A source network load storage cluster power distribution method

Abstract

This invention relates to a power allocation method for a source-grid-load-storage cluster, comprising: acquiring power scheduling requirements and power scheduling preferences; determining adjustable resource nodes within the source-grid-load-storage cluster and the adjustment capacity range of each adjustable resource node; traversing the adjustable resource nodes, taking the power decision of each adjustable resource node as a dynamic programming decision stage, and solving for the optimal power decision of the adjustable resource node corresponding to each decision stage, forming an optimal power decision sequence that satisfies the power scheduling requirements for power allocation; wherein, the optimal power decision is within the stage feasible region and minimizes the stage cumulative total cost; the stage feasible region is formed by pruning the adjustment capacity range using state reachability conditions; the stage cumulative total cost is calculated based on the stage cost function, which is constructed according to the power scheduling preferences. This invention can effectively realize the power scheduling requirements of the source-grid-load-storage cluster and improve the real-time performance and reliability of scheduling.

Description

Technical Field

[0001] This invention relates to the field of power system automation and optimal dispatching technology, and in particular to a power allocation method for a power generation, grid, load and storage cluster. Background Technology

[0002] With the large-scale integration of distributed photovoltaic, wind power, energy storage, and controllable loads, the operation of distribution networks exhibits characteristics such as multi-source heterogeneity, strong uncertainty, and coupled complexity. Current power allocation in source-grid-load-storage clusters primarily employs techniques such as optimal power flow, model predictive control, or multi-timescale dynamic programming, which have the following inherent drawbacks:

[0003] Existing dynamic programming methods often use time scales as the basis for dividing stages or adopt a unified, all-resource coupled optimization model. This makes it impossible to verify in advance whether the decision will lead to the inability of subsequent resources to meet the overall scheduling requirements. It is easy to find invalid calculations that are not feasible until the end, resulting in high computing power overhead and difficulty in adapting to the time delay requirements of real-time scheduling of distribution networks.

[0004] For scenarios where a single node connects to multiple adjustable devices, either modeling each device individually leads to an explosion of optimization dimensions, or simple aggregation lacks standardized device instruction decomposition rules, which can easily result in scheduling instructions becoming unexecutable. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a power allocation method for a source-grid-load-storage cluster, which can effectively meet the power scheduling requirements of the source-grid-load-storage cluster and improve the real-time performance and reliability of scheduling.

[0006] The technical solution adopted by this invention to solve its technical problem is: to provide a power allocation method for a source-grid-load-storage cluster, comprising the following steps:

[0007] Obtain power scheduling requirements and power scheduling preferences;

[0008] Determine the adjustable resource nodes within the source-grid-load-storage cluster and the adjustment capacity range of each adjustable resource node;

[0009] The process iterates through the adjustable resource nodes, treating the power decision of each node as a dynamic programming decision stage. Following the traversal order, the optimal power decision for each adjustable resource node at each decision stage is solved, forming an optimal power decision sequence that satisfies power scheduling requirements. Then, a power command set is generated based on this optimal power decision sequence to control the adjustable resource nodes for power allocation.

[0010] The optimal power decision for each adjustable resource node is within the stage feasible region of the corresponding decision stage, and minimizes the stage cumulative total cost of the corresponding decision stage.

[0011] The stage feasible region is formed by trimming the adjustment capability range of the adjustable resource node corresponding to the current decision stage using the state reachability condition. The state reachability condition is used to ensure that the power decision of the adjustable resource node corresponding to the subsequent decision stage can meet the remaining power scheduling requirements of the current decision stage.

[0012] The total cumulative cost of a stage is calculated based on the stage cost function, which is constructed according to power scheduling preferences.

[0013] Furthermore, the optimal power decision for the adjustable resource node corresponding to each decision stage is solved separately, including the following steps:

[0014] Establish a relationship model between the minimum cumulative total cost and the remaining power scheduling demand in the current decision-making stage, and form a dynamic programming recursive equation for the current decision-making stage; where the minimum cumulative total cost is the sum of the decision execution cost in the current decision-making stage and the minimum stage cumulative total cost in the next decision-making stage, and the decision execution cost in the current decision-making stage is calculated based on the stage cost function of the current decision-making stage.

[0015] Solve the dynamic programming recursive equation for the current decision stage within the feasible region of the current stage to obtain the optimal power decision for the adjustable resource node corresponding to the current decision stage.

[0016] Furthermore, the minimum cumulative total cost of the next decision stage is calculated using dynamic programming recursive equations based on the remaining power scheduling requirements of the next decision stage; where,

[0017] The remaining power scheduling requirement for the next decision-making stage is calculated based on the remaining power scheduling requirement of the current decision-making stage and the power scheduling requirement covered by the decision output of the adjustable resource nodes corresponding to the current decision-making stage.

[0018] Furthermore, the dynamic programming recurrence relation is expressed as:

[0019]

[0020] in, Indicates the current decision-making stage The corresponding adjustable resource node's decision-making output, Indicates the current decision-making stage The remaining power scheduling requirements, Indicates the current decision-making stage The minimum cumulative total cost of the stage. Indicates the current decision-making stage The cost of implementing the decision, Indicates the next decision-making stage The minimum cumulative total cost of the stage. Indicates the current decision-making stage The overall output efficiency of the corresponding adjustable resource nodes. Indicates the current decision-making stage The corresponding feasible region for each stage.

[0021] Optionally, the stage cost function can be constructed based on power scheduling preferences, including the following steps:

[0022] If users have a flexible scheduling preference, then the power scheduling requirements are mapped to several optimization objectives and assigned corresponding weights to form a stage cost function;

[0023] If the user does not have a preference for flexible scheduling, a stage cost function is constructed that includes operating costs, power adjustment range, and equipment depreciation.

[0024] Furthermore, the power scheduling requirements are mapped to several optimization objective models, including:

[0025] The economic demand is mapped to a model that minimizes the cost of purchasing and selling electricity;

[0026] Map safety requirements to voltage deviation metrics and / or line loss metrics;

[0027] Map the demand for new energy utilization rate to the renewable energy consumption index;

[0028] Map reliability requirements to a risk function;

[0029] The carbon emission optimization requirements are mapped to the carbon emission index and / or pollutant emission index corresponding to the unit power.

[0030] Optionally, for an adjustable resource node with more than one access adjustable device, the node aggregated power is used as the power decision object for the adjustable resource node, and the optimal power decision of the adjustable resource node is transformed into power commands for each access adjustable device through the device decomposition operator.

[0031] Optionally, the following steps may also be included:

[0032] The generated power command set is subjected to consistency verification to verify whether the power scheduling requirements can be met when the power command set is used to control the adjustable devices of each access adjustable resource node.

[0033] If the consistency check passes, the set of power instructions will be executed.

[0034] Furthermore, it also includes the following steps:

[0035] Based on the execution results of the power command set, update the adjustment capability range of the adjustable resource node;

[0036] The optimal power decision sequence for the next scheduling cycle is generated based on the updated adjustment capability range.

