Dynamic polymer visual multi-level load collaborative management method

Abstract

The invention relates to a dynamic polymer visual multi-level load collaborative management method. Firstly, power system operation, user energy consumption behaviors and environment parameters are obtained and subjected to structured processing to obtain a data set; performing hierarchical division according to resource characteristics and spatial distribution characteristics in the data set to generate a dynamic polymer; extracting time domain and frequency domain characteristic parameters of each level of load, and constructing a layered load behavior model; then, based on the dynamic polymer topology, the hierarchical load behavior model and the constraint conditions, a multi-objective optimization function is constructed to generate a collaborative management strategy; and finally, a strategy is optimized by means of real-time feedback, virtual verification and a long-term learning mechanism, a layered visual mapping rule is constructed, and the dynamic aggregate, the layered load behavior model and a strategy execution result are graphically output. Compared with the prior art, the method has the advantages of being good in data integration performance, high in dynamic adaptability, capable of achieving multi-target tradeoff and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system load management, and particularly relates to a dynamic aggregation body visualized multi-level load collaborative management method. BACKGROUND

[0002] Current power system load management generally faces core challenges such as low coordination efficiency caused by data heterogeneity, difficulty in adapting to dynamic resource characteristics due to static hierarchical division, and insufficient multi-objective strategy optimization capability. In the prior art, traditional load management systems (such as centralized control based on SCADA) are difficult to realize the linkage analysis of user behavior and environmental parameters due to single data source and lack of cross-dimension information fusion capability; the distributed management scheme based on fixed hierarchical division (such as hierarchical demand response system) is limited by the rigid topology and cannot dynamically adjust the hierarchical relationship according to the controllability of devices and the response timeliness, resulting in low resource coordination efficiency; single target optimization strategy (such as only energy balance or delay suppression) is easy to fall into local optimum in complex scenarios and lacks multi-objective trade-off mechanism; in addition, the existing system generally lacks dynamic calibration capability, and the strategy generation depends on static parameter model, which cannot optimize the hierarchical division threshold and coordination rules through real-time feedback.

[0003] Current prior art solutions include: Dynamic aggregation body management technology: some studies use clustering algorithm to divide load aggregation body, but only based on historical load characteristics to statically divide the hierarchy, resulting in disconnection between hierarchical division and actual operating conditions. Multi-objective optimization strategy generation: existing studies combine energy efficiency, delay and other targets into a single objective function through weighted summation method, but the fixed weight leads to rigid strategy and cannot adapt to the multi-objective dynamic game demand in disturbance scenarios. Hierarchical load characteristic analysis: existing methods analyze load rules through time domain statistics (such as mean, variance), but ignore frequency energy entropy and cross-level correlation weight indicators, making it difficult to quantify the mutation response characteristics and coupling strength between levels.

[0004] A multi-level power market user profile dynamic collaborative management method and system are disclosed in the invention patent with publication number CN120013088A, but it only manages the association between user portraits and power consumption mode models, and there is no collaborative management mechanism combining multi-objective optimization and real-time feedback calibration, and there is no construction of a full-process closed-loop system covering data processing, dynamic aggregation, load modeling to strategy optimization. The scheme only expands around the association of user portraits and power consumption modes, and does not establish an integration mechanism for the heterogeneity of data, resulting in fragmented multi-source information, which cannot provide data support for cross-dimensional load evolution analysis; and its hierarchical division relies on static classification of user types, lacks dynamic adjustment logic based on resource characteristics such as device controllability and response timeliness, and is difficult to adapt to real-time changes in device status and energy demand in the power system, which is prone to hierarchical instruction transmission redundancy or failure; at the same time, its strategy optimization only focuses on the flexibility of power supply strategy, without considering the dynamic game between multiple objectives such as response delay and system stability, and without introducing calibration mechanisms such as virtual verification and long-term learning, the strategy robustness is insufficient in disturbance scenarios such as load mutation and device failure.

[0005] In summary, the existing technology is difficult to establish a cross-dimensional load evolution model due to the fragmentation of multi-source information caused by data heterogeneity (such as the lack of association between power system operation data and user behavior data), making collaborative decision-making rely on local optimal strategies; at the same time, static hierarchical division (such as fixed tree topology) cannot dynamically adapt to changes in resource characteristics such as device controllability and response timeliness, causing redundancy or failure in hierarchical instruction transmission; in addition, single-objective optimization strategies (such as only energy balance) ignore multi-objective dynamic games (such as the conflict between delay suppression and stability guarantee), and lack real-time feedback calibration mechanisms, with significantly decreased strategy robustness in disturbance scenarios. The current technology has defects in data integration, dynamic adaptability and multi-objective trade-off. SUMMARY

[0006] The purpose of the present application is to overcome the defects of the prior art and provide a dynamic aggregation visualization multi-level load collaborative management method.

[0007] The purpose of the present application can be achieved by the following technical solutions: According to one aspect of the present application, a dynamic aggregation visualization multi-level load collaborative management method is provided, characterized in that the method steps include: S1, obtaining power system operation data, user energy consumption behavior data and environmental parameters, and structurally processing the three types of data to obtain a structured test data set; S2, according to the resource characteristics and spatial distribution characteristics in the structured test data set, hierarchical division is performed to generate a tree or net topology structure, and a dynamic aggregation is generated; S3, extract the time domain and frequency domain characteristic parameters of the load of each level in the dynamic aggregate, and construct a hierarchical load behavior model based on the characteristic parameters; S4, based on the dynamic aggregate topology and the hierarchical load behavior model, combined with the constraint conditions, construct a multi-objective optimization function, and generate a collaborative management strategy based on the multi-objective optimization function; S5, use real-time feedback, virtual verification and long-term learning mechanisms to optimize the collaborative management strategy, and construct a hierarchical visualization mapping rule, which uses the visualization mapping rule to output the execution results of the dynamic aggregate, the hierarchical load behavior model and the collaborative management strategy as graphical expressions.

