A flexible excitation system reactive power control parameter optimization setting method

Abstract

The application belongs to the technical field of power system control, and discloses a method for optimizing and setting reactive power control parameters of a flexible excitation system, which comprises the following steps: d Starting from the axis transient equation, explicit mapping of the control parameters of the flexible excitation system and transient reactive power is derived and established, the control parameters including a strong excitation trigger threshold, a strong excitation exit threshold, a strong excitation multiple, a reactive power channel input depth, a reactive power channel input rate, a direct current side voltage set value and a reference trajectory climbing slope; a normalized weighted comprehensive performance index function is constructed, the index function containing a reactive power support intensity index and a voltage overshoot index in the transient process of the synchronous generator; the control parameters to be optimized are divided into two subsets according to the physical channels to which the control parameters belong, and the control parameters in the two subsets are alternately optimized by using an alternating iteration dimension reduction optimization algorithm, so as to minimize the normalized weighted comprehensive performance index function, and the optimal control parameter group is obtained for the transient control of the flexible excitation system.

Description

Technical Field

[0001] This invention belongs to the field of power system control technology, specifically relating to a method for optimizing and setting reactive power control parameters of a flexible excitation system. Background Technology

[0002] The excitation system is the core of the electrical control of a synchronous generator. It controls the terminal voltage (U) by adjusting the generator excitation current. t The system addresses reactive power (Q) and maintaining the power angle stability of the power system. Traditional self-excited static excitation systems typically employ thyristor (semi-controlled rectifier) ​​rectifier bridges to achieve closed-loop control of power drawn from the generator terminals and rectified before being supplied to the rotor windings. Their excitation peak voltage is heavily dependent on the generator terminal voltage. When a severe power system fault causes a drop in generator terminal voltage, the strong excitation capability of the traditional excitation system decreases sharply, easily leading to a passive predicament where the more reactive power support is needed, the less capable the excitation system is of providing it.

[0003] To address this challenge, the Flexible Excitation System (FES) was developed. The FES utilizes fully controllable power devices such as IGBTs (Insulated Gate Bipolar Transistors) to construct a two-stage power conversion topology: a front-stage voltage source converter (VSC) + a rear-stage DC chopper. The front and rear stages are connected via an intermediate DC-link capacitor loop, forming a stable and controllable DC voltage source. The flexible excitation topology is as follows: Figure 1 As shown in the diagram. Through this architecture, the flexible excitation system achieves physical barrier decoupling between the excitation peak voltage and the generator terminal voltage, maintaining a strong excitation output capability even when the grid voltage drops significantly. Simultaneously, the flexible excitation system possesses independently operating excitation voltage channels (channel A) and reactive current channels (channel B), thus providing the synchronous generator with stronger transient voltage support and a wealth of broadband oscillation suppression methods.

[0004] However, while the introduction of flexible excitation systems significantly improves control flexibility, it also constructs a complex control system involving multiple coupled parameters, including the forced excitation trigger threshold, forced excitation exit threshold, forced excitation multiple, reactive power channel engagement depth, reactive power channel engagement rate, DC-side voltage setpoint, and reference trajectory ramp slope. In current engineering practice, the tuning of these control parameters still heavily relies on trial and error based on manual experience or independent scanning of single parameters, leading to the following shortcomings in existing technologies: 1. The lack of an analytical functional relationship between control parameters and transient performance makes the tuning process a black box. Existing methods typically use electromagnetic transient simulation waveforms as the basis for adjustment. For each set of parameters, engineers must run a complete electromagnetic transient simulation to obtain the corresponding reactive power output and voltage recovery performance. The mapping relationship between parameters and performance is an implicit black box, making it impossible to directly and quantitatively predict the direction and magnitude of parameter adjustment from the physical equations.

[0005] 2. Significant multi-parameter coupling effects lead to the dimensionality curse of exponentially increasing computational load. Several core control parameters of the flexible excitation system exhibit strong cross-coupling relationships. For example, the optimization effect of the forced excitation multiple is highly dependent on the depth of the reactive power channel, while the optimal forced excitation exit threshold is directly affected by the forced excitation multiple. Traditional serial debugging or single-parameter scanning cannot account for these cross-influences, and exhaustive mesh scanning in a multi-dimensional space will cause the simulation computational load to increase exponentially with the number of parameters.

[0006] 3. There is an inherent contradiction between the speed and stability of transient voltage support, and a unified quantitative trade-off framework is lacking. If rapid reactive power support is desired, a larger excitation multiple and a later excitation exit timing are required; if suppressing overshoot during voltage recovery is desired, a moderate excitation strength and an earlier exit timing are needed. Current technologies typically treat maximizing reactive power support capability and minimizing voltage overshoot as independent steps. Due to the lack of a unified quantitative objective function, the final tuning result highly depends on the engineer's personal experience, resulting in extremely poor reproducibility and difficulty in approximating the global near-optimal solution. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method for optimizing and tuning reactive power control parameters of a flexible excitation system. By establishing an explicit mapping model between the control parameters and transient reactive power of the flexible excitation system, a dimensionality reduction decomposition alternating iterative optimization algorithm based on channel attribution is used to solve for subsets of control parameters. Furthermore, the method quantifies the trade-off between reactive power support strength and voltage overshoot, thereby achieving the global optimal tuning of control parameters for a multi-parameter coupled flexible excitation system.

