A microgrid energy efficiency analysis and auxiliary decision management system based on big data

Abstract

This invention discloses a microgrid energy efficiency analysis and auxiliary decision management system based on big data, belonging to the field of microgrid energy management technology. The invention includes a data acquisition and real-time mapping module, which constructs a real-time digital mapping model and evaluates its reliability; a proximity analysis module and a proximity group construction module, used to perform correlation analysis on parameter nodes and construct proximity groups; a strategy game and decision-making module, which establishes bidirectional interaction with a virtual sandbox simulation module, generating a global strategy through weighted game iteration; a safety and efficiency analysis module, which evaluates the effectiveness and risk of the strategy and assesses the risk level; a hierarchical decision-making and execution module, which executes automatic issuance, manual confirmation, or early warning and prevention operations according to the risk level; and a bottom-line verification and rollback module, which performs bottom-line verification and triggers rollback and manual intervention when the bottom line is exceeded. This invention achieves safe, efficient, and collaborative optimization of microgrid operation through closed-loop optimization and multi-level security protection mechanisms.

Description

Technical Field

[0001] This invention relates to the field of microgrid energy management technology, specifically to a microgrid energy efficiency analysis and auxiliary decision management system based on big data. Background Technology

[0002] With the large-scale integration of renewable energy and the in-depth implementation of the "dual carbon" strategy, microgrids, as an important carrier for the efficient utilization of distributed energy, have become a hot topic in industry research in terms of energy efficiency management and carbon emission optimization.

[0003] Optimal scheduling technology for microgrids has made some progress. Chinese patent application CN121036192A discloses a flexible networking system and method for multi-user, multi-energy collaborative microgrids. This method constructs a multi-user, multi-energy collaborative microgrid model and uses a combination of particle swarm optimization and Nash game iteration to obtain the optimal networking scheme, thus realizing flexible networking and collaborative operation of multiple users within a microgrid cluster.

[0004] The aforementioned technologies employ an independent game-theoretic optimization approach, lacking a hybrid architecture that combines independent optimization at edge nodes with collaborative game-theoretic interaction at the central node. Furthermore, the generation and verification processes of optimization strategies are fragmented, and the game-theoretic process lacks a real-time simulation and verification feedback mechanism, making it difficult to handle extreme scenarios such as drastic fluctuations in wind and solar power output. When assessing the risks of optimization strategies, only a single strategy result is typically output. Decision execution is mostly fully automated, lacking a risk-level-based tiered decision-making process, a manual confirmation step for medium-risk strategies, and a fallback / circuit breaker mechanism triggered by bottom-line constraints. This fails to achieve a good balance between ensuring system security and improving optimization efficiency. Summary of the Invention

[0005] Based on this, the purpose of this invention is to provide a microgrid energy efficiency analysis and auxiliary decision management system based on big data, so as to solve the technical problems of the separation between the optimization strategy generation and verification process and the lack of real-time feedback, the lack of a hierarchical manual confirmation mechanism in the decision-making process and the lack of a bottom-line constraint circuit breaker mechanism.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a microgrid energy efficiency analysis and auxiliary decision management system based on big data, comprising a data acquisition and real-time mapping module configured to acquire multi-source heterogeneous data of the microgrid and construct a real-time digital mapping model that operates synchronously with the physical microgrid; a proximity analysis module configured to perform correlation analysis on multiple parameter nodes in the real-time digital mapping model and calculate a proximity index; a proximity group construction module configured to construct corresponding internal devices into proximity groups based on the proximity index; and an edge grouping optimization module deployed on the edge computing nodes corresponding to each proximity group, configured to independently perform local energy efficiency optimization calculations based on the real-time mapping data corresponding to its proximity group. The system includes: a strategy game and decision-making module, configured to perform strategy game on multiple candidate local optimization strategies to generate a global candidate optimization strategy; a virtual sandbox simulation module, configured to simulate the execution of the global candidate optimization strategy in a virtual sandbox environment independent of the real-time digital mapping model, generating simulation results; a security and performance analysis module, configured to analyze the simulation results, evaluate the execution effect and security risks, and assess the risk level; a graded decision-making and execution module, configured to execute decision-making operations based on the risk level assessment; and a bottom-line verification and rollback module, configured to perform bottom-line verification, triggering a rollback mechanism and issuing an optimization danger alarm when the bottom-line conditions are exceeded.

[0007] By adopting the above technical solutions, this invention constructs a complete closed loop from data acquisition, affinity analysis, affinity group construction, edge optimization, strategy game, sandbox simulation, performance analysis to hierarchical decision-making, realizing the refined management and safe and efficient operation of microgrids.

[0008] The present invention is further configured such that the data acquisition and real-time mapping module is connected to the proximity analysis module, and the data acquisition and real-time mapping module further includes a credibility evaluation unit configured to calculate the synchronization error between the real-time digital mapping model and the physical microgrid in real time and generate a credibility index; the edge grouping optimization module is connected to the strategy game and decision module, and is further configured to perform the local energy efficiency optimization calculation when the credibility index is higher than a preset threshold.

[0009] By adopting the above technical solution, the credibility assessment unit ensures that the optimization process is triggered only when the model is reliable, effectively avoiding the security risks of making decisions based on distorted data.

[0010] The present invention is further configured such that the intimacy analysis module is connected to the intimacy group construction module, the real-time digital mapping model includes multiple parameter nodes, each parameter node corresponding to a monitoring device in the microgrid; the intimacy analysis module is further configured to comprehensively determine the intimacy index based on at least one factor among the electrical distance between parameter nodes, data correlation coefficient, and historical cooperative change rate; the intimacy group construction module is further configured to configure a set of bottom-line conditions for each intimacy group, the bottom-line conditions including at least one insurmountable operating constraint among equipment capacity limit, safe operation threshold, carbon emission quota, and voltage frequency boundary.

[0011] By adopting the above technical solution, the proximity analysis module comprehensively evaluates the correlation strength between devices from three dimensions: electrical distance, data correlation, and historical co-change rate, and realizes the optimized unit division with highly coupled operating characteristics.

[0012] The present invention is further configured such that the intimacy analysis module is further configured to obtain the electrical distance between any two parameter nodes in the real-time digital mapping model, the electrical distance being calculated based on the impedance matrix between the nodes; obtain the data correlation coefficient of the two parameter nodes within a preset time window, the data correlation coefficient being calculated based on the Pearson correlation coefficient or mutual information; obtain the co-change rate of the two parameter nodes in historical operating data, the co-change rate being calculated based on the frequency of the two parameter nodes changing in the same direction at the same time point; and generate the intimacy index by weighted summation of the electrical distance, data correlation coefficient, and co-change rate. The intimacy group construction module is further configured to construct an association network graph with each parameter node as a vertex and the intimacy index as the edge weight; divide the association network graph using a community detection algorithm, identifying the node groups with tightly connected internal connections and sparse external connections as the intimacy groups; when the size of the identified intimacy group exceeds a preset maximum size threshold, the intimacy group is further divided until the size of each intimacy group meets the processing capacity of the edge computing node.

[0013] By adopting the above technical solutions, the calculation of the intimacy index integrates multi-dimensional factors and achieves accurate quantification through normalization and weighted summation; the community detection algorithm can automatically identify closely related device groups, and the secondary partitioning ensures the load balance of edge computing nodes.

[0014] The invention is further configured such that the virtual sandbox simulation module is connected to the strategy game and decision-making module, and is further configured to capture the current running state data of the real-time digital mapping model as the initial boundary conditions of the virtual sandbox environment; in the virtual sandbox environment, the execution of the global candidate optimization strategy is driven according to a preset simulation time scale and event sequence, and the state change data of each key node during the execution process is recorded to form the simulation result; the virtual sandbox simulation module includes an adversarial verification engine, configured to inject preset extreme disturbance scenarios into the virtual sandbox environment to stress test the global candidate optimization strategy, the extreme disturbance scenarios including at least one of the following: sudden drop in new energy output, sudden increase in load, communication interruption, and equipment failure; when there are multiple global candidate optimization strategies, the virtual sandbox simulation module is also configured to run multiple independent virtual sandbox environments in parallel, and simulate the execution of multiple global candidate optimization strategies respectively, so as to generate multiple simulation results in parallel.

