A city-level agent resource collaborative scheduling method and system

Abstract

The application discloses a kind of urban intelligent agent resource collaborative scheduling method and system, collect multi-agent interaction data to carry out conflict feature mining, from resource access sequence, generate competition hotspot set to identify high-frequency access resource, extract implicit dependency between intelligent agents to generate dependency transmission chain and construct resource conflict relationship graph;Resource conflict relationship graph is hierarchically analyzed, conflict convergence node is located to trace back conflict source and determine arbitration key point, and high-intensity conflict node is preferentially degraded in reverse;For influence source evaluation positioning core intelligent agent and auxiliary intelligent agent, extract load gain and loss and compensation ability to generate complementary pairing relationship;Extract response delay difference to generate delay compensation parameters for advance scheduling, identify idle time slice in execution window for dynamic compression, according to window scheduling table, generate resource allocation scheme by time sequence interleaving, realize the efficient collaborative scheduling and peak load shifting of urban multi-agent energy resources.

Description

Technical Field

[0001] This invention relates to the field of energy conservation and carbon reduction technology, and in particular to a city-level intelligent agent resource collaborative scheduling method and system. Background Technology

[0002] City-level energy conservation and carbon reduction systems involve the collaborative work of intelligent entities across multiple sectors, including building energy consumption, power supply, heat allocation, and transportation charging. These intelligent entities frequently compete and depend on each other during the acquisition and use of energy. Existing energy allocation methods often employ fixed priority ranking and single-period coordination strategies, which are insufficient to cope with complex situations such as a large number of intelligent entities, large fluctuations in energy demand, and significant time differences in execution response. This leads to problems such as energy supply and demand mismatch, concentrated peak electricity consumption, and insufficient allocation efficiency.

[0003] Traditional energy allocation methods lack in-depth analysis of the supply-demand network, fail to identify key energy nodes and competitive transmission paths, and do not adequately consider the execution time differences of different entities, resulting in a mismatch between auxiliary energy supply and core energy demand in time. Furthermore, there are numerous unused intervals within the task scheduling period, and the simultaneous startup of multiple entities leads to instantaneous load superposition, causing energy infrastructure overload. Therefore, a new city-level intelligent entity-based energy collaborative allocation method is needed to address at least one of the above problems. Summary of the Invention

[0004] This invention discloses a city-level intelligent agent resource collaborative scheduling method and system. It constructs a resource conflict relationship graph by collecting multi-agent interaction data and mining conflict features. It performs hierarchical analysis of the conflict relationship graph to locate key nodes and implements dynamic priority adjustment. It extracts the load surplus and deficit and compensation capabilities of core intelligent agents and auxiliary intelligent agents to establish complementary pairing. It performs advance scheduling compensation based on response delay differences, dynamically compresses the execution window, and generates a resource allocation scheme through time-series interleaving optimization.

[0005] The first aspect of this invention proposes a city-level intelligent agent resource collaborative scheduling method, comprising the following steps:

[0006] Collect multi-agent interaction data during the city-level energy conservation and carbon reduction operation, and construct a resource conflict relationship graph by mining the conflict features of the interaction data;

[0007] Based on the resource conflict relationship diagram, hierarchical analysis is performed to generate arbitration key points. Conflict intensity features are extracted from the arbitration key points to generate intensity identification parameters. High-intensity conflict nodes in the intensity identification parameters are identified and priority is reversed to generate suppression parameters. The intensity identification parameters and the suppression parameters are used to perform mutual exclusion partitioning to determine the exclusive scheduling domain.

[0008] An influence tracing assessment is performed on the exclusive scheduling domain to generate tracing weights. Based on the tracing weights, the core intelligent agent and the auxiliary intelligent agent are located from the exclusive scheduling domain. The load surplus and deficit of the core intelligent agent and the compensation capability of the auxiliary intelligent agent are extracted to generate a complementary pairing relationship. Based on the complementary pairing relationship, resources are cross-allocated to construct a scheduling topology table.

[0009] The scheduling topology table is negotiated and cross-parsed to construct cross-decision nodes. The response delay difference of the auxiliary agent is extracted from the cross-decision nodes to generate delay compensation parameters. The delay compensation parameters are used to pre-schedule the core agent to generate a response delay spectrum.

[0010] The response delay spectrum is analyzed by back-calculation of scheduling time to generate a time base sequence. The execution window of each agent is determined based on the time base sequence. Idle time segments in the execution window are identified and dynamically compressed to generate a window scheduling table. A resource allocation scheme is generated by time-series interleaving based on the window scheduling table.

[0011] A second aspect of this invention proposes a city-level intelligent agent resource collaborative scheduling system, comprising:

[0012] The data acquisition module is used to collect interaction data of multiple agents during the city-level energy conservation and carbon reduction operation, and to mine conflict features from the interaction data to construct a resource conflict relationship graph.

[0013] The conflict resolution module is used to perform hierarchical resolution based on the resource conflict relationship diagram to generate arbitration key points, extract conflict intensity features from the arbitration key points to generate intensity identifier parameters, identify high-intensity conflict nodes in the intensity identifier parameters to perform priority reverse degradation to generate suppression parameters, and use the intensity identifier parameters and the suppression parameters to perform mutual exclusion partitioning to determine the exclusive scheduling domain.

[0014] The agent matching module is used to perform influence tracing assessment on the exclusive scheduling domain to generate tracing weights, locate the core agent and auxiliary agent from the exclusive scheduling domain based on the tracing weights, extract the load surplus / deficit of the core agent and the compensation capability of the auxiliary agent to generate a complementary pairing relationship, and perform cross-allocation of resources based on the complementary pairing relationship to construct a scheduling topology table.

[0015] The delay compensation module is used to negotiate and cross-parse the scheduling topology table to construct cross-decision nodes, extract the response delay difference of the auxiliary agent from the cross-decision nodes to generate delay compensation parameters, and use the delay compensation parameters to pre-schedule the core agent to generate a response delay spectrum.

[0016] The window optimization module is used to perform scheduling time back-analysis on the response delay spectrum to generate a time base sequence, determine the execution window of each agent based on the time base sequence, identify idle time segments in the execution window and dynamically compress them to generate a window scheduling table, and generate a resource allocation scheme by performing time-series interleaving based on the window scheduling table.

[0017] The beneficial effects of this invention are reflected in the following points: First, by identifying competition hotspots from resource access sequences and extracting dependency propagation chains to construct a resource conflict relationship graph, and then decomposing the conflict relationship graph into a main conflict layer and branch conflict layers, the technical path of tracing the source of conflict along the conflict convergence node to determine the arbitration key point and performing reverse priority downgrading on high-intensity conflict nodes can systematically present the energy resource competition and supply-demand dependence relationship in a city-level system, identify key energy bottleneck nodes and conflict propagation paths, and reduce the frequency of resource contention at high-intensity conflict nodes by dynamically adjusting priorities, thereby alleviating the energy conflict pressure of the city-level system. Second, by using influence tracing assessment to generate tracing weights to locate core intelligent agents and auxiliary intelligent agents, the load surplus / deficit of the core intelligent agent and the compensation capability of the auxiliary intelligent agent are extracted for matching degree assessment to establish a complementary pairing relationship. Based on the complementary pairing relationship, resources are cross-allocated to construct a scheduling topology table, enabling the core intelligent agent and auxiliary intelligent agent to form complementary collaboration based on the matching of load surplus / deficit and compensation capability, thereby improving the utilization rate of energy resources and the supply reliability of the system. Finally, by combining response delay differences to generate delay compensation parameters for the core agents to schedule in advance, and by performing bidirectional time-axis scanning of the execution window to locate idle time segments and identify reverse filling opportunities, and by elastically shrinking the filling candidate set to generate a compact execution window, the tasks of multiple agents that were originally concentrated at the same time are staggered according to the window scheduling table. This bridges the problem of energy compensation not being timely due to the execution time difference of different agents, while reducing the peak of instantaneous energy demand, and providing a refined multi-agent collaborative scheduling capability for city-level energy-saving and carbon reduction systems.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0020] Unless otherwise specified or defined, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0021] Figure 1 This is a flowchart illustrating a city-level intelligent agent resource collaborative scheduling method according to the present invention.

[0022] Figure 2 This is a structural block diagram of a city-level intelligent agent resource collaborative scheduling system according to the present invention. Detailed Implementation

[0023] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0024] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0025] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0026] The technical solutions of the embodiments of this application will be described below.

[0027] like Figure 1 As shown, this embodiment of the invention provides a city-level intelligent agent resource collaborative scheduling method, including the following steps S110-S150:

[0028] Step S110: Collect multi-agent interaction data during the city-level energy conservation and carbon reduction operation, and construct a resource conflict relationship graph by mining the conflict features of the interaction data.

