Reconstruction of optimal strategy competition relationship of power system and method for tracing decision link

Abstract

The present application relates to the optimal strategy competition relationship reconstruction and decision link tracing method of power system, through the generation of the optimal regulation strategy of the current stage by using dynamic programming, and based on the stage cost function and state transition function of dynamic programming, the local value function of the stage is constructed, the key competitive strategy affecting the stability of the optimal regulation strategy is extracted according to its characteristics, and the competition relationship reconstruction is realized.On this basis, the stage comprehensive regulation value difference function between the optimal strategy and the key competitive strategy is constructed, and the strategy switching boundary is solved; according to the recursive order of dynamic programming, the stage comprehensive regulation value difference function and the strategy switching boundary are connected in series, and the whole link decision tracing is realized.The present application reconstructs the implicit competition relationship between the optimal strategy and the competitive strategy into the explicit tracing link containing the value difference function and the switching boundary without changing the dynamic programming solver, so that the regulation decision process is transparent, interpretable and traceable.

Description

Technical Field

[0001] This invention relates to the field of dynamic programming technology for power systems, and in particular to a method for reconstructing the competitive relationship of optimal strategies and tracing the decision-making link in power systems based on dynamic programming. Background Technology

[0002] With the large-scale integration of distributed power sources, energy storage, flexible loads, electric vehicles, and other adjustable resources into power distribution systems, the objects to be regulated exhibit characteristics of multi-type, multi-level, multi-constraint, and strong dynamic coupling. To address these issues, dynamic programming, due to its ability to handle stage recursion relationships, state transition relationships, and multi-stage optimal decision-making problems, has been widely used in scenarios such as power distribution system scheduling, energy management, coordinated operation of power generation, grid, load, and storage, and multi-stage control optimization.

[0003] However, existing dynamic programming methods in engineering applications typically only output the optimal control action for a certain stage, making it difficult to explain the reasons for the formation of this optimal solution, its competitive advantage over other candidate solutions, and the conditions under which the optimal solution switches as the operating state changes. In actual power distribution system control, the feasible region of a stage, boundary constraints, resource allocation structure, and state transition relationships may all change with the operating state and external conditions. If only a single optimal control result is output, without identifying key competing strategies that pose a threat to the current optimal control strategy, without quantifying the stage value gap between the optimal strategy and competing strategies and its evolution with state changes, then the stability range of the optimal strategy cannot be determined, and it is difficult to support strategy interpretation, switching warnings, and operator verification in engineering scenarios. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a method for reconstructing the competitive relationship of optimal strategies and tracing the decision link in a power system based on dynamic programming. Without changing the main structure of the dynamic programming solver, the single optimal solution is extended into a structured result that includes key competitive strategies, stage-based integrated control value difference function, switching boundary and stage-based tracing link, thereby improving the interpretability, traceability and engineering usability of the dynamic programming control results.

[0005] The technical solution adopted by this invention to solve its technical problem is: to provide a method for reconstructing the competitive relationship of optimal strategies in a power system based on dynamic programming, comprising the following steps:

[0006] The optimal control strategy for the current stage is solved using a dynamic programming decision model. The dynamic programming decision model is based on the stage cost function and the state transition function, and it recursively solves the optimal control strategy within the feasible region of each decision stage.

[0007] A stage local value function is constructed; the stage local value function is constructed based on the stage cost function and the state transition function, and is used to calculate the stage comprehensive control value of the control strategy. The stage comprehensive control value includes the sum of the direct value of the current stage and the total value transmitted in subsequent stages.

[0008] Candidate strategy points are determined based on the extreme value characteristics of the local value function of the stage and the boundary characteristics of the feasible region of the stage.

[0009] Candidate strategy points that affect the stability of the optimal control strategy are extracted as key competitive strategies in the current stage.

[0010] Furthermore, it also includes the following steps:

[0011] The collected multi-source heterogeneous information from the power system is converted into structured data;

[0012] Decision-driving factors are extracted from structured data and mapped to corresponding dynamic programming decision model components to form the structured input of the dynamic programming decision model; wherein, the dynamic programming decision model includes a stage cost function, a stage feasible region, and a state transition function.

[0013] The generated structured input is fed into the dynamic programming decision model.

[0014] Furthermore, mapping the extracted decision-driving factors to the corresponding dynamic programming decision model components includes the following steps:

[0015] Map the target configuration decision-driving factors to the target components of the stage cost function;

[0016] Map the constraint boundary decision-making drivers to the boundary components of the stage feasible region;

[0017] Map the state transition decision driving factors to the state transition function's state transition components.

[0018] Furthermore, the candidate strategy points include stage feasible region boundary points, stage feasible region boundary intersection points, stage feasible region boundary representative points, stationary points of local value functions within the current stage feasible region, and extreme points indicated by derivative features.

[0019] Furthermore, before the step of extracting candidate strategy points that affect the stability of the optimal control strategy as key competitive strategies in the current stage, a step of reducing the candidate strategy points is also included, specifically including:

[0020] When the local value function of a stage satisfies the boundary optimality condition, candidate strategy points corresponding to the boundary endpoints and / or intersections are preferentially retained.

[0021] When a stationary point exists in the feasible domain of a stage and may constitute the required extreme value, the candidate policy point corresponding to the stationary point is retained.

[0022] When multiple candidate strategy points correspond to the same or similar resource allocation structures, or the same or similar boundary dominance patterns, the representative candidate strategy point is retained.

[0023] Furthermore, the extraction of candidate strategy points that affect the stability of the optimal control strategy as key competitive strategies in the current stage includes the following steps:

[0024] When there is only one candidate strategy point, that unique candidate strategy point shall be taken as the key competitive strategy.

[0025] When there is not a unique candidate strategy point, the first candidate strategy point that satisfies the structural proximity condition with the current stage's optimal control strategy and the second candidate strategy point that satisfies the equivalent proximity condition are extracted as key competitive strategies. The structural proximity condition includes that there is only a single-dimensional difference with the current stage's optimal control strategy, and the equivalent proximity condition includes that the difference in the stage's comprehensive control value with the current stage's optimal control strategy is less than a set value difference threshold. The stage comprehensive control value difference is determined by the difference in value obtained by substituting the current stage's optimal control strategy and the key competitive strategy into the stage's local value function.

[0026] Furthermore, the step of extracting candidate strategy points that satisfy the set proximity conditions with the current optimal control strategy as key competitive strategies also includes the following steps:

[0027] When the number of candidate strategy points exceeds the set threshold for the number of candidate strategies, the third candidate strategy point that meets the boundary proximity condition is extracted as the key competitive strategy. The boundary proximity condition includes the distance to the set boundary being less than the first distance threshold. The set boundary includes the current stage feasible domain boundary, the constraint activation boundary, or the strategy switching boundary. The strategy switching boundary is determined based on the strategy point when the stage comprehensive control value difference is equal to zero.

[0028] Furthermore, the step of extracting candidate strategy points that satisfy the set proximity conditions with the current optimal control strategy as key competitive strategies also includes the following steps:

[0029] When there are multiple candidate strategy points, and the distance between the current optimal control strategy and the constraint activation boundary is less than the second distance threshold, candidate strategy points that satisfy the boundary proximity condition are extracted first as key competitive strategies.

[0030] Furthermore, the single-dimensional differences include different proportions of single resource responsibilities, different start / stop states of single resources, different depths of single resource calls, different activation states of single constraint boundaries, or different allocation shares of single resource groups or node groups.

[0031] Furthermore, when the number of extracted key competitive strategies exceeds a set threshold, the method also includes a step of reducing the extracted key competitive strategies, specifically including:

[0032] The key competitive strategies are sorted according to the set sorting rules;

[0033] Based on the ranking results, retain the top-ranked key competitive strategies by a set number, and remove the remaining key competitive strategies.

[0034] The sorting rules include at least one of the following rules:

[0035] The key competitive strategies are ranked according to the difference in comprehensive regulatory value between the key competitive strategies and the optimal regulatory strategy at the current stage. The smaller the difference in comprehensive regulatory value between the key competitive strategies, the higher the priority of the ranking.

[0036] The key competitive strategies are ranked based on their proximity to the current optimal control strategy in terms of resource allocation structure or control action structure. The closer the key competitive strategies are, the higher their ranking priority.

[0037] The ranking is based on the location of key competitive strategies, with higher priority given to key competitive strategies that are closer to a set boundary.

[0038] This invention also provides a dynamic programming-based power system decision-making link tracing method, comprising the following steps:

[0039] The optimal control strategy for the current stage is solved using a dynamic programming decision model; the dynamic programming decision model includes a stage cost function, a stage feasible region, and a state transition function.

[0040] Based on the stage cost function and the state transition function, a stage local value function is constructed; the stage local value function is used to calculate the stage comprehensive control value of the control strategy, and the stage comprehensive control value includes the sum of the direct value of the current stage and the total value transmitted in subsequent stages;

[0041] Candidate strategy points are determined based on the extreme value characteristics of the local value function of the stage and the boundary characteristics of the feasible region of the stage.

[0042] Extract candidate strategy points that affect the stability of the optimal control strategy as key competitive strategies in the current stage;

[0043] Calculate the difference between the local value function of the current stage key competitive strategy and the optimal control strategy, establish the stage comprehensive control value difference function, and solve the strategy switching boundary of the current stage based on the stage comprehensive control value difference function;

[0044] Following the recursive order of the dynamic programming decision-making model, the stage-specific comprehensive regulation value difference function and strategy switching boundary corresponding to each stage are linked together to form a decision tracing link;

[0045] Based on the current operational status, the decision-making basis for each stage is traced back along the decision tracing link to obtain the decision tracing result.

[0046] Furthermore, the method of solving the current stage strategy switching boundary based on the stage comprehensive regulation value difference function includes the following steps:

[0047] Based on the phase-based comprehensive regulation value difference function, the phase-based comprehensive regulation value difference between the current phase's optimal regulation strategy and each key competitive strategy is calculated respectively.

[0048] The strategy point at which the comprehensive control value difference in the solution stage equals zero is used to form the strategy switching boundary for the current stage.

[0049] Furthermore, the decision-making chain tracing results include the optimal control strategy, key competitive strategy, stage comprehensive control value difference function, direct value of the current stage, total value transmitted in subsequent stages, and strategy switching boundary for each stage.

[0050] Furthermore, the decision link tracing result also includes the local strategy region of the current operating state relative to each key competitive strategy, wherein the local strategy region includes the strategy region when the difference between the current stage's optimal control strategy and the corresponding key competitive strategy is greater than zero.

[0051] Furthermore, the decision-making link tracing result also includes the overall dominant region of the current operating state relative to all key competitive strategies, wherein the overall dominant region includes the strategy region where the difference between the current stage's optimal control strategy and the stage's comprehensive control value between each key competitive strategy is greater than zero.

[0052] Furthermore, the dynamic programming decision model uses structured input, and the processing method for structured input includes the following steps:

[0053] The collected multi-source heterogeneous information from the power system is converted into structured data;

[0054] Decision-driving factors are extracted from structured data and mapped to the corresponding dynamic programming decision-making model components to form the structured input of the dynamic programming decision-making model.

[0055] Furthermore, mapping the extracted decision-driving factors to the corresponding dynamic programming decision model components includes the following steps:

[0056] Map the target configuration decision-driving factors to the target components of the stage cost function;

[0057] Map the constraint boundary decision-making drivers to the boundary components of the stage feasible region;

[0058] Map the state transition decision driving factors to the state transition function's state transition components.

[0059] Furthermore, it also includes the following steps:

[0060] Bind a corresponding anchor point identifier to each structured data item;

[0061] When interpreting the decision-making basis in stages, the original data information is obtained by reverse positioning through anchor point markers.