[0015] By adopting the above technical solutions, the virtual sandbox simulation module verifies the effectiveness of the strategy in an independent environment, the adversarial verification engine performs stress tests by injecting extreme disturbance scenarios, and the parallel simulation technology significantly improves the verification efficiency.

[0016] The present invention is further configured such that the safety and efficiency analysis module is connected to the virtual sandbox simulation module and includes an effect evaluation unit configured to compare the simulation results with preset energy efficiency improvement targets and calculate the expected energy efficiency benefits of the global candidate optimization strategy; a safety analysis unit configured to check whether there are at least one preset abnormal operating state among equipment overload, voltage over-limit, frequency instability, and carbon emission over-limit in the simulation results, and quantify the severity of the abnormal operating state; a risk rating unit configured to comprehensively evaluate the risk level of the global candidate optimization strategy based on the expected energy efficiency benefits and the severity of the abnormal operating state; and a decision tracing unit configured to record the key factors affecting the evaluation and their weights during the risk level evaluation process, and generate a decision tracing diagram, wherein the key factors... The sub-module includes abnormal operating states that lead to an increase in risk level and the sources of expected energy efficiency benefits. The hierarchical decision-making and execution module is connected to the safety and efficiency analysis module and is further configured to automatically send the global candidate optimization strategy to the corresponding field execution device via an encrypted communication protocol when the risk level is assessed as low risk; when the risk level is assessed as medium risk, a detailed report containing the global candidate optimization strategy, the expected energy efficiency benefits, the abnormal operating states, and the decision tracing diagram is generated, pushed to the management terminal, and waits for confirmation or modification instructions from the management terminal before sending the global candidate optimization strategy; when the risk level is assessed as high risk, the warning information is generated and pushed to the management terminal, and the sending of the global candidate optimization strategy is prevented.

[0017] By adopting the above technical solutions, the safety and efficiency analysis module comprehensively evaluates strategies from two dimensions: energy efficiency benefits and safety. The decision traceability unit provides interpretable decision-making basis, and the hierarchical decision-making mechanism achieves a reasonable balance between automation and manual intervention.

[0018] The present invention is further configured such that the strategy game and decision-making module is connected to the edge group optimization module and establishes a two-way interactive channel with the virtual sandbox simulation module. The strategy game and decision-making module is further configured to take the candidate local optimization strategies uploaded by each close group as the initial strategy set of the game participants, and configure an initial game weight coefficient for each game participant. The initial game weight coefficient is calculated based on at least one parameter among the importance level, security risk level and historical reliability factor of the corresponding close group.

[0019] During the game, the candidate global optimization strategy for the current round is sent to the virtual sandbox simulation module for pre-simulation, and the pre-simulation results fed back by the virtual sandbox simulation module are received. The pre-simulation results include the estimated value of the strategy execution effect and the security risk assessment value.

[0020] A weighted game model is constructed, wherein the objective function of the weighted game model is to maximize the sum of the weighted utilities of each close group. The game weight coefficient of each close group is dynamically adjusted during the game iteration process based on the pre-simulation results fed back by the virtual sandbox simulation module: when the candidate strategy corresponding to a close group shows a high security risk in the pre-simulation, the game weight coefficient of the close group is reduced; when the candidate strategy corresponding to a close group shows a high energy efficiency gain in the pre-simulation, the game weight coefficient of the close group is increased.

[0021] The weighted Nash equilibrium solution algorithm is adopted to determine the global candidate optimization strategy that can achieve equilibrium among each close group through iterative calculation. In each iteration, the virtual sandbox simulation module is called to pre-simulate and verify the current candidate strategy until the convergence condition is met or the preset maximum number of iterations is reached.

[0022] The strategy game and decision-making module also includes a bottom-line pre-verification unit, configured to perform a bottom-line pre-verification before sending the candidate global optimization strategy to the virtual sandbox simulation module after generating the candidate global optimization strategy in each round of game iteration: compare the candidate global optimization strategy with the bottom-line condition set of each close group; if any control instruction or running parameter in the candidate global optimization strategy exceeds the bottom-line condition of the corresponding close group, the bottom-line verification and rollback module is immediately triggered to terminate the current game iteration, roll back the strategy to the previous round of feasible solution, and issue an optimization danger alarm requiring manual intervention;

[0023] The weighted Nash equilibrium solution algorithm includes generating a global candidate optimization strategy using weighted cooperative game theory when there is no conflict in the optimization objectives of each close group; and generating a global candidate optimization strategy using weighted non-cooperative game theory when there is a conflict in the optimization objectives of each close group.

[0024] The strategy game and decision-making module is further configured to: identify the coupling relationship, conflict points, and strategy dependency strength among close groups during the strategy game process; for close groups with strong coupling relationships, perform weighted joint solution based on their dynamically adjusted game weight coefficients to generate collaborative optimization sub-strategies, wherein close groups with higher game weight coefficients enjoy higher decision priority in the joint solution process; for close groups with conflicts, introduce a weighted coordination compensation mechanism to adjust the optimization target weights of each close group or allocate compensation coefficients to make the strategy game converge to a feasible solution; in the weighted coordination compensation mechanism, the allocation of compensation coefficients is positively correlated with the dynamically adjusted game weight coefficients; and also includes a security constraint pre-verification unit, configured to perform rapid security pre-verification before sending candidate strategies to the virtual sandbox simulation module, and directly remove strategies that clearly exceed the security boundary from the game strategy set of the current round.

[0025] The virtual sandbox simulation module is further configured such that, after performing the pre-simulation, the returned pre-simulation results also include strategy sensitivity analysis data. The strategy sensitivity analysis data is used to identify the key parameter nodes that have the greatest impact on the results during the simulation process. Based on the strategy sensitivity analysis data, the strategy game and decision-making module prioritizes adjusting the strategies of the close groups corresponding to the key parameter nodes in subsequent iterations.

[0026] By adopting the above technical solutions, the strategy game and decision-making module forms a closed-loop iterative optimization mechanism with the sandbox simulation module through a two-way interactive channel. The weighted Nash equilibrium solution ensures the coordination of the global strategy, the bottom-line pre-verification unit eliminates dangerous strategies in the game stage, and sensitivity analysis guides the precise adjustment in subsequent iterations.

[0027] The present invention is further configured to include a strategy execution feedback and self-evolution module, which is connected to the hierarchical decision and execution module and the affinity analysis module, respectively. It is configured to continuously collect actual running data during the execution period after the optimization strategy is issued and executed, compare the actual running data with the simulation results of the corresponding strategy in the virtual sandbox environment, and generate an execution deviation report.

[0028] The intimacy analysis module is further configured to dynamically update the calculation factor weights of the intimacy index based on the execution deviation report, so that the division of intimacy groups can adapt to changes in the system's operating status.

[0029] By adopting the above technical solutions, the strategy execution feedback and self-evolution module form a complete closed loop of "simulation-execution-feedback-optimization", enabling the system to have continuous learning and self-evolution capabilities.

[0030] The invention is further configured such that the bottom-line verification and rollback module is connected to the strategy game and decision-making module, the virtual sandbox simulation module, and the hierarchical decision-making and execution module, respectively, and is further configured to store the bottom-line condition set of each close group. The bottom-line conditions include absolute bottom lines and warning bottom lines. The absolute bottom line is an insurmountable hard constraint, and the warning bottom line is a soft constraint that allows short-term overshoot but requires triggering a warning. During the strategy game, when a candidate global optimization strategy is detected to exceed the absolute bottom line of any close group, a forced rollback is immediately triggered, rolling the strategy back to the previous round's feasible solution, and generating an optimization danger alarm containing the type of exceeding the bottom line, the corresponding close group information, and suggested adjustment direction, which is pushed to the administrator terminal, requiring manual intervention to reset the game parameters or adjust the bottom-line conditions. After the virtual sandbox simulation is completed, when the running parameters of any close group in the simulation results are detected to exceed its absolute bottom line, a forced rollback is triggered, rolling the strategy back to the previous round's feasible solution, and generating an optimization danger alarm containing the type of exceeding the bottom line, the corresponding close group information, and suggested adjustment direction, which is pushed to the administrator terminal, requiring manual intervention to reset the game parameters or adjust the bottom-line conditions. When the bottom line is reached, a simulation rollback mechanism is triggered, marking the global candidate optimization strategy as invalid and rolling back to the strategy game and decision-making module to regenerate the strategy. At the same time, a detailed report including the bottom line breakthrough time, breakthrough magnitude, and chain reaction analysis during the simulation process is generated and pushed to the manager's terminal. When the simulation result is detected to exceed the warning bottom line but not the absolute bottom line, the bottom line verification and rollback module is configured not to trigger rollback, but to generate a warning message, which is sent to the hierarchical decision-making and execution module along with the risk level assessment result of the safety and effectiveness analysis module as an auxiliary basis for manual confirmation of medium-risk strategies. The bottom line verification and rollback module also includes a rollback counter, configured to record the number of consecutive rollbacks within the same optimization cycle. When the number of consecutive rollbacks exceeds a preset threshold, it is determined that the current optimization problem has no feasible solution, the automatic optimization process is forcibly suspended, and an emergency alarm is issued requiring deep manual intervention.