[0008] This invention provides the following technical solution: A method for optimizing and tuning reactive power control parameters of a flexible excitation system includes: From synchronous generator d Starting from the shaft transient equation, an explicit mapping between the control parameters of the flexible excitation system and the transient reactive power is derived and established. The control parameters include the forced excitation trigger threshold, the forced excitation exit threshold, the forced excitation multiple, the reactive power channel input depth, the reactive power channel input rate, the DC side voltage setpoint, and the reference trajectory ramp slope. A normalized weighted comprehensive performance index function is constructed, which includes the reactive power support strength index and voltage overshoot index during the transient process of synchronous generator; Based on the physical channel to which the control parameters belong, the control parameters to be optimized are divided into two subsets. An alternating iterative dimensionality reduction optimization algorithm is used to alternately optimize the control parameters in the two subsets. With the goal of minimizing the normalized weighted comprehensive performance index function, the optimal control parameter set is obtained for the transient control of the flexible excitation system.

[0009] An explicit mapping model between control parameters and transient reactive power is established, providing a calculable and analytical mathematical foundation for the parameter-to-performance mapping relationship. Mathematical calculations can quantitatively predict the direction and magnitude of parameter adjustments, transforming control parameter tuning from empirical trial-and-error to theoretical analysis. The full-space search is decomposed into two subsets by channel dimensionality reduction, and iterative optimization is performed alternately, significantly compressing the search volume in six-dimensional space and making multi-parameter global optimization engineering-feasible. By constructing a normalized weighted comprehensive performance index function, the two contradictory objectives of maximizing reactive power support and minimizing voltage overshoot are unified into a single scalar optimization problem. This resolves the fundamental contradiction between the rapidity of transient reactive power support and the stability of voltage recovery, providing reproducible and quantifiable optimality criteria.

[0010] Preferably, the transient reactive power parameterization model includes a three-stage excitation voltage control logic: When the terminal voltage is less than or equal to the product of the forced excitation trigger threshold and the rated terminal voltage, it is triggered and enters the forced excitation state. When the terminal voltage is greater than the product of the forced excitation trigger threshold and the rated terminal voltage, and less than or equal to the product of the forced excitation exit threshold and the rated terminal voltage, the forced excitation state is maintained to achieve delayed exit. When the generator terminal voltage exceeds the product of the forced excitation exit threshold and the rated generator terminal voltage, the forced excitation state is exited, and the excitation voltage decays exponentially.

[0011] Breaking away from the traditional limitation of binding the excitation triggering and exit to the same threshold, a three-stage model is constructed by introducing an independent excitation exit threshold. By utilizing the generator's delayed exit state during the voltage recovery phase, a significant amount of transient reactive power is injected into the grid, significantly enhancing the system's transient voltage support strength.

[0012] Preferably, the normalized weighted comprehensive performance index function is in the form of: using weighting coefficients to weight and combine the normalized reactive power support strength index and the normalized voltage overshoot index, wherein the reactive power support strength index has a negative weight in the comprehensive index. The reactive power support strength index is the cumulative reactive power injected into the system by the unit within the observation window after the fault is cleared; the voltage overshoot index is the maximum extent by which the generator terminal voltage exceeds the rated generator terminal voltage during the voltage recovery process after the fault is cleared.

[0013] Reactive power support and voltage overshoot differ significantly in dimension and order of magnitude. By normalizing the benchmark value, the overwhelming effect of a large-order-of-magnitude indicator on a small-order-of-magnitude indicator is eliminated, ensuring that the voltage overshoot indicator can be correctly perceived during optimization. By introducing weighting coefficients, the tuning method can flexibly adapt to the needs of different power grids (such as a relaxed power grid that favors maximizing reactive power, or a voltage-sensitive power grid that favors strictly suppressing overshoot).

[0014] Preferably, when calculating the voltage overshoot index, a reactive power-voltage sensitivity mapping model is used to generate the recovery trajectory of the generator terminal voltage. A fault severity correction factor is introduced into the mapping model, and a safety upper limit truncation constraint is applied to the generated generator terminal voltage trajectory. The fault severity correction factor is determined by the ratio of the initial voltage after fault clearance to the reference recovery voltage.

[0015] A fault severity correction factor is introduced to ensure that faults of different severity receive differentiated voltage overshoot penalties in the optimization objective, avoiding the potential for severe voltage overshoot due to blindly pursuing reactive power under minor faults. Applying a safety upper limit cutoff constraint not only conforms to the actual operating specifications of power system overvoltage protection but also prevents unrealistic voltage spikes caused by mathematical extrapolation in simulation calculations.

[0016] Preferably, the following is also included before parameter tuning: The physical lower bound constraint of the forced excitation trigger threshold is determined by analytical derivation, and the physical lower bound constraint is determined by the voltage fluctuation amplitude during normal operation. The reference trajectory ramp slope is parameterized as an explicit functional expression with respect to the forced excitation trigger threshold, thereby eliminating the independent search dimension of the reference trajectory ramp slope in parameter tuning.

