New energy station and grid-constructing unit joint optimization configuration method

Abstract

The application provides a new energy field station and network construction unit combined optimization configuration method, and belongs to the technical field of new energy grid connection planning and operation control of power systems. The application quantifies the stable demand of the system by constructing a four-dimensional dynamic grid connection strength deficiency tensor of nodes, frequency bands, dynamic directions and operation scenarios, constructs a conservatively provable grid connection support ability polyhedron to depict the unit support boundary in combination with physical constraints, establishes support contract constraints through support propagation operators and inter-unit interaction loss calculation, carries out unit configuration and control parameter collaborative optimization, and outputs the final configuration scheme through iterative correction of the back-feeding model of the cutting constraint generated by the instability counterexample. The application can improve the accuracy of grid resource configuration and the adaptability of weak grid working conditions, and guarantee the engineering realizability of the configuration scheme.

Description

Technical Field

[0001] This invention belongs to the field of power system new energy grid connection planning and operation control technology, and particularly relates to a method for joint optimization configuration of new energy power plants and grid-connected generating units. Background Technology

[0002] As the proportion of power electronic interface power sources such as wind power and photovoltaics in the power system continues to increase, the functional positioning of new energy power plants is gradually shifting from simple power injection units to active grid-connected units that need to undertake tasks such as voltage establishment, frequency support, oscillation suppression, and fault recovery. Grid-connected converters typically adopt a control architecture with phase-locked loops, an inner current loop, and an outer power loop. Their operational stability is significantly dependent on the strength of the external power grid. Under conditions of weak grids and multiple converters in parallel coupling, they are prone to problems such as synchronous instability, wide-frequency oscillations, and difficulties in fault recovery. Grid-connected converters can actively establish grid voltage and frequency by forming internal voltage phase and amplitude references through droop control, virtual synchronous machines, power synchronization control, or virtual oscillators. However, their available support capacity is limited by multiple constraints, including connection location, control parameters, current limiting, thermal margin of power devices, DC-side energy reserves, and primary energy reserve levels.

[0003] Existing research on grid resource allocation primarily uses grid capacity ratio, system frequency deviation, or single characteristic value damping ratio as constraints, focusing on addressing two main issues: how much grid capacity to allocate and at which nodes. However, this approach has three limitations in practical applications. First, single short-circuit ratio or minimum damping ratio indicators cannot distinguish between different types of stability requirements, such as low-frequency synchronous support, mid-frequency control coupling suppression, and fault transient support, making precise allocation of grid resources difficult. Second, equipment type, capacity, and access location are usually determined in advance, with virtual inertia, damping coefficient, droop coefficient, and phase-locked loop parameters of grid-connected units subsequently tuned empirically. This can easily lead to discrepancies between the planning results and the realizable domain of control parameters, making it difficult to implement the planning scheme. Third, conventional typical scenario verification methods cannot fully cover the combined risks arising from continuous changes in grid impedance, equipment outages, current saturation, and fault recovery sequences, easily overlooking low-probability, high-loss extreme weak grid conditions. Summary of the Invention

[0004] To address the shortcomings and deficiencies of existing technologies in grid-based resource allocation, such as the use of single evaluation indicators, disconnect between planning and control, and insufficient coverage of extreme operating conditions, this invention provides a joint optimization configuration method for new energy power plants and grid-connected generating units. First, a unified multi-port dynamic model of the new energy power plant is established, constructing a four-dimensional dynamic grid connection strength deficit tensor that includes nodes, frequency bands, dynamic directions, and operating scenarios, thereby accurately characterizing the system's stability support requirements in different dimensions. Second, considering physical constraints such as unit current, heat, and energy, a provably conservative grid-connection support capability polyhedron is constructed through key operating point coverage and differential convex internal approximation, quantifying the available support boundary of the units. Furthermore, the support propagation operator and the control coupling interaction loss between units are calculated to establish a support contract constraint for the strength deficit corresponding to the effective support coverage dimension. Finally, a collaborative optimization model incorporating this contract constraint is solved to obtain candidate schemes, and stability verification is performed. Identified instability counterexamples are transformed into cutting constraints and fed back into the optimization model for iterative correction until the stability requirements are met, at which point the final configuration scheme is output. This invention achieves full-process collaboration from demand identification, capability certification, propagation quantification to configuration decision-making and closed-loop verification, providing a systematic solution for the grid transformation and stable operation of high-proportion new energy power stations.

[0005] The present invention specifically adopts the following technical solution:

[0006] To address the shortcomings of existing grid resource allocation methods, such as single evaluation indicators, capability characterization detached from physical constraints, sequential design of planning and control, and insufficient coverage of extreme weak grid conditions, this invention provides a joint optimization allocation method for grid-connected and grid-connected generating units at new energy power plants. This method improves the accuracy and engineering feasibility of grid resource allocation while meeting the multi-dimensional stability requirements of the system. The method includes: establishing a unified multi-port dynamic model for grid-connected, grid-connected, and collection networks at new energy power plants; constructing a four-dimensional dynamic grid connection strength deficit tensor for nodes, frequency bands, dynamic direction, and operating scenarios to form a deficit distribution characterizing the system's stability support requirements; and combining unit physical and operational constraints to achieve coverage of key operating points. A provably conservative network support capability polyhedron is constructed using differential convex internal approximation to obtain a support capability model characterizing the available support boundary of the generating units. The support propagation operator and the control coupling interaction loss between generating units are calculated to establish support contract constraints for the corresponding dimensional strength gaps in effective support coverage. Using the generating unit control type, capacity, access location, multi-stage operation role, and control parameters as decision variables, a configuration and control co-optimization model incorporating the aforementioned support contract constraints is solved to obtain candidate configuration schemes. The stability of the candidate configuration schemes is verified. If an unstable counterexample is identified, a corresponding cutting constraint is generated and fed back into the co-optimization model for iterative correction until the stability requirements are met, at which point the final configuration scheme is output.

[0007] Furthermore, when constructing the four-dimensional dynamic grid-connected strength deficit tensor, clustering is performed based on the system's closed-loop eigenvalue frequency and the control loop's open-loop crossover frequency. Multiple dynamic frequency bands are adaptively divided, and the band boundaries can be dynamically updated according to the operating mode and control parameters, adapting to the bandwidth differences of different machine models and controllers, thus focusing the strength assessment on the system's actual dominant mode. Stability characteristics of different dynamic action directions are extracted through the directional projection matrix, decoupling the dynamic interactions of different physical directions. The directional stability margin for each dimension is calculated based on the minimum singular value of the projected loop gain matrix; this minimum singular value corresponds to the weakest stable channel within the direction. The calculated directional stability margin is compared with a preset safety margin to obtain the strength deficit in the corresponding dimension.

