A non-directional and directional fusion ACUC unit start-stop optimization method

Abstract

The present invention discloses a non-directional and directional fusion ACUC unit start-stop optimization method. Aiming at the unit combination problem under AC power flow constraints, the existing non-directional optimization algorithm is improved, and the search is guided by a priori physical rules. The basic process includes: using evolutionary algorithms to make preliminary decisions on unit start-stop variables; judging whether the system meets the conditions for starting directional optimization; if satisfied, screening the problem nodes in the system; according to the electrical distance of the nodes, directional start of the units near the above-mentioned problem nodes and updating the population; returning to the large cycle to continue iteration according to the constraint violation; ending the directional optimization stage. On the basis of the existing evolutionary algorithm, the present invention adds a directional population update strategy, which can significantly improve the speed at which the total constraint violation of large-scale power systems in terms of power balance, node voltage, and line power flow converges to zero, solves the problem of system convergence difficulties due to continuous constraint violations in some small areas, and improves economic dispatch efficiency.

Description

Technical Field

[0001] The present invention relates to the field of power system optimization and dispatching, and in particular to a non-directional and directional integrated ACUC unit start-stop optimization method. Background Art

[0002] In the dispatching and operation of modern power systems, the unit commitment problem (UC) has always been a complex optimization problem. Since its mathematical model contains a large number of integer variables and nonlinear constraints, direct solution has extremely high computational difficulty. Therefore, traditional methods often use the unit commitment under DC power flow constraints (DC constrained Unit Commitment, DCUC) to conditionally simplify the problem. However, the DC power flow model ignores factors such as voltage amplitude, reactive power, and network loss, and cannot accurately reflect the operating characteristics of the actual power system. Its optimization results often need to be verified afterwards in actual applications, which is difficult to meet the needs of safe and economical operation of large-scale power systems.

[0003] The AC constrained Unit Commitment (ACUC) problem takes into account the influence of active and reactive power flows, node voltage amplitude constraints, and line transmission capacity constraints, and is closer to actual operating conditions. However, as a non-convex mixed integer nonlinear programming (MINLP) problem, it is an NP-hard problem, which also brings higher difficulty in solving. Therefore, studying efficient optimization algorithms to solve ACUC problems is of great practical significance for improving the operating efficiency and economy of power systems.

[0004] Traditional non-directional optimization algorithms, such as genetic algorithms and sparse evolutionary algorithms that introduce sparse initialization methods, can handle complex scheduling problems to a certain extent, but due to their long calculation time and slow convergence of objective functions, they have become bottlenecks in large-scale power system scheduling. One of the problems is that there is a large amount of waste of iterations before the algorithm finds a feasible solution. The significance of a feasible solution is that the penalty coefficient for violating the system constraints is high enough. Only when the values ​​of the unit start and stop variables make the optimal power flow problem (OPF) have a solution that does not violate the constraints, the system has physical feasibility, and the economic dispatch that minimizes the total operating cost of the system can proceed normally on this basis. Therefore, it is necessary to find a set of feasible solutions for unit combinations that meet the system constraints as soon as possible during the algorithm implementation process.

[0005] On the other hand, for the situation where the unit combination scheduling scheme is applied to the system and causes violation of system constraints, in order to facilitate quantitative description, the total system constraint violation degree VC can be defined by comprehensively considering system power balance, node voltage and line flow. Constraint violation has a high penalty coefficient and is optimized by the algorithm first. In the evolutionary algorithm, VC decreases with the increase of population iteration number, and the rate of change of VC's convergence descent curve also gradually decreases. Especially for large-scale systems, when VC drops to a certain value, there is a non-convergence interval with a slow rate of change. The reason is that the generator sets that should be turned on near some specific small areas in the system are not turned on in the original scheduling plan, and power needs to be transmitted from a longer distance. Considering the limitations of the actual system network topology, network constraint violations are caused. The performance is that the net input power of the local area node is less than the load power required by the node. At the same time, the voltage of the internal node of the area is lower than its voltage lower limit, and the voltage of the peripheral node is forced to rise. Even if it is higher than its voltage upper limit, it still cannot meet the power demand of the local area, causing the system to have voltage imbalance or line power overload problems in these areas, so that the overall system constraints do not converge. At this time, if there is no effective measure to solve the problem in a targeted manner, such as relying only on non-directional random search and natural selection mechanism for iteration, without considering the essential cause of the problem node constraint violation and the fundamental solution, that is, starting the unit in a targeted manner based on the electrical distance, then the system constraint violation VC will be maintained at the constraint violation VC output relative to the first iteration for a long time. （1） The constraint violation in the local area is solved only when the above-mentioned specific unit scheduling is turned on due to the randomness of the mutation in the evolutionary algorithm. Therefore, the traditional non-directional optimization method often fails to converge effectively in multiple iterations of solving ACUC of large-scale power systems, or requires a large amount of computing resources, resulting in unsatisfactory optimization results. Summary of the invention

[0006] The present invention aims to solve the technical problems that, although the existing evolutionary algorithm can perform global search when dealing with the unit combination problem under the AC power flow constraints of large-scale power systems, the algorithm is inefficient when the system constraint violation is close to convergence due to the lack of utilization of the specific structure of the problem. In particular, when the system has a power imbalance problem in a small local area, a large number of iterations are required to find a feasible solution, resulting in convergence difficulties, waste of computing resources and difficulty in eliminating local constraints.

[0007] Global search is to explore the global solution space through random search and natural selection mechanisms.