[0017] By deriving the physical lower bound determined by the normal voltage fluctuation amplitude, a reasonable search boundary is set for the optimization algorithm, fundamentally eliminating the situation where small fluctuations during normal operation trigger strong excitation, protecting the fully controlled IGBT devices in the subsequent chopper circuit, and reducing unnecessary switching losses and junction temperature rise.

[0018] Preferably, the physical channel includes an excitation voltage channel and a reactive current channel; The control parameter subset includes a first subset belonging to the excitation voltage channel and a second subset belonging to the reactive current channel; The control parameters for the first subset include the excitation multiple, the DC side voltage setpoint, and the excitation trigger threshold. The control parameters of the second subset include reactive power channel activation depth, reactive power channel activation rate, and forced excitation exit threshold.

[0019] Subset partitioning is performed based on the natural separation of the excitation voltage channel and reactive current channel of the flexible excitation system on the time scale. This fully utilizes the physical characteristics of the independent and coordinated operation of the two channels, providing reasonable physical boundary guidance for the accurate iteration of the subsequent dimensionality reduction optimization algorithm.

[0020] Preferably, the alternating iterative dimensionality reduction optimization algorithm for alternately tuning the control parameters in the two subsets includes: Initialize all control parameters in the first and second subsets; With the second subset of control parameters fixed, the first subset of control parameters that minimizes the normalized weighted comprehensive performance index function is solved by scanning within a constrained 3D mesh. The first subset of control parameters obtained from the current iteration round is fixed, and the second subset of control parameters that minimize the normalized weighted comprehensive performance index function is solved within the constrained 3D mesh. The first subset of control parameters and the second subset of control parameters are solved alternately until the normalized weighted comprehensive performance index function converges or the maximum number of iterations is reached, so as to obtain the optimal set of control parameters.

[0021] Preferably, the alternating iterative dimensionality reduction optimization algorithm also includes a grid resolution adaptive adjustment strategy: after the initial iteration, the grid search range of subsequent iterations is narrowed to the neighborhood of the previous round's optimal value, and the search step size is reduced.

[0022] By adopting a strategy of initial coarse mesh positioning and subsequent neighborhood shrinking with halved step size, the computational cost of subsequent iterations is reduced to about 1 / 4 of that of the initial iteration. This approach utilizes the coarse mesh to ensure global search coverage while achieving high-precision local convergence through the fine mesh, effectively improving the engineering efficiency of optimization tuning.

[0023] Preferably, during the grid scan solution of the first subset control parameters and the second subset control parameters, a penalty function method is used to process the candidate parameter points that violate the voltage safety constraint; the voltage safety constraint condition is that the terminal voltage is less than or equal to the maximum voltage.

[0024] Preferably, the method further includes: performing transient simulation verification on the optimal control parameter set in a fault condition set, wherein the fault condition set includes multiple fault scenarios determined by fault type, voltage drop depth and fault duration; if a condition does not meet the constraints, the condition is added to the training set and the control parameters are re-optimized and solved.

[0025] Closed-loop verification and retraining are performed on a set of operating conditions consisting of different fault types, drop depths and durations to overcome the limitation that single-condition optimization results may fail in other scenarios. This ensures that the final output of the optimal control parameter set has high adaptability and versatility when facing complex and ever-changing actual power grid faults.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: Establish an explicit mapping model between control parameters and transient reactive power, so that the mapping relationship between parameters and performance has a calculable and analytical mathematical basis. The direction and magnitude of parameter adjustment can be quantitatively predicted through mathematical calculations, transforming control parameter tuning from empirical trial and error to theoretical analysis.

[0027] The full-space search is decomposed into two subsets by channel dimensionality reduction and iteratively optimized alternately, which greatly reduces the search volume in the six-dimensional space and makes the multi-parameter global optimization feasible in engineering.

[0028] By constructing a normalized weighted comprehensive performance index function, the two contradictory objectives of maximizing reactive power support strength and minimizing voltage overshoot are unified into a scalar optimization problem, which solves the essential contradiction between the rapidity of transient reactive power support and the stability of voltage recovery, and provides a reproducible and quantifiable criterion for optimality. Attached Figure Description

[0029] Figure 1 It is a flexible excitation topology; Figure 2 A flowchart of the method for optimizing and setting reactive power control parameters of a flexible excitation system provided by the present invention; Figure 3 A comparison chart of the terminal voltage recovery trajectory under extreme fault conditions; Figure 4 A comparison chart of transient reactive power output of the system under extreme fault conditions; Figure 5 A comparison chart of the recovery trajectory of the terminal voltage under severe fault conditions; Figure 6 This is a comparison chart of transient reactive power output of the system under severe fault conditions; Figure 7 A comparison chart of the recovery trajectory of the terminal voltage under moderate fault conditions; Figure 8 This is a comparison chart of the transient reactive power output of the system under moderate fault conditions. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0031] The present invention will now be described in further detail with reference to the accompanying drawings: Example 1: like Figure 2 As shown in the figure, this embodiment provides a method for optimizing and tuning the reactive power control parameters of a flexible excitation system. The flexible excitation topology is as follows. Figure 1 As shown, the method includes the following steps: From synchronous generator d Starting from the shaft transient equations, an explicit mapping between the control parameters and transient reactive power of the flexible excitation system is derived and established to construct a parameterized model. The control parameters include the forced excitation trigger threshold. Strong incentive exit threshold Incentive multiplier Depth of reactive power channel investment Reactive power channel input rate DC side voltage setting value and the slope of the reference trajectory ; Constructing a normalized weighted comprehensive performance index function The index function includes the reactive power support strength index during the transient process of the synchronous generator. With voltage overshoot index ; Based on the physical channels to which the control parameters belong, the control parameters to be optimized are divided into two subsets. An alternating iterative dimensionality reduction optimization algorithm is used to alternately optimize the control parameters in the two subsets to minimize the normalized weighted comprehensive performance index function. To achieve the objective, the optimal control parameter set is obtained by solving the problem. Transient control for flexible excitation systems.