[0008] Furthermore, when constructing the polyhedron for network support capabilities, a candidate pool of key operating points, including operational quantiles, physical critical points, modal critical points, and topological boundary points, is selected. The weighted maximum-minimum distance method is used to screen key operating points, taking into account operational probability distribution, physical boundary constraints, and dynamic criticality, ensuring that the selected key points effectively cover the actual operating space of the unit. Differential convex decomposition is performed on the nonlinear constraints within the neighborhood of each key operating point. Combined with the upper bound of the interval Hessian matrix, a convex upper bound for the constraints is generated, and projected to obtain a locally conservative polyhedron, transforming the complex nonlinear physical boundary into a linear inequality form that can be directly used for optimization. The conservatism of the polyhedron is ensured through a three-layer mechanism: analytical upper bound verification, numerical back-substitution verification, and engineering boundary contraction. Analytically, the convex upper bound function is guaranteed to be no less than the original nonlinear constraint function within the trust region. Numerically, the original constraints are back-substituted for all vertices and edge midpoints of the polyhedron, and the maximum remainder term within the unit is verified through interval arithmetic. At the engineering level, the polyhedron boundary is contracted by a fixed ratio, reserving additional safety margins. If the back-substitution check finds a constraint violation, the corresponding element is bisected along the direction of maximum curvature and then re-convexized.

[0009] Furthermore, the propagation operator is used to derive the analytical sensitivity of eigenvalues ​​to support components based on the left and right eigenvectors of the critical mode. Combined with the generation of minimum singular value perturbation relationships, a quantitative mapping relationship is established from control parameter changes to the stability margin increment of the target node. A second-order correction term is introduced to compensate for linear truncation errors under large-scale parameter changes, improving the propagation calculation accuracy in scenarios with large parameter adjustments. The control coupling interaction loss is obtained by quantizing the maximum singular value of node mutual impedance and unit control closed-loop gain. It characterizes the worst-case interaction channel strength under control loop coupling when multiple grid-connected units are running simultaneously, and is used to reserve margin in the configuration to suppress the risk of single-unit stability but multi-unit parallel instability.

[0010] Furthermore, the specific constraint of the support contract is as follows: the total amount of network support propagated through the network, after deducting the control coupling and interaction loss between units, shall not be less than the sum of the strength deficit and reserved safety margin under the corresponding node, frequency band, dynamic action direction and operating scenario, so as to achieve the matching of stable demand and network support in four dimensions: spatial location, spectrum band, physical direction and operating scenario.

[0011] Furthermore, the unit control types are divided into three categories: fixed grid-following type, fixed grid-connecting type, and dual-mode (grid-following and grid-connecting). The configuration variables of the three types satisfy mutual exclusion constraints, ensuring that a single unit corresponds to only one configuration type at any given time. For dual-mode units, three operational roles are set: normal operation, fault process, and fault recovery. Phase pre-synchronization constraints, voltage pre-synchronization constraints, power pre-synchronization constraints, and minimum dwell time constraints are configured for mode switching. These constraints limit the power angle, voltage, and power jump amplitude and the shortest operating time during the switching process, respectively. This ensures that the planned role reconfiguration strategy can be smoothly executed by the actual controller, avoiding system disturbances caused by switching shocks.

[0012] Furthermore, the collaborative optimization model employs the Wasserstein fuzzy optimization framework to handle operational uncertainties. It constructs an empirical distribution using historical samples and establishes a Wasserstein fuzzy set. The conditional value of risk under the worst-case distribution is used as a stability risk indicator. Under limited sample conditions, it balances historical operational patterns with distributional bias protection, reducing the risk of missed detections in low-probability, high-loss extreme operating conditions. The radius of the Wasserstein fuzzy set is determined through initial resampling combined with validation set calibration. The minimum radius is selected while meeting the target risk level, balancing the robustness and economy of the configuration scheme and avoiding overly conservative or insufficiently risky approaches derived from empirical values.

[0013] Furthermore, the collaborative optimization model embeds local dissipative stability certificate constraints, constructs linear matrix inequalities based on the Kalman-Yakubovich-Popov lemma, transforms the local stability conditions of the control parameters into directly solvable convex constraint forms, and limits the optimized control parameters to be within the stable feasible domain, thereby realizing parallel constraints on equipment configuration and control parameters and avoiding the disconnect between the planning scheme and the control feasible domain.

[0014] Furthermore, instability counterexamples are categorized into four types: phase-locked loop (PLL) lockout, power angle instability of network units, inter-network coupling oscillations, and mode switching impacts. Each type of instability counterexample corresponds to a generation mechanism-matched cutting constraint, specifically excluding configuration regions that cause the corresponding instability problem. Each cutting constraint only takes effect within the corresponding parameter's trust region. If a new scheme is outside the trust region, the cutting constraint is regenerated to avoid incorrectly excluding feasible configuration regions far from the counterexample. Iterative convergence simultaneously satisfies the triple criteria of upper and lower bound gaps, counterexample violation degree, and discrete configuration invariance. Discrete configuration invariance means that the discrete configuration schemes remain consistent across multiple iterations without any new counterexamples with different topologies. Additionally, an anti-dead-loop termination mechanism is implemented, including a maximum iteration limit, duplicate cutting deduplication, invalid configuration exclusion, adaptive trust region shrinkage, and cutting aging deletion, ensuring stable convergence of the algorithm within a finite number of iterations.

[0015] The present invention also provides a non-transitory computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned method for joint optimization configuration of grid-connected and grid-connected generating units in new energy power plants.

[0016] Compared to existing technologies, this invention and its preferred solutions extend grid connection intensity assessment from a single scalar to a multi-dimensional tensor form, enabling precise differentiation of different types of stability requirements and accurate allocation of grid resources to avoid support mismatches. It expands the characterization of grid support capabilities from rated capacity to conservative boundaries under multiple constraints, closely aligning with the actual physical limits of equipment and effectively improving the engineering feasibility of planning schemes. It achieves coordinated optimization of unit configuration and control parameters, avoiding implementation deviations caused by sequential planning and control design, and enhances the scheme's adaptability to operational uncertainties through distributed processing. It constructs a counterexample-driven closed-loop optimization mechanism, effectively covering extreme weak grid conditions and reducing the risk of operational instability. It can improve the utilization efficiency of grid resources and optimize the overall economic efficiency of configuration while meeting system stability requirements. Attached Figure Description

[0017] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0018] Figure 1 This is a diagram illustrating the overall technical architecture of the joint optimization configuration method according to an embodiment of the present invention.