[0062] Beneficial effects

[0063] By adopting the above-described technical solution, the present invention has the following advantages and positive effects compared with the prior art:

[0064] After the dynamic programming solver outputs the optimal control strategy for the current stage, this invention reconstructs the competitive relationship between the optimal strategy and the key competitive strategy through an external method, without changing the internal structure of the dynamic programming solver, and is easy to integrate with existing dynamic programming control systems.

[0065] This invention generates a set of candidate strategy points and extracts key competitive strategies using structural proximity and equivalent proximity conditions. It replaces the exhaustive comparison of all feasible strategies with a finite number of representative competitive strategies, thereby controlling the computational scale of subsequent value difference analysis and boundary switching solution while ensuring that the competitive objects have engineering representativeness and analytical value.

[0066] This invention constructs a stage comprehensive value difference function by taking the comprehensive value difference between the optimal strategy and the key competitive strategy as a stage comprehensive value difference function, and decomposes the stage comprehensive value difference function into two parts: the benefit difference of the current stage and the value difference of the subsequent stages. In this way, the qualitative question of "why the current strategy is optimal and why not choose the alternative" is transformed into a calculable and comparable structured expression, so that the source of the advantage of the optimal strategy can be presented item by item.

[0067] This invention determines the switching boundary between the current optimal strategy and the key competitive strategy by solving the critical condition that the phase comprehensive regulation value difference function is equal to zero. This switching boundary explicitly marks the stable range of the current optimal strategy and the critical condition for the reversal of the strategy advantage, providing early warning basis for the scheduler to switch strategies.

[0068] This invention expands the original single optimal solution output of dynamic programming into a structured competitive evidence chain that includes the optimal strategy, key competitive strategies, stage comprehensive regulation value difference function, strategy switching boundary, and strategy region. This makes the formation logic of regulation decision and competitive landscape explicit, which is convenient for post-audit and review verification.

[0069] This invention connects the competitive relationship analysis results of each stage into a decision tracing link by recursively following the dynamic programming sequence. Combined with anchor point identification, it realizes the backtracking from the decision node to the original information source. This tracing link reveals that the current optimal strategy depends not only on the direct benefits of this stage, but also on the impact of state transitions on the cumulative value of subsequent stages, forming a complete causal explanation of staged recursion. Attached Figure Description

[0070] Figure 1 This is a flowchart illustrating the first embodiment of the present invention;

[0071] Figure 2 This is a flowchart illustrating the second embodiment of the present invention. Detailed Implementation

[0072] 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.

[0073] To avoid inconsistencies in dynamic programming-related symbols throughout the manual, the following conventions are adopted for the main symbols:

[0074] The stage numbering is uniformly adopted This indicates that the final stage number is uniformly recorded as follows: ;

[0075] stage The initial state variables are uniformly denoted as The next stage state after the implementation of stage-based decision-making is uniformly denoted as... ;

[0076] stage The decision variables are uniformly denoted as The optimal control strategy obtained through dynamic programming at the current stage is denoted as... ;

[0077] To describe the competitive relationship between the optimal control strategy and other possible strategies, the first... The key competitive strategies are collectively denoted as: The feasible region at the current stage is uniformly denoted as The local value function of a stage is uniformly denoted as ; used to characterize the regulatory strategy employed in stage k. The corresponding stage of comprehensive regulation value.

[0078] Current optimal control strategy Compared to the first Key competitive strategies The phase-based comprehensive regulation value difference function is uniformly denoted as: ;in,

[0079] The differences in the current stage objectives are uniformly denoted as The value difference in subsequent stages is uniformly recorded as Both conditions are met: ;

[0080] The current switching boundary between the current optimal control strategy and the key competitive strategy is denoted as . ;

[0081] The set of candidate strategy points is uniformly denoted as The set of key competitive strategies is denoted as... ;

[0082] parameter , , A unified parameter representing the comprehensive regulatory benefits of resource groups at different stages;

[0083] parameter , , A unified representation of the contribution coefficient of different resource groups to the unit equivalent adjustment demand;

[0084] parameter The unified expression represents the total equivalent adjustment demand.

[0085] The first embodiment of the present invention relates to a method for reconstructing the competitive relationship of optimal strategies in a power system based on dynamic programming, comprising the following steps:

[0086] The optimal control strategy for the current stage is solved using a dynamic programming decision model.

[0087] Extract the key competitive strategies at the current stage that affect the stability of the optimal control strategy, and reconstruct the competitive relationship between the key competitive strategies at the current stage and the optimal control strategy at the current stage.

[0088] The dynamic programming decision model is built upon the dynamic programming algorithm. It is used to recursively solve for the optimal control strategy within the feasible region of each decision stage, based on the stage cost function and the state transition function.

[0089] The stage cost function is used to characterize the cost of implementing the control strategy in the current stage, the stage feasible region is used to limit the feasible range of the control strategy in the current stage, and the state transition function is used to characterize the state transition relationship after the decision is made in the current stage and the next stage is entered.

[0090] In some preferred embodiments, the optimal control strategy for each current stage not only lies within the stage-feasible region of the corresponding decision stage, but also minimizes the cumulative total cost of the corresponding decision stage. The cumulative total cost of the stage can be calculated based on a stage cost function, which can be constructed according to scheduling preferences.

[0091] To minimize the cumulative total cost of the corresponding decision-making stage during the solution process, a relationship model between the minimum cumulative total cost of the current decision-making stage and the remaining power scheduling demand can be established. This forms the dynamic programming recursive equation for the current decision-making stage, deriving the optimal control strategy for the current stage. The minimum cumulative total cost is the sum of the decision execution cost of the current decision-making stage and the minimum cumulative total cost of the next decision-making stage. The decision execution cost of the current decision-making stage can be calculated based on the stage cost function of the current decision-making stage.

[0092] In some preferred embodiments, to ensure the determinism, verifiability, and traceability of subsequent candidate strategy point generation and key competitive strategy extraction processes, the input data structure can be standardized and traceable anchor points can be constructed. For example... Figure 1 As shown, the specific steps include:

[0093] Step 1: Collect multi-source information related to control and decision-making, and perform structural standardization processing on the multi-source information to construct a traceable anchor point system;

[0094] Step 2: Extract the driving factors that enter the dynamic programming solution process from the structurally standardized input. The driving factors include at least the target configuration driving factors, constraint boundary driving factors, and state transition influence driving factors, and map them to the stage cost function, stage feasible region, and state transition function of the dynamic programming, respectively.

[0095] Step 3: Based on the current stage's optimal control strategy obtained through dynamic programming, a set of candidate strategy points is formed based on the current stage state, the current stage feasible region, and the stage local value function. Then, focusing on the current optimal control strategy, a set of key competitive strategies is extracted from the candidate strategy point set to reconstruct the competitive relationship between the current stage's key competitive strategies and the current stage's optimal control strategy.

[0096] Step 1: Information collection, structural standardization, and traceable anchor point construction: By collecting multi-source information related to the power distribution system's control decisions, heterogeneous inputs are processed in a unified structure, and a traceable anchor point system for dynamic programming control processes is constructed. This provides a consistent and traceable input foundation for subsequent solutions such as input driving factor extraction, key competitive strategy extraction, construction of stage-based comprehensive control value difference function, and current switching boundary solution.

[0097] In this embodiment, the control-related information collected by the system includes at least the following five categories: First, operating status information, including node voltage, branch current, active power, reactive power, energy storage state of charge, temperature, switch status, interconnection status, capacity boundary, and over-limit alarm status; Second, equipment parameter information, including equipment type, rated capacity, power upper and lower limits, efficiency parameters, dynamic characteristic parameters, start-stop constraint parameters, safe operating boundaries, and control mode parameters; Third, network topology information, including node connection relationships, line connection relationships, transformer connection relationships, switch and interconnection relationships, transformer area and feeder association relationships, branch parameters, and impedance parameters; Fourth, constraint boundary information, including node voltage boundaries, line current boundaries, equipment power boundaries, energy storage SOC boundaries, temperature boundaries, operating mode boundaries, safety boundaries, and scheduling rule boundaries; Fifth, natural language control demand information, including descriptions of control objectives, operating preferences, scenario instructions, and constraint preferences input by dispatchers, users, or external control systems.

[0098] The system performs structural standardization processing on the information collected above. For structured data, the system directly performs field mapping, field completion, unit standardization, timestamp standardization, and format standardization; for unstructured text or chart data, the system extracts parameters related to regulation and converts them into a structured parameter set; for network connection relationship data, the system constructs a description of the relationship between resources, nodes, lines, and constraints, thereby forming a structured input set that can be used for subsequent dynamic planning regulation and link tracing.

[0099] Preferably, the structured input set formed after structural standardization includes at least: a resource state parameter set, a device parameter set, a network topology parameter set, a constraint boundary parameter set, a state transition impact parameter set, and a regulation requirement parameter set.

[0100] After completing the above structural standardization process, the system further constructs a traceable anchor point system. The traceable anchor point system includes at least: object anchor points, used to establish unique identifiers for resource objects, node objects, line objects, constraint objects, output objects, and feedback objects; control scenario anchor points, used to establish unique identifiers for control scenarios, scheduling rounds, stage numbers, time slices, and operating modes; and event anchor points, used to establish unique identifiers for constraint trigger events, boundary change events, state transition events, optimal solution switching events, output generation events, and feedback update events.

[0101] The system binds various data items in the structured input set with object anchors, control context anchors, and event anchors to form a unified anchor binding relationship. Preferably, at least the following binding relationships are completed: device parameters are bound to resource object anchors, network boundaries are bound to node or line object anchors, operating status is bound to control context anchors, feedback information is bound to event anchors, and natural language requirement parsing results are bound to object anchors and control context anchors.

[0102] In some preferred embodiments, a large model parsing module can be set up to perform semantic parsing and parameter extraction on natural language processing requirements and unstructured device data, thereby improving input processing efficiency and scenario adaptability. It should be noted that the large model parsing module in this step is only a limited-use parsing auxiliary module; its output does not directly enter the subsequent solution process. Instead, it must undergo anchor binding verification, field integrity verification, parameter range verification, and solution compatibility verification before it can be used in subsequent steps.

[0103] After the above processing, the following data is output: standardized structured input set, object anchor set, regulation context anchor set, event anchor set, and anchor binding relationship set, which serve as the input basis for the next step (structured extraction and mapping of dynamic programming driving factors).

[0104] Step 2, Structured Extraction and Mapping of Dynamic Programming Driving Factors: This step extracts driving factors from the standardized structured input set obtained in the previous step and uses a restricted call configuration and hierarchical mapping mechanism to map the driving factors to the stage cost function, stage feasible region, and state transition function of the dynamic programming. This forms the unified solution input basis required for subsequent key competitive strategy extraction, construction of stage comprehensive control value difference function, and current switching boundary solution.

[0105] In this embodiment, the driving factors entering the dynamic programming solution process include at least the following three categories: target configuration driving factors. Constraint boundary driving factors and state transition influence driving factors Among them, target configuration source factors are used to construct the cost function in the construction phase, constraint boundary source factors are used to construct the feasible region in the construction phase, and state transition influence source factors are used to construct the state transition function. For candidate source factors generated by large models or assisted generation, they must undergo anchor point binding verification, parameter range verification, and dynamic programming compatibility verification before they can be included in the formal source factor set. The three types of driving factors do not exist merely as explanatory labels, but are structured and extracted and verified as formal input sources for subsequent dynamic programming solution components.

[0106] (1) The target configuration parameters used to characterize the construction of the stage cost function include at least the target item identifier, target weight parameters, target activation flag, and scene correction coefficient. Preferably, the target configuration driving factors are mapped to the stage cost function:

[0107]

[0108] in, Indicates the target item identifier. Indicates the first Sub-objective functions corresponding to each objective item Indicates the target activation flag. This represents the scene correction factor. This represents the target weight parameter.