[0029] Specifically, the system collects interaction data from multiple agents during city-level energy conservation and carbon reduction operations. Data acquisition modules are deployed at each node of the city-level energy conservation and carbon reduction system to monitor agent behavior events in real time. The acquisition frequency is set to 100 milliseconds, capturing interactions between agents such as energy coordination requests, resource allocation requests, load adjustment instructions, and status synchronization messages. Agents include building energy management agents, power grid dispatch agents, heating system agents, and charging dispatch agents. Each agent's interaction behavior during task execution includes fields such as timestamp, source agent identifier, target agent identifier, interaction type, and involved resource identifiers. The acquisition module records interactions between building energy management agents and power grid dispatch agents requesting power resource allocation, recording the request initiation time, resource type, power demand, and request priority. In city-level power dispatch scenarios, agents frequently request and release power resources; the acquisition module captures these power allocation and recovery interaction data. In district heating systems, multiple building agents concurrently request heating resources; the acquisition module records the application, supply, and adjustment processes for heating allocation. The collected raw interaction data is preprocessed to filter out non-business-related interaction data such as heartbeat detection, and retain the valid interaction data involving energy resource access and task collaboration.

[0030] In some embodiments, the step of mining conflict features from the interaction data to construct a resource conflict relationship graph includes: extracting resource access sequences from the interaction data; identifying high-frequency access resources from the resource access sequences to generate a set of competitive hotspots; extracting implicit dependencies between agents in the set of competitive hotspots to generate dependency transit chains; and constructing a resource conflict relationship graph based on the dependency transit chains.

[0031] Resource access sequences are extracted from the interaction data. The interaction data is sorted by timestamp to form a global time-series interaction event stream. The interaction event stream is traversed to extract the energy resource access events for each agent. Energy resource access events include access time, agent identifier, resource identifier, access type, and energy demand. Access types are divided into four categories: electricity request, heat request, cooling request, and load reduction. For resource access events of the same agent, they are concatenated in chronological order to form the resource access sequence for that agent. Building energy consumption agent A requests 150kW of power from power distribution resource R1 at time T1, requests 80GJ of heat from heating resource R2 at time T2, and requests 200kW of power from power distribution resource R1 again at time T3, forming the access sequence [R1@T1:150kW,R2@T2:80GJ,R1@T3:200kW]. The access sequence of power grid dispatching agent B is [R1@T1.5:100kW, R3@T2.5:300kW, R1@T3.5:180kW]. The frequency of each energy resource appearing in the resource access sequences of all agents is counted to form resource access frequency statistics.

[0032] This process identifies frequently accessed resources from resource access sequences to generate a set of competitive hotspots. Access records for energy resources are extracted from the resource access sequences of all agents, and the frequency of each energy resource in the access sequence is statistically analyzed. Energy resources are then sorted in descending order of access frequency. A high-frequency access threshold is set as the 75th quantile of the access frequency distribution. Energy resources with access frequencies exceeding the threshold are identified and marked as high-frequency access resources. In city-level power grid dispatching, the access frequency of distribution transformers in commercial areas is 8500 times, and that of distribution transformers in industrial parks is 7200 times, both exceeding the threshold and thus marked as high-frequency access resources. The concurrent access characteristics of high-frequency access resources in the resource access sequences are analyzed, and the number of times each high-frequency resource is simultaneously requested by multiple agents within a unit time window is statistically analyzed. This number reflects the intensity of competition for energy resources. For distribution transformers in commercial areas, an average of 12 building energy consumption agents concurrently request them within a 1-second time window. The energy competition intensity index of the resource is calculated; the competition intensity index is equal to the product of the number of concurrent requests and the access frequency. Energy resources with a competition intensity index exceeding a threshold are extracted as competition hotspots and compiled into a competition hotspot set. This set includes the access frequency, concurrent request count, and competition intensity index for each energy resource.

[0033] Extract implicit dependencies between agents in a competitive hotspot set to generate dependency transit chains. For each energy resource in the competitive hotspot set, analyze the access timing relationships of all agents accessing that resource. When agent A accesses power distribution resource R and occupies power at time T1, and agent B requests access to the same power distribution resource R at time T2, and T2-T1 is less than the average power supply cycle of the resource, it is determined that B has a power waiting dependency relationship with A. In the power distribution scenario of a commercial complex, building energy consumption agent A obtains 300kW power from the central air conditioning system at 10:00:00, and building energy consumption agent B requests 150kW power from the elevator system at 10:00:02, but the power distribution capacity is insufficient and it is blocked. B only successfully obtains power after agent A reduces the air conditioning power at 10:05:00, thus identifying B's power waiting dependency on A. Analyze the energy supply cooperation mode between agents in the competitive hotspot set. When the waste heat recovery output of agent A is the heating input of agent B, establish an energy supply dependency relationship from A to B. By integrating the dependencies on electricity waiting and energy supply, a dependency network among intelligent agents is constructed. The propagation path of these dependencies is traced; when agent A depends on B, B depends on C, and C depends on D, a dependency propagation chain is formed: A→B→C→D.

[0034] A resource conflict graph is constructed based on dependency transitive chains. The graph is created where each node corresponds to an agent, and node attributes include agent identifier, managed energy type, and average power consumption. The dependency transitive chain set is traversed, and for adjacent agents in a chain, a directed edge is created in the conflict graph. The edge direction points from the dependent agent to the dependent agent, and the edge weight is set according to the energy dependency strength. Energy dependency strength is calculated by combining power waiting time and dependency frequency. If agent B waits an average of 500 seconds for agent A to release power capacity, and this waiting occurs 30 times within the observation period, the energy dependency strength is set to 15000. For the dependency transitive chain A→B→C, edges A→B and B→C are created in the conflict graph, with edge weights corresponding to the strength of each energy dependency relationship. In addition to the directed edges generated by the dependency transitive chains, concurrent competition relationships in concentrated competition hotspots also need to be handled. For energy resources in concentrated competition hotspots, agent pairs accessing the resource simultaneously are identified, and undirected edges are created between these agents to represent energy resource competition relationships. The weight of an undirected edge represents the frequency of power competition. When agents D and E engage in 45 concurrent competitions for the distribution transformer within the observation period, an undirected edge with a weight of 45 is created. Integrating energy dependency edges and energy competition edges forms a complete resource conflict graph. In this graph, directed edges reflect the power waiting dependency and energy supply dependency among agents, while undirected edges reflect concurrent competition for energy resources, comprehensively showcasing the overall picture of energy resource conflicts among multiple agents in a city-level energy conservation and carbon reduction system.

[0035] Step S120: Based on the resource conflict relationship diagram, perform hierarchical analysis to generate arbitration key points, extract conflict intensity features from the arbitration key points to generate intensity identification parameters, identify high-intensity conflict nodes in the intensity identification parameters and perform priority reverse downgrading to generate suppression parameters, and use the intensity identification parameters and suppression parameters to perform mutual exclusion partitioning to determine the exclusive scheduling domain.

[0036] In some embodiments, the step of generating arbitration key points by hierarchical parsing based on the resource conflict relationship diagram includes: decomposing the resource conflict relationship diagram into a main conflict layer and a branch conflict layer; locating conflict convergence nodes from the main conflict layer and the branch conflict layer; tracing the conflict source location along the conflict convergence node; and determining arbitration key points based on the conflict source location.

[0037] The resource conflict graph is decomposed into a backbone conflict layer and branch conflict layers. The weights of all edges in the graph are extracted, and their statistical distribution characteristics are calculated. A backbone edge threshold is set, typically the 60th quantile of the edge weight distribution. For a resource conflict graph with 500 edges, the threshold corresponds to the 300th value after the edge weights are sorted. Edges in the resource conflict graph whose weights exceed the backbone edge threshold are identified, and these edges and their connected nodes are extracted to form the backbone conflict layer. The backbone conflict layer contains the most frequent and intense energy resource conflicts in the system, typically consisting of dependencies and competition between core energy-consuming agents. In a city-level energy conservation and carbon reduction scenario, the energy dependency edge weight between the building energy consumption agent and the grid dispatch agent is 8500, and the thermal competition edge weight between the heating system agent and the building energy consumption agent is 7200, both exceeding the threshold of 6000 and thus classified into the backbone conflict layer. Edges with weights lower than the threshold of the main edge in the resource conflict graph and the nodes they connect to are extracted as branch conflict layers. The branch conflict layers contain energy resource conflict relationships with low frequency or small impact.

[0038] The conflict convergence nodes are located in the main conflict layer and branch conflict layers. The in-degree of each node in the main conflict layer is analyzed; in-degree represents the number of edges pointing to that node. A high in-degree threshold of 3 is set. When the in-degree of a node in the main conflict layer is greater than or equal to 3, that node is identified as a conflict convergence node in the main layer. In the city-level power grid dispatch scenario, the distribution transformer node receives energy-dependent edges from three nodes: the building energy consumption agent, the charging dispatch agent, and the industrial energy consumption agent. With an in-degree of 3, it is identified as a conflict convergence node in the main layer. The in-degree of each node in the branch conflict layer is analyzed, and a high in-degree threshold of 2 is set for the branch layer. When the in-degree of a node in the branch layer is greater than or equal to 2, it is identified as a conflict convergence node in the branch layer. The heating station node receives heat demand input from the office building agent and the commercial building agent. With an in-degree of 2, it is identified as a conflict convergence node in the branch layer. The conflict convergence nodes in the main conflict layer and the branch conflict layer are merged to form a complete set of conflict convergence nodes. This set contains all key nodes whose in-degree reaches the threshold standard.