[0019] Figure 2 This is a flowchart illustrating the implementation steps of the joint optimization configuration method according to an embodiment of the present invention. Detailed Implementation

[0020] To make the features and advantages of the present invention more apparent and understandable, specific embodiments are described below in conjunction with the accompanying drawings:

[0021] I. Overall Methodological Framework

[0022] This embodiment proposes a joint optimization configuration method for new energy power plants and grid-connected units (grid-connected and grid-connected units), especially a joint optimization configuration method for grid-connected / grid-connected units driven by dynamic grid connection intensity deficit. The core idea is to start from the system stability requirements, sequentially identify the support deficits of each node, frequency band, and dynamic direction of the power plant, characterize the available grid-connected support capability of the units under physical constraints, quantify the actual stability contribution of the support after propagation through the power plant network, and finally achieve coordinated decision-making on unit type, capacity, location, operating role, and control parameters through optimized matching, and form a closed-loop correction mechanism through instability counterexample feedback.

[0023] The overall technical framework of this embodiment is as follows: Figure 1 As shown, a unified multi-port model of the power station-grid type units, the grid-connected type units, and the collection network is first established. Based on this, the dynamic grid connection strength deficit for each node, frequency band, dynamic direction, and scenario is calculated. Simultaneously, a polyhedron of the unit's grid-connection support capability is constructed in conjunction with physical constraints, and the actual contribution of support after propagation through the network is calculated. Then, support contract constraints are established based on the strength deficit, and collaborative optimization of unit configuration and control parameters is carried out. The configuration results are verified through stability certificates and transient simulations. Instability counterexamples identified in the transient verification can be transformed into explicit optimization constraints and returned to the optimization stage. After correction, the solution is re-solved to ensure that the final configuration scheme meets the stability requirements under various operating conditions.

[0024] II. Unified Dynamic Modeling of New Energy Power Stations

[0025] 2.1 Station Structure and Uncertain Parameters

[0026] The new energy power station includes A group of converter equivalent machines and There are n electrical nodes, and the set of nodes and the set of lines are denoted as follows: and Candidate fleets can be configured into three types: fixed network-following, fixed network-building, or a dual-mode network-following / network-building. All uncertainties during station operation are uniformly represented as an uncertain parameter vector:

[0027]

[0028] In the formula, and These are the active and reactive power operating points of the new energy source, respectively. and Equivalent resistance and reactance of the external power grid The voltage at the grid connection point. For control or communication delay, This refers to the on / off status vectors of lines, transformers, and generating units.

[0029] 2.2 Unified multi-port model with network topology / structure

[0030] In a synchronously rotating coordinate system, the first The port small-signal model of a network cluster can be represented as:

[0031]

[0032] In the formula, For small signal disturbances in port current. For small signal disturbances in port voltage. To match the port admittance matrix of the network cluster, The control parameter vector for the grid type includes the phase-locked loop bandwidth, current loop proportional-integral coefficient, reactive power voltage control coefficient, and current limiting parameter.

[0033] No. The port small-signal model of a network-type cluster can be represented as:

[0034]

[0035] In the formula, For the port admittance matrix of a network-type cluster, The control parameter vector for the network structure must contain at least virtual inertia. Damping coefficient Active droop coefficient reactive power droop coefficient Virtual resistance and virtual inductance .

[0036] definition For the cluster Given the equivalent network capacity ratio, the unified port admittance of the cluster can be expressed as a weighted sum of the admittances of the network-based and network-structured types:

[0037]

[0038] Assemble the port admittances of the entire cluster into a block diagonal matrix. The nodal impedance matrix is ​​obtained by combining the parameters of the line, transformer, and external power grid. The multi-port loop gain matrix of the power station can be constructed:

[0039]

[0040] In the formula, This is a joint decision vector that includes unit control type, capacity, access location, operating role, and control parameters. The impedance model is mainly used to describe the system's broadband dynamic interaction characteristics, while the corresponding state-space model is used for eigenvalue sensitivity calculation, dissipative stability certificate construction, and transient initialization. The two types of models maintain consistency through the same operating point and parameter mapping relationship.

[0041] III. Dynamic Grid Connection Intensity Shortfall Tensor

[0042] 3.1 Adaptive Dynamic Frequency Band Allocation

[0043] Fixed frequency band allocation is difficult to adapt to the bandwidth differences between different models and controllers. Therefore, this embodiment dynamically determines the frequency band boundary based on the closed-loop characteristic frequency, mode participation factor, and open-loop crossover frequency of the control loop. Let the candidate mode set be... The set of open-loop crossover frequencies of the control loop is Then the set of frequency band boundaries can be defined as:

[0044]

[0045] In the formula, This is a clustering algorithm (an unsupervised clustering algorithm used to group frequency points to divide frequency bands, commonly such as K-means clustering). Through clustering, three typical frequency bands can be formed: low-frequency synchronization band, mid-frequency control coupling band, and high-frequency current-delay band. This division method is dynamically updated according to the operating mode and control parameters, so that subsequent intensity indicators can focus on the actual dominant mode of the system, rather than a preset fixed frequency range.

[0046] 3.2 Directional stability margin and strength deficit

[0047] For nodes ,frequency band , Dynamic direction of action and operating scenarios Define the corresponding directional projection matrix. and Then the directional stability margin can be expressed as the minimum of the minimum singular values ​​of the projected loop matrix within the corresponding frequency band:

[0048]

[0049] In the formula, It is the smallest singular value of the matrix. For the scene The frequency domain loop gain matrix, For the first Each frequency band corresponds to a frequency range. The dynamic action direction includes at least the active-frequency direction, the reactive-voltage direction, the power angle-active direction, the phase-locked loop-current direction, and the current saturation recovery direction.

[0050] Given a preset safety and stability margin requirement Dynamic grid connection intensity deficit can be defined for the corresponding dimension:

[0051]

[0052] The deficiencies of all nodes, frequency bands, directions, and scenarios are combined into a fourth-order tensor. Unlike the traditional short-circuit ratio index, this deficit tensor can not only determine whether the entire power station is in a weak network state, but also clearly indicate the location of the weak node, the dominant frequency band, the type of missing dynamic support, and the corresponding operating scenario, providing a basis for the precise allocation of network resources.

[0053] IV. Network Support Capability Polyhedron

[0054] 4.1 Network Support Vectors and Physical Constraints

[0055] For candidate units Define its frequency band The following are the network support vectors:

[0056]

[0057] In the formula, For equivalent synchronous stiffness, For equivalent damping, For voltage stiffness, For incremental dissipation margin, To ensure sustainable support over time.

[0058] The network support vector is determined by both control parameters and equipment status, and is also subject to three types of physical constraints. The first type is current and junction temperature constraints:

[0059]

[0060] In the formula, , These are the instantaneous values ​​of the d-axis and q-axis currents, respectively. This is the upper limit of the allowable current for the unit. For power device junction temperature, This represents the upper limit of the allowable junction temperature.