[0037] Furthermore, updating the adjustable resource node's adjustment range includes the following steps:

[0038] If the adjustable devices connected to the adjustable resource node include energy storage units, then based on the current state of charge, temperature, health status, and safe operating range of the energy storage units, their maximum charging power and maximum discharging power are updated, thereby updating the corresponding adjustable resource node's adjustment capability range; and / or,

[0039] If the adjustable equipment connected to the adjustable resource node includes renewable energy equipment, then based on the irradiance prediction, wind speed prediction, operating status, and current available power output information of the renewable energy equipment, its maximum power generation capacity is updated, thereby updating the upper boundary of the corresponding adjustable resource node's adjustment capacity range; and / or,

[0040] If the adjustable devices connected to the adjustable resource node include controllable loads, then based on the load plan, user demand forecast, current operating status, and interruptibility of the controllable loads, their minimum and maximum load response capabilities are updated, thereby updating the adjustment capability range of the corresponding adjustable resource node; and / or,

[0041] If the adjustable devices connected to the adjustable resource node include inverter-type devices, then based on the current operating status, thermal limit information, operating boundaries and protection status given by the manufacturer, the maximum output power or maximum adjustable power range of the inverter-type devices is updated, and the adjustment capability range of the corresponding adjustable resource node is updated accordingly.

[0042] Beneficial effects

[0043] By adopting the above-mentioned technical solution, the present invention has the following advantages and positive effects compared with the prior art:

[0044] This invention uses an adjustable resource traversal order as the decision-making stage of dynamic programming and utilizes state reachability conditions to pre-cut the equipment's adjustment capacity range. This mechanism ensures that the remaining demand after each decision can be met by subsequent resources, completely avoiding invalid calculations, improving the real-time performance of scheduling, and ensuring that scheduling requirements are effectively implemented.

[0045] This invention flexibly configures the stage cost function according to scheduling preferences, which can adapt to different optimization objectives in different scenarios without modifying the underlying solution logic, thereby reducing the cost of multi-scenario adaptation.

[0046] This invention reduces the optimization dimension by first using the node aggregation power as the decision object when multiple devices are connected to a single node, and then uses the device decomposition operator to standardize the mapping into single device instructions, thereby controlling the computational complexity while ensuring the executability of device-level instructions.

[0047] This invention ensures that the scheduling results conform to the actual operating constraints by performing a consistency review before execution, and dynamically updates the equipment adjustment capacity range based on the operating feedback after execution, forming a complete scheduling closed loop, avoiding deviations between the model and actual operation, and improving the stability and reliability of long-term operation. Attached Figure Description

[0048] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0049] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0050] The embodiments of the present invention relate to a power allocation method for a source-grid-load-storage cluster, such as... Figure 1 As shown, it includes the following steps:

[0051] Determine power scheduling requirements and power scheduling preferences;

[0052] Obtain the adjustable resource nodes within the source-grid-load-storage cluster and the adjustment capability range of each adjustable resource node;

[0053] The process iterates through the adjustable resource nodes, treating the power decision of each node as a dynamic programming decision stage. Following the traversal order, the optimal power decision for each adjustable resource node at each decision stage is solved, forming an optimal power decision sequence that satisfies power scheduling requirements. Then, a power command set is generated based on this optimal power decision sequence to control the adjustable resource nodes for power allocation.

[0054] The optimal power decision for each adjustable resource node is within the stage feasible region of the corresponding decision stage, and minimizes the stage cumulative total cost of the corresponding decision stage.

[0055] The stage feasible region is formed by trimming the adjustment capability range of the adjustable resource node corresponding to the current decision stage using the state reachability condition. The state reachability condition is used to ensure that the power decision of the adjustable resource node corresponding to the subsequent decision stage can meet the remaining power scheduling requirements of the current decision stage.

[0056] The total cumulative cost of a stage is calculated based on the stage cost function, which is constructed according to power scheduling preferences.

[0057] Power scheduling requirements, power scheduling scenarios, and power scheduling preferences can be extracted by configuring a unified large model base, denoted as:

[0058]

[0059] It is used for semantic understanding, relation extraction, parameter suggestion generation, and interpretive output generation of multimodal power data, user natural language input, scheduling summary information, operation status summary information, and security domain summary information. The unified large model base does not directly replace power grid topology modeling, security domain construction, state machine switching, or dynamic programming solution. Instead, it serves as a global enhancement support module, which is called through different functional interfaces and provides gated structured output as needed.

[0060] By calling the unified large model base Group building interface By understanding multimodal power grid data and user natural language input regarding cluster construction, it can assist in generating candidate clusters, generating candidate resource ownership, and generating explanatory information for cluster construction. It also constructs the basic topology and resource ownership relationships of source-grid-load-storage clusters. Cluster construction is based on power grid structure, equipment connection relationships, and mutual support capabilities, and can perform cluster boundary adjustments or resource ownership corrections through user natural language interaction.

[0061] Using a unified large model base The system's semantic parsing capabilities enable automatic conversion from user natural language expressions, superior dispatch instructions, and grid operating status to optimization objective functions and cluster power demands. The large model performs semantic analysis on the user's input natural language information. For example, if expressions such as "improve renewable energy utilization," "limit energy storage cycle count," "reduce electricity costs," and "improve power supply reliability" are present, it assumes the user has flexible dispatch preferences and converts these preferences into corresponding optimization objective categories to guide power dispatch decisions.

[0062] The large model also determines the target power that the cluster needs to provide or absorb externally and / or the power transfer requirements between nodes within the cluster based on the superior scheduling instructions and the internal voltage control requirements of the cluster, forming the cluster power requirements. It then divides the scheduling requirements into three categories—cluster-level, region-level, and node-level—by calling the scope determination function, thus forming the corresponding scheduling scenarios. The scope determination function is used for:

[0063] Map external exchange or higher-level scheduling requirements to cluster-level requirement objects;

[0064] Local demand for facilitation, regional mutual assistance, and regional balance is mapped to regional-level demand objects.

[0065] The requirements for equipment life protection, node reference values, and local joint adjustment are mapped to node-level requirement objects.

[0066] The following describes the use of the unified large model base. This allows for the extraction of structured objects and the generation of optional implementation methods for power scheduling requirements, scheduling scenarios, and scheduling preferences.

[0067] Unified Large Model Base This can be achieved through a combination of "domain pre-training, supervised fine-tuning, agent scheduling, and scenario skill module collaboration." Specifically, a unified large model base is used to complete the semantic understanding of power dispatching, the agent selects the corresponding scenario skill module based on the current input, and the scenario skill module converts unstructured input and multi-source operational data into structured objects that can be called for dynamic programming solutions.

[0068] In the domain pre-training phase, towards a unified large model base Injecting multimodal power grid data, including power grid operation procedures, dispatch terminology, equipment operation boundaries, historical dispatch logs, power flow calculation results, load forecasts, photovoltaic forecasts, energy storage operation records, historical node voltage data, historical branch power flow data, fault handling records, and typical control cases, enables the model to learn the correspondence between power grid operation constraints, equipment relationships, dispatch language semantics, and typical control scenarios.

[0069] During the supervised fine-tuning phase, "operational data input, scheduling decisions, and control effects" are used as supervised sample pairs for training. Operational data input includes superior scheduling instructions, cluster gateway power targets, real-time power flow status, node voltage status, branch load status, energy storage charge status, adjustable resource boundaries, load forecasts, and new energy forecasts. Scheduling decisions include historically formed power scheduling demands, scheduling scenario labels, scheduling preference weights, and corresponding control strategies. Control effects include voltage recovery results, power flow limit mitigation results, network loss changes, new energy absorption changes, energy storage cycle count changes, and power supply reliability changes. Through this training process, the unified large model base can generate structured semantic results corresponding to power dispatching tasks based on the input data.

