A power grid resource scheduling method and system

Abstract

This invention discloses a power grid resource scheduling method and system, applied in the field of power grid resource scheduling technology. The method includes: aggregating electrical nodes in the initial topology map based on an aggregation scale determined by the topological features of the acquired target power grid to obtain a regional topology map; analyzing real-time operating data of the target power grid to determine power grid scheduling requirements; decomposing the power grid scheduling requirements based on the node characteristics of each regional node in the regional topology map to determine the sub-scheduling requirements of each regional node; extracting each electrical node corresponding to each regional node from the initial topology map and determining the power equipment corresponding to each electrical node; obtaining the equipment scheduling strategy corresponding to each power equipment and controlling the target power grid to execute the target scheduling scheme determined by the equipment scheduling strategy. The power grid resource scheduling method and system provided by this invention can schedule power grid resources in real time, effectively improving the stability of power grid operation.

Description

Technical Field

[0001] This invention relates to the field of power grid resource scheduling technology, and in particular to a power grid resource scheduling method and system. Background Technology

[0002] By scheduling adjustable resources in the power grid, such as through precise peak-shaving of thermal power generation by adjusting fuel levels, real-time deviations in power generation and consumption can be effectively mitigated. This not only improves the absorption of new energy sources such as wind and solar power but also reduces system redundancy reserves and fossil fuel consumption, achieving optimal resource allocation and ensuring the safe and stable operation of the power grid.

[0003] However, existing dispatching technologies mostly employ a day-ahead planning system, whereby dispatching departments formulate the next day's power generation plan in advance based on historical data and load forecasts, and strictly execute it throughout the day. Taking thermal power dispatching as an example, once the day-ahead power generation plan is determined, the output curves of each thermal power unit are fixed. This results in thermal power units being unable to adjust in time when there is a sudden increase in renewable energy output or a sharp decrease in electricity load, forced to curtail large amounts of wind and solar power, ultimately leading to a serious waste of power resources and severely reducing the stability of the power grid operation. Summary of the Invention

[0004] This invention provides a power grid resource scheduling method and system to solve the technical problem that the prior art uses day-ahead planning for scheduling, which leads to serious waste of power resources and severely reduces the stability of power grid operation. The invention aims to achieve real-time scheduling of power grid resources and improve the stability of power grid operation.

[0005] To address the aforementioned technical problems, this invention provides a power grid resource scheduling method and system, the method comprising: Based on the aggregation scale determined by the topological features of the initial topology map of the target power grid, the electrical nodes in the initial topology map are aggregated to obtain a regional topology map; Analyze the real-time operation data of the target power grid to determine the power grid dispatching requirements; Based on the node characteristics of each regional node in the regional topology map, the power grid dispatching requirements are decomposed to determine the sub-dispatch requirements of each regional node. Extract each electrical node corresponding to each region node from the initial topology graph, and determine the power equipment corresponding to each electrical node; Taking the sub-scheduling requirements as the optimization objective, the optimization algorithm is used to solve the scheduling parameters of each power device to obtain the equipment scheduling strategy corresponding to each power device, and the target power grid is controlled to execute the target scheduling scheme determined by each equipment scheduling strategy.

[0006] Preferably, the aggregation of electrical nodes in the initial topology map based on the aggregation scale determined by the topological features of the acquired target power grid's initial topology map to obtain a regional topology map includes: Analyze the connectivity characteristics of each electrical node in the initial topology graph to determine the topology characteristics; The aggregation boundary is determined based on the topological features, and the aggregation scale is determined based on the aggregation boundary; Based on the aggregation scale, the electrical nodes in the initial topology map are aggregated to obtain the regional topology map.

[0007] Preferably, the step of decomposing the power grid dispatching requirements based on the node characteristics of each regional node in the regional topology map to determine the sub-dispatch requirements of each regional node includes: Analyze the power grid dispatch requirements to determine the resources to be regulated; Analyze each electrical node corresponding to each of the said regional nodes to determine the characteristics of each node; Based on the resources to be controlled and the characteristics of each node, a decomposition scheme is obtained; Based on the decomposition scheme, the sub-scheduling requirements of each of the aforementioned regional nodes are obtained.

[0008] Preferably, the step of using each of the sub-scheduling requirements as the optimization objective, employing an optimization algorithm to solve for the scheduling parameters of each of the power devices, obtaining the equipment scheduling strategy corresponding to each of the power devices, and controlling the target power grid to execute the target scheduling scheme determined by each of the equipment scheduling strategies includes: Extract scheduling constraint information from the power grid scheduling requirements; For each of the sub-scheduling requirements, the scheduling constraint information is used as a constraint condition, and the scheduling parameters of each of the power devices are solved using the alternating direction multiplier technique to obtain the device scheduling strategy. The solution uses Markov decision technology to quantify the scheduling process. The target scheduling scheme for the target power grid is determined based on the scheduling strategies of each of the aforementioned devices; Control the target power grid to execute the target scheduling scheme.

[0009] Preferably, after taking each of the sub-scheduling requirements as the optimization objective, using an optimization algorithm to solve for the scheduling parameters of each of the power devices, obtaining the equipment scheduling strategy corresponding to each of the power devices, and controlling the target power grid to execute the target scheduling scheme determined by each of the equipment scheduling strategies, the process includes: The target scheduling scheme and the power grid scheduling requirements are analyzed using deep learning technology. A scheduling model is constructed based on the analysis results. The scheduling model is used to generate scheduling schemes at least.