[0008] In order to solve the above problems, the present invention provides a non-directional and directional fusion ACUC unit start-stop optimization method, which comprises the following steps:

[0009] Step A: Initialize the system and use the evolutionary algorithm to make preliminary decisions on the start and stop variables of the unit;

[0010] Step B. After each population iteration, the constraint violation degree VC of the system is calculated. The constraint violation degree VC is the weighted sum of the power imbalance value, the voltage over-limit value and the line power flow over-limit value. If VC drops below 5% of its first iteration output value and VC≠0, the directional optimization stage is entered;

[0011] Step C. For each period of the original scheduling plan, detect the power inflow and outflow of each node, and filter out the problem nodes whose net input power is less than the load power;

[0012] Step D. For each problem node, retrieve the electrical distance between the problem node and the units that are not started in this scheduling plan based on the node impedance matrix, start the nearby units in ascending order of electrical distance, and repair the violations of the timing constraints of the unit start and stop, including the minimum start time and the minimum shutdown time; the number of units to be started needs to meet the condition that the sum of the maximum active output power of these units is not less than the difference between the node net input power and the load power calculated in the previous step;

[0013] Step E. After the directional adjustment, recalculate the constraint violation degree VC. If VC≠0, return the adjusted unit start-stop plan to the iteration, repeat steps B to D until VC=0, end the directional optimization process, and optimize the economic dispatch on this basis.

[0014] In step A, the system data is initialized including: system topology; type, capacity and start / stop status of generators; load demand; physical constraints such as line flow constraints and node voltage limits; and calculation of node impedance matrix. The purpose of calculating the node impedance matrix is ​​to facilitate the subsequent steps to retrieve the mutual impedance elements between nodes to characterize the electrical distance.

[0015] In step A, the non-directional optimization algorithm is an evolutionary algorithm, such as a genetic algorithm (GA) or a sparse evolutionary algorithm (Sparse EA), wherein the sparse evolutionary algorithm refers to an algorithm that introduces sparse initialization technology on the basis of a traditional evolutionary algorithm to obtain better initial start and stop variables. It is called non-directional because its optimization process mainly relies on random search and natural selection mechanisms, rather than actively using problem-specific structures or prior knowledge to guide the search direction, which is different from the directional optimization method proposed in the present invention.

[0016] In step A, the relevant parameters of the non-directional optimization algorithm, i.e., the genetic algorithm or the sparse evolutionary algorithm, including the initial population size, the upper limit of the number of iterations, etc., are initialized, and the initial unit start and stop state is set, and then any of the above algorithms is used to perform population iteration on the population composed of the initialized unit start and stop state variables.

[0017] In step B, the constraint violation of the system is calculated after each population iteration using the evolutionary algorithm. If VC drops to the constraint violation VC output in the first iteration, （1） If VC≠0, the system meets the conditions for entering the directional optimization stage. The reason why VC drops to the constraint violation VC output of the first iteration is （1） The directional optimization stage is entered only when the VC value is less than 5%. This is because the problem that the present invention focuses on solving is that the generator sets that should be turned on near the small area in the large-scale system are not turned on in the original scheduling plan, resulting in the continuous non-convergence of VC. It is necessary to enter the directional optimization stage when the VC is close to convergence to be targeted; on the contrary, if the directional optimization stage is entered when the VC value is high, it does not meet the problem to be solved by the directional optimization method because of the widespread constraint violations in the system. Therefore, it is not meaningful to enter the directional optimization stage at this time.

[0018] In step B, although the problem addressed by the present invention is based on ACUC, active power is a direct measure of energy consumption, and the active power constraint is focused on. In addition, the reactive power constraint is implicitly processed when solving the optimal power flow (OPF) problem by using the interior point method (IPM). Therefore, the constraint violation degree VC is calculated by combining the following three factors, specifically including: In order to examine the degree of satisfaction of the power balance constraint, the difference between the total output power of the system units and the total load power is calculated, and the system is obtained in the time period The power imbalance value ; To examine the satisfaction degree of node voltage constraints, check whether the node voltage exceeds the set limit, and obtain the system in the period The voltage exceeds the limit ; To examine the satisfaction degree of line flow constraints, check whether the line flow exceeds the rated capacity, and obtain the system The line power flow exceeds the limit The calculation formula of constraint violation degree VC is:

[0019] ,

[0020] ,

[0021] ,

[0022] ,

[0023] Among them, T represents the 24 time periods of the day-ahead scheduling plan, N represents the node set of the system, Indicates that the system is in the period The power imbalance value, Indicates that the system is in the period The voltage exceeds the limit value, Indicates the system time period α1 represents the weight coefficient of the power imbalance value, α2 represents the weight coefficient of the voltage excess value, and α3 represents the weight coefficient of the line flow excess value. Since the power imbalance value, the voltage excess value, and the line flow excess value have different physical dimensions, the per-unit value is used for normalization to make the violation degrees comparable. The weight coefficients α1, α2, and α3 can be set according to the actual situation to ensure that the influence of different constraints on the total violation degree is reasonable. If there is no special requirement, they are set to equal weights, that is, α1=α2=α3=1. ; Indicates the system The units at the nodes are in the period The power generation capacity, Indicates the system Nodes in the time period The load power, Indicates that the system is in the period The network loss, Indicates that the system is in the period The system spinning reserve power is 10% of the sum of the total system load power and network loss; Indicates the system Nodes in the period The voltage, Indicates the system The voltage upper limit of each node, Indicates the system The lower voltage limit of each node, Indicates the system The voltage rating of each node; Indicates the system Lines in the period The meritorious trend, Indicates the system The upper limit of active power flow of the line, Indicates the system Active power flow rating of the line.

[0024] In step C, the program calculates the difference between the net input power and the load power of each node in each period of the original scheduling plan. The calculation formula is: ,in, Indicates Nodes in the period The difference between the node net input power and the load power, Indicates Nodes in the period The net input power, Indicates Nodes in the period Furthermore, the program screens out the problem nodes in the scheduling plan for each period of the day, and the screening conditions are: <0, then Nodes in the period It is a problem node when the net input power of the node is less than the load power required by the node.