[0008] The power system operation data in S1 includes node voltage, frequency, line power and device state; the user energy consumption behavior data includes user side load curve; the environmental parameters include temperature, humidity and illumination intensity; the structured processing includes: converting the above three types of data into a unified tensor representation, detecting and cleaning abnormal values, unifying data format, modeling node spatial correlation, supplementing missing data, verifying data correlation and checking data integrity.

[0009] As a preferred technical solution, the specific process of S2 includes: based on the structured test data set, establishing resource characteristic quantitative indicators and spatial distribution characteristics; wherein, the resource characteristic quantitative indicators include controllability indicators and response timeliness indicators; the spatial distribution characteristics include communication cost and electrical correlation degree; based on the resource characteristic quantitative indicators and the spatial distribution characteristics, a tree-shaped hierarchy is generated, the node with the highest controllability indicator and the optimal geographical centrality is selected as the root node of the entire hierarchy structure, and according to the communication cost and the electrical correlation degree, the nodes except the root node are classified into sub-trees through K-means clustering; on the basis of the tree-shaped structure, long-range connections across levels are added through reconnection probability to form a mesh structure; after the hierarchical structure is established, when running tasks and transferring information on it; follow the task allocation mechanism and information transmission protocol; The task allocation mechanism is specifically: calculate the load index of each node, when allocating tasks, first sort the node priority according to the load index, and preferentially select the nodes with load index less than or equal to the preset load threshold; then update the node priority according to the real-time response timeliness index; The information transmission protocol specifically includes intra-layer communication rules, cross-layer communication rules and information compression rules, the intra-layer communication rules are broadcast mode between nodes in the same level, and the bandwidth of intra-layer communication accounts for 70% of the total bandwidth resources; the cross-layer communication rules are unicast mode between upper and lower nodes, and the bandwidth of cross-layer communication accounts for 30% of the total bandwidth resources, and the time delay of cross-layer communication is less than or equal to 100 ms .

[0010] As a preferred technical solution, the time domain and frequency domain characteristic parameters of the load of each level extracted by S3 include: the time domain characteristic parameters include steady fluctuation law and sudden response characteristics, the steady fluctuation law includes steady fluctuation amplitude and periodicity intensity; the sudden response characteristics include sudden amplitude and sudden recovery time; the frequency domain characteristic parameters include frequency band energy distribution and frequency domain energy entropy; and the time domain and frequency domain characteristic parameters further include correlation weight indexes, the correlation weight indexes include cross-level correlation and spatial correlation weight.

[0011] As a preferred technical solution, the hierarchical load behavior model in S3 is a hierarchical structure, the model includes a first level and a second level, in the first level, a load behavior model of a device level is established to describe single device start-stop law; in the second level, a hierarchical aggregation model is constructed by superimposing the first level and introducing a synergistic effect term; in the hierarchical load behavior model, the time scale of minute level and below uses an autoregressive moving average model to predict the load recovery trend after mutation, and the time scale of hour level and above uses an LSTM network to capture nonlinear time sequence dependence.

[0012] As a preferred technical solution, the multi-objective optimization function in S4 includes a response delay penalty term and a system stability penalty term; the specific formula of the multi-objective optimization function is: ; ; ; Wherein, is an optimization decision variable; and are weight coefficients; is a response delay penalty term, the larger the delay, the higher the penalty value; is the total number of devices; k is the device serial number; is the current time; is the response time of the k th device; is an exponential decay coefficient; is a system stability penalty term; and are weight coefficients; is the system frequency deviation at time t ; is the voltage qualified value at time t ; is the system rated voltage; is a voltage deviation degree term.

[0013] As a preferred technical solution, the cooperative management strategy in S4 is a recursive strategy, specifically including a first decision level and a second decision level, and the cooperative management strategy further includes a negotiation mechanism based on game theory; the input of the first decision level is global load prediction and system constraints, and a global strategy is generated based on the input; the goal of the global strategy is to minimize the system stability penalty term; the global strategy is then converted into a set of inter-level cooperative instructions and output; the input of the second decision level is level weight and local constraints, after receiving the set of inter-level cooperative instructions, the second decision level combines the input to decompose the task to a sub-level, thereby calculating a local strategy for each device or sub-area, and finally converting the local strategy into a device-level operation instruction and output; the negotiation mechanism based on game theory: when the inter-level instructions conflict, a non-cooperative game model is constructed, and the optimal strategy is finally solved through Nash equilibrium.