[0070] In the unified large model framework, agents can invoke corresponding scenario skill modules in parallel or sequentially based on the current input data. Scenario skill modules include power scheduling requirement extraction skills, scheduling scenario recognition skills, scheduling preference extraction skills, and scope determination skills. These different skill modules are not mutually exclusive but can be invoked in parallel based on the input content at the same time. For example, when both a superior scheduling instruction and a user's natural language preference exist simultaneously, the agent can simultaneously invoke the power scheduling requirement extraction skill and the scheduling preference extraction skill to form rigid power scheduling requirement objects and flexible scheduling preference weights, respectively.

[0071] The power dispatch demand extraction skill is used to generate power dispatch demand objects based on superior dispatch instructions, cluster gateway power targets, virtual power plant trading power, node voltage over-limit information, branch power flow over-limit information, regional load forecast results, photovoltaic and new energy forecast results, energy storage charge status, adjustable resource boundaries, and real-time power flow operation status. A power dispatch demand object can be represented as:

[0072]

[0073] in, Indicates the type of requirement. This indicates the target power adjustment or control target. Indicates the scope of application. This indicates priority. For example, when the input includes the cluster's external power exchange target, a cluster-level active power tracking requirement is generated; when the input includes information on heavy loads on lines in a certain area, a region-level power flow diversion requirement is generated; when the input includes information on node voltages below the lower limit, a node-level voltage support requirement is generated.

[0074] The unified large model base is mainly used to identify the semantic type, target, scope, and priority of power dispatching requirements. The specific target adjustment power value is calculated by the physical calculation module based on the real-time operating status of the power grid, prediction results, and safety domain boundaries. For regional mutual assistance or power flow diversion requirements, the physical calculation module determines the regional target support power based on the regional net load deficit, branch power flow exceeding limits, remaining capacity of interconnection channels, and remaining adjustable resource capacity within the region. For node support requirements, the physical calculation module determines the node-level target support power based on node voltage deviation, node reference power deviation, node safety domain boundaries, and active / reactive voltage sensitivity.

[0075] The scheduling scenario identification skill is used to generate scheduling scenario objects based on the current power grid topology, node voltage status, branch power flow status, safety domain construction results, set of over-limit nodes, set of over-limit branches, fault alarm information, superior scheduling objectives, and power scheduling demand objects. For example, when there is a cluster-wide output tracking objective, it is identified as a cluster-level scheduling scenario; when there is a local power flow over-limit and cross-regional mutual assistance demand, it is identified as a regional-level scheduling scenario; and when there is a critical node voltage support demand, it is identified as a node-level scheduling scenario.

[0076] The dispatch preference extraction skill is used to generate corresponding dispatch preference weights based on user natural language input, power grid enterprise operation evaluation indicators, economic operation rules, new energy consumption rules, energy storage life constraints, grid loss assessment indicators, power quality indicators, and power supply reliability indicators. The dispatch preference weights can be expressed as:

[0077]

[0078] in, Represents reliability weight, Indicates network loss weight. Indicates voltage quality weight. Indicates economic weight, Indicates the weight of new energy consumption. This indicates the weighting of energy storage lifespan. For example, when a user inputs "improve the utilization rate of new energy sources," the weighting of new energy consumption is increased; when a user inputs "reduce the number of energy storage cycles," the weighting of energy storage lifespan is increased; and when a user inputs "reduce operating costs," the weighting of economic efficiency is increased. The specific weighting can be set according to the actual indicators, rules, and constraints used, and is not limited to the above.

[0079] The scope determination skill is used to hierarchically classify power scheduling request objects based on their scope of influence, control objects, and physical areas of impact. Preferably, the scope determination function is called:

[0080]

[0081] Among them, external power exchange, cluster overall active power tracking, or virtual power plant overall output requirements are mapped to cluster-level requirements; local power flow diversion, regional mutual assistance, regional balancing, or feeder zone support requirements are mapped to regional-level requirements; and node voltage support, key node reference value tracking, equipment life protection, or local joint regulation requirements are mapped to node-level requirements.

[0082] Based on the AC power flow model of the distribution network, the following property can be achieved by constructing the feasible AC power interval within the node's convex zone as the adjustment capability interval of the corresponding adjustable resource node, and constructing the cluster adjustable power range formed by linear projection of this interval: regardless of how the node power is combined within the node interval, as long as the total power is within the cluster adjustable range, the corresponding AC power flow operating point satisfies voltage and current constraints, thus ensuring the system's safe operation by achieving consistency between the adjustable power interval at the node level and the adjustable power range at the cluster level. Specifically, a linearized power flow model, a polyhedral in-body approximation method, a robust sensitivity domain construction method, or an inner approximation set based on scenario sampling can be used. As long as a safe power interval satisfying AC constraints can be given at the node level, and the cluster adjustable power range can be obtained through linear projection, it should be considered an equivalent implementation of this embodiment.

[0083] By constructing a node-level AC convex inner security domain and linearly projecting it, the adjustable power range of the cluster layer is obtained. A monotonicity verification mechanism and power flow solvability constraints are embedded during the interval construction process, thereby establishing a consistent mapping relationship between the node security domain and the cluster adjustment range. Thus, the node convex inner security domain constitutes an inner approximation set of the feasible AC power flow domain. When a node power vector belongs to the security domain, the corresponding AC power flow solution satisfies the node voltage and branch current constraints and remains feasible without introducing additional coordination optimization steps.

[0084] Taking a linearized power flow model as an example, a power flow model of the grid branch, relating node voltage, branch current, and branch power, can be established. Convex relaxation of the branch current term in the model can then be applied to generate a linear function of the branch current operating boundary. For each adjustable resource node, based on the grid topology information, functional representations of the branch power boundary and node voltage operating boundary are generated using the linear function of the branch current operating boundary. This leads to the construction and solution of a linear programming problem aimed at maximizing and minimizing power injection at the current adjustable resource node. The regulation capacity range is determined based on the obtained maximum and minimum feasible power injections, forming the node power safety region.

[0085] Under the constraints of the established strict security domain, and based on the scope of the demand object, a dynamic programming subproblem that matches the current power scheduling scenario can be constructed, and the device-level power allocation result can be solved.

[0086] For each adjustable resource The node adjustable power boundary in strict security domain mode is given as its adjustment capability range:

[0087]

[0088] in, Indicates the first The active power output or equivalent active power adjustment of an adjustable resource in the current scheduling cycle; the boundary represents the adjustable power boundary of the corresponding node of the resource obtained under the conditions of the security domain construction, combined with the current physical boundary of the equipment and the current operating conditions.

[0089] Therefore, procedure-type constraints, safety domain boundaries, and state reachability conditions are all directly embedded as hard constraints into the feasible domain of dynamic programming, rather than being treated as flexible optimization objectives.

[0090] In some preferred embodiments, appropriate dynamic programming subproblems can be constructed based on the scope of the demand object. Specifically, the adjustment scenarios can be divided into the following three categories:

[0091] Cluster-level scenarios: These correspond to scenarios such as upper-level scheduling instructions, virtual power plant trading instructions, and overall cluster power exchange targets.

[0092] Regional-level scenarios: These correspond to scenarios such as local area mutual assistance, cross-regional power flow diversion, regional balancing, and formal support between platforms.

