Power grid dispatching method, device and equipment, storage medium and program product

Abstract

The application discloses a power grid dispatching method and device, equipment, a storage medium and a program product, and belongs to the technical field of power system dispatching automation. The method comprises the following steps: constructing an evolution knowledge graph representing risk semantic association based on extreme weather risk characteristics; performing multi-day rolling trend sensing based on the evolution knowledge graph to obtain a multi-day scene set containing risk scores, influence ranges and confidence levels; mapping the multi-day scene set into a feedforward control amount; combining the feedforward control amount, constructing and solving a multi-objective optimization problem to obtain a self-consistent feedforward strategy template; executing the self-consistent feedforward strategy template to obtain actual power grid response data and feeding back the actual power grid response data to the evolution knowledge graph to drive online evolution of the evolution knowledge graph. The embodiment of the application can improve the operation safety of the power grid under extreme weather and the robustness of power grid stability.

Description

Technical Field

[0001] This application relates to the field of power system dispatch automation technology, and in particular to a power grid dispatching method, device, equipment, storage medium and program product. Background Technology

[0002] With the continuous increase in the proportion of new energy sources such as wind power and photovoltaics in the power system, the inertia and adjustability of the power grid have significantly decreased, and its sensitivity to extreme weather such as typhoons, severe convection, and cold waves has increased dramatically. Traditional dispatching methods mainly rely on daily forecasts and post-event feedback, which are difficult to cope with complex disturbances lasting for several days. They also suffer from lag and mismatch risks in the coordination of reserve configuration and energy paths, which not only affect the safety and stability of the system but also reduce the economic efficiency of operation. Existing technologies often simplify risks into numerical disturbances or probabilistic scenarios, and their decision-making relies on manual analysis and has a lag in response, resulting in low security. Summary of the Invention

[0003] The purpose of this application is to provide a power grid dispatching method, apparatus, equipment, storage medium, and program product that can effectively improve the operational safety of the power grid under extreme weather conditions and the robustness of power grid stability.

[0004] To achieve the above objectives, a first aspect of this application provides a power grid dispatching method, comprising: Constructing an evolutionary knowledge graph representing semantic associations of risks based on the characteristics of extreme weather risks; Based on the evolutionary knowledge graph, multi-day rolling situational awareness is performed to obtain a multi-day scenario set including risk score, impact range and confidence level; Map the multi-day scene set to feedforward control variables; By combining the aforementioned feedforward control variables, a multi-objective optimization problem is constructed and solved to obtain a self-consistent feedforward strategy template. The self-consistent feedforward strategy template is executed to obtain actual power grid response data, and the actual power grid response data is fed back to the evolutionary knowledge graph to drive the online evolution of the evolutionary knowledge graph.

[0005] Compared with existing technologies, the power grid dispatching method provided in this application has the following advantages: By constructing an evolutionary knowledge graph that represents the semantic association of risks, a systematic semantic expression of power grid risks under extreme weather conditions is realized, providing unified knowledge support for multi-day risk perception; based on the evolutionary knowledge graph, multi-day rolling situational awareness is carried out to obtain a multi-day scenario set containing risk scores, impact ranges, and confidence levels, which can identify the distribution characteristics and impact boundaries of power grid risks in multiple time periods in advance, effectively solving the response lag problem caused by traditional dispatching relying on single-day predictions; mapping the multi-day scenario set to feedforward control variables and obtaining a self-consistent feedforward strategy template by constructing a multi-objective optimization problem can make the dispatching strategy adapt to multi-day risk changes and ensure the matching of various dispatching parameters, avoiding the situation of mismatch between reserve configuration and energy path; at the same time, the actual power grid response data is fed back to the evolutionary knowledge graph and driven to evolve online, forming a closed-loop dispatching mechanism of "perception-dispatch-feedback-optimization", which can continuously accumulate dispatching experience and optimize subsequent dispatching strategies, steadily improving the initiative, safety, and economy of power grid dispatching under high-risk scenarios, thereby improving the robustness of power grid stability.

[0006] In some embodiments, the construction of an evolutionary knowledge graph representing semantic associations of risks based on extreme weather risk characteristics includes: Based on the entity classes and relation classes pre-defined to characterize the characteristics of extreme weather risks, a graph structure is established to represent the relationships between weather events, risk types, power grid objects, and control strategies. The evolutionary knowledge graph is obtained by assigning the validity period attribute and confidence attribute to the relationships in the graph structure.

[0007] In some embodiments, the multi-day rolling situational awareness based on the evolutionary knowledge graph, resulting in a multi-day scenario set including risk scores, impact ranges, and confidence levels, includes: Meteorological forecast data and power grid operation measurement data covering a multi-day time range are input into the evolutionary knowledge graph. Based on the evolutionary knowledge graph and preset risk criteria, a risk score covering a multi-day time range is calculated. Based on the entity relationship chain in the evolutionary knowledge graph, the scope of the risk's impact is deduced. The confidence level of the risk is derived based on the reliability of the source of the input data; Based on risk scores, scope of impact, and confidence levels, a multi-day scenario set is formed.

[0008] In some embodiments, mapping the multi-day scene set to feedforward control variables includes: Based on the risk scores from the multi-day scenario cluster, the following are generated: backup demand correction, energy storage status reference trajectory correction, and system control parameter level switching suggestions. Based on the impact range of the multi-day scenario concentration, upper and lower envelope constraint parameters for new energy output are set; The feedforward control quantity is formed based on the backup demand correction amount, the energy storage state reference trajectory correction amount, the system control parameter level switching suggestion, and the upper and lower envelope constraint parameters.

[0009] In some embodiments, the step of combining the feedforward control variable to construct and solve a multi-objective optimization problem to obtain a self-consistent feedforward strategy template includes: Using feedforward control variables as boundary conditions, and with operating costs, safety penalties, and resilience indicators as objectives, a multi-objective optimization problem covering a future multi-day time range is constructed. The multi-objective evolutionary algorithm is used to solve the multi-objective optimization problem, and a self-consistent feedforward policy template distributed on the Pareto front is obtained.

[0010] In some embodiments, the multi-objective evolutionary algorithm is a multi-objective hoarfrost optimization algorithm, which generates an initial population by simulating a deposition operator, performs individual evolution by a dendrite growth operator, and performs constraint repair and population screening by an ice peeling operator.

[0011] To achieve the above objectives, a second aspect of this application provides a power grid dispatching device, the device comprising: The module is used to construct an evolutionary knowledge graph representing semantic associations of risks based on the characteristics of extreme weather risks; The perception module is used to perform multi-day rolling situational awareness based on the evolutionary knowledge graph, and obtain a multi-day scene set including risk score, impact range and confidence level; The mapping module is used to map the multi-day scene set into feedforward control variables; The solution module is used to combine the feedforward control quantity to construct and solve a multi-objective optimization problem and obtain a self-consistent feedforward strategy template. The feedback module is used to execute the self-consistent feedforward strategy template, obtain actual power grid response data, and feed the actual power grid response data back to the evolutionary knowledge graph to drive the online evolution of the evolutionary knowledge graph.

[0012] To achieve the above objectives, a third aspect of this application provides an electronic device, the electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the method described in the first aspect.