[0109] In this invention, semantic understanding is not a necessary step in target configuration, but rather an optional preprocessing method when the upstream target is not given in the form of structured parameters; what ultimately enters the dynamic programming solution process are always the formal target configuration parameters consistent with the stage cost function.

[0110] (2) The constraint boundary parameters used to characterize the construction of the feasible region in each stage include at least the constraint term identifier, the object to which the constraint applies, the constraint boundary value, the constraint direction indicator, and the stage activation indicator. Preferably, the constraint boundary driving factors are mapped to the stage feasible region:

[0111]

[0112] in, Representation phase feasible domain, This represents the constraint function constructed from the corresponding constraint boundary driving factors.

[0113] In some preferred embodiments, candidate constraint boundaries from specification texts, operating rule texts, or equipment description texts can also be extracted with the assistance of the large model parsing module. However, candidate boundary conditions must not be directly included in the formal feasible domain construction process. Instead, they must first be converted into constraint item identifiers, constraint boundary values, stage activation flags, and applicable objects. Only after anchor point binding and solution compatibility verification can they be included in the formal constraint boundary driving factor set. .

[0114] (3) Parameters used to characterize state transitions in the impact phase include at least the impact item identifier, the impact object identifier, the impact parameter value, the phase activation flag, and the correction method identifier. Preferably, the state transition impact driving factors are mapped to a state transition function:

[0115]

[0116] in, Representation phase Initial state variables Represents the decision variables for each stage. This indicates the state of the next stage.

[0117] In some preferred embodiments, candidate parameters in the state transition impact driving factors can be extracted by the large model analysis module in conjunction with equipment specifications, efficiency curve data, operating condition data, or operational feedback data. However, the output of this extraction should not be directly included in the formal solution process. Instead, it must undergo anchor point binding and solution compatibility verification before being incorporated into the formal set of state transition impact driving factors. .

[0118] (4) For candidate driving factors generated by a large model or assisted generation, this step uses a restricted call configuration. Let the candidate results be... Only those factors that satisfy anchor point binding validation, parameter range validation, and dynamic programming compatibility validation are allowed to enter the formal driver factor set. The admission criteria are:

[0119]

[0120] Otherwise, the candidate results are only kept as candidate records and are not included in the subsequent dynamic programming formal solution process.

[0121] Through the above processing, the information related to objectives, constraints, and state transitions is uniformly converted into structured inputs that can be called by dynamic programming. This limits the role of the large model to the generation of candidate driving factors and structured assistance, rather than directly participating in the recursive solution of dynamic programming. This ensures that the subsequent processes of generating candidate strategy points and extracting key competitive strategies are deterministic, verifiable, and traceable.

[0122] This step does not simply classify the inputs, nor does it interpret the objective, constraints, and efficiency as independent links. Instead, it transforms the upstream factors affecting the dynamic programming solution into three categories of driving factors, restricted call configuration, and hierarchical mapping mechanism: the objective component of the cost function of the entry stage, the boundary component of the feasible domain of the entry stage, and the state transition component of the state transition function of the entry stage.

[0123] Therefore, this step provides a unified, structured, and verifiable input layer for subsequent key competitive strategy extraction.

[0124] Step 3, Key Competitive Strategy Extraction: Based on the current optimal control strategy obtained through dynamic programming, a set of candidate strategy points is first generated through local value function analysis. Then, focusing on the current optimal control strategy, key competitive strategies that can form an actual competitive relationship with the current optimal control strategy are extracted from the candidate strategy point set. This provides comparison objects for the construction of the comprehensive control value difference function, the solution of the current switching boundary, and the interpretation of the current link in subsequent stages. It mainly includes the following steps:

[0125] Based on the stage cost function and the state transition function, a local value function is constructed. The local value function is used to calculate the stage-wide comprehensive control value of the control strategy. The stage-wide comprehensive control value includes the sum of the direct value of the current stage and the total value transmitted in subsequent stages.

[0126] Candidate strategy points are determined based on the extreme value characteristics of the local value function and the boundary characteristics of the stage feasible region;

[0127] Candidate strategy points that satisfy the set proximity conditions with the current optimal control strategy are extracted as key competitive strategies.

[0128] It should be noted that this step does not re-perform global optimization, nor does it exhaustively enumerate all allowed control strategies. Instead, it processes "candidate point generation" and "competitor extraction" in two levels: the first level addresses "which points might become competitors in the current stage," and the second level addresses "which competitors are worth further comparison with the current optimal control strategy." Therefore, the subsequent value difference analysis and boundary analysis in this invention are based on finite local competitive relationships, rather than on arbitrary comparisons in an infinite strategy space.

[0129] Step 3 (1), Candidate strategy point generation: used in the known phase Current state The current feasible domain and local value functions of the stage Under these conditions, generate a set of candidate strategy points that may become the optimal control strategy or key competitive strategy at the current stage. The candidate policy point set is not required to cover all feasible solutions in the current stage, but rather to extract a finite number of representative decision points that may influence the optimal policy choice under the local value function structure in the current stage.

[0130] A. Determination of local value functions at each stage

[0131] stage Let the current stage state be The stage decision variables are The local value function in the definition phase is:

[0132]

[0133] in, Indicates the current state. Given the established conditions, if regulatory decisions are adopted at the current stage... Then, the comprehensive regulatory value of this regulatory decision in the recursive sense of dynamic programming is determined. It should be noted that the local value function of the stage... It does not simply represent the immediate benefits or costs generated by the current stage's objectives, but rather includes both of the following:

[0134] 1. Current stage of regulatory decisions The value directly generated in this stage;

[0135] 2. Current stage of regulatory decisions The effect of state transition on the next stage state on the optimal cumulative value of subsequent stages, i.e., the total value passed on to subsequent stages.

[0136] Therefore, the local value function of the stage This can be understood as: making a decision on a specific candidate regulation at the current stage. A comprehensive evaluation function that combines the direct effects of this stage with the effects passed on to subsequent stages. Dynamic programming in stages Choosing the optimal strategy essentially involves comparing different strategies within the feasible region at the current stage. The magnitude of the value determines which regulatory decision is better in the Bellman recursive sense.

[0137] It should be further noted that the local value function of the stage defined in this invention... This differs fundamentally from the basic recurrence equations in dynamic programming. In solving dynamic programming problems, the basic recurrence equations describe the state from the current stage. Starting from this point, and under the condition of satisfying the stage-specific feasibility constraints, optimize the decision-making process for the current stage. The recursive relationship for obtaining the optimal cumulative value result in future stages is usually expressed as follows:

[0138]

[0139] in, Indicates the state from the stage The optimal cumulative value that can be obtained by starting. This indicates the stage value generated by the decisions made at the current stage. This represents the optimal cumulative value in subsequent stages after a state transition.

[0140] The local value function of the stage in this invention In the current stage state Given a predetermined condition and without performing an optimal search operation, a specific control decision is made. The constructed value evaluation function is used to characterize the comprehensive value of this specific regulatory decision in the current stage and its subsequent state transition process:

[0141]

[0142] Therefore, the following relationship exists between the two:

[0143]

[0144] That is, the dynamic programming recursive equation obtains the optimal cumulative value of the current stage by optimizing the local value function corresponding to different control decisions; while the local value function of the stage retains the specific control decision. Information is used to compare the value differences between different control strategies. The dynamic programming recurrence equation solves the problem of: "In the current state, which of all feasible control strategies can obtain the optimal cumulative value?", while the stage local value function solves the problem of: "What kind of comprehensive value can a specific control strategy generate in the current state, and what value difference does this strategy have relative to other competing strategies?"

[0145] This invention utilizes a stage-specific local value function to retain value information corresponding to different control strategies, and further constructs a stage-specific comprehensive control value difference function between the current optimal control strategy and key competitive strategies, thereby enabling the tracing of the causes of the optimal dynamic programming result and the conditions for strategy switching. In this invention, subsequent steps need to revolve around the current optimal control strategy. Key competitive strategies The competitive relationship between them, and the construction stage of the comprehensive regulation value difference function

[0146]

[0147] Furthermore, the current switching boundary is solved, therefore this invention uses the local value function of the subsequent comparable stage. This serves as the basis for analysis, rather than simply focusing on the stage-optimal results given by the basic recursive equations of dynamic programming.

[0148] B. Determining the feasible region at the current stage

[0149] Let the feasible region at the current stage be In the case of one-dimensional decision-making, the feasible region can be represented as:

[0150]

[0151] in, and Each represents a stage Centralized regulation of decision variables The lower and upper bounds.

[0152] In a more general implementation, the current stage feasible domain It can also be represented as a set of feasible solutions jointly defined by multiple boundary conditions. Boundary conditions include, but are not limited to, the upper and lower limits of equipment output, the upper and lower limits of energy storage state of charge, node voltage boundaries, line current boundaries, transformer area regulation capacity boundaries, feeder regulation capacity boundaries, and other safe operation constraint boundaries.

[0153] C. Hierarchical Formation Mechanism of Candidate Strategy Point Set: Based on the optimal control strategy already determined at the current stage, and combining the structural characteristics of the local value function and the boundary characteristics of the feasible region at the current stage, a finite set of candidate strategy points that may become the current optimal strategy or key competitive strategy is formed. .

[0154] It should be noted that the set of candidate strategy points It is not directly composed of all feasible solutions at the current stage, but is gradually generated through a hierarchical formation mechanism to balance the completeness of extreme value candidates with the finiteness of subsequent competitive comparisons. Preferably, the hierarchical formation mechanism includes at least the formation of a theoretical candidate layer and the formation of a representative reduction layer.

[0155] (1) Theoretical candidate layer formation: used to form the boundary endpoints, boundary intersections, boundary representative points, feasible stationary points, and representative extreme points indicated by derivative characteristics or boundary optimal conditions of the current stage.

[0156] First, a set of theoretical candidate points for the current stage is formed. Preferably, the theoretical candidate points can come from at least one of the following:

[0157] The boundary endpoints, boundary intersections, or boundary representative points of the feasible domain at the current stage;

[0158] The feasible stationary point of the local value function in the current stage within the feasible region;

[0159] Representative extreme points indicated by changes in the sign of the first derivative, characteristics of the second derivative, or boundary optimal conditions.

[0160] The boundary optimality condition refers to the first-order necessary condition that must be satisfied when the potential extremum of the local value function of a stage is located at the boundary of the feasible region of the current stage. For example, for a minimization problem, if there exists a corresponding non-negative multiplier at a feasible boundary point such that the gradient of the local value function at that point satisfies the first-order stationary relationship with the gradient of the activation constraint at that point, then that point is considered to satisfy the boundary optimality condition and is regarded as a candidate boundary extremum point.

[0161] Thus, a theoretical candidate layer is formed that covers the local value function at the current stage and the possible extreme values ​​or competition points within the feasible region at the current stage.

[0162] (2) Representative Reduction Layer Formation: This layer is used to reduce theoretical candidate points by combining the monotonicity, convexity / concavity, stationary point effectiveness, and boundary optimality characteristics of the local value function at the current stage, forming a finite set of candidate strategy points. Specifically, it includes:

[0163] When the local value function satisfies the boundary optimality condition, candidate strategy points corresponding to the boundary endpoints and / or intersections are preferentially retained.

[0164] When a stationary point exists in the feasible domain of a stage and may constitute the required extreme value, the candidate policy point corresponding to the stationary point is retained.

[0165] When multiple candidate strategy points correspond to the same or similar resource allocation structures, or the same or similar boundary dominance patterns, the representative candidate strategy point is retained.

[0166] After forming the theoretical candidate layer, the theoretical candidate points are further reduced by combining the monotonicity, convexity and concavity, stationary point effectiveness and boundary optimality characteristics of the local value function at the current stage. Points that are theoretically questionable but cannot become the extreme values ​​required at the current stage or have no significance for subsequent competitive comparison are eliminated, and a limited number of representative points that may become the optimal strategy at the current stage or the extraction objects of subsequent key competitive strategies are retained.