[0032] The following describes the specific method and process: Taking a single-unit infinite bus system of a synchronous generator as the object, neglecting the transient processes of the damping winding and stator winding, the synchronous generator... d The shaft transient equation is expressed as:

[0033] in, for q shaft transient electromotive force, This is the excitation voltage. for d Shaft synchronous reactance, for d Shaft transient reactance, for d Axis current components, for d Axis transient open-circuit time constant.

[0034] The excitation peak voltage of a traditional self-excited system is proportional to the generator terminal voltage. The excitation voltage increment of a traditional self-excited static excitation system is expressed as:

[0035] in, For the excitation voltage increment of a traditional self-excited static excitation system, it represents the additional excitation voltage that the system can actually provide under fault transient conditions; This is the rated excitation voltage of the synchronous generator. This refers to the terminal voltage of the synchronous generator. This is the rated terminal voltage of the synchronous generator. When When it drops to 0.3 pu, The excitation capability is negative. The excitation system not only fails to increase the excitation current, but also experiences a net attenuation of the excitation current, leading to a predicament where traditional excitation becomes increasingly ineffective under severe faults. In contrast, the flexible excitation system achieves physical decoupling between the excitation peak voltage and the generator terminal voltage through a two-stage conversion via a front-end voltage source converter, an intermediate DC capacitor circuit, and a rear-end DC chopper.

[0036] in, This represents the excitation voltage increment of the flexible excitation system, signifying the additional excitation voltage that the flexible excitation system can stably and actively provide under fault transient conditions.

[0037] IGBT junction temperature It is a strong incentive multiple. The key constraint for determining the value is that the conduction loss of the IGBT during the forced excitation period is proportional to the square of the excitation current.

[0038] At ambient temperature Radiator thermal resistance Under these conditions, the IGBT junction temperature constraint gives Engineering upper limit: Under typical parameters (maximum junction temperature) On resistance ), incentive multiple Theoretical maximum value Therefore, the incentive multiplier will be increased. The search space constraint is [1.5, 3.0], rather than any theoretically arbitrary value.

[0039] The establishment of the excitation current is also affected by the LC filtering stage of the DC chopper. The complete excitation current transfer function contains a second-order stage:

[0040] in, and These are the filter inductor and filter capacitor at the output of the DC chopper, respectively. The equivalent resistance of the generator rotor winding. For the Laplace operator, This is the generator rotor excitation current. This represents the increment of the excitation voltage.

[0041] In typical parameter values ​​( Under these conditions, the cutoff frequency of the second-order element is approximately 2.3 kHz, which is much higher than the second-level timescale of the strong excitation. Therefore, in the parameterized model, it is taken as... The simplification is reasonable.

[0042] The circuit parameter matrix (DCMatrix) constraints of the DC-DC Chopper circuit are given. Physical upper limit :

[0043] in, This represents the physical upper limit of the reactive current channel connection rate. This refers to the IGBT switching frequency (typically 5kHz). This represents the maximum change in duty cycle in a single step. It is the rated excitation current. This refers to the VSC DC bus voltage.

[0044] By decoupling the triggering and deactivation of the forced excitation into two independent parameters, the following three-stage excitation voltage is formed. Model:

[0045] in, It is a strong constraint, meaning the strong excitation exit threshold must not be lower than the strong excitation trigger threshold. The existence of the second paragraph is one of the core features of this invention: when It has been restored to (This indicates the fault has been cleared and the voltage is recovering), but it has not yet reached its target. During this period, the strong excitation continues, and delaying the exit from this phase will inject additional reactive power into the grid, which is a significant contributor to J1 (reactive power support intensity). The larger the value, the longer the delay, the larger the J1 (reactive power support strength), and the higher the risk of J2 (voltage overshoot peak). When At that time, the above model degenerates into existing engineering practices, with strong incentive triggering and exit tied to the same threshold, losing the freedom to optimize.

[0046] Based on this Model, for d By performing a Laplace transform on the axis incremental equation, the incremental stator current component generated by the forced excitation is obtained. :

[0047] in, This is the trigger moment for strong excitation.

[0048] The flexible excitation system has two independent and coordinated control channels.

[0049] Excitation voltage channel (channel A): This channel controls the excitation current of the rotor winding through the front-end VSC and the rear-end chopper, thereby adjusting the generator air gap flux and terminal voltage. The response time of this channel is 100~500ms, and it is mainly used for steady-state voltage regulation and transient forced excitation support.