[0031] By adopting the above technical solution, the bottom-line verification and backoff module sets two layers of constraints: an absolute bottom line and an early warning bottom line. Verification is performed during the game phase and after the simulation is completed. The backoff counter prevents infinite loop optimization, forming a four-layer security protection system of "bottom-line pre-verification - game iteration - sandbox simulation - bottom-line post-verification - backoff circuit breaker".

[0032] In summary, the present invention has the following main beneficial effects:

[0033] This invention utilizes an edge grouping optimization module to independently perform local optimization on the edge computing nodes corresponding to each close group. It then establishes a two-way interactive channel between a strategy game and decision-making module and a virtual sandbox simulation module. During the game iteration process, the sandbox module is invoked in real-time for pre-simulation verification, and the game weight coefficients are dynamically adjusted based on the pre-simulation results. Simultaneously, an adversarial verification engine injects extreme perturbation scenarios for stress testing, forming a closed-loop iterative optimization mechanism of "game-simulation-re-game," significantly improving the robustness of the optimization strategy under complex conditions. By setting up a decision tracing unit to record key factors affecting the assessment and their weights and generating a decision tracing graph, it provides managers with interpretable decision-making basis. Furthermore, the hierarchical decision and execution module manages low-level... The system employs tiered operations for medium- and high-risk strategies, including automatic issuance, issuance after manual confirmation, and early warning prevention. The manual confirmation mechanism for medium-risk strategies, in particular, achieves a good balance between automated decision-making and manual safeguards. By establishing two layers of constraints—an absolute bottom line and an early warning bottom line—through a bottom line verification and rollback module, pre-verification and post-simulation verification of the bottom line are performed during the strategy game phase and after sandbox simulation, respectively. When the absolute bottom line is exceeded, a forced rollback is triggered, and an optimization danger alarm is issued. Simultaneously, a rollback counter records the number of consecutive rollbacks, and the automatic optimization process is urgently suspended when the threshold is exceeded. This forms a four-layer security protection system: "bottom line pre-verification - game iteration - sandbox simulation - bottom line post-verification - rollback circuit breaker," fundamentally ensuring the safety of microgrid operation. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0035] Figure 2 This is a flowchart of the intimacy analysis and intimacy group construction process of the present invention;

[0036] Figure 3 This is a flowchart illustrating the strategy game and decision-making process of the present invention.

[0037] Figure 4 This is a flowchart of the bottom line verification and rollback process of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0039] The embodiments of the present invention will now be described.

[0040] A microgrid energy efficiency analysis and auxiliary decision management system based on big data, such as Figure 1-4As shown, it includes a data acquisition and real-time mapping module, an intimacy analysis module, an intimacy group construction module, an edge grouping optimization module, a strategy game and decision-making module, a virtual sandbox simulation module, a security and efficiency analysis module, a hierarchical decision-making and execution module, and a bottom-line verification and rollback module.

[0041] The data acquisition and real-time mapping module is configured to collect multi-source heterogeneous data from the microgrid and construct a real-time digital mapping model that operates synchronously with the physical microgrid. Specifically, in this embodiment, the microgrid includes 3 photovoltaic power generation units, 2 wind power generation units, 2 sets of energy storage battery units, and 5 key load nodes.

[0042] The data acquisition and real-time mapping module collects data such as voltage, current, power, energy storage state of charge, ambient irradiance, and wind speed at a sampling frequency of 1Hz to 10Hz through a multi-sensor network deployed at each distributed energy node and load node. This data is then transmitted to edge computing nodes via a 5G private network and a fiber optic ring network. The real-time digital mapping model is constructed using digital twin technology and includes the microgrid's topology, equipment parameters, and real-time operating status. In this embodiment, the model update cycle is 1 minute, and an extended Kalman filter algorithm is used to correct the sensor data in real time, ensuring that the synchronization error with the physical microgrid is controlled within 2%.

[0043] The data acquisition and real-time mapping module also includes a reliability assessment unit, configured to calculate the synchronization error between the real-time digital mapping model and the physical microgrid in real time and generate a reliability index. In this embodiment, the reliability assessment unit calculates the root mean square error as the synchronization error by comparing the physical measurement value and the model simulation value at the same time. The reliability index is defined as 1 minus the ratio of the root mean square error to the reference error, but is limited to between 0 and 1. The reference error is dynamically set according to the equipment accuracy and communication delay. In this embodiment, the voltage measurement reference error is 2%, and the power reference error is 3%. When the reliability index is lower than 0.85, the system issues a model calibration warning; when the reliability index is higher than 0.95, the model is deemed reliable, allowing the subsequent optimization process to be triggered. In this embodiment, the reliability index is maintained between 0.92 and 0.98 during normal operation.

[0044] The affinity analysis module is connected to the data acquisition and real-time mapping module and is configured to perform correlation analysis on multiple parameter nodes in the real-time digital mapping model to calculate the affinity index. In this embodiment, the real-time digital mapping model contains 15 parameter nodes, each corresponding to a monitoring device in the microgrid, including photovoltaic inverters, wind turbine converters, energy storage battery management systems, and smart meters.

[0045] The intimacy analysis module determines the intimacy index based on the electrical distance between parameter nodes, data correlation coefficients, and historical rate of coordinated change. Specifically:

[0046] The electrical distance is calculated based on the impedance matrix between nodes and reflects the degree of electrical connection between two nodes in the power grid topology. Taking node A (PV1) and node B (PV2) as an example, their electrical distance is 0.25 pu; the electrical distance between node A and node C (energy storage 1) is 0.48 pu.

[0047] The data correlation coefficient is used to obtain the correlation of power data between two parameter nodes within a 24-hour sliding window, calculated using the Pearson correlation coefficient. The correlation coefficient between the power data of node A and node B is 0.92, and the correlation coefficient between node A and node C is 0.31.

[0048] The historical co-change rate is calculated based on the frequency of co-change between two parameter nodes in historical operational data. It is defined as the proportion of the number of times the two parameter nodes change in the same direction at the same time point to the total number of samples. The co-change rate between node A and node B is 87%, and the co-change rate between node A and node C is 23%.

[0049] Before calculating the intimacy index, each factor is first normalized to eliminate dimensional differences. This embodiment uses a min-maximum normalization method to map electrical distance, data correlation coefficient, and cooperative change rate to the interval between 0 and 1. For electrical distance, the normalized value is the distance minus the minimum electrical distance between all node pairs, divided by the difference between the maximum and minimum values. For the data correlation coefficient, since its range is -1 to 1, it is first added by 1 and then divided by 2 to convert it to a non-negative value. The cooperative change rate itself is between 0 and 1 and does not require additional normalization.

[0050] After normalization, the intimacy index is generated through weighted summation. In this embodiment, the weight coefficients are determined using the analytic hierarchy process (AHP). First, a judgment matrix is ​​constructed, and three experts compare the importance of electrical distance, data correlation coefficient, and historical rate of collaborative change pairwise. After calculating and normalizing the eigenvectors, preliminary weights are obtained, which pass a consistency test. Considering that data correlation has a more significant impact on intimacy in practical applications, the weights are adjusted to: electrical distance 0.3, data correlation coefficient 0.4, and rate of collaborative change 0.3. The intimacy index between node A and node B is 0.683, and the intimacy index between node A and node C is 0.337.