[0167] Preferably, when the local value function in the current stage satisfies the boundary optimality condition, the boundary endpoints, boundary intersections, or representative boundary points are preferentially retained; when the local value function in the current stage has a feasible stationary point in the feasible domain of the current stage and the stationary point may constitute the extreme value required by the current problem, the feasible stationary point is retained; when multiple candidate points correspond to the same or approximately the same resource-sharing structure, boundary dominance mode, or competitive significance, one or more representative points are preferably retained.

[0168] After the above reduction, the set of candidate strategy points for the current stage is formed:

[0169]

[0170] in, Each point represents a finite number of points within the feasible region of the current stage where the local value function may become the current optimal strategy or the extraction object of subsequent key competitive strategies.

[0171] Through the hierarchical formation mechanism of the candidate strategy point set described above, this invention achieves the following: without relying on exhaustive comparison of all feasible solutions, it first ensures that no basic candidate objects of the current stage's extreme values ​​or competitive points are omitted, and then reduces the candidate points based on the structural and boundary characteristics of the local value function at the current stage, thereby forming a finite, clear, and subsequently competitive set of candidate strategy points. .

[0172] It should be noted that the extraction of candidate strategy points can be achieved by combining first derivatives, second derivatives, monotonicity determination, convexity / concavity determination, or boundary optimality conditions. However, this invention does not limit the specific extreme value algorithm itself, but rather uses the set of candidate strategy points as the input object for subsequent key competitive strategy extraction. (Candidate strategy point set) The formation of the key competitive strategy solves the problem of "which points may become the extreme values ​​or competitive objects in the current stage", while the subsequent key competitive strategy extraction solves the problem of "which of these candidate points are worth further comparison around the current optimal control strategy". The two are functionally connected but not mutually substitutable.

[0173] Step 3 (2), Key Competitive Strategy Extraction: Used to extract key competitive strategies from the existing candidate strategy set. Based on the current optimal control strategy Centered on this, we extract a set of key competitive strategies that can actually compete with the current optimal control strategy. It should be noted that this step is not a re-execution of the dynamic programming optimal solution, nor is it a re-search for a new optimal control strategy. Rather, it is based on the current optimal control strategy already obtained through dynamic programming, and identifies the local competitive relationships among candidate extreme strategies. The purpose is to extract from the candidate extreme strategies a limited number of competing objects that may truly affect the validity of the current optimal control strategy, serving as comparison objects for subsequent stages such as constructing the comprehensive control value difference function, solving the current switching boundary, and tracing the current link, rather than analyzing all candidate extreme strategies one by one.

[0174] Preferably, the key competitive strategy extraction mechanism includes steps such as candidate number determination, structural proximity extraction, equivalent proximity extraction, boundary association extraction, and multi-candidate reduction.

[0175] A. Reading the current optimal control strategy

[0176] Assume the dynamic programming solver is in stage The current optimal control strategy has been obtained:

[0177]

[0178] in, Indicates the current state. The current feasible domain And the optimal control action determined by the dynamic programming recursive relationship under the current parameter conditions.

[0179] It should be noted that this step does not replace the dynamic programming solver in resolving for the optimal solution, but rather is performed after the current optimal control strategy has been obtained. Based on this, competitive relationships are identified among surrounding or related candidate strategies.

[0180] B. Local comparison of candidate strategy points and determination of activation conditions

[0181] For the candidate strategy set any candidate strategy point Calculate the corresponding local value function for that stage:

[0182]

[0183] in, Indicates the current state. The following candidate strategy is adopted. The corresponding stage of comprehensive regulation value.

[0184] Furthermore, based on the current optimization direction of the dynamic programming problem, the comprehensive control value of each candidate strategy point at each stage is compared with the current optimal control strategy. The corresponding stage-based comprehensive regulatory values ​​are compared, and the competitive probability of candidate strategy points is determined based on the degree of value similarity, strategy structure characteristics, and boundary correlation characteristics.

[0185] Preferably, for any candidate strategy point If the following conditions are met: If the candidate strategy point is considered to have a strong value similarity to the current optimal control strategy, it can be used as a candidate for extracting key competitive strategies.

[0186] in, The threshold for the difference in comprehensive regulatory value at the preset stage.

[0187] C. Rules for extracting key competitive strategies and their activation conditions

[0188] In the set of candidate strategy points Based on this, and focusing on the current optimal control strategy Preferably, key competitive strategies are extracted based on local value proximity, structural proximity, and boundary correlation. It should be noted that each extraction rule is not activated in all cases, but rather selectively activated based on the number of candidate strategy points, the distribution characteristics of candidate strategy points, and the complexity of competitive relationships.

[0189] (1) Determination of candidate number and rule degradation

[0190] First, based on the candidate extreme value strategy set The number of non-optimal candidate strategies determines the initiation method for subsequent extraction rules.

[0191] When the candidate extreme value strategy set In addition to the current optimal control strategy When there is only one non-optimal candidate strategy, the candidate strategy is directly determined as the key competitive strategy, and subsequent structural proximity extraction, equivalent proximity extraction, and multi-candidate reduction rules are no longer initiated.

[0192] When the number of candidate extreme value strategies is small, it is preferable to start structural proximity extraction and equivalent proximity extraction; if the number of candidate strategies is still small after extraction, multi-candidate reduction will not be performed.

[0193] When there are a large number of candidate extreme value strategies, structural proximity extraction, equivalent proximity extraction, boundary association extraction, and multi-candidate reduction rules are initiated sequentially to obtain a set of key competitive strategies of a limited size.

[0194] By determining the number of candidates as described above, we can avoid mechanically executing all extraction rules when the number of candidate strategies is already small, thereby improving computational efficiency and ensuring the completeness of competition relationship identification.

[0195] (2) Structural Proximity Extraction Rules

[0196] When the control strategy can be represented as a resource undertaking structure, resource allocation structure, or resource call combination, it is preferable to first activate the structural proximity extraction rule.

[0197] Specifically, using the current optimal control strategy For reference, candidate extreme value strategies that differ from the current optimal control strategy only in local structure will be preferentially retained, including but not limited to:

[0198] The proportion of resources borne by a single entity has changed;

[0199] The start / stop status of a single resource has changed;

[0200] The depth of a single resource call has changed;

[0201] The activation state of a single constraint boundary changes;

[0202] The share of a single resource group or node group changes.

[0203] Preferably, only candidate strategies with limited changes in the local structure are retained as key competitive strategy candidates.

[0204] The scientific basis for this is that, in actual regulation processes, the solutions that are truly likely to replace the current optimal regulation strategy are usually not entirely different, but rather local adjustment solutions with a high degree of structural similarity to the current optimal regulation strategy. Therefore, structurally proximate strategies are better able to reflect the true competitive relationship formed by the current optimal regulation strategy.

[0205] (3) Equivalent Proximity Extraction Rule

[0206] After structural proximity extraction is completed, further extraction is performed based on the local value difference between candidate strategies and the current optimal control strategy.

[0207] For the candidate strategy set any candidate strategy point They are considered as potential key competitive strategy candidates and screened based on their value proximity to the current optimal control strategy.

[0208] Specifically, for candidate strategy points If the following conditions are met:

[0209]

[0210] If the candidate strategy point is considered to have a strong competitive proximity to the current optimal control strategy, it will be preferentially retained as a key competitive strategy candidate. The preset value difference threshold is used.

[0211] (4) Boundary association extraction rules

[0212] Furthermore, the boundary attributes corresponding to the candidate strategies are extracted.

[0213] When a candidate strategy corresponds to a representative location near the current stage's feasible domain boundary, constraint activation boundary, or strategy switching boundary (with a distance from the boundary less than a first distance threshold), the candidate strategy is preferably retained.

[0214] Especially when the current optimal control strategy is dominated by the constraint boundary (the distance to the constraint activation boundary is less than the second distance threshold), the boundary association extraction rule is activated first.

[0215] The scientific basis for this is that changes in actual control strategies usually occur near the constraint boundary. Candidate strategies near the boundary can more accurately reflect the formation mechanism of the current optimal control strategy under the influence of constraints, and can provide representative competing objects for the subsequent construction of the current switching boundary.

[0216] (5) Multiple candidate reduction rules

[0217] If the number of key competitive strategy candidates is still large after the above extraction, it is preferable to activate the multi-candidate reduction rule.

[0218] Specifically, the ranking is based on at least one of the following indicators:

[0219] (1) The difference in the overall regulatory value between candidate strategies and the current optimal regulatory strategy at different stages;

[0220] (2) The structural proximity between the candidate strategy and the current optimal control strategy;

[0221] (3) The importance of the boundary position corresponding to the candidate strategy.

[0222] Based on the ranking results, only a limited number of the most competitive candidate strategies are retained to form a key competitive strategy set.

[0223]

[0224] Its purpose is to avoid having too many comprehensive control value difference functions, current switching boundaries, and current link traceability objects in subsequent stages, thereby improving interpretation efficiency and highlighting the most representative competitive relationships.

[0225] By using the above extraction rules and their activation conditions, we can avoid mechanically activating all extraction rules when the number of candidate points is already small, thus ensuring that this step is both complete and simple in terms of engineering. At the same time, in actual dynamic programming control problems, although candidate strategy points have been reduced by step three (1), there may still be a large number of them, with similar structures or different boundary meanings. If all candidate points are included in the subsequent comparison, it is easy to cause redundancy in the construction of the stage comprehensive control value difference function, too many current switching boundaries, and lack of prominence in current link tracing. Therefore, this invention retains the candidate points with the most practical competitive significance through rules such as structural proximity, equivalent proximity, boundary association, and multi-candidate reduction.

[0226] D. Formation of a set of key competitive strategies

[0227] Suppose that the set of key competitive strategies obtained after extraction is:

[0228]

[0229] in, Representation phase The Middle A key competitive strategy This indicates the number of key competitive strategies at the current stage.

[0230] For any key competitive strategy To further construct its relationship with the current optimal control strategy The value difference function of comprehensive regulation between stages:

[0231]

[0232] The comprehensive regulation value difference function at this stage is used to characterize the current optimal regulation strategy relative to the first... The degree of advantage of each key competitive strategy, and the subsequent current switching boundaries. This provides the foundation for solving the problem.

[0233] In the subsequent analysis at this stage, this invention will no longer use the entire set of candidate strategy points. Instead of using it as a comparison object, the current optimal control strategy is used. With key competitive strategy set The competitive relationship between them serves as the basis for constructing the phase-based comprehensive regulation value difference function, solving the current switching boundary, and tracing the decision-making link.

[0234] E. Technical Function of this Step

[0235] Through this step, the solution result of the current stage of dynamic programming is no longer simply represented as the output of a single optimal control strategy, but is reconstructed into the following structured result:

[0236]

[0237] That is, a current optimal control strategy, a set of key competitive strategies extracted from candidate strategy points, and the relationship between the current optimal control strategy and the stage-based comprehensive control value difference between each key competitive strategy.

[0238] Therefore, the dynamic programming result is expanded from "only giving the optimal action" to "giving the optimal action and the source of its advantage relative to the key competitive strategy". This local competitive relationship can be used for the decomposition of the comprehensive regulation value difference function in subsequent stages, the solution of the current switching boundary, and the tracing of the regulation decision link, thereby improving the interpretability, traceability, and verifiability of the dynamic programming regulation result.

[0239] The second embodiment of the present invention relates to a power system decision link tracing method based on dynamic programming, comprising the following steps:

[0240] The optimal control strategy for the current stage is solved using a dynamic programming decision model; the dynamic programming decision model includes a stage cost function, a stage feasible region, and a state transition function.