[0025] In step D, the goal of the program is to adjust the start and stop of the units in a targeted manner in the areas near the problem nodes detected in the previous step, so as to solve the problem of unreasonable power flow distribution in the system caused by the fact that the generators in these local areas were not started in the original scheduling plan, and thus caused the system constraint violation problem. The process of targeted start-up of the units includes:

[0026] Based on the system topology and power flow path, the electrical distances between these problematic nodes and the unstarted units in the system are retrieved:

[0027] ,

[0028] in, is the node in the system node impedance matrix With Node The mutual impedance elements between Represented as a node With Node The correction coefficient between them is set to 1 by default. Since the electrical distance is used as a qualitative indicator in the present invention to locate the area near the problem node, the correction coefficient can be set to 1. When the algorithm is actually running, since the electrical distance can be approximately considered to be only related to the system parameters, and the node impedance matrix has been calculated when the system is initialized, this step retrieves the mutual impedance elements between nodes through the node impedance matrix, and then characterizes the electrical distance between nodes.

[0029] Furthermore, the units with a closer electrical distance to the problem node are preferentially turned on, thereby adjusting the local power distribution in order to eliminate the local constraint violation. Specifically, a series of units with electrical distances from the problem node from near to far are retrieved in ascending order of electrical distance, and the program turns on these units in sequence until the sum of the maximum active output power of the series of units turned on is not less than the difference between the node net input power and the load power calculated in the previous step. , that is, the series of units that are turned on must at least have the Nodes in the period Necessary conditions for node power shortfall.

[0030] Furthermore, according to the adjusted start and stop status of the unit, the timing constraint violations of the new scheduling plan are repaired so that the scheduling plans of the same unit in each period meet the operating restrictions of the minimum start and stop time, including the minimum opening time MUT and the minimum closing time MDT. During the repair, the variable values ​​​​directedly modified in the previous step are kept unchanged. Specifically, for each unit that is directionally turned on, after the unit start and stop variables are directed to be modified, if the operating restrictions of the minimum start and stop time between the scheduling plans of the unit in each period have been met, there is no need to repair its timing constraints; if the operating restrictions of the minimum start and stop time between the scheduling plans of the unit in each period are not met, it is necessary to repair its timing constraints. Taking the time period where the start and stop variables are directed to be modified as the center, the time periods on both sides are searched in turn, and the units are moved to the same time period in turn. The time periods that were not turned on in the original scheduling plan of the unit are changed to turned on until the time periods that were turned on in the original scheduling plan of the unit are searched. At this time, the start and stop variables of the unit in the search range on both sides of the central time period are in the turned-on state. Since the original scheduling plan must meet the timing constraints to appear in the algorithm's population, the time periods after the central time period must also meet the timing constraints. However, the time periods before the central time period may not meet the timing constraints because the scheduling plan of the previous day cannot be modified. If the scheduling plan still does not meet the timing constraints after the start and stop status of the time periods before the central time period is modified, the unit will be canceled, and the next unit will be turned on in the order of the units retrieved from the nearest to the farthest electrical distance, and the above repair steps will be repeated until the principle is met: a series of turned-on units have the ability to make up for the first Nodes in the period Necessary conditions for node power difference.

[0031] In step E, the directional optimization result of step D is used as the new population, and the system constraint violation VC is recalculated. If VC=0, the directional optimization stage is ended directly; if VC≠0, the new population returns to the genetic algorithm or sparse evolutionary algorithm, and continues to iterate in a large loop until the system constraint violation is reduced to 0, and the directional optimization process is ended. Since the system constraint violation is close to convergence when the directional optimization program is started, the number of iterations required in this stage is much smaller than the number of iterations required by the original non-directional method.

[0032] After the directed optimization process is completed, the system has preliminary physical feasibility. On this basis, the evolutionary algorithm is continued to be used to optimize economic scheduling and minimize the total operating cost of the system. The decision variables include: the switching state of the unit, the active power output of the unit, the reactive power output of the unit, and the node voltage amplitude. When the maximum number of iterations is reached, the algorithm selects the individual with the highest fitness from the current population as the final solution and outputs the final decision variables.

[0033] Beneficial effects: The present invention appropriately integrates the characteristics of the physical system with the existing evolutionary algorithm through reasonable analysis of electrical distance and directional adjustment of the start-stop state of the unit, adds directional "navigation" to the evolutionary algorithm, and uses a priori physical rules to guide the search, which can effectively eliminate the local small area constraint violation of the large-scale power grid, improve the efficiency of the unit start-stop scheduling, and is suitable for large-scale and complex topological systems. The unit combination problem under the AC power flow constraint of the large-scale power system is solved. On the basis of non-directional optimization methods such as genetic algorithms and sparse evolutionary algorithms that improve the initialization method of traditional evolutionary algorithms, the directional optimization strategy is combined to improve the system convergence speed and reduce the constraint violation degree, thereby realizing efficient power system scheduling. The present invention mainly solves the problem that the VC in the random search of the traditional evolutionary algorithm does not converge for a long time due to insufficient power supply in the local area of ​​the large-scale power system. This scheme quickly fills the power gap by directional starting of the unit and reduces the waste of computing resources by dynamic stage switching. Compared with the prior art, the present invention integrates non-directional and directional strategies, combines the global search capability of the evolutionary algorithm and the local directional adjustment guided by physical rules, breaks through the randomness limitations of the traditional non-directional algorithm, and significantly improves the convergence speed. Secondly, in the dynamic switching optimization stage, directional optimization is started only when the constraints are close to convergence to avoid premature local optimization that leads to a decrease in the quality of the global solution. The problem nodes are precisely located, and the start-stop strategy of the unit is adjusted based on the electrical distance to effectively solve the local power imbalance problem and reduce network constraint violations. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a flow chart of the optimization method proposed in the present invention. DETAILED DESCRIPTION

[0035] The specific implementation of the present invention will be further described in conjunction with the accompanying drawings of the specification, as well as an IEEE-118 node system and a 354-bus power system obtained by merging three IEEE-118 node systems as examples.