[0050] Reactive current channel (channel B): This channel directly controls the reactive current component injected into the generator stator side via an AVR (Automatic Field Controller), achieving rapid reactive power injection or absorption in less than 20ms. This channel bypasses the limitation of the large time constant of the rotor magnetic field, directly and rapidly compensating for system reactive power at the electrical level, equivalent to a fast-response static synchronous compensator (STATCOM).

[0051] The reactive power component of the excitation channel (channel A, response time 100~500ms) for:

[0052] The reactive power output component of the reactive power channel (channel B, response time <20ms, response rate limited by ramp slope) for:

[0053] Comprehensive reactive power output for:

[0054] This expression establishes seven parameters. to transient reactive power The explicit mapping is the mathematical foundation for subsequent optimizations throughout the paper. Among them, This is the rated reactive power capacity of the excitation channel. This represents the maximum reactive power capacity of the reactive power channel. This refers to the initial reactive power before the fault occurs.

[0055] A reactive power support strength index and a voltage overshoot index are constructed, and the normalized reactive power support strength index and the normalized voltage overshoot index are weighted and combined using weighting coefficients to form a normalized weighted comprehensive performance index function.

[0056] Index 1 - Reactive Power Support Strength (J1), representing the observation window [0, ...] after fault clearing. Within this range, the cumulative reactive power injected into the system by the generator unit is calculated using the following formula:

[0057] Indicator 2 - Voltage overshoot peak value (J2) indicates the voltage recovery process after fault clearance. The maximum extent by which the terminal voltage exceeds the rated value is calculated using the following formula:

[0058] These two objectives have different dimensions and opposite trends, so a weighting coefficient is used. The weighted combination and normalized weighted comprehensive performance index function is expressed as follows:

[0059] in, This is the weighting coefficient (default value is 0.5). The baseline reactive power support strength for traditional excitation. This represents the maximum allowable overshoot.

[0060] In the parameter optimization framework, the corresponding terminal voltage for each set of candidate parameter vectors is obtained through simulation. and doing nothing Then, the comprehensive performance index is calculated. The generation of the generator terminal voltage adopts reactive power-voltage sensitivity mapping, which is divided into three progressive steps: fixed during the fault period, linear mapping during the recovery period, and safe cut-off.

[0061] 1. Voltage model during fault operation During the fault ( t < T fault The terminal voltage is directly determined by external faults and does not participate in optimized control.

[0062] in, This represents the voltage drop on the high-voltage side during the fault (typically 0.2~0.7 pu). This refers to the fault clearing time. After fault clearing, there is a brief transition period (approximately 10ms) during which the voltage jumps to the initial recovery value after the fault. :

[0063] in, The topology of the line after the fault is cleared is determined. Taking a three-phase short circuit as an example, during the fault period... Approximately 0.2~0.5 pu, after excision It is approximately 0.60~0.90 pu, with the specific value depending on the fault depth and grid strength.

[0064] 2. Reactive power-voltage linear mapping during the recovery period After the fault is cleared, the reactive power injection from the generator continues to drive the generator terminal voltage back up. A linearized reactive power-voltage sensitivity mapping model is used:

[0065]

[0066] in, Q total ( t )= Q exc ( t )+ Q flex ( t ), which is the sum of the reactive power output of the excitation channel and the reactive power channel; Indicates the unit reactive power of the excitation channel How much voltage can be increased by injection? This is a correction factor.

[0067] The correction factor is normalized to a baseline of 0.70, reflecting the physical characteristic that mild faults are more sensitive to overshoot. The higher the voltage (the less severe the fault), the greater the voltage rise caused by the same reactive power injection, and the easier it is to overshoot.

[0068] Serious malfunction ( =0.70): Correction factor = 1.0, No magnification Moderate fault ( =0.80): Correction factor = 1.143, Magnified by 14.3% This ensures that faults of different severity receive differentiated J2 penalties in the optimization objective, with milder faults incurring higher overshoot costs.

[0069] To ensure safety during the voltage recovery process, two constraints are introduced to prevent the terminal voltage from rising indefinitely.

[0070] Voltage safety constraints:

[0071] Hard cap (physical cutoff):

[0072] 1.15 pu is a common operating threshold for voltage protection in power systems; exceeding this value will trigger overvoltage protection tripping. In the simulation, any voltage value exceeding 1.15 pu is directly truncated.

[0073] Before iteratively solving for the parameters, the forced excitation trigger threshold is determined. The lower bound of the parsing is constrained. The voltage fluctuation range must not be lower than the normal range. The physical lower limit is determined; otherwise, minute voltage fluctuations during normal operation will falsely trigger forced excitation, increasing unnecessary switching losses of the IGBT. Specifically:

[0074] The analytical lower bound is not optimal. A higher... This means earlier triggering of strong excitation and more reactive energy injection, but it also means a longer duration of strong excitation and a greater risk of overshoot. The optimal balance depends on... (Strong excitation intensity) (Strong excitation setup speed) and fault depth (total reactive power required). Therefore Cannot be Fixed, must be included in subsequent optimization and solution steps, only as This serves as a lower bound constraint.