[0061] The second category is energy and active power reserve constraints:

[0062]

[0063] In the formula, For the DC side available energy, To support incremental growth, This is the active power reserve capacity.

[0064] By conservatively convexizing the aforementioned nonlinear physical boundaries at several key operating points, the unit can be obtained. The network support capability of the polyhedron:

[0065]

[0066] In the formula, It is a multifaceted system number matrix. The right-hand term of the polyhedron is determined by parameters such as the upper limit of current, junction temperature margin, available DC energy, and active power reserve. This polyhedron can reflect the differences in the support capabilities of units with the same rated capacity but different current margins, heat capacity, DC energy, and primary energy reserves, avoiding the limitation of measuring grid support capabilities solely by rated capacity.

[0067] 4.2 Critical Operating Point Selection Principles

[0068] Candidate aircraft group The original running-control variables are First, the variables in each dimension are dimensionless based on the median and interquartile range. Then, a candidate pool of critical operating points is constructed. The candidate pool includes the 5%, 50%, and 95% quantiles of historical samples, power flow boundary points, current critical points with a current ratio greater than or equal to 0.90, thermal critical points with a junction temperature distance upper limit of less than 5°C, modal critical points with a damping ratio close to the safety threshold, and topological boundary points after the commissioning or decommissioning of lines or units.

[0069] Key points are selected point by point using the weighted maximum-minimum distance method:

[0070]

[0071] In the formula, For the critical runpoint candidate pool, This is a weighted diagonal matrix, where the diagonal elements are the reciprocals of the robustness scales of each variable. The first critical point is selected as the rated operating point, and the superscript... This represents the selected key running points. Iterative selection continues until the coverage radius (the maximum normalized distance between the selected key point and all sample points in the candidate pool, used to measure the coverage of the running space by the key point) meets the threshold requirement. As a preferred implementation, the upper limit of the coverage radius can be 0.20~0.30, and all points with constraint activity (characterizing the degree to which a running point is close to the constraint boundary; the closer the activity is to 1, the closer the running point is to the constraint critical state) exceeding 0.90 or the modal participation factor exceeding 0.15 must be retained to balance running probability, physical boundaries, and dynamic criticality.

[0072] 4.3 Differential Convex Approximation and Conservatism Guarantee

[0073] The current, junction temperature, energy, and local stability region constraints are uniformly expressed as... In the form of a differential convex decomposition of the constraint function in the neighborhood of each key point, a difference function is performed. ,in and All are convex functions. For A first-order lower bound approximation is used, and an interval Hessian matrix upper bound is added to the unmodeled second-order terms. We can obtain the convex upper bound of the original constraint:

[0074]

[0075] In the formula, For the first Key operating points, for The gradient vector.

[0076] By applying convex upper bound constraints within the trust region and projecting the affine mapping between the support variables and the original running-control variables, the second-order cone constraint, and the linear device boundary into linear inequality forms, the locally conservative polyhedron for the corresponding partition can be obtained.

[0077]

[0078] In the formula, For the first Locally conservative polyhedra of each partition, Binary variables are selected for partitioning. During optimization, binary variables are used to select partitions that match the current scenario, ensuring that feasible points always belong to a certified inner approximation polyhedron, thus avoiding conservative distortion caused by directly taking the convex hull of different partitions.

[0079] The conservatism of the network-supported polyhedron is guaranteed by a three-layer mechanism: At the analytical level, the convex upper bound function is always greater than or equal to the original nonlinear constraint function within the trust region; at the numerical level, the original nonlinear constraints are back-substituted and verified for all vertices and midpoints of each edge of the polyhedron, and the maximum remainder term within each element is verified using interval arithmetic; at the engineering level, the polyhedron boundary is uniformly shrunk according to a fixed ratio. If a constraint violation is found during the back-substitution process, the corresponding element is bisected along the direction of maximum curvature and then re-convexized. When the margin of the convex upper bound is not less than the upper bound of the model error, any support vector within the polyhedron can satisfy the original nonlinear constraint requirements.

[0080] 4.4 Calculation Examples of Polyhedral Parameters under Typical Working Conditions

[0081] Taking a standardized 100MVA, 35kV unit as an example, the rated operating point active power is 0.80 per unit, reactive power is 0.10 per unit, the upper limit of current is 1.20 per unit, the junction temperature margin is 25℃, the available energy on the DC side is 8MJ, and the active power reserve is 0.15 per unit. A normalized support vector is defined, with each component corresponding to the normalized values ​​of synchronization stiffness, damping, voltage stiffness, buffer energy, and support time. The upper limit of each normalized component is [0.70, 0.75, 0.80, 0.60, 0.80]. After calculation using the above internal approximation method and performing 5% boundary contraction, the linear constraint form of the polyhedron can be obtained.

[0082] The coefficient row vector corresponding to the peak current constraint is [1.00, 0.15, 0.70, 0.10, 0.00], with a right-hand value of 0.92, which physically means limiting the current peak exceeding the limit caused by the superposition of synchronization stiffness and voltage stiffness; the coefficient row vector corresponding to the thermal margin constraint is [0.20, 0.90, 0.15, 0.55, 0.10], with a right-hand value of 0.85, which physically means limiting the device loss temperature rise caused by damping injection and continuous energy release; the coefficient row vector corresponding to the energy and reserve constraint is [0.35, 0.20, 0.10, 1.00, 0.65], with a right-hand value of 0.88, which physically means limiting the consumption of DC side energy and primary energy reserve; the coefficient row vector corresponding to the power angle safety constraint is [0.80, 0.35, 0.00, 0.25, [0.00], with a right-hand value of 0.90, physically meaning to ensure the deceleration area margin after fault clearance; the coefficient row vector corresponding to the local stability domain constraint is [0.15, 0.75, 0.55, 0.00, 0.10], with a right-hand value of 0.86, physically meaning to ensure the local dissipation rate and damping ratio requirements.

[0083] The normalized support vector is taken as [0.35, 0.30, 0.45, 0.25, 0.35], corresponding to the actual parameters of synchronous stiffness 1.40 per radian, damping 0.75 per radian per second, voltage stiffness 1.35 per radian per unit, buffer energy 2.0 MJ, and support time 1.75 seconds. Substituting these parameters into the constraint calculation, the residual margins of the five constraints are 0.185, 0.270, 0.175, 0.453, and 0.300, respectively, with a minimum margin of 0.175. Substituting this point back into the original model for peak current, thermal network, and energy verification, all constraints are satisfied. Therefore, this example can be used as a benchmark input for program unit testing. The above values ​​are only a standardized example provided for understanding this method. In actual engineering applications, the polyhedral parameters should be calculated and determined based on the specific site's equipment test parameters and operating data.