[0241] Identify the key competitive strategies at each stage that affect the stability of the optimal control strategy;

[0242] Based on the difference between the local value functions of the corresponding key competitive strategies and the optimal control strategies, a phase-wide comprehensive control value difference function is constructed, and the strategy switching boundary is solved.

[0243] Following the recursive order of the dynamic programming decision-making model, the stage-specific comprehensive regulation value difference function and strategy switching boundary corresponding to each stage are linked together to form a decision tracing link;

[0244] Based on the current operational status, the decision-making basis for each stage is traced back along the decision tracing link to obtain the decision tracing result.

[0245] The dynamic programming decision model can use structured input, and the processing method for structured input includes the following steps:

[0246] The collected multi-source heterogeneous information from the power system is converted into structured data;

[0247] Decision-driving factors are extracted from structured data and mapped to the corresponding dynamic programming decision-making model components to form the structured input of the dynamic programming decision-making model.

[0248] In some preferred embodiments, each structured data item can be bound to a corresponding anchor identifier. This allows for reverse tracing of the original data information by using the anchor identifiers during the phased interpretation of the decision-making basis. The entire decision-making chain tracing method, such as... Figure 2 As shown, it includes the following steps:

[0249] Step 1: Collect multi-source information related to control and decision-making, and perform structural standardization processing on the multi-source information to construct a traceable anchor point system;

[0250] Step 2: Extract the driving factors that enter the dynamic programming solution process from the structurally standardized input. The driving factors include at least the target configuration driving factors, constraint boundary driving factors, and state transition influence driving factors, and map them to the stage cost function, stage feasible region, and state transition function of the dynamic programming, respectively.

[0251] Step 3: Based on the current stage's optimal control strategy obtained through dynamic programming, form a set of candidate strategy points based on the current stage state, the current stage feasible region, and the stage local value function. Then, extract a set of key competitive strategies from the set of candidate strategy points, focusing on the current optimal control strategy.

[0252] Step 4: Construct a phase-based comprehensive control value difference function for the current optimal control strategy relative to each key competitive strategy, and solve the current switching boundary between the current optimal control strategy and each key competitive strategy based on the phase-based comprehensive control value difference function, thereby determining the strategy region to which the current running point belongs.

[0253] Step 5: Based on the obtained stage comprehensive control value difference function, current switching boundary and strategy area determination results, and combined with the stage recursive relationship of dynamic programming, explain the formation reasons, stability interval and crossing conditions of the current optimal control strategy in each stage in stages, and associate the explanation results with the current control output, execution feedback and the state update of the next cycle to form the decision link traceability results.

[0254] Steps one through three can be performed using the method described in the first embodiment, and will not be repeated in this embodiment. Steps four and five will be further explained below.

[0255] Step 4: Constructing the phased comprehensive regulation value difference function, solving the current switching boundary, and determining the current strategy region:

[0256] This step involves constructing a phase-specific comprehensive control value difference function relative to the key competitive strategy, based on the current optimal control strategy and key competitive strategy extracted in step three. This function is then used to solve for the current switching boundary between the current optimal control strategy and the key competitive strategy, further determining the strategy region to which the current operating point belongs. Through this step, the current stage solution of dynamic programming no longer merely concludes with "the current optimal control strategy has been obtained," but further reveals why the current optimal control strategy holds true, why the key competitive strategy was not adopted, and which side of the current switching boundary the current operating point is located on.

[0257] It should be noted that the "current switching boundary" in this step is not the boundary between all global strategies, nor is it the first critical crossing boundary to be reached in the future. Rather, it is the local equivalence boundary between the current optimal control strategy and the key competing strategy in the Bellman sense at the current stage. The current switching boundary is only used to trace the reasons for the formation of the optimal solution at the current stage, the reasons why the current alternative was not adopted, and the strategy region to which the current running point belongs.

[0258] Step 4 (1) The difference function of the stage-based comprehensive control value between the current optimal control strategy and the key competitive strategy

[0259] For any key competitive strategy:

[0260]

[0261] Define the current optimal control strategy The stage-based comprehensive regulatory value difference function between the key competitive strategies mentioned above:

[0262]

[0263] in:

[0264] This indicates the overall regulatory value of the current optimal regulatory strategy at this stage.

[0265] Indicates the first The comprehensive regulatory value of each key competitive strategy at each stage.

[0266] When the local value function of a stage is represented by a revenue-type value:

[0267]

[0268] This indicates that the current optimal control strategy has a higher overall value compared to the key competitive strategy;

[0269]

[0270] This indicates that the current optimal control strategy and the key competitive strategy are in a locally equivalent state, corresponding to the current switching boundary;

[0271]

[0272] This indicates that under the corresponding state or parameter conditions, the key competitive strategy has a higher comprehensive value. This region corresponds to the candidate region where the current optimal control strategy may be switched, and is used to trigger strategy updates or boundary reassessments.

[0273] The meaning of the symbol is determined based on the specific optimization objective. When the local value function represents a revenue-type value function... This indicates that the current optimal control strategy has higher value; when the local value function of a stage represents a cost-type cost function, the sign should be adjusted accordingly based on the minimization objective. In a preferred embodiment of the present invention, the stage comprehensive control value difference function is further decomposed into:

[0274]

[0275] in, This represents the difference in objectives directly formed by the current optimal control strategy and the key competitive strategy at the current stage. This indicates the difference in value between the current optimal control strategy and the key competitive strategy in subsequent stages due to different decisions made at the current stage.

[0276] Therefore, this invention not only compares the current optimal control strategy with the key competitive strategy to determine which is better, but also further judges whether its advantage mainly comes from the current stage objective or from the cumulative value after the state is passed down in subsequent stages.

[0277] When multiple key competitive strategies exist, a set of local comparison results relative to each key competitive strategy can be formed. :

[0278]

[0279] Step 4 (2), Solving the current switching boundary: When there is at least one key competitive strategy with actual competitive significance in the current stage in addition to the current optimal control strategy, and the stage comprehensive control value difference function of the current optimal control strategy relative to the key competitive strategy will be reversed with the change of state, parameters or boundary conditions, the current switching boundary is constructed to characterize the local equivalence conditions between the current optimal control strategy and the key competitive strategy in the Bellman sense.

[0280] For any key competitive strategy Define the current optimal control strategy The current switching boundary conditions between the key competitive strategy and the current strategy are:

[0281]

[0282] Boundary conditions are expressed as follows: Under the current stage state, current stage solution structure, and current strategy comparison framework, the current optimal control strategy and the key competitive strategy are locally equivalent in the Bellman sense.

[0283] Furthermore, the corresponding current switching boundary is defined as a set:

[0284]

[0285] in, This represents a vector of combined parameters, states, or boundary variables relevant to the current stage and comparison. Therefore, the current switching boundary... It can be expressed as:

[0286] A critical value for a certain parameter;

[0287] A critical value for a certain state;

[0288] A critical value for the proportion of a certain resource's contribution;

[0289] A set of local equivalence conditions that several states / parameters satisfy.

[0290] in, This represents a vector of variable combinations related to the strategy comparison results at the current stage. This represents the value difference function of the comprehensive regulation at each stage. The set of variable combinations that equal zero, i.e., the current optimal control strategy and the first... The current set of switching boundaries between key competitive strategies.

[0291] Step 4 (3), current strategy region determination

[0292] A. Local strategy area relative to a single key competitive strategy

[0293] For any key competitive strategy According to the stage-based comprehensive regulation value difference function: A vector of combinations of state variables, parameter variables, or boundary variables: Based on the changing relationships, the current stage strategy space is divided into three local regions.

[0294] (1) The current optimal control strategy maintains the region

[0295]

[0296] in, This represents the set of states, parameters, and constraint boundary variables that affect the competitive relationship of strategies at the current stage.

[0297] when:

[0298]

[0299] When, it indicates that the current running point is relative to the current optimal control strategy relative to the first... The current optimal control strategy still has a competitive advantage in the region where key competitive strategies are maintained.

[0300] (2) Current switching boundary area

[0301]

[0302] Defined as:

[0303]

[0304] Right now:

[0305]

[0306] This region represents the local equivalence between the current optimal control strategy and the key competitive strategy in the Bellman sense, and is the critical boundary for the strategy switch between the two.

[0307] when:

[0308]

[0309] When this occurs, it indicates that the current running point is in a critical state of switching between two strategies.

[0310] (3) Potential areas of dominance for key competitive strategies

[0311]

[0312] This region indicates that as the current stage state, parameter conditions, or constraint boundaries change, the [missing information]... One key competitive strategy may surpass the current optimal control strategy and become the new optimal control strategy.

[0313] when:

[0314]

[0315] When this occurs, it indicates that the current operating conditions have crossed the current switching boundary, and the original optimal control strategy no longer maintains its advantage, requiring a re-execution of dynamic programming or a strategy update.

[0316] B. Overall retention area relative to all key competitive strategies

[0317] When multiple key competitive strategies exist:

[0318]

[0319] At that time, the overall retention region of the current optimal control strategy relative to all key competitive strategies is defined as:

[0320]

[0321] Right now:

[0322]

[0323] in:

[0324]

[0325] Representation phase Number of key competitive strategies.

[0326] The integrated retention region indicates that only when the current operating point simultaneously satisfies... At that time, the current optimal control strategy Only then can we maintain a competitive advantage relative to all key competitive strategies.

[0327] When multiple key competitive strategies exist, define the set of comprehensive competitive boundaries of the current optimal control strategy relative to all key competitive strategies:

[0328]

[0329] in:

[0330]

[0331] This represents the set of boundaries where the current optimal control strategy achieves a locally equivalent state relative to at least one key competitive strategy.

[0332] When the running point approaches or crosses the comprehensive competitive boundary, it indicates that the current strategy may be affected by key competitive strategies, and the risk of strategy change needs to be closely monitored.

[0333] C. Range-based strategy region determination

[0334] In an analytical implementation, if the current switching boundary can be simplified to the critical value of a certain parameter or state variable, then the current policy region can be further written in interval form. For example, when a certain single variable... When deciding on the comparison result, it can be written as:

[0335] The current optimal control strategy is dominant within a certain range, for example:

[0336]

[0337] Equivalent boundary points, for example:

[0338]

[0339] Key competitive strategies dominance intervals, for example:

[0340]

[0341] The specific direction of the inequality sign is determined by... The direction of the sign change is determined.

[0342] Step 4 (4): Formation of basic information for current link tracing

[0343] The phase-based comprehensive control value difference function, current switching boundary, and strategy area determination results output in this step serve as the basis for generating the current link tracing results in subsequent steps.

[0344] Preferably, the basic information includes at least:

[0345] (1) Current optimal control strategy:

[0346]

[0347] (2) Key competitive strategies:

[0348]

[0349] (3) Stage-based comprehensive regulation value difference function:

[0350]

[0351] (4) Differences in current stage objectives:

[0352]

[0353] (5) Value differences in subsequent stages:

[0354]

[0355] (6) Current switching boundary:

[0356]

[0357] (7) Local strategy area relative to a single key competitive strategy:

[0358]

[0359] (8) Overall retention area relative to all key competitive strategies:

[0360]

[0361] The technical role of this step is not to replace dynamic programming in searching for a new optimal solution, but to reconstruct the local competitive relationship between the current optimal control strategy and the key competitive strategy, based on the current optimal control strategy already obtained by dynamic programming.

[0362] pass:

[0363]

[0364] Dynamic programming, which originally only outputs... The single result is expanded into a competitive evidence structure consisting of the current optimal control strategy, key competitive strategy, stage difference function, current switching boundary, local strategy region, and comprehensive maintenance region.

[0365] This provides a foundation for subsequent steps to generate decision-making traceability results, such as "why this solution is currently adopted, why a key competitive solution is not adopted, which strategy zone we are currently in, and when a strategy switch may occur."

[0366] Step 5: Result tracing, stability interval determination, and feedback closed-loop update based on the stage recursive relationship.