[0036] like Figure 1 As shown, the present invention is a non-directional and directional fusion ACUC unit start-stop optimization method, the method comprising the following steps:

[0037] Step A: Initialize the system and use the evolutionary algorithm to make preliminary decisions on the start and stop variables of the unit;

[0038] Step B. After each population iteration, the constraint violation degree VC of the system is calculated. The constraint violation degree VC is the weighted sum of the power imbalance value, the voltage over-limit value and the line power flow over-limit value. If VC drops below 5% of its first iteration output value and VC≠0, the directional optimization stage is entered;

[0039] Step C. For each period of the original scheduling plan, detect the power inflow and outflow of each node, and filter out the problem nodes whose net input power is less than the load power;

[0040] Step D. For each problem node, retrieve the electrical distance between the problem node and the units that are not started in this scheduling plan based on the node impedance matrix, start the nearby units in ascending order of electrical distance, and repair the violations of the timing constraints of the unit start and stop, including the minimum start time and the minimum shutdown time; the number of units to be started needs to meet the condition that the sum of the maximum active output power of these units is not less than the difference between the node net input power and the load power calculated in the previous step;

[0041] Step E. After the directional adjustment, recalculate the constraint violation degree VC. If VC≠0, return the adjusted unit start-stop plan to the iteration, repeat steps B to D until VC=0, end the directional optimization process, and optimize the economic dispatch on this basis.

[0042] In step A, the system data is initialized including: system topology; type, capacity and start / stop status of generators; load demand; physical constraints such as line flow constraints and node voltage limits; and calculation of node impedance matrix. The purpose of calculating the node impedance matrix is ​​to facilitate the subsequent steps to retrieve the mutual impedance elements between nodes to characterize the electrical distance.

[0043] In step A, the non-directional optimization algorithm is an evolutionary algorithm, such as a genetic algorithm (GA) or a sparse evolutionary algorithm (Sparse EA), wherein the sparse evolutionary algorithm refers to an algorithm that introduces sparse initialization technology on the basis of a traditional evolutionary algorithm to obtain better initial start and stop variables. It is called non-directional because its optimization process mainly relies on random search and natural selection mechanisms, rather than actively using problem-specific structures or prior knowledge to guide the search direction, which is different from the directional optimization method proposed in the present invention.

[0044] Step A of the present invention initializes the power system data (topology, unit parameters, load demand, etc.), and generates the initial unit start-stop plan using a non-directional optimization algorithm (such as a genetic algorithm or a sparse evolutionary algorithm). This stage explores the global solution space through random search and natural selection mechanisms, providing a basis for subsequent optimization.

[0045] In step A, the relevant parameters of the non-directional optimization algorithm, i.e., the genetic algorithm or the sparse evolutionary algorithm, including the initial population size, the upper limit of the number of iterations, etc., are initialized, and the initial unit start and stop state is set, and then any of the above algorithms is used to perform population iteration on the population composed of the initialized unit start and stop state variables.

[0046] In step B, the constraint violation of the system is calculated after each population iteration using the evolutionary algorithm. If VC drops to the constraint violation VC output in the first iteration, （1） If VC≠0, the system meets the conditions for entering the directional optimization stage. The reason why VC drops to the constraint violation VC output of the first iteration is （1） The directional optimization stage is entered only when the VC value is less than 5%. This is because the problem that the present invention focuses on solving is that the generator sets that should be turned on near the small area in the large-scale system are not turned on in the original scheduling plan, resulting in the continuous non-convergence of VC. It is necessary to enter the directional optimization stage when the VC is close to convergence to be targeted; on the contrary, if the directional optimization stage is entered when the VC value is high, it does not meet the problem to be solved by the directional optimization method because of the widespread constraint violations in the system. Therefore, it is not meaningful to enter the directional optimization stage at this time.

[0047] Step B of the present invention calculates the total system constraint violation (including power balance, voltage overrun, and line flow overrun) to determine whether the conditions for entering directional optimization are met (VC drops below 5% of the first iteration value and VC≠0).

[0048] Through dynamic threshold control, directional optimization is started only when the constraints are close to convergence, avoiding inefficient adjustments in the stage of global constraint confusion and reducing the waste of computing resources.

[0049] In step B, although the problem addressed by the present invention is based on ACUC, active power is a direct measure of energy consumption, and the active power constraint is focused on. In addition, the reactive power constraint is implicitly processed when solving the optimal power flow (OPF) problem by using the interior point method (IPM). Therefore, the constraint violation degree VC is calculated by combining the following three factors, specifically including: In order to examine the degree of satisfaction of the power balance constraint, the difference between the total output power of the system units and the total load power is calculated, and the system is obtained in the time period The power imbalance value ; To examine the satisfaction degree of node voltage constraints, check whether the node voltage exceeds the set limit, and obtain the system in the period The voltage exceeds the limit ; To examine the satisfaction degree of line flow constraints, check whether the line flow exceeds the rated capacity, and obtain the system The line power flow exceeds the limit The calculation formula of constraint violation degree VC is:

[0050] ,

[0051] ,

[0052] ,

[0053] ,

[0054] Among them, T represents the 24 time periods of the day-ahead scheduling plan, N represents the node set of the system, Indicates that the system is in the period The power imbalance value, Indicates that the system is in the period The voltage exceeds the limit value, Indicates the system time period α1 represents the weight coefficient of the power imbalance value, α2 represents the weight coefficient of the voltage excess value, and α3 represents the weight coefficient of the line flow excess value. Since the power imbalance value, the voltage excess value, and the line flow excess value have different physical dimensions, the per-unit value is used for normalization to make the violation degrees comparable. The weight coefficients α1, α2, and α3 can be set according to the actual situation to ensure that the influence of different constraints on the total violation degree is reasonable. If there is no special requirement, they are set to equal weights, that is, α1=α2=α3=1. ; Indicates the system The units at the nodes are in the period The power generation capacity, Indicates the system Nodes in the time period The load power, Indicates that the system is in the period The network loss, Indicates that the system is in the period The system spinning reserve power is 10% of the sum of the total system load power and network loss; Indicates the system Nodes in the period The voltage, Indicates the system The voltage upper limit of each node, Indicates the system The lower voltage limit of each node, Indicates the system The voltage rating of each node; Indicates the system Lines in the period The meritorious trend, Indicates the system The upper limit of active power flow of the line, Indicates the system Active power flow rating of the line.