[0051] The intimacy group construction module is connected to the intimacy analysis module and is configured to construct corresponding internal devices into intimacy groups based on the intimacy index. In this embodiment, the intimacy index threshold is set to 0.5. The intimacy group construction module constructs an association network graph with each parameter node as a vertex and the intimacy index as the edge weight. The community detection algorithm (Louvain algorithm) is used to divide the association network graph, and the node groups with tightly connected internal connections and sparse external connections are identified as the intimacy groups. After calculation, the intimacy index between nodes A, B, and D (photovoltaic 3) is greater than 0.6, and they are divided into the first intimacy group (photovoltaic group); the intimacy index between nodes C and E (energy storage 2) is 0.55, and they are divided into the second intimacy group (energy storage group); the intimacy index between nodes F to J (5 load nodes) is 0.7 to 0.9, and they are divided into the third intimacy group (load group).

[0052] When the size of the identified intimate group exceeds the preset maximum size threshold (set to 8 nodes in this embodiment), the intimate group is further divided until the size of each intimate group meets the processing capacity of the edge computing node. In this embodiment, the size of each intimate group is less than 8, so secondary division is not required.

[0053] The intimate group construction module also configures a set of baseline conditions for each intimate group. These baseline conditions include equipment capacity limits (e.g., the maximum output power of a photovoltaic inverter does not exceed 110% of its rated value), safe operation thresholds (e.g., the state of charge of energy storage is not less than 10% and not more than 95%), carbon emission quotas (e.g., daily carbon emissions do not exceed 5 tons), and voltage and frequency boundaries (e.g., node voltage deviation does not exceed ±5%, and frequency deviation does not exceed ±0.5Hz). The baseline conditions are divided into absolute baselines and warning baselines. Absolute baselines are hard constraints that cannot be infringed, while warning baselines are soft constraints that allow short-term infringement but require triggering a warning.

[0054] The edge grouping optimization module is deployed on the edge computing nodes corresponding to each close group. It is configured to independently perform local energy efficiency optimization calculations based on the real-time mapping data corresponding to the close group when the confidence index is higher than a preset threshold (0.9 in this embodiment), generating candidate local optimization strategies. In this embodiment, the edge computing nodes use ARM architecture industrial-grade processors and deploy lightweight neural network models.

[0055] The first close-knit group (photovoltaic cluster) aims to maximize power generation efficiency and prediction accuracy. The objective function comprehensively considers both photovoltaic power generation efficiency and power prediction accuracy, balancing them through preset weights. Constraints include that the output of each photovoltaic unit does not exceed its rated capacity, and the rate of change of output between adjacent time points does not exceed an allowable value. A particle swarm optimization algorithm is used to solve the problem, generating three candidate strategies.

[0056] The second close-knit group (energy storage cluster) aims to optimize charge and discharge efficiency and extend battery life. The objective function comprehensively considers both charge and discharge efficiency and battery aging. A dynamic programming algorithm is used to solve the problem, and candidate strategies include peak shaving and valley filling, demand response, and reserve capacity reservation.

[0057] The third close-knit group (load cluster) aims at peak shaving and demand response, with the objective function comprehensively considering peak power reduction and user compensation costs. A genetic algorithm is used to solve the problem, and candidate strategies include off-peak operation, flexible load shedding, and emergency shutdown.

[0058] Each edge computing node operates independently, generates candidate local optimization strategies, and then uploads them to the strategy game and decision-making module.

[0059] The strategy game and decision-making module is connected to the edge grouping optimization module and establishes a two-way interactive channel with the virtual sandbox simulation module. It is configured to perform strategy games on multiple candidate local optimization strategies to generate a global candidate optimization strategy. In this embodiment, the strategy game and decision-making module is deployed on the server of the regional scheduling center.

[0060] The candidate local optimization strategies uploaded by each close-knit group are used as the initial strategy set for the game participants, and an initial game weight coefficient is assigned to each participant. The initial game weight coefficient is calculated based on the importance level, security risk level, and historical reliability factor of the corresponding close-knit group. In this embodiment, the weight of the first close-knit group (photovoltaic cluster) is 0.35, the weight of the second close-knit group (energy storage cluster) is 0.40, and the weight of the third close-knit group (load cluster) is 0.25.

[0061] The utility function of the i-th close-knit group is defined as a comprehensive function of economic benefits, carbon emission reduction benefits, and safety risks. Economic benefits are calculated based on electricity prices and power changes; carbon emission reduction benefits are calculated based on the ratio of carbon emissions to the baseline emission allowance; and the safety risk coefficient is quantified comprehensively based on factors such as voltage deviation and equipment load rate. In this embodiment, the preference coefficients for economic benefits, carbon emission reduction benefits, and safety risks are set to 0.4, 0.3, and 0.3, respectively.

[0062] A weighted game model is constructed, with the objective function being to maximize the sum of weighted utilities of each close group. The game weight coefficients of each close group are dynamically adjusted during game iterations based on pre-simulation results from the virtual sandbox simulation module: when a candidate strategy corresponding to a close group exhibits a high security risk in the pre-simulation, the game weight coefficient of that close group is decreased; when it exhibits a high energy efficiency benefit, the weight coefficient is increased. The adjustment magnitude is related to the rate of change of energy efficiency benefit and the rate of change of security risk, and a preset adjustment coefficient controls the adjustment step size.

[0063] In this embodiment, after the first round of pre-simulation, the first intimate group strategy has a low safety risk but a moderate energy efficiency benefit, and its weight is adjusted from 0.35 to 0.33; the second intimate group strategy shows good energy efficiency benefit (7.2%), and its weight is adjusted from 0.40 to 0.44; the third intimate group strategy has a certain safety risk, and its weight is adjusted from 0.25 to 0.23.

[0064] A weighted Nash equilibrium solution algorithm is employed to determine the global candidate optimization strategies that can achieve equilibrium among the close groups through iterative calculation. The algorithm steps include: first, initializing the strategies of each close group as the strategies with the highest utility values ​​among the candidate strategies; then, in each iteration, updating the strategies of each close group sequentially, selecting the strategy that maximizes the utility of the current group under the current strategies of other close groups; calculating the change in strategy between the current and previous rounds, and if the change is less than the convergence threshold (0.01 in this embodiment), it is determined that convergence has reached Nash equilibrium, and the current strategy set is output; if the number of iterations exceeds the maximum number of iterations (20 in this embodiment) and convergence has not yet occurred, the current optimal strategy set is output.

[0065] When there is no conflict in the optimization objectives of the close-knit groups, a weighted cooperative game is used to generate a global candidate optimization strategy; when there is a conflict, a weighted non-cooperative game is used to generate a global candidate optimization strategy. In this embodiment, the charging and discharging strategy of the energy storage group and the power generation strategy of the photovoltaic group are time-coupled. The global candidate optimization strategy is obtained by solving the weighted cooperative game, which includes the output power setpoint of each photovoltaic unit, the charging and discharging power curve of each energy storage unit, and the control commands of each load node.

[0066] The strategy game and decision-making module also includes a bottom-line pre-verification unit, configured to perform a bottom-line pre-verification before sending the candidate global optimization strategy to the virtual sandbox simulation module after generating iterative candidate strategies in each round of game. This involves comparing the candidate global optimization strategy with the bottom-line condition sets of each close group. If any control command or operating parameter in the candidate global optimization strategy exceeds the bottom-line condition of the corresponding close group, the bottom-line verification and rollback module is immediately triggered, terminating the current game iteration, rolling back the strategy to the previous feasible solution, and issuing an optimization danger alarm requiring manual intervention. For example, in a strategy generated in a certain iteration, the target value of the energy storage unit's state of charge is set to 96%, exceeding the absolute bottom line of 95%, triggering the bottom-line pre-verification, and the strategy is rejected.

[0067] The strategy game and decision-making module is further configured to: identify the coupling relationships, conflict points, and strategy dependency strengths among close groups during the strategy game process. For close groups with strong coupling relationships, a weighted joint solution is performed based on their dynamically adjusted game weight coefficients to generate collaboratively optimized sub-strategies, where close groups with higher game weight coefficients enjoy higher decision priority in the joint solution. For close groups with conflicts, a weighted coordination and compensation mechanism is introduced, adjusting the optimization objective weights of each close group or allocating compensation coefficients to bring the strategy game to a feasible solution; the allocation of compensation coefficients is positively correlated with the dynamically adjusted game weight coefficients.