[0367] This step, based on the comprehensive control value difference function, current switching boundary, and current strategy region determination results obtained in Step 4, combines the stage recursive relationship of dynamic programming with the parameterized coefficients in the model and their physical meanings to trace the formation process of the current optimal control strategy in stages. It further explains within what parameter range the current optimal control strategy remains unchanged in each stage, and under what conditions it will cross the current switching boundary, leading to a change in the stage strategy. Finally, the stage-by-stage tracing results are correlated with the current control output, execution feedback, and the state update of the next cycle, thus forming a tracing-execution-feedback closed loop oriented towards the optimal solution of the current stage.

[0368] It should be noted that the object of tracing in this step is not the isolated overall result, but rather the recursive formation process of the optimal strategy at each stage of dynamic programming. Therefore, this step preferably traces the current optimal control strategy at each stage according to the stage sequence of dynamic programming, and links the tracing results of each stage through the state transmission relationship between stages to form a complete current link tracing. The stage tracing object includes at least the current stage state, the current stage optimal control strategy, the current stage key competitive strategy, the stage comprehensive control value difference function, the current stage objective item difference, the value difference of subsequent stages, the current switching boundary, and the current stage strategy region determination result.

[0369] Step 5 (1), constructing the object in stages: for any stage of dynamic programming The current stage state has already been obtained in step four. Optimal control strategy at the current stage Key competitive strategies at the current stage The current stage of comprehensive regulation value difference function Differences in current stage objectives Value differences in subsequent stages Current switching boundary And the current stage strategy area determination results. Based on this, the construction phase... traceable objects ,in This indicates the policy region information to which the current stage operation point belongs.

[0370] Step 5 (2), based on the phased results traceability of the phased comprehensive regulation value difference function: for any phase , for stage The advantage of the current optimal control strategy relative to the key competitive strategy can be decomposed into the contribution in the current stage and the contribution in subsequent stages. When, it indicates that the current optimal strategy mainly stems from the direct control of profit advantages in this stage; when When this occurs, it indicates that the current optimal strategy primarily stems from the cumulative value advantage generated after the state transition. (Stage) The reason why the current optimal control strategy holds true can be traced back to the following: .

[0371] Step 5 (3), cross-stage tracing and connection based on stage recursion: Since dynamic programming satisfies stage recursion, therefore the stages The tracing process cannot be conducted in isolation; it should be linked together with the results of subsequent tracing stages. Preferably, the stages... Value differences in subsequent stages It should go through stages Value function or stage The traceability of objects is further supported. That is, the stage The validity of the current optimal control strategy depends not only on the current stage, but also on the state it induces in the next stage. Whether to make the stage The optimal control strategy maintains a higher cumulative value. Therefore, this invention preferably follows a dynamic programming recursive direction, connecting the traceable objects of each stage from back to front. This results in a phased, recursive tracing of the current link.

[0372] Step 5 (4), determination of the stable interval based on parameterized coefficients: Based on the difference function of the comprehensive control value of each stage and the corresponding current switching boundary, determine the stable interval of the current optimal control strategy in each stage and the conditions for the current extreme value to be crossed.

[0373] For any stage Based on the comprehensive regulation value difference function at this stage and the corresponding current switching boundary Further determination stage The stable range within which the current optimal control strategy remains unchanged. Preferably, the so-called stage The stable range refers to the period during which... When the parameterized coefficients or state variables change within a certain range, there is still Thus, the stage The stable interval is the range of parameters or states that remain unchanged under the current optimal control strategy. The stable interval can be represented by the inequality range of stage benefit coefficients, the inequality range of stage contribution coefficients, the upper or lower bound range of stage demand parameters, the effective interval of stage state transition coefficients, and the position range of stage boundary parameters relative to the current switching boundary.

[0374] Step 5 (5), determine the phased crossing conditions based on the current switching boundary: for any phase and any key competitive strategy The current optimal extreme value crossing condition is determined by Given. This condition indicates: stage. The current optimal control strategy and the key competitive strategy are locally equivalent in the Bellman sense, and the system is in a critical state for strategy switching at this stage. When the stage parameters change further, ... If the current optimal control strategy is reversed, it means that the competitive relationship between the current optimal control strategy and the key competitive strategy has been reversed. The current operating state has entered the strategy switching candidate region, and it is necessary to re-execute the strategy evaluation or dynamic programming solution to determine the new optimal control strategy.

[0375] Step 5 (6), generation of current link traceability results in stages: associate the stage traceability results with the current control output object, execution feedback object and the status update results of the next cycle to form a traceability-execution-feedback closed loop.

[0376] Based on the traceability objects at each stage The results of determining the stable intervals and crossing conditions at each stage are used to generate a complete phased current link tracing result. Preferably, the phased current link tracing result includes at least: the current optimal control strategy at each stage, the key competitive strategy at each stage, the comprehensive control value difference function at each stage, the difference in target items at the current stage at each stage, the value difference in subsequent stages at each stage, the current switching boundary at each stage, the stable interval at each stage, the crossing conditions at each stage, the strategy region to which the operating point at each stage belongs, and the state transmission tracing relationship between each stage.

[0377] Step 5 (7), Current Control Output, Execution Feedback, and Closed-Loop Update: After generating the phased current link tracing results, extract the current control output object and associate it with the execution feedback object. Preferably, the current control output object includes at least the device-level control instructions corresponding to the optimal control strategy for each phase, the resource allocation results for each phase, and the status output results of key nodes or key lines for each phase. The execution feedback object includes at least the actual equipment response results for each phase, the actual node voltage change results for each phase, the actual line current change results for each phase, the actual energy storage SOC update results for each phase, the execution deviation for the current cycle, and the status correction for the next cycle, and updates the status for the next cycle based on the execution feedback.

[0378] Step 5 (8), subsequent critical crossing boundary support in preferred embodiments: In some preferred embodiments, for the current switching boundary between the current optimal control strategy and the key competition strategy, the minimum disturbance amount required to reach each current switching boundary along the disturbance path can be calculated by combining the current disturbance path of concern, and the switching boundary corresponding to the minimum disturbance amount is taken as the critical crossing boundary to characterize the condition under which the current optimal control strategy is most likely to switch first. It should be noted that the critical crossing boundary analysis is only retained as a preferred extension in this main invention to support the subsequent application of the current switching boundary, and does not constitute the main inventive line of this main invention.

[0379] Step 5 (9), this step outputs: the current link tracing results in each stage; the parameterization basis for the current optimal control strategy in each stage; the stable interval in which the current optimal control strategy remains unchanged in each stage; the conditions for the current extreme value to be crossed in each stage; the current control output object. ; Current execution feedback object The initial state update result for the next cycle.

[0380] The following specific embodiments illustrate the reconstruction of the optimal strategy competition relationship, the solution of the current switching boundary, and the phased link in the equivalent scheduling problem of the transformer area / feeder layer in this embodiment.

[0381] In this embodiment, the comprehensive adjustment tasks that a certain distribution area / feeder layer needs to complete within a scheduling cycle are abstracted into a total amount of The equivalent regulation requirement. The equivalent regulation requirement can be understood as: the total amount of regulation tasks that need to be completed within the current cycle to meet power balance, voltage support, network boundary mitigation, or upper-level scheduling response. In the numerical example of this embodiment, we take:

[0382]

[0383] The adjustable resources within the system are divided into three resource groups according to a preset scheduling decomposition order, corresponding to three stages:

[0384] The first phase resource group is an energy storage resource group, and the adjustment amount is denoted as... ;

[0385] The second-stage resource group is a flexible load resource group, and the adjustment amount is denoted as... ;

[0386] The third-stage resource group consists of distributed power sources or other rapid support resource groups, with the adjustment amount denoted as... .

[0387] Here, "stage" refers to the order of decision decomposition in dynamic programming, and does not necessarily correspond to the absolute temporal order of physical device actions.

[0388] Let the contribution coefficients of the three resource groups to the unit equivalent adjustment demand be respectively The comprehensive regulation benefit parameters formed after the participation of the three types of resource groups in regulation are as follows: Its physical meaning is as follows:

[0389] These represent the contribution capabilities of the energy storage resource group, the flexible load resource group, and the third-stage resource group to the unit equivalent regulation demand, respectively. The larger the value, the more equivalent adjustment demand can be offset by the unit amount of this type of resource;

[0390] These represent the intensity of the comprehensive regulatory benefits formed after the three types of resource groups undertake regulatory tasks, which can comprehensively reflect economic benefits, supporting value, regulatory efficiency, or operational benefits.

[0391] These represent the amount of adjustment tasks assigned to the corresponding resource groups in each of the three stages.

[0392] Therefore, the equivalent scheduling problem for this transformer area / feeder layer can be expressed as:

[0393]

[0394]

[0395] In the numerical example of this embodiment, we take:

[0396]

[0397]

[0398] Therefore, the specific problem is:

[0399]

[0400] The purpose of this embodiment is not simply to find the optimal solution, but to illustrate how, based on the current optimal control strategy already obtained by dynamic programming, key competitive strategies are extracted, a phase-specific comprehensive control value difference function is constructed, the current switching boundary is solved, and a link tracing result with parameter physical meaning and cross-phase recursive relationship is formed.

[0401] Define the phase state For the stage If the remaining equivalent adjustment demand that was not initially met by the resource pool in the current and subsequent stages is:

[0402]

[0403] After the first phase of decision-making:

[0404]

[0405] After the second phase of decision-making:

[0406]

[0407] After the third phase of decision-making:

[0408]

[0409] Therefore, the recursive relationship of the three-stage dynamic programming is as follows:

[0410] Phase Three

[0411]

[0412] Phase Two

[0413]

[0414] Phase 1

[0415]

[0416] in, Representation phase In state The optimal value function for the next stage.

[0417] In the third stage, the local value function is:

[0418]

[0419] Its physical meaning is: when the system recursively reaches the third stage, the comprehensive regulatory value formed by the remaining equivalent regulatory demand borne by the resource group in the third stage. Among these, This indicates the value intensity of the regulatory task undertaken by the resource group in the third stage. This indicates the amount of regulation undertaken by the resource group in the third stage.

[0420] because ,and exist The interval is monotonically increasing, therefore the optimal solution is taken from the upper boundary:

[0421]

[0422] thereby:

[0423]

[0424] Its physical meaning is: when the system enters the third stage, all remaining equivalent adjustment demand is borne by the third stage resource group, and the amount borne is equal to the remaining demand. And the contribution capacity of the third-stage resource group They are inversely proportional.

[0425] Since the third stage only has a single boundary-optimal strategy and no longer has alternatives with practical competitive significance, a set of key competitive strategies is not constructed in this embodiment. Therefore, the target for tracing in the third stage can be denoted as:

[0426]

[0427] The purpose of this traceability object is to provide direct support for the second stage, "value differences in subsequent stages".

[0428] Substituting the optimal value function of the third stage into the function of the second stage, we get:

[0429]

[0430] Substitute the parameters into this embodiment:

[0431]

[0432] The feasible region for the second phase is:

[0433]

[0434] The physical meaning of this formula is:

[0435] First item The comprehensive regulation benefits generated by the flexible load resource group directly undertaking regulation tasks in the current stage;

[0436] Second item The remaining adjustment needs not undertaken in the current stage are passed on to the third stage, and the value formed by the resource group in the third stage is borne by the resource group in the subsequent stage.

[0437] Therefore, the decision in the second stage is not simply about choosing "who should bear more responsibility at present," but rather about comparing the combined value of the two options, "directly bearing the responsibility in the current stage" and "reserving it for the next stage," in the Bellman sense.