[0055] In step C, the program calculates the difference between the net input power and the load power of each node in each period of the original scheduling plan. The calculation formula is: ,in, Indicates Nodes in the period The difference between the node net input power and the load power, Indicates Nodes in the period The net input power, Indicates Nodes in the period Furthermore, the program screens out the problem nodes in the scheduling plan for each period of the day, and the screening conditions are: <0, then Nodes in the period It is a problem node when the net input power of the node is less than the load power required by the node.

[0056] Since the key nodes that cause constraint violations can be quickly identified in large-scale systems, the node net input power difference analysis ( ), combined with time period dimension screening, to ensure the precise spatial and temporal positioning of the problem node.

[0057] Step C of the present invention is to calculate the difference between the net input power and the load power of each node for each time period, and screen out nodes with insufficient net input power (i.e., problem nodes). This can accurately locate the local power gap area, clarify the optimization target, and avoid global invalid adjustments.

[0058] In step D, the goal of the program is to adjust the start and stop of the units in a targeted manner in the areas near the problem nodes detected in the previous step, so as to solve the problem of unreasonable system power flow distribution caused by the fact that the generators in these local areas were not started in the original scheduling plan, thereby causing the system constraint violation problem.

[0059] The process of directional start-up of the unit includes:

[0060] Based on the system topology and power flow path, the electrical distances between these problematic nodes and the unstarted units in the system are retrieved: ,in, is the node in the system node impedance matrix With Node The mutual impedance elements between Represented as a node With Node The correction coefficient between them is set to 1 by default. Since the electrical distance is used as a qualitative indicator in the present invention to locate the area near the problem node, the correction coefficient can be set to 1. When the algorithm is actually running, since the electrical distance can be approximately considered to be only related to the system parameters, and the node impedance matrix has been calculated when the system is initialized, this step retrieves the mutual impedance elements between nodes through the node impedance matrix, and then characterizes the electrical distance between nodes.

[0061] Since the traditional topological distance cannot well reflect the power transmission efficiency, it is necessary to quantify the electrical coupling strength between nodes. Therefore, the present invention is based on the mutual impedance element of the node impedance matrix ( ) defines the electrical distance ( ), focusing on key areas.

[0062] Furthermore, the units with a closer electrical distance to the problem node are preferentially turned on, thereby adjusting the local power distribution in order to eliminate the local constraint violation. Specifically, a series of units with electrical distances from the problem node from near to far are retrieved in ascending order of electrical distance, and the program turns on these units in sequence until the sum of the maximum active output power of the series of units turned on is not less than the difference between the node net input power and the load power calculated in the previous step. , that is, the series of units that are turned on must at least have the Nodes in the period Necessary conditions for node power shortfall.

[0063] Furthermore, according to the adjusted start and stop status of the unit, the timing constraint violations of the new scheduling plan are repaired so that the scheduling plans of the same unit in each period meet the operating restrictions of the minimum start and stop time, including the minimum opening time MUT and the minimum closing time MDT. During the repair, the variable values ​​​​directedly modified in the previous step are kept unchanged. Specifically, for each unit that is directionally turned on, after the unit start and stop variables are directed to be modified, if the operating restrictions of the minimum start and stop time between the scheduling plans of the unit in each period have been met, there is no need to repair its timing constraints; if the operating restrictions of the minimum start and stop time between the scheduling plans of the unit in each period are not met, it is necessary to repair its timing constraints. Taking the time period where the start and stop variables are directed to be modified as the center, the time periods on both sides are searched in turn, and the units are moved to the same time period in turn. The time periods that were not turned on in the original scheduling plan of the unit are changed to turned on until the time periods that were turned on in the original scheduling plan of the unit are searched. At this time, the start and stop variables of the unit in the search range on both sides of the central time period are in the turned-on state. Since the original scheduling plan must meet the timing constraints to appear in the algorithm's population, the time periods after the central time period must also meet the timing constraints. However, the time periods before the central time period may not meet the timing constraints because the scheduling plan of the previous day cannot be modified. If the scheduling plan still does not meet the timing constraints after the start and stop status of the time periods before the central time period is modified, the unit will be canceled, and the next unit will be turned on in the order of the units retrieved from the nearest to the farthest electrical distance, and the above repair steps will be repeated until the principle is met: a series of turned-on units have the ability to make up for the first Nodes in the period Necessary conditions for node power difference.

[0064] Table 1 Directional algorithm modification table

[0065]

[0066] As shown in Table 1, in the upper part of the table, the day-ahead scheduling plan for the i-th unit lists 24 time periods, and also lists the scheduling plan for the last 8 time periods of the previous day for the unit, where the previous scheduling plan cannot be changed. In each time period, 0 and 1 represent the start / stop status, 0 represents shutdown, and 1 represents start. In the original plan, a problem node is found in time period t, which is bolded in the table. Near the problem node, it is found that the i-th unit in this time period is in the shutdown state, and then the start / stop plan of the unit in this time period is modified in a targeted manner, and the targeted modification is made to the time t of the unit. The start-stop state of the segment is changed to open, and then it is determined whether the scheduling plan violates MUT (assuming MUT=3, that is, it must be opened for more than 3 periods in a row) and MDT (assuming MDT=3, that is, it must be closed for more than 3 periods in a row). If it is violated (it can be seen that both MUT and MDT are violated at this time), it is repaired, with the period as the center, and the period on both sides is searched in turn, and the period that was not opened in the original scheduling plan of the unit is changed to open in turn, until the period that was opened in the original scheduling plan of the unit is found, so that the repaired scheduling plan meets MUT and MDT.