[0068] In addition, a security constraint pre-verification unit is included, configured to perform rapid security pre-verification before sending candidate strategies to the virtual sandbox simulation module, directly eliminating strategies that clearly exceed the security boundary from the game strategy set for the current round. After performing the pre-simulation, the virtual sandbox simulation module returns pre-simulation results including strategy sensitivity analysis data, used to identify key parameter nodes that have the greatest impact on the results during the simulation. Based on this data, the strategy game and decision-making module prioritizes adjusting the strategies of the close groups corresponding to the key parameter nodes in subsequent iterations.

[0069] The virtual sandbox simulation module is connected to the strategy game and decision-making module, and is configured to simulate the execution of the global candidate optimization strategy in a virtual sandbox environment independent of the real-time digital mapping model, generating simulation results. In this embodiment, the virtual sandbox environment uses a simulation engine based on model predictive control, with a simulation step size of 1 minute and a simulation duration of 24 hours.

[0070] First, the current operating status data of the real-time digital mapping model is captured as the initial boundary conditions of the virtual sandbox environment, including the voltage, power, and energy storage charge status of each node at the current moment.

[0071] In the virtual sandbox environment, the global candidate optimization strategy is driven to execute according to the preset simulation time scale and event sequence, and the state change data of each key node during the execution process are recorded, including voltage curves, power flow, energy storage state of charge changes, carbon emission accumulation, etc., to form simulation results.

[0072] The virtual sandbox simulation module includes an adversarial verification engine configured to inject preset extreme disturbance scenarios into the virtual sandbox environment for stress testing. In this embodiment, the extreme disturbance scenarios include: a sudden drop in renewable energy output simulating cloud cover causing a 60% decrease in photovoltaic output within 5 minutes; a sudden load surge simulating concentrated charging of electric vehicles causing a 40% increase in load within 10 minutes; a communication interruption simulating a 30-minute communication interruption between an edge node and the control center; and equipment failure simulating a sudden failure and shutdown of an energy storage unit. The adversarial verification engine injects the above disturbance scenarios sequentially to simulate the performance of the strategy under extreme conditions.

[0073] When there are multiple global candidate optimization strategies, the virtual sandbox simulation module runs multiple independent virtual sandbox environments in parallel to simulate the execution of each strategy. In this embodiment, three global candidate optimization strategies, namely strategy A, strategy B, and strategy C, are generated, and three sandbox environments are run in parallel for simulation.

[0074] The security and performance analysis module is connected to the virtual sandbox simulation module and is configured to analyze the simulation results, evaluate the execution effect and security risks, and assess the risk level.

[0075] The performance evaluation unit compares the simulation results with the preset energy efficiency improvement targets and calculates the expected energy efficiency gains of each global candidate optimization strategy.

[0076] The baseline value is determined as follows: Operational data from the past 30 days without optimization intervention is collected, and after removing outliers, the average energy efficiency is calculated and used as the baseline energy efficiency. In this embodiment, the baseline energy efficiency is 82%. The baseline carbon emissions are determined similarly: daily carbon emissions data for the same period are collected, and the median is taken as the baseline. In this embodiment, the baseline carbon emissions are 5.2 tons / day.

[0077] Comprehensive energy efficiency improvement calculation: Comprehensive energy efficiency is defined as the sum of total effective output energy and net energy discharge of energy storage divided by the sum of total input energy and net energy charging of energy storage, and then multiplied by 100%. The comprehensive energy efficiency improvement is the comprehensive energy efficiency minus the baseline energy efficiency.

[0078] Carbon emission reduction calculation: Total carbon emissions are the sum of the output of each power source multiplied by its corresponding carbon emission factor, then multiplied by the time period, plus grid loss carbon emissions. Carbon emission reduction is the baseline carbon emission amount minus the total carbon emissions amount, divided by the baseline carbon emissions amount, and then multiplied by 100%. The carbon emission factors are set as follows: thermal power 0.85 kg / kWh, photovoltaic 0.05 kg / kWh, and wind power 0.02 kg / kWh.

[0079] Comprehensive energy efficiency benefit score: The ratio of comprehensive energy efficiency improvement to energy efficiency improvement target (3% in this example) and the ratio of carbon emission reduction to carbon emission reduction target (5% in this example) are weighted and summed, with each weight being 0.5.

[0080] Calculations show that Strategy A's expected energy efficiency gains are a 2.8% increase in overall energy efficiency and a 4.2% reduction in carbon emissions, with an overall score of 0.885; Strategy B's expected energy efficiency gains are a 3.5% increase in overall energy efficiency and a 6.1% reduction in carbon emissions, with an overall score of 1.195; and Strategy C's expected energy efficiency gains are a 4.1% increase in overall energy efficiency and a 7.3% reduction in carbon emissions, with an overall score of 1.415.

[0081] The safety analysis unit checks the simulation results for abnormal operating states such as equipment overload, voltage exceeding limits, frequency instability, and excessive carbon emissions, and quantifies the severity of these abnormal operating states. Severity is quantified using a weighted scoring method: the ratio of the measured value of each abnormal parameter to the safety threshold minus 1, multiplied by the corresponding weight, and then summed. For example, for voltage exceeding limits, the ratio of the voltage deviation exceeding 5% to 5% is multiplied by a weight of 0.4. When the severity is greater than 1.0, it is considered an absolute bottom line breach; when the severity is between 0 and 1.0, it is considered a warning bottom line breach; when the severity is 0, there is no abnormality.

[0082] In the simulation results of Strategy A, under a certain load surge disturbance, the current of a certain feeder reached 98% of the rated value, which was within the limit and the severity was 0. In the simulation results of Strategy B, under the disturbance of a sudden drop in renewable energy output, the node voltage dropped to a minimum of 0.94 pu, which was within the limit and the severity was 0. In the simulation results of Strategy C, under the disturbance of equipment failure, the risk of overcharging of the energy storage unit caused the voltage to temporarily rise to 1.08 pu, exceeding the absolute bottom line of 1.05 pu, lasting for 12 seconds, with a severity of 1.2.

[0083] The risk rating unit comprehensively assesses the risk level based on the expected energy efficiency benefits and the severity of abnormal operating conditions. The assessment rules are as follows: a comprehensive score ≥ 0.9 and no abnormalities are assessed as low risk; a comprehensive score ≥ 0.9 and a breach of the warning threshold are assessed as medium risk; a comprehensive score ≥ 0.9 and an breach of the absolute threshold are assessed as high risk; a comprehensive score < 0.9 and no abnormalities are assessed as low risk; a comprehensive score < 0.9 and any breach of the threshold are assessed as medium or high risk.

[0084] Strategy A has a comprehensive score of 0.885 and no abnormalities, and is rated as low risk; Strategy B has a comprehensive score of 1.195 and no abnormalities, and is rated as low risk; Strategy C has a comprehensive score of 1.415 and there is an absolute bottom line breach, and is rated as high risk.

[0085] During the risk level assessment process, the decision sourcing unit records the key factors influencing the assessment and their weights, generating a decision sourcing diagram. For strategy C, the key factors influencing the assessment include: initial state of charge of the energy storage unit (weight 0.35), disturbance type (weight 0.25), voltage response time (weight 0.20), and control command execution delay (weight 0.20). The decision sourcing diagram displays the contribution path of each factor to the risk assessment in a tree structure, providing managers with interpretable decision-making basis.

[0086] The hierarchical decision-making and execution module is connected to the safety and efficiency analysis module and is configured to execute decision-making operations based on risk level assessment. Low-risk strategies are automatically sent to the corresponding field execution devices via an encrypted communication protocol. Strategies A and B are automatically sent, and devices such as photovoltaic inverters, energy storage converters, and load controllers receive and execute the instructions.

[0087] The medium-risk strategy generates a detailed report including global candidate optimization strategies, expected energy efficiency gains, abnormal operating states, and a decision tracing diagram. This report is pushed to the administrator's terminal, and the strategy awaits confirmation or modification instructions from the administrator's terminal before being issued. This embodiment does not include a medium-risk strategy.

[0088] High-risk strategies generate warning messages and push them to the administrator's terminal, while preventing further distribution. Strategy C is marked as high-risk, a warning message is pushed, and the administrator intervenes to adjust the strategy parameters and then re-optimizes the game.