[0438] right Taking the second derivative, we get:

[0439]

[0440] Because in Shang Hengyou Therefore:

[0441]

[0442] Therefore, the local value function in the second stage is a convex function within the feasible region. Furthermore, since this embodiment is a maximization problem, the desired extrema preferentially appear at the boundary of the feasible region. Thus, the set of candidate strategy points for the second stage is formed as follows:

[0443]

[0444] in:

[0445] This indicates that the flexible load resource group will not undertake adjustment tasks in the second stage, and will reserve all remaining demand for the third stage resource group;

[0446] This indicates that the flexible load resource group will assume all remaining demand in the second phase.

[0447] Due to the set of candidate strategy points In this case, there is only one non-optimal candidate point besides the optimal strategy. Therefore, the second stage belongs to the unique candidate degenerate case, and the unique non-optimal candidate point is directly determined as the key competitive strategy.

[0448] Calculate the local values ​​of the two candidate points for each stage:

[0449] when hour,

[0450]

[0451] when hour,

[0452]

[0453] Therefore, the optimal control strategy and key competitive strategy for the second stage are determined as follows:

[0454]

[0455]

[0456] Their physical meanings are as follows:

[0457] In the second phase, the flexible load resource group will undertake all remaining equivalent adjustment needs.

[0458] The second phase does not utilize the flexible load resource group, but instead reserves all remaining demand to be handled by the third phase resource group.

[0459] According to the unified definition of this invention, the value difference function of the second-stage comprehensive regulation is:

[0460]

[0461] Right now:

[0462]

[0463] Summarized as follows:

[0464]

[0465] Furthermore, it can be broken down into:

[0466]

[0467] in:

[0468]

[0469] This indicates the comprehensive regulatory benefits directly generated in this stage by the current optimal regulation strategy in the second stage;

[0470]

[0471] This indicates that if a key competitive strategy is chosen, the remaining demand will be reserved for the resource group in the third stage, and the value of the subsequent stage formed will correspond to the value in the current comparison.

[0472] Its physical meaning is:

[0473] This indicates that the immediate comprehensive value generated by the flexible load resource group directly assuming demand in the second phase is positive.

[0474] This indicates that while reserving demand for the third-stage resource group will also create value for subsequent stages, this value is insufficient to surpass the value created by the flexible load resource group directly borne by the current stage under the parameters of this embodiment.

[0475] Set the value difference function of the second-stage comprehensive regulation to zero:

[0476]

[0477] The current switching boundary for the second phase is obtained as follows:

[0478]

[0479] In this embodiment, the following parameters are substituted:

[0480]

[0481] Therefore, the second-stage strategy area is:

[0482] The current optimal control strategy is dominant within the following range:

[0483]

[0484] Iso-boundary:

[0485]

[0486] Key competitive strategy dominance range:

[0487]

[0488] The physical meaning of this boundary is:

[0489] When entering the second stage of residual demand When the load is relatively small, it is better to have the flexible load resource group directly bear all the demand.

[0490] when When the value increases to a critical value, the combined value of the two options in the second stage, "directly assuming responsibility" and "reserving it for the third stage to assume responsibility," is equal.

[0491] when As the resource pool continues to grow, the cumulative value generated by reserving it for the third stage resource group is actually higher, so the strategy should be switched.

[0492] Since this embodiment will subsequently prove that the first stage has Therefore, when entering the second stage:

[0493]

[0494] It is obviously satisfied:

[0495]

[0496] Therefore, the current operating point is located within the region where the current optimal control strategy dominates in the second stage.

[0497] Therefore, the target for the second phase of tracing can be written as:

[0498]

[0499] This means that the current optimal control strategy in the second stage is valid not only because the flexible load resource group can generate higher immediate benefits by directly assuming demand in this stage, but also because it is linked to the objects traced in the third stage. The comparison shows that reserving the demand for the third phase and the resulting subsequent value is insufficient to overturn the current plan for the second phase.

[0500] Because the second stage satisfies the parameter conditions in this embodiment. Therefore:

[0501]

[0502] Substituting into the first stage, we get:

[0503]

[0504] Substitute the parameters into this embodiment:

[0505]

[0506] The feasible region for the first phase is:

[0507]

[0508] The physical meaning of this formula is:

[0509] First item The immediate comprehensive regulation benefits generated by the energy storage resource cluster directly undertaking regulation tasks in the first stage;

[0510] Second item The remaining demand not undertaken in the first phase is reserved for the second phase and then borne by the flexible load resource group, forming the value of the subsequent phase.

[0511] Therefore, the choice in the first phase is essentially a comparison:

[0512] The first phase will immediately mobilize energy storage resources to undertake the task;

[0513] In the first phase, energy storage resources will not be utilized, and the demand will be reserved for flexible load resources in the second phase.

[0514] right Find the second derivative:

[0515]

[0516] In this embodiment:

[0517]

[0518] Therefore, the local value function in the first stage is a convex function within the feasible region. Since this embodiment is a maximization problem, the required extreme points preferentially appear at the boundary endpoints. Thus, the set of candidate strategy points for the first stage is:

[0519]

[0520] in:

[0521] This indicates that the first phase of the energy storage resource group will not undertake regulation tasks;

[0522] This indicates that the first-stage energy storage resource cluster undertakes all the regulation tasks it can handle.

[0523] Due to the first phase candidate strategy point set In addition to the optimal strategy, there is only one non-optimal candidate point. Therefore, the first stage also belongs to the unique candidate degeneracy case, and the unique non-optimal candidate point is directly identified as the key competitive strategy.

[0524] Calculate the local values ​​of the two candidate points for each stage:

[0525] when hour,

[0526]

[0527] when hour,

[0528]

[0529] Therefore, the optimal control strategy and key competitive strategy for the first stage are determined as follows:

[0530]

[0531]

[0532] Their physical meanings are as follows:

[0533] In the first phase, energy storage resource clusters will not be prioritized for use.

[0534] In the first phase, the energy storage resource cluster will undertake all the regulation tasks it can handle.

[0535] According to the unified definition of this invention, the value difference function of the first stage comprehensive regulation is:

[0536]

[0537] Right now:

[0538]

[0539] Summarized as follows:

[0540]

[0541] Furthermore, it can be traced back to:

[0542]

[0543] in:

[0544]

[0545] This indicates the corresponding item in the current comparison if the energy storage resource group is immediately mobilized to undertake the task in the first stage;

[0546]

[0547] This indicates the value of the subsequent stages formed by the flexible load resource group in the second stage if the energy storage resource group is not used in the first stage and the demand is reserved for the second stage.

[0548] Its physical meaning is:

[0549] It reflects the intensity of the comprehensive regulation benefits of energy storage resource clusters on a unit equivalent regulation demand;

[0550] It reflects the intensity of the comprehensive regulation benefit of flexible load resource groups on a unit equivalent regulation demand;

[0551] when Higher than In this case, it is better to reserve the demand for the second phase of flexible load resources.

[0552] when Higher than In such cases, the first phase should prioritize utilizing energy storage resource clusters to undertake the tasks.

[0553] Set the first stage comprehensive regulation value difference function to zero:

[0554]

[0555] Obtain the current switching boundary for the first phase:

[0556]

[0557] Therefore, the first phase strategy area is:

[0558] The current optimal control strategy is dominant in the following regions:

[0559]

[0560] Iso-boundary:

[0561]

[0562] Key competitive strategies have an advantage in the following areas:

[0563]

[0564] Under the parameters of this embodiment:

[0565]

[0566] Right now:

[0567]

[0568] Therefore, the current operating point in the first phase is located in the dominant area of ​​"not prioritizing the use of energy storage resource clusters in the first phase".

[0569] Therefore, the first phase of tracing can be written as:

[0570]

[0571] This means that the current optimal control strategy in the first stage is valid not only because energy storage resources can be mobilized in the first stage, but also because of the tracking of the target in the second stage. Does the value of the subsequent stages it supports increase?

[0572] Because dynamic programming satisfies the recursive relationship between stages, the tracing of each stage cannot be performed in isolation, but should be linked together in a recursive order from back to front.

[0573] In this embodiment, the sequential relationship of the objects traced in stages is as follows:

[0574]

[0575] Its meaning is:

[0576] Third phase of tracing targets Explanation: When the system enters the final stage, how the remaining demand is handled by the third-stage resource group, and thus the basis for determining the value of subsequent stages of the second stage;

[0577] Second phase of tracing targets Note: Under the support of the third phase, the second phase's approach of "directly assuming all remaining demand by the flexible load resource group" is better than "reserving the demand for the third phase to assume."

[0578] First phase of tracing targets Note: Under the support of the second phase, the first phase's "not prioritizing the use of energy storage resource clusters" is better than "the first phase immediately using energy storage resource clusters to undertake tasks".

[0579] Therefore, the final optimal scheduling result in this embodiment is:

[0580]

[0581] It is not determined in isolation by a single stage, but by:

[0582] The optimal result for the third stage single boundary;

[0583] A comparison of the direct value of the current stage in the second phase with the subsequent value in the third phase;

[0584] A comparison of the direct value of the current stage in the first phase with the subsequent value in the second phase;

[0585] The chained result is formed step by step along the Bellman recursive relation.

[0586] From the value difference function of the first stage comprehensive regulation, we can see that:

[0587]

[0588] therefore:

[0589] when

[0590]

[0591] At that time, the current optimal control strategy for the first stage remains as follows:

[0592]

[0593] when

[0594]

[0595] At that time, the first stage is in a critical state of equivalent value;

[0596] when

[0597]

[0598] When the current extreme value in the first stage crosses the current switching boundary, the optimal strategy switches to:

[0599]

[0600] Its physical meaning is: when the overall regulation efficiency of the energy storage resource group on the unit equivalent regulation demand exceeds that of the flexible load resource group, the first-stage strategy switches from "reserving demand for subsequent stages" to "immediately calling upon the energy storage resource group".

[0601] From the value difference function of the second stage comprehensive regulation, we can see that:

[0602]

[0603] therefore:

[0604] when

[0605]

[0606] At that time, the optimal control strategy for the second stage remains as follows:

[0607]

[0608] when

[0609]

[0610] At that time, the second stage is in a critical state of equivalent value;

[0611] when

[0612]

[0613] When the current extreme value in the second stage crosses the current switching boundary, the optimal strategy switches to:

[0614]

[0615] Its physical meaning is: when the remaining equivalent adjustment demand when entering the second stage is large enough, the cumulative value formed by the resource group in the third stage will surpass the current stage value formed by the flexible load resource group in the second stage.

[0616] Based on the above analysis, the final optimal scheduling result of this embodiment is as follows:

[0617]

[0618] Its unified physical meaning is:

[0619] In the current total equivalent adjustment demand In the next phase, there is no need to prioritize the use of energy storage resources.

[0620] Reserving demand for the second phase of flexible load resources can generate higher comprehensive regulation value;

[0621] The second phase has already undertaken all the adjustment tasks, so the resource group does not need to intervene in the third phase.

[0622] In other words, under the parameter conditions of this embodiment:

[0623] The first phase of operation points are located in the dominant area where "energy storage resource clusters are not prioritized";

[0624] The second phase of operation is located in the dominant area where "flexible load resource clusters can handle all remaining demand";

[0625] The third stage retains only the single-boundary optimal solution;

[0626] Therefore, the overall optimal solution can be traced back to: the system prioritizes using the equivalent comprehensive regulation capacity of the flexible load resource group to complete the current regulation task, rather than immediately releasing the capacity of the energy storage resource group in the first stage or postponing the task to the third stage.

[0627] In this embodiment:

[0628] The equivalent comprehensive regulation value intensity of the first-stage energy storage resource cluster is:

[0629]

[0630] The equivalent comprehensive adjustment value intensity of the second-stage flexible load resource group is:

[0631]

[0632] The equivalent comprehensive regulatory value intensity of the resource group in the third stage is:

[0633]

[0634] It should be noted that the switching boundary conditions between the first and second stages are essentially a comparison of the equivalent comprehensive regulatory value intensity of different resource groups under the caliber of unit equivalent regulatory demand; while the switching conditions between the second and third stages depend on... and In addition to its relative size, it is further affected by the current remaining equivalent adjustment demand. Therefore, the maintenance range and switching boundary of the current optimal control strategy at each stage in this embodiment not only reflect the differences in the comprehensive adjustment benefit parameters of the resource group, but also reflect the comprehensive comparative relationship under the combined effect of the resource group's demand contribution capacity and the current remaining demand scale.