[0067] In another case, in the lower part of the table, the day-ahead dispatch plan for the j-th unit lists 24 time periods, and also lists the dispatch plan for the last 8 time periods of the previous day for the unit. The previous dispatch plan cannot be changed. In the original plan, it is found that there is a problem node period in the x period, which is displayed in bold in the table. The j-th unit is found to be in a closed state near the problem node, and then the unit is turned on in a directional manner during this period. Then, it is determined whether the dispatch plan violates MUT and MDT (assuming MUT=MDT=3). It can be seen that the start-stop state of the period to the right of the central period is already on, that is, there is no need to continue searching to the right, and the dispatch plans on the right side of the central period (later) all meet MUT and MDT, but the left side of the central period The scheduling plan does not satisfy the MDT. The time periods on the left are searched in sequence, and the time periods that were not turned on in the original scheduling plan of the unit are changed to turned on in sequence. However, since the scheduling plan of the previous day cannot be modified, the timing constraints are still not met after the start and stop status of the time periods before the central time period is modified (at this time, the shutdown plan of the last time period in the scheduling plan of the previous day will not satisfy the MDT, that is, there is only one continuous 0). Therefore, in the current x time period, this unit does not have the conditions to repair the violation of the start and stop timing constraints after the directional start. Then cancel the start of the unit, and start the next unit in the order of the units with the electrical distance from the nearest to the farthest that have been retrieved, and repeat the above repair steps until the series of units that have been turned on meet the necessary conditions to make up for the power difference of the problem node.

[0068] Step D of the present invention is based on the directional optimization of electrical distance:

[0069] Electrical distance analysis: The node impedance matrix is ​​used to calculate the electrical distance between the problem node and the unstarted unit, reflecting the power transmission efficiency.

[0070] Directional start-up of units: start nearby units in ascending order of electrical distance until their maximum total active output makes up for the power gap.

[0071] Timing constraint fix: Adjust the start and stop schedule to meet the minimum start and stop time (MUT / MDT) and keep the variables of the directed modification stable.

[0072] Therefore, local constraints can be eliminated quickly, power gaps can be solved in a targeted manner, and VC convergence to zero can be accelerated; short-distance units can be started first to reduce long-distance transmission losses; frequent start and stop of units can be avoided by repairing timing constraints, in line with actual operating restrictions, and the optimized variable values ​​can be kept unchanged to ensure the targetedness and feasibility of the adjustment.

[0073] In step E, the directional optimization result of step D is used as the new population, and the system constraint violation VC is recalculated. If VC=0, the directional optimization stage is ended directly; if VC≠0, the new population returns to the genetic algorithm or sparse evolutionary algorithm, and continues to iterate in a large loop until the system constraint violation is reduced to 0, and the directional optimization process is ended. Since the system constraint violation is close to convergence when the directional optimization program is started, the number of iterations required in this stage is much smaller than the number of iterations required by the original non-directional method.

[0074] After the directed optimization process is completed, the system has preliminary physical feasibility. On this basis, the evolutionary algorithm is continued to be used to optimize economic scheduling and minimize the total operating cost of the system. The decision variables include: the switching state of the unit, the active power output of the unit, the reactive power output of the unit, and the node voltage amplitude. When the maximum number of iterations is reached, the algorithm selects the individual with the highest fitness from the current population as the final solution and outputs the final decision variables.

[0075] There is often a balance between timing constraints and optimization goals. Directed start-up of units may destroy the original start-stop timing constraints (such as MUT / MDT). Therefore, after the directed modification, only the timing constraints are repaired and the optimization variable values ​​are kept unchanged to ensure the pertinence and feasibility of the adjustment.

[0076] Step E of the present invention is iterative verification and economic dispatch optimization:

[0077] Verify the VC value after directional optimization. If it does not converge, return to step B for iteration. If it converges, enter economic dispatch optimization to minimize the total operating cost of the system (power generation cost, start-up and shutdown cost, etc.). A closed-loop optimization is achieved, and the global main constraints are fully satisfied through iteration. Further optimize the cost based on the physically feasible solution to achieve a balance between safety and economy.

[0078] In addition, multi-stage optimization coordination is also required, mainly to coordinate the conflict between non-directional global search and directional local adjustment schemes; the present invention controls stage switching through a dynamic threshold (VC drops to 5% of the initial value) to ensure that directional optimization is only triggered at the critical stage to avoid deviation of the global solution.

[0079] The specific implementation method is as follows: First, in the system initialization stage, it is necessary to input the topology of the power system, unit type, generator capacity, load demand, line flow constraints, and voltage limit information, and calculate the node impedance matrix (the MATPOWER toolkit can be used to find the system node admittance matrix, and then the inverse matrix is ​​obtained to obtain the node impedance matrix). If the IEEE-118 system is used for the experiment, all unit and node information comes from the actual power system data; second, in the non-directional optimization stage, a genetic algorithm (GA) or a sparse evolutionary algorithm (Sparse EA) is used to make preliminary decisions on the start and stop of the unit. The optimization goal is to minimize the total operating cost including the power generation cost and the start and stop cost, while satisfying the constraints of power balance, voltage limit, and line flow. Third, after each iteration of the non-directional optimization, the convergence of the system is evaluated and the constraint violation is calculated. If the constraint violation is less than 5% of the constraint violation output of the first iteration, the directional optimization stage is entered.