[0093] Node-level scenarios: These include scenarios such as supporting key nodes, tracking reference output of specified nodes, adjusting local node combinations, and reconfiguring local power of key nodes.

[0094] You can construct only one type of dynamic programming subproblem as the main implementation method based on the current scenario, or you can construct joint state variables for coordinated optimization of the two levels of requirements.

[0095] In this implementation, cluster-level, region-level, and node-level scenarios all refer to formal adjustment scenarios when the system is currently within a strict security domain. Specifically, cluster-level scenarios primarily correspond to the scheduling-tracking state and the scheduling-preference coordination state; they can also correspond to the autonomous optimization state when there is no external scheduling but an overall autonomous goal exists. Region-level scenarios primarily correspond to the scheduling-preference coordination state and the autonomous optimization state; they can also correspond to the scheduling-tracking state with region constraints. Node-level scenarios primarily correspond to the scheduling-preference coordination state and the autonomous optimization state; they can also correspond to the scheduling-tracking state with node constraints.

[0096] Let the set of adjustable resources within the group be:

[0097]

[0098] The first The active power output of each adjustable resource is denoted as the stage decision variable:

[0099]

[0100] Then the first The basic feasible region of a stage is defined as follows:

[0101]

[0102] To ensure that subsequent stages can still complete the control tasks under the current scenario after the decision is made in the current stage, it is also necessary to introduce state reachability conditions induced by state variables to form the feasible region of the stage:

[0103]

[0104] in, This represents the state variable corresponding to the current scenario; This represents the set of reachable states that, given the current state variables, guarantee the completion of the control task in subsequent stages.

[0105] Let the current decision-making stage be... The current remaining power scheduling requirement is The power decision for the adjustable resource node at the current stage is as follows: The feasible decision domain for each stage is The stage cost function is The contribution parameter of power decision to scheduling demand is: The remaining power scheduling requirement for the next decision-making stage is:

[0106]

[0107] The dynamic programming recurrence relation is expressed as:

[0108]

[0109] in, Indicates from the first The decision-making phase begins with the remaining power scheduling demand being... The minimum cumulative total cost of completing subsequent scheduling tasks under the given conditions; Indicates the power decision to be executed in the current decision-making stage. The resulting costs; This represents the decision-making output of the adjustable resource node corresponding to the current decision-making stage. This represents the overall output efficiency of the node. Multiplying the two together yields the power scheduling requirements covered by the node's decision-making output. This represents the minimum cumulative total cost of the next decision stage for the updated remaining power scheduling demand after the current power decision is executed; that is, the remaining power scheduling demand for the next decision stage. In the state, from the first The minimum cumulative total cost from the start of the phase to the end of the phase; This represents the stage feasible region that satisfies equipment capacity constraints, operational safety constraints, and power regulation constraints under the current remaining power scheduling demand state.

[0110] For the current decision-making stage, the dynamic programming recurrence equation is solved within the feasible region of the stage to obtain the optimal power decision for the corresponding adjustable resource node. The optimal power decision for the adjustable resource node at each decision stage is solved sequentially according to the traversal order, forming an optimal power decision sequence that satisfies power scheduling requirements.

[0111] When multiple adjustable devices are aggregated and connected to the same node, in order to control the scale of the dynamic programming problem, it is preferable to use the aggregated power of the node as the decision object of the optimization layer. Let the node be... The set of devices is as follows:

[0112]

[0113] Then node The polymerization power is defined as:

[0114]

[0115] Find the optimal power command for node aggregation Then, the operator is decomposed by the device. Decompose it into device-level instructions:

[0116]

[0117] in, This represents the decomposition parameters of devices within a node, including device availability boundary, priority, response speed, lifetime cost coefficient, and device type parameters. The decomposition result satisfies:

[0118]

[0119] And satisfy the boundary constraints of each device:

[0120]

[0121] Based on the dynamic programming recurrence equation corresponding to the current scenario, perform inverse recurrence to obtain the optimal decision at each stage:

[0122]

[0123] If node aggregation is used for solving, then the operators are further decomposed through devices. Generate device-level optimal power allocation results:

[0124]

[0125] in, This indicates the total number of actual devices in the system.

[0126] The final output is:

[0127]

[0128] in:

[0129] This is either the optimal sequence for stage decision-making or the optimal sequence for node aggregation power.

[0130] Assign a device-level power vector;

[0131] This serves as the current scenario label, used to identify whether the current problem belongs to a cluster-level, region-level, or node-level dynamic programming sub-problem.

[0132] These are the valid template parameters after gating.

[0133] Therefore, under strict security domain constraints, dynamic programming subproblems at the cluster, regional, or node levels are constructed based on the selected formal scenario scope, and the device-level power allocation results are obtained by solving them. Furthermore, when the final demand object is a cluster-level demand object, a cluster-level dynamic programming subproblem is constructed; when the final demand object is a regional-level demand object, a regional-level dynamic programming subproblem is constructed; and when the final demand object is a node-level demand object, a node-level dynamic programming subproblem is constructed. The cluster-level scenario is mainly applicable to cluster-to-external power exchange commands issued by the upper-level dispatcher, execution of virtual power plant transaction power, and cluster-wide active power tracking tasks; the regional-level scenario is mainly applicable to local regional mutual assistance, cross-regional power flow guidance, regional balancing, and formal support between distribution stations; and the node-level scenario is mainly applicable to critical node support, reference output tracking of designated nodes, node combination adjustment, and local power reconfiguration of critical nodes.

[0134] The following sections will provide further explanations for cluster-level, region-level, and node-level scenarios.

[0135] When the final requirement object is a cluster-level requirement object, construct a cluster-level dynamic programming subproblem. This scenario is mainly applicable to:

[0136] The cluster's external power exchange command issued by the superior dispatcher;

[0137] Execution of virtual power plant transaction power;

[0138] Cluster-wide active tracking task.

[0139] First, the cluster-level power requirements are trimmed down to the cluster safety domain, and the trimmed cluster-level requirements are treated as hard constraints that must be met. The cluster safety domain is determined based on the node safety domains. The sum of the upper and lower bounds of the safety domains of all schedulable nodes in the cluster is calculated to obtain the total adjustable range of cluster power, which is then used as the cluster safety domain.

[0140] Define cluster-level state variables as follows:

[0141]

[0142] It indicates the first At the start of the phase, subsequent resources still need to meet the cluster-level total power requirements. The initial state is:

[0143]

[0144] in, This represents the total cluster-level power requirement output from step three. The state recursion relationship is as follows:

[0145]

[0146] in, Indicates the first The contribution coefficient of each resource to the total power demand of the cluster.

[0147] Define the minimum / maximum residual contribution that subsequent resources can achieve in a cluster-level scenario as follows:

[0148]

[0149]

[0150] The set of state reachability conditions under cluster-level state variables is defined as follows:

[0151]

[0152] Therefore, the first The feasible domain of the stage is:

[0153]

[0154] Since the cluster-level requirements are already a hard constraint, the stage cost function is only used to select a more engineering-optimal allocation scheme from the set of feasible solutions that meet the total power requirements of the cluster level.

[0155] When users have flexible preferences, the first can be defined. The cost function for each stage is:

[0156]

[0157] in , which is the flexible target weight after gating. , , , , These are the economic objective function, the new energy utilization rate objective function, the renewable energy utilization rate objective function, the safety objective function, and the carbon emission optimization objective function.