[0084] V. Interaction Loss of Support Propagation Operator and Network Construction

[0085] The network support capability needs to be propagated through the power collection network of the substation before it can actually contribute to the stability margin of the target nodes. This applies to the system's critical modes. Its relative to the supporting component The sensitivity can be calculated using the eigenvalue perturbation formula:

[0086]

[0087] In the formula, and These are the right and left eigenvectors of the critical mode, respectively. Let be the system state space matrix.

[0088] By combining the differentiable perturbation relation of minimum singular values, a scenario can be constructed. The propagation coefficient below:

[0089]

[0090] The propagation coefficients corresponding to all nodes, units, frequency bands, directions, and scenarios are combined to form the supporting propagation operator. The incremental node stability margin brought about by network support can be expressed as:

[0091]

[0092] In the formula, This is a second-order correction term used to correct the calculation error of linear sensitivity under a wide range of parameter variations.

[0093] To describe the negative effects of power loop, droop loop, and virtual impedance coupling when multiple grid-connected units are put into operation simultaneously, a frequency band is defined. Lower unit With the unit Network interaction loss between them:

[0094]

[0095] In the formula, The largest singular value of the matrix. For nodes With nodes mutual impedance between This is the closed-loop transfer function from control parameters to port response. The interactive loss term is used to suppress the risk of inter-machine oscillations that occur when multiple machines are running simultaneously but the single machine is stable.

[0096] Example of derivation of control parameters for propagation operator

[0097] Taking the active-frequency channel of the virtual synchronous machine as an example, its locally linearized model satisfies the second-order oscillation equation, and the eigenvalues ​​satisfy the quadratic equation relationship. From this, the analytical sensitivity of the eigenvalues ​​to the damping coefficient and synchronization stiffness can be derived:

[0098]

[0099] In the formula, The equivalent inertia coefficient corresponding to the virtual inertia. The damping coefficient is... For synchronous stiffness.

[0100] make For local attenuation margin, with Indicates frequency band Lower unit For the target node Given the network-modal participation coefficient, the support propagation coefficient corresponding to the damping parameter can be expressed as the product of the participation coefficient and the real part of the damping sensitivity. The network-modal participation coefficient is obtained by normalizing the mutual impedance, directional projection vector, and left and right eigenvectors of the critical mode.

[0101]

[0102] In the formula, For node observation vectors, This is the unit's modal participation vector.

[0103] Take typical parameters , , The eigenvalues ​​can be calculated to be -0.150 ± j0.8529, and the real part of the damping sensitivity is 0.125. If the network-modal participation coefficient corresponding to the target node is 0.62, then the corresponding propagation coefficient is 0.0775, meaning that when the damping coefficient increases by 0.4, the first-order prediction increment of the target node's attenuation margin is 0.031. As a preferred implementation, if the deviation between the high-fidelity recalculation result and the linear prediction exceeds 5%, the parameter confidence region is reduced and the second-order correction term is updated to ensure the accuracy of the propagation calculation.

[0104] VI. Strength Deficit - Network Support Contract

[0105] Based on the aforementioned quantitative results of the deficit tensor, support capacity, and propagation loss, a support contract constraint for supply and demand matching can be established as the core constraint of the optimization model.

[0106] set up Representing a scene Lower unit Whether to assume the role of network construction To reserve a safety margin, the interaction loss corresponding to different dynamic directions and operating scenarios can be calculated by substituting the direction projection matrix and scene parameters, while keeping the dimension aligned with the dynamic grid connection strength deficit tensor. To extend the network construction interaction loss to include the dynamic direction r and the runtime scenario k, supporting contract constraints can be established:

[0107]

[0108] The physical meaning of this constraint is that the effective support, after propagation through the field network and deducting the interaction losses between network units, must cover the strength gaps and reserved safety margins under the corresponding nodes, frequency bands, dynamic directions, and operating scenarios. Compared with the traditional capacity constraint that only requires the total network capacity to reach a threshold, the support contract simultaneously completes location matching, spectrum matching, dynamic direction matching, and operating scenario matching, and is the core constraint of the joint configuration model.

[0109] VII. Co-optimization model for configuration and control parameters

[0110] 7.1 Decision Variables and Multi-Stage Roles

[0111] Set a fixed network type variable for each candidate unit. Fixed network type variables and dual-mode variables The three types of variables satisfy mutual exclusion constraints:

[0112]

[0113] For dual-mode units, role variables are set for three phases: normal operation, fault process, and fault recovery. The mode transition process must meet pre-synchronization constraints for phase, voltage, and power.

[0114]

[0115] In the formula, , , These represent the power angle, voltage, and power jump values ​​before and after the mode switch, respectively, with the superscript 'max' indicating the allowable upper limit for the corresponding jump. Minimum dwell time, controller integral state mapping, and current limiting state inheritance constraints are also set to ensure that the role reconfiguration strategy output by the planning model can be executed by the actual controller.

[0116] 7.2 Optimization Objectives

[0117] The joint decision vector can be represented as a combination of unit type variables, stage role variables, network capacity variables, support vectors, network-following control parameters, and network-structure control parameters:

[0118]

[0119] Taking into account equipment upgrades, network backup, operational losses, role switching, control coupling, and stability risks, the objective function is set to minimize the weighted sum of total cost and risk.

[0120]

[0121] In the formula, To increase the investment cost of new network-type equipment, Costs of dual-mode retrofitting of existing generating units, For operating losses and energy curtailment costs, To cover the backup costs of network construction, The price of role transformation To mitigate the risks and costs associated with inter-grid coupling between network units, To stabilize risk indicators, This is the risk weighting coefficient.

[0122] 7.3 Wasserstein risk constraints

[0123] From historical samples Empirical distributions can be constructed:

[0124]

[0125] In the formula, Let be the Dirac function corresponding to the sample point.

[0126] Centered on this empirical distribution, establish a Wasserstein fuzzy set:

[0127]

[0128] In the formula, The first-order Wasserstein distance. Let be the radius of the fuzzy set.

[0129] Define the amount of contract breach. The system stability risk is then represented by the conditional value of risk under the worst-case distribution:

[0130]

[0131] In the formula, Confidence level The model utilizes empirical data and protects against distribution bias under limited sample conditions, thus avoiding the omission of low-probability, high-loss weak network conditions by typical scenario methods.

[0132] Fuzzy set radius selection principle

[0133] The radius of the fuzzy set is not set directly based on experience, but is determined using a four-step method: robust normalization, effective sample size correction, bootstrap initial selection, and validation set calibration. First, the uncertain variables are robustly normalized using the median and interquartile range. For samples with time correlation, the effective sample size is corrected using the autocorrelation coefficient. Then... With replacement, resampling is performed, and the first-order Wasserstein distance between each resampled empirical distribution and the original empirical distribution is calculated. Quantiles as the initial radius:

[0134]

[0135] In the formula, For the first The empirical distribution obtained by resampling.

[0136] On the independent validation set, from the candidate radii that are not less than the initial radius, the minimum value that makes the upper confidence bound of the constraint violation rate not exceed the target risk level is selected as the optimal radius, to avoid unnecessary investment conservatism caused by an excessively large radius.

[0137]

[0138] In the formula, The upper confidence bound for the violation rate, This represents the constraint violation rate on the validation set.

[0139] Taking a typical scenario with a normalized dimension of 6, 240 original samples, 168 valid samples, 1000 resampling iterations, and a quantile level of 0.95 as an example, Bootstrap initially selects an initial radius of 0.118. With a target risk level of 5%, the optimal radius is selected as 0.13 after validation set calibration. The validation set violation rate corresponding to a candidate radius of 0.08 is 8.8%, and the configuration cost increment is 0.9%, exceeding the risk limit; the violation rate corresponding to a candidate radius of 0.10 is 6.7%, and the cost increment is 1.6%, still posing a risk; the violation rate corresponding to a candidate radius of 0.13 is 3.8%, and the cost increment is 2.8%, meeting the risk requirement; the violation rate corresponding to a candidate radius of 0.16 is 2.9%, and the cost increment is 4.5%, which is overly conservative. As a preferred implementation, in actual engineering, cost-risk sensitivity within a radius range of ±20% can be analyzed simultaneously, and recalibration is performed if the sample distribution drifts.

[0140] 7.4 Local dissipation stability certificate

[0141] To ensure that the optimized control parameters have an achievable stability region, the first... The Taiwanese unit establishes an incremental dissipation inequality:

[0142]

[0143] In the formula, For energy storage function, Here is the dissipation coefficient. , These are the port voltage and current increment vectors, respectively.

[0144] Linearized state-space model of the unit , According to the Kaman-Yakubovich-Popov (KYP) lemma, the corresponding linear matrix inequality can be constructed:

[0145]

[0146] In the formula, This is the coefficient matrix of the state-space model. Let Lyapunov be the positive definite matrix to be solved. For stability margin coefficient, Let be the identity matrix. The control parameters that satisfy this inequality constitute the stable feasible region. Joint optimization requires that the control parameters be within this feasible region, thereby achieving parallel constraints on equipment configuration and control parameters and avoiding the disconnect between the planning scheme and the control feasible region.

[0147] VIII. Counterexample-Driven Stable Certificate Decomposition Algorithm

[0148] The aforementioned joint optimization model includes integer configuration variables, nonlinear support propagation relations, bibliometric risk terms, and semidefinite stability certificate constraints. Direct solution involves significant computational costs. Therefore, a stability certificate-driven generalized Benders decomposition method is adopted, and a counterexample-driven refinement approach is introduced. The algorithm as a whole consists of four parts: the configuration master problem, the control certificate subproblem, the worst-case scenario subproblem, and the electromagnetic transient verifier.

[0149] The main problem is configured with a fixed set of generated scenarios and segments. The solution involves assigning unit type, capacity, location, stage roles, and supporting tasks to obtain the [missing information - likely a specific problem or solution]. Candidate configuration schemes for the wheel. The main problem is approximated by a mixed-integer second-order cone form, which represents the power flow, current, and support contract constraints, while retaining the global combination decision of investment and operating costs.

[0150] The control certificate subproblem, under the premise of fixed discrete configuration and network capacity, involves solving for the control parameters and corresponding Lyapunov matrices of each unit. If the subproblem is infeasible, a certificate infeasibility cutoff is generated based on the dual variables of semidefinite programming.

[0151]

[0152] In the formula, , Let be the cutting coefficients derived from the dual variables of the semidefinite programming problem. This is the cutoff threshold. This cutoff excludes configuration areas with sufficient capacity but no stable control parameters, preventing the main problem from repeatedly generating unfeasible solutions.

[0153] The worst-case scenario search addresses the continuous-discrete hybrid adversarial problem by identifying the operational scenario that minimizes the support contract margin:

[0154]

[0155] In the formula, To support the contractual residual margin, The feasible region is defined by uncertain parameters. Continuous optimization variables include external grid impedance, renewable energy output, and control delay, while discrete variables include line disconnection, generator disconnection, and fault location. The search process utilizes differentiable stable sensitivity for gradient search of continuous variables, combined with a branch-and-select method to handle discrete topological variables, which can locate critical weak network states faster than random enumeration.

[0156] Electromagnetic transient verification is performed on the most unfavorable scenario obtained from the search, checking for phenomena such as phase-locked loop (PLL) lockout, grid-connected unit power angle instability, current saturation, DC overvoltage, junction temperature exceeding limits, and mode switching impact. If the verification fails, the corresponding instability counterexamples are extracted:

[0157]

[0158] In the formula, For current-saturated generator sets, When saturation occurs, For the collection of off-grid units, This is the critical instability mode. These are the parameters for the scenario corresponding to instability.

[0159] Depending on the different instability mechanisms, cuts can be generated for insufficient support, current saturation, mesh interaction, or mode switching. Taking the current saturation counterexample as an example, the corresponding cut constraints are:

[0160]

[0161] In the formula, A collection of available units, This is the available current margin of the unit. For the current required by the scene, To increase the safety margin, newly generated scenarios and segmentation constraints are added to the configuration master problem and iterated until no counterexamples violating the supporting contract and transient safety requirements can be found.

[0162] 8.1 Cutting Structures of Four Types of Instability Counterexamples

[0163] A safety margin is defined for any counterexample, and the safety margin is greater than or equal to 0 in steady state. Using the moment when the limit is first exceeded and the dominant unit or mode in the transient trajectory as active features, the gradient is calculated using the adjoint sensitivity or central difference method. A local linear cut is added within the trust region, with a unified cut template as follows:

[0164]

[0165] In the formula, For the safety margin at the counterexample point, , These are the gradients of the decision variable and the control parameter, respectively. , For the decision and parameter points corresponding to the counterexamples, Activate binary variables for counterexample scenarios. For safety margin, M is a large constant. To avoid erroneously excluding feasible regions far from counterexamples, each cut only takes effect within the corresponding parameter's trust region. If the new solution is outside the trust region, the simulation is re-done and a new cut is generated, without extrapolating the old gradient information.

[0166] Counterexamples of phase-locked loop (PLL) lockout are characterized by the moment when the phase angle error first reaches the threshold. Sensitivity is calculated for the synchronous stiffness, damping support, and PLL bandwidth on the grid-connecting side. The cutting constraint is that the combination of synchronous support increment and PLL bandwidth adjustment must meet the lockout margin requirement. Parameter mapping is based on the control law of the synchronous rotating coordinate system PLL. The PLL bandwidth can be calculated from the proportional and integral coefficients, satisfying... , The correspondence, where, This represents the d-axis voltage component.

[0167] Counterexamples of power angle instability in grid-connected generating units are characterized by the maximum power angle after fault clearance or insufficient deceleration area. The power angle stability margin can be improved by increasing the contribution of synchronous stiffness, damping, or inertia support. Parameter mapping is based on fundamental physical relationships, and the equivalent inertia and virtual inertia satisfy... The corresponding relationship is that damping includes two parts: virtual damping and sag equivalent damping. The synchronous stiffness can be approximated as... .

[0168] For counterexamples of inter-machine coupled oscillations in the network, the inter-machine mode with the minimum damping is selected. The parameter sensitivity is calculated using the left and right eigenvectors, while penalizing the product of node mutual impedance and control closed-loop gain. The parameter mapping employs a linearization method to handle sensitivity, and an upper bound is applied to the interaction term. We can take the product of the mutual impedance norm and the norm of the closed-loop transfer function of both units, i.e. The local upper bound of the unit state product term is handled using the McCormick relaxation method, where, Let be the closed-loop transfer function of unit i.

[0169] Mode switching impact counterexamples are characterized by the first peak current or voltage after the switch, mapping the pre-synchronization error and integral state discontinuities to a second-order cone-cutting constraint. The parameter mapping includes the integral state mapping matrix. It is used to achieve a smooth transition of the integral state of the grid-type and network-type controllers. The synchronization hold time can be included as a discrete or piecewise linear variable in the constraint. The jump values ​​of power angle, voltage and power are all from the pre-synchronization window before switching.

[0170] 8.2 Convergence Criteria and Termination Mechanism

[0171] Let the lower bound of the objective value of the configuration master problem be . The optimal objective value of the verified feasible solution is an upper bound. The worst-case violation is The algorithm is considered convergent when the following conditions are met: the relative gap between the upper and lower bounds is less than a set threshold, the worst-case violation is less than a set threshold, and the discrete configuration remains unchanged for multiple consecutive rounds without any new counterexamples with different topologies:

[0172]

[0173] As a preferred implementation, the relative gap threshold can be taken as follows: The violation threshold can be taken as follows: The per-unit value is 3 for consecutive rounds.

[0174] The mechanism to prevent iteration from entering an infinite loop includes five aspects. First, a maximum number of iterations and a maximum number of electromagnetic transient calls are set as hard constraints on computational resources. Second, when the normalized distance between two scenarios is less than a threshold and the cosine similarity of the cutting normal vectors is higher than a threshold, it is judged as a duplicate cut, and only the one with stronger constraints is retained. Third, when the same integer configuration generates the same type of counterexample for the second time, an invalid cut is added to exclude that configuration. Fourth, when the decrease in violation degree is insufficient after multiple consecutive rounds, the trust region radius is halved to improve cutting accuracy. Fifth, long-term inactive cuts are aged out and deleted, while certificate-infeasible cuts are permanently retained. If the convergence condition is not met even after reaching the resource limit, the current optimal feasible solution, upper and lower bound gaps, and a list of unresolved counterexamples are output, while a preset conservative network construction backup scheme is activated, and infinite iteration is not performed. Under the conditions of a finite set of discrete configurations, a compact continuous uncertainty domain, and each new counterexample cut excluding the current violation point, the algorithm can achieve finite convergence with a given tolerance. The deduplication and trust region mechanisms ensure that the numerical implementation will not cycle between equivalent counterexamples.

[0175] Based on the above design, the specific implementation process of this method is as follows: Figure 2 As shown, the specific implementation of each step is as follows:

[0176] Step 1: Data Acquisition and Unified Modeling of Power Plants. Acquire data on the topology, line and transformer parameters, unit capacity, control parameters, grid impedance, and operating scenarios of new energy power plants, and establish a unified multi-port dynamic model for grid-connected units, grid-connected units, and power plant collection networks.

[0177] Step 2: Calculate the dynamic grid connection strength deficit. Calculate the directional stability margin for each dimension according to the station node, control frequency band, dynamic action direction, and operating scenario, and compare it with the preset safety margin to form the dynamic grid connection strength deficit tensor.

[0178] Step 3: Determine the grid support capacity of the units. Based on the rated capacity, control parameters, current margin, junction temperature margin, available DC energy, and active power reserve of each candidate unit, construct a multi-faceted model of their grid support capacity.

[0179] Step 4: Calculate the propagation effect of network support. Calculate the contribution of different types of network support from candidate units to the stability margin of each node in the station, forming a support propagation operator. At the same time, calculate the control coupling loss between network units based on the network mutual impedance and controller dynamic characteristics.

[0180] Step 5: Establish grid connection support contract. Link dynamic grid connection strength deficit, unit support capacity, support propagation effect, and control coupling loss together to establish support contract constraints, ensuring that effective grid connection support is not less than the strength deficit at each node, frequency band, dynamic direction, and operating scenario.

[0181] Step 6: Conduct joint optimization of configuration and control parameters. Using the unit's control type, capacity, location, normal operation, fault process, and fault recovery stages, as well as the control parameters for grid-connected and grid-connected types, as decision variables, establish and solve a joint optimization model that includes support contracts, equipment operation constraints, stability certificate constraints, and uncertainty constraints to obtain candidate configuration schemes.

[0182] Step 7: Perform stability verification and worst-case scenario search. For each candidate configuration, search for the operating scenario that minimizes the grid connection support contract margin. Verify the stability of the candidate schemes using stability certificates, frequency domain models, and electromagnetic transient models.

[0183] Step 8: Counterexample Feedback and Result Output. If the verification reveals that the candidate scheme has problems such as insufficient support, current saturation, control coupling, or mode switching instability, the corresponding instability counterexamples are extracted and cutting constraints are generated. The process is then repeated in Step 6 to solve the optimization model again. This process is iterated until there are no scenarios that violate the stability requirements. Finally, the number, capacity, location, control parameters, and role reconfiguration strategies of the following network types, fixed network types, and dual-mode units are output.

[0184] The beneficial effects of the method provided in this embodiment are as follows:

[0185] First, this method expands the grid connection strength evaluation index from a single scalar to a four-dimensional deficit tensor with node, frequency band, dynamic direction and operation scenario attributes. It can accurately distinguish different types of stability requirements such as low-frequency synchronization support, voltage support, control coupling suppression and fault recovery, realize the precise location of weak network weaknesses, provide a quantitative basis for the differentiated configuration of network resources, and avoid the support mismatch problem caused by traditional single index.

[0186] Second, this method expands the measurement dimension of network support capacity from rated capacity to a polyhedral form constrained by current, junction temperature, energy, primary energy reserve and local stability certificate. By screening key operating points, differential convex inner approximation and interval remainder term certification, it can realize the provable conservative capacity boundary characterization, which can truly reflect the actual available support level of different equipment and avoid planning schemes that cannot be implemented due to ignoring physical constraints.