[0075] The slope of the voltage recovery trajectory after forced excitation is removed. It must match the decay rate of the forced excitation component in the stator current; the decay follows... ,exist Internal decay to , The parameterized expression is as follows:

[0076] Notice: In the parameterized expression It is not a constant value, but rather a variable to be optimized in subsequent iterative optimization steps. Instead of directly taking a fixed value, it is used as... A parameterized function embedding optimization framework. Once Determined in SP1, Therefore, it was determined that no separate grid search was needed. k By setting the value to 1.5 to 2.0, the forced excitation component is reduced to 10% to 22% of its initial value, and the residual reactive power no longer drives significant overshoot.

[0077] The remaining four parameters There is cross-coupling between them, for example The contribution to J1 depends on The computational cost of direct search in the six-dimensional full space is unacceptable, therefore a dimensionality reduction decomposition strategy is proposed.

[0078] The six-parameter optimization space is decomposed into two sub-problems according to channel affiliation ( Not as an independent search dimension, but rather following By parameterized expression (Automatic determination)

[0079] The above six control parameters are solved by dimensionality reduction decomposition and alternating iterative optimization. .

[0080] Input: Generator parameters Fault conditions (voltage drop depth) (fault duration), weight Convergence tolerance .

[0081] Output: Optimal parameter vector

[0082] S0 initialization: by As The lower bound of the analytic expression; Set initial parameter values: (Take a slight position above the lower bound to balance trigger speed and margin for false triggers). ; Set the iteration counter: .

[0083] S1 Excitation Channel Subproblem (SP1): fixed ; Solving with 3D mesh scanning :

[0084] constraint: .

[0085] It does not participate in the grid search, but is instead... By parameterized expression Calculate directly.

[0086] Each parameter grid point performs a parameterized transient simulation based on the parameterized model to calculate the normalized weighted comprehensive performance index function (i.e., the J value); where the terminal voltage... Exceeding the maximum overshoot Candidate points are directly eliminated. The grid point with the smallest J and its four adjacent grid points are selected, and local quadratic interpolation is performed to refine the optimal value to avoid grid discretization errors.

[0087] S2 reactive power channel subproblem (SP2): fixed , This was subsequently determined; Solving using 3D mesh scanning ( ):

[0088] in, Constrained .

[0089] S3 convergence criterion: Calculate the objective function value corresponding to the current optimal control parameter set. ; like or If the maximum number of iterations (default is 3) is reached, the iteration terminates and the output is given. The optimal control parameter set is used; otherwise, Return to S1.

[0090] S4 parameter output: Output the final control parameter set and record the sub-target values ​​J1 (cumulative reactive power support energy) and J2 (maximum voltage overshoot) for verification.

[0091] In the grid scans of SP1 and SP2, for candidate points that violate voltage safety constraints (i.e., exist) t Make The penalty function method is used instead of direct elimination to avoid discretization artifacts at the constraint boundary.

[0092] For the initial iteration (m=0), a 3D mesh scan is used to quickly locate the optimal region. In the second and subsequent iterations ( In this process, the grid range is shrunk to a neighborhood of the previous optimal value, and the step size is halved. For example, if the previous round... The search range for this round is then narrowed to [2.5, 2.9], with a step size of 0.05. This strategy significantly reduces the number of simulations required in subsequent iterations (to about 1 / 4 of the initial number) while ensuring that the search accuracy improves with each iteration.

[0093] S4 obtained The result is obtained under a single fault condition. To ensure generality, it is verified one by one under the following fault conditions. The fault type can be selected as three-phase short circuit, single-phase grounding, two-phase short circuit, etc.; the voltage drop depth can be selected as 0.2, 0.3, 0.5, or 0.7 pu; and the fault duration can be selected as 0.05, 0.1, or 0.2 s. If any operating condition does not meet the constraints, that condition is added to the training set, and the control parameters are re-optimized and solved. This verification ensures that the final output parameter set has universality to adapt to various fault scenarios.

[0094] Example 2: The proposed parameter tuning method is verified using a single 640MW synchronous generator infinite bus system. The system topology is that the generator is connected to a 500kV infinite bus via a step-up transformer (0.15pu reactance) and a double-circuit transmission line (0.30pu reactance). The fault point is located on the high-voltage side of the transformer.

[0095] The simulation employs a parameterized equivalent model based on the d-axis transient time constant Td0′, which preserves the core dynamic characteristics of strong excitation current establishment and dual-channel reactive power output. The main parameters are: =6.0s, X d =1.25 pu, =0.32 pu; Rated reactive power capacity of flexible excitation channel Q exc,nom =450Mvar, maximum reactive power capacity of reactive power channel Q flex,max =180Mvar; simulation step size is 0.5ms, total duration is 2.5s. Six fault conditions were selected for optimization verification, covering typical ranges of voltage dip depth 0.20 and duration 0.2s, to comprehensively test the adaptability and versatility of the optimization parameters under different fault severity levels. The fault type is three-phase metallic short circuit on the high-voltage side of the transformer. Severe faults (depth 0.2~0.3pu) allow moderate voltage overshoot, J2≤0.05pu; moderate faults (depth 0.5pu) have extremely high voltage quality requirements, J2≤0.01pu. This differentiated constraint is derived from the actual operation specifications of the power grid, where users have much higher requirements for power supply quality under shallow voltage dips than under deep dips. The optimization weight λ=0.5 (equal weight for reactive power support and overshoot suppression), convergence tolerance ε=10. 4 The parameter optimization range is shown in Table 1: Table 1

[0096] The empirical parameters are: =2.0, =0.85, =0.92, =0.5, =0.2, =1.0. The calculated optimal parameter values ​​are: =2.0, =0.90, =0.95, =0.7, =1.0, =1.0. The simulation results of empirical and optimized parameters under different fault conditions are shown in Table 2: Table 2

[0097] Conclusion: Optimizing the parameters reduced J1 from a mean of 585. Increased to 685 Mvar·s (+17.2%), with a cumulative increase of 603 Mvar·s reactive power across six operating conditions. Meanwhile, all faults' J2 values ​​satisfy the differentiation constraint, with the maximum J2 value being 0.0041 pu.