[0158] Specifically, a large model can be used to perform semantic analysis on the natural language information input by the user, such as expressions like "improve the utilization rate of new energy sources," "limit the number of energy storage cycles," "reduce electricity costs," and "improve power supply reliability," and convert them into corresponding optimization objective categories. The system predefines a set of basic objective functions such as economic efficiency, safety and stability, renewable energy utilization rate, power supply reliability, and carbon emission optimization. The large model identifies the category of the user's intent through semantic vector matching and generates corresponding flexible objective weight vectors.

[0159]

[0160] Used to characterize the relative importance of targets.

[0161] After completing the target type identification, the large model maps the user intent into a mathematically oriented objective function:

[0162] 1) Economic objectives

[0163] To achieve the economic operation objective, a model is constructed that minimizes the costs of electricity purchase and sale as well as the operating costs of controllable power sources:

[0164]

[0165] in:

[0166] and Time periods The electricity prices at which the cluster purchases electricity from the main grid and sells electricity to the main grid;

[0167] and Time periods Cluster power purchase and power sales;

[0168] This is the operating cost function of the controllable power supply.

[0169] In scenarios where energy storage participates in electricity price arbitrage, the energy storage arbitrage profit target can be further constructed:

[0170]

[0171] in:

[0172] and These are the energy storage discharge power and the charging power, respectively.

[0173] 2) Voltage and network loss targets

[0174] For voltage quality objectives, the objective function is constructed using node voltage deviation:

[0175]

[0176] in:

[0177] For nodes During the period The per-unit voltage value;

[0178] This is the node reference voltage.

[0179] For the network loss optimization objective, the objective function is constructed using the total active power loss of the distribution network:

[0180]

[0181] in:

[0182] For time period The sum of active power losses of all branches within the cluster.

[0183] 3) New energy consumption target

[0184] For the renewable energy consumption target, the form of minimizing the curtailment rate is adopted:

[0185]

[0186] in:

[0187] To maximize the available power output of new energy sources;

[0188] This represents the actual power absorbed.

[0189] 4) Reliability Objectives

[0190] For the power supply reliability objective, a power supply risk function is used:

[0191]

[0192] in:

[0193] For time period The quantitative value of power supply risk.

[0194] Preferably, the power supply risk quantification value can be based on:

[0195] Lack of power supply;

[0196] Risk of loss of supply to critical loads;

[0197] The degree of inadequacy in security domain margin;

[0198] Probability of failure to recover from fault;

[0199] At least one of them is determined.

[0200] 5) Carbon emission targets

[0201] For the carbon emission optimization target, the following approach is adopted:

[0202]

[0203] This indicates the system's carbon emissions.

[0204] in:

[0205] Describes the amount of carbon emissions or pollutant emissions per unit of power. This is a carbon emission function constructed based on various power source outputs and carbon emission factors.

[0206] Preferably, when the system does not detect a power scheduling preference explicitly input by the user, the unified large model base automatically generates a default scheduling preference object based on the power grid operation procedures, enterprise operation evaluation indicators, historical scheduling records and current operating status, and uses the default scheduling preference object as the target construction basis for subsequent optimization solutions.

[0207] The default scheduling preference object can be represented as:

[0208]

[0209] in, Indicates the power supply reliability weight. Indicates voltage quality weight. Indicates the weight for network loss optimization. Indicates the weight of economic operation. Indicates the weight of new energy consumption. This represents the weighting of energy storage lifetime protection. Among them, evaluation indicators such as power supply reliability, voltage quality, comprehensive network loss rate, economic operation, renewable energy absorption capacity, and energy storage lifetime are preferably characterized by the definitions, calculation methods, and evaluation criteria in power industry standards, dispatching operation procedures, or the current evaluation system of power grid enterprises. The focus of this invention is to use a unified large model base and dynamic programming to achieve dynamic coordination and adaptive optimization among the above objectives, without limiting the specific definition of each evaluation indicator.

[0210] Furthermore, the default scheduling preference object is adaptively adjusted according to the current operating scenario. When the system is under heavy load, in fault recovery, or with insufficient safety margin, the weight of power supply reliability and voltage quality is increased; when the system is under high renewable energy output or reverse power flow risk, the weight of renewable energy absorption is increased; when the energy storage state of charge is low, the cycle count is high, or the equipment health status is declining, the weight of energy storage life protection is increased; when the system is in normal operation and there is no safety risk, the weight of economic operation and network loss optimization is increased.

[0211] Therefore, even when users do not explicitly specify their scheduling preferences, the system can still generate executable autonomous optimization objectives based on power grid operation procedures and operation evaluation indicators, thereby ensuring the continuity and reproducibility of the power scheduling decision-making process.

[0212] The recurrence relation for cluster-level dynamic programming can be written as:

[0213]

[0214] The termination condition is defined as:

[0215]

[0216] This requires that the total power demand at the cluster level be strictly met during the final stage. When the final demand object is a region-level demand object, a region-level dynamic programming subproblem is constructed. This scenario is mainly applicable to:

[0217] Localized mutual assistance;

[0218] Cross-regional flow diversion;

[0219] Regional balance;

[0220] Official support has been provided between Taiwan and mainland China.

[0221] First, the regional power demand is clipped to a regional safety domain, and this clipped regional demand is treated as a hard constraint that must be fulfilled. The regional safety domain is determined based on the node safety domains. The sum of the upper and lower bounds of the safety domains of all schedulable nodes within the region is calculated to obtain the total adjustable range of regional power, which serves as the regional safety domain.

[0222] Let the size of the selected region set be... Define the region-level state variables as follows:

[0223]

[0224] It indicates the first At the start of the phase, each region still needs to meet its regional-level power requirements. The initial state is as follows:

[0225]

[0226] in, This represents the regional demand vector. The state recursion is as follows:

[0227]

[0228] in, Indicates the first The contribution vector of each resource to the state variables of each region.

[0229] Define the reachable set of subsequent resources in a region-level scenario as follows:

[0230]

[0231] Then the set of state reachability conditions under the region-level state variables is:

[0232]

[0233] Therefore, the first The feasible region of a stage is defined as follows:

[0234]

[0235] Consistent with cluster-level scenarios, regional requirements serve as hard constraints, and the stage cost function is only used to select a more reasonable allocation scheme from an engineering perspective, provided that the regional state terminal conditions are met.

[0236] When the user provides valid flexible preferences, a stage cost function is constructed based on the flexible objective function and its weights; when the user does not provide valid flexible preferences, the default secondary objective function is automatically activated.

[0237]

[0238] in, Indicates the first Stage power decision The corresponding running cost function, Indicates the first Stage power decision The corresponding reference output (such as current output or planned output). Indicates the first Stage power decision The corresponding equipment depreciation (lifespan loss) function, , , These represent the corresponding weighting coefficients.

[0239] The recurrence relation for regional dynamic programming can be written as:

[0240]

[0241] The termination condition is defined as:

[0242]

[0243] This means that the selected regional requirements must be strictly fulfilled during the termination phase.

[0244] When the final requirement object is a node-level requirement object, construct a node-level dynamic programming subproblem. This scenario is mainly applicable to:

[0245] Key node support;

[0246] Specify node reference output tracking;

[0247] Node combination adjustment;

[0248] Local power reconfiguration at critical nodes.

[0249] In this implementation, node-level requirement objects are used as effective control targets in the current scenario, and selected node-level requirements are treated as hard constraints that must be completed.