[0014] As a preferred technical solution, the real-time feedback in S5 is: for the current measurement, multi-dimensional feedback data is collected in real time, including instruction completion rate, response delay distribution and load deviation; performance degradation detection is performed based on the multi-dimensional feedback data, when the instruction completion rate is less than a preset completion rate threshold or the response delay distribution is greater than a preset delay distribution threshold, it is determined that performance degradation occurs, an isolated algorithm is used to detect abnormal feedback data points, an abnormal score is calculated, and a strategy optimization process is performed; at this time, the strategy optimization process includes: updating the strategy parameters using online gradient descent, adjusting the global strategy based on the federated learning framework, and optimizing the local strategy based on reinforcement learning; virtual verification is: constructing a power system simulation model containing a disturbance scenario, injecting the current strategy into the simulation model, running Monte Carlo simulation , and calculating the strategy failure probability; if the strategy failure probability is higher than a preset failure probability threshold, the strategy optimization process is performed; at this time, the strategy optimization process is: adding a robustness penalty term to the optimization objective, optimizing the strategy based on adversarial training, and finally solving the adversarial sample through projected gradient descent; The long-term learning is: constructing a data lake, storing real-time running data in time series; calculating the volatility of the hierarchical weight and the error between the predicted load and the actual load based on the real-time running data; constructing an incremental self-learning model, training the self-learning model based on the federated learning framework using cross-hierarchical devices, and using the self-learning model to complete the optimization process of the strategy; constructing a closed-loop calibration and optimization process, periodically comparing the deviation of the predicted load and the actual load, adjusting the hierarchical division threshold according to the deviation, and adjusting the parameters based on the real-time running data using reinforcement learning, and updating the strategy through the PPO algorithm; based on the real-time running data, fitting the recession curve of the performance index of the calibrated strategy, aligning the cross-period data, aligning the load feature distribution of different time periods, and testing the data consistency through the maximum mean difference; when the recession rate of the recession curve at a certain place is greater than the preset recession rate threshold or the data consistency is less than the preset consistency threshold, the above incremental self-learning model is used to optimize the strategy.

[0015] As a preferred technical solution, the visualization mapping rule in S5 specifically includes an aggregate topology mapping rule, a load state mapping rule, and a strategy execution result mapping rule; the aggregate topology mapping rule is used to output the dynamic aggregate as a graphical expression, which is specifically: based on the hierarchical division result of the dynamic aggregate, using force-directed graph or hierarchical layout algorithm to dynamically render nodes and edges; wherein the attributes of the node include node size and node color, the node size maps the hierarchical weight, and the node color encodes the response timeliness score of the node; the attributes of the edge include edge thickness and solid and dashed lines, the edge thickness represents the communication bandwidth occupancy rate, and the solid and dashed lines are used to distinguish the inter-level coordination rules; node dragging and zooming are supported, and local topology detail expansion is supported; the load state mapping rule is used to output the hierarchical load behavior model as a graphical expression, including a hierarchical superposition process of time series load curves and a mutation event highlighting process, wherein when the time series load curves are hierarchically superimposed, the hierarchical load curve and the device load curve are superimposed and displayed, and transparency is used; when the mutation event is highlighted, the mutation response event is marked with a pulse waveform, and the corresponding device state label is associated; the strategy execution result mapping rule is used to output the execution result of the coordination management strategy as a graphical expression, specifically using a multi-objective optimization index dashboard and a strategy execution heat map; wherein the multi-objective optimization index dashboard dynamically displays the delay penalty and stability index of the current strategy, using a ring progress bar or a heat map, and comparing the preset reference threshold with the actual value after the execution of the current strategy; the strategy execution heat map superimposes the strategy execution intensity matrix on the aggregate topology graph.

[0016] The specific formula of the strategy execution intensity matrix is: wherein, is the policy execution intensity value of the topology position of i, j in the aggregation topology at time t; N is the total number of devices participating in policy execution in the aggregation; is the weight of device k; is the real-time active power of device k at time t; is the rated maximum active power of the device.

[0017] According to another aspect of the present application, a dynamic aggregation visualization multi-level load collaborative management system is provided, the system comprising a data processing and integration module, a dynamic aggregation generation module, a hierarchical load modeling module, a collaborative strategy optimization module and a strategy evaluation and visualization module; The data processing and integration module is used to obtain power system operation data, user energy consumption behavior data and environmental parameters, and to structure process the three types of data to obtain a structured test data set; The dynamic aggregation generation module is used to generate a tree or mesh topology structure according to the resource characteristics and spatial distribution characteristics in the structured test data set, and to generate a dynamic aggregation; The hierarchical load modeling module is used to extract time domain and frequency domain characteristic parameters of each hierarchical load in the dynamic aggregation, and to construct a hierarchical load behavior model based on the characteristic parameters; The collaborative strategy optimization module is used to construct a multi-objective optimization function based on the dynamic aggregation topology and the hierarchical load behavior model, in combination with constraint conditions, and to generate a collaborative management strategy based on the multi-objective optimization function.

[0018] The strategy evaluation and visualization module is used to optimize the collaborative management strategy using three mechanisms of real-time feedback, virtual verification and long-term learning, and to construct hierarchical visualization mapping rules, and to output the execution results of the dynamic aggregation, the hierarchical load behavior model and the collaborative management strategy as graphical expressions using the visualization mapping rules.

[0019] Compared with the prior art, the present application has the following beneficial effects: 1、In the present application, power system operation data, user energy consumption behavior data and environmental parameters are obtained, and the three types of data are structured processed to construct a hierarchical load behavior model and a multi-objective optimization function, a collaborative management strategy is generated based on the multi-objective optimization function, and the collaborative management strategy is optimized using three mechanisms of real-time feedback, virtual verification and long-term learning; an end-to-end, visual multi-level load collaborative management scheme is provided, full-process automation from data processing, structure generation, model construction, strategy optimization to result display is realized, and the intelligent level and management efficiency of the system are improved. The management method has good data integration performance, strong dynamic adaptability and realizes multi-objective trade-off, completes the intelligentization of load collaborative management, and greatly improves the adaptability of the load collaborative management strategy.

[0020] 2、In the application, by establishing resource characteristic quantitative indicators and spatial distribution characteristics, a tree hierarchy is generated based thereon, and finally, long-range connections across the hierarchy are added through reconnection probability based on the tree structure to form a mesh structure, combining the clear hierarchy of the tree structure and the high robustness of the mesh structure, so that the method realizes the dual advantages of multi-level load management in efficient cooperation and stable anti-disturbance, and adapts to the complex and variable operation scenarios of the power system. In the application, after the hierarchical structure is established, the task allocation mechanism and the information transmission protocol are followed when running tasks and transmitting information on the hierarchical structure; the task allocation mechanism realizes global load balancing by preferentially selecting low-load nodes and dynamically adjusting the priority in combination with real-time response timeliness, avoids the situation that some devices are overloaded while other devices are idle, and maximizes resource utilization; and the information transmission protocol ensures the orderliness and efficiency of information transmission. Both of them significantly improve the operation efficiency and cooperation ability.