[0013] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium comprising a stored computer program, wherein the computer program, when executed, controls the device containing the computer-readable storage medium to perform the method described in the first aspect.

[0014] To achieve the above objectives, a fifth aspect of the present application provides a computer program product, which includes a computer program or computer instructions, wherein the computer program or computer instructions, when executed by a processor, implement the method described in the first aspect. Attached Figure Description

[0015] Figure 1 This is a flowchart of a power grid dispatching method provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of the evolutionary knowledge graph provided in an embodiment of this application; Figure 3 This is a schematic diagram of the power grid dispatching device provided in an embodiment of this application; Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

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

[0019] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0020] With the continuous increase in the proportion of new energy sources such as wind power and photovoltaics in the power system, the inertia and adjustability of the power grid have significantly decreased, and its sensitivity to extreme weather such as typhoons, severe convection, and cold waves has increased dramatically. Traditional dispatching methods mainly rely on daily forecasts and post-event feedback, which are difficult to cope with complex disturbances lasting for several days. They also suffer from lag and mismatch risks in the coordination of reserve configuration and energy paths, which not only affect the safety and stability of the system but also reduce the economic efficiency of operation. Existing technologies often simplify risks into numerical disturbances or probabilistic scenarios, and their decision-making relies on manual analysis and has a lag in response, resulting in low security.

[0021] Based on this, embodiments of this application provide a power grid dispatching method, apparatus, equipment, storage medium, and program product, which can effectively improve the operational safety of the power grid under extreme weather conditions and the robustness of power grid stability.

[0022] Please see Figure 1 , Figure 1 This is an optional flowchart of the power grid dispatching method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0023] Step S101: Construct an evolutionary knowledge graph representing the semantic associations of risks based on the characteristics of extreme weather risks; Step S102: Perform multi-day rolling situational awareness based on evolutionary knowledge graph to obtain a multi-day scenario set including risk score, impact range and confidence level; Step S103: Map the multi-day scene set to feedforward control variables; Step S104: Combine the feedforward control variable to construct and solve the multi-objective optimization problem, and obtain the self-consistent feedforward strategy template. Step S105: Execute the self-consistent feedforward strategy template to obtain actual power grid response data and feed the actual power grid response data back to the evolutionary knowledge graph to drive the online evolution of the evolutionary knowledge graph.

[0024] Steps S101 to S105 of this application embodiment, by constructing an evolutionary knowledge graph representing the semantic association of risks, realize a systematic semantic expression of power grid risks under extreme weather conditions, providing unified knowledge support for multi-day risk perception; based on the evolutionary knowledge graph, multi-day rolling situational awareness is carried out to obtain a multi-day scenario set including risk scores, impact ranges, and confidence levels, which can identify the distribution characteristics and impact boundaries of power grid risks in multiple time periods in advance, effectively solving the response lag problem caused by traditional scheduling relying on single-day predictions; mapping the multi-day scenario set to feedforward control variables and obtaining a self-consistent feedforward strategy template by constructing a multi-objective optimization problem can make the scheduling strategy adapt to multi-day risk changes and ensure the matching of various scheduling parameters, avoiding the situation of mismatch between backup configuration and energy path; at the same time, the actual power grid response data is fed back to the evolutionary knowledge graph and driven to evolve online, forming a closed-loop scheduling mechanism of "perception-scheduling-feedback-optimization", which can continuously accumulate scheduling experience and optimize subsequent scheduling strategies, steadily improve the initiative, safety, and economy of power grid scheduling under high-risk scenarios, and thus improve the robustness of power grid stability.

[0025] In step S101 of some embodiments, the extreme weather risk characteristics can be the various risk manifestations of the power grid when extreme weather (such as typhoons, strong convection, cold waves) acts on the power grid; the evolutionary knowledge graph can be a graph used to characterize the semantic association between power grid risks and has the ability to be updated online.

[0026] By combining the risk characteristics of the power grid under extreme weather conditions, an evolutionary knowledge graph that can bear risk semantic associations and support subsequent updates is constructed to provide a knowledge foundation for subsequent situational awareness.

[0027] In some embodiments, an evolutionary knowledge graph representing semantic associations of risks is constructed based on extreme weather risk characteristics, including: Based on the entity classes and relation classes pre-defined to characterize the characteristics of extreme weather risks, a graph structure is established to represent the relationships between weather events, risk types, power grid objects, and control strategies. By assigning the validity period attribute and confidence attribute to the relationships in the graph structure, an evolutionary knowledge graph is obtained.

[0028] Specifically, the preset entity classes are designed to adapt to the risk characterization needs of a high proportion of renewable energy power grids under extreme weather conditions. They include five types of entities: WeatherEvent, Region, Asset, Constraint, and Strategy. Among them, WeatherEvent specifically refers to meteorological processes with clear temporal and spatial range and intensity, such as severe convection / thunderstorms, typhoons, cold waves / heat waves, and long-term cloudy / dust storms, which are also the main causes of power grid risks under extreme weather conditions; Region specifically refers to the power grid operating area (including zones, feeders, key sections, and converter station coverage areas), with geographical boundaries, topology, and power flow sensitivity parameters; Asset specifically refers to power grid equipment such as conventional units, wind / solar RES, energy storage ESS, and FACTS / DC devices, with parameters such as capacity, ramp rate, and status; Constraint refers to frequency, RoCoF, voltage / section / N-1, and reserve coverage (Security Constraint); Strategy template refers to strategy levels that can be quickly issued, such as inertia enhancement, droop / AGC bandwidth, fast reserve, demand response, and damping support.

[0029] The power grid objects correspond to the regions and equipment in the entity class, and the control strategies correspond to the strategy templates in the entity class, specifically referring to strategy levels that can be quickly issued, such as inertia enhancement, droop / AGC bandwidth, fast backup, demand response, and damping support; the risk types are three preset types based on the risk characteristics of the power grid under extreme weather conditions, namely Pulse, Step, and Self.

[0030] Among them, "pulse" refers to sudden disturbances on a timescale of seconds to minutes, manifested as rapid jumps / impacts in power output or frequency within a short time window. Typical triggers include thunderstorms / strong convection, gusts, and sudden changes in weather such as overcast skies and clear skies. To accurately quantify the probability and impact of this type of risk, this embodiment constructs a quantitative expression for a pulse risk score. This risk score is calculated by weighting parameters such as the regional net power output change, frequency jump amplitude, and radar echo intensity. The calculation formula can be expressed as follows:

[0031] in, It is a cutoff function that restricts the weighted sum calculated within the brackets to the interval [0,1], ensuring that the final risk score is always a reasonable value between 0 and 1, and avoiding the calculation result from exceeding the effective range of risk assessment. For the region Net output (renewable + conventional) (load) in window The amount of change, You can choose 60s or 300s; It is 0.8–1.0 times the net ramp rate of the same period and hour in history or the maximum ramp allowed by grid connection. The frequency jump amplitude within a 10–30s window. This is the permissible frequency jump threshold for the region, reflecting the regional power grid's tolerance to frequency fluctuations. When the frequency jump amplitude exceeds this threshold, the corresponding component will significantly increase the risk score. It is set to 0.05–0.2 Hz (based on rigid regional frequency tuning). For regional radar echo intensity, The starting point for convection triggering (e.g., 35dBZ). The span of the starting point for strong convection (e.g., 55–60 dBZ); when there is no radar, the instantaneous drop ratio of the upper quantile of gust wind speed and total solar radiation (GHI) can be used as a substitute and adjusted accordingly. Weight Allocation based on regional characteristics (the allocation can be increased in areas with severe convection). The frequency vulnerability zone can be increased. ).when The pulse pattern is determined to be valid; Mark as a note to follow.