[0250] Let the size of the selected node set be... Define the node-level state variables as follows:

[0251]

[0252] It indicates the first At the start of the phase, each node still needs to meet its node-level power requirements. The initial state is:

[0253]

[0254] in, This represents the node-level demand vector. The state recursion relationship is as follows:

[0255]

[0256] in, Indicates the first The contribution vector of each resource to the state variables of each node.

[0257] Node-level state variables can be formed by selecting, aggregating, projecting, equipping, or mapping node-level requirement objects to control the state dimension of dynamic programming.

[0258] Define the reachable set of subsequent resources in a node-level scenario as follows:

[0259]

[0260] Then the set of state reachability conditions under the node-level state variables is:

[0261]

[0262] Therefore, the first The feasible region of a stage is defined as follows:

[0263]

[0264] Node-level requirements are a hard constraint in the current scenario, and the stage cost function is only used to select a better device allocation scheme under the premise of satisfying the node-level state terminal conditions.

[0265] When the user provides a valid flexible preference, the stage cost function is constructed using the flexible objective function and its weights; when the user does not provide a valid flexible preference, the system automatically uses the default secondary objective function.

[0266]

[0267] in, Indicates the first Stage power decision The corresponding running cost function, Indicates the first Stage power decision The corresponding reference output (such as current output or planned output). Indicates the first Stage power decision The corresponding equipment depreciation (lifespan loss) function, , , These represent the corresponding weighting coefficients.

[0268] In node-level scenarios, the default secondary objective can also be understood as: minimizing the range of equipment movements, reducing adjustment costs, or reducing equipment wear, while satisfying the node reference tracking task or the node combination adjustment task.

[0269] The node-level dynamic programming recurrence equation can be written as:

[0270]

[0271] The termination condition is defined as:

[0272]

[0273] This means that the selected node-level requirements must be strictly completed during the termination phase.

[0274] Based on the device-level power allocation results, a consistency review can be performed to improve the robustness of the project implementation.

[0275] Specifically, it can receive device-level power allocation vectors:

[0276]

[0277] And generate a set of executable control instructions:

[0278]

[0279] in, This indicates the output control command for the corresponding equipment.

[0280] In this implementation, since the decision variables are always constrained by the node security domain boundary, the theoretically obtained results should meet the operational requirements of the strict security domain mode. However, in engineering applications, considering factors such as measurement errors, parameter drift, approximation errors, and equipment dynamic behavior, it is preferable to perform a consistency check on the equipment-level power allocation results.

[0281] The consistency review includes one or more of the following:

[0282] (1) Node voltage verification:

[0283]

[0284] in, Represents a node In power allocation vector Voltage amplitude under, , Representing nodes respectively The lower and upper limits for safe operation of voltage amplitude.

[0285] (2) Branch current verification:

[0286] in, and Representing branch roads In power allocation vector The current amplitude and its upper limit.

[0287] (3) Equipment output boundary verification:

[0288] (4) Requirements Consistency Verification: Verify the degree to which formal requirements are met at the cluster, region, and node levels.

[0289] When the consistency verification passes, the output will be:

[0290]

[0291] in, For device-level power allocation vectors, For a set of control instructions, This is the output mode label.

[0292] Preferably, when the consistency verification fails, the system does not directly issue control commands, but instead sends the verification result back to the previous steps:

[0293] If the verification results show that the current solution deviates from the strict security domain but still has a strict recovery path, then the system flows back to the strict security domain recovery mode. This mode has no security constraint relaxation and must return to the security domain completely.

[0294] If the verification results indicate that there is no strict recovery path for the current solution, the system reverts to the minimum default recovery mode, which allows for limited relaxation of low-priority security constraints.

[0295] Therefore, this step, as an optional engineering enhancement, can improve the stability and feasibility of the adjustment results in actual operation.

[0296] After the power allocation strategy is issued and executed, the system updates the available adjustment capability boundary of each device for the next scheduling cycle based on real-time measurement data, device operating status, environmental conditions and execution results, and forms the state input and boundary input required for the strict safety domain reconstruction and optimization solution of the next cycle accordingly.

[0297] Preferably, the update process includes the collection and processing of at least one or more of the following information: equipment state of charge, temperature, health status, inverter limit, predicted renewable energy available power, load plan, operating mode status, protection status, and manual interlock status. Based on the above information, the system updates the upper and lower limits of the adjustable power of each device in the next cycle.

[0298] For energy storage units, based on their current state of charge, temperature, health status, and safe operating range, their maximum charging power and maximum discharging power for the next cycle are updated, and corresponding device power boundaries are formed:

[0299]

[0300] For renewable energy equipment such as photovoltaic and wind power, the maximum power generation capacity for the next cycle is updated based on irradiance prediction, wind speed prediction, operating status and current available power output information, and this is used as the upper bound of the corresponding equipment power boundary.

[0301] For controllable loads, update the minimum and maximum load response capabilities for the next cycle based on load plans, user demand forecasts, current operating status, and interruptibility.

[0302] For inverter-type equipment, based on its current operating status, thermal limit information, operating boundaries and protection status given by the manufacturer, update the maximum allowable output power or maximum adjustable power range for the next cycle.

[0303] This forms the set of available power boundaries for each device in the next power scheduling cycle:

[0304]

[0305] Boundary set It is not directly equivalent to the strict node security domain of the next cycle, but serves as the device boundary input condition for reconstructing the node security domain and cluster security domain in the next power scheduling cycle, and as the basic input for constructing the dynamic programming feasible domain in the next power scheduling cycle.

[0306] Simultaneously, the system compresses the execution results of this cycle, the current measurement status, and the updated device capability boundaries to form the set of state variables for the next cycle:

[0307]

[0308] in, Represents the set of state variables for the current period. This indicates the control instructions that have been executed in the current cycle. This indicates the measurement results and environmental input for the current period. This indicates a state update mapping.

[0309] Through the above update mechanism, at the end of each scheduling cycle, the system compresses historical execution results, device state changes, and environmental changes into the current state and boundary inputs for the next scheduling cycle, so that the optimization solution for the next scheduling cycle depends only on the updated set of state variables. and boundary set This allows for the fulfillment of the aftereffects requirement of dynamic programming without having to revisit all historical information from earlier periods.

[0310] The so-called "no aftereffect" property means that the optimization solution for the next scheduling cycle depends only on the set of state variables updated at the end of the current cycle and the boundary inputs, and does not depend on the complete trajectory of earlier historical decision processes; the influence of earlier history has already been mapped through state updates. It is compressed into the current state.

[0311] Thus, this step realizes a closed-loop connection mechanism of "execution result - state update - boundary reconstruction - solution in the next cycle", providing a consistent, up-to-date and computable input basis for the construction of the strict safety domain and the optimization of standard dynamic programming in the next scheduling cycle.

[0312] The following example, using three adjustable resource nodes, illustrates how to map multiple adjustable resource nodes sequentially to multiple decision stages in a target power demand solution scenario. Based on the safety domain and state reachability conditions corresponding to the target scope, the adjustment capability range of each stage is trimmed, and then the optimal power decision sequence that satisfies the target power demand is recursively solved.

[0313] The target scope can be cluster-level, region-level, or node-level. In other words, this embodiment can be used for cluster-level total power demand decomposition, region-level power demand decomposition, and node-level power demand decomposition. The difference between different target scopes lies in the meaning of the target power demand and the contribution coefficient of each resource node to that target power demand, but the dynamic programming modeling and recursive solution process are the same.