[0021] 3、In the application, both time domain and frequency domain characteristics are considered, which more comprehensively depicts the static law of the load; and the correlation weight index is introduced, which breaks the isolation of the loads at each level and can quantify and capture the mutual influence between the loads at different levels and different spatial positions. The rich feature dimension makes the model not sensitive to single abnormal data, and improves the robustness. In the application, a hierarchical load behavior model of the first level and the second level is constructed, the first level starts from the most basic unit, accurately describes the start-stop and operation law of a single device, and provides high-quality data for upper aggregation; and the second level is superimposed on the basis of the first level, and introduces a coordination effect term to model the mutual influence between devices, so that the aggregated load prediction is more accurate. And by distinguishing the multi-scale prediction of minute level and hour level, each takes the best at different time scales, realizing the best balance between prediction performance and calculation efficiency.

[0022] 4、In the application, by constructing a multi-objective optimization function including a response delay penalty term and a system stability penalty term, the balance optimization of multiple objectives is realized while meeting the system constraints. By adopting a recursive collaborative management strategy including a first decision level and a second decision level, and constructing a negotiation mechanism based on game theory, the global optimality, local feasibility and effective coordination in conflict scenarios of the strategy are ensured.

[0023] 5、In the present application, in real-time feedback, performance degradation detection is realized by real-time collection of multi-dimensional feedback data, which realizes dynamic monitoring and rapid adjustment of the effect of policy execution, ensures that the system can adapt to changes in time and maintain optimal performance. In virtual verification, through digital twin technology, the strategy is fully tested in a virtual environment, potential risks are discovered in advance, and the robustness and reliability of the strategy in real complex scenarios are significantly enhanced. In long-term learning, by constructing a data lake and a self-learning model, the strategy has the ability of self-evolution and continuous optimization, which can adapt to environmental changes and performance degradation in long-term operation, prolonging the effective life cycle of the strategy.

[0024] 6、In the present application, by constructing a visual mapping rule including a polymer topology mapping rule, a load state mapping rule and a policy execution result mapping rule, visualization is realized, and complex topology structure, load state and policy execution result are presented in an intuitive graphical way, greatly improving the readability of information and the transparency of decision-making, facilitating managers to understand, analyze and intervene. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 A method step schematic diagram of a dynamic polymer visual multi-level load collaborative management method in the present application; Figure 2 A flowchart of data structured processing in the present application; Figure 3 A flowchart of generating a dynamic polymer by level division in the present application. DETAILED DESCRIPTION

[0026] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0027] Aiming at the core pain points existing in the current power system load management, such as low coordination efficiency caused by data heterogeneity, rigid hierarchical division difficult to adapt to dynamic resource characteristics, insufficient prediction deviation caused by insufficient load characteristic analysis precision, lack of global optimization ability in multi-objective strategy generation, inability of static strategy to cope with real-time disturbance, and lack of system adaptive ability, a dynamic aggregation visualization multi-level load coordination management method is proposed. Through a multi-channel data fusion framework, cross-dimensional information correlation is realized, a dynamic aggregation hierarchical structure is constructed to optimize resource coordination efficiency, a hierarchical feature analysis technology is used to improve the quantification precision of load evolution law, a multi-objective optimization coordination management strategy generation mechanism is combined, and a dynamic optimization and closed-loop calibration process based on real-time feedback is designed. Finally, the intelligent and adaptive coordination management of the power system load is realized, breaking through the technical limitations of traditional methods in dynamic, robustness and multi-objective trade-off, and solving the efficiency and reliability problems of load coordination management in complex scenarios.

[0028] The dynamic aggregation refers to a hierarchical structure unit based on resource characteristics (controllability, response timeliness) and spatial distribution characteristics, which realizes multi-level coordination through tree or network topology, supports efficient aggregation of load resources and cross-level instruction transmission. Multi-level load characteristic analysis: by extracting time domain fluctuation law, frequency energy entropy and correlation weight index, a hierarchical load behavior model is constructed to quantify the load evolution characteristics at different time scales and provide accurate input for strategy generation. Coordination management strategy generation: combined with load prediction and system constraint conditions, a multi-objective optimization function (energy efficiency balance, delay suppression, stability guarantee) is constructed, and cross-level instruction coordination and priority allocation are realized through hierarchical recursive decision-making process. Strategy dynamic optimization and execution verification: based on real-time feedback data, parameter correction and local iteration are triggered, and the robustness of the strategy in disturbance scenarios is verified through virtual simulation deduction, forming a "perception-decision-verification" closed-loop optimization mechanism. System adaptive calibration mechanism: use long-term operation data to train self-learning model, dynamically adjust hierarchical division threshold and coordination rule parameters, and realize continuous precision optimization of feature analysis and strategy generation through comparison of predicted and actual load deviation.

[0029] Embodiment 1 In this embodiment, a dynamic aggregation visualization multi-level load coordination management method is adopted, and the method steps are as shown in Figure 1 The specific steps include: S1, obtaining power system operation data, user energy consumption behavior data and environmental parameters, and structuring the three types of data to obtain a structured test data set; S2, according to the resource characteristics and spatial distribution characteristics in the structured test data set, hierarchical division is performed to generate a tree or network topology structure, and a dynamic aggregation is generated; S3, extract the time domain and frequency domain characteristic parameters of each level load in the dynamic aggregate, and construct a hierarchical load behavior model based on the characteristic parameters; S4, based on the dynamic aggregate topology and the hierarchical load behavior model, combined with the constraint conditions, construct a multi-objective optimization function, and generate a collaborative management strategy based on the multi-objective optimization function; S5, use real-time feedback, virtual verification and long-term learning mechanisms to optimize the collaborative management strategy, and construct a hierarchical visualization mapping rule, which uses the visualization mapping rule to output the execution results of the dynamic aggregate, the hierarchical load behavior model and the collaborative management strategy as graphical expressions.

[0030] The main steps of the method are as follows: 1. Data acquisition and structured processing: Capture the basic data stream of power system operation, user energy behavior and environmental parameters through multi-channel input interface, and construct a unified data representation form. Establish data correlation mapping rules to realize logical correlation and dynamic supplement of cross-dimension information.

[0031] (1) Data source definition and access, through multi-channel input interface (such as sensor network, SCADA system, smart meter, user behavior log interface) to capture three types of data stream in real time: Power system operation data: including node voltage (V t ), frequency (f t ), line power (P ij , Q ij ), device state (such as transformer load rate ). User energy behavior data: covering user side load curve (s u,k is the device start-stop state, P u,k is the device power), time-of-use price response coefficient . Environmental parameters: temperature (T t ), humidity (H ), light intensity (I t ), obtained through weather sensors or public API. Data synchronization mechanism: use timestamp alignment algorithm (such as linear interpolation method) to ensure the consistency of time resolution of multi-source data (synchronization error Δt≤1s).

[0032] The specific processing process of data standardization and structured processing is shown in Figure 2 , including converting heterogeneous data into a unified tensor representation, detecting and cleaning outliers, unifying data formats, data correlation, hierarchical storage and data integrity verification. Through standardization and structured processing of multi-source heterogeneous data, the data quality and consistency are ensured, which provides a solid data foundation for subsequent accurate analysis and decision-making.

[0033] (2) Data representation: convert heterogeneous data into unified tensor representation: where m is the data dimension (e.g., power parameters, environmental parameters, etc.), and n is the time series length.

[0034] (3) Outlier detection and cleaning: Grubbs test method: identify outliers, hypothesis testing statistics , when G>G threshold , it is determined as an outlier. 0.95

[0035] Sliding window mean filling: bad data is filled with (window width w=5).

[0036] (4) Data format unification, unstructured data processing: parse user behavior logs into structured fields, for example: User behavior label: b u =[appliance type, usage period, energy consumption level].

[0037] Dynamic feature extraction: calculate the short-time Fourier transform (STFT) frequency energy entropy of the load: where X k is the STFT coefficient, quantifying the complexity of load fluctuations (the larger H(f), the more disordered the fluctuations).

[0038] (5) Spatio-temporal correlation rule, spatial correlation modeling: based on the sensitivity of power flow calculation , define the electrical distance matrix between nodes ( is the attenuation coefficient).

[0039] Dynamic replenishment mechanism: when the data source is missing, it is supplemented by Markov Chain Monte Carlo (MCMC): where X miss is the missing data, and X obs is the known data.

[0040] (6) Data correlation verification, consistency check: evaluate the consistency of multi-source data through Cross-Entropy: When CE>1.5, trigger an abnormal alarm and start redundant data source switching.

[0041] (7) Hierarchical storage architecture, raw data layer: store raw time series data (compression ratio 80%, storage period 1 year).​

[0042] Feature data layer: Extract load features (such as peak-valley difference ), storage period 3 years.

[0043] API interface design: Provide standardized data calling protocol: def fetch_data(start_time, end_time, data_type="power"): return X_t[start_time:end_time][data_type] Support flexible calling by time range and data type, interface response time <100ms$.

[0044] (8) Data integrity check, time series continuity test: Calculate the mean and variance of the difference between adjacent timestamps through sliding window: When , it is determined that the data is broken.

[0045] Abnormal feedback mechanism: Input the data integrity check result (such as CE value, to the system adaptive calibration module, dynamically adjust the data cleaning parameters (such as sliding window width W).

[0046] 2, Dynamic aggregation body level division: According to the resource characteristics (such as controllability, response timeliness) and spatial distribution characteristics, define the level division principle of aggregation body, form tree or net topology structure. Design the coordination rules between levels, and clearly define the interaction logic of different levels in task allocation, information transmission and decision authority.

[0047] (1) Resource characteristic quantitative index, controllability index: Based on the historical data of device controllable state, calculate the response probability p ctrl : The number of response instructions The controllability threshold is set to , and the devices below this value are classified into the low response level.

[0048] Response timeliness index: Calculate the timeliness score through instruction execution delay data : Decay coefficient ( ≥0.1) Timeliness score is positively correlated with the weight of level division.

[0049] (2) Spatial distribution feature modeling, geographical distance weight: Based on node coordinates Compute Euclidean distance , convert to communication cost (ε is a constant to prevent zero).

[0050] Electrical coupling strength: use power flow sensitivity matrix , define spatial correlation degree , high correlation nodes are preferentially divided into the same level.

[0051] (3) Tree level generation, top-down division: Root node selection: select the node with the highest controllability and the optimal geographical centrality (such as betweenness centrality ) as the root node.

[0052] Sub-node allocation: according to the communication cost and electrical correlation degree , the remaining nodes are divided into sub-trees by K-means clustering, and the objective function is: .

[0053] Where w is the weight coefficient, is the communication cost of the node to the root, is the correlation degree of the node and the parent node.

[0054] (4) Optimization of mesh level, modeling of small world network: adjust the network topology through reconnection probability p rewire Balance local connection and global propagation efficiency.

[0055] Reconnection probability (p rewire ) is the core parameter for adjusting the network topology structure (value is 0~1), p rewire The determination method of the value is as follows: Method one: experience setting Power network optimization: p=0.1~0.3 (balance local power supply stability and fault propagation control) Method two: dynamic optimization Reinforcement learning: take p as a strategy parameter, dynamically adjust through reward function (such as network efficiency, robustness). Adaptive algorithm: adjust p in real time according to network load or fault event.

[0056] Method three: data driven Historical data analysis: fit the optimal value of p through historical network performance (such as propagation delay, failure rate).

[0057] Machine learning model: use regression or neural network to predict p, input variables include node degree distribution, network structure, etc Local connection weight: keep edges between high correlation nodes (A ij >θA ).

[0058] Random long-range connection: with probability p rewire Add cross-area edges to enhance system robustness.

[0059] (5) Task allocation mechanism, load balancing based on resource capacity: define node load index ( for the current task amount, for the maximum capacity), task allocation preferentially selects nodes with load index L i ≤ 0.8.

[0060] Dynamic priority adjustment: update node priority according to real-time response delay τ real (t): where, is the average delay of the system.

[0061] (6) Information transmission protocol, intra-layer communication: broadcast mode between nodes in the same layer, bandwidth allocation ratio . Cross-layer communication: unicast mode between upper and lower nodes, bandwidth , and needs to meet the delay constraint .

[0062] Information compression rule: differential encoding is used for non-critical data (such as environmental parameters), compression rate .

[0063] (7) Adaptive re-partitioning trigger condition, performance degradation detection: trigger reconstruction when the average response delay within the hierarchy or the node failure rate .

[0064] is the average response delay within the hierarchy, i.e. the average response time of all nodes in a hierarchy to process requests (such as the sum of network delay and computation delay).

[0065] is a preset threshold.

[0066] According to the judgment logic of the formula, when the average response delay within the hierarchy is greater than the preset threshold, reconstruction is triggered.

[0067] The failure rate is the ratio of the number of failed nodes in the current hierarchy or system to the total number of nodes. By regularly detecting node status (such as the heartbeat mechanism), the number of failed nodes N fail and the total number of nodes N total ; The average response delay within a hierarchy, i.e., the average response time for all nodes within a hierarchy to process a request (such as the sum of network delay, computation delay). The preset threshold. According to the judgment logic of the formula, when the average response delay within the hierarchy is greater than the preset threshold, reconstruction is triggered.

[0068] Incremental adjustment algorithm: only local optimization is performed on the affected sub-tree, reducing the computational overhead: where S is the set of affected nodes, and the goal is to minimize .

[0069] represents the amount of energy change or error change, i.e., the change in the overall objective function (such as the energy function, loss function) after the local optimization of the affected sub-tree by the algorithm. The incremental algorithm avoids recalculating the entire system's objective function by calculating this local change, thereby reducing the computational overhead. represents the weight coefficient of node i, which measures the importance or influence of the node in local optimization. The weight can be set based on the feature saliency of the node, historical optimization experience or domain knowledge. A larger weight means that the change of the node has a more significant impact on the overall objective function. represents the local energy value or local error value of node i after local optimization. It reflects the optimal state (such as the minimum energy configuration, minimum prediction error) of the adjusted node and its associated sub-tree. represents the local energy value or local error value of node i before optimization, i.e., the initial state before adjustment. The subscript i belongs to the set S, where S is defined as the set of affected sub-tree nodes. The incremental algorithm only calculates the local area (rather than the entire system) directly related to the current adjustment operation, and the range of S is determined by the locality constraints of the algorithm (such as dependency relationships, neighborhood rules), ensuring the efficiency of the calculation.

[0070] (8) Topology evolution model, policy update based on reinforcement learning: define the state space , action space A = {merge hierarchy, split hierarchy}, and the reward function is: where is the delay improvement, is the change in communication cost, is the robustness gain.

[0071] 3. Multi-level load feature analysis: extract time-domain and frequency-domain feature parameters of each level of load, including steady-state fluctuation rules, sudden response characteristics and correlation weight indicators. Based on the feature parameters, build a hierarchical load behavior model to quantify the load evolution rules at different time scales.

[0072] 3.1 Multi-level load feature extraction: (1) Time-domain feature parameters, steady fluctuation pattern: define load curve L u (t) steady fluctuation amplitude σ steady and periodicity intensity C period : where, reflects the long-term stability of the load, quantifies the periodicity characteristics (such as daily cycle, weekly cycle).

[0073] Mutation response characteristics: detect load mutation points , and calculate mutation amplitude and recovery time : where, is the mutation start time, is the time when the load recovers to within the threshold.

[0074] (2) Frequency-domain feature parameters, frequency band energy distribution: wavelet transform (Daubechies wavelet basis ) on load curve, calculate each frequency band energy proportion: where, is the j-th layer wavelet coefficient, indicates the energy contribution rate of frequency band j.

[0075] Frequency energy entropy: quantifies the complexity of load fluctuation: The larger, the more disordered the load fluctuation (such as sudden appliance start-stop).

[0076] (3) Correlation weight index, cross-level correlation: calculate the load correlation between level i and parent level p(i) : where, is the covariance, is the standard deviation.

[0077] Spatial correlation weight: combine electrical distance matrix and time sequence correlation , define the comprehensive correlation weight: 3.2 Hierarchical load behavior modeling: (1) Hierarchical model structure, low level (device level): Establish device-level load behavior model , describe single-device start-stop law: where, is the amplitude of the mth active mode of device k, is the decay coefficient, is the noise.

[0078] Mid-high level (regional / system level): By superimposing low-level models and introducing collaborative effect terms, build hierarchical aggregation model: where, is the hierarchical weight (such as formula ), represents the cross-level collaborative effect (such as demand response strategy).

[0079] (2) Time scale quantification, short time scale (minute level): Adopt autoregressive moving average model (ARMA) to predict the load recovery trend after mutation: where p is the autoregressive order and q is the moving average order.

[0080] Long time scale (hour level and above): Capture nonlinear time series dependence based on LSTM network: where h t is the hidden state and c t is the cell state.

[0081] 3.3 Dynamic calibration of relevance weight: (1) Weight update mechanism, based on Granger causality test: Test the prediction ability of level i to j, dynamically adjust the weight W ij : where RSS is the residual sum of squares, p and q are the number of model parameters. If F Granger >F threshold , then enhance W ij .

[0082] Weight optimization based on reinforcement learning: Define state space , action space , and reward function as: Update weight policy by Q-learning: 3.4 Cross-layer transfer of features and policies: (1) Feature transfer rule, low level → middle-high level: transfer device-level features by weight after aggregation, add delay compensation term to the upper layer: where, is the communication delay, is the compensation function (such as linear interpolation).

[0083] (2) Policy decomposition logic, middle-high level → low level: decompose the optimization policy into device-level instructions , meet resource constraints: where, is the execution cost penalty term of device k.

[0084] 4. Collaborative management strategy generation: combine load prediction results and system constraint conditions to build a multi-objective optimization function, covering energy efficiency balance, response delay suppression, and system stability guarantee. Design a hierarchical decision-making process to realize the coordinated generation and priority allocation of cross-level instructions through recursive policy decomposition.

[0085] 4.1 Multi-objective optimization function construction: (1) Objective function design, based on load prediction results and system constraint conditions, build a multi-objective optimization function, covering the following core indicators: Response delay suppression: limit instruction response time , introduce delay penalty term: where, is the actual response time of device k.

[0086] System stability guarantee: constrain system stability through frequency deviation and voltage qualification rate : where, is the weight coefficient.

[0087] Comprehensive optimization goal: where, For multi-objective weights, meet .

[0088] (2) Constraints, device capacity constraints: Communication bandwidth constraints: Cross-layer instruction transmission rate Power balance constraints: 4.2 Hierarchical decision-making process design: (1) Recursive strategy decomposition, high-level (system-level) decision-making: input global load prediction and system constraints, generate initial strategy , the goal is to balance stability under constraints: Output is a set of inter-layer coordinated instructions .

[0089] Low-level (regional / device-level) decision-making: according to the level weight and local constraints, decompose global instructions into local strategies : Output is a device-level operation instruction .

[0090] (2) Priority allocation mechanism, priority based on response timeliness: define the priority score of the device i ( The historical average response time is ), high-score devices are executed first.

[0091] Dynamic adjustment rule: when a sudden increase in response delay of device i is detected ), trigger priority reordering: 4.3 Cross-layer instruction coordination: (1) Instruction mapping and compensation, inter-layer instruction mapping: decompose high-level instructions to sub-level according to weight : Communication delay compensation: introduce a time delay compensation term for cross-layer instructions , the corrected instruction is: wherein, (t) is the instruction transmission delay, and a is a compensation coefficient.

[0092] (2) Conflict resolution strategy, negotiation mechanism based on game theory: when there is a conflict between instructions at different levels (such as a parent node requiring load reduction and a child node needing to maintain power supply), a non-cooperative game model is constructed: , is the utility function of level i, is the conflict penalty coefficient. The optimal strategy is solved by Nash equilibrium.

[0093] 4.4 Strategy dynamic adjustment interface: (1) Real-time feedback access, deviation detection: calculate the actual load and the residual error between the predicted value : When trigger the strategy correction.

[0094] (2) Recursion optimization of strategy in time window using model predictive control (MPC) framework: wherein, is the optimal strategy at the next time.

[0095] 5. Strategy dynamic optimization and execution verification: based on real-time feedback data, dynamically evaluate the execution effect of the strategy, trigger parameter correction and local strategy iteration. Establish a virtual verification mechanism to verify the effectiveness of the strategy in the disturbance scenario through simulation deduction, and optimize the robustness of the strategy.

[0096] 5.1 Real-time feedback data evaluation and strategy performance quantification: (1) Multi-dimensional feedback data collection, execution effect index: Instruction completion rate: calculate the device-level instruction execution success rate η exec : wherein, is the indicator function, is the allowable error range.

[0097] Response delay distribution: statistics of the percentile (such as ) of instruction response time : Load deviation penalty: calculate the actual load and the predicted value Root Mean Square Error (RMSE) of the model: (2) Performance degradation detection, dynamic threshold triggering mechanism: when or , trigger strategy optimization. For example, set .

[0098] Abnormal pattern recognition: use the Isolation Forest algorithm to detect abnormal feedback data points, calculate the anomaly score : where μ and are the mean and standard deviation of the features, n is the feature dimension. When , it is marked as abnormal.

[0099] 5.2 Dynamic correction of parameters and local policy iteration: (1) Online learning and parameter update, incremental model update: based on real-time feedback data, use online gradient descent (OGD) to update strategy parameters : where is the loss function, is the learning rate, is the real-time data.

[0100] Federal collaborative optimization: in cross-level policy adjustment, use the federal learning framework to protect privacy and only share model gradients: where is the differential privacy noise intensity.

[0101] (2) Local strategy re-optimization, fast adjustment based on reinforcement learning: define the state space , action space , reward function as: Update the policy network through the Proximal Policy Optimization (PPO) algorithm: where is the ratio of new and old policy probabilities, ε is the clipping coefficient.

[0102] 5.3 Virtual verification mechanism and robustness enhancement: (1) Construction of digital twin simulation environment, dynamic modeling of the system: establish a power system simulation model, including load mutation, equipment failure and other disturbance scenarios: where, is a random perturbation term, is a system parameter.

[0103] Policy injection and deduction: inject the policy to be verified into the simulation model, run Monte Carlo simulation , and calculate the probability of policy failure: (2) Robustness optimization method, robustness constraint embedding: increase the robustness penalty term in the optimization objective, for example, minimize the cost under the worst-case scenario: where, is a set of perturbations, is a robustness weight.

[0104] Policy enhancement based on adversarial training: generate adversarial perturbations to simulate extreme scenarios and optimize the anti-interference ability of the policy: s.t Solve the adversarial sample by projected gradient descent (PGD).

[0105] 5.4 Closed-loop feedback and policy version management: (1) Policy version iteration record, version comparison and rollback mechanism: record the performance indicators (such as ) of the policy version , and automatically roll back when the performance of the new version decreases by more than a threshold .

[0106] Incremental policy patch: only update the local part of the sub-module with performance degradation, reduce the global policy fluctuation. For example, if the regional policy fails, only adjust the weight of the region.

[0107] (2) Long-term performance evaluation, decay trend analysis: fit the decay curve of policy performance over time (such as exponential decay model): where, is the decay rate, is a noise term.

[0108] Adaptive calibration trigger: when the decay rate , trigger the adaptive calibration mechanism of the seventh part of the system, globally update the hierarchical division rules and coordination parameters.

[0109] ​6. Visualization and interactive logic design: Construct hierarchical visualization mapping rules to convert the aggregated topology, load status and policy execution results into graphical expressions. Design an interactive analysis framework to support users to drive policy optimization direction through parameter adjustment and hypothesis scenario input.

[0110] 6.1 Hierarchical visualization mapping rules: (1) Aggregated topology visualization, tree / mesh structure rendering: Based on the dynamic aggregated hierarchy division results (such as tree or mesh topology), use Force-Directed Graph or hierarchical layout algorithm (such as Reingold-Tilford algorithm) to dynamically render nodes and edges: Node attributes: Node size maps hierarchy weight W ij , color coding responds to timeliness score s time (such as cold tone indicating high delay).

[0111] Edge attributes: Edge thickness represents communication bandwidth occupancy , solid and dashed lines distinguish inter-level coordination rules (such as solid lines for mandatory constraints and dashed lines for recommended strategies).

[0112] Dynamic interaction: Support node dragging and zooming to trigger local topology detail expansion (such as clicking on a parent node to display child node load distribution).

[0113] (2) Load status visualization, hierarchical superposition of time series load curve: Aggregated load curve by level and device-level load curve are superimposed and displayed, using transparency layering (Alpha Blending) to avoid visual confusion: Where k is the adjustment factor, is the average load of the system.

[0114] Mutation event highlighting: Use pulse waveform markers for mutation response events (such as ) and associate with device status labels (such as fault code ).

[0115] 3) Strategy execution result visualization, multi-objective optimization indicator dashboard: Dynamically display the delay penalty and stability indicators of the current strategy, using ring progress bars or heat maps to compare threshold values and actual values.

[0116] Strategy execution heat map: Superimpose the strategy execution intensity matrix on the aggregated topology graph, with the formula: where, is the weight of device k, is the real-time power.

[0117] 6.2 Interactive analysis framework design: (1) Parameter adjustment interface, slider bar and input box linkage: users can adjust the multi-objective optimization weight through the slider bar , real-time trigger strategy regeneration (such as formula ), and display the Pareto frontier changes through three-dimensional surface graph.

[0118] Hypothetical scenario input module: support users to define disturbance scenario parameters (such as extreme temperature , price fluctuation ), generate load prediction deviation through Monte Carlo simulation, and automatically trigger strategy robustness evaluation.

[0119] (2) Dynamic deduction and feedback loop, real-time strategy simulator: after users adjust the parameters, call the fifth part of the strategy generation module, run incremental MPC optimization (time window ) in the virtual simulation environment, and return the predicted load curve and system stability index.

[0120] Interactive labeling and annotation: allow users to mark abnormal events (such as clicking on load mutation points to add annotations) in the visualization interface, and feed back the annotation data to the seventh part of the calibration mechanism for optimizing feature analysis threshold.

[0121] 6.3 Data-driven visualization logic: (1) Multi-source data mapping rules, time series data to color mapping: normalize the environmental parameters (such as temperature ) and map them to background color temperature (cool color corresponds to low temperature, warm color corresponds to high temperature), formula: Hierarchical weight to layout deformation: dynamically adjust node positions according to hierarchical weights , high-weight nodes are automatically clustered to the topological center (such as the force-directed parameter in the Fruchterman-Reingold algorithm).

[0122] (2) User behavior feedback modeling, interactive hotspot analysis: count the visualization areas frequently operated by users (such as weight parameters adjusted multiple times ), generate an interaction frequency matrix , and guide interface optimization: where, is the indicator function, Let i be the operating region of user i at time t.

[0123] 6.4 Visualization and System Closed-Loop Linkage: (1) Visualized results drive strategy optimization and user preference learning: Record users' preferences for adjusting specific strategies (such as reducing multiple times). User profile vectors are generated using a collaborative filtering algorithm. Enter the sixth part to automatically generate strategy suggestions.

[0124] Anomaly alarm linkage: When the visual interface detects a load deviation An automatic pop-up strategy correction suggestion panel will recommend adjusting the parameter range (e.g., ...). ).

[0125] (2) Long-term performance feedback to the calibration mechanism, and archiving of visualized performance indicators: This includes policