Power distribution network flexibility considering power transmission and distribution pre-decision scheduling method and system

Abstract

The present application belongs to the technical field of power system operation optimization and intelligent dispatching, and provides a power transmission and distribution pre-decision dispatching method and system considering the flexibility of distribution network, which obtains original data of flexible operation domain by using an improved Markov chain Monte Carlo algorithm, trains a physical information neural network with physical consistency constraints, solves a day-ahead optimal power flow of the distribution network, defines a distribution network output and cost model based on a cost increment of the flexible operation domain and a deviation from an optimal power flow operating point, performs photovoltaic output prediction based on a bidirectional long short-term memory neural network and a convolutional neural network, solves an economic dispatching pre-decision upper problem of the power transmission network under participation of the distribution network, solves a lower problem of a dynamic flexible operation domain considering time sequence constraints, and iteratively obtains an optimal economic dispatching scheme. The present application can better represent the power support that the distribution network can provide from the perspective of the power transmission network, and is more in line with system operation conditions.

Description

Technical Field

[0001] This invention belongs to the field of power system operation optimization and intelligent dispatching technology, and particularly relates to a pre-decision dispatching method and system for power transmission and distribution that takes into account the flexibility of the distribution network. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the increasing penetration of new energy sources such as wind power and photovoltaics in the power system, and the rapid development of flexible resources such as energy storage, smart soft switching, and controllable loads on the distribution side, the operating characteristics of the power system have undergone profound changes. The traditional hierarchical dispatching model of "centralized optimization on the transmission side and local execution on the distribution side" faces challenges such as increased uncertainty, cross-layer coupling of constraints, and resource distribution sinking.

[0004] To address these challenges, the concept of Flexible Operation Region (FOR) has emerged. FOR is defined as the set of all feasible operating states that a distribution network can achieve using its internal controllable resources within a selected variable space, given network topology, equipment capacity, and safety constraints. It is typically represented by the power exchange range at the Point of Common Coupling (PCC) between the distribution and transmission networks. By efficiently solving the FOR, the distribution network can be aggregated into a "virtual power plant" with specific regulation capabilities from the transmission network's perspective. This allows the internal operating constraints and regulation capabilities of the distribution network to be considered in the transmission network's day-ahead economic dispatch and reserve capacity allocation decisions. This effectively solves the problem of the distribution network being unable to upload full models and data due to privacy or technical reasons, thereby improving the system's security, reliability, and economy.

[0005] Against this backdrop, existing technologies have focused on transmission and distribution coordinated dispatch. These studies primarily concentrate on centralized optimization of the transmission network side, master-slave collaborative iteration, or coordination through ancillary service markets. Research on FOR (Forward-Oriented Operations), especially dynamic FOR, in transmission and distribution coordinated dispatch remains limited. Current FOR modeling methods are mostly designed for single time segments, emphasizing the static regulation capability of the distribution network at a specific moment. These methods neglect key time-series factors, making it difficult to accurately describe the dynamic flexibility of the distribution network over continuous time scales. Furthermore, a balance between solution speed and accuracy is often difficult to achieve, thus failing to meet the needs of power dispatch. Summary of the Invention

[0006] To address at least one of the technical problems mentioned above, this invention provides a scheduling pre-decision method and system that considers the equivalent flexibility cost of the distribution network. This method better characterizes the power support that the distribution network can provide from the perspective of the transmission network; it is more consistent with the system operating conditions; it has a clear architecture; and it offers better versatility, privacy, and speed.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: A first aspect of the present invention provides a transmission and distribution pre-decision scheduling method that takes into account the flexibility of the distribution network, comprising the following steps: The photovoltaic output prediction results are obtained based on the acquired historical power data and weather factor data. Construct a set of action vectors containing the actions of controllable resources in the distribution network, sample power flow samples in the action vector space, calculate the corresponding exchange power at the common connection point between the distribution network and the transmission network based on forward and backward power flow calculation, use the exchange power as the variable space of the operating domain, and perform feasibility judgment on the power flow samples according to the established operating constraints of the distribution network to obtain the original dataset of the flexible operating domain of the distribution network. Based on the constructed photovoltaic output prediction results, the previous distribution network state and the previous distribution network state, a physical information neural network is trained to learn the mapping from system state to flexible operating domain parameters, so that when the system state of a given time period is input, it outputs the flexible operating domain boundary parameters of that time period. Within the distribution network, a day-ahead optimal power flow model considering flexibility resources is solved to obtain the common connection reference operating point and its reference cost, and an equivalent flexibility model of the distribution network is constructed using the incremental cost deviating from the reference operating point. Based on the equivalent flexibility model of the distribution network, a hierarchical iterative solution framework is adopted, and time constraints are introduced to dynamically update the boundary parameters of the flexible operating domain of the distribution network in the next time period, so as to obtain the scheduling pre-decision results.

[0008] Furthermore, an improved Markov chain Monte Carlo algorithm is used to sample the action vector set, specifically including: A multi-strategy progressive approach is used for initial point search; Optimization problems are established in the four directions of the PQ axis, and the gradient is estimated by forward difference along each dimension of the action space. Gradient ascent iteration is performed to solve the power flow samples at the extreme points of PQ. Introduce directional perturbations toward the boundary to advance the PCC power in multiple directions, and increase the sample density near the direction when a feasible point is found to be close to the constraint activation. The ConvexHull function is used for convex hull edge filling, and state-space interpolation is performed on the e-th edge of the convex hull. The classic Metropolis-Hastings algorithm is used to sample power flow samples within the operating domain, thereby enhancing the sample set while verifying the operating domain boundary.

[0009] Furthermore, when training the physical information neural network using the original dataset, the loss function is composed of a weighted average of the data fitting term and the physical consistency term: , , , , , , in, This represents the regression loss used for supervised fitting of the parsed sample output. This represents the residual term after incorporating constraints such as power flow conservation, power balance, and PCC power definition consistency. This indicates that a soft penalty is applied to the inequality constraints, allowing the network output to naturally satisfy the boundary conditions of voltage / current / power / SOC constraints. , These represent the penalty weights for violating physical constraints and inequality constraints, respectively. , This indicates the weights of the boundary points and the area regression loss. This represents the boundary point regression loss. Indicates area regression loss, Indicates the number of time intervals. This represents the true radial radius in the FOR obtained by simulation. Indicates the radius of the FOR iteration. Indicates the radial division number, This represents the true area obtained by the simulation method for FOR. Indicates the area of ​​the FOR iteration. Represents the physical loss term. Indicates tidal residual, Indicates a PCC interface consistency item. Indicates a time-consistent term. , , The weights represent power flow residuals, PCC interface, and timing consistency errors. This indicates the penalty for exceeding the node voltage limit. This indicates the penalty for exceeding the branch current limit.

[0010] Furthermore, the equivalent flexibility cost model for the distribution network is expressed as follows: ,in, Indicates the equivalent incremental cost. This represents the function for calculating the cost of the power distribution network. and This represents the active power exchange and reactive power exchange at the PCC junction point at time t for sample k. This indicates internal operating costs.

[0011] Furthermore, the dynamic updating of the flexible operating domain boundary parameters of the distribution network over a given period, considering time-series constraints, includes: When economic dispatch imposes requirements on the distribution network's flexible output at a certain moment, the operating domain for the next time series shrinks systematically to accommodate the dynamics of energy storage SOC and component ramp-up constraints. If feasible, the distribution network confirms the plan and updates the flexible operating domain for the next moment. If not feasible, revised feasible boundary information is generated based on the reasons for constraint activation, and the information used by the upper layer in the next round is updated. and cost parameters.

[0012] Furthermore, the iterations continue until any of the following preset convergence conditions are met: Condition 1: The PCC plan given by the upper layer is feasible in all time periods of the lower layer, and the change in the objective function is less than the threshold. Condition 2: The FOR area update magnitude is less than the threshold; Condition 3: Reach the maximum number of iterations.

[0013] A second aspect of the present invention provides a transmission and distribution pre-decision dispatching system that takes into account the flexibility of the distribution network, comprising: The prediction module is used to predict photovoltaic output based on the acquired historical power data and weather factor data; The operation domain construction module is used to construct a set of action vectors containing the actions of controllable resources in the distribution network, sample power flow samples in the action vector space, calculate the corresponding exchange power at the common connection point between the distribution network and the transmission network based on forward and backward power flow calculation, use the exchange power as the operation domain variable space, and perform feasibility judgment on the power flow samples according to the established distribution network operation constraints to obtain the original dataset of the flexible operation domain of the distribution network. The flexible operating domain boundary determination module is used to train a physical information neural network based on the constructed photovoltaic output prediction results, the previous distribution network state, and the previous distribution network state. It learns the mapping from the system state to the flexible operating domain parameters, so that when the system state of a given time period is input, it outputs the flexible operating domain boundary parameters for that time period. The equivalent flexibility model construction module is used to solve the day-ahead optimal power flow model considering flexibility resources within the distribution network to obtain the common connection reference operating point and its reference cost, and to construct the distribution network equivalent flexibility model with the incremental cost deviating from the reference operating point. The pre-decision output module is used to dynamically update the boundary parameters of the flexible operating domain of the distribution network over a period of time based on the equivalent flexibility model of the distribution network, using a hierarchical iterative solution framework and introducing time constraints, and to obtain the scheduling pre-decision results.

[0014] A third aspect of the present invention provides a computer-readable storage medium.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the transmission and distribution pre-decision scheduling method that takes into account the flexibility of the distribution network as described above.

[0016] A fourth aspect of the present invention provides a computer device.

[0017] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the transmission and distribution pre-decision scheduling method that takes into account the flexibility of the distribution network as described above.

[0018] Compared with the prior art, the beneficial effects of the present invention are: This invention takes into account the timing constraints of distribution network operation and constructs a dynamic and flexible operating domain for distribution networks. Compared with the time-section operating domain, the dynamic and flexible operating domain can use volume to characterize the flexibility of time series and can better characterize the power support that the distribution network can provide from the perspective of the transmission network. This invention takes into account the uncertainty of new energy output and comprehensively considers the asymmetric, non-monotonic, and nonlinear mapping relationship between prediction error and decision offset in the coordinated economic scheduling of transmission and distribution. It is more in line with the system operating conditions, with a clear architecture and better versatility.

[0019] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0020] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0021] Figure 1 This is a flowchart of a scheduling pre-decision-making method that considers the equivalent flexibility cost of the distribution network, provided in an embodiment of the present invention. Figure 2 This is the photovoltaic prediction neural network provided in the embodiments of the present invention; Figure 3 This is the equivalent flexibility cost model for distribution networks provided in the embodiments of the present invention; Figure 4This is the open-loop and closed-loop pre-decision framework provided in the embodiments of the present invention; Figure 5 This is the hierarchical scheduling framework for coordinated transportation and distribution provided in the embodiments of the present invention; Figure 6 This is the IEEE 33-node system provided in the embodiments of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0025] To address the problems mentioned in the background technology, this invention is based on the concept of flexible operating domain. First, an improved analytical method is used to obtain the original data of the flexible operating domain. Then, a PINN neural network with physical consistency constraints is trained to solve the day-ahead optimal power flow of the distribution network. The power output and cost model of the distribution network is defined with the flexible operating domain. Photovoltaic power output is predicted based on bidirectional long short-term memory neural network (BiLSTM) and convolutional neural network (CNN). The upper-level problem of pre-decision economic dispatch of the transmission network with the participation of the distribution network is solved. The lower-level problem of dynamic flexible operating domain is solved considering time constraints. The optimal transmission and distribution coordinated dispatch scheme is obtained iteratively.

[0026] Example 1 like Figure 1 As shown, this embodiment provides a scheduling pre-decision method that considers the equivalent flexibility cost of the distribution network. It performs upper-level economic scheduling pre-decision based on the distribution network's dynamic flexible operating domain (FOR) and equivalent flexibility cost model, and achieves plan feasibility verification and dynamic updating of the operating domain through upper and lower-level iterations. The method includes the following steps: Step 1: Based on the acquired historical power data and weather factor data, predict the photovoltaic output results; like Figure 2As shown, for photovoltaic output on the distribution network side, a hybrid feature input of weather factors and historical power sequences is constructed. The historical power sequences are input into a CNN to extract local patterns of multi-dimensional weather features (irradiance, cloud cover, temperature, humidity, wind speed, etc.). The weather factors are input into a BiLSTM to capture the bidirectional temporal correlation of historical sequences, and output the photovoltaic forecast values ​​for each time period before the current day. This is used for scenario modeling of FOR construction and upper-level scheduling.

[0027] Unlike traditional photovoltaic short-term forecasting frameworks, the forecasting framework described in this invention no longer uses the minimum of statistical quantities such as the root mean square error between the predicted output and the actual output as the objective for gradient descent. Instead, it uses the gradient chain rule to minimize the economic dispatch cost of the predicted output, thereby enabling the forecasting model to learn and update the weight parameters of each layer in a decision-oriented manner.

[0028] Step 2: Construct action vectors containing the actions of controllable resources in the distribution network, sample power flow samples in the action vector space, and calculate the corresponding exchange power at the common connection point between the distribution network and the transmission network based on forward and backward power flow calculations. This invention samples the flexible resource action space by improving the Markov Chain Monte Carlo algorithm (MCMC), solves the power flow based on the forward-backward substitution method, then constructs a distribution network operation constraint discriminator to determine the boundary of the flexible operation domain, and finally uses a physical information neural network to fit the mapping from the distribution network state to the parameters of the flexible operation domain, thereby achieving efficient solution of the flexible operation domain.

[0029] Step 201: Define the flexible operating domain; The flexible operating domain is defined as the set of all feasible operating states that a distribution network can achieve with controllable resources in a selected variable space, given network topology, equipment capacity and operating constraints. It is generally represented by the power exchange range at the point of connection (PCC).

[0030] A flexible operating domain can be represented as follows: , in, Represents the time function of the flexible operating domain. This indicates the exchangeable active power range at PCC. This indicates the range of reactive power that can be exchanged at PCC. Indicates the spatial dimension of the device's motion. Represents the power flow calculation function. This represents the voltage at node i. Indicates the lower limit of the node voltage. Indicates the upper limit of the node voltage. This represents the current in branch ij. Indicates the upper limit of the branch current. Represents the device action vector. Indicates the device's operating space; Step 202: Construct an action vector containing the actions of controllable resources in the distribution network. ; Controllable and flexible resources in a distribution network include photovoltaic (PV) reactive power, energy storage devices (ESS), on-load tap changers (OLTC), capacitor switching (SC), smart soft switches (SOP), and adjustable loads (AL). This invention improves the MCMC algorithm to achieve sampling of continuous and discrete action spaces. It constructs action vectors for controllable resources in the distribution network. It includes the reactive power output of the PV inverter, the charging and discharging power of the ESS, the adjustment of the OLTC tap, the switching of the SC, the direction and magnitude of the SOP power flow, the AL delay time and load, and the load reduction.

[0031] This invention constructs multiple types of spatiotemporal constraints to restrict the action space set of controllable flexibility resources in the distribution network. Specifically, each controllable flexibility resource in t The action space at any given moment is formed considering the following spatiotemporal constraints: PV active power constraints: , PV reactive power constraints: , PV ramping constraint: , ESS power constraints: , ESS capacity constraints: , ESS climbing constraints: , SC Action Constraints: , SVG power constraints: , SVG ramping constraints: , SOP capacity constraints: , SOP ramping constraints: , OLTC Motion Constraints: , AL transferable load constraints: , Load constraints can be reduced: , In the formula, , Let represent the active power output of the photovoltaic system at node i at time t and time t-1. This indicates the rated active power of photovoltaic power. This represents the efficiency of photovoltaic power at time t. , This represents the reactive power output of the photovoltaic system at node i at time t and time t-1. Indicates the rated capacity of photovoltaic power. This represents the rated active power ramp rate of the photovoltaic system at node i. This represents the rated reactive power ramp rate of the photovoltaic system at node i. Indicates the rated power of energy storage. , This represents the exchange power of the energy stored at node i at time t and time t-1. This represents the state of charge of the energy stored at node i at time t. This represents the rated ramp rate of energy storage at node i. This indicates the switchable capacitor bank's position at time t. This indicates the switching action of the switchable capacitor bank at time t. Indicates the total number of switches for the capacitor bank. This indicates the maximum number of times a switchable capacitor bank can be switched on and off per day. This indicates the maximum reactive power compensation amount of the SVG. , This represents the reactive power compensation output of the SVG at node i at time t and time t-1. This represents the rated reactive ramp rate of the SVG on node i. This indicates that the SOP at node i has active power flow at time t. This indicates the reactive power flow at time t of the SOP on node i. This represents the apparent power of SOP at node i at time t-1. Indicates the SOP rated capacity. This represents the rated apparent power ramp rate of SOP at node i. This indicates the tap position of the on-load tap-changing transformer at time t. This indicates the tap changer operation of the on-load tap-changing transformer at time t. This indicates the maximum tap value of an on-load tap-changing transformer. This indicates the maximum number of daily operations for an on-load tap-changing transformer. Indicates load transfer delay. Indicates the maximum load transfer delay. Indicates the transfer of active power load. This represents the maximum transferable active power load. This indicates the transfer of reactive load. This represents the maximum transferable reactive load. This indicates a reduction in active power load. This indicates the maximum active power load that can be reduced. This indicates a reduction in reactive power load. This indicates the maximum amount of reactive load that can be reduced. Step 203: Sample and generate power flow samples in the action vector space; In this embodiment, an improved Markov chain Monte Carlo algorithm is used to sample the action set of flexibility adjustment resources. The improved boundary-enhanced Markov chain Monte Carlo sampling generates action sequences, and the hierarchical sampling configuration is as follows: , in, The number of power flow samples in the flexible operating domain at each moment. This represents the number of exchangeable active and reactive power extreme point samples at PCC. The number of boundary point samples. The number of samples fitted to the convex hull. The number of samples traversed within the operating domain. This represents the total number of trend samples.

[0032] The specific sampling steps include: Step 2031: Find the initial point of the trend; In this embodiment, a multi-strategy cascading approach is proposed to set four initial point search strategies. When the initial point of one strategy is not feasible, the approach is switched to the next strategy to ensure the feasibility of the initial point.

[0033] The four strategies are ordered by priority as follows: ① Time continuity warm-up, using the center point of the previous moment as inspiration; ② Conservative state detection ensures that all flexible devices operate at a conservative level; ③ Zero-state detection, shut down all flexible and controllable devices; ④ Random brute-force search: randomly search for feasible points within the action space.

[0034] Step 2032: Active search for extreme points: Establish four optimization problems and solve for the boundary values ​​of active and reactive power respectively, that is, find a rectangular domain that is tangent to the flexible operating domain; Specifically, optimization problems are established in the four directions of the PQ axis, and the gradient is estimated by forward difference along each dimension of the action space. Gradient ascent iteration is performed to solve the power flow samples at the extreme points of PQ.