[0089] The bottom line verification and rollback module is connected to the strategy game and decision-making module, the virtual sandbox simulation module and the hierarchical decision-making and execution module, respectively, and is configured to perform bottom line verification. When the bottom line condition is exceeded, the rollback mechanism is triggered and an optimization danger alarm is issued.

[0090] The storage system comprises a set of baseline conditions for each close-knit group, including absolute baselines and warning baselines. Absolute baselines are hard constraints that cannot be infringed, such as equipment capacity limits and voltage / frequency boundaries. Warning baselines are soft constraints that are allowed to be exceeded in the short term but require an immediate warning, such as the energy storage state of charge being allowed to slightly exceed the limit for a short period of time but requiring an immediate warning.

[0091] During the strategic game, when a candidate global optimization strategy is detected to exceed the absolute threshold of any close group, a forced rollback is immediately triggered, reverting the strategy to the feasible solution of the previous round, and generating an optimization danger alarm that is pushed to the manager's terminal. In this embodiment, the aforementioned energy storage state of charge target value of 96% exceeds the 95% absolute threshold, triggering a forced rollback.

[0092] After the virtual sandbox simulation is completed, when the running parameters of any close group in the simulation results exceed its absolute limit, a simulation rollback mechanism is triggered. The global candidate optimization strategy is marked as invalid, and the simulation is rolled back to the strategy game and decision-making module to regenerate the strategy. In this embodiment, strategy C experiences a voltage spike to 1.08 pu during the simulation, exceeding the absolute limit of 1.05 pu, triggering a simulation rollback. The strategy is marked as invalid, and the game module regenerates the strategy. Simultaneously, a detailed report is generated, including the time of the limit breach (3 hours and 15 minutes of simulation), the breach magnitude (0.03 pu), and a chain reaction analysis (causing voltage fluctuations in adjacent nodes), and is pushed to the administrator terminal.

[0093] When the simulation result is detected to exceed the warning threshold but not the absolute threshold, no rollback is triggered, but a warning message is generated and sent to the hierarchical decision-making and execution module along with the risk level assessment result, as an auxiliary basis for manual confirmation of medium-risk strategies.

[0094] The bottom-line verification and rollback module also includes a rollback counter, configured to record the number of consecutive rollbacks within the same optimization cycle. The rollback counter threshold is determined based on statistical analysis of historical system operation data. Optimization operation data from the past 90 days is collected, and the distribution of the maximum number of consecutive rollbacks per day is statistically analyzed. The number corresponding to 95% of the cumulative frequency is selected as the threshold to ensure that an emergency shutdown is not triggered under 95% normal operating conditions. In this embodiment, the number of consecutive rollbacks corresponding to 95% of the cumulative frequency is 3. When the number of consecutive rollbacks exceeds 3, the current optimization problem is determined to be unsolvable, the automatic optimization process is forcibly paused, and an emergency alarm is issued requiring deep manual intervention.

[0095] The system also includes a strategy execution feedback and self-evolution module, which is connected to the hierarchical decision-making and execution module and the affinity analysis module, respectively. It is configured to continuously collect actual operational data during execution after the optimized strategy is issued and executed, compare the actual operational data with the simulation results of the corresponding strategy in a virtual sandbox environment, and generate an execution deviation report. In this embodiment, the actual comprehensive energy efficiency improvement after the execution of strategy A is 2.9%, a deviation of 0.1% from the simulated value of 2.8%; carbon emissions are reduced by 4.0%, a deviation of 0.2% from the simulated value of 4.2%. The execution deviation report records these deviations.

[0096] The intimacy analysis module dynamically updates the weights of the intimacy index calculation factors based on the execution deviation report. The dynamic weight update employs gradient descent, with the loss function defined as the sum of squared differences between predicted and actual values. The weights are updated by calculating the partial derivatives of the loss function with respect to each weight, combined with the learning rate (0.05 in this embodiment). The updated weights must satisfy the constraint that the sum of all weights is 1 and all are non-negative. Due to the large prediction deviation in carbon emissions, the weights of carbon emission-related factors are adjusted from 0.3 to 0.35, allowing the division of intimacy groups to adapt to changes in the system's operating state.

[0097] Example 2:

[0098] This embodiment is basically the same as Embodiment 1, except that the weighted game model in the strategy game and decision-making module adopts a different weight coefficient adjustment strategy. In this embodiment, the initial game weight coefficients are calculated based on the historical reliability factors of the corresponding close groups, and the reliability factors are dynamically updated according to indicators such as equipment failure rate and communication success rate. As shown in Table 1, the initial weight coefficients and historical reliability data of the three close groups are presented.

[0099] Table 1:

[0100] Intimate Group Equipment failure rate (times / year) Communication success rate (%) Historical reliability factor Initial weighting coefficients Photovoltaic cluster 0.5 99.2 0.95 0.32 Energy storage cluster 1.2 98.5 0.88 0.38 load group 0.8 99.5 0.92 0.30

[0101] During the game iteration process, the weight coefficients are dynamically adjusted based on the pre-simulation results fed back by the virtual sandbox simulation module. This embodiment sets up 5 rounds of iteration, and the changes in the weight coefficients in each round are shown in Table 2.

[0102] Table 2:

[0103] Iteration rounds Photovoltaic cluster weight Energy storage cluster weight Load group weight Utility function value 1 0.32 0.38 0.30 78.5 2 0.30 0.42 0.28 82.3 3 0.28 0.45 0.27 85.1 4 0.27 0.46 0.27 85.8 5 0.27 0.46 0.27 85.9

[0104] After five rounds of iteration, the utility function value converged to 85.9. Due to the higher energy efficiency (8.5%) and lower safety risk shown in the pre-simulation, the weight coefficient of the energy storage group increased from 0.38 to 0.46, while the weights of the photovoltaic group and the load group decreased accordingly.

[0105] Example 3:

[0106] This embodiment is basically the same as Embodiment 1, except for the number of virtual sandbox environments running in parallel in the virtual sandbox simulation module and the setting of the stress test scenario. In this embodiment, the microgrid scale is expanded to include 10 photovoltaic units, 5 wind power units, 8 energy storage units, and 20 load nodes, totaling 43 parameter nodes. After affinity analysis, 6 affinity groups are formed. The virtual sandbox simulation module is configured to run 6 independent virtual sandbox environments in parallel, simulating the execution of 6 candidate global optimization strategies respectively.

[0107] The extreme disturbance scenarios injected into the adversarial verification engine have been expanded to eight types, including: a sudden drop in renewable energy output (PV output decreases by 70% within 5 minutes), a sudden increase in renewable energy output (wind power output increases by 80% within 10 minutes), a sudden load surge (electric vehicle charging is concentrated, and the load increases by 50% within 15 minutes), a sudden load drop (industrial users experience sudden shutdowns, and the load decreases by 40% within 5 minutes), a communication interruption (communication between edge nodes and the control center is interrupted for 20 minutes), equipment failure (critical energy storage unit fails and exits the system), grid-side failure (upstream grid voltage drops by 30%), and extreme weather (continuous rainy weather causes PV output to remain below the rated value by 20%). The simulation results are shown in Table 3.

[0108] Table 3:

[0109] Strategy Number Overall energy efficiency improvement carbon emissions reduction Number of abnormal states Bottom line break Risk level Strategy 1 3.2% 4.5% 0 none Low risk Strategy 2 4.1% 5.8% 1 none Low risk Strategy 3 5.3% 7.2% 2 Early warning bottom line Medium risk Strategy 4 4.8% 6.5% 1 none Low risk Strategy 5 6.1% 8.3% 4 absolute bottom line High risk Strategy 6 3.5% 4.8% 0 none Low risk

[0110] Strategy 5 was flagged as high-risk due to voltage exceeding the absolute limit during simulation, triggering the limit verification and rollback module, forcing a rollback, and issuing an optimization hazard alarm. Strategy 3 was assessed as medium-risk due to a potential limit breach and was pushed to the administrator terminal for confirmation. All other strategies were assessed as low-risk and executed automatically.