[0080] In the directional optimization stage, the specific implementation method is as follows: first, evaluate the power inflow and outflow of each node in each time period, and detect the problem node: the net input power of the problem node is less than the load power required by the node. At the same time, the LAM_P† multiplier in the node data (mpc.bus) in the MATPOWER toolkit can be used for reference and comprehensive verification; secondly, the algorithm determines which units near the above-mentioned problem nodes are not turned on based on the electrical distance analysis, and performs directional start and stop adjustments. The mutual impedance elements in the node impedance matrix calculated in the system initialization phase are used to characterize the electrical distance; after modifying the start and stop of the unit, it is necessary to perform timing repair on the start and stop scheduling plan of the unit, keep the variable values ​​​​directly modified in the previous step unchanged, so that it meets the start and stop constraints, that is, the minimum start time MUT and the minimum close time MDT.

[0081] The core of the directed optimization process is to detect problem nodes and analyze electrical distances, and to prioritize the start-up of units close to the constraint violation area based on these distances, thereby adjusting the local power flow and eliminating local constraint violations. After directed optimization, the directed optimization results are used as a new population to recalculate the system constraint violation VC. If VC=0, the directed optimization stage ends directly; if VC≠0, the new population returns to the genetic algorithm or sparse evolutionary algorithm and continues to iterate in a large cycle until the system constraint violation is reduced to 0, ending the directed optimization process. The unit start-stop plan after VC convergence will be used for economic dispatch. Through further optimization calculations, the total operating cost of the system is minimized. The decision variables include: the switch state of the unit, the active power output of the unit, the reactive power output of the unit, and the node voltage amplitude; until the maximum number of iterations is reached, the algorithm selects the individual with the highest fitness from the current population as the final solution and outputs the final decision variable.

[0082] The present invention properly integrates the characteristics of the physical system with the existing evolutionary algorithm through reasonable analysis of electrical distance and directional adjustment of the start and stop status of the unit, adds directional "navigation" to the evolutionary algorithm, and uses a priori physical rules to guide the search, which can effectively eliminate the local small area constraint violation of the large-scale power grid and improve the efficiency of the unit start and stop scheduling. It is suitable for large-scale and complex topological systems. The unit combination problem under the AC power flow constraint of large-scale power systems is solved. On the basis of non-directional optimization methods such as genetic algorithms and sparse evolutionary algorithms that improve the initialization method of traditional evolutionary algorithms, combined with directional optimization strategies, the system convergence speed is improved and the constraint violation degree is reduced, thereby realizing efficient power system scheduling.

[0083] The present invention mainly solves the problem of long-term non-convergence of VC in random search of traditional evolutionary algorithms due to insufficient power supply in local areas of large-scale power systems. This solution quickly fills the power gap by starting the units in a directional manner and reduces the waste of computing resources by dynamic stage switching. Compared with the prior art, the present invention integrates non-directional and directional strategies, combines the global search capability of evolutionary algorithms and local directional adjustments guided by physical rules, breaks through the random limitations of traditional non-directional algorithms, and significantly improves the convergence speed. Secondly, in the dynamic switching optimization stage, directional optimization is started only when the constraints are close to convergence, avoiding premature local optimization that leads to a decrease in the quality of the global solution. The problem nodes are finely located, and the start-stop strategy of the unit is adjusted based on the electrical distance, which effectively solves the problem of local power imbalance and reduces network constraint violations.

[0084] The present invention solves the problems of slow convergence, waste of computing resources, and difficulty in eliminating local constraints in large-scale power systems existing in traditional methods through a phased optimization strategy (global search → local directional adjustment → economic dispatch) and a directional mechanism guided by physical rules. It not only improves the quality of the solution, but also significantly reduces the computing cost, providing an efficient solution for the safe and economical operation of complex power systems.

[0085] The above is only a preferred embodiment of the present invention, and does not limit the structure of the present invention in any form. Any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A non-directional and directional fusion ACUC unit start-stop optimization method, characterized in that: The following steps are involved: Step A. Initialize system data and evolutionary algorithm, and use non-directional optimization algorithm to make preliminary decisions on unit start and stop variables; Step B. After each population iteration, the constraint violation degree VC of the system is calculated. The constraint violation degree VC is the weighted sum of the power imbalance value, the voltage over-limit value and the line power flow over-limit value. If VC drops below 5% of its first iteration output value and VC≠0, the directional optimization stage is entered; Step C. For each period of the original scheduling plan, calculate the difference between the net input power and the load power of each node, and filter out the problem nodes. The problem nodes must meet the following conditions: the net input power of the node is less than the load power of the node; Step D. For each problem node, retrieve the electrical distance between the problem node and the units that are not started in this scheduling plan based on the node impedance matrix, start the nearby units in ascending order of electrical distance, and repair the violations of the timing constraints of the unit start and stop, including the minimum start time and the minimum shutdown time; the number of units to be started needs to meet the condition that the sum of the maximum active output power of these units is not less than the difference between the node net input power and the load power calculated in the previous step; Step E. After the directional adjustment, recalculate the constraint violation degree VC. If VC≠0, return the adjusted unit start-stop plan to the iteration, repeat steps B to D until VC=0, end the directional optimization process, and optimize the economic dispatch on this basis; Among them, the difference between the net input power and the load power of each node in each period of the original scheduling plan is calculated, and the calculation formula is: ,in, Indicates Nodes in the period The difference between the node net input power and the load power, Indicates Nodes in the period The net input power, Indicates Nodes in the period Load power; Filter out the problem nodes in the scheduling plan for each period of the day, and the screening conditions are: <0, then Nodes in the period It is a problem node when , that is, the net input power of the node is less than the load power required by the node; The non-directional optimization algorithm includes: a genetic algorithm and a sparse evolutionary algorithm, wherein the sparse evolutionary algorithm refers to an algorithm that introduces sparse initialization technology on the basis of a traditional evolutionary algorithm to obtain a better initialized start and stop variable.