[0035] Taking maximum active search as an example, the optimization problem is defined as follows: For this objective function The numerical gradient estimate is The gradient uses an adaptive step size With the addition of a restart mechanism for ascending iterations, the gradient iteration formula is: The step size update rule is as follows: , in, This represents the action vector that achieves maximum work output. This represents the maximum active power boundary point of the flexible operating domain. Indicates a set of possible actions. Indicates action d, This represents the incremental change in the action vector. This represents the infinitesimal increment of action d. Represents the action vector. , This represents an approximate solution for the action vector at the k-th and (k+1)-th iterations. This represents a projection operator that guarantees the action vector lies within the feasible region. , This represents the step size in the k-th and (k+1)-th iterations; Step 2033: Traverse the boundary points starting from the extreme points; Traditional MCMC performs boundary interpolation and extrapolation based on the initial boundary points, and iteratively updates the boundary state pool. This invention proposes to introduce directional perturbations toward the boundary during the sampling process, advancing the PCC power in multiple directions (radial angles). When a feasible point is found to be close to the constraint activation (voltage / current / power / SOC, etc.), the sample density in the vicinity of that direction is increased to accelerate the convergence of the boundary points.

[0036] , in, This indicates the addition of new boundary test samples. Indicates the reference point. This indicates a random perturbation. This indicates that the random perturbation follows a multivariate standard Gaussian distribution; Step 2034: Convex hull edge filling: The algorithm is discrete, and to generate continuous and flexible operating domain boundaries, convex hull filling is needed to ensure that the edges are differentiable; The ConvexHull function is used for convex hull edge filling. State space interpolation is performed on the e-th edge of the convex hull. This invention adds a normal perturbation to refine the convex hull edge of FOR.

[0037] , in, Indicates the edge reference point. Indicates normal perturbation; Step 2035: MCMC Internal Walk: Based on the classic Metropolis-Hastings algorithm, power flow samples within the operating domain are sampled to enhance the sample set while verifying the operating domain boundary.

[0038] Step 204: Obtain the corresponding exchange power at the common connection point between the distribution network and the transmission network based on forward and backward power flow calculations. And form sample pairs; For typical radial or weak loop distribution network structures, this invention uses the forward-backward substitution method to calculate the power flow at sample points, obtaining node voltages, branch currents / power flow, network losses, and power exchange values ​​at the PCC, and forming sample pairs: ,in express t The system status at any given moment (load, photovoltaic, energy storage SOC, previous action status, etc.). express t System actions at any given moment This represents the active power exchange at the PCC connection point at time t. This represents the reactive power exchange at the PCC connection point at time t. It should be noted that the forward-backward substitution method is similar to the normal power flow calculation. First, the initial values ​​of the voltages of each node are set, then the power flow is calculated, and then the voltage is updated by inversely calculating the voltage through the power flow. The process is iterated with the new voltage values ​​until the convergence threshold is reached. The forward push process is as follows: , , , , The process of back-substitution is as follows:

[0039] in, This represents the active power loss of branch ij. This represents the resistance of branch ij. This represents the active power flowing out of node j through branch jk. This represents the active load at node j. This represents the reactive power flowing out of node j through branch jk. This represents the reactive load at node j. This represents the voltage at node j. This represents the reactive power loss of branch ij. Indicates the reactance of branch ij. This represents the active power flowing out of node i through branch ij. This represents the reactive power flowing out of node i through branch ij. Let j represent the set of nodes directly connected to node j. This represents the current correction value on branch ij. This represents the voltage correction value for node i. This represents the voltage correction value at node j; Step 205: Switch the power at the corresponding point of common coupling (PCC). As the operational domain variable space, based on the operational constraints of the distribution network, a set of feasible exchangeable powers that can be realized by the controllable and flexible resources of the distribution network is obtained, forming the original dataset of the flexible operational domain of the distribution network.

[0040] In this embodiment, the established distribution network operation constraint discriminator performs feasibility judgment on the node voltage, branch current / power flow, equipment power, energy storage state of charge (SOC) and ramp and action number constraints of the power flow sample to obtain the original dataset of the flexible operation domain of the distribution network. Compared to simulation methods, analytical methods rely on rigorous mathematical derivations and offer higher solution accuracy. However, the computational complexity of analytical methods increases exponentially with system size. This invention uses the point of connection (PCC) between the distribution network and the transmission network as the interface, selecting the switching power at the PCC (…). As the operational domain variable space, under given topology, equipment capacity and operational constraints, it obtains the set of feasible exchange power that can be realized by the controllable and flexible resources of the distribution network, forming the FOR original sample dataset.

[0041] Step 3: Train PINN using the original dataset of the flexible operation domain of the distribution network to learn the mapping from system state to flexible operation domain parameters, so that it outputs the FOR boundary parameters for a given time period when the state input is given. Machine learning algorithms have been widely applied in power distribution networks, but in computational simulations of complex graphical information such as flexible operating domains, appropriate problem decomposition and robust model support are required. To achieve rapid day-ahead multi-period invocation, this invention uses the obtained FOR raw sample dataset to train a Physical Information Neural Network (PINN) with physical consistency constraints, enabling it to quickly output the key parameters and boundaries of FOR given a state input.

[0042] Specifically, the steps include the following: For each time period t, PINN inputs the system state. ,in, It includes the photovoltaic output, energy storage output, SVG output, and SOP output from the previous moment to adapt to timing constraints such as ramp-up constraints. This represents the active power load vector at time t. This represents the reactive load vector at time t. This represents the photovoltaic output vector at time t. This represents the node's SOC vector at time t; To facilitate incorporating physical consistency into differentiable constraints, a two-output network approach is adopted: FOR output network: Using polar coordinates, it outputs the parameterized radius of the FOR boundary for each time interval t. and angle , And thus, the PCC boundary points are obtained: , in, This represents the k-th radial PCC active power boundary point at time t. This indicates that the center point of PCC has active power at time t. This indicates that the center point of PCC is reactive at time t; Physical variable output network: For each time period t, output the internal operating variables of the distribution network. Used to constrain power flow and limit residuals, denoted as This includes parameters such as node voltage square, branch active / reactive power flow, branch current square, energy storage charging and discharging power, reactive power compensation / inverter reactive power, etc.