[0010] Another aspect of the present invention provides a power grid resource scheduling system, comprising: The aggregation module is used to aggregate the electrical nodes in the initial topology map based on the aggregation scale determined by the topological features of the initial topology map of the acquired target power grid, so as to obtain a regional topology map. The analysis module is used to analyze the real-time operating data of the target power grid and determine the power grid dispatching requirements; The determination module is used to decompose the power grid dispatching requirements based on the node characteristics of each regional node in the regional topology map, and determine the sub-dispatch requirements of each regional node. An extraction module is used to extract each electrical node corresponding to each region node from the initial topology map, and to determine the power equipment corresponding to each electrical node; The solution module is used to solve the scheduling parameters of each of the power devices by using optimization algorithms with each of the sub-scheduling requirements as the optimization objective, to obtain the equipment scheduling strategy corresponding to each of the power devices, and to control the target power grid to execute the target scheduling scheme determined by each of the equipment scheduling strategies.

[0011] Preferably, the aggregation module includes: A connectivity unit is used to analyze the connectivity characteristics of each electrical node in the initial topology graph and determine the topology characteristics; Boundary units are used to determine the aggregation boundary based on the topological features and to determine the aggregation scale based on the aggregation boundary; A topology graph unit is used to aggregate the electrical nodes in the initial topology graph based on the aggregation scale to obtain the regional topology graph.

[0012] Preferably, the determining module includes: The resource unit is used to analyze the power grid dispatching demand and determine the resources to be regulated. A feature unit is used to analyze each electrical node corresponding to each of the regional nodes and determine the features of each node. The scheme unit is used to obtain a decomposition scheme based on the resource to be controlled and the characteristics of each node; The decomposition unit is used to obtain the sub-scheduling requirements of each of the regional nodes based on the decomposition scheme.

[0013] Preferably, the solution module includes: A constraint unit is used to extract scheduling constraint information from the power grid scheduling requirements; The parameter unit is used to solve the scheduling parameters of each of the power devices for each of the sub-scheduling requirements, with the scheduling constraint information as the constraint condition, using the alternating direction multiplier technique to obtain the device scheduling strategy. The solution uses Markov decision technology to quantify the scheduling process. The target unit is used to determine the target scheduling scheme of the target power grid based on the scheduling strategies of each of the devices. An execution unit is used to control the target power grid to execute the target scheduling scheme.

[0014] Preferably, after the solution module, the following is included: The model unit is used to analyze the target scheduling scheme and the power grid scheduling requirements using deep learning technology, and to construct a scheduling model based on the analysis results. The scheduling model is used to generate a scheduling scheme at least.

[0015] Compared with the prior art, the beneficial effects of the present invention are at least one of the following: This invention first determines the aggregation scale based on the topological characteristics of the initial topology of the target power grid, aggregating each electrical node into a regional topology to reduce the problem scale. Then, it analyzes real-time operational data to determine power grid dispatching requirements and decomposes the total demand into sub-dispatch requirements for each region using the node characteristics of the regional nodes. Next, it extracts the specific electrical nodes and power equipment corresponding to each region from the initial topology. Finally, using each sub-dispatch requirement as the optimization objective, it employs an optimization algorithm to solve the dispatching parameters of each power equipment layer by layer, forming a globally coordinated equipment dispatching strategy and controlling the power grid execution. This invention, through real-time operational data-driven dynamic decomposition and solution, breaks through the time scale limitations of day-ahead planning, enabling adjustable resources such as thermal power to flexibly adjust their output according to the real-time state of the power grid. Simultaneously, the aggregation and partitioning strategy significantly reduces the computational complexity of large-scale power grid collaborative optimization, allowing refined and differentiated sub-dispatch requirements to be efficiently solved and implemented. This effectively enhances the power grid's response to power fluctuations and operational stability while improving the absorption of new energy sources and reducing resource waste. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a power grid resource scheduling method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a power grid resource scheduling system in one embodiment of the present invention; Figure label: The module consists of: 11. Aggregation module; 12. Analysis module; 13. Determination module; 14. Extraction module; and 15. Solving module. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0019] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0021] Flexible scheduling of grid-adjustable resources can smooth out power generation and consumption deviations, improve the absorption of new energy sources, reduce fossil fuel consumption, and ensure grid stability. However, existing scheduling systems mostly adopt day-ahead planning, which fixes the output plans of thermal power units in advance. This makes it impossible to adjust in a timely manner according to real-time fluctuations such as sudden increases in new energy output or sudden decreases in load, leading to forced curtailment of wind and solar power, resulting in waste of power resources and reduced operational stability.

[0022] One embodiment of the present invention provides a power grid resource scheduling method; for details, please refer to [link to specific documentation]. Figure 1 , Figure 1 The diagram shown is a flowchart illustrating a power grid resource scheduling method according to one embodiment of the present invention, including: S1. Based on the aggregation scale determined by the topological features of the initial topology map of the target power grid, the electrical nodes in the initial topology map are aggregated to obtain the regional topology map. S2. Analyze the real-time operation data of the target power grid to determine the power grid dispatching requirements; S3. Based on the node characteristics of each regional node in the regional topology map, the power grid dispatching requirements are decomposed to determine the sub-dispatch requirements of each regional node. S4. Extract each electrical node corresponding to each region node from the initial topology graph, and determine the power equipment corresponding to each electrical node; S5. Taking the sub-scheduling requirements as the optimization objective, the optimization algorithm is used to solve the scheduling parameters of each power equipment to obtain the equipment scheduling strategy corresponding to each power equipment, and the target power grid is controlled to execute the target scheduling scheme determined by the scheduling strategy of each equipment.