[0098] Extreme Failure U t =0.2, duration is 0.15s, the influence of empirical parameters and optimized parameters on terminal voltage is as follows: Figure 3 As shown, the effects of empirical parameters and optimized parameters on reactive power output are as follows: Figure 4 As shown. By Figure 3 and Figure 4 It can be seen that the optimized terminal voltage recovers faster, and the reactive power injection is higher than the empirical parameters in both the early and late stages.

[0099] Serious Fault U t =0.3, duration is 0.15s, the influence of empirical parameters and optimized parameters on terminal voltage is as follows: Figure 5 As shown, the effects of empirical parameters and optimized parameters on reactive power output are as follows: Figure 6 As shown, since the voltage drop rates of severe faults and extreme faults are not significantly different, the simulation results for severe faults and extreme faults are not significantly different.

[0100] Medium fault U t =0.5, duration is 0.15s, the influence of empirical parameters and optimized parameters on terminal voltage is as follows: Figure 7 As shown, the effects of empirical parameters and optimized parameters on reactive power output are as follows: Figure 8 As shown.

[0101] Depend on Figures 3-8 It can be seen that the optimized parameters are more effective than the empirical parameters.

[0102] Example 3: This embodiment provides a method for parametric evaluation based on a high-order detailed generator model, as an alternative to the parametric model in Embodiment 1.

[0103] In this embodiment, the simulation of the generator transient behavior adopts a 7th-order generator mathematical model that includes the complete damping winding, the stator winding transient full dynamic process, and the magnetic saturation effect. During the parameter evaluation stage, each candidate control parameter group is directly input into a professional electromagnetic transient simulation software (such as MATLAB / Simulink or PSD-BPA) with this 7th-order model for full-time domain simulation. The generator terminal voltage and reactive power curves output by the simulation are directly extracted, and then the normalized weighted comprehensive performance index function is calculated.

[0104] The advantage of this embodiment is that it fully preserves the dynamic suppression effect of the damping winding in the early stage of high-frequency transients, and provides a more accurate characterization of the microscopic electromagnetic characteristics of large-capacity units. However, due to the loss of the analytical interpretability brought by the parameterization unique in Embodiment 1, and the fact that the evaluation of each set of parameter vectors requires calling the underlying simulation software to run a complete set of nonlinear differential equations, the single calculation takes a long time, making it suitable for high-precision verification scenarios with extremely abundant computing resources.

[0105] Example 4: This embodiment provides a parameter optimization method based on a global heuristic intelligent optimization algorithm, as an alternative to the dimensionality reduction decomposition strategy in Embodiment 1. This embodiment directly... As a whole, a six-dimensional joint vector space is used to introduce Bayesian optimization or genetic algorithm to perform direct search of the six-dimensional full space, with the goal of minimizing the normalized weighted comprehensive index, and to solve for the optimal parameter combination.

[0106] This embodiment completely eliminates the reliance on the physical coupling mechanism of the generator's dual channels, exhibiting strong algorithm versatility and the ability to be extended to complex tuning scenarios with other additional controllers. However, because it does not rely on physical dimensionality reduction, the total number of simulation iterations for the full-space search typically ranges from several thousand to tens of thousands, resulting in lower engineering efficiency compared to the dimensionality reduction decomposition iterative optimization strategy provided in Embodiment 1.

[0107] Example 5: Besides the fixed weighting coefficient λ, this embodiment directly employs a multi-objective optimization algorithm, treating the maximization of reactive power support strength and the minimization of voltage overshoot as two completely independent conflicting optimization objectives. The algorithm ultimately solves for and outputs a uniformly distributed Pareto optimal frontier curve. On this frontier curve, each point represents a set of optimal control parameters under a specific trade-off. Engineers can dynamically select the operating point based on the actual operating specifications of the current power grid: if the current power grid is a weak grid highly sensitive to voltage overshoot, then parameter points close to the minimum value of the voltage overshoot index are selected on the frontier curve; if the current power grid is a coarse-featured system with relaxed voltage quality requirements, then parameter points close to the maximum value of the reactive power support strength index are selected. This embodiment provides extremely rich technical trade-off information, but the computational load and complexity are higher.

[0108] Example 6: To address the highly random and uncertain technical characteristics of fault location, short-circuit impedance, and transition resistance in actual power grid operation, Monte Carlo sampling can be used to randomly generate a large number of fault scenarios for statistical testing, or the Bruker bar optimization method can be used to explicitly consider parameter uncertainties during the design phase, thereby expanding the fixed enumeration fault condition set in Example 1.