[0043] The loss function during PINN training is a weighted sum of the data fitting term and the physical consistency term: , in, This represents the regression loss used for supervised fitting of the parsed sample output. This represents the residual term that incorporates constraints such as power flow conservation, power balance, and consistency of PCC power definition (using a differentiable approximation based on sample power flow mapping). This indicates that a soft penalty is applied to the inequality constraints, allowing the network output to naturally satisfy the boundary conditions of voltage / current / power / SOC constraints. , These represent the penalty weights for violating physical constraints and inequality constraints, respectively.

[0044] Among them, regression loss It is expressed as a weighted sum of boundary point regression loss and area regression loss. ,in, , The weights of the boundary point regression loss and the area regression loss are indicated. Assuming the analytical sample provides FOR true boundary points ( and true radius and FOR area .

[0045] Boundary point regression loss (radius-supervised) It can be represented as: , Area regression loss It can be represented as: ,in, Indicates the total number of moments. This represents the iterative value of the PCC area. This represents the true value of the PCC area; loss function The physical loss term in the equation is represented as a weighted sum of the power flow equation residuals, the PCC interface consistency term, and the timing consistency term. All of them use the Distflow power flow algorithm, which is consistent with the analytical method. , , Represents the weights of the power flow equation residual terms, PCC interface consistency terms, and timing consistency terms; The residual current can be expressed as: , Branch power recursive residual: , , Voltage drop residual: , Current-power relationship residuals: , in, This represents the active power flowing from node i to node j at time t in the k-th sample. Let represent the active power flowing from node j to its directly connected nodes at time t in the k-th sample. This represents the net active power injection / consumption of node j at time t in the k-th sample. This represents the resistance of branch l. This represents the current on line l at time t in the k-th sample. This represents the reactive power flowing from node i to node j at time t in the k-th sample. This represents the reactive power flowing from node j to its directly connected nodes at time t in the k-th sample. This represents the net reactive power injection / consumption of node j at time t in the k-th sample. Indicates the reactance of branch l. This represents the predicted squared voltage value of node j at time t and sample k. This represents the predicted squared voltage value of node i at time t and sample k. This indicates the prevention of very small positive numbers with a denominator of 0; The PCC interface consistency term (ensuring the correspondence between FOR boundary points and internal power flow) can be represented as: , in, This represents the PCC exchange active power prediction value for sample k at time t. This represents the true value of PCC exchange active power for sample k at time t. This represents the PCC (Positive Voltage Calibration) reactive power prediction value for sample k at time t. This represents the true value of PCC switching reactive power for sample k at time t; The timing consistency term can be expressed as the sum of the energy storage state of charge and the ramp constraint loss: , Energy storage state of charge loss: , Climbing constraint loss: , in, Represents the residuals of the SOC dynamic equation. This represents the charging and discharging power of the k-th sample at time t. This represents the charging and discharging power of the k-th sample at time t-1. Indicates the maximum rate of ascent. Indicates the maximum descent rate of the slope; The inequality constraint soft penalty in the loss function is represented by a weighted sum of the node voltage and branch current over-limit penalties: ; The penalty for exceeding the node voltage limit can be expressed as: , Branch current over-limit penalty can be expressed as: , in, Indicates the maximum node voltage. This represents the minimum node voltage. Represents the set of nodes in a distribution network; Step 4: Solve for the day-ahead optimal power flow considering flexibility resources within the distribution network to obtain the common connection reference operating point. and its reference cost And construct an equivalent flexibility cost model for the distribution network using the incremental cost deviating from the reference operating point; To enable the distribution network as a whole to participate in the day-ahead economic dispatch of the transmission network as a dispatchable resource, this invention proposes an equivalent generation cost model of the distribution network from the perspective of the transmission network, based on the optimal power flow within the distribution network. This allows the transmission network to characterize the marginal cost of utilizing the flexibility of the distribution network without needing to know a large number of equipment and network details within the distribution network.

[0046] Specifically, the steps include the following: This invention assumes that the objective of the distribution network during day-ahead dispatch is always to minimize its own operating costs. Considering the adjustability margin of the distribution network's flexibility resources, an optimal power flow model for the distribution network is established: Objective function: , in, For distribution network losses, denoted as , Time-of-use pricing; The component cost includes source side (cured light cost), grid side (SOP loss), and energy storage side (energy storage charging and discharging degradation cost). This is a voltage offset penalty for the distribution network, used to guide the model to approach the rated voltage; Penalize costs for insufficient flexibility in the distribution network; The weighting coefficients representing network loss, voltage, component cost, and flexibility requirements can be adjusted. This characterizes the flexibility requirements of the transmission network for the active distribution network.

[0047] The model adopts the Distflow power flow model, which can be transformed into a mixed integer second-order cone programming solution. The model constraints are consistent with the constraints in step 203. The equivalent flexibility cost model at the PCC point of the distribution network can be represented as a convex polyhedron, and its projection onto the active and reactive power plane is the flexible operating domain.

[0048] In this embodiment, as Figure 3 As shown, an equivalent flexibility cost model for the distribution network is established using the incremental cost deviating from the optimal power flow point as the flexibility call cost. This includes first obtaining the common connection reference operating point within the distribution network based on the optimal power flow. Its corresponding internal operating cost is .

[0049] When the upper-layer scheduling requires the PCC power to deviate from this reference point, the equivalent incremental cost is defined as follows: ,in, This represents the distribution network cost calculation function. By calling the optimal power flow, it can calculate the minimum cost and set of flexibility resource actions under the output condition after the upper-level transmission network determines the flexibility output of the distribution network.

[0050] Step 5: Based on the equivalent flexibility model of the distribution network, a hierarchical iterative solution framework is adopted to output the scheduling pre-decision results; Photovoltaic forecasting errors can lead to deviations in economic dispatch decisions. To address this, this invention proposes a pre-decision strategy for transmission and distribution coordinated economic dispatch that considers short-term photovoltaic forecasting errors. The traditional open-loop forecasting and decision-making framework can be represented as: "Forecasted photovoltaic load → Flexible operating domain cost model → Transmission and distribution coordinated economic dispatch," while the closed-loop pre-decision strategy uses the economic dispatch objective function to feedback and correct the photovoltaic forecasting model, thereby solving the asymmetric, nonlinear, and non-monotonic problems between forecasting errors and decision deviations.

[0051] In this embodiment, based on the equivalent flexibility model of the distribution network, and with economic dispatch cost as the objective, gradient training is performed on the parameters of the dispatch model, PINN network, and prediction network to obtain the optimal network parameter solution for dispatch. A hierarchical iterative solution framework is adopted, and time constraints are introduced to dynamically update the boundary parameters of the flexible operating domain of the distribution network for the next period, thus obtaining the dispatch pre-decision result. To simultaneously consider "economic dispatch solvability, prediction error, and time constraints," this embodiment adopts a hierarchical iterative solution framework: the upper layer uses the transmission network as the main model, aggregating the adjustable resources of each distribution network into a flexible operating domain to solve the day-ahead economic dispatch problem; the lower layer maps the economic dispatch results to the actions of the flexible resources within the distribution network, and dynamically updates the flexible operating domain of the distribution network for the next period, considering time constraints.

[0052] like Figure 4 and Figure 5 As shown, it specifically includes: The upper layer establishes a mapping from photovoltaic prediction values ​​to flexible operating domain parameters through a trained PINN neural network, and then constructs a day-ahead economic dispatch (ED) model for the transmission network to minimize the total day-ahead operating cost of the entire system. The model is then fed back to the prediction model via gradient through FOR as a link to correct the parameters of the prediction model. 1) The prediction model based on BiLSTM and CNN can be expressed as: , The FOR solver based on PINN can be expressed as: , in, This represents a vector of historical output and weather factors. This represents the predicted photovoltaic power output. Indicates the radial parameters of the flexible operating domain. 2) The decision variables of the economic scheduling model can be expressed as: ,in, These represent the transmission network state variables, conventional unit output, and PCC injection in each distribution network, respectively.

[0053] The objective function of economic dispatch is to minimize the weighted sum of the equivalent costs of conventional generating units and the distribution network. , in, The secondary power generation cost function representing the convexity of conventional generating units. This represents the equivalent cost of the flexible operating domain deviating from the optimal power flow. This represents the active power output of unit g at time t. This represents the coefficient (constant) of the quadratic term in the cost function. This represents the coefficient (constant) of the linear term of the cost function. This represents the constant term of the cost function. The constraints that need to be considered in economic dispatch include power flow / power balance of the transmission network, unit operating boundaries and ramping, line power flow and safety constraints, etc. The internal constraints of the distribution network no longer appear explicitly, but are equivalently borne by the flexible operating domain and its cost function.

[0054] 3) According to the chain rule, the gradient of the pre-decision framework can be expressed as:

[0055] By constructing the Lagrangian function, the optimal original variables given by the economic scheduling model are analyzed. and dual variables The gradient can be expressed computationally using KKT conditions: Based on this gradient, the parameters of the prediction network are updated to achieve model update and correction.

[0056] The lower layer performs feasibility verification and dynamic updates of the PCC plan given by the upper layer within the distribution network: when economic dispatch makes demands on the flexibility output of the distribution network at a certain moment, the operating domain of the next time series shrinks according to a certain pattern to adapt to the dynamics of energy storage SOC, component ramp-up constraints, etc. If If feasible, the distribution network confirms the plan and updates the flexible operating domain for the next time step; if not feasible, it generates revised feasible boundary information based on the constraint activation reasons and updates the upper layer's usage for the next round. and cost parameters.

[0057] Repeat the upper-level solution and lower-level verification update process until a preset convergence condition is met. The preset convergence condition is that the upper and lower level iterations continue until any of the following conditions are met: 1) The PCC plan given by the upper layer is feasible in all time periods of the lower layer, and the change in the objective function is less than the threshold. 2) The FOR area update magnitude is less than the threshold; 3) Reach the maximum number of iterations.

[0058] The following examples further illustrate the invention. This embodiment takes a typical active distribution network as the research object, and constructs and verifies the effectiveness of the method according to the process of "analytical generation of FOR samples—PINN training for rapid FOR solution—equivalent cost modeling—photovoltaic prediction—transmission and distribution hierarchical iterative pre-decision making," as shown below. Figure 6 As shown. Specific measures are as follows: 1. Calculation System and Parameter Settings The IEEE 33-node distribution network was selected as the active distribution network (ADN) simulation system, and the IEEE 9-node transmission network was selected as the economic dispatch simulation system. Baseline capacity was used. Reference voltage The system is connected to the upstream transmission network at node 1 (substation bus) via a point of common coupling (PCC). The day-ahead dispatch cycle is T=24h, with a time granularity of [missing information]. .

[0059] The distribution network is configured according to the typical active distribution network. Photovoltaic power is connected to nodes 18 / 25 / 30, with installed capacities of 1.5 / 1.0 / 0.8MW respectively, and the inverter capacity is 1.1 times the installed capacity. Energy storage is connected to nodes 13 / 24, with a rated power of 1.0MW and a rated capacity of 2.0MWh. The on-load tap-changing transformer of the distribution network is set to -8 to 8 levels, with a maximum of 10 level changes per day. Capacitors are connected to nodes 14 / 30, with four groups connected to each node, each group having a capacity of 0.15Mvar, and a maximum of 8 switching operations per day.

[0060] In the improved MCMC solution for the flexible operating domain, 2000 samples are generated for each time segment. The number of extreme points, boundary points, convex hull filling points, and internal walk samples are allocated according to the weights described in this invention. FOR is quickly solved by fitting with PINN, and the boundary is parameterized using polar coordinates. (For every 1 degree), FOR is approximately represented by the FOR area, boundary point radius, and angle. After training, the FOR boundary parameters for each time period can be output in milliseconds and quickly called by the upper layer.

[0061] 2. Strategy Setting and Result Analysis To verify the effectiveness of the "closed-loop pre-decision" and "dynamic FOR" methods described in this invention, three strategies were set up to compare indicators such as total operating cost, curtailment, and load shedding: 1) Strategy A: Open-loop forecasting and decision making The process is as follows: PV prediction → PINN generates FOR (built only according to the predicted state, without lower-level feedback contraction) → Transmission network ED → AC-PF verification → Distribution side execution / rollback. When prediction errors cause plan deviations, they tend to approach the FOR boundary, triggering lower-level rollback and high-cost corrections.