[0039] For example, tracing the source location of a conflict along the conflict convergence node includes: identifying the multi-branch in-degree connection relationship of the conflict convergence node; performing parallel reverse tracing along the multi-branch in-degree connection relationship to generate a multi-path tracing set; evaluating the propagation intensity of the multi-path tracing set to extract the dominant propagation path; and locating the source location of the conflict based on the dominant propagation path.

[0040] Identify the multi-branch in-degree connections of conflict-affected convergence nodes. For a given conflict-affected convergence node, extract the set of all incoming edges pointing to that node. Traverse the set of incoming edges, recording the source node, edge weight, and edge type of each incoming edge. Count the number of incoming edges; the number of incoming edges is the in-degree of the convergence node. For a convergence node N1 with an in-degree of 5, its five incoming edges come from nodes A, B, C, D, and E, with edge weights of 120, 95, 80, 150, and 110, respectively, and edge types including electricity dependence and thermal competition. Analyze the source node types of the incoming edges; the source node may be another convergence node, a general energy-consuming node, or a conflict source node. Group the incoming edges according to the upstream hierarchy and topological relationship of the source nodes to form multi-branch in-degree connections. Multi-branch in-degree connections describe the hierarchical structure, weight differences, and topological distribution among the incoming edges, providing a path basis and branch guidance for parallel reverse tracing.

[0041] A multi-path tracing set is generated by parallel reverse tracing along the multi-branch in-degree connection relationship. Based on the multi-branch in-degree connection relationship, for each incoming edge branch of the sink node, the traversal proceeds upstream from the source node of the incoming edge. When a node with an in-degree of 0 is reached, this node is marked as the end of the path and the tracing stops. For branch 1, tracing starts from node A, passes through nodes F and K, and reaches node S1 with an in-degree of 0, forming path P1: N1←A←F←K←S1, with a path length of 4 and a cumulative edge weight of 320. For branch 2, tracing starts from node B, passes through node G, and reaches node S1 with an in-degree of 0, forming path P2: N1←B←G←S1, with a path length of 3 and a cumulative edge weight of 205. For branch 3, tracing starts from node C, passes through nodes H, L, and M, and reaches node S2 with an in-degree of 0, forming path P3: N1←C←H←L←M←S2, with a path length of 5 and a cumulative edge weight of 240. All branches along the multi-branch in-degree connection relationship are summarized to form a multi-path tracing set, which contains 5 complete tracing paths starting from the aggregation node N1.

[0042] The propagation intensity of the multi-path tracing set is assessed to extract the dominant propagation path. For each path in the multi-path tracing set, the energy propagation intensity index is calculated. The energy propagation intensity index comprehensively considers the cumulative edge weight and path length of the path, and the calculation formula is I=W / L, where I is the propagation intensity, W is the cumulative edge weight of the path, and L is the path length. For path P1, the cumulative edge weight is 320, the path length is 4, and the propagation intensity is 320 / 4=80. For path P2, the cumulative edge weight is 205, the path length is 3, and the propagation intensity is 205 / 3=68.33. For path P3, the cumulative edge weight is 240, the path length is 5, and the propagation intensity is 240 / 5=48. The propagation intensity of all paths in the multi-path tracing set is ranked, and the path with the highest propagation intensity is identified as the dominant propagation path. In the current case, path P1 has the highest propagation intensity of 80 and is extracted as the dominant propagation path from the multi-path tracing set.

[0043] The source of conflict is located based on the dominant propagation path. The endpoint node of the dominant propagation path is extracted, and this endpoint node is the source of conflict. For the dominant propagation path P1: N1←A←F←K←S1, the endpoint node S1 is the source of conflict. The in-degree of the node corresponding to the source of conflict is verified, confirming that its in-degree is 0, indicating that the node does not depend on other nodes and is the upstream starting point of energy conflict propagation. The functional type and energy characteristics of the node corresponding to the source of conflict are analyzed to identify the root cause of the conflict at this location. Node S1 corresponds to the user's electricity demand access agent. This agent generates a large number of electricity demand requests during peak electricity consumption periods, leading to intensified energy competition in the downstream power dispatch link. For the conflict convergence node N2, the endpoint node S2 of its dominant propagation path P4 corresponds to the industrial park agent. This agent generates a large amount of heat demand during peak production periods, forming another source of conflict. The distribution of conflict source locations in the system is statistically analyzed: electricity-related conflict sources account for 45%, heat-related conflict sources account for 35%, and comprehensive energy-related conflict sources account for 20%. The locations of the conflict sources of all conflict convergence nodes are aggregated to form a system-level distribution of conflict source locations.

[0044] Arbitration key points are determined based on the location of the conflict source. The distribution of conflict source locations is analyzed, and the number of downstream convergence nodes affected by each conflict source location is counted. The more convergence nodes affected, the wider the influence range of the conflict source location. Conflict source locations affecting more than a threshold of 3 are identified. When a conflict source location affects 3 or more convergence nodes, the corresponding node is marked as a candidate arbitration key point. Conflict source location S1 affects three convergence nodes: a distribution transformer, a heat exchange station, and a charging station, and is therefore marked as a candidate arbitration key point. The distribution of candidate nodes in the main conflict layer and branch conflict layers is analyzed, and nodes corresponding to conflict source locations in the main conflict layer are preferentially selected as arbitration key points. A comprehensive score is given to the candidate arbitration key points, considering three dimensions: influence range, energy conflict intensity, and scheduling controllability. The influence range is measured by the number of downstream convergence nodes, the energy conflict intensity is measured by the cumulative weight of the dominant propagation path, and scheduling controllability is measured by the node's load adjustment capability. The node with the highest comprehensive score is selected as the arbitration key point, forming the arbitration key point set.

[0045] The system extracts conflict intensity features from key arbitration points to generate intensity identification parameters. For each key arbitration point, it analyzes the energy conflict path information passing through that node. It counts the number of energy requests flowing through the key arbitration point, reflecting the load intensity of that node. It calculates the average waiting time at the key arbitration point, which refers to the length of time the agent waits for energy resource allocation at that node; a longer waiting time indicates more intense energy competition at that node. It extracts the energy type distribution of the key arbitration points to identify the main energy types involved, such as electricity, heat, cooling, or combined energy. For the distribution transformer arbitration point, it mainly involves electricity resource allocation and load balancing. It analyzes the frequency of energy conflicts at key arbitration points in different time periods to identify periods of concentrated conflict. Finally, it integrates the number of energy requests, average waiting time, energy type distribution, and conflict frequency distribution to construct an intensity identification parameter vector, which provides a quantitative description of the energy conflict intensity for each key arbitration point.

[0046] In some embodiments, the step of identifying high-intensity conflict nodes in the intensity identifier parameters and performing priority downgrading to generate suppression parameters includes: performing threshold filtering on the intensity identifier parameters to identify high-intensity conflict nodes; extracting the resource contention frequency of the high-intensity conflict nodes; determining the priority downgrading magnitude based on the resource contention frequency; and generating suppression parameters based on the priority downgrading magnitude.

[0047] High-intensity conflict nodes are identified by threshold screening of intensity identifier parameters. An intensity identifier parameter vector is extracted, with each dimension corresponding to a different energy conflict intensity characteristic. Statistical analysis is performed on each intensity characteristic dimension, calculating the mean, standard deviation, and quantile distribution of the values ​​for that dimension. For the energy request quantity dimension, the average number of requests across all arbitration key points is calculated to be 1200 requests / hour, the standard deviation to be 450 requests / hour, and the 75th quantile to be 1500 requests / hour. High-intensity thresholds are set for each dimension, with the threshold set to the mean plus one standard deviation. For the energy request quantity dimension, the high-intensity threshold is set to 1200 + 450 = 1650 requests / hour. For the average waiting time dimension, the high-intensity threshold is set to the mean of 300 seconds plus one standard deviation of 150 seconds, i.e., 450 seconds. Multi-dimensional threshold screening is performed on each arbitration key point; when a node exceeds the high-intensity threshold in at least two dimensions, the node is identified as a high-intensity conflict node. Arbitration key point N5 had 1800 energy requests per hour and an average waiting time of 520 seconds, exceeding the threshold in both dimensions and was identified as a high-intensity conflict node.

[0048] Extract the resource contention frequency of high-intensity conflict nodes. For each high-intensity conflict node, analyze the energy resource contention events of that node within the observation period. An energy resource contention event refers to multiple agents requesting access to the energy resources controlled by that node within the same time window, and the total demand exceeding the supply capacity. Set the time window size to 1 second, and count the number of concurrent energy requests occurring within each time window. When the total energy demand of concurrent requests exceeds the available resource, it is determined to be a resource contention event. For a distribution transformer node with a transformer capacity of 2000kW, resource contention occurs when the total power of concurrent requests exceeds 2000kW within 1 second. Count the number of resource contention events occurring within the observation period, which is usually set to 1 hour or 1 day. Within a 1-hour observation period, high-intensity conflict node N5 experienced 320 resource contention events. Calculate the resource contention frequency, which is the number of contention events divided by the total number of time windows within the observation period. The observation period of 1 hour contains 3600 time windows. The resource contention frequency of node N5 is 320 / 3600=0.089, which means that resource contention occurs in 8.9% of the time windows.