[0111] Example 4:

[0112] This embodiment provides a microgrid energy efficiency analysis and auxiliary decision management method based on big data, applied to the aforementioned system, including the following steps:

[0113] Step S1: Collect multi-source heterogeneous data from the microgrid and construct a real-time digital mapping model that operates synchronously with the physical microgrid. Specifically, deploy IoT sensors to collect data such as voltage, current, power, energy storage state of charge, ambient irradiance, and wind speed at sampling frequencies from 1Hz to 10Hz. Use digital twin technology to construct a real-time digital mapping model with a model update cycle of 1 minute.

[0114] Step S2: Perform correlation analysis on multiple parameter nodes in the real-time digital mapping model and calculate the affinity index between any two parameter nodes. Specifically, obtain the electrical distance, data correlation coefficient, and historical co-variance rate between nodes, and calculate the affinity index by weighted summation, and pre-process each factor with min-max normalization.

[0115] Step S3: Based on the intimacy index, construct an intimacy group for the internal devices corresponding to parameter nodes whose intimacy index exceeds a preset threshold, forming multiple independent intimacy groups, and configure a set of bottom-line conditions for each intimacy group. Specifically, construct an association network graph with parameter nodes as vertices and intimacy index as edge weights, and use a community detection algorithm to divide it into intimacy groups.

[0116] Step S4: The edge computing nodes corresponding to each close group independently perform local energy efficiency optimization calculations to generate candidate local optimization strategies for the corresponding close group. Specifically, each edge node, based on the real-time mapping data of its group, aims to maximize operating efficiency and minimize carbon emissions, and uses particle swarm optimization, dynamic programming, or genetic algorithms to solve the problem, generating 3 to 5 candidate local optimization strategies.

[0117] Step S5: Under a unified game framework, conduct strategy games on multiple candidate local optimization strategies to generate global candidate optimization strategies. Specifically, the strategies uploaded by each close group are used as the strategy set of the game participants. Initial game weight coefficients are configured, a weighted game model is constructed, and a weighted Nash equilibrium solution algorithm is used for iterative solution. During the game, bidirectional interaction is performed with the virtual sandbox simulation module, and the weight coefficients are dynamically adjusted based on the pre-simulation results.

[0118] Step S6: Simulate the execution of the global candidate optimization strategy in an independent virtual sandbox environment, and inject preset extreme perturbation scenarios into the virtual sandbox environment for stress testing, generating simulation results. Specifically, capture the current running state as the initial boundary condition, drive the strategy execution with a simulation step size of 1 minute, and inject at least 3 extreme perturbation scenarios for stress testing.

[0119] Step S7: Analyze the simulation results, evaluate the effectiveness and risk of the global candidate optimization strategy, record the key factors affecting the assessment and their weights to generate a decision source map, and assess the risk level. Specifically, calculate the expected energy efficiency benefits (including overall energy efficiency improvement and carbon emission reduction), detect abnormal operating states, and comprehensively assess low, medium, and high risk levels.

[0120] Step S8: Based on the risk level assessment results, execute the hierarchical decision-making operation of automatically issuing and generating auxiliary decision-making schemes with the decision tracing diagram for manual confirmation or generating early warning information and preventing the issuance of such schemes.

[0121] The temporal sequence of the above method steps is as follows: Step S1 is a continuously running data acquisition and mapping process; Steps S2 to S3 are executed every 15 minutes based on Step S1; Steps S4 to S5 are triggered when the confidence index is higher than the threshold, with an optimization cycle of 5 minutes; Steps S6 to S7 are executed immediately after Step S5 is completed; Step S8 is executed after Step S7 is completed. A closed-loop control process of "perception-analysis-optimization-simulation-decision-execution" is formed between each step, with a total optimization cycle not exceeding 10 minutes.

[0122] In summary, this invention, through the aforementioned multiple embodiments, elaborates in detail the specific implementation of a microgrid energy efficiency analysis and auxiliary decision management system based on big data. This includes the credibility assessment of the data acquisition and real-time mapping module, the affinity analysis and affinity group construction, edge group optimization, the bidirectional interaction between the strategy game and decision-making module and the virtual sandbox simulation module, the effect evaluation and risk rating in safety and efficiency analysis, the low / medium / high risk graded operation of the graded decision-making and execution module, the four-layer security protection mechanism of the bottom-line verification and rollback module, and the continuous optimization capability of the strategy execution feedback and self-evolution module.

[0123] Although embodiments of the present invention have been shown and described, these specific embodiments are merely explanations of the invention and are not intended to limit it. The specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. After reading this specification, those skilled in the art may make modifications, substitutions, and variations to the embodiments as needed without departing from the principles and spirit of the invention, but such modifications, substitutions, and variations are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A microgrid energy efficiency analysis and auxiliary decision management system based on big data, characterized in that, include: The data acquisition and real-time mapping module is configured to acquire multi-source heterogeneous data from the microgrid and construct a real-time digital mapping model that operates synchronously with the physical microgrid. The intimacy analysis module is configured to perform correlation analysis on multiple parameter nodes in the real-time digital mapping model and calculate the intimacy index. The intimacy group construction module is configured to construct the corresponding internal devices into an intimacy group based on the intimacy index; The edge grouping optimization module is deployed on the edge computing nodes corresponding to each close group. It is configured to perform local energy efficiency optimization calculations independently based on the real-time mapping data corresponding to its close group and generate candidate local optimization strategies. The strategy game and decision-making module is configured to perform strategy games on multiple candidate local optimization strategies to generate global candidate optimization strategies. The virtual sandbox simulation module is configured to simulate the execution of the global candidate optimization strategy in a virtual sandbox environment independent of the real-time digital mapping model, and generate simulation results. The safety and performance analysis module is configured to analyze the simulation results, evaluate the execution effect and safety risks, and assess the risk level. The hierarchical decision-making and execution module is configured to perform decision-making operations based on the risk level assessment. The bottom line verification and rollback module is configured to perform bottom line verification. When the bottom line condition is exceeded, a rollback mechanism is triggered and an optimization danger alarm is issued.

2. The system according to claim 1, characterized in that, The data acquisition and real-time mapping module is connected to the proximity analysis module. The data acquisition and real-time mapping module also includes a credibility assessment unit, configured to calculate the synchronization error between the real-time digital mapping model and the physical microgrid in real time and generate a credibility index. The edge grouping optimization module is connected to the strategy game and decision-making module, and is further configured to perform the local energy efficiency optimization calculation when the credibility index is higher than a preset threshold.

3. The system according to claim 1, characterized in that, The intimacy analysis module is connected to the intimacy group construction module. The real-time digital mapping model contains multiple parameter nodes, each corresponding to a monitoring device in the microgrid. The intimacy analysis module is further configured to comprehensively determine the intimacy index based on at least one of the following factors: electrical distance between parameter nodes, data correlation coefficient, and historical cooperative change rate. The intimacy group construction module is further configured to configure a set of bottom-line conditions for each intimacy group. The bottom-line conditions include at least one insurmountable operational constraint among equipment capacity limit, safe operation threshold, carbon emission quota, and voltage frequency boundary.

4. The system according to claim 3, characterized in that, The intimacy analysis module is further configured as follows: The electrical distance between any two parameter nodes in the real-time digital mapping model is obtained, and the electrical distance is calculated based on the impedance matrix between the nodes. The data correlation coefficient of the two parameter nodes within a preset time window is obtained, and the data correlation coefficient is calculated based on the Pearson correlation coefficient or mutual information. The coordinated change rate of the two parameter nodes in historical running data is obtained, and the coordinated change rate is calculated based on the frequency of the two parameter nodes changing in the same direction at the same time point. The intimacy index is generated by combining the electrical distance, data correlation coefficient, and cooperative change rate through a weighted summation method. The intimate group construction module is further configured as follows: using each parameter node as a vertex and the intimacy index as the edge weight, a relational network graph is constructed; a community detection algorithm is used to divide the relational network graph, and the node groups with tight internal connections and sparse external connections are identified as the intimate groups; when the size of the identified intimate group exceeds the preset maximum size threshold, the intimate group is divided a second time until the size of each intimate group meets the processing capacity of the edge computing node.