[0062] 2) Strategy B: Closed-loop pre-decision Guided by the goal of minimizing the system's objective cost (ED), gradient / KKT feedback is used to correct the prediction network parameters, making the predictions more biased towards the region where "decision-making is feasible and low-cost." Simultaneously, upper and lower layer iterations are introduced, utilizing lower-layer feedback to dynamically shrink the FOR (Forward Error). This effectively reduces the asymmetric losses caused by "prediction error and decision bias."

[0063] 3) Strategy C: Do not consider dynamic FOR Using only the static FOR (time-section FOR) for each time period, ignoring the contraction effect of SOC dynamics, ramp-up, and discrete action frequencies on the next time period; the upper-level ED directly calls the static FOR and clears. On the surface, the ED is feasible, but when executing the timing sequence, SOC overdraft or action conflicts are more likely to occur, resulting in additional costs (curtailment / load shedding / emergency power purchase) or the need for frequent readjustment. Table 1 shows the comparison results of the three strategies (same load day, same equipment parameters, same PV error level).

[0064] Table 1 Comparison of the daytime execution effects under the three strategies

[0065] After comparison, the closed-loop pre-decision method (Strategy B) described in this invention has the lowest cost. The closed loop feeds the prediction error back to the prediction model through scheduling objectives, making the prediction results more inclined to reduce aggressive commitments to PCC power during evening ramp-up periods and SOC tight-constraint periods. Simultaneously, dynamic FOR contraction information is introduced in the upper and lower layer iterations, resulting in fewer critical ED solutions. Therefore, under the same equipment conditions, Strategy B exhibits the following characteristics: the fewest infeasible backoffs (1 time / day), significantly reduced curtailment and load shedding, and zero AC-PF check exceeding limits. In contrast, the open-loop method (Strategy A) is only guided by minimizing statistical prediction error and does not explicitly perceive the asymmetric penalty of scheduling costs on prediction errors. When the actual PV is lower than the prediction and during periods of high electricity prices / high load, it forces the system to increase the output of high-cost units, trigger deeper energy storage discharge on the distribution side, or undergo uneconomical corrections such as minor load shedding / curtailment. Therefore, the total cost and penalty are higher than Strategy B. Without considering the apparent adjustability of dynamic FOR (strategy C), it has the highest execution cost. Static FOR ignores the pattern of energy storage SOC and ramp-up affecting the contraction of the operating domain in the next time period, which can easily lead to cross-time period infeasibility. For example, excessive discharge during the day to support power transmission can result in insufficient SOC when reactive / active power support is needed in the evening. Limited switching frequency of OLTC / capacitors can lead to insufficient voltage support. Therefore, strategy C often triggers rollback and emergency costs frequently during the execution phase, which is reflected in the highest number of rollbacks (7 times / day), the highest curtailment / load shedding, and the highest total cost.

[0066] In summary, this invention, based on a flexible operating domain representation method for distribution networks, uses an improved Markov chain Monte Carlo algorithm to obtain the original data of the flexible operating domain, trains a physical information neural network with physical consistency constraints, solves the day-ahead optimal power flow of the distribution network, defines the power output and cost model of the distribution network using the cost increment of the flexible operating domain and the deviation from the optimal power flow operating point, predicts photovoltaic power output based on a bidirectional long short-term memory neural network and a convolutional neural network, solves the upper-level problem of pre-decision economic dispatch of the transmission network with the participation of the distribution network, and solves the lower-level problem of the dynamic flexible operating domain considering time constraints, iteratively obtaining the optimal economic dispatch scheme. Finally, using an economic dispatch pre-decision model of the equivalent flexibility cost of the distribution network constructed using the IEEE 33-bus system, the impact of distribution network flexibility resources on the economic dispatch of the transmission network is quantitatively simulated, verifying the effectiveness of the proposed method. The numerical examples show that the constructed model fully considers the speed of solution, model accuracy, and method comprehensiveness, and is more consistent with the characteristics of power dispatch automation systems.

[0067] Example 2 This embodiment provides a power transmission and distribution pre-decision dispatching system that considers the flexibility of the distribution network, including: The prediction module is used to predict photovoltaic output based on the acquired historical power data and weather factor data; The operation domain construction module is used to construct a set of action vectors containing the actions of controllable resources in the distribution network, sample power flow samples in the action vector space, calculate the corresponding exchange power at the common connection point between the distribution network and the transmission network based on forward and backward power flow calculation, use the exchange power as the operation domain variable space, and perform feasibility judgment on the power flow samples according to the established distribution network operation constraints to obtain the original dataset of the flexible operation domain of the distribution network. The flexible operating domain boundary determination module is used to train a physical information neural network based on the constructed photovoltaic output prediction results, the previous distribution network state, and the previous distribution network state. It learns the mapping from the system state to the flexible operating domain parameters, so that when the system state of a given time period is input, it outputs the flexible operating domain boundary parameters for that time period. The equivalent flexibility model construction module is used to solve the day-ahead optimal power flow model considering flexibility resources within the distribution network to obtain the common connection reference operating point and its reference cost, and to construct the distribution network equivalent flexibility model with the incremental cost deviating from the reference operating point. The pre-decision output module is used to dynamically update the boundary parameters of the flexible operating domain of the distribution network over a period of time based on the equivalent flexibility model of the distribution network, using a hierarchical iterative solution framework and introducing time constraints, and to obtain the scheduling pre-decision results.

[0068] It should be noted that the specific implementation of the transmission and distribution pre-decision scheduling system considering the flexibility of the distribution network in this embodiment of the invention is similar to the specific implementation of the transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network in this embodiment of the invention. For details, please refer to the description in the method section. In order to reduce redundancy, it will not be repeated here.

[0069] Example 3 This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the power transmission and distribution pre-decision scheduling method that considers the flexibility of the distribution network as described above.

[0070] Example 4 This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the power transmission and distribution pre-decision scheduling method that considers the flexibility of the distribution network as described above.