[0032] A step-type load refers to a sustained plateau or significant shift on an hourly to daily timescale, characterized by prolonged periods of high / low power output or load increase / decrease. Typical triggers include typhoon plateau periods, prolonged cloudy / dust storms, cold waves / heat waves, etc. Its risk score is calculated by weighting parameters such as net power output or load shift amplitude, duration of continuous plateau, and temperature change amplitude. The calculation formula can be expressed as follows:

[0033] in: It is a truncation function that restricts the calculation results to the interval [0,1] to ensure the rationality of the risk score. The offset of net output or load relative to a slowly varying baseline (such as 24-hour moving median / day-ahead plan). Take the historical platform offset P90 or the planned acceptable offset; To meet Continuous platform duration, Data is collected over 3–24 hours (based on regional statistics). Temperature at window The range of variation (6–24h) Based on historical seasonal temperature variations. Weighting. Weighted by the platform amplitude, persistence, and importance of slow meteorological variables. When The step type is determined to be valid; Mark as a note to follow.

[0034] Self-excited (Self) refers to a situation where, under high-stress conditions caused by extreme weather (such as decreased effective inertia, reserve strain, power flow reconfiguration, AGC / droop loop overload, etc.), control loops or group couplings are induced or amplified, resulting in or tending to exhibit low-frequency oscillations of 0.1–1 Hz or a high-risk ready state. This is characterized by insufficient damping and significant coherence between multiple power source units. Its risk score is calculated by combining measured oscillation priority, oscillation indication, and high-stress ready index. The calculation formula can be expressed as follows:

[0035] in: It is a truncation function that restricts the calculation results to the interval [0,1] to ensure the rationality of the risk score. For "measured oscillation priority", it is recommended =0.7, used to prioritize the acceptance of direct evidence of measured oscillations: if the dominant mode is detected in the 0.1–1Hz band with insufficient damping, or if the in-band spectral peaks are significant, then take 0.7. =1, otherwise take =0.

[0036] Oscillation Indicator A value of 1 indicates that a low-frequency oscillation meeting the criteria has been detected, while 0 indicates that it has not been detected. High Stress Readiness Index It reflects the stress level of the system under extreme weather conditions and is a weighted composite of the inertia gap and the upper reserve tension.

[0037] Among them: inertia gap:

[0038] The equivalent inertia of the region; The lower limit of the region is the index that reflects the degree of inertia deficiency, and the result is normalized to [0, 1].

[0039] Increase the reserve tension level: ; For future use; Demand calculated based on risk criteria; To prevent division by zero for extremely small quantities.

[0040] when The self-excited type is determined to be valid; This is categorized as a high-risk ready state. The risk scoring formulas and judgment logic for the aforementioned three risk types together constitute the preset risk criteria.

[0041] The predefined relationship classes are used to characterize the semantic associations between the aforementioned entities and risk types, specifically including six relationships: Triggering relationship (weather event → risk type), which means that a weather event triggers / increases the likelihood of a certain type of risk occurring within its spatial projection and timeframe; Influence relationship (risk type → region / equipment / operational constraint), which means that the risk has a measurable impact on a region, equipment, or operational constraint; Mitigation relationship (strategy template → risk type), which means that a strategy template has a mitigating effect on a specific risk; Subordination relationship (equipment → region), which means that the equipment belongs to or primarily operates in a certain region; Constraint relationship (operational constraint → region / equipment), which means that operational constraints impose operational boundaries on a region or equipment; and Relevance relationship (equipment → risk type ... The presence of significant coherence between devices in a specific frequency band is an important structural clue for identifying self-excited risks.

[0042] When constructing the graph structure, the aforementioned entities and three risk types are first used as nodes, and the six relationships are used as edges connecting the nodes. Simultaneously, based on a two-part causal chain of "weather—risk—object," derivation rules are constructed. The semantic chain of "weather event triggers risk type" (a weather event triggers a certain type of risk (e.g., pulse / step / self-excited) within its spatiotemporal range) and "risk type affects target object" (the risk affects an object (the object can be a region, equipment, or safety constraint)) is synthesized into the derived association of "weather event affects target object." Derivation follows three unified criteria: time-based, intensity-based, and confidence-based. Furthermore, combining various risk assessment methods, risk scoring results are integrated into the associations between corresponding nodes and edges, ensuring that the graph structure can completely and accurately represent the semantic relationships between weather events, risk types, power grid objects, and control strategies, and accurately reflect the power grid risk characteristics under extreme weather conditions.

[0043] For example, if there is a risk type triggered by a certain weather event:

[0044] The impact of this risk on the target object (area / equipment / constraints):

[0045] Then the derivation:

[0046] The three unified standards that should be followed when deriving can be expressed as follows:

[0047] It should be noted that, regarding the time frame: derived facts only hold true within the intersection of the validity periods of the two relationships, avoiding exceeding the validity period of either original fact. Regarding the strength frame: the derived strength is the product of the strengths of the two relationships, compressed to the [0,1] interval, indicating that "the stronger the trigger and the stronger the impact, the greater the overall impact." Regarding the confidence frame: the derived confidence is the joint reliability of the confidence levels of the two relationships, defaulting to the product or configured to a more conservative minimum value.

[0048] Furthermore, by associating time validity and confidence attributes with the relationships in the graph structure, an evolutionary knowledge graph is obtained. In this graph, all relationships (including the original six relationships and derived relationships) are associated with both time validity and confidence attributes. These attributes are set based on a knowledge graph model with time and confidence attributes, defined as G=(V,E,R,A), where V is the set of nodes comprising all the aforementioned entities and risk types, and E... V×R×V is the semantic relation set consisting of all relations, R is the relation type set, and A is the attribute set of entities and relations; each relation instance uses extended triples. This indicates that there is a relation type r between the subject entity h and the object entity t, and this relation is factual. Effective within the period, The start time, The end time is defined as w∈[0,1], the influence strength (weight) of the relationship is defined as w, and the evidence confidence level of the relationship is defined as c∈[0,1]. The confidence level is dynamically determined based on the type of evidence: high confidence level when PMU / authoritative meteorological data is used as evidence, medium confidence level when SCADA / short-term data is used as evidence, low confidence level when proxy quantity is used as evidence, increased confidence level when multiple sources of evidence are consistent, and marked as divergent information and with reduced confidence level weight when multiple sources of evidence are contradictory. The influence strength (weight) is allocated based on the risk score results, the strength definition of the derived rules, and regional characteristics.

[0049] By associating the aforementioned time validity period and confidence attributes with all relationships in the graph structure, the graph structure can not only represent the semantic relationships between nodes, but also reflect the timeliness and reliability of the relationships. At the same time, combined with the entity definition, risk assessment and derivation rules mentioned above, an evolutionary knowledge graph with semantic definition capabilities, risk identification capabilities and dynamic relationship adjustment capabilities is formed. This knowledge graph can uniformly express various power grid risks caused by extreme weather, and provide a solid knowledge foundation for subsequent multi-day rolling situational awareness and its own online evolution.