[0635] The strategy maintenance intervals, switching boundaries, and physical meanings for each stage are as follows:

[0636] I. First Phase

[0637] 1. Current optimal control strategy

[0638]

[0639] Its physical meaning is: in the first stage, the energy storage resource group is not prioritized for use, and the total equivalent regulation demand is reserved for the resource group to be processed in subsequent stages.

[0640] 2. Value difference function of phased comprehensive regulation

[0641]

[0642] in:

[0643] Total equivalent adjustment demand within the current scheduling cycle;

[0644] Parameters for the comprehensive regulation and adjustment benefits of energy storage resource clusters;

[0645] Parameters for the comprehensive regulation benefits of flexible load resource groups;

[0646] : The contribution coefficient of energy storage resource clusters to unit equivalent regulation demand;

[0647] : The contribution coefficient of flexible load resource group to unit equivalent adjustment demand.

[0648] Its physical meaning is to compare the comprehensive value of the two schemes, "immediately calling on energy storage to undertake the task in the first stage" and "reserving the demand for flexible loads in the second stage", under the unit equivalent demand adjustment caliber.

[0649] 3. Strategy maintenance range

[0650]

[0651] Equivalent to

[0652]

[0653] Its physical meaning is: the equivalent comprehensive regulation value intensity formed by the flexible load resource group under the unit equivalent regulation demand is higher than that of the energy storage resource group. Therefore, the strategy of "not prioritizing the use of the energy storage resource group" will be maintained in the first stage.

[0654] 4. Switch Boundaries

[0655]

[0656] Its physical meaning is: in the first stage, the two schemes of "immediately calling on the energy storage resource group to undertake the task" and "reserving the demand to the flexible load resource group in the second stage" are exactly equal in comprehensive value, and the system enters the critical state of the first stage strategy switching.

[0657] 5. Conditions after switching

[0658] when

[0659]

[0660] At that time, it is equivalent to

[0661]

[0662] This indicates that the equivalent comprehensive regulation value of the energy storage resource group under the "unit equivalent regulation demand caliber" is higher than the equivalent comprehensive regulation value of the flexible load resource group under the same caliber. Therefore, the first-stage strategy has been switched from "reserving demand for subsequent stages" to "immediately calling upon the energy storage resource group".

[0663] The corresponding strategy becomes:

[0664]

[0665] II. Second Phase

[0666] 1. Current optimal control strategy

[0667]

[0668] Its physical meaning is: in the second stage, the flexible load resource group will undertake all the remaining equivalent adjustment demand.

[0669] 2. Value difference function of phased comprehensive regulation

[0670]

[0671] in:

[0672] The remaining equivalent adjustment requirements upon entering the second stage;

[0673] Parameters for the comprehensive regulation benefits of flexible load resource groups;

[0674] Parameters for the comprehensive adjustment benefits of resource clusters in the third stage;

[0675] : The contribution coefficient of flexible load resource group to unit equivalent adjustment demand;

[0676] The contribution coefficient of the resource cluster in the third stage to the unit equivalent adjustment demand.

[0677] Its physical meaning is: to compare the comprehensive value difference between the two schemes: "the second phase is directly undertaken by the flexible load resource group to bear all the remaining demand" and "the second phase does not bear the demand and reserves the demand to the third phase resource group".

[0678] 3. Strategy maintenance range

[0679]

[0680] The physical meaning is that the remaining equivalent adjustment demand when entering the second stage is not large enough to be borne by the resource group in the third stage. It is better for the flexible load resource group to directly bear the demand in the current stage. Therefore, the second stage maintains the strategy of "bearing all the remaining demand itself".

[0681] 4. Switch Boundaries

[0682]

[0683] Its physical meaning is: the combined value of the two options in the second stage, "directly assuming the remaining demand" and "reserving it for the third stage to assume", is exactly equal, and the system enters the critical state of switching strategies in the second stage.

[0684] 5. Conditions after switching

[0685]

[0686] Its physical meaning is: representing the critical condition of residual demand that keeps the current optimal control strategy unchanged in the second stage. Where, on the left... This indicates the remaining equivalent adjustment demand upon entering the second stage; the right side represents this. This represents the comprehensive adjustment benefit parameters of the second-stage flexible load resource group. Parameters of the comprehensive regulation benefits of the resource cluster in the third stage and the contribution coefficients of the two types of resource groups The critical demand value is jointly determined. When When the value is below this critical threshold, it is better for the second stage to directly handle the remaining demand; when If the threshold is exceeded, it is better to reserve the remaining demand for the third phase.

[0687] The corresponding strategy becomes:

[0688]

[0689] III. Third Stage

[0690] 1. Current optimal control strategy

[0691]

[0692] Its physical meaning is: the resource group in the third stage undertakes all the remaining equivalent adjustment needs.

[0693] 2. Features Description

[0694] In the third stage, there is only a single boundary optimal strategy, and there are no longer any key competitive strategies with practical competitive significance. Therefore, in this embodiment, no further strategy switching boundaries are constructed in the third stage.

Claims

1. A method for reconstructing the competitive relationship of optimal strategies in a power system, characterized in that, Includes the following steps: The optimal control strategy for the current stage is solved using a dynamic programming decision model. The dynamic programming decision model is based on the stage cost function and the state transition function, and it recursively solves the optimal control strategy within the feasible region of each decision stage. A stage local value function is constructed; the stage local value function is constructed based on the stage cost function and the state transition function, and is used to calculate the stage comprehensive control value of the control strategy. The stage comprehensive control value includes the sum of the direct value of the current stage and the total value transmitted in subsequent stages. Candidate strategy points are determined based on the extreme value characteristics of the local value function of the stage and the boundary characteristics of the feasible region of the stage. Candidate strategy points that affect the stability of the optimal control strategy are extracted as key competitive strategies in the current stage, thereby reconstructing the competitive relationship between the key competitive strategies and the optimal control strategy in the current stage.

2. The optimal strategy competition relationship reconstruction method according to claim 1, characterized in that, The extraction of candidate strategy points that affect the stability of the optimal control strategy as key competitive strategies in the current stage includes the following steps: When there is only one candidate strategy point, that unique candidate strategy point shall be taken as the key competitive strategy. When there is not a unique candidate strategy point, candidate strategy points that satisfy the structural proximity condition with the current stage's optimal control strategy and candidate strategy points that satisfy the equivalent proximity condition are extracted as key competitive strategies. Among them, the structural proximity condition includes having only a single-dimensional difference with the current stage's optimal control strategy, and the equivalent proximity condition includes the difference in stage comprehensive control value between the current stage's optimal control strategy and the current stage's optimal control strategy being less than a set value difference threshold. The stage comprehensive control value difference is determined based on the difference in value obtained by substituting the current stage's optimal control strategy and the key competitive strategy into the stage's local value function.

3. The optimal strategy competition relationship reconstruction method according to claim 2, characterized in that, The extraction of candidate strategy points that affect the stability of the optimal control strategy as key competitive strategies in the current stage also includes the following steps: When the number of candidate strategy points exceeds the set threshold, candidate strategy points that meet the boundary proximity condition are extracted as key competitive strategies. The boundary proximity condition includes the distance from the set boundary being less than a first distance threshold. The set boundary includes the current stage feasible domain boundary, the constraint activation boundary, and the strategy switching boundary. The strategy switching boundary is determined based on the strategy points when the stage comprehensive control value difference is equal to zero.

4. The optimal strategy competition relationship reconstruction method according to claim 3, characterized in that, The extraction of candidate strategy points that affect the stability of the optimal control strategy as key competitive strategies in the current stage also includes the following steps: When there are multiple candidate strategy points, and the distance between the current optimal control strategy and the constraint activation boundary is less than the second distance threshold, candidate strategy points that satisfy the boundary proximity condition are extracted first as key competitive strategies.

5. The optimal strategy competition relationship reconstruction method according to claim 2, characterized in that, When the number of extracted key competitive strategies exceeds a set threshold, the process also includes a step of reducing the extracted key competitive strategies, specifically including: The key competitive strategies are sorted according to the set sorting rules; Based on the ranking results, retain the top-ranked key competitive strategies by a set number, and remove the remaining key competitive strategies. The sorting rules include at least one of the following rules: The key competitive strategies are ranked according to the difference in comprehensive regulatory value between the key competitive strategies and the optimal regulatory strategy at the current stage. The smaller the difference in comprehensive regulatory value between the key competitive strategies, the higher the priority of the ranking. The key competitive strategies are ranked based on their proximity to the current optimal control strategy in terms of resource allocation structure or control action structure. The closer the key competitive strategies are, the higher their ranking priority. The ranking is based on the location of key competitive strategies, with higher priority given to key competitive strategies that are closer to a set boundary.

6. The optimal strategy competition relationship reconstruction method according to claim 1, characterized in that, Before the step of extracting candidate strategy points that affect the stability of the optimal control strategy as the key competitive strategy in the current stage, the method further includes a step of reducing the candidate strategy points, specifically including: When the local value function of a stage satisfies the boundary optimality condition, candidate strategy points corresponding to the boundary endpoints and / or intersections are preferentially retained. When a stationary point exists in the feasible domain of a stage and may constitute the required extreme value, the candidate policy point corresponding to the stationary point is retained. When multiple candidate strategy points correspond to the same or similar resource allocation structures, or the same or similar boundary dominance patterns, the representative candidate strategy point is retained.

7. A method for tracing decision-making links in a power system, characterized in that, Includes the following steps: Obtain the key competitive strategy for the current stage, wherein the key competitive strategy for the current stage is obtained by reconstructing the optimal strategy competitive relationship as described in any one of claims 1-6; Calculate the difference between the local value function of the current stage key competitive strategy and the optimal control strategy, establish the stage comprehensive control value difference function, and solve the strategy switching boundary of the current stage based on the stage comprehensive control value difference function; Following the recursive order of the dynamic programming decision-making model, the stage-specific comprehensive regulation value difference function and strategy switching boundary corresponding to each stage are linked together to form a decision tracing link; Based on the current operational status, the decision-making basis for each stage is traced back along the decision tracing link to obtain the decision tracing result.

8. The decision-making link tracing method according to claim 7, characterized in that, The method for solving the current stage strategy switching boundary based on the stage-based comprehensive regulation value difference function includes the following steps: Based on the phase-based comprehensive regulation value difference function, the phase-based comprehensive regulation value difference between the current phase's optimal regulation strategy and each key competitive strategy is calculated respectively. The strategy point at which the comprehensive control value difference in the solution stage equals zero is used to form the strategy switching boundary for the current stage.

9. The decision-making link tracing method according to claim 8, characterized in that, The decision-making chain tracing results include the optimal control strategy, key competitive strategy, stage comprehensive control value difference function, direct value of the current stage, total value transmitted in subsequent stages, and strategy switching boundary for each stage.

10. The decision-making link tracing method according to claim 8, characterized in that, The decision-making link tracing results also include the local strategy region of the current operating state relative to each key competitive strategy. The local strategy region includes the strategy region when the difference between the stage comprehensive control value between the optimal control strategy and the corresponding key competitive strategy is greater than zero.

11. The decision-making link tracing method according to claim 8, characterized in that, The decision-making link tracing results also include the overall dominant region of the current operating state relative to all key competitive strategies. The overall dominant region includes the strategy region where the difference in the stage comprehensive control value between the optimal control strategy and each key competitive strategy is greater than zero.

Images