[0049] The priority degradation range is determined based on the frequency of resource contention. The relationship between resource contention frequency and system energy-saving performance is analyzed. A higher contention frequency indicates more intense competition for energy resources, requiring a larger priority degradation range to suppress conflicts and achieve load balancing and peak shaving. A mapping rule is established between contention frequency and degradation range, dividing contention frequency into three levels: Low-frequency contention (5%-8%), corresponding to a degradation level of 1; Medium-frequency contention (8%-12%), corresponding to a degradation level of 2; and High-frequency contention (above 12%), corresponding to a degradation level of 3. For node N5, a resource contention frequency of 8.9% falls under medium-frequency contention, and the priority degradation range is determined to be level 2. For node N8, a resource contention frequency of 12.5% ​​falls under high-frequency contention, and the priority degradation range is determined to be level 3. For node N12, a resource contention frequency of 6.7% falls under low-frequency contention, and the priority degradation range is determined to be level 1. The rationality of the degradation range is verified to ensure that the downgraded priority is not lower than the minimum priority allowed by the system. The system priority is divided into 6 levels from P0 to P5. Node N5 originally had a priority of P1, and after being downgraded 2 levels, it became P3.

[0050] Suppression parameters are generated based on the priority downgrade magnitude. For each high-intensity conflict node, its node identifier, current priority, and priority downgrade magnitude are extracted. The new priority after downgrading is calculated, which equals the current priority value plus the downgrade magnitude. Node N5's current priority P1 corresponds to a value of 1, and its downgrade magnitude is 2; the new priority is P3, corresponding to a value of 3. The rationality of the downgraded priority is verified to ensure that the new priority does not exceed the system-defined priority range. The system priority is divided into 6 levels, from P0 to P5, and the new priority P3 is within a reasonable range. For nodes whose downgrade magnitude is too large, causing the new priority to exceed P5, the new priority is limited to P5. A suppression parameter structure is constructed, which includes the node identifier, original priority, new priority, downgrade magnitude, and suppression effective time. The suppression effective time refers to the start time of the priority downgrade measure, usually set to the next cycle of the scheduling policy update. For node N5, the suppression parameter is {node_id:N5,original_priority:P1,new_priority:P3,downgrade_level:2,effective_time:T+100ms}. For node N8, the suppression parameters are {node_id:N8,original_priority:P2,new_priority:P5,downgrade_level:3,effective_time:T+100ms}. The suppression parameters of all high-intensity conflict nodes are integrated into a suppression parameter set, which provides a basis for priority adjustment in the allocation of exclusive scheduling domains.

[0051] Exclusive scheduling domains are determined through mutual exclusion partitioning using intensity identification parameters and suppression parameters. Intensity identification and suppression parameters are applied to classify all arbitration key points. High-intensity conflict nodes and low-intensity conflict nodes are identified based on the intensity identification parameters. Degraded nodes and non-degraded nodes are identified based on the suppression parameters. Suppression parameters are applied to high-intensity conflict nodes, lowering their scheduling priority and delaying their processing during energy resource allocation, thus achieving peak shaving and valley filling. Low-intensity conflict nodes retain their original priority, allowing them to participate normally in energy competition. Mutually exclusive partitioning is performed based on priority differences, assigning downgraded high-intensity nodes to the restricted scheduling domain and non-degraded low-intensity nodes to the priority scheduling domain. Nodes in the restricted scheduling domain are placed at the back of the queue during energy resource allocation and are only processed after nodes in the priority scheduling domain have completed their energy allocation. For nodes requiring exclusive energy resources, exclusivity analysis is performed based on their energy type and energy consumption scenario. When a node belongs to a critical guarantee scenario, the node and the energy resources it manages are assigned to the exclusive scheduling domain. Nodes within an exclusive scheduling domain exclusively utilize specific energy resources during execution, without sharing them with other nodes, ensuring the reliability of energy supply in critical scenarios. In a medical building scenario, the power supply nodes for operating rooms and ICUs are assigned to exclusive scheduling domains, exclusively controlling power distribution capacity and backup power. All arbitration critical points are categorized into priority scheduling domains, restricted scheduling domains, and exclusive scheduling domains, resulting in a three-domain partitioning outcome.

[0052] Step S130: Perform an influence tracing assessment on the exclusive scheduling domain to generate tracing weights. Based on the tracing weights, locate the core intelligent agent and the auxiliary intelligent agent in the exclusive scheduling domain. Extract the load surplus / deficit of the core intelligent agent and the compensation capability of the auxiliary intelligent agent to generate complementary pairing relationships. Based on the complementary pairing relationships, perform cross-allocation of resources to construct a scheduling topology table.

[0053] Specifically, an impact tracing assessment is performed on the exclusive scheduling domain to generate tracing weights. All agent nodes within the exclusive scheduling domain are extracted, and the position and role of each node in the energy supply chain are analyzed. For each agent in the exclusive scheduling domain, its impact on the energy supply of downstream agents and its dependence on the energy demand of upstream agents are traced. The number of downstream affected nodes of an agent is counted, representing the scope of its energy supply impact. For a data center power supply agent, a power outage directly affects three types of downstream equipment: server clusters, cooling systems, and network equipment; the number of downstream affected nodes is 3. The total energy supply of the agent is calculated, reflecting its energy contribution to the system. In a city-level heating scenario, the daily heat supply of a combined heat and power (CHP) agent reaches 50,000 GJ, becoming the main energy source for regional heating. The failure propagation risk of the agent is analyzed; when the agent experiences an energy supply outage, the impact on the entire system is assessed. An outage of a hospital power supply agent would render critical medical facilities such as operating rooms and ICUs inoperable, resulting in a high failure propagation risk rating. Integrating three dimensions—the number of downstream impact nodes, total energy supply, and fault propagation risk—an influence score is calculated for each agent. The influence score is then normalized to generate a source tracing weight, with a value ranging from 0 to 1. A higher value indicates greater influence of the agent within the exclusive scheduling domain.

[0054] Core and auxiliary agents are located from the exclusive scheduling domain based on their source weights. The source weights of all agents in the exclusive scheduling domain are extracted and sorted in descending order. A core agent threshold is set, typically the 60th quantile of the source weight distribution. For an exclusive scheduling domain containing 50 agents, the threshold corresponds to the 30th ranked value after sorting the source weights. Agents with source weights exceeding the core agent threshold are identified and marked as core agents. Core agents are those with high energy demand, high supply importance, and wide impact in the system, typically including energy consumption agents for commercial complexes, power supply agents for data centers, power supply agents for hospitals, and power supply agents for rail transit. In a smart city, the source weight of a commercial complex is 0.85, the data center's is 0.78, and the hospital's is 0.82, all exceeding the threshold of 0.65, and are therefore identified as core agents. Agents with source weights below the core agent threshold are identified and marked as auxiliary agents. Auxiliary agents are those in the system with relatively flexible energy demand and capable of participating in load regulation. They typically include general office building agents, energy storage device agents, waste heat recovery agents, and interruptible load agents. Office buildings are identified as auxiliary agents because their traceability weight is 0.45, and energy storage devices are identified as auxiliary agents because their traceability weight is 0.38.

[0055] In some embodiments, the step of extracting the load surplus / deficit of the core agent and the compensation capability of the auxiliary agent to generate a complementary pairing relationship includes: quantifying the resource requirements of the core agent to obtain a load demand vector; extracting idle resource capacity from the auxiliary agent to generate a compensation capability reserve; evaluating the matching degree between the load demand vector and the compensation capability reserve to obtain a pairing score; and establishing a complementary pairing relationship based on the pairing score.

[0056] Resource requirements are quantified for core agents to obtain load demand vectors. For each core agent, its energy resource requirements at the current time and in future time periods are extracted. Energy resource requirements include various energy types such as electricity, heat, cooling, and gas. Analyzing electricity demand, the total electricity demand of the commercial complex is 3000kW, of which 2000kW is uninterruptible load. The energy gap of the core agents is calculated, equal to the demand minus the current supply. The power gap of the commercial complex at 14:00 is 500kW, and the predicted power gap at 16:00 is 600kW. Data center power demand remains high on weekdays and decreases at night. The power gap, heat gap, time period information, and priority are integrated into a load demand vector, which is [power gap, heat gap, time period identifier, priority]. The load demand vector of the commercial complex at 14:00 is [500kW, 0GJ, T14, P2].

