentum acceleration technique, which improved the iteration speed of the traditional gradient descent method.

[0141] Traditional heavy-ball momentum acceleration technology is applied to time-varying n-node information networks to make decisions about the current moment based on information obtained from past time periods. Then, the loss function that node i did not know before. Known by node i and for decision A loss is incurred, the magnitude of which is Within each time interval, the node only knows the loss function value of the previous time interval and does not know the loss function value of the subsequent time interval. At time t, the sum of the loss functions of all nodes in the information network is:

[0142] (3-5)

[0143] loss function Known only to node i and not to other nodes, nodes in the network need to exchange information to obtain the global optimization loss function. The total loss function of the entire information network is:

[0144] (3-6)

[0145] T represents the time period. Based on this loss function, the traditional heavy-ball acceleration technique is derived as follows:

[0146] (3-7)

[0147] in and This represents an intermediate variable. The expression can be equivalent to:

[0148] (3-8)

[0149] in, Indicates the learning rate. This represents the coefficient of the momentum term in the heavy ball type, which determines the degree of influence of past gradient changes on current gradient changes.

[0150] This embodiment improves upon the traditional heavy ball acceleration technique. For a time-varying information network with n variables, given the initial values ​​of the variables, it is necessary to determine the values ​​of each variable when the information network is stable.

[0151] set up i Indicates the first i Message i Given a time period T, each time interval is labeled with its starting time, and the iteration process of each variable is also labeled.

[0152] Start time of each time period For each time period t Each variable makes decisions based on information from previous time periods. surface Show time period t Inner i The value of the information at the k-th iteration. For the time period. t The convergence speed of variable values ​​within the current time period can be accelerated by referencing the changes of the same variable during the iteration process. The values ​​of the same variable obtained through iteration within time period t-1 are fitted as a function of the number of iterations; let's assume this function is... .

[0153] The obtained function is used as the function for obtaining the gradient using the gradient descent method. Furthermore, a heavy-ball momentum term is introduced into the gradient descent method, resulting in a new heavy-ball acceleration technique. This is expressed as:

[0154] (3-9)

[0155] in Indicates the first i Messages within a time period t Intermediate variables generated during the (k+1)th iteration. Indicates the first i Messages within a time period t Inner k Intermediate variables generated in the next iteration.

[0156] The expression can be equivalent to:

[0157] (3-10)

[0158] in, Indicates the learning rate. This represents the coefficient of the momentum term in a heavy ball, which determines the degree of influence of information from the past time period on information from the current time period.

[0159] This paper applies the improved heavy ball acceleration technique to the adaptive penalty parameter of the ADMM algorithm during the iteration process, making the penalty parameter more reasonable during the iteration process, thereby accelerating the convergence speed of the algorithm.

[0160] The heavy ball momentum acceleration takes into account the descent direction in the past time. Under the premise that the current descent direction is consistent with the past descent direction, it can accelerate the convergence of the algorithm.

[0161] In this embodiment, to further improve the solution speed of the ADMM algorithm and reduce the number of iterations, a heavy-ball acceleration technique is applied to the ADMM algorithm. This involves changing the penalty parameter during the iteration process of the traditional ADMM algorithm, i.e., introducing an adaptive penalty parameter for the heavy-ball term. This method improves the solution speed of the ADMM algorithm by continuously and dynamically changing the penalty parameter in each iteration.

[0162] This embodiment improves upon the traditional heavy ball acceleration technique, assuming the penalty parameter of the ADMM algorithm is... Given a time period T, each time interval is labeled with its start time, and the iteration process of the penalty parameter is also marked. The start time of each time interval is... For each time period t The penalty parameters are all based on the penalty parameters of the previous time period to make decisions. Indicates time period t The value of the internal penalty parameter at the k-th iteration.

[0163] For time period t The penalty parameter within the time interval can be referenced to the changes of the same variable during the iteration process, which can accelerate the convergence speed of the ADMM algorithm. The value of the penalty parameter obtained from iterations within the time interval t-1 is fitted as a function of the number of iterations; let this function be denoted as . .

[0164] The obtained function is used as the function for obtaining the gradient by the gradient descent method. At the same time, a heavy ball momentum term is introduced on the basis of the gradient descent method to obtain a new heavy ball acceleration technology.

[0165] Represented as:

[0166] (3-11)

[0167] in, This represents the intermediate variable generated by the penalty parameter during the (k+1)th iteration within the time period t. This represents the intermediate variable generated by the penalty parameter during the k-th iteration within the time period t.

[0168] The expression can be equivalent to:

[0169] (3-12)

[0170] in, Indicates the learning rate. This represents the coefficient of the momentum term in a heavy ball, which determines the degree of influence of information from the past time period on information from the current time period.