[0314] 1. Target scope setting

[0315] Let the scope of the currently selected target be:

[0316]

[0317] in, It can represent cluster-level, region-level, or node-level target objects.

[0318] Let the target power requirement for output be:

[0319]

[0320] Among them, when When it is at the cluster level, Indicates the total power requirement at the cluster level; when When it is at the regional level, Indicates regional power demand; when When at the node level, This indicates the power support requirements of a specified node or equivalent node.

[0321] The system contains 3 adjustable resource nodes, which correspond to 3 dynamic programming decision stages according to a preset scheduling decomposition order. The corresponding decision variables for each stage are denoted as follows:

[0322]

[0323] in, These represent the power adjustment amounts undertaken by the three resource nodes during the current scheduling cycle.

[0324] 2. Relationship between contribution coefficient and target power balance

[0325] Assume three resource nodes are associated with the target scope. The contribution coefficients are as follows:

[0326]

[0327] in, Indicates the first Adjustment amount of each resource node Target power requirements The equivalent contribution relationship.

[0328] For ease of explanation, the following is taken in this embodiment:

[0329]

[0330] Therefore, the target power balance relationship is:

[0331]

[0332] In practical engineering applications, if the target scope is at the cluster level, then This can represent the contribution coefficient of a resource node to the total power of the cluster; if the target scope is at the regional level, then... This can represent the contribution coefficient of resource nodes to regional power deficit or regional switching power; if the target scope is at the node level, then... It can represent the contribution coefficient of a resource node to the power support of a specified node or the equivalent node demand.

[0333] 3. Target security domain pruning

[0334] Let the safe adjustment intervals of the three resource nodes in the current scheduling period be as follows:

[0335]

[0336]

[0337]

[0338] Right now:

[0339]

[0340]

[0341]

[0342] Based on the target scope The resource contribution coefficient is used to calculate the target power safety region as follows:

[0343]

[0344] Substituting the parameters into this embodiment, we have:

[0345]

[0346]

[0347] Therefore, the target power safety domain is:

[0348]

[0349] because:

[0350]

[0351] Therefore, the target power requirement in this embodiment does not need to be further reduced, and the reduced target power requirement remains as follows:

[0352]

[0353] The pruned target power requirement serves as a hard constraint in the dynamic programming solution process.

[0354] 4. Stage Cost Function

[0355] Since the target power requirement has been treated as a hard constraint, the stage cost function is not used to determine whether the target requirement is met. Instead, it selects the more engineering-optimized power allocation scheme from the set of feasible solutions that meet the target requirement.

[0356] In this embodiment, the following stage cost function is constructed:

[0357]

[0358]

[0359]

[0360] Therefore, the original optimization problem can be expressed as:

[0361]

[0362] Among them, constraints This means that the target power requirement after the trimming must be strictly met.

[0363] 5. Dynamic Programming Modeling

[0364] Define the target scope The state variables are:

[0365]

[0366] It indicates the first At the start of the phase, subsequent resources still need to meet the target power requirements.

[0367] The initial state is:

[0368]

[0369] The state recursion relationship is as follows:

[0370]

[0371] In this embodiment, due to Therefore:

[0372]

[0373] Specifically:

[0374]

[0375]

[0376]

[0377] The termination phase requires that the target power demand be strictly met, therefore:

[0378]

[0379] set up Indicates from the first Phase begins, with the remaining target power requirement being The minimum cumulative cost that can be achieved at any time is given by the dynamic programming recurrence relation as follows:

[0380]

[0381] in:

[0382]

[0383] The termination condition is:

[0384]

[0385] 6. State reachability conditions and stage feasible domain pruning

[0386] Definition of the first Subsequent resources after the phase are in the target scope The minimum and maximum residual contributions that can be realized are as follows:

[0387]

[0388]

[0389] Then the first The set of state reachability conditions for a stage is:

[0390]

[0391] Therefore, the first The actual feasible region of the stage is:

[0392]

[0393] In other words, the decision at the current stage It is necessary not only to satisfy the security adjustment range of the resource node itself, but also to ensure that the remaining target demand formed after the execution of the decision can still be completed by subsequent resources.

[0394] 7. Reverse reasoning solution process

[0395] 7.1 Solving in the third stage

[0396] There are no further resources after Phase 3, therefore:

[0397]

[0398] The conditions for reaching the third stage state are:

[0399]

[0400] Right now:

[0401]

[0402] Meanwhile, the security adjustment range for resource nodes in Phase 3 is as follows:

[0403]

[0404] Therefore, only when:

[0405]

[0406] At that time, Phase 3 is feasible.

[0407] This leads to the optimal value function for the third stage:

[0408]

[0409] The corresponding optimal decision is:

[0410]

[0411] 7.2 Solving in the second stage

[0412] After Phase 2, only Phase 3 resources remain, therefore:

[0413]

[0414]

[0415] The conditions for reaching the second stage state are:

[0416]

[0417] Meanwhile, the security adjustment range for resource nodes in Phase 2 is as follows:

[0418]

[0419] therefore:

[0420]

[0421] The recurrence relation for the second stage is:

[0422]

[0423] Substituting into the optimal value function for stage 3, we get:

[0424]

[0425] make:

[0426]

[0427] right Differentiate:

[0428]

[0429] Right now:

[0430]

[0431] Setting the first derivative to zero, we obtain the unconstrained extreme points:

[0432]

[0433] Combination Based on the state reachability conditions, the optimal decision function for the second stage can be obtained as follows:

[0434]

[0435] The corresponding optimal value function is:

[0436]

[0437] 7.3 Solution in Stage 1

[0438] After Phase 1, there are remaining resources for Phase 2 and Phase 3, therefore:

[0439]

[0440] The conditions for reaching the state in stage 1 are:

[0441]

[0442] because:

[0443]

[0444] Therefore:

[0445]

[0446] Meanwhile, the security adjustment range of the resource nodes in Phase 1 is as follows:

[0447]

[0448] Therefore, the practical feasible region for Phase 1 is:

[0449]

[0450] The recurrence relation for the first stage is:

[0451]

[0452] Depend on:

[0453]

[0454] and:

[0455]

[0456] It can be known that:

[0457]

[0458] Therefore, only the first stage needs to consider middle The part.