[0071] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0072] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0073] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0074] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0076] The above description is merely a preferred embodiment of the present invention and is not intended to limit the 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 transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network, characterized in that, Includes the following steps: The photovoltaic output prediction results are obtained based on the acquired historical power data and weather factor data. Construct a set of action vectors containing the actions of controllable resources in the distribution network, sample the action vector set to generate power flow samples, calculate the corresponding exchange power at the common connection point between the distribution network and the transmission network based on forward and backward power flow calculation, use the exchange power as the variable space of the operating domain, and perform feasibility judgment on the power flow samples according to the established operating constraints of the distribution network to obtain the original dataset of the flexible operating domain of the distribution network. The physical information neural network is trained based on the photovoltaic output prediction results and the distribution network state at the previous moment. It learns the mapping from the system state to the flexible operating domain parameters, so that when the system state for a given time period is input, it outputs the flexible operating domain boundary parameters for that time period. Within the distribution network, a day-ahead optimal power flow model considering flexibility resources is solved to obtain the common connection reference operating point and its reference cost, and an equivalent flexibility model of the distribution network is constructed using the incremental cost deviating from the reference operating point. Based on the equivalent flexibility model of the distribution network, a hierarchical iterative solution framework is adopted, and time constraints are introduced to dynamically update the boundary parameters of the flexible operating domain of the distribution network in the next time period, so as to obtain the scheduling pre-decision results. Specifically, the improved Markov chain Monte Carlo algorithm is used to sample the action vector set, including: A multi-strategy progressive approach is adopted to find the initial point of power flow; optimization problems are established in the four directions of the PQ axis, and the gradient is estimated by forward difference along each dimension of the action space. Gradient ascent iteration is performed to solve the power flow samples of PQ extreme points. Introduce directional perturbations toward the boundary to advance the PCC power in multiple directions, and increase the sample density near the direction when a feasible point is found to be close to the constraint activation. The ConvexHull function is used for convex hull edge filling, and state-space interpolation is performed on the e-th edge of the convex hull. The classic Metropolis-Hastings algorithm is used to sample power flow samples within the operating domain, thereby enhancing the sample set while verifying the operating domain boundary.

2. The transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network as described in claim 1, characterized in that, During the training of a physical information neural network, the loss function It is composed of a weighted average of the data fitting term and the physical consistency term: ， ， ， ， ， ， in, This represents the regression loss used for supervised fitting of the parsed sample output. This represents the residual term after incorporating constraints such as power flow conservation, power balance, and PCC power definition consistency. This indicates that a soft penalty is applied to the inequality constraints, allowing the network output to naturally satisfy the boundary conditions of voltage / current / power / SOC limits. These represent the penalty weights for violating physical constraints and inequality constraints, respectively. , This indicates the weights of the boundary point regression loss and the area regression loss. This represents the boundary point regression loss. Indicates area regression loss, Indicates the number of time steps. Indicates the true radius of FOR. Indicates the prediction radius of FOR. This represents the number of samples at each time step. Indicates the actual area. Indicates the predicted area for FOR. Indicates tidal residual, Indicates a PCC interface consistency item. Indicates a time-consistent term. , This represents the weights of power flow residuals, PCC interface consistency items, and timing consistency items. This indicates the penalty for exceeding the node voltage limit. This indicates the penalty for exceeding the branch current limit.

3. The transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network as described in claim 1, characterized in that, The equivalent flexibility cost model for distribution networks is expressed as follows: ,in, Indicates the equivalent incremental cost. This represents the function for calculating the cost of the power distribution network. and This represents the active power exchange and reactive power exchange at the PCC junction point at time t for sample k. This indicates internal operating costs.

4. The transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network as described in claim 1, characterized in that, The aforementioned distribution network equivalent flexibility model employs a hierarchical iterative solution framework. The solution process includes: the upper layer uses the day-ahead economic dispatch of the transmission network as the objective function, aggregates the adjustable resources of each distribution network into a flexible operating domain, and solves for the planned exchange power of the distribution network PCC in each time period; the lower layer maps the solution results to the actions of the flexible resources within the distribution network, considers time-series constraints, and dynamically updates the boundary parameters of the flexible operating domain of the distribution network for the next time period until the preset convergence condition is met, and outputs the scheduling pre-decision result.

5. The transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network as described in claim 1, characterized in that, The introduction of time-series constraints to dynamically update the boundary parameters of the flexible operating domain of the distribution network over a period of time includes: When economic dispatch imposes requirements on the distribution network's flexible output at a certain moment, the operating domain for the next time series shrinks systematically to accommodate the dynamics of energy storage SOC and component ramp-up constraints. If feasible, the distribution network confirms the plan and updates the flexible operating domain for the next moment. If not feasible, revised feasible boundary information is generated based on the reasons for constraint activation, and the information used by the upper layer in the next round is updated. and cost parameters.

6. The transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network as described in claim 4, characterized in that, Iterate through the upper and lower layers until any of the following preset convergence conditions are met: Condition 1: The PCC plan given by the upper layer is feasible in all time periods of the lower layer, and the change in the objective function is less than the threshold. Condition 2: The FOR area update magnitude is less than the threshold; Condition 3: Reach the maximum number of iterations.

7. A power transmission and distribution pre-decision dispatching system considering the flexibility of the distribution network, characterized in that, The transmission and distribution pre-decision scheduling method considering the flexibility of the distribution network as described in any one of claims 1-6 includes: The prediction module is used to predict photovoltaic output based on the acquired historical power data and weather factor data; The operation domain construction module is used to construct a set of action vectors containing the actions of controllable resources in the distribution network, sample power flow samples from the set of action vectors, calculate the corresponding exchange power at the common connection point between the distribution network and the transmission network based on forward and backward power flow calculation, use the exchange power as the operation domain variable space, and perform feasibility judgment on the power flow samples according to the established distribution network operation constraints to obtain the original dataset of the flexible operation domain of the distribution network. The flexible operating domain boundary determination module is used to train a physical information neural network based on the photovoltaic output prediction results and the distribution network state at the previous moment, learn the mapping from the system state to the flexible operating domain parameters, and output the flexible operating domain boundary parameters for a given period when the system state for that period is input. The equivalent flexibility model construction module is used to solve the day-ahead optimal power flow model considering flexibility resources within the distribution network to obtain the common connection reference operating point and its reference cost, and to construct the distribution network equivalent flexibility model with the incremental cost deviating from the reference operating point. The pre-decision output module is used to dynamically update the boundary parameters of the flexible operating domain of the distribution network over a period of time based on the equivalent flexibility model of the distribution network, using a hierarchical iterative solution framework and introducing time constraints, and to obtain the scheduling pre-decision results.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the transmission and distribution pre-decision scheduling method that takes into account the flexibility of the distribution network as described in any one of claims 1-6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the transmission and distribution pre-decision scheduling method that takes into account the flexibility of the distribution network as described in any one of claims 1-6.

Images