[0171] Considering the special case, for the first time interval, i.e., t=0, there are no penalty parameters from previous time intervals as a reference; at this time, only the momentum term of the heavy ball exists. The formula is somewhat special:

[0172] (3-13)

[0173] The specific solution steps are as follows: Figure 4 As shown. The improved ADMM algorithm incorporates a heavy-ball acceleration technique, making its iteration process different from the traditional ADMM algorithm. The optimization scheduling solution for smart microgrids consists of the following steps: First, the smart microgrid is partitioned, and the combined branches and parameters are determined. Second, initial values ​​are given for the algorithm iteration. Third, each independent sub-region is solved independently using a cplex solver, and the values ​​of intermediate variables are obtained according to the algorithm formula. Fourth, the system error is calculated using the error formula. Fifth, the penalty parameters are updated using the formula. Sixth, it is determined whether the error is less than the set error. If it is less, the optimized scheduling result is output, and the iteration ends. If it is greater than the set error, it is necessary to return to step three and repeat the iteration until the output error is within the set value.

[0174] Based on the above steps, this embodiment establishes a partitioned optimization model, introduces the abstract principle of applying distributed optimization algorithms to intelligent microgrid optimization scheduling, and introduces the parameters and variables required for distributed optimization algorithms. Then, this embodiment introduces the traditional ADMM algorithm, including the establishment of the augmented Lagrangian function, the iterative steps of the algorithm, and specifies the error calculation method. A flowchart then describes the solution steps and methods of the algorithm. Next, the classic heavy-ball acceleration technique is introduced, its origin and principle are explained, and the principle of the improved heavy-ball acceleration technique is discussed. Then, the improved heavy-ball acceleration technique is applied to the adaptive penalty parameter of the traditional ADMM algorithm to accelerate the convergence speed, and a flowchart is used to design the specific solution process of the improved algorithm.

[0175] In one embodiment of the present invention, based on the above-described intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0176] The simulation examples in this embodiment strive to closely approximate real-world scenarios, using a smart microgrid example with multiple distributed power sources that consider time-varying characteristics. Analysis of the examples reveals the changes in voltage stability, network loss, number of iterations, and convergence speed before and after optimization. This embodiment can run on MATLAB software, using the Cplex solver on the yalmip platform. The algorithm is compiled on an AMD Ryzen 7 4800H processor with 16GB of RAM.

[0177] This embodiment uses an improved IEEE 33-node smart microgrid for simulation calculations. The node diagram of the smart microgrid is as follows: Figure 5 As shown, the area is divided into three zones, each containing two generator units, two distributed photovoltaic power sources, and one distributed wind power source. The voltage reference value is 12.66kV, and the smart microgrid is also equipped with two reactive power compensation devices. The specific zoning is shown in the figure. The specific locations of each power source and device are shown in Table 1.

[0178] Table 1 Power Supply and Device Node Locations

[0179]

[0180] Considering the different power generation of photovoltaic and wind power at different times of the day, as well as the different active and reactive power consumed by the load at different times, the example is a time-varying smart microgrid system, which takes into account the photovoltaic power generation and wind power generation data at different times of the day, as well as the load at different times of each node.

[0181] Table 2. Line parameters for the IEEE 33-node example.

[0182]

[0183] Table 3 Real-time power output of photovoltaic and wind power in the example

[0184]

[0185] The initial preset parameters for the ADMM algorithm are shown in Table 4.

[0186] Table 4 Initial preset parameters of the algorithm

[0187]

[0188] In this embodiment, yalmip is a powerful toolkit for MATLAB, which allows users to solve engineering problems through MATLAB operations and calls. It's a tool for building models; in fact, it can be considered similar to a programming language. Models are built using its language, and then other solvers (such as CPLEX, GROBI, etc.) are called to solve them. Essentially, it acts as a language converter, transforming the "yalmip language" into the "languages" of other solvers.

[0189] Cplex is an optimization tool engine developed by IBM. It is widely used in various fields, including but not limited to electrical engineering and mechanical engineering. This tool can be used to solve quadratic programming, linear programming, integer programming and other problems, and can quickly solve many industry problems.

[0190] The included IBM ILOG Cplex Optimization Studio allows for programming using the built-in language and also provides interfaces for many popular languages, offering broad application prospects.

[0191] like Figure 6 and 7 The figures show the 24-hour time-varying network loss diagrams of the smart microgrid obtained using the traditional ADMM algorithm and the improved ADMM algorithm with the introduction of a heavy ball term, respectively.

[0192] Calculations showed that the average hourly network loss was 56kW before optimization and 32kW after optimization. The optimization algorithm significantly reduced the network loss of the smart microgrid.