[0057] Idle resource capacity is extracted from auxiliary agents to generate compensation capacity reserves. For each auxiliary agent, its current energy resource usage status and adjustable space are analyzed. Idle energy capacity of the auxiliary agent is extracted, which refers to the amount of energy resources that the agent is currently not using or can release. The energy storage device currently stores 800kWh of electricity, has a maximum discharge power of 500kW, and has an idle energy capacity of 800kWh. The load reduction capability of the auxiliary agent is analyzed, which refers to the amount of energy demand that the agent can reduce through optimized control. The office building can reduce the power load by 230kW by adjusting the air conditioning and lighting control. The waste heat recovery capability of the auxiliary agent is extracted, which refers to the recoverable heat energy generated by industrial equipment or refrigeration equipment. The production equipment in the industrial park generates 500GJ of waste heat per hour, of which 300GJ is recoverable. The energy resource types and supply durations of the auxiliary agents are statistically analyzed. The energy storage device can discharge continuously for 1.6 hours, and the office building can reduce its load continuously for 4 hours. By integrating idle energy capacity, load reduction capacity, waste heat recovery capacity, and supply duration, a compensation capacity reserve vector is generated, which is [electricity compensation, heat compensation, and sustainable duration].

[0058] The matching score is obtained by evaluating the matching degree between the load demand vector and the compensation capacity reserve. The load demand vector of the core agent and the compensation capacity reserve vector of the auxiliary agent are extracted, and a dimension-by-dimensional matching analysis is performed. For the power dimension, the matching degree is calculated as follows: when the power compensation capacity of the auxiliary agent is greater than or equal to the power gap of the core agent, the power matching degree is 1; otherwise, the matching degree is the ratio of compensation capacity to gap. For the thermal dimension, the matching degree is calculated as follows: when the thermal compensation capacity is greater than or equal to the thermal gap, the thermal matching degree is 1; otherwise, it is the ratio of compensation capacity to gap. For the time dimension, the matching degree is calculated as follows: when the sustainable supply duration of the auxiliary agent is greater than or equal to the demand duration of the core agent, the time matching degree is 1; otherwise, it is the ratio of supply duration to demand duration. For geographical distance, the matching degree is calculated as follows: the closer the auxiliary agent and the core agent are, the smaller the transmission loss and the higher the matching degree. The distance matching degree is calculated using an exponential decay function. The overall matching degree is calculated as the pairing score. The pairing score formula is S=w_e·M_e+w_h·M_h+w_t·M_t+w_d·M_d, where S is the pairing score, M_e, M_h, M_t, and M_d are the matching degrees of the four dimensions of electricity, heat, time, and distance, respectively, and w_e, w_h, w_t, and w_d are the corresponding weights, and the sum of the weights is 1.

[0059] Complementary pairing relationships are established based on pairing scores. The pairing scores of all core agents and auxiliary agents are ranked, and agent pairs with pairing scores exceeding a threshold are identified. A pairing score threshold of 0.6 is set; when a pairing score is greater than or equal to 0.6, the core agent and auxiliary agent are considered to have a complementary pairing relationship. For example, the pairing score between a commercial complex and an energy storage device is 0.65, exceeding the threshold, thus establishing a complementary pairing relationship. Similarly, the pairing score between a data center and an office building is 0.72, exceeding the threshold, also establishing a complementary pairing relationship. The pairing score between a hospital and waste heat recovery in an industrial park is 0.58, not exceeding the threshold, so no pairing relationship is established. In cases where a core agent may match multiple auxiliary agents, pairing relationships are established by prioritizing the auxiliary agent with the highest score, ranked from highest to lowest. For instance, the pairing score between a commercial complex and an energy storage device is 0.65, with office building A having a pairing score of 0.62 and office building B having a pairing score of 0.58; therefore, the commercial complex is prioritized for pairing with the energy storage device. In cases where an auxiliary agent may be matched with multiple core agents, allocation is based on the priority of the core agents and the size of the load gap. The total compensation capacity of the energy storage equipment is 500kW. The commercial complex requires 500kW (priority P2), and the hospital requires 300kW (priority P1). The hospital's 300kW requirement is prioritized, and the remaining 200kW is allocated to the commercial complex. All established complementary pairings are compiled into a complementary pairing list.

[0060] A scheduling topology table is constructed based on complementary pairing relationships for cross-allocation of resources. The complementary pairing list is extracted, and for each pairing, the transmission path and allocation scheme of energy resources are analyzed. The transmission link of energy resources from the auxiliary agent to the core agent is determined, including power distribution lines, heat pipe networks, or virtual energy trading channels. The energy storage device is connected to the power distribution bus of the commercial complex via distribution cabinet DC01, with the transmission link being energy storage device → DC01 → commercial complex power distribution bus. The cross-allocation amount of energy resources is calculated, determined based on the load gap of the core agent and the compensation capability of the auxiliary agent. The energy storage device allocates 500kW of power to the commercial complex for 1 hour, with a cross-allocation of 500kWh. The office building provides indirect power support to the data center by reducing the load by 230kW, with a cross-allocation of 230kW. The industrial park supplies 300GJ of heat to the hospital through a waste heat recovery network, with a cross-allocation of 300GJ of heat. The execution time and priority of energy resource cross-allocation are set, with the execution time aligned with the occurrence time of the load gap of the core agent, and the priority set according to the importance of the core agent. The commercial complex's electricity compensation is executed from 14:00 to 15:00, with priority P2. The hospital's heating compensation is executed throughout the day, with priority P1. Construct a scheduling topology table, which includes fields such as source agent, target agent, energy type, allocation amount, transmission path, execution time, and priority. Each row corresponds to one energy cross-allocation instruction.

[0061] Step S140: Perform negotiation protocol cross-parsing on the scheduling topology table to construct cross-decision nodes. Extract the response delay difference of the auxiliary agents from the cross-decision nodes to generate delay compensation parameters. Use the delay compensation parameters to pre-schedule the core agents to generate response delay spectra.

[0062] Specifically, the scheduling topology table is parsed using negotiation protocols to construct cross-decision nodes. The scheduling topology table is loaded, and the execution dependencies of each energy cross-allocation instruction are analyzed. Cross-related energy allocation instructions in the scheduling topology table are identified; when multiple instructions involve the same energy resources or the same agent, a cross-relationship is determined. The negotiation protocol information of the cross-related instructions is extracted, including priority rules and timing constraints for energy allocation. In a city-level power grid dispatch scenario, energy storage devices simultaneously provide power compensation to a commercial complex and a hospital. Two allocation instructions are cross-related, requiring priority protocols to first satisfy the hospital's P1 level needs before satisfying the commercial complex's P2 level needs. The timing constraints in the negotiation protocol are analyzed; when the execution of one instruction depends on the completion of another, a timing dependency is established. Load reduction in office buildings needs to be executed 5 minutes before the data center load increases, forming a timing constraint. Cross-decision nodes are created at the cross-relationship locations in the scheduling topology table. These nodes are responsible for coordinating the execution order and resource allocation ratios of multiple instructions.

[0063] Response latency differences of auxiliary agents are extracted at cross-decision nodes to generate latency compensation parameters. For each cross-decision node, the response characteristics of the involved auxiliary agents are analyzed. The time delay from receiving the scheduling command to actually executing energy compensation for the auxiliary agents associated with the cross-decision node is statistically analyzed; this delay is called response latency. The response latency of energy storage equipment from receiving the discharge command to starting to discharge is 10 seconds; the response latency of office buildings from receiving the load reduction command to completing the air conditioning temperature adjustment is 180 seconds; and the response latency of industrial parks from receiving the waste heat recovery command to heat output is 300 seconds. The response latency differences between different auxiliary agents in the cross-decision nodes are identified; these differences reflect inconsistencies in the agents' execution speed. The impact of response latency differences on the energy supply of the core agent is analyzed. When the response latency of the auxiliary agents is too long, the core agent may experience insufficient energy supply during the waiting period. A commercial complex experiences a 500kW power shortage at 14:00; if it waits for the office buildings to complete the 180-second load reduction, there will be a continuous power shortage during this period. The delay compensation parameters are constructed by integrating the identifiers and response delay values ​​of each auxiliary agent, and the delay compensation parameters of all cross-decision nodes in the system are summarized to form a delay compensation parameter set.

[0064] In some embodiments, the step of using the delay compensation parameter to pre-schedule the core agent to generate a response delay spectrum includes: determining the pre-schedule time window of the core agent based on the delay compensation parameter; shifting the execution time of the core agent forward by the pre-schedule time window to generate a shifted execution sequence; performing time-series feature analysis on the shifted execution sequence to generate time-series response features; and generating a response delay spectrum based on the time-series response features.

[0065] For example, determining the advance scheduling time window of the core intelligent agent based on the delay compensation parameter includes: performing quantile analysis on the delay compensation parameter to obtain multi-level delay thresholds; extracting historical load characteristics of the core intelligent agent to generate a load fluctuation coefficient; constructing an adaptive safety margin based on the multi-level delay thresholds and the load fluctuation coefficient; and fusing the adaptive safety margin with the multi-level delay thresholds to determine the advance scheduling time window of the core intelligent agent.