[0050] In an alternative embodiment, such as Figure 2 As shown, Figure 2This is a structural diagram of the evolutionary knowledge graph provided in this application embodiment. The evolutionary knowledge graph includes six major entity modules: weather events, risk types, strategy templates, regions, equipment, and operational constraints. The weather event entity covers extreme weather processes such as thunderstorms / severe convection, typhoon plateau periods, cold waves / heat waves / long cloudy periods; the risk type entity includes three types of power grid risks: pulse-type, step-type, and self-excited; the strategy template entity includes directly deployable scheduling strategies such as inertia enhancement / droop adjustment and rapid standby / frequency regulation capabilities; the region entity includes power grid operation zones such as nearshore transmission areas and load center areas; the equipment entity includes power grid equipment such as wind farms, photovoltaic power stations, and energy storage power stations; and the operational constraint entity includes power grid safety boundaries such as frequency / RoCoF constraints and cross-section / N-1 constraints. Simultaneously, entity association is achieved through six semantic relationship links: triggering, impact, mitigation, membership, coherence, and constraint. The triggering relationship connects weather events and risk types to reflect... The map explores the inducing effects of extreme weather on risks, connecting risk types with regions / equipment / constraints to illustrate the path of risk impact on power grid objects, mitigation relationships with strategy templates and risk types to clarify the risk mitigation effect of dispatch strategies, affiliation relationships with equipment and regions to indicate the operating area of ​​equipment, coherence relationships with equipment to characterize the dynamic coupling characteristics between equipment, and constraint relationships with operational constraints and regions / equipment to define the operational limitations of safety constraints on power grid objects. Overall, this map integrates scattered meteorological, risk, power grid, and strategy information into an interpretable semantic network through structured entities and semantic relationships. It supports risk tracing from "extreme weather → risk type → affected object" to precise dispatch matching from "risk type → strategy template," providing a unified knowledge foundation for subsequent multi-day situational awareness and feedforward dispatch.

[0051] In step S102 of some embodiments, multi-day rolling situational awareness can be a process of capturing and analyzing the power grid risk situation in a rolling manner with multiple consecutive dates as the cycle; the multi-day scenario set includes a collection of multi-day power grid risk-related information, including risk score, impact range, and confidence level; the risk score can be a quantitative value of the risk level of the power grid at various time periods over multiple days; the impact range can be the range of power grid areas, equipment, etc. affected by the risk; and the confidence level can be the reliability of information such as risk score and impact range.

[0052] Based on the constructed evolutionary knowledge graph, a multi-day rolling approach is adopted to carry out power grid situational awareness. Through the risk semantic association carried by the graph, the risk score of each time period is quantified, the scope of risk impact is clarified, and the confidence level of relevant information is determined. Finally, the data is integrated to form a multi-day scenario set.

[0053] In some embodiments, multi-day rolling situational awareness is performed based on an evolutionary knowledge graph to obtain a multi-day scenario set including risk score, impact range, and confidence level, including: Meteorological forecast data and power grid operation measurement data covering a multi-day time range are input into an evolutionary knowledge graph. Based on the evolutionary knowledge graph and preset risk criteria, a risk score covering a multi-day time range is calculated. Based on the entity relationship chain in the evolutionary knowledge graph, the scope of the risk's impact is deduced. The confidence level of the risk is derived based on the reliability of the source of the input data; Based on risk scores, scope of impact, and confidence levels, a multi-day scenario set is formed.

[0054] Specifically, firstly, power grid operation measurement data (including short-term real-time data such as PMU measurements and SCADA data) covering a multi-day time range (e.g., the next 72 hours) and meteorological forecast data (e.g., radar echoes, temperature, and wind speed forecast data) are rolled out in a rolling scenario. Taking 15 minutes as the time resolution, the data is mapped to the dynamic changes of wind and solar power output, load, reserve, and inertia, and then input into the evolutionary knowledge graph. Based on the semantic association of "weather - risk type (pulse / step / self-excited) - object (region / equipment / constraint)" in the evolutionary knowledge graph, within the 72-hour rolling time domain (days D+1, D+2, and D+3), the risk scores are calculated using the preset risk criteria for the aforementioned three types of risks, resulting in risk scores corresponding to the three risk types (pulse type, step type, and self-excited type).

[0055] Simultaneously, at different daily levels (D+1, D+2, and D+3), following a weighted fusion strategy of "relying more on actual conditions for near-term data and more on forecasts for distant-term data," the forecast risk level is weighted and fused with the near-actual / short-term risk levels to obtain the risk level (i.e., risk score) for each daily level. To standardize this fusion calculation process, the risk scores for each daily level and the three risk types are implemented using the following normalization formula:

[0056] in, This represents the final merged risk score; r represents the power grid region, such as a coastal zone or a key section coverage area; t represents the time point, corresponding to a 15–60 minute time step in the scheduling; (d) represents the daily level, d∈{1,2,3} corresponding to D+1 (near end), D+2 (middle end), and D+3 (far end), respectively; RT represents the risk type, covering three types of risks: pulse-type, step-type, and self-excited. For the three types of risks—pulse-type, step-type, and self-excited—the normalization formula is substituted into each risk type to perform fusion calculation, and the final output risk score is... Each risk type will correspond to a specific risk level. When When the aforementioned judgment threshold is reached (≥0.6 indicates a risk is established, 0.4–0.6 indicates a concern), "Trigger (Weather → Risk)" and "Impact (Risk → Object)" are directly written / updated in the corresponding daily snapshot, and "Weather Impact Objects" are derived from this to provide input for "when, where, how strong, and how credible" in the future.

[0057] Next, relying on the entity relationships in the evolutionary knowledge graph, the scope of risk impact is deduced: first, the risk score is projected onto the three types of safety screens of concern to the dispatcher, clarifying the specific dimensions of the risk's impact on grid security, including frequency and reserve coverage (assessing effective inertia, sufficient upper / lower reserve and master / slave regulation bandwidth availability), ramp accessibility (assessing the coverage capability of wind, solar and energy storage available ramps to net load changes), and network security (assessing critical section / PTDF exposure, N...). 1. Voltage and Over-Limit Risks); Furthermore, by utilizing the evolutionary knowledge graph's concepts of "affiliation (device → region)" and "coherence (assets)", The entity relationships such as "asset, self-excited special" and "constraint (constraint → object)" extend the scope of influence from the regional level to equipment and cross sections. Among them, pulse-type and step-type risks extend according to the geographical and topological exposure range, while self-excited risks only spread among assets with "significant coherence and low damping" and the impact level decreases according to the coherence intensity to avoid excessive amplification of risk impact.

[0058] Subsequently, the confidence level of the risk score and the scope of impact is updated based on the reliability of the source of the input data: confidence levels are divided according to the data source, with PMU / authoritative meteorological data being high confidence, SCADA / short-term data being medium confidence, and proxy data being low confidence; when multiple sources of data are consistent, the confidence level is increased, and when multiple sources of data are contradictory, they are marked as "disagreement information" and participate in subsequent derivation with only low weight.

[0059] In an optional embodiment, this method also drives the evolutionary knowledge graph to achieve self-evolution through a multi-time-dimensional rolling snapshot and iteration mechanism, so as to ensure the dispatch center's forward-looking control over the entire process of extreme weather affecting the power grid. The system maintains risk situation snapshots at three daily levels (D+1, D+2, and D+3) with a rolling time step of 15–60 minutes. At each rolling time point, it automatically absorbs new power grid operation measurement data and higher-resolution meteorological forecast data, and gently updates the influence strength and confidence of existing semantic relationship edges in the graph.

[0060] During the snapshot iteration process, the system dynamically optimizes the risk representation based on the evolution characteristics of the risk: for risks that are outdated or significantly weakened, they will be automatically forgotten or downgraded; for risks that continue to strengthen and exceed the threshold, they will be automatically upgraded and their scope of influence mapping will be tightened; when faced with conflicts in multi-source data, the system will follow the principle of "high confidence + latest priority" to process them, mark the second-best facts with "disputed" labels and restrict their spread to downstream links, thereby ensuring the reliability of risk information.

[0061] At the same time, the system will write back the actual execution effect of the self-consistent feedforward strategy template (such as whether frequency overruns occur, whether low-frequency oscillations are alleviated, and whether the reserve capacity is fully covered) to the evolutionary knowledge graph, and transform it into the semantic relationship of "strategy → risk (mitigation / ineffectiveness)". This allows the graph to continuously accumulate interpretable scheduling experience in the closed-loop iteration of "risk identification - strategy execution - effect verification", and gradually form a scheduling knowledge system that can be directly reused.

[0062] To standardize the iterative process of the graph, the system adopts a unified online evolutionary approach:

[0063] When new evidence arrives, the fusion coefficient is used. The intensity of the old influence With the strength of the new recommendations Convex combinations are used to avoid abrupt changes in risk scores; confidence levels are based on source reliability. Based on the improvement and at time intervals Exponential decay (forgetting rate) (It can be adjusted on a monthly replay basis) to quickly bring the confidence level closer to the actual system state. After the snapshot is updated, the D+1 daily snapshot mainly supports intraday and day-ahead scheduling presets, while the D+2 and D+3 daily snapshots are mainly used for the early layout of prefabricated energy storage SOC platforms, planning of reserve rolling bandwidth, and cross-regional collaborative scheduling.

[0064] Finally, the dynamically updated results of risk scores, impact range, and confidence levels are integrated to form a multi-day scenario set: risk snapshots are generated on days D+1, D+2, and D+3. When the risk score reaches the threshold in step one (≥0.6 indicates a risk is established, 0.4–0.6 indicates a risk of concern), the semantic relationship between "trigger (weather → risk)" and "impact (risk → object)" is written / updated in the corresponding day snapshot, and the association of "weather-affected objects (affects)" is derived, providing precise "when—where—how strong—how reliable" information for subsequent scheduling. The system features three daily snapshots that are automatically synchronized. The D+1 snapshot is primarily used for intraday and day-ahead scheduling presets, while the D+2 / D+3 snapshots are primarily used for the advance planning of pre-built energy storage SOC platforms, reserve rolling bandwidth, and cross-regional collaboration. Simultaneously, the execution effects of the self-consistent feedforward strategy template (such as whether over-limits occur, whether oscillations are mitigated, and whether reserves are covered) are written back as a semantic relationship of "strategy → risk (mitigation / ineffectiveness)," enabling the evolutionary knowledge graph to continuously accumulate scheduling experience in the closed loop of "identification—execution—verification," further optimizing the accuracy and practicality of multi-day scenario sets.

[0065] In step S103 of some embodiments, the feedforward control quantity can be a control parameter that is determined in advance based on a multi-day scenario set and is used to guide subsequent power grid dispatch optimization.

[0066] The multi-day scenario set (including risk score, impact range, and confidence level) is transformed and mapped into feedforward control quantities that can be directly used in subsequent power grid dispatching, thereby realizing the transformation of risk information into dispatching control parameters.

[0067] In some embodiments, mapping a multi-day scene set to feedforward control variables includes: Based on risk scores from multiple days of scenario clustering, the system generates backup demand correction, energy storage status reference trajectory correction, and system control parameter level switching suggestions. Based on the impact range of multiple days of concentrated scenarios, set the upper and lower envelope constraint parameters for new energy output; The feedforward control quantity is formed based on the backup demand correction, the energy storage state reference trajectory correction, the system control parameter level switching suggestion, and the upper and lower envelope constraint parameters.

[0068] Specifically, mapping a multi-day scenario set to feedforward control variables is a crucial step in transforming risk semantics into executable scheduling instructions. This process relies on risk scores (e.g., risk score sequences for the three risk types) corresponding to different risk types (e.g., impulsive, step-type, and self-excited) within the multi-day scenario set. Based on the risk and its impact range, various control parameters are generated in a differentiated manner to ensure that the scheduling system can respond to specific risks in advance and accurately.

[0069] In specific mapping, the generation of reserve requirement adjustments follows differentiated objectives based on risk type: pulse-type risks focus on short-term, rapid reserve requirement adjustments; step-type risks focus on long-term reserve requirement adjustments; and self-excited risks focus on adjustments to equivalent inertia and vibration damping capacity requirements, based on a pre-set reserve baseline. Inertia baseline Starting from the risk score, the values ​​are amplified to form reserve and inertia target values ​​that match the risk level.

[0070]

[0071] in, This is the revised backup demand value, which is the final backup target issued by the scheduler. The superscript ↑ indicates backup, req represents the demand, and t is the scheduling time point (corresponding to a time step of 15–60 minutes). Use it as a scene / day level identifier (e.g., D+1 / D+2 / D+3). The baseline reserve requirement, i.e. the initial reserve target value under the condition of no extreme weather risk, serves as the benchmark for risk adjustment. , These are risk amplification weighting coefficients, corresponding to the impact levels of pulse-type and step-type risks, respectively. Their values ​​are determined by the risk tolerance characteristics of the regional power grid (e.g., they can be increased in areas with strong convection). ). This is a pulse-type risk score, where r represents the power grid region and t represents a time point. For the scenario / day level, Pulse represents pulse-type risk, reflecting the risk level of short-term power output mutation (range 0–1). For step-type risk scoring, Step represents step-type risk and reflects the degree of risk of long-term output deviation (range 0–1).

[0072] The revised lower bound target for equivalent inertia is the minimum equivalent inertia value required by the scheduler. `min` represents the lower bound, and `t` is the time point. For scene / day level identification. The baseline equivalent inertia, i.e. the initial target inertia value under conditions of no extreme weather risk, serves as the benchmark for risk correction. , The risk amplification weighting coefficients correspond to the degree of impact of self-excited and impulse-type risks, respectively (e.g., the weighting coefficient can be increased in the sensitive area of ​​low-frequency oscillations). ). For self-excited risk scoring, Self represents self-excited risk and reflects the risk level of low-frequency oscillations in the system (range 0–1).