[0193] like Figure 8 and Figure 9 The figures show the per-unit voltage values ​​of 33 nodes 24 hours before and after optimization. It can be seen from the figures that the per-unit voltage values ​​of each node remained within the acceptable range before and after optimization, and the voltages were relatively stable.

[0194] Regarding the number of iterations and convergence speed: The traditional ADMM algorithm requires 8 iterations, while the improved ADMM algorithm with a heavy ball term achieves the same error accuracy as the traditional ADMM algorithm after 6 iterations. The traditional ADMM algorithm has an average iteration time of 5.79 seconds. Since this paper uses a distributed algorithm, when solving independently in the three sub-regions, the actual average iteration time is 1.93 seconds, with a total of 8 iterations, so the solution takes 15.44 seconds to converge. The improved ADMM algorithm with a heavy ball term has an average iteration time of 5.94 seconds. Due to the use of a distributed optimization algorithm, the actual average iteration time is 1.98 seconds, with a total of 6 iterations to achieve the same accuracy as the traditional ADMM algorithm, so the solution takes 11.88 seconds.

[0195] Based on the above method, although the average iteration time of the improved ADMM algorithm with the introduction of a heavy ball term is slightly higher than that of the traditional ADMM algorithm, the overall convergence performance of the improved algorithm is higher than that of the traditional algorithm due to the reduction in the number of iterations.

[0196] Thus, this embodiment applied the ADMM distributed algorithm and the improved algorithm using the heavy ball acceleration technique with an improved time-varying IEEE 33-node example. Finally, the example results for both algorithms were obtained, demonstrating that the intelligent microgrid multi-zone cooperative scheduling method based on heavy ball acceleration distributed optimization ensures the stability of system node voltages.

[0197] The following are embodiments of the intelligent microgrid multi-zone collaborative scheduling system based on heavy ball acceleration distributed optimization provided in this disclosure. This system and the intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the intelligent microgrid multi-zone collaborative scheduling system based on heavy ball acceleration distributed optimization, please refer to the embodiments of the intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization described above.

[0198] The system includes: a microgrid output power determination module, a constraint definition module, a power flow solution constraint definition module, and an optimization control module.

[0199] The microgrid output power determination module is used to define the minimum network loss objective function of the sum of active power losses in the smart microgrid model, thereby determining the generator output power.

[0200] The constraint definition module is used to define power flow constraints, active and reactive power constraints of distributed power sources, active and reactive power output constraints of generator sets, reactive power compensation constraints, and per-unit voltage and current constraints.

[0201] The power flow solution constraint definition module is used to process non-convex functions using second-order cone relaxation to obtain the constraints required for power flow solution.

[0202] The optimization control module improves the algorithm based on the ADMM distributed algorithm and the heavy ball acceleration technology to optimize the control of the smart microgrid model; the optimization control parameters include network loss, convergence speed, and number of iterations.

[0203] The intelligent microgrid multi-zone cooperative scheduling system based on heavy-ball accelerated distributed optimization disclosed herein comprises the units and algorithm steps of various examples described in conjunction with the embodiments disclosed herein. These can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0204] The readable storage medium of the present invention stores a program product capable of implementing the intelligent microgrid multi-zone cooperative scheduling method based on heavy ball acceleration distributed optimization described above in this specification. In some possible embodiments, various aspects of this disclosure can also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps according to the various exemplary embodiments of this disclosure described in the "Exemplary Methods" section above.

[0205] The readable storage medium of the present invention can be any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0206] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-zone collaborative scheduling method for intelligent microgrids based on heavy-ball accelerated distributed optimization, characterized in that, The methods include: S101: Define the minimum network loss objective function for the sum of active power losses in the smart microgrid model, and use it to determine the generator output power; S102: Define power flow constraints, active and reactive power constraints of distributed power sources, active and reactive power output constraints of generator sets, reactive power compensation constraints, and per-unit voltage and current constraints. S103: Use second-order cone relaxation to process non-convex functions to obtain the constraints required for power flow solution; S104: An improved algorithm based on ADMM distributed algorithm and heavy ball acceleration technology is used to optimize the control of the smart microgrid model; The ADMM distributed algorithm in step S104 adopts the standard ADMM distributed optimization algorithm, and the Lagrangian function of the sub-region p is defined as follows: Let G be the square of the current in subregion q, and G be the set of all branches connected to subregion p. For the Lagrange multiplier vectors in the iteration, For penalty parameters; Each independent sub-region is solved separately. After obtaining the results, iterative calculations are performed using the following steps: In the formula, k is the number of iterations; Lagrange multiplier vectors of independent subregions and boundary conditions The convergence is determined by the approximation of zero, and the data error is considered. The calculation formula is determined by the Lagrange multiplier vector and the boundary conditions, and is as follows: , These represent the residuals of the Lagrange multipliers and the boundary conditions after the k-th iteration, respectively. The algorithm stops iterating and outputs the optimal solution when the value is less than a certain minimum value 'a'. The improved algorithm for referencing the heavy ball acceleration technique in step S104 includes: Let the penalty parameter of the ADMM distributed algorithm be... Given a time period T, each time period is labeled with its starting time, and the iteration process of the penalty parameter is also marked. Start time of each time period The penalty parameter for each time period t is determined based on the penalty parameters of the previous time periods. This represents the value of the penalty parameter in the k-th iteration within the time period t; The heavy ball acceleration technology is represented as: in, This represents the intermediate variable generated by the penalty parameter during the (k+1)th iteration in time period t; This represents the intermediate variable generated by the penalty parameter during the (k-1)th iteration within the time period t; This represents the intermediate variable generated by the penalty parameter during the k-th iteration within the time period t; Indicates the learning rate. This represents the coefficient of the momentum term for a heavy ball.