[0066] Quantile analysis was performed on the delay compensation parameters to obtain multi-level delay thresholds. Based on the delay compensation parameters, response delay data of all auxiliary agents in the system were extracted to form a response delay dataset. Statistical analysis was performed on the response delay dataset to calculate the 25th, 50th, 75th, and 90th quantiles, which correspond to four levels: fast response, medium response, slow response, and extremely slow response, respectively. In a system containing 100 auxiliary agents, the 25th quantile of response delay is 30 seconds, the 50th quantile is 120 seconds, the 75th quantile is 240 seconds, and the 90th quantile is 420 seconds. Each quantile was set as a multi-level delay threshold, with the thresholds increasing from 30 seconds to 420 seconds. For each auxiliary agent, a corresponding delay threshold was matched based on its response delay value, with the matching rule being to select the smallest threshold greater than or equal to the response delay. The energy storage device's response delay is 10 seconds, less than all thresholds, corresponding to the first-level delay threshold of 30 seconds. The response latency for office buildings is 180 seconds, falling between 120 and 240 seconds, corresponding to the third-tier latency threshold of 240 seconds. The response latency for industrial parks is 300 seconds, falling between 240 and 420 seconds, corresponding to the fourth-tier latency threshold of 420 seconds. These multi-tiered latency thresholds provide graded latency benchmarks for auxiliary agents with different response speeds.

[0067] Historical load characteristics of the core agent are extracted to generate a load fluctuation coefficient. The energy load variation patterns of the core agent over historical periods are analyzed to extract the load fluctuation amplitude and frequency. The load fluctuation amplitude is calculated as the ratio of the difference between the maximum and minimum load values ​​to the average load. The average daily power load of the commercial complex is 2000kW, the peak load is 3000kW, and the valley load is 1200kW, with a load fluctuation amplitude of 0.9. The load fluctuation frequency is calculated as the number of significant load changes per unit time. The commercial complex experienced 8 load changes exceeding 200kW within one hour, resulting in a load fluctuation frequency of 8 times / hour. The load fluctuation coefficient is generated by weighting the normalized fluctuation amplitude and frequency, with a weight of 0.6 for amplitude and 0.4 for frequency. The load fluctuation coefficient for the commercial complex is 0.86, while that for the hospital is 0.35.

[0068] An adaptive safety margin is constructed based on multi-level delay thresholds and load fluctuation coefficients. The multi-level delay thresholds corresponding to the auxiliary agent and the load fluctuation coefficient of the core agent are extracted. An adaptive safety margin calculation model is constructed, and the adaptive safety margin is dynamically adjusted according to the multi-level delay thresholds and load fluctuation coefficients. The adaptive safety margin formula is T_s=T_th·(1+β·C_f), where T_s is the adaptive safety margin, T_th is the corresponding delay threshold in the multi-level delay thresholds, C_f is the load fluctuation coefficient, and β is the adjustment coefficient, typically ranging from 0.3 to 0.5. For a scenario where an office building (multi-level delay threshold corresponding to 120 seconds) provides compensation to a commercial complex (load fluctuation coefficient 0.86), the adjustment coefficient is taken as 0.4, and the adaptive safety margin is 161 seconds. For a scenario where an industrial park (multi-level delay threshold corresponding to 240 seconds) provides thermal compensation to a hospital (load fluctuation coefficient 0.35), the adjustment coefficient is taken as 0.3, and the adaptive safety margin is 265 seconds. The adaptive safety margin reflects the advance scheduling buffer time after taking into account the uncertainty of load fluctuations.

[0069] The advance scheduling time window of the core agent is determined by fusing adaptive safety margin and multi-level delay thresholds. The actual response delay and adaptive safety margin of the auxiliary agent are extracted. The advance scheduling time window is calculated using the formula T_w = T_d + T_s, where T_w is the advance scheduling time window, T_d is the actual response delay of the auxiliary agent, and T_s is the adaptive safety margin. For providing compensation to commercial complexes from office buildings (response delay 180 seconds, adaptive safety margin 161 seconds), the advance scheduling time window is 341 seconds. For providing heating compensation to hospitals from industrial parks (response delay 300 seconds, adaptive safety margin 265 seconds), the advance scheduling time window is 565 seconds. The rationality of the advance scheduling time window is verified, ensuring that the time window is not less than the response delay of the auxiliary agent and not greater than the maximum acceptable advance time of the core agent. The maximum acceptable advance time for commercial complexes is 600 seconds, and the 341-second advance scheduling time window is within a reasonable range. Multi-level delay thresholds are integrated into the determination process of the advance scheduling time window through the calculation of adaptive safety margins.

[0070] The execution times of the core agents are shifted forward by an advance scheduling time window to generate the offset execution sequence. The original scheduled execution times and corresponding advance scheduling time windows of all core agents are extracted. For each core agent, the execution time offset is calculated one by one. The formula for the offset execution time is T_e' = T_e - T_w, where T_e' is the offset execution time, T_e is the original scheduled execution time, and T_w is the advance scheduling time window. For example, the commercial complex was originally scheduled to perform load adjustment at 14:00:00, with an advance scheduling time window of 341 seconds, and the offset execution time is 13:54:19. The hospital was originally scheduled to initiate heating demand at 15:00:00, with an advance scheduling time window of 565 seconds, and the offset execution time is 14:50:35. The data center was originally scheduled to request power at 16:00:00, with an advance scheduling time window of 280 seconds, and the offset execution time is 15:55:20. The execution time of the auxiliary agents involved is synchronously offset. The offset time of the auxiliary agents is calculated based on their response delay to ensure that the energy compensation of the auxiliary agents arrives before the core agents' needs. For example, for an office building paired with a commercial complex, with a response delay of 180 seconds, the offset execution time is 13:54:19, and the actual energy arrival time is 13:57:19, meeting the requirement of arriving before 14:00:00. For an industrial park paired with a hospital, with a response delay of 300 seconds, the offset execution time is 14:50:35, and the actual heat arrival time is 14:55:35, meeting the requirement of arriving before 15:00:00. The rationality of the offset execution time is verified to ensure that the offset time is not earlier than the current system time and not later than the originally scheduled execution time. The offset execution times of all core agents and auxiliary agents are arranged in chronological order, and the agent identifier, execution time, and priority are recorded to form the offset execution sequence.

[0071] Timing feature analysis was performed on the post-offset execution sequence to generate timing response features. The distribution of execution times for each agent in the post-offset execution sequence was analyzed. The total response time from the issuance of the scheduling command to the actual arrival of energy was statistically analyzed for each agent, including command transmission time, agent processing time, and energy transmission time. The commercial complex issued a request at 13:54:19, the office building completed load reduction at 13:57:19, and electricity arrived at the commercial complex at 13:57:29, with a total response time of 190 seconds. The density features of the post-offset execution sequence were analyzed to identify groups of agents executing concentratedly within a short period. Between 13:54:00 and 13:55:00, 5 core agents and 8 auxiliary agents simultaneously executed scheduling commands, forming a dense execution segment. The timing interval features of the post-offset execution sequence were extracted, and the average and minimum intervals between the execution times of adjacent agents were calculated. The average execution interval of the post-offset execution sequence was 45 seconds, and the minimum execution interval was 8 seconds. The priority distribution characteristics of the execution sequence after offset were analyzed, and the dispersion of execution times for agents of different priorities was statistically analyzed. The execution times of P1-level agents were distributed between 13:50:00 and 13:55:00, P2-level agents between 13:54:00 and 13:58:00, and P3-level agents between 13:56:00 and 14:00:00. The temporal response characteristics encompassed four dimensions: total response time, density, temporal interval, and priority distribution, reflecting the overall temporal distribution pattern of the execution sequence.

[0072] A response delay spectrum is generated based on temporal response characteristics. Temporal response characteristics reflect the overall temporal distribution of the execution sequence after offset. The identifiers and response delay data of each auxiliary agent are extracted from these characteristics, and the distribution patterns of response delays for different types of auxiliary agents are analyzed. Auxiliary agents are classified and statistically analyzed according to their response delays, forming a delay distribution histogram. Energy storage agents have short response delays, concentrated in the 10-30 second range; load shedding agents have moderate response delays, distributed in the 120-200 second range; waste heat recovery agents have longer response delays, distributed in the 280-350 second range; and interruptible load agents have moderate response delays, distributed in the 80-150 second range. The average response delays for each type of agent are calculated: energy storage agents average 18 seconds, load shedding agents average 165 seconds, waste heat recovery agents average 310 seconds, and interruptible load agents average 115 seconds. A response delay spectrum chart is constructed, with the horizontal axis representing the auxiliary agent type and the vertical axis representing the response delay duration. The chart labels the delay distribution range, average value, and standard deviation for each type of agent. The response delay spectrum records the specific response delay values ​​for each auxiliary agent in detail. The response delay of the energy storage device is 10 seconds, the response delay of the office building air conditioning system is 180 seconds, the response delay of the industrial park waste heat recovery system is 300 seconds, and the response delay of the industrial load interruptible system is 120 seconds.

[0073] Step S150: The response delay spectrum is analyzed by back-calculation of scheduling time to generate a time base sequence. The execution window of each agent is determined based on the time base sequence. Idle time segments in the execution window are identified and dynamically compressed to generate a window scheduling table. A resource allocation scheme is generated by time-series interleaving based on the window scheduling table.