[0073] The generation of energy storage state reference trajectory correction is dynamically adjusted based on the characteristics of risk type. When the step risk score is high, the energy storage SOC target is raised to the platform state to cover long-term output deviation. When the pulse risk score is high, short-term ramp-up margin is reserved for energy storage power stations and the upper limit of SOC is limited to ensure that energy storage can respond quickly to output gaps. The system control parameter level switching suggestion is automatically generated based on the risk threshold. For example, when the self-excited risk is determined to be established, it switches to the vibration suppression level to increase the virtual inertia and damping level. When the pulse risk is determined to be established, the AGC bandwidth and droop sensitivity are relaxed. Moreover, the level switching suggestion is matched with the reserve demand correction amount and inertia target value to avoid the scheduling contradiction of "having indicators but no ability".

[0074] Based on the impact range of multi-day scenarios, upper and lower envelope constraint parameters for renewable energy output are set. For renewable energy power plants within the clearly defined risk impact range of multi-day scenarios, the upper and lower envelope constraints for wind and solar power output are set using the calculation method of "mean ± (risk coefficient × uncertainty band)". The higher the risk score, the wider the envelope range and the more conservative the value, in order to adapt to the output fluctuations caused by risks. At the same time, based on the "attention / establishment" window corresponding to the risk score, demand response (DR) is triggered in stages, and the capacity of demand response is decoupled from the standby metering to avoid duplicate metering affecting the accuracy of scheduling resource allocation.

[0075] Based on the reserve demand correction, energy storage status reference trajectory correction, system control parameter level switching suggestions, and upper and lower envelope constraint parameters, a feedforward control quantity is formed. This feedforward control quantity integrates multi-dimensional dispatch instructions such as reserve, energy storage, control strategy, and new energy constraints, and can be directly issued to the dispatch execution end to realize feedforward active defense against grid risks under extreme weather conditions.

[0076] In step S104 of some embodiments, the multi-objective optimization problem is to build a scheduling optimization model of the demand solution with the demand of power grid scheduling as the objective; the self-consistent feedforward strategy template is a standardized scheduling scheme that adapts to the feedforward control quantity and whose scheduling parameters are matched with each other and can be directly used for power grid scheduling.

[0077] By combining feedforward control variables, a multi-objective optimization model that meets the needs of power grid dispatch is built. By solving this model, a self-consistent feedforward strategy template that matches each dispatch parameter and can be directly executed is obtained.

[0078] In some embodiments, a multi-objective optimization problem is constructed and solved by combining feedforward control variables to obtain a self-consistent feedforward policy template, including: Using feedforward control variables as boundary conditions, and with operating costs, safety penalties, and resilience indicators as objectives, a multi-objective optimization problem covering a future multi-day time range is constructed. A multi-objective evolutionary algorithm is used to solve the multi-objective optimization problem, and a self-consistent feedforward policy template distributed on the Pareto front is obtained.

[0079] Specifically, this multi-objective optimization problem covers a time domain of 72 hours, with a time resolution set to 15–60 minutes, and is calculated based on a multi-day scenario set. The optimization objective is to construct a multi-objective system oriented towards "economy, safety, and resilience," solving for time-segmented reserve allocation, unit output and start-up / shutdown, ESS charging / discharging and SOC trajectories, control levels, and wind / solar / reverse rotation execution quantities. The objective vector is defined as follows:

[0080] Among them: The expected operating costs (fuel, start-stop, ESS loss, DR costs, etc. aggregated by scenario weight); For safety penalties, a summary of various over-limit / gap penalties (frequency / section over-limit, load loss, reserve gap, etc., calculated using the maximum value or weighted value); This refers to resilience indicators (minimum ramp margin, SOC position, cross-regional mutual aid availability, etc., the larger the better, hence the negative sign).

[0081] Let the rolling optimization window be 72h, and the time step be... The total number of steps is: .

[0082] At each time step The feedforward parameters required for scheduling decisions include: traditional unit output and reserve, energy storage power and SOC, control level, composite inertia coefficient, droop coefficient, AGC participation factor, renewable output envelope coefficient, demand response trigger parameters (trigger threshold, available capacity, etc.).

[0083] Stacking all the decisions for a given candidate template over time can be represented as a decision matrix:

[0084] in: The total dimension of the single-step decision variables; For the first The column vector of all decision variables. This is the mathematical representation of a feedforward policy template.

[0085] The self-consistent feedforward strategy template, by definition self-consistent, means that the quantities in the decision matrix X must be mutually matched, namely: consistency with various physical constraints (power balance, network power flow, frequency indicators, etc.), inherent dynamic constraints (SOC evolution, minimum start-up / shutdown, ramp-up constraints), and feedforward objectives (reserve requirements, minimum inertia, etc.). The equation is expressed as:

[0086] Therefore, the constructed multi-objective optimization problem is to find a set of non-dominated solutions within the aforementioned constraint set:

[0087] These X's are the candidate "self-consistent feedforward strategy templates".

[0088] In some embodiments, a multi-objective evolutionary algorithm is used as a multi-objective hoarfrost optimization algorithm. The multi-objective hoarfrost optimization algorithm generates an initial population by simulating a deposition operator, performs individual evolution by a dendrite growth operator, and performs constraint repair and population screening by an ice peeling operator.

[0089] Furthermore, a multi-objective rime optimization algorithm is used to solve the problem. The multi-objective "rime optimization" is essentially a multi-objective evolutionary algorithm, using the imagery of "rime" to characterize three key operators: deposition (generation), dendrite growth (mutation / crossover), and ice peeling (repair / screening).

[0090] (1) Sedimentation operator: Population initialization Let the population size of each generation be N. The population size of the k-th generation is denoted as:

[0091] First, construct a baseline template X. base Using feedforward control variables (reserve requirements, inertia threshold, ESS reference trajectory, control gear, etc.), a complete T×D matrix is ​​given according to the principle of "safety first" or "economy first" without violating constraints.

[0092] Then, "droplet deposition" is performed: for each individual Generate a perturbation matrix ,For example:

[0093] in For the first The standard deviation of each decision component (which can be set according to capacity or experience) is then used to perform boundary projection on the perturbed template:

[0094] This means "projecting onto the basic boundary set" (such as the upper and lower limits of output, SOC, and reserve limits), ensuring that the initial individuals are not physically out of line. This yields the initial population P(0) of generation 0.

[0095] (2) Target calculation and constraint default determination For each individual X, calculate the target vector F(X), and simultaneously calculate a constraint violation scalar V(X) to quantify the degree of constraint violation:

[0096] If V(X)=0, it indicates self-consistency, that is, the constraints are satisfied; If V(X) > 0, it indicates that there is a certain degree of breach (exceeding the limit, conservation imbalance, etc.).

[0097] During sorting, V(X) is used as the criterion for "ice peeling": templates with more severe out-of-bounds errors are peeled off first.

[0098] (3) Ranking of multi-objective dominance relationships In a multi-objective context, Pareto dominance is used to define individual merit. Given two new modules X and Y, we define:

[0099] in" Pareto Domination The condition for "is:

[0100] Based on the above rules, the population can be divided into several "non-dominated layers" (front): the first layer is the optimal template that is not dominated by any individual; the second layer is the optimal template that is "dominated only by the first layer"; and so on. In the rime ice generation algorithm, the "dendritic skeleton" is formed by the individuals at the front, which are the current "rime ice branches".