[0023] First, based on the aggregation scale determined by the topological features of the initial topology map of the target power grid, the electrical nodes in the initial topology map are aggregated to obtain a regional topology map. The connectivity features of each electrical node in the initial topology map are analyzed to determine the topological features; aggregation boundaries are determined based on the topological features, and the aggregation scale is determined based on the aggregation boundaries; finally, the electrical nodes in the initial topology map are aggregated based on the aggregation scale to obtain the regional topology map.

[0024] The initial topology diagram is a complete network model that directly reflects the original physical connections of the target power grid, where electrical nodes correspond to actual electrical connection points such as buses and substations. Connectivity features characterize the tightness of electrical coupling between electrical nodes, while topology features are structural indicators extracted after statistical analysis of connectivity features. Aggregation boundaries are partitioning criteria or threshold conditions determined based on topology features, and aggregation scales are the aggregation granularity or number of region divisions set under the guidance of these boundaries. The regional topology diagram is a simplified network model obtained by merging the original electrical nodes based on the aggregation scale, where regional nodes represent an equivalent set of aggregated electrical nodes.

[0025] An initial topology map of the target power grid is obtained, which fully depicts the physical connections of the grid using electrical nodes and branches. Based on this, the connectivity characteristics between each electrical node are analyzed one by one. This involves calculating quantitative indicators such as impedance magnitude, power flow transfer sensitivity, or voltage / frequency influence factor between nodes to assess the tightness of electrical coupling between any two nodes. Subsequently, community detection algorithms, such as spectral clustering, are used to statistically analyze these connectivity characteristics, extracting topological features that reflect the natural partitioning tendency of the power grid, such as the similarity matrix between nodes or the community structure of the network. Next, the aggregation boundary is determined based on these topological features, i.e., a partitioning threshold or objective function is set, such as maximizing the coupling degree within a region or minimizing the coupling degree between regions. The aggregation scale is then determined accordingly. This scale is specifically reflected in the final number of regions or the upper limit of the number of nodes allowed in each region, and is the core parameter controlling the coarsness of the aggregation.

[0026] After determining the aggregation scale, it is used as the input parameter for the clustering algorithm to perform aggregation operations on all electrical nodes in the initial topology graph. Specifically, hierarchical clustering or K-means clustering algorithms based on topological features can be used. The number of regions K in the aggregation scale is used as the number of cluster centers, and each electrical node is assigned to the region corresponding to the cluster center with which it is most electrically coupled. This divides the original large number of electrical nodes into K non-overlapping node sets. For each node set, all electrical nodes within it are equivalent to a region node. The attributes of this region node, such as equivalent load, equivalent generating capacity, and equivalent inertia, are obtained by weighted summation or aggregation of the corresponding attributes of each node within the set. Simultaneously, the connectivity between regions, such as equivalent line impedance or transmission capacity, is obtained by aggregating the electrical parameters of all original branches between sets. Finally, based on these region nodes and inter-region connectivity relationships, a regional topology graph with a significantly reduced scale but retaining key topological association information is constructed for subsequent hierarchical scheduling.

[0027] Preferably, the real-time operating data of the target power grid is analyzed to determine the grid dispatch requirements. Real-time operating data includes dynamic monitoring information that can be collected within seconds to minutes, such as bus voltage, branch power flow, generator output, load power, renewable energy generation output, and frequency. Grid dispatch requirements are global control objectives determined after analyzing the aforementioned real-time operating data. Specifically, these can be quantified as the total power deficit or surplus that the system needs to adjust, the frequency deviation that needs to be mitigated, the line over-limit risks that need to be eliminated, and the renewable energy absorption that needs to be maximized. These serve as the overall input instructions for subsequent hierarchical decomposition and equipment strategy solution.

[0028] Various operational data are collected in real time at sampling frequencies of seconds or milliseconds by data acquisition and monitoring systems and phasor measurement units deployed at various monitoring points in the power grid. This data includes the voltage amplitude and phase angle of each bus, active and reactive power flow of each branch, actual output of each generator, power consumption of each load node, actual power generation of new energy sources, and system frequency. Next, the collected raw data is sent to a data preprocessing module for cleaning operations such as bad data identification and removal, measurement data normalization, and time-scale alignment to ensure data integrity and consistency. Subsequently, based on the preprocessed real-time data, state estimation calculations are performed to optimally estimate the current operating state of the power grid, complete missing measurements, and filter out noise interference, resulting in a complete and reliable set of full-network operational profile data. Based on this, the operational cross-sectional data is compared item by item with pre-defined power grid safety operation boundaries. These boundaries include line thermal stability limits, bus voltage upper and lower limits, frequency allowable deviation range, and generator regulation upper and lower limits. Simultaneously, the current renewable energy absorption margin and spinning reserve capacity are comprehensively assessed to identify various deviations or over-limit risks currently existing in the system, such as frequency deviation, power imbalance, line overload, or voltage over-limit. Finally, with the goal of eliminating the aforementioned deviations and risks, and using system safety constraints as boundaries, the overall control target that the power grid needs to execute at the current moment is calculated, i.e., the power grid dispatch demand. This demand can be specifically quantified as the total active power value that the system needs to adjust, the target value of frequency deviation that needs to be smoothed, the power adjustment amount of specific over-limit lines that need to be eliminated, and the additional renewable energy generation that needs to be absorbed or restricted. These indicators serve as the overall input instructions for subsequent hierarchical decomposition and strategy solution.

[0029] Preferably, the power grid dispatching demand is decomposed based on the node characteristics of each regional node in the regional topology map to determine the sub-dispatch demand of each regional node. The power grid dispatching demand is analyzed to determine the resources to be regulated; the characteristics of each electrical node corresponding to each regional node are determined; based on the resources to be regulated and the characteristics of each node, a decomposition scheme is obtained; and based on the decomposition scheme, the sub-dispatch demand of each regional node is obtained.

[0030] A regional topology map is a simplified network model obtained by merging the original electrical nodes of the power grid according to the aggregation scale. Regional nodes represent the equivalent set of a group of aggregated electrical nodes. Node characteristics refer to the attribute indicators possessed by each regional node, specifically including the equivalent total load, equivalent generating capacity, current power deficit or surplus, upper limit of regulation rate, and internal line transmission margin within the region. Power grid dispatching demand is a global overall control target determined based on real-time operational data analysis. The resources to be controlled refer to all flexible resources available for adjustment to achieve this overall target, including thermal power units whose output can be increased or decreased. The system includes energy storage devices that can be charged and discharged, interruptible loads that can be reduced or transferred, and new energy power stations that can limit or increase power generation. The decomposition scheme is a strategy formed by comprehensively considering the global total constraints of the resources to be regulated and the node characteristics of each regional node. It allocates the total scheduling demand in a differentiated manner according to the actual regulation capacity of each region. This scheme ultimately determines each sub-scheduling demand, that is, the specific regulation task that each regional node needs to undertake in this scheduling cycle, such as the value of active power to be increased or decreased, the magnitude of reactive power adjustment, or the contribution of frequency support, etc., as local inputs for subsequent equipment-level strategy solutions.

[0031] In practice, the overall control target type and total adjustment value are extracted from the grid dispatch demand. It is then determined whether the current dispatch requires adjustment of active or reactive power, and the total increase or decrease required. Based on this, the entire network's power equipment ledger information is traversed to select all flexible resources with adjustment capabilities as resources to be controlled. These include the current output and adjustable upper and lower limits of each thermal power unit, the current state of charge and charging / discharging power limits of energy storage devices, the contracted capacity and response time of interruptible loads, and the current output and potential for limiting renewable energy generation at renewable energy plants. This forms a complete list of adjustable resources, with the original electrical node to which each resource belongs marked.

[0032] After identifying the resources to be regulated, for each regional node in the regional topology map, it is first back-mapped to all the original electrical nodes it contains. Then, the information of the resources to be regulated corresponding to these electrical nodes is summarized to calculate the node characteristics of the region. Specifically, the calculation method is as follows: the sum of the adjustable capacities of all thermal power units in the region is used as the upper and lower limit of the regulation capacity; the available capacity of all interruptible loads is accumulated as the load-side regulation potential; the charging and discharging power limits of energy storage devices are accumulated and the net value is taken as the energy storage regulation capacity; simultaneously, the difference between the equivalent total load and the equivalent power generation output in the region is calculated as the basic imbalance degree. These indicators are combined to form the regulation capacity vector of the region. Next, using the total regulation amount in the grid dispatch demand as the total constraint, and the regulation capacity in the node characteristics of each regional node as the allocation weight, the total dispatch demand is decomposed into specific values ​​that each region must bear using rules such as allocation by capacity ratio or allocation by marginal cost. Finally, the regulation value allocated to each region, along with the regulation direction and regulation time requirements, is encapsulated into the sub-dispatch demand of that region, serving as an independent input instruction for subsequent equipment-level optimization solutions.

[0033] Preferably, each electrical node corresponding to each region node is extracted from the initial topology graph, and the power equipment corresponding to each electrical node is determined.

[0034] The initial topology diagram is a complete network model that directly reflects the original physical connections of the target power grid. The regional nodes are equivalent sets formed by merging multiple original electrical nodes in the aforementioned aggregation process. Each regional node corresponds to a subset composed of several original electrical nodes. The electrical nodes specifically refer to actual electrical connection points such as buses and substations. Power equipment refers to various controllable physical objects that are attached to each electrical node or connected to the node through branches, including generators, energy storage devices, transformers, reactive power compensation devices, and adjustable loads. Determining the power equipment corresponding to each electrical node involves consulting the power grid equipment ledger or topology association data to establish a mapping relationship from each original electrical node to the list of specific equipment it is connected to.

[0035] The mapping table between regional nodes and original electrical nodes generated during the aggregation phase is analyzed. This table records the IDs of which original electrical nodes each regional node contains. By directly querying this table, each regional node can be traversed and all its corresponding electrical nodes can be extracted. Next, for each extracted electrical node, using its unique ID as an index, a search is performed in the power grid topology database or equipment ledger management system to retrieve all records of power equipment directly associated with that electrical node through bus connection points, switches, or disconnectors. These devices include generator sets, energy storage devices, reactive power compensation capacitors, reactors, and adjustable loads connected to that node. Finally, the query results are organized and stored according to a hierarchical structure of "regional node - electrical node - power equipment" to form a complete equipment mapping list. This ensures that each electrical node can locate its corresponding regional node and all connected equipment information for direct retrieval during subsequent equipment-level scheduling parameter solving.

[0036] Finally, taking each sub-scheduling demand as the optimization objective, an optimization algorithm is used to solve for the scheduling parameters of each power device, obtaining the corresponding device scheduling strategy, and controlling the target power grid to execute the target scheduling scheme determined by each device scheduling strategy. Scheduling constraint information is extracted from the power grid scheduling demand; for each sub-scheduling demand, using the scheduling constraint information as constraints, the alternating direction multiplier technique is used to solve for the scheduling parameters of each power device, obtaining the device scheduling strategy. The solution uses Markov decision technology to quantify the scheduling process; based on the scheduling strategies of each device, the target scheduling scheme of the target power grid is determined; and the target power grid is controlled to execute the target scheduling scheme.

[0037] Sub-scheduling requirements are the specific adjustment tasks that each regional node needs to undertake in this scheduling; optimization algorithms are calculation methods used to solve for optimal scheduling parameters; power equipment are controllable physical devices connected to each electrical node; scheduling parameters refer to the operating variables that need to be determined for each power device; equipment scheduling strategies refer to the specific output plans or action instructions obtained after optimization for each power device; the target grid is the specific power system network for executing scheduling; the target scheduling scheme is a globally coordinated execution scheme formed by summarizing all equipment scheduling strategies; scheduling constraint information is the safe operation boundary condition extracted from the grid scheduling requirements; alternating direction multiplier technique is a distributed optimization algorithm; Markov decision technique is a mathematical framework for characterizing time-series decision-making processes, used to quantify state transitions and strategy evaluations during the scheduling process.

[0038] This paper analyzes the safety boundary conditions attached to the power grid dispatching requirements, and compiles the upper and lower limits of safe operation of bus voltage amplitude, the thermal stability limits of active power of each transmission line and transformer, the upper and lower limits of output regulation of each generator unit and the ramp rate limit, the safe state-of-charge range and charging and discharging power limits of energy storage devices, and the allowable frequency deviation range. These constraints are uniformly transformed into mathematical expressions, serving as equality and inequality constraints in the subsequent solution process. Based on this, the sub-dispatch requirements of each regional node are transformed into optimization objective functions that the dispatching parameters of each power equipment in its region should satisfy. For example, the objective might be to minimize the generation cost of the region or maximize the absorption of new energy. Variables such as generator output, energy storage charging and discharging power, and adjustable load power in the region are defined as dispatching parameters to be solved, forming independent optimization sub-problems for each region.

[0039] For each region's optimization subproblem, a distributed solution is employed using the alternating direction multiplier technique. Specifically, the coupling variables and Lagrange multipliers for each region are initialized. Then, in each iteration, the following steps are executed sequentially: each region independently solves its own equipment scheduling parameter optimization subproblem, then passes the calculation results of the boundary variables between regions to neighboring regions. Each region receives the boundary variables from neighboring regions, updates its own Lagrange multipliers, and repeats the above solution and exchange process until the global residual meets the preset convergence accuracy threshold. During this solution process, considering the temporal recursive nature of scheduling decisions, Markov decision technology is used to quantitatively model the scheduling process. Specifically, the scheduling parameters of each power device at different times are defined as state variables, the iterative steps in the optimization algorithm or the power adjustment actions within the actual scheduling period are defined as action variables, and the degree of satisfaction of sub-scheduling requirements and the penalty for constraint violation are defined as reward functions. The contribution of each adjustment step to the final goal is quantified by evaluating the cumulative expected value of state-action-reward, guiding the selection of the optimization direction and the determination of the iteration step size.

[0040] After the alternating direction multiplier technique reaches convergence, the optimal scheduling parameter solutions for each power device in each region are output. These solutions include the active and reactive power output setpoints of generators, the planned charging and discharging power curves of energy storage, the power adjustment amounts of adjustable loads, and the control levels of reactive power compensation devices. All equipment scheduling parameters for all regions are aggregated and integrated according to equipment ID to form a set of equipment scheduling strategies covering the entire network. Subsequently, based on this set, a complete target scheduling scheme for the target power grid within the current scheduling cycle is developed. This scheme clearly defines the execution instructions for each power device at each time point during the scheduling period. Finally, through the remote control channel of the dispatch automation system, the instructions for each device in the target scheduling scheme are sent as remote control commands to the automatic control systems of each power plant, substation, and energy storage station. The execution terminals then perform actual operations such as output adjustment and switching actions according to the received instructions, thereby realizing the execution of the scheduling scheme by the target power grid.

[0041] After obtaining the target scheduling scheme, deep learning technology is used to analyze the target scheduling scheme and power grid scheduling requirements. Based on the analysis results, a scheduling model is constructed, which is used to generate at least the scheduling scheme.

[0042] The target scheduling scheme is a globally coordinated execution scheme that includes specific execution instructions for each power device, obtained after optimization. The grid scheduling demand is a global overall control target determined based on real-time operation data analysis. Deep learning technology is a data-driven modeling method based on multi-layer neural networks, which can automatically extract complex nonlinear mapping relationships from a large number of input and output samples. The analysis result refers to the mapping rules or parameter weights obtained after training and learning the intrinsic relationship between the target scheduling scheme and the grid scheduling demand through a deep learning model. The scheduling model is a proxy model built based on the above analysis results. This model takes real-time grid operation data or scheduling demand as input and can directly output the corresponding scheduling scheme. It at least has the function of replacing the original optimization solution process to quickly generate feasible scheduling schemes.

[0043] After obtaining the target scheduling scheme through optimization algorithms each time, the scheme is paired with the current input grid scheduling demand to form a set of training samples from the input scheduling demand to the output scheduling scheme. A large amount of sample data is continuously accumulated over multiple scheduling cycles to build a sample library covering various grid operating conditions. Next, these samples undergo data preprocessing. Various indicators in the grid scheduling demand, such as power deficit, frequency deviation, and the location and degree of exceeding limits, are encoded into multi-dimensional input feature vectors for the deep learning model. The scheduling parameters of each power device in the target scheduling scheme are encoded into output label vectors. The sample library is then divided into training, validation, and test sets. Next, a deep learning network structure suitable for this task is designed, such as a multi-layer fully connected neural network or a temporal neural network. The input feature vectors are mapped to the output label vectors, and the loss function and optimizer are determined. The network is iteratively trained using the training set samples, and the network weights are continuously updated through backpropagation until the model's prediction error on the validation set meets the accuracy requirements. Finally, the trained network model and its weight parameters are solidified and saved to form a deployable scheduling model. When the model receives a new power grid scheduling request, it can directly output the corresponding scheduling scheme through forward propagation calculation without having to re-execute the time-consuming iterative optimization solution process, thus achieving the generation of scheduling schemes in seconds or milliseconds.

[0044] The above embodiments provide a general technical framework for power grid dispatching based on topology aggregation and hierarchical decomposition. By determining the regional topology map through aggregation scale, decomposing dispatching requirements, solving equipment strategies, and constructing a deep learning dispatching model, hierarchical control from the global to the equipment level is achieved. The following embodiments are specific implementations of this general framework in the distribution network scenario. It concretizes the regional topology map into a stepped partition structure, refines regional nodes into a three-level hierarchical architecture of regional agents, local agents, and unit agents, quantifies power grid dispatching requirements into a multi-objective optimization function, concretizes the optimization algorithm into a solution strategy combining the alternating direction multiplier method and NSGA-II (Nondominated Sorting Genetic Algorithm II), concretizes deep learning technology into a GRU (Gated Recurrent Unit) gated recurrent network, and connects each link into a practically deployable system through a multi-timescale closed-loop mechanism, thus fully realizing the entire process from theoretical architecture to engineering implementation of the above embodiments.

[0045] Another embodiment of this invention proposes a tiered source-load adjustable resource multi-agent collaborative regulation and optimization method and system, aiming to address the increasingly complex regulation and control needs of distribution networks under the high proportion of distributed energy and new load access. First, in the spatial dimension, the method divides the entire distribution network into three tiered zones based on indicators such as distribution area boundaries, electrical topology, resource distribution density, and regulation capacity: a global overall planning area, an intermediate collaborative sub-area, and a basic autonomous distribution area. On this basis, a three-level hierarchical architecture of regional agents, local agents, and unit agents is established vertically, thereby forming a three-dimensional regulation and control structure with intra-area autonomy, cross-area collaboration, and vertical linkage.

[0046] At the model construction level, this embodiment of the invention uses the power supply and demand balance of the distribution network, inter-regional power interaction, node voltage security, and the operational boundaries of various adjustable resources as the main constraints to construct a distributed collaborative optimization model that takes into account global comprehensive benefits, regional power coordination, and local equipment autonomy. This model adopts a multi-objective optimization framework, comprehensively considering four indicators: economic cost, renewable energy absorption rate, network loss, and voltage quality. Through a Softmax adaptive weighting mechanism, the priority of each objective is dynamically adjusted based on real-time electricity prices, renewable energy output, and load fluctuations, thereby improving the model's adaptability to complex operating scenarios.

[0047] To achieve efficient solution, this embodiment of the invention models the regulation process as a Markov decision process, explicitly defining elements such as environment, state, action, policy, reward, and outcome, characterizing the dynamic characteristics of source resources from instantaneous response to long-term evolution. In terms of solution strategy, the alternating direction multiplier method is used to decompose the global optimization problem into multiple local subproblems, which are solved independently by each agent and iterated collaboratively until the original residual and dual residual converge, outputting a baseline regulation solution. Based on this baseline solution, a high-quality initial population is generated by combining Haldane low-discrepancy sequences and normal distribution perturbations. Further iterative optimization using NSGA-II is then performed to obtain the Pareto optimal policy set, thereby providing diverse feasible solutions among multiple conflicting objectives.

[0048] Finally, this embodiment of the invention introduces a multi-agent deep learning model based on a GRU-gated recurrent network. Taking historical electricity prices, renewable energy output, and load time-series data as input, and optimal control parameters as output, it fits the nonlinear mapping relationship from scenario to strategy through supervised learning, achieving millisecond-level strategy generation under real-time operating conditions. Simultaneously, the system establishes a multi-timescale closed-loop control mechanism covering second-level local response, hourly rolling correction, daily scheme generation, weekly parameter updates, and monthly / quarterly strategy evolution, ensuring continuous optimization and evolution of the control strategy. Overall, this method combines refined zoning, hierarchical collaboration, distributed computing, and intelligent decision-making capabilities, significantly improving the operational economy, security, and adaptability of the distribution network in scenarios with high renewable energy penetration.

[0049] This invention utilizes a 10kV distribution network in the central urban area of ​​a city as an application scenario to build a hierarchical architecture. This network covers 3 streets and 12 power supply substations, connecting 8.6MW of distributed photovoltaic power, 1.2MW of wind power, 5MWh of energy storage, 3.8MW of flexible and controllable loads, and 210 charging piles. First, based on electrical topology, resource distribution, and load characteristics, a tiered partitioning system is established, dividing the area into 1 overall coordinated region, 3 intermediate collaborative sub-regions (Eastern New Energy Agglomeration Area, Central Load Core Area, and Western Energy Storage Supporting Area), and 12 basic autonomous substations. Simultaneously, a three-level intelligent agent hierarchical architecture (regional, local, and unit) is built, configuring four types of communication protocols. Cross-level communication latency is controlled within 8 seconds, forming a hierarchical and clearly defined control system. Based on this, a distributed collaborative optimization model is constructed, setting constraints such as node voltage (0.93~1.07 pu), energy storage state of charge (0.2~0.95), distributed power output, and cross-regional power interaction. A four-dimensional target system is established, encompassing economic efficiency, grid absorption, network loss, and voltage. Softmax adaptive weights are used to dynamically adjust target priorities based on electricity prices, renewable energy sources, and load fluctuations. The control process is modeled using a Markov decision process, defining multi-level state and action spaces, with a reward mechanism that integrates positive incentives and penalties for exceeding limits. The solution employs an alternating direction multiplier method for distributed iteration, with each agent independently solving its local problem and collaboratively updating. The original residual and dual residual are less than 10. -4 The system converges and outputs a baseline solution. Then, an initial population of size 100 is generated centered on this solution. After 50 generations of evolution using NSGA-II, a Pareto (Pareto Optimality) optimal policy set is obtained. Finally, a GRU network is used to train the mapping relationship between time-series data and control parameters, achieving millisecond-level real-time decision-making. The system establishes a multi-timescale closed-loop mechanism encompassing second-level local response, hourly rolling correction, daily scheme generation, weekly parameter updates, and monthly / quarterly policy evolution. Actual trial operation shows that this method effectively improves source-load imbalance, wind and solar power curtailment, and voltage fluctuation problems, significantly enhancing the economic efficiency and security of the distribution network, verifying its feasibility and superiority.

[0050] Another embodiment of the present invention provides a power grid resource scheduling system; for details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram shown is a structural schematic of a power grid resource scheduling system according to one embodiment of the present invention, comprising: Aggregation module 11 is used to aggregate each electrical node in the initial topology map based on the aggregation scale determined by the topology features of the initial topology map of the acquired target power grid, so as to obtain a regional topology map. Analysis module 12 is used to analyze the real-time operating data of the target power grid and determine the power grid dispatching requirements; The determination module 13 is used to decompose the power grid dispatching requirements based on the node characteristics of each regional node in the regional topology map, and determine the sub-dispatch requirements of each regional node. Extraction module 14 is used to extract each electrical node corresponding to each region node from the initial topology map and determine the power equipment corresponding to each electrical node; The solution module 15 is used to solve the scheduling parameters of each power equipment with each sub-scheduling requirement as the optimization objective, and obtain the equipment scheduling strategy corresponding to each power equipment, and control the target power grid to execute the target scheduling scheme determined by each equipment scheduling strategy.

[0051] Preferably, the aggregation module 11 includes: Connectivity units are used to analyze the connectivity characteristics of each electrical node in the initial topology graph and determine the topology characteristics. Boundary units are used to determine aggregation boundaries based on topological features and to determine aggregation scales based on aggregation boundaries; Topology graph cells are used to aggregate electrical nodes in the initial topology graph based on the aggregation scale to obtain a regional topology graph.

[0052] Preferably, the determining module 13 includes: Resource unit, used to analyze power grid dispatching demand and determine the resources to be regulated; Feature units are used to analyze each electrical node corresponding to each regional node and determine the characteristics of each node. The scheme unit is used to obtain a decomposition scheme based on the resources to be controlled and the characteristics of each node; The decomposition unit is used to obtain the sub-scheduling requirements of each regional node based on the decomposition scheme.

[0053] Preferably, the solver module 15 includes: The constraint unit is used to extract scheduling constraint information from power grid scheduling requirements; The parameter unit is used to solve the scheduling parameters of each power equipment for each sub-schedule requirement, with scheduling constraint information as the constraint condition, and adopts the alternating direction multiplier technique to obtain the equipment scheduling strategy. The solution adopts the Markov decision technique to quantify the scheduling process. The target unit is used to determine the target scheduling scheme for the target power grid based on the scheduling strategies of each device. The execution unit is used to control the target power grid to execute the target scheduling scheme.

[0054] Preferably, after the solver module 15, the following is included: The model unit is used to analyze the target scheduling scheme and power grid scheduling requirements using deep learning technology, and to build a scheduling model based on the analysis results. The scheduling model is used to generate scheduling schemes at least.

[0055] Accordingly, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform steps in the power grid resource scheduling method of the above embodiments, for example... Figure 1 Steps S1 to S5 as described above.

[0056] This invention does not directly perform global unified modeling of massive power equipment. Instead, it first determines the aggregation scale based on the connectivity characteristics of the initial power grid topology, abstracting the vast power grid into a hierarchical regional topology, thus achieving a reduction in scale by breaking down the entire system into smaller parts. Based on this, relying on online analysis of real-time operational data, the global scheduling requirements are differentiated according to the inherent characteristics of each regional node, enabling each region to obtain a sub-scheduling objective matching its regulation capacity. Subsequently, it backtracks layer by layer to the specific electrical nodes and power equipment in the initial topology, and, guided by the sub-scheduling requirements, uses optimization algorithms to solve the equipment-level scheduling parameters in a distributed manner, ultimately merging them into a globally collaborative target scheduling scheme. This invention enables adjustable resources such as thermal power to no longer be limited by a fixed day-ahead curve, but can flexibly adjust output based on real-time fluctuations. Simultaneously, through a hierarchical solution structure, the originally difficult-to-handle ultra-large-scale non-convex optimization problem is decomposed into multiple sub-problems of controllable scale, significantly improving solution efficiency and engineering feasibility. Ultimately, while ensuring the safety boundary of the power grid, it maximizes the capacity for renewable energy absorption and minimizes resource consumption.

[0057] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A power grid resource scheduling method, characterized in that, include: Based on the aggregation scale determined by the topological features of the initial topology map of the target power grid, the electrical nodes in the initial topology map are aggregated to obtain a regional topology map; Analyze the real-time operation data of the target power grid to determine the power grid dispatching requirements; Based on the node characteristics of each regional node in the regional topology map, the power grid dispatching requirements are decomposed to determine the sub-dispatch requirements of each regional node. Extract each electrical node corresponding to each region node from the initial topology graph, and determine the power equipment corresponding to each electrical node; Taking the sub-scheduling requirements as the optimization objective, the optimization algorithm is used to solve the scheduling parameters of each power device to obtain the equipment scheduling strategy corresponding to each power device, and the target power grid is controlled to execute the target scheduling scheme determined by each equipment scheduling strategy.

2. The power grid resource scheduling method as described in claim 1, characterized in that, The aggregation scale, determined based on the topological features of the initial topology map of the acquired target power grid, aggregates the electrical nodes in the initial topology map to obtain a regional topology map, including: Analyze the connectivity characteristics of each electrical node in the initial topology graph to determine the topology characteristics; The aggregation boundary is determined based on the topological features, and the aggregation scale is determined based on the aggregation boundary; Based on the aggregation scale, the electrical nodes in the initial topology map are aggregated to obtain the regional topology map.

3. The power grid resource scheduling method as described in claim 1, characterized in that, The step of decomposing the power grid dispatching requirements based on the node characteristics of each regional node in the regional topology map, and determining the sub-dispatch requirements of each regional node, includes: Analyze the power grid dispatch requirements to determine the resources to be regulated; Analyze each electrical node corresponding to each of the said regional nodes to determine the characteristics of each node; Based on the resources to be controlled and the characteristics of each node, a decomposition scheme is obtained; Based on the decomposition scheme, the sub-scheduling requirements of each of the aforementioned regional nodes are obtained.

4. The power grid resource scheduling method as described in claim 1, characterized in that, The step of using each of the sub-scheduling requirements as the optimization objective, employing an optimization algorithm to solve for the scheduling parameters of each of the power devices, obtaining the equipment scheduling strategy corresponding to each of the power devices, and controlling the target power grid to execute the target scheduling scheme determined by each of the equipment scheduling strategies includes: Extract scheduling constraint information from the power grid scheduling requirements; For each of the sub-scheduling requirements, the scheduling parameters of each of the power devices are solved using the scheduling constraint information as a constraint condition, and the device scheduling strategy is obtained by using the alternating direction multiplier technique. The solution is quantified by Markov decision technology. The target scheduling scheme for the target power grid is determined based on the scheduling strategies of each of the aforementioned devices; Control the target power grid to execute the target scheduling scheme.

5. The power grid resource scheduling method as described in claim 1, characterized in that, After taking each of the sub-scheduling requirements as the optimization objective, solving the scheduling parameters of each of the power devices using an optimization algorithm to obtain the equipment scheduling strategy corresponding to each of the power devices, and controlling the target power grid to execute the target scheduling scheme determined by each of the equipment scheduling strategies, the process includes: The target scheduling scheme and the power grid scheduling requirements are analyzed using deep learning technology. A scheduling model is constructed based on the analysis results. The scheduling model is used to generate scheduling schemes at least.

6. A power grid resource scheduling system, characterized in that, include: The aggregation module is used to aggregate the electrical nodes in the initial topology map based on the aggregation scale determined by the topological features of the initial topology map of the acquired target power grid, so as to obtain a regional topology map. The analysis module is used to analyze the real-time operating data of the target power grid and determine the power grid dispatching requirements; The determination module is used to decompose the power grid dispatching requirements based on the node characteristics of each regional node in the regional topology map, and determine the sub-dispatch requirements of each regional node. An extraction module is used to extract each electrical node corresponding to each region node from the initial topology map, and to determine the power equipment corresponding to each electrical node; The solution module is used to solve the scheduling parameters of each of the power devices by using optimization algorithms with each of the sub-scheduling requirements as the optimization objective, to obtain the equipment scheduling strategy corresponding to each of the power devices, and to control the target power grid to execute the target scheduling scheme determined by each of the equipment scheduling strategies.

7. The power grid resource scheduling system as described in claim 6, characterized in that, The aggregation module includes: A connectivity unit is used to analyze the connectivity characteristics of each electrical node in the initial topology graph and determine the topology characteristics; Boundary units are used to determine the aggregation boundary based on the topological features and to determine the aggregation scale based on the aggregation boundary; A topology graph unit is used to aggregate the electrical nodes in the initial topology graph based on the aggregation scale to obtain the regional topology graph.

8. The power grid resource dispatching system as described in claim 6, characterized in that, The determining module includes: The resource unit is used to analyze the power grid dispatching demand and determine the resources to be regulated. A feature unit is used to analyze each electrical node corresponding to each of the regional nodes and determine the features of each node. The scheme unit is used to obtain a decomposition scheme based on the resource to be controlled and the characteristics of each node; The decomposition unit is used to obtain the sub-scheduling requirements of each of the regional nodes based on the decomposition scheme.

9. The power grid resource dispatching system as described in claim 6, characterized in that, The solution module includes: A constraint unit is used to extract scheduling constraint information from the power grid scheduling requirements; The parameter unit is used to solve the scheduling parameters of each of the power devices for each of the sub-scheduling requirements, with the scheduling constraint information as the constraint condition, using the alternating direction multiplier technique to obtain the device scheduling strategy. The solution uses Markov decision technology to quantify the scheduling process. The target unit is used to determine the target scheduling scheme of the target power grid based on the scheduling strategies of each of the devices. An execution unit is used to control the target power grid to execute the target scheduling scheme.

10. The power grid resource scheduling system as described in claim 6, characterized in that, Following the solution module, the following is included: The model unit is used to analyze the target scheduling scheme and the power grid scheduling requirements using deep learning technology, and to construct a scheduling model based on the analysis results. The scheduling model is used to generate a scheduling scheme at least.

Images