[0074] Specifically, the response delay spectrum is analyzed using a time-based timeline to generate a time reference sequence. The response delay spectrum is loaded, and the response delay distribution data for various auxiliary agents are extracted. For each core agent's energy demand time, a timeline calculation is performed based on the response delays of its paired auxiliary agents in the response delay spectrum. This timeline calculation involves working backward from the core agent's energy demand time to determine when the auxiliary agent should start. For example, a commercial complex needs power compensation at 14:00, and its paired office building has a response delay of 180 seconds; therefore, the office building should initiate load reduction at 13:57:00. A hospital needs heat supply at 15:00, and its paired industrial park has a response delay of 300 seconds; therefore, the industrial park should initiate waste heat recovery at 14:55:00. A timeline analysis is performed on the pairing relationships of all core and auxiliary agents in the system to calculate the latest start time for each agent. The latest start time refers to the time before which the agent must start execution to meet energy supply timeliness requirements. The data center requires power at 16:00, and the paired energy storage device has a response delay of 10 seconds, with the latest startup time being 15:59:50. Arrange the latest startup times of all agents in chronological order to form a time-based sequence.

[0075] The execution window for each agent is determined based on the time base sequence. The latest start time for each agent is extracted from the time base sequence. An execution window is set for each agent, referring to the time range within which the agent can initiate and execute tasks. The execution window includes the earliest and latest start times. The latest start time is given by the time base sequence, and the earliest start time is determined based on the agent's ability to execute tasks ahead of time. For the office building, the latest start time is 13:57:00. Considering that its air conditioning system can pre-cool 30 minutes in advance, the earliest start time is 13:27:00, and the execution window is 13:27:00-13:57:00. The latest start time for the energy storage device is 15:59:50. The energy storage device can start discharging at any time, and the execution window can be advanced to 2 hours, making the execution window 13:59:50-15:59:50. The latest start time for waste heat recovery in the industrial park is 14:55:00. Waste heat recovery requires the operation of production equipment, and the earliest start time, constrained by the production plan, is 14:00:00. The execution window is 14:00:00-14:55:00. The overlapping relationships of the execution windows of each agent are analyzed to identify agent groups that can undergo time-series coordination. The execution windows of the commercial complex, data center, and hospital overlap within the time period of 13:50:00-14:00:00, allowing for staggered time-series scheduling.

[0076] In some embodiments, the step of identifying idle time segments in the execution window and dynamically compressing them to generate a window scheduling table includes: performing a bidirectional time axis scan of the execution window to locate idle time segments; identifying reverse filling opportunities from the idle time segments to generate a filling candidate set; performing elastic boundary shrinking of the filling candidate set to generate a compact execution window; and generating a window scheduling table based on the compact execution window.

[0077] A bidirectional timeline scan is performed on the execution window to locate idle time segments. A timeline representation is established for the execution window of each agent, marking the time periods for scheduled tasks and idle time periods. A bidirectional timeline scan is performed on the execution window to identify the start and end points of the window and the idle time segments between tasks. The execution window for the office building starts at 13:27:00, the first task starts at 13:50:00, and the idle time segment from 13:27:00 to 13:50:00 is identified. The execution window for the energy storage device ends at 15:59:50, the last task ends at 15:30:00, and the idle time segment from 15:30:00 to 15:59:50 is identified. The industrial park performs waste heat recovery tasks from 14:10:00 to 14:25:00 and a second waste heat recovery task from 14:40:00 to 14:55:00, and the idle time segment from 14:25:00 to 14:40:00 is identified. Count the number of idle time segments and the total idle time of each agent's execution window.

[0078] Reverse filling opportunities are identified from idle time segments to generate a candidate set for filling. The overlap between idle time segments and the execution windows of other agents is analyzed. When an agent's idle time segment overlaps with another agent's execution window, it is identified as a reverse filling opportunity. Reverse filling refers to moving tasks originally scheduled for the latter part of the execution window to the earlier part of the idle time segment. The idle time segment of the office building from 13:27:00 to 13:50:00 partially overlaps with the energy demand period of the commercial complex from 13:40:00 to 14:00:00, identifying it as a reverse filling opportunity. The feasibility of reverse filling is assessed, considering the energy compensation capacity of the agents and the matching degree of the core agent's energy demand. The office building can provide 230kW load reduction, and the commercial complex needs 200kW of power compensation at 13:40:00; the capabilities match, and reverse filling is feasible. Feasible reverse filling opportunities are prioritized, with opportunities having higher energy demand urgency being filled first. Reverse filling opportunities for P1-level demand in hospitals ranked first, followed by reverse filling opportunities for P2-level demand in commercial complexes. All identified reverse filling opportunities were compiled into a filling candidate set, which included information such as the source agent, target agent, filling time period, and filling energy amount.

[0079] A compact execution window is generated by elastically shrinking the candidate set of fillers. Each filler record in the candidate set is extracted, and the execution window of the source agent is shrunk. Boundary shrinking refers to reducing the execution window range of the agent based on the reverse filling arrangement of the candidate set, making task execution more compact. For example, the original execution window for the office building was 13:27:00-13:57:00. Under the arrangement of the candidate set, the load reduction task at 13:40:00 is executed earlier. The front boundary of the execution window remains unchanged at 13:27:00, while the back boundary shrinks to 13:50:00, resulting in a new execution window of 13:27:00-13:50:00. Similarly, the original execution window for the energy storage device was 13:59:50-15:59:50. After filling multiple charging scheduling tasks, the front boundary shrinks to 14:30:00, while the back boundary remains unchanged at 15:59:50, resulting in a new execution window of 14:30:00-15:59:50. Analyzing the changes in execution window width after boundary shrinkage reveals that moderate boundary shrinkage can improve window utilization, while excessive shrinkage reduces the system's ability to handle unexpected events. The shrunken execution windows are then organized into a compact set.

[0080] A window scheduling table is generated based on compact execution windows. The compact execution windows of each agent are extracted, and a window scheduling table is constructed by combining this with the task arrangements used to fill the candidate set. The window scheduling table records the specific execution plan of each agent within the compact execution window. The window scheduling table includes fields such as agent identifier, start and end time of the compact execution window, list of scheduled tasks, task execution time, and task duration. The window scheduling table record for office buildings is as follows: {Agent: Office Building A, Window: 13:27:00-13:50:00, Task: Load Reduction, Execution Time: 13:40:00, Duration: 10 minutes}. The window scheduling table record for energy storage devices is as follows: {Agent: Energy Storage Device B, Window: 14:30:00-15:59:50, Task: Discharge × 3, Execution Time: 14:30:00 / 15:15:00 / 15:45:00, Duration: 15 minutes × 3}. The overall temporal distribution characteristics of the window scheduling table are analyzed, and the number of concurrent executions by agents in each time period is statistically analyzed. At 14:00:00, eight agents were executing simultaneously, approaching the system's concurrent processing limit. Some tasks need to be rescheduled to execute after 14:30:00. The window scheduling table needs load balancing optimization to prevent too many agents from executing simultaneously during a particular time period, which could overload the energy system.

[0081] A resource allocation scheme is generated by staggering the execution times of tasks based on the window scheduling table. The window scheduling table is loaded, and the task execution time distribution of each agent is analyzed. Agent groups with overlapping task execution times are identified in the window scheduling table. When multiple agents execute tasks at the same or similar times, energy resource competition may occur. At 14:00:00, three core agents—a commercial complex, a data center, and a hospital—simultaneously require power compensation. Correspondingly, office buildings A and B, and energy storage device C, execute tasks simultaneously, resulting in overlapping execution times. The overlapping task execution times are staggered, meaning tasks that were originally executed simultaneously are moved to different times to reduce instantaneous peak energy demand. The load reduction time for office building A is adjusted from 14:00:00 to 13:55:00, office building B remains at 14:00:00, and energy storage device C is adjusted to 14:05:00, with the three execution times staggered by 5 minutes. The load curve after time-series staggering is calculated. Before staggering, the peak instantaneous power demand at 14:00:00 is 2500kW, and after staggering, the peak value drops to 2000kW. Based on the window scheduling table execution arrangement after time-series staggering, a resource allocation scheme is formulated. The resource allocation scheme includes the type, quantity, and duration of energy resources obtained by each agent at a specific time. The commercial complex receives 230kW of power compensation from office building A for 15 minutes at 13:55:00, the data center receives 200kW of power compensation from office building B for 20 minutes at 14:00:00, and the hospital receives 500kW of power compensation from energy storage device C for 15 minutes at 14:05:00. Through time-series staggering optimization, the resource allocation scheme achieves peak shaving and valley filling of energy demand and efficient collaborative scheduling among multiple agents.