2. According to claim 1, a non-directional and directional fusion ACUC unit start-stop optimization method is characterized in that: In step A, the system data are initialized including: system topology; type, capacity and start / stop status of generators; load demand; physical constraints including line flow constraints and node voltage limits; and node impedance matrix is ​​calculated.

3. According to claim 1, a non-directional and directional fusion ACUC unit start-stop optimization method is characterized in that: In step A, the relevant parameters of the evolutionary algorithm, i.e., the genetic algorithm or the sparse evolutionary algorithm, including the initial population size and the upper limit of the number of iterations, are initialized, and the initial unit start and stop state is set, and then the genetic algorithm or the sparse evolutionary algorithm is used to perform population iteration on the population composed of the initialized unit start and stop state variables.

4. According to claim 1, a non-directional and directional fusion ACUC unit start-stop optimization method is characterized in that: In step B, the constraint violation VC of the system is calculated after each population iteration using the evolutionary algorithm. If VC drops to the constraint violation VC output in the first iteration, （1） If the constraint violation degree VC is less than 5% and VC≠0, the system meets the conditions for entering the directional optimization stage. The calculation formula of the constraint violation degree VC is: ， ， ， ， ， Among them, T represents the 24 time periods of the day-ahead scheduling plan, N represents the node set of the system, Indicates that the system is in the period The power imbalance value, Indicates that the system is in the period The voltage exceeds the limit value, Indicates the system time period α1 represents the weight coefficient of the power imbalance value, α2 represents the weight coefficient of the voltage over-limit value, and α3 represents the weight coefficient of the line power flow over-limit value, α1=α2=α3=1; Indicates the system The units at the nodes are in the period The power generation capacity, Indicates the system Nodes in the time period The load power, Indicates that the system is in the period The network loss power, Indicates that the system is in the period The system spinning reserve power is 10% of the sum of the total system load power and the network loss power; Indicates the system Nodes in the period The voltage, Indicates the system The voltage upper limit of each node, Indicates the system The lower voltage limit of each node, Indicates the system The voltage rating of each node; Indicates the system Lines in the period The meritorious trend, Indicates the system The upper limit of active power flow of the line, Indicates the system Active power flow rating of the line.

5. According to claim 1, a non-directional and directional fusion ACUC unit start-stop optimization method is characterized in that: In step D, the electrical distance is determined by analyzing the power flow path and network topology between nodes, and is reflected by the magnitude of the mutual impedance element in the node impedance matrix. The calculation formula is: ， in, is the node in the system node impedance matrix With Node The mutual impedance elements between Represented as a node With Node The default correction factor is 1.

6. The ACUC unit start-stop optimization method of non-directional and directional fusion according to claim 1 is characterized in that: In step D, a series of units with electrical distances from the problem node from near to far are retrieved in ascending order of electrical distance, and these units are turned on in sequence until the sum of the maximum active output power of the series of units turned on is not less than the difference between the node net input power and the load power calculated in the previous step. , that is, the series of units that are turned on must at least have the Nodes in the period Necessary conditions for node power difference.

7. The ACUC unit start-stop optimization method of non-directional and directional fusion according to claim 1 is characterized in that: In the step D, after the units close to the problem nodes violating the constraints are preferentially started according to the electrical distance, the timing constraint violations of the new scheduling plan are repaired according to the adjusted start and stop status of the units, so that the operation restrictions of the minimum start and stop time are met between the scheduling plans of the same unit in each period, including the minimum start time MUT and the minimum close time MDT, and the variable values ​​modified in a directionally manner are kept unchanged during the repair; specifically, for each unit that is directionally started, after the start and stop variables of the unit are directionally modified, if the operation restrictions of the minimum start and stop time have been met between the scheduling plans of the unit in each period, then there is no need to repair its timing constraints; if the operation restrictions of the minimum start and stop time have not been met between the scheduling plans of the unit in each period, then it is necessary to repair its timing constraints, with the time period where the start and stop variables modified in a direction as the center, to the two sides. The time periods on both sides are searched in sequence, and the time periods that were not turned on in the original scheduling plan of the unit are changed to be turned on in sequence, until the time period that was turned on in the original scheduling plan of the unit is searched. At this time, the start and stop variables of the unit in the search range on both sides of the central time period are in the turned-on state. Since the original scheduling plan must meet the timing constraints to appear in the algorithm's population, the time periods after the central time period must also meet the timing constraints. However, the time periods before the central time period may not meet the timing constraints because the scheduling plan of the previous day cannot be modified. If the scheduling plan still does not meet the timing constraints after the start and stop status of the time period before the central time period is modified, the unit will be canceled, and the next unit will be turned on in the order of the units retrieved from the nearest to the farthest electrical distance, and the above repair steps will be repeated until the principle is met: a series of turned-on units have the ability to make up for the first Nodes in the period Necessary conditions for node power difference.

8. The ACUC unit start-stop optimization method of non-directional and directional fusion according to claim 1 is characterized in that: In step E, the directed optimization result of step D is used as the new population, and the system constraint violation VC is recalculated. If VC=0, the directed optimization stage is directly ended; if VC≠0, the new population returns to the genetic algorithm or sparse evolutionary algorithm, and continues to perform large loop iterations until the system constraint violation is reduced to 0, and the directed optimization process is ended.

9. The ACUC unit start-stop optimization method of non-directional and directional fusion according to claim 1 is characterized in that: In the step E, After the directional optimization process is completed, the evolutionary algorithm is used to optimize the economic dispatch on this basis to minimize the total operating cost of the system. The decision variables include: the switching state of the unit, the active power output of the unit, the reactive power output of the unit, and the node voltage amplitude. When the maximum number of iterations is reached, the algorithm selects the individual with the highest fitness from the current population as the final solution and outputs the final decision variables.

Images