5. The system according to claim 1, characterized in that, The virtual sandbox simulation module is connected to the strategy game and decision-making module, and is further configured as follows: The current running status data of the real-time digital mapping model is captured and used as the initial boundary conditions of the virtual sandbox environment; In the virtual sandbox environment, the global candidate optimization strategy is executed according to the preset simulation time scale and event sequence, and the state change data of each key node during the execution process is recorded to form the simulation results; The virtual sandbox simulation module includes an adversarial verification engine, configured to inject preset extreme disturbance scenarios into the virtual sandbox environment to stress test the global candidate optimization strategy. The extreme disturbance scenarios include at least one of the following: sudden drop in new energy output, sudden increase in load, communication interruption, and equipment failure. When there are multiple global candidate optimization strategies, the virtual sandbox simulation module is also configured to run multiple independent virtual sandbox environments in parallel, and simulate and execute multiple global candidate optimization strategies respectively, so as to generate multiple simulation results in parallel.

6. The system according to claim 1, characterized in that, The security and performance analysis module is connected to the virtual sandbox simulation module and includes: The effect evaluation unit is configured to compare the simulation results with the preset energy efficiency improvement target and calculate the expected energy efficiency benefits of the global candidate optimization strategy. The safety analysis unit is configured to check whether there is at least one preset abnormal operating state among equipment overload, voltage over-limit, frequency instability and carbon emission over-limit in the simulation results, and to quantify the severity of the abnormal operating state. The risk rating unit is configured to comprehensively assess the risk level of the global candidate optimization strategy based on the expected energy efficiency gains and the severity of the abnormal operating state. The decision tracing unit is configured to record the key factors affecting the assessment and their weights during the risk level assessment process, and generate a decision tracing diagram. The key factors include abnormal operating conditions that lead to an increase in the risk level and the sources of contribution to the expected energy efficiency benefits. The hierarchical decision-making and execution module is connected to the safety and efficiency analysis module and is further configured as follows: when the risk level is assessed as low risk, the global candidate optimization strategy is automatically sent to the corresponding field execution device through an encrypted communication protocol; when the risk level is assessed as medium risk, a detailed report containing the global candidate optimization strategy, the expected energy efficiency benefits, the abnormal operating status, and the decision tracing diagram is generated, pushed to the management terminal, and waits for confirmation or modification instructions from the management terminal before sending the global candidate optimization strategy; when the risk level is assessed as high risk, the early warning information is generated and pushed to the management terminal, and the sending of the global candidate optimization strategy is prevented.

7. The system according to claim 1, characterized in that, The strategy game and decision-making module is connected to the edge grouping optimization module and establishes a two-way interaction channel with the virtual sandbox simulation module. The strategy game and decision-making module is further configured as follows: The candidate local optimization strategies uploaded by each close group are used as the initial strategy set for the game participants, and an initial game weight coefficient is configured for each game participant. The initial game weight coefficient is calculated based on at least one parameter among the importance level, security risk level and historical reliability factor of the corresponding close group. During the game, the candidate global optimization strategy for the current round is sent to the virtual sandbox simulation module for pre-simulation, and the pre-simulation results fed back by the virtual sandbox simulation module are received. The pre-simulation results include the estimated value of the strategy execution effect and the security risk assessment value. A weighted game model is constructed, wherein the objective function of the weighted game model is to maximize the sum of the weighted utilities of each close group. The game weight coefficient of each close group is dynamically adjusted during the game iteration process based on the pre-simulation results fed back by the virtual sandbox simulation module: when the candidate strategy corresponding to a close group shows a high security risk in the pre-simulation, the game weight coefficient of the close group is reduced; when the candidate strategy corresponding to a close group shows a high energy efficiency gain in the pre-simulation, the game weight coefficient of the close group is increased. The weighted Nash equilibrium solution algorithm is adopted to determine the global candidate optimization strategy that can achieve equilibrium among each close group through iterative calculation. In each iteration, the virtual sandbox simulation module is called to pre-simulate and verify the current candidate strategy until the convergence condition is met or the preset maximum number of iterations is reached. The strategy game and decision-making module also includes a bottom-line pre-verification unit, configured to perform a bottom-line pre-verification before sending the candidate global optimization strategy to the virtual sandbox simulation module after generating the candidate global optimization strategy in each round of game iteration: compare the candidate global optimization strategy with the bottom-line condition set of each close group; if any control instruction or running parameter in the candidate global optimization strategy exceeds the bottom-line condition of the corresponding close group, the bottom-line verification and rollback module is immediately triggered to terminate the current game iteration, roll back the strategy to the previous round of feasible solution, and issue an optimization danger alarm requiring manual intervention; The weighted Nash equilibrium solution algorithm includes: when there is no conflict in the optimization objectives of each close group, a weighted cooperative game is used to generate a global candidate optimization strategy; when there is a conflict in the optimization objectives of each close group, a weighted non-cooperative game is used to generate a global candidate optimization strategy. The strategy game and decision-making module is further configured to: identify the coupling relationship, conflict points, and strategy dependency strength among close groups during the strategy game process; for close groups with strong coupling relationships, perform weighted joint solution based on their dynamically adjusted game weight coefficients to generate collaborative optimization sub-strategies, wherein close groups with higher game weight coefficients enjoy higher decision priority in the joint solution process; for close groups with conflicts, introduce a weighted coordination compensation mechanism to adjust the optimization target weights of each close group or allocate compensation coefficients to make the strategy game converge to a feasible solution; in the weighted coordination compensation mechanism, the allocation of compensation coefficients is positively correlated with the dynamically adjusted game weight coefficients; and also includes a security constraint pre-verification unit, configured to perform rapid security pre-verification before sending candidate strategies to the virtual sandbox simulation module, and directly remove strategies that clearly exceed the security boundary from the game strategy set of the current round. The virtual sandbox simulation module is further configured such that, after performing the pre-simulation, the returned pre-simulation results also include strategy sensitivity analysis data. The strategy sensitivity analysis data is used to identify the key parameter nodes that have the greatest impact on the results during the simulation process. Based on the strategy sensitivity analysis data, the strategy game and decision-making module prioritizes adjusting the strategies of the close groups corresponding to the key parameter nodes in subsequent iterations.

8. The system according to claim 1, characterized in that, It also includes a strategy execution feedback and self-evolution module, which is connected to the hierarchical decision and execution module and the affinity analysis module, respectively. It is configured to continuously collect actual running data during the execution period after the optimization strategy is issued and executed, compare the actual running data with the simulation results of the corresponding strategy in the virtual sandbox environment, and generate an execution deviation report. The intimacy analysis module is further configured to dynamically update the calculation factor weights of the intimacy index based on the execution deviation report, so that the division of intimacy groups can adapt to changes in the system's operating status.

9. The system according to claim 1, characterized in that, The bottom-line verification and rollback module is connected to the strategy game and decision-making module, the virtual sandbox simulation module, and the hierarchical decision-making and execution module, respectively, and is further configured as follows: Store the set of bottom line conditions for each intimate group. The bottom line conditions include absolute bottom lines and warning bottom lines. The absolute bottom lines are hard constraints that cannot be crossed, and the warning bottom lines are soft constraints that allow short-term crossing but require triggering a warning. During the strategic game, when a candidate global optimization strategy is detected to exceed the absolute bottom line of any close group, a forced rollback is immediately triggered, rolling the strategy back to the previous round's feasible solution. An optimization danger alarm is generated, which includes the type of exceeding the bottom line, the corresponding close group information, and the suggested adjustment direction. This alarm is then pushed to the administrator terminal, requiring manual intervention to reset the game parameters or adjust the bottom line conditions. After the virtual sandbox simulation is completed, when it is detected that the running parameters of any close group in the simulation results exceed its absolute bottom line, the simulation rollback mechanism is triggered, the global candidate optimization strategy is marked as invalid, and the strategy game and decision module is rolled back to regenerate the strategy. At the same time, a detailed report containing the bottom line breakthrough time, breakthrough magnitude and chain reaction analysis during the simulation process is generated and pushed to the administrator terminal. When the simulation result is detected to exceed the warning threshold but not the absolute threshold, the threshold verification and rollback module is configured not to trigger rollback, but to generate a warning message, which is sent to the graded decision and execution module along with the risk level assessment result of the safety and effectiveness analysis module, as an auxiliary basis for manual confirmation of medium-risk strategies. The bottom-line verification and rollback module also includes a rollback counter, configured to record the number of consecutive rollbacks within the same optimization cycle. When the number of consecutive rollbacks exceeds a preset threshold, it is determined that the current optimization problem has no feasible solution, the automatic optimization process is forcibly paused, and an emergency alarm is issued to require deep manual intervention.

Images