[0101] (4) Dendrite growth operator: offspring generation For each parent individual The generation of offspring is achieved through a growth operator. A typical form is:

[0102] in, As a guiding template, it is usually selected from the current non-dominant frontier (e.g., representatives in the direction of security priority or economic priority). For the first The "growth coefficient" of each generation gradually decreases with each iteration, achieving "early exploration and later convergence"; For random disturbance terms, It is a standard normal matrix. The scaling matrix for each decision component, " indicates element-wise multiplication. The result is..." Then, perform two steps of "ice peeling / repair": The first step is to perform boundary clipping; first, perform a boundary projection:

[0103] Ensure that all variables fall within the physically permissible range (output, SOC, gear, etc. do not exceed basic limits).

[0104] The second step is to perform feasibility repairs: For individuals that still exceed limits, repairs are made using engineering rules, such as: increasing reserves, increasing inertia, tightening the envelope, and delaying gear shifts. If the degree of violation is reduced after repair... If the value is less than the threshold, it is retained; otherwise, it is considered "ice peeling" and does not enter the next round of candidates.

[0105] All offspring that pass the repair process constitute the offspring population. .

[0106] (5) Selection and Diversity Parent generation with offspring Merged into a mixed population :

[0107] right Perform a non-dominated sort to obtain several levels of the frontier. When selecting the next generation of the population: First, select according to the leading edge layer: starting from the first layer, add non-dominant layers to the next generation in sequence until the population size is about to be exceeded. .

[0108] In the last included non-dominated layer, selection is based on "crowding distance" to maintain the "dendritic morphology diversity" of the leading edge. Crowding distance is defined as (calculated individually for each target, then summed):

[0109] For an individual in a non-dominated layer, according to the goal Sort in ascending order; These are the target values ​​of the individual's neighboring individuals in this target dimension; the crowding distance at the extreme point is set to infinity to ensure that the boundary solution is not eliminated. Individuals with a large crowding distance have "few neighbors" in the target space, which helps to prevent solutions with different safety-economy-resilience trade-offs from being overwhelmed.

[0110] The next generation is obtained through the above selection process:

[0111] Iterate until convergence (e.g., when the frontier changes fall below a threshold after several generations, or when the maximum number of generations is reached).

[0112] (6) Output a set of self-consistent feedforward strategy templates After the algorithm converges, select the first non-dominated layer from the final population:

[0113] Each They are all one A matrix that satisfies: in all scenarios Above, constraints It is not dominated by any other template in the target space. This is called a cluster of self-consistent feedforward strategy templates, possessing the characteristic of "self-consistency": each The internal variables (reserve, inertia, ESS trajectory, gear, envelope, DR) are matched in terms of time, scenario, and feedforward objectives; "Feedforward": These templates do not rely on post-event feedback, but are based on multi-day rolling situational awareness, providing parameter settings for every moment within 72 hours in advance; "Diversity": Different templates correspond to different safety-economy-resilience trade-offs, and dispatchers can select appropriate templates according to operational preferences or superior strategies.

[0114] In addition, the results of scheduling execution (whether the limit is exceeded, the lowest frequency point, reserve use, cost level, etc.) are written back to the knowledge graph through the "strategy → risk / constraint" edge. They are absorbed by the previous rolling snapshot and iteration mechanism, so that the initial population of the next round of rime optimization is closer to the regional reality and scheduling preferences, thereby continuously refining the self-consistent feedforward strategy template.

[0115] In step S105 of some embodiments, the actual power grid response data is various types of data generated by the actual operation of the power grid after the self-consistent feedforward strategy template is applied to the actual power grid dispatch (such as operating status, risk mitigation effect, etc.); online evolution is the process by which the evolutionary knowledge graph updates its semantic associations and risk-related information in real time based on the feedback of the actual power grid response data, and continuously improves itself.

[0116] The self-consistent feedforward strategy template is applied to actual power grid dispatching. The actual response data of the power grid after dispatching is collected and fed back to the evolutionary knowledge graph. The knowledge graph is updated and improved through the feedback data, thus realizing its online evolution.

[0117] This method employs a closed-loop process of "constructing an evolutionary knowledge graph—generating a multi-day scenario set—mapping feedforward control variables—solving a self-consistent scheduling template—feedback-driven graph evolution," using the evolutionary knowledge graph as the core carrier. This enables the early perception and quantification of multi-day risks to the power grid under extreme weather conditions, transforming risk information into implementable scheduling strategies. Through feedback iteration, the knowledge graph and scheduling template are continuously optimized, effectively solving the problems of delayed response and configuration mismatch in traditional scheduling. This significantly improves the initiative, safety, and economy of power grid scheduling under extreme weather conditions, while also achieving continuous accumulation and optimization of scheduling knowledge.

[0118] In one embodiment, a coastal power grid is used as the application object. The installed capacity of new energy sources accounts for about 60% of the grid, which is equipped with centralized wind farms, photovoltaic power stations and electrochemical energy storage power stations. It is significantly affected by extreme weather such as typhoons, strong convection and cold waves, and is prone to problems such as large fluctuations in net load and insufficient system inertia. The specific implementation process is as follows: The dispatch center first constructs an evolutionary knowledge graph based on the characteristics of extreme weather risks, defining six entities: weather events (thunderstorms, typhoons, cold waves, long cloudy days), risk types (pulse-type, step-type, self-excited type), region, equipment, constraints, and strategies. This is achieved through mechanisms such as "triggering (weather → risk)," "impacting (risk → region / equipment / constraint)," "mitigating (strategy → risk)," "belonging (equipment → region)," and "coherence (equipment)." The graph structure is established for relationships such as "equipment", and the validity period and confidence level attributes are associated with all relationships, so that the causal relationship between extreme weather, risk type, affected objects and control strategies has a clear semantic expression.

[0119] Based on this evolutionary knowledge graph, the dispatch center conducts 72-hour multi-day rolling situational awareness, unfolding rolling scenarios with a 15-minute time resolution. Meteorological forecast data and power grid operation measurement data are input into the evolutionary knowledge graph. Risk scores are calculated at each time point according to three preset risk criteria. The scope of risk impact is deduced through entity relationship chains in the graph, and the confidence level is determined according to the reliability of data sources, forming a multi-day scenario set containing risk scores, scope of impact, and confidence levels. At the same time, a rolling snapshot and iteration mechanism with multiple time dimensions (D+1, D+2, D+3) is established. At each rolling time point, new measurement data and high-resolution forecasts are absorbed, and the impact strength and confidence level of graph relationships are dynamically updated. Expired or weakened risks are automatically forgotten, and enhanced risks are automatically upgraded. Multi-source conflicts are handled according to "high confidence + latest priority", realizing the self-evolution of the knowledge graph.

[0120] Subsequently, the dispatch center maps the multi-day scenario set into feedforward control variables: based on risk scores, it generates reserve demand correction variables, energy storage status reference trajectory correction variables, and system control parameter level switching suggestions. Pulse-type risks focus on short-term rapid reserve, step-type risks focus on long-term reserve and energy storage SOC platform settings, and self-excited risks focus on equivalent inertia and vibration suppression levels. Based on the impact range, it sets upper and lower envelope constraint parameters for new energy output, with higher risks resulting in more conservative envelopes, and triggers demand response in stages while decoupling it from reserve metering. Using these as boundary conditions, a multi-objective optimization problem is constructed with operating costs, safety penalties, and resilience indicators as objectives. The multi-objective hoarfrost optimization algorithm is used to solve this problem: an initial population is generated through a deposition operator, individual evolution is achieved through a dendrite growth operator, constraint repair and screening are performed using an ice peeling operator, and population diversity is maintained by combining non-dominated sorting and crowding distance, ultimately obtaining a self-consistent feedforward strategy template set on the Pareto front.