[0109] By expanding the fault condition set, the final output parameter set has extremely high resistance to system parameter fluctuations and can adapt to the highly uncertain environment of the new power system under the high proportion of new energy access, but its engineering implementation complexity increases significantly.

[0110] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing and tuning reactive power control parameters of a flexible excitation system, characterized in that, include: From synchronous generator d Starting from the shaft transient equation, an explicit mapping between the control parameters of the flexible excitation system and the transient reactive power is derived and established. The control parameters include the forced excitation trigger threshold, the forced excitation exit threshold, the forced excitation multiple, the reactive power channel input depth, the reactive power channel input rate, the DC side voltage setpoint, and the reference trajectory ramp slope. A normalized weighted comprehensive performance index function is constructed, which includes the reactive power support strength index and voltage overshoot index during the transient process of synchronous generator; Based on the physical channel to which the control parameters belong, the control parameters to be optimized are divided into two subsets. An alternating iterative dimensionality reduction optimization algorithm is used to alternately optimize the control parameters in the two subsets. With the goal of minimizing the normalized weighted comprehensive performance index function, the optimal control parameter set is obtained for the transient control of the flexible excitation system.

2. The method according to claim 1, characterized in that, The transient reactive power parameterization model includes a three-stage excitation voltage control logic: When the terminal voltage is less than or equal to the product of the forced excitation trigger threshold and the rated terminal voltage, it is triggered and enters the forced excitation state. When the terminal voltage is greater than the product of the forced excitation trigger threshold and the rated terminal voltage, and less than or equal to the product of the forced excitation exit threshold and the rated terminal voltage, the forced excitation state is maintained to achieve delayed exit. When the generator terminal voltage exceeds the product of the forced excitation exit threshold and the rated generator terminal voltage, the forced excitation state is exited, and the excitation voltage decays exponentially.

3. The method according to claim 1, characterized in that, The normalized weighted comprehensive performance index function is in the form of: using weight coefficients to weight and combine the normalized reactive power support strength index and the normalized voltage overshoot index, wherein the reactive power support strength index takes a negative weight in the comprehensive index; The reactive power support strength index is the cumulative reactive power injected into the system by the unit within the observation window after the fault is cleared; the voltage overshoot index is the maximum extent by which the generator terminal voltage exceeds the rated generator terminal voltage during the voltage recovery process after the fault is cleared.

4. The method according to claim 3, characterized in that, When calculating the voltage overshoot index, a reactive power-voltage sensitivity mapping model is used to generate the recovery trajectory of the generator terminal voltage. A fault severity correction factor is introduced into the mapping model, and a safety upper limit truncation constraint is applied to the generated generator terminal voltage trajectory. The fault severity correction factor is determined by the ratio of the initial voltage after fault clearance to the reference recovery voltage.

5. The method according to claim 1, characterized in that, Before parameter tuning, the following is also included: The physical lower bound constraint of the forced excitation trigger threshold is determined by analytical derivation, and the physical lower bound constraint is determined by the voltage fluctuation amplitude during normal operation. The reference trajectory ramp slope is parameterized as an explicit functional expression with respect to the forced excitation trigger threshold, thereby eliminating the independent search dimension of the reference trajectory ramp slope in parameter tuning.

6. The method according to claim 5, characterized in that, The physical channels include excitation voltage channels and reactive current channels; The control parameter subset includes a first subset belonging to the excitation voltage channel and a second subset belonging to the reactive current channel; The control parameters for the first subset include the excitation multiple, the DC side voltage setpoint, and the excitation trigger threshold. The control parameters of the second subset include reactive power channel activation depth, reactive power channel activation rate, and forced excitation exit threshold.

7. The method according to claim 6, characterized in that, The alternating iterative dimensionality reduction optimization algorithm used to alternately tune the control parameters in the two subsets includes: Initialize all control parameters in the first and second subsets; With the second subset of control parameters fixed, the first subset of control parameters that minimizes the normalized weighted comprehensive performance index function is solved by scanning within a constrained 3D mesh. The first subset of control parameters obtained from the current iteration round is fixed, and the second subset of control parameters that minimize the normalized weighted comprehensive performance index function is solved within the constrained 3D mesh. The first subset of control parameters and the second subset of control parameters are solved alternately until the normalized weighted comprehensive performance index function converges or the maximum number of iterations is reached, so as to obtain the optimal set of control parameters.

8. The method according to claim 7, characterized in that, The alternating iterative dimensionality reduction optimization algorithm also includes a grid resolution adaptive adjustment strategy: after the initial iteration, the grid search range of subsequent iterations is narrowed to the neighborhood of the previous round's optimal value, and the search step size is reduced.

9. The method according to claim 7, characterized in that, During the mesh scan solution of the first subset control parameters and the second subset control parameters, the penalty function method is used to process the candidate parameter points that violate the voltage safety constraints. The voltage safety constraint condition is that the terminal voltage is less than or equal to the maximum voltage.

10. The method according to claim 1, characterized in that, The method further includes: performing transient simulation verification on the optimal control parameter set in the fault condition set, wherein the fault condition set includes multiple fault scenarios determined by fault type, voltage drop depth and fault duration; if there is a condition that does not meet the constraints, the condition is added to the training set and the control parameters are optimized and solved again.

Images