[0082] To implement the city-level intelligent agent resource collaborative scheduling method corresponding to the above method embodiments, in order to achieve the corresponding functions and technical effects. See also Figure 2 , Figure 2 This diagram illustrates a structural block diagram of a city-level intelligent agent resource collaborative scheduling system 200 provided in an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The city-level intelligent agent resource collaborative scheduling system 200 provided in this embodiment includes:

[0083] Data acquisition module 201 is used to collect multi-agent interaction data during the city-level energy conservation and carbon reduction operation, and to perform conflict feature mining on the interaction data to construct a resource conflict relationship graph;

[0084] The conflict resolution module 202 is used to perform hierarchical resolution based on the resource conflict relationship diagram to generate arbitration key points, extract conflict intensity features from the arbitration key points to generate intensity identifier parameters, identify high-intensity conflict nodes in the intensity identifier parameters to perform priority reverse degradation to generate suppression parameters, and use the intensity identifier parameters and the suppression parameters to perform mutual exclusion partitioning to determine the exclusive scheduling domain.

[0085] The agent matching module 203 is used to perform influence tracing assessment on the exclusive scheduling domain to generate tracing weights, locate the core agent and the auxiliary agent from the exclusive scheduling domain based on the tracing weights, extract the load surplus and deficit of the core agent and the compensation capability of the auxiliary agent to generate a complementary pairing relationship, and perform cross-allocation of resources based on the complementary pairing relationship to construct a scheduling topology table.

[0086] The delay compensation module 204 is used to perform negotiation protocol cross-parsing on the scheduling topology table to construct cross-decision nodes, extract the response delay difference of the auxiliary agent from the cross-decision nodes to generate delay compensation parameters, and use the delay compensation parameters to pre-schedule the core agent to generate a response delay spectrum.

[0087] The window optimization module 205 is used to perform scheduling time back-analysis on the response delay spectrum to generate a time base sequence, determine the execution window of each agent based on the time base sequence, identify idle time segments in the execution window and dynamically compress them to generate a window scheduling table, and generate a resource allocation scheme by performing time-series interleaving based on the window scheduling table.

[0088] The aforementioned city-level intelligent agent resource collaborative scheduling system 200 can implement a city-level intelligent agent resource collaborative scheduling method according to the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0089] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0090] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A city-level intelligent agent resource collaborative scheduling method, characterized in that, include: Collect multi-agent interaction data during the city-level energy conservation and carbon reduction operation, and construct a resource conflict relationship graph by mining the conflict features of the interaction data; Based on the resource conflict relationship diagram, hierarchical analysis is performed to generate arbitration key points. Conflict intensity features are extracted from the arbitration key points to generate intensity identification parameters. High-intensity conflict nodes in the intensity identification parameters are identified and priority is reversed to generate suppression parameters. The intensity identification parameters and the suppression parameters are used to perform mutual exclusion partitioning to determine the exclusive scheduling domain. An influence tracing assessment is performed on the exclusive scheduling domain to generate tracing weights. Based on the tracing weights, the core intelligent agent and the auxiliary intelligent agent are located from the exclusive scheduling domain. The load surplus and deficit of the core intelligent agent and the compensation capability of the auxiliary intelligent agent are extracted to generate a complementary pairing relationship. Based on the complementary pairing relationship, resources are cross-allocated to construct a scheduling topology table. The scheduling topology table is cross-parsed using a negotiation protocol to construct cross-decision nodes. Response latency differences of the auxiliary agents are extracted from these cross-decision nodes to generate latency compensation parameters. These latency compensation parameters are then used to pre-schedule the core agent, generating a response latency spectrum. This process includes: determining the pre-schedule time window for the core agent based on the latency compensation parameters, including: performing quantile analysis on the latency compensation parameters to obtain multi-level latency thresholds; extracting historical load characteristics of the core agent to generate a load fluctuation coefficient; constructing an adaptive safety margin based on the multi-level latency thresholds and the load fluctuation coefficient; fusing the adaptive safety margin with the multi-level latency thresholds to determine the pre-schedule time window for the core agent; shifting the execution time of the core agent forward by the pre-schedule time window to generate a shifted execution sequence; performing time-series feature analysis on the shifted execution sequence to generate time-series response features; and generating a response latency spectrum based on the time-series response features. The response delay spectrum is analyzed by back-calculation of scheduling time to generate a time base sequence. The execution window of each agent is determined based on the time base sequence. Idle time segments in the execution window are identified and dynamically compressed to generate a window scheduling table. A resource allocation scheme is generated by time-series interleaving based on the window scheduling table.

2. The method according to claim 1, characterized in that, The step of mining conflict features from the interaction data to construct a resource conflict relationship graph includes: Resource access sequence extraction is performed on the interactive data; A set of competitive hotspots is generated by identifying frequently accessed resources from the resource access sequence; Extract the implicit dependencies between agents in the set of competitive hotspots to generate dependency transit chains; Construct a resource conflict relationship graph based on the dependency transitive chain.

3. The method according to claim 1, characterized in that, The generation of arbitration key points through hierarchical parsing based on the resource conflict relationship diagram includes: The resource conflict relationship diagram is decomposed into a main conflict layer and a branch conflict layer; Locate the conflict convergence node from the main conflict layer and the branch conflict layer; Tracing the location of the conflict source along the conflict convergence node; The key points of arbitration are determined based on the location of the source of the conflict.

4. The method according to claim 1, characterized in that, The process of identifying high-intensity conflict nodes in the intensity identifier parameters and performing priority reverse degradation to generate suppression parameters includes: High-intensity conflict nodes are identified by threshold filtering of the intensity identification parameters; Extract the resource contention frequency of the high-intensity conflict nodes; The priority downgrade magnitude is determined based on the frequency of resource contention. Suppression parameters are generated based on the priority downgrade magnitude.

5. The method according to claim 1, characterized in that, The step of extracting the load surplus / deficit of the core agent and the compensation capability of the auxiliary agent to generate a complementary pairing relationship includes: The resource requirements of the core intelligent agent are quantified to obtain the load requirement vector; Idle resource capacity is extracted from the auxiliary intelligent agent to generate a compensation capacity reserve; The matching score is obtained by evaluating the matching degree between the load demand vector and the compensation capacity reserve. Complementary pairings are established based on the pairing scores.

6. The method according to claim 1, characterized in that, The step of identifying idle time segments in the execution window and dynamically compressing them to generate a window scheduling table includes: Perform a bidirectional timeline scan of the execution window to locate idle time segments; Reverse filling opportunities are identified from the idle time segments to generate a filling candidate set; The candidate set for filling is elastically shrunk to generate a compact execution window; A window scheduling table is generated based on the compact execution window.

7. The method according to claim 3, characterized in that, Tracing the source location of the conflict along the conflict convergence node includes: Identify the multi-branch in-degree connection relationships of the conflict convergence nodes; Parallel reverse tracing is performed along the multi-branch in-degree connection relationship to generate a multi-path tracing set; The propagation intensity of the multi-path tracing set is assessed to extract the dominant propagation path; The source of the conflict is located based on the dominant propagation path.

8. A city-level intelligent agent resource collaborative scheduling system, characterized in that, include: The data acquisition module is used to collect interaction data of multiple agents during the city-level energy conservation and carbon reduction operation, and to mine conflict features from the interaction data to construct a resource conflict relationship graph. The conflict resolution module is used to perform hierarchical resolution based on the resource conflict relationship diagram to generate arbitration key points, extract conflict intensity features from the arbitration key points to generate intensity identifier parameters, identify high-intensity conflict nodes in the intensity identifier parameters to perform priority reverse degradation to generate suppression parameters, and use the intensity identifier parameters and the suppression parameters to perform mutual exclusion partitioning to determine the exclusive scheduling domain. The agent matching module is used to perform influence tracing assessment on the exclusive scheduling domain to generate tracing weights, locate the core agent and auxiliary agent from the exclusive scheduling domain based on the tracing weights, extract the load surplus / deficit of the core agent and the compensation capability of the auxiliary agent to generate a complementary pairing relationship, and perform cross-allocation of resources based on the complementary pairing relationship to construct a scheduling topology table. The delay compensation module is used to perform negotiation protocol cross-parsing on the scheduling topology table to construct cross-decision nodes, extract the response delay differences of the auxiliary agents from the cross-decision nodes to generate delay compensation parameters, and use the delay compensation parameters to pre-schedule the core agents to generate a response delay spectrum. This includes: determining the pre-schedule time window of the core agents based on the delay compensation parameters, including: performing quantile analysis on the delay compensation parameters to obtain multi-level delay thresholds; extracting historical load characteristics of the core agents to generate load fluctuation coefficients; constructing an adaptive safety margin based on the multi-level delay thresholds and the load fluctuation coefficients; fusing the adaptive safety margin and the multi-level delay thresholds to determine the pre-schedule time window of the core agents; shifting the execution time of the core agents forward by the pre-schedule time window to generate a shifted execution sequence; performing time-series feature analysis on the shifted execution sequence to generate time-series response features; and generating a response delay spectrum based on the time-series response features. The window optimization module is used to perform scheduling time back-analysis on the response delay spectrum to generate a time base sequence, determine the execution window of each agent based on the time base sequence, identify idle time segments in the execution window and dynamically compress them to generate a window scheduling table, and generate a resource allocation scheme by performing time-series interleaving based on the window scheduling table.

Images