[0187] Third, this method establishes an analytical sensitivity chain of control parameters, local modes, network propagation, and target node stability margin, forming an interpretable support propagation operator. It also explicitly considers the control coupling interaction loss between multiple network support units, which can accurately quantify the actual stability contribution of different locations and types of network support, realize the spatial optimization of network resources, and improve resource utilization efficiency.

[0188] Fourth, this method achieves coordinated optimization of unit type, capacity, location, multi-stage operation roles and control parameters. It directly embeds local dissipative stability certificates as constraints into the optimization model, so that equipment configuration and control parameter tuning are mutually constrained in the same process. This fundamentally avoids the infeasibility problem caused by the serial design of planning and control, and improves the engineering feasibility of the configuration results.

[0189] Fifth, this method constructs a counterexample-driven closed-loop optimization architecture. For four typical instability scenarios, namely phase-locked loop loss, power angle instability, inter-machine coupling oscillation, and mode switching impact, it constructs mechanism-specific cutting constraints and transforms the electromagnetic transient simulation verification results into an explicit optimization constraint reinjection model. This enables continuous refinement of the configuration scheme, effectively covers extreme weak network conditions, and reduces the risk of operational instability.

[0190] Sixth, this method uses a combination of Bootstrap and validation set calibration to determine the radius of the Wasserstein fuzzy set. Under finite sample conditions, it balances the robustness of the fuzzy set with the economy of configuration, avoiding the problems of excessive conservatism or insufficient risk caused by empirical values. The accompanying finite termination mechanism ensures the numerical stability and engineering practicality of the algorithm.

[0191] The present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the method for joint optimization configuration of new energy power stations and grid-connected units as described above.

[0192] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for joint optimization configuration of new energy power plants and grid-connected generating units, characterized in that, include: Establish a unified multi-port dynamic model for new energy power plants, grid-connected units, grid-structured units, and power collection networks; construct a four-dimensional dynamic grid connection strength deficit tensor of nodes, frequency bands, dynamic directions, and operating scenarios; and form a deficit distribution that characterizes the system's stability support requirements. Combining the physical and operational constraints of the unit, a provably conservative network support capacity polyhedron is constructed through key operating point coverage and differential convex internal approximation, resulting in a support capacity model characterizing the available support boundary of the unit. The calculation of network support propagation operators and control coupling interaction losses between units is used to establish support contract constraints for the corresponding dimensional strength gaps in effective support coverage. Using unit control type, capacity, access location, multi-stage operation role and control parameters as decision variables, the configuration and control collaborative optimization model containing the supporting contract constraints is solved to obtain candidate configuration schemes; Stability verification is performed on candidate configuration schemes. If an unstable counterexample is identified, the corresponding cutting constraint is generated and fed back into the collaborative optimization model for iterative correction until the stability requirements are met, and then the final configuration scheme is output. The steps for constructing a four-dimensional dynamic grid-connected strength deficit tensor include: clustering based on the closed-loop characteristic frequency and the open-loop crossover frequency of the control loop, adaptively dividing multiple dynamic frequency bands; extracting the stability characteristics of different dynamic action directions through the directional projection matrix, calculating the directional stability margin of each dimension based on the minimum singular value of the projected loop gain matrix, and comparing it with the preset safety margin to obtain the corresponding strength deficit. The propagation operator is derived from the left and right eigenvectors of the critical mode to obtain the analytical sensitivity of the eigenvalues ​​to the support components. It is generated by combining the minimum singular value perturbation relationship and introducing a second-order correction term to compensate for the linear error under a wide range of parameter changes. The control coupling interaction loss is obtained by quantizing the maximum singular value of the node mutual impedance and the unit control closed-loop gain. The specific constraint of the support contract is that the total amount of network support transmitted through the network, after deducting the control coupling and interaction loss between units, shall not be less than the sum of the strength deficit and reserved safety margin under the corresponding node, frequency band, dynamic action direction and operating scenario.

2. The method for joint optimization configuration of new energy power plants and grid-connected generating units according to claim 1, characterized in that: The steps for constructing the network support capability polyhedron include: selecting a candidate pool of key operating points containing operating quantiles, physical critical points, modal critical points, and topological boundary points; screening key operating points using the weighted maximum-minimum distance method; performing differential convex decomposition on the nonlinear constraints in the neighborhood of each key operating point; generating a constraint convex upper bound by combining the upper bound of the interval Hessian matrix; and projecting the upper bound to obtain a locally conservative polyhedron. The conservatism of the polyhedron is ensured through a three-layer mechanism of analytical upper bound verification, numerical back-substitution verification, and engineering boundary shrinkage.

3. The method for joint optimization configuration of new energy power plants and grid-connected generating units according to claim 1, characterized in that: The unit control types are divided into three categories: fixed grid-following type, fixed grid-connecting type, and dual-mode type (grid-following and grid-connecting). The configuration variables of the three types satisfy mutual exclusion constraints. For dual-mode units, three operating roles are set for normal operation, fault process, and fault recovery, and phase pre-synchronization constraints, voltage pre-synchronization constraints, power pre-synchronization constraints, and minimum dwell time constraints are configured for mode switching.

4. The method for joint optimization configuration of new energy power plants and grid-connected generating units according to claim 1, characterized in that: The collaborative optimization model uses the Wasserstein sparse bar optimization framework to handle operational uncertainties. It constructs an empirical distribution based on historical samples and establishes a Wasserstein fuzzy set, using the conditional risk value under the worst distribution as a stability risk indicator. The radius of the Wasserstein fuzzy set is determined by a combination of resampling initial selection and validation set calibration, selecting the minimum radius while meeting the target risk level.

5. The method for joint optimization configuration of new energy power plants and grid-connected generating units according to claim 1, characterized in that: The collaborative optimization model embeds local dissipative stability certificate constraints and constructs linear matrix inequalities based on the Kalman-Jakubovich-Popov lemma to ensure that the optimized control parameters are within the stable feasible region.

6. The method for joint optimization configuration of new energy power plants and grid-connected generating units according to claim 1, characterized in that: Instability counterexamples include four categories: phase-locked loop lockout, power angle instability of grid-connected units, coupling oscillation between grid-connected units, and mode switching impact. Each type of instability counterexample corresponds to a cutting constraint that matches the generation mechanism. Each cutting constraint is effective only within the corresponding parameter trust region. Iterative convergence satisfies the triple criteria of upper and lower bound gaps, counterexample violation degree, and discrete configuration invariance, and a mechanism to prevent infinite loop termination is set.

7. A non-transitory computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the method for joint optimization configuration of new energy power stations and grid-connected generating units as described in any one of claims 1 to 6.

Images