2. The intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization according to claim 1, characterized in that, Step S101 also includes: The objective function expression for minimizing the sum of active power losses is: (2-1) Indicates the branch resistance. This represents the square of the branch current.

3. The intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization according to claim 1, characterized in that, The power flow constraint is: Inject active power into the nodes. Active power is consumed by node loads; Inject reactive power into the node. The reactive power consumed by the node load These are the active power and reactive power of the branch circuit, respectively. , Let them represent the total active power and total reactive power of the set of branches with node j as the first end, respectively; The following method represents the active and reactive power constraints of distributed generation sources; Photovoltaic active and reactive power constraints Wind turbine active and reactive power constraints i This indicates the node where distributed power sources such as photovoltaics and wind turbines are located; , These represent the upper and lower limits of the output active power and the upper and lower limits of the output reactive power of the photovoltaic distributed power source, respectively. These represent the upper and lower limits of the output active power and the upper and lower limits of the output reactive power of wind power distributed generation, respectively. The active and reactive power output constraint methods for generator sets are as follows: i Indicates the node where the generator set is located; , These represent the upper and lower limits of the generator set's output active power and output reactive power, respectively. The reactive power compensation constraint method is as follows: These represent the upper and lower limits of the static var compensator; The per-unit constraint method for voltage and current is as follows: These represent the upper and lower limits of the square of the voltage, respectively. This represents the upper limit of the square of the current.

4. The intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization according to claim 1, characterized in that, The second-order cone relaxation form in step S103 is as follows: 。 5. The intelligent microgrid multi-zone cooperative scheduling method based on heavy ball acceleration distributed optimization according to claim 1, characterized in that, Step S104 further includes: The topology of the smart microgrid is partitioned; the smart microgrid model is divided into m sub-regions, and for any region p, it is connected to its adjacent region q by a combined line. Based on the conditions that the current flowing through the combined line is consistent, the active power and reactive power flowing through it are consistent, and the voltage at the same node is the same, the model is continuously iterated to keep the error of the smart microgrid model within a preset range, so as to achieve global optimization scheduling of the smart microgrid model. Step S104 further includes: the constraint satisfied by region p is: Let be the objective function for subregion p, i.e., the minimum network loss for subregion p. It refers to the equality constraint among all constraints. It refers to inequality constraints among all constraints. This indicates that the associativity conditions in subregions p and q are equal.

6. A smart microgrid multi-zone collaborative scheduling system based on heavy ball acceleration distributed optimization, characterized in that, The system is used to implement the intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization as described in any one of claims 1 to 5; The system includes: a microgrid output power determination module, a constraint definition module, a power flow solution constraint definition module, and an optimization control module; The microgrid output power determination module is used to define the minimum network loss objective function of the sum of active power losses in the smart microgrid model, thereby determining the generator output power; The constraint definition module is used to define power flow constraints, active and reactive power constraints of distributed power sources, active and reactive power output constraints of generator sets, reactive power compensation constraints, and per-unit voltage and current constraints. The power flow solution constraint definition module is used to process non-convex functions using second-order cone relaxation to obtain the constraints required for power flow solution. The optimization control module improves the algorithm based on the ADMM distributed algorithm and the heavy ball acceleration technology to optimize the control of the smart microgrid model.

7. An electronic device 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 the steps of the intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization as described in any one of claims 1 to 5.

8. A readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent microgrid multi-zone collaborative scheduling method based on heavy ball acceleration distributed optimization as described in any one of claims 1 to 5.

Images