[0121] Dispatchers select and execute appropriate templates based on extreme weather risk levels and operational preferences. After the rolling cycle ends, actual power grid response data (frequency response, reserve mobilization, over-limit situations, operating costs, etc.) are written back to the evolutionary knowledge graph in the form of "strategy → risk" and "strategy → constraint". The data is absorbed by the rolling snapshot and iteration mechanism, forming a closed loop of "identification - perception - scheduling - feedback - evolution", which continuously improves the safety and economy of power grid dispatch under extreme weather conditions.

[0122] Please see Figure 3 This application also provides a power grid dispatching device that can implement the above-described power grid dispatching method. The device includes: Module 301 is used to construct an evolutionary knowledge graph representing semantic associations of risks based on the characteristics of extreme weather risks; The perception module 302 is used to perform multi-day rolling situational awareness based on the evolutionary knowledge graph, and obtain a multi-day scene set including risk score, impact range and confidence level. Mapping module 303 is used to map multi-day scene sets into feedforward control variables; The solver module 304 is used to combine the feedforward control quantity to construct and solve a multi-objective optimization problem and obtain a self-consistent feedforward strategy template. Feedback module 305 is used to execute the self-consistent feedforward strategy template, obtain actual power grid response data, and feed the actual power grid response data back to the evolutionary knowledge graph to drive the online evolution of the evolutionary knowledge graph.

[0123] The specific implementation of this power grid dispatching device is basically the same as the specific implementation of the power grid dispatching method described above, and will not be repeated here.

[0124] Thirdly, embodiments of this application provide an electronic device, see [link to relevant documentation]. Figure 4 The diagram shown is a structural schematic of an electronic device provided in this application.

[0125] like Figure 4 As shown, the device includes: Memory 31 is used to store computer programs; Processor 32 is used to execute computer programs; When the processor 32 executes the computer program, it implements the power grid dispatching method as described in any of the above embodiments.

[0126] For example, a computer program may be divided into one or more modules / units, one or more of which are stored in memory 31 and executed by processor 32 to complete this application. One or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in an electronic device.

[0127] The processor 32 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0128] The memory 31 can be used to store computer programs and / or modules. The processor 32 implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory 31 and calling the data stored in the memory 31. The memory 31 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 31 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0129] It should be noted that the aforementioned electronic devices include, but are not limited to, processors and memory, as will be understood by those skilled in the art. Figure 4 The structural diagram is merely an example of the electronic device described above and does not constitute a limitation on the electronic device. It may include more components than shown in the diagram, or combine certain components, or use different components.

[0130] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed, implements the power grid dispatching method of any of the above embodiments.

[0131] It should be understood that the implementation of all or part of the processes in the above-described power grid dispatching method can also be accomplished by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the above-described power grid dispatching method. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the relevant jurisdiction. For example, in some relevant jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0132] Fifthly, embodiments of this application also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the power grid dispatching method of any of the above embodiments.

[0133] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0134] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. A power grid dispatching method, characterized in that, include: Constructing an evolutionary knowledge graph representing semantic associations of risks based on the characteristics of extreme weather risks; Based on the evolutionary knowledge graph, multi-day rolling situational awareness is performed to obtain a multi-day scenario set including risk score, impact range and confidence level; Map the multi-day scene set to feedforward control variables; By combining the aforementioned feedforward control variables, a multi-objective optimization problem is constructed and solved to obtain a self-consistent feedforward strategy template. The self-consistent feedforward strategy template is executed to obtain actual power grid response data, and the actual power grid response data is fed back to the evolutionary knowledge graph to drive the online evolution of the evolutionary knowledge graph.

2. The method according to claim 1, characterized in that, The evolutionary knowledge graph that constructs semantic associations of risk based on extreme weather risk characteristics includes: Based on the entity classes and relation classes pre-defined to characterize the characteristics of extreme weather risks, a graph structure is established to represent the relationships between weather events, risk types, power grid objects, and control strategies. The evolutionary knowledge graph is obtained by assigning the validity period attribute and confidence attribute to the relationships in the graph structure.

3. The method according to claim 1, characterized in that, The multi-day rolling situational awareness based on the evolutionary knowledge graph yields a multi-day scenario set including risk scores, impact ranges, and confidence levels, including: Meteorological forecast data and power grid operation measurement data covering a multi-day time range are input into the evolutionary knowledge graph. Based on the evolutionary knowledge graph and preset risk criteria, a risk score covering a multi-day time range is calculated. Based on the entity relationship chain in the evolutionary knowledge graph, the scope of the risk's impact is deduced. The confidence level of the risk is derived based on the reliability of the source of the input data; Based on risk scores, scope of impact, and confidence levels, a multi-day scenario set is formed.

4. The method according to claim 1, characterized in that, The step of mapping the multi-day scene set to feedforward control variables includes: Based on the risk scores from the multi-day scenario cluster, the following are generated: backup demand correction, energy storage status reference trajectory correction, and system control parameter level switching suggestions. Based on the impact range of the multi-day scenario concentration, upper and lower envelope constraint parameters for new energy output are set; The feedforward control quantity is formed based on the backup demand correction amount, the energy storage state reference trajectory correction amount, the system control parameter level switching suggestion, and the upper and lower envelope constraint parameters.

5. The method according to claim 1, characterized in that, The process of combining the feedforward control variable to construct and solve a multi-objective optimization problem yields a self-consistent feedforward strategy template, including: Using feedforward control variables as boundary conditions, and with operating costs, safety penalties, and resilience indicators as objectives, a multi-objective optimization problem covering a future multi-day time range is constructed. The multi-objective evolutionary algorithm is used to solve the multi-objective optimization problem, and a self-consistent feedforward policy template distributed on the Pareto front is obtained.

6. The method according to claim 5, characterized in that, The multi-objective evolutionary algorithm used is a multi-objective hoarfrost optimization algorithm. The multi-objective hoarfrost optimization algorithm generates an initial population by simulating a deposition operator, performs individual evolution by a dendrite growth operator, and performs constraint repair and population screening by an ice peeling operator.

7. A power grid dispatching device, characterized in that, include: The module is used to construct an evolutionary knowledge graph representing semantic associations of risks based on the characteristics of extreme weather risks; The perception module is used to perform multi-day rolling situational awareness based on the evolutionary knowledge graph, and obtain a multi-day scene set including risk score, impact range and confidence level; The mapping module is used to map the multi-day scene set into feedforward control variables; The solution module is used to combine the feedforward control quantity to construct and solve a multi-objective optimization problem and obtain a self-consistent feedforward strategy template. The feedback module is used to execute the self-consistent feedforward strategy template, obtain actual power grid response data, and feed the actual power grid response data back to the evolutionary knowledge graph to drive the online evolution of the evolutionary knowledge graph.

8. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the power grid dispatching method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the power grid dispatching method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which, when executed by a processor, implement the power grid dispatching method as described in any one of claims 1 to 6.

Images