[0459] when:

[0460]

[0461] Right now:

[0462]

[0463] At that time, there were:

[0464]

[0465] Substitution ,get:

[0466]

[0467] This function is Within range As it increases, it decreases; therefore, the optimal boundary value of this interval corresponds to:

[0468]

[0469] The target value is:

[0470]

[0471] when:

[0472]

[0473] Right now:

[0474]

[0475] At that time, there were:

[0476]

[0477] Substitution ,get:

[0478]

[0479] Summarized as follows:

[0480]

[0481] This function is Within range It increases and decreases, therefore we take:

[0482]

[0483] The corresponding target value is:

[0484]

[0485] Comparing the target values ​​under the two state intervals above, the target value in the second case is smaller. Therefore, the optimal decision for the first stage is:

[0486]

[0487] 8. Reverse optimization strategy

[0488] From the optimal decision in stage 1, we can obtain:

[0489]

[0490] Therefore, the remaining target power requirement at the start of Phase 2 is:

[0491]

[0492] because:

[0493]

[0494] According to the optimal decision function for stage 2, we have:

[0495]

[0496] Therefore, the remaining target power requirement at the start of Phase 3 is:

[0497]

[0498] Phase 3 is directly determined by the termination conditions:

[0499]

[0500] Therefore, the optimal power decision sequence is obtained as follows:

[0501] 9. Result Verification

[0502] Substituting the optimal solution into the hard constraint of target power balance, we have:

[0503]

[0504] Meet the target power requirements after trimming:

[0505]

[0506] At the same time, all resource nodes meet the node safety adjustment range:

[0507]

[0508]

[0509] Substituting the optimal solution into the stage cost function, we have:

[0510]

[0511] Right now:

[0512]

[0513]

[0514] Therefore, the minimum cumulative adjustment cost obtained in this embodiment is:

[0515] Although this embodiment uses scalar requirements This explanation does not limit the target scope to the cluster level. In practical applications, the target scope... The required objects can be determined based on the output of step three:

[0516]

[0517] When the target scope is cluster-level, state variables This indicates the total cluster-level power requirements that subsequent resources still need to meet; when the target scope is at the region level, the state variable... This indicates the regional power requirements that subsequent resources still need to meet; when the target scope is at the node level, the state variables... This indicates that subsequent resources still need to meet the node-level power support requirements.

[0518] Under different target scopes, only the target requirement needs to be replaced. Resource contribution coefficient The dynamic programming state recursion, state reachability conditions, stage feasible domain pruning, and recursive solution process are kept consistent with the corresponding safety domain boundaries and the corresponding safety domain boundaries.

Claims

1. A power allocation method for a source-grid-load-storage cluster, characterized in that, Includes the following steps: Obtain power scheduling requirements and power scheduling preferences; Determine the adjustable resource nodes within the source-grid-load-storage cluster and the adjustment capacity range of each adjustable resource node; The process iterates through the adjustable resource nodes, treating the power decision of each node as a dynamic programming decision stage. Following the traversal order, the optimal power decision for each adjustable resource node at each decision stage is solved, forming an optimal power decision sequence that satisfies power scheduling requirements. Then, a power command set is generated based on this optimal power decision sequence to control the adjustable resource nodes for power allocation. The optimal power decision for each adjustable resource node is within the stage feasible region of the corresponding decision stage, and minimizes the stage cumulative total cost of the corresponding decision stage. The stage feasible region is formed by trimming the adjustment capability range of the adjustable resource node corresponding to the current decision stage using the state reachability condition. The state reachability condition is used to ensure that the power decision of the adjustable resource node corresponding to the subsequent decision stage can meet the remaining power scheduling requirements of the current decision stage. The total cumulative cost of a stage is calculated based on the stage cost function, which is constructed according to power scheduling preferences.

2. The power distribution method according to claim 1, characterized in that, Solve for the optimal power decision of the adjustable resource node for each decision stage, including the following steps: Establish a relationship model between the minimum cumulative total cost and the remaining power scheduling demand in the current decision-making stage, and form a dynamic programming recursive equation for the current decision-making stage; where the minimum cumulative total cost is the sum of the decision execution cost in the current decision-making stage and the minimum stage cumulative total cost in the next decision-making stage, and the decision execution cost in the current decision-making stage is calculated based on the stage cost function of the current decision-making stage. Solve the dynamic programming recursive equation for the current decision stage within the feasible region of the current stage to obtain the optimal power decision for the adjustable resource node corresponding to the current decision stage.

3. The power distribution method according to claim 2, characterized in that, The minimum cumulative total cost of the next decision stage is calculated using dynamic programming recursive equations based on the remaining power scheduling requirements of the next decision stage; where, The remaining power scheduling requirement for the next decision-making stage is calculated based on the remaining power scheduling requirement of the current decision-making stage and the power scheduling requirement covered by the decision output of the adjustable resource nodes corresponding to the current decision-making stage.

4. The power distribution method according to claim 2, characterized in that, The dynamic programming recurrence equation is expressed as: ， in, Indicates the current decision-making stage The corresponding adjustable resource node's decision-making output, Indicates the current decision-making stage The remaining power scheduling requirements, Indicates the current decision-making stage The minimum cumulative total cost of the stage. Indicates the current decision-making stage The cost of implementing the decision, Indicates the next decision-making stage The minimum cumulative total cost of the stage. Indicates the current decision-making stage The overall output efficiency of the corresponding adjustable resource nodes. Indicates the current decision-making stage The corresponding feasible region for each stage.

5. The power distribution method according to claim 1, characterized in that, Constructing the stage cost function based on power scheduling preferences includes the following steps: If users have a flexible scheduling preference, then the power scheduling requirements are mapped to several optimization objectives and assigned corresponding weights to form a stage cost function; If the user does not have a preference for flexible scheduling, a stage cost function is constructed that includes operating costs, power adjustment range, and equipment depreciation.

6. The power distribution method according to claim 5, characterized in that, Power scheduling requirements are mapped to several optimization objective models, including: The economic demand is mapped to a model that minimizes the cost of purchasing and selling electricity; Map safety requirements to voltage deviation metrics and / or line loss metrics; Map the demand for new energy utilization rate to the renewable energy consumption index; Map reliability requirements to a risk function; The carbon emission optimization requirements are mapped to the carbon emission index and / or pollutant emission index corresponding to the unit power.

7. The power distribution method according to claim 1, characterized in that, For an adjustable resource node with more than one connected adjustable device, the node aggregated power is used as the power decision object for that adjustable resource node, and the optimal power decision of the adjustable resource node is transformed into power commands for each connected adjustable device through the device decomposition operator.

8. The power distribution method according to claim 1, characterized in that, It also includes the following steps: The generated power command set is subjected to consistency verification to verify whether the power scheduling requirements can be met when the power command set is used to control the adjustable devices of each access adjustable resource node. If the consistency check passes, the set of power instructions will be executed.

9. The power distribution method according to claim 8, characterized in that, It also includes the following steps: Based on the execution results of the power command set, update the adjustment capability range of the adjustable resource node; The optimal power decision sequence for the next scheduling cycle is generated based on the updated adjustment capability range.

10. The power distribution method according to claim 9, characterized in that, Updating the adjustable capacity range of adjustable resource nodes includes the following steps: If the adjustable devices connected to the adjustable resource node include energy storage units, then based on the current state of charge, temperature, health status, and safe operating range of the energy storage units, their maximum charging power and maximum discharging power are updated, thereby updating the corresponding adjustable resource node's adjustment capability range; and / or, If the adjustable equipment connected to the adjustable resource node includes renewable energy equipment, then based on the irradiance prediction, wind speed prediction, operating status, and current available power output information of the renewable energy equipment, its maximum power generation capacity is updated, thereby updating the upper boundary of the corresponding adjustable resource node's adjustment capacity range; and / or, If the adjustable devices connected to the adjustable resource node include controllable loads, then based on the load plan, user demand forecast, current operating status, and interruptibility of the controllable loads, their minimum and maximum load response capabilities are updated, thereby updating the adjustment capability range of the corresponding adjustable resource node; and / or, If the adjustable devices connected to the adjustable resource node include inverter-type devices, then based on the current operating status, thermal limit information, operating boundaries and protection status given by the manufacturer, the maximum output power or maximum adjustable power range of the inverter-type devices is updated, and the adjustment capability range of the corresponding adjustable resource node is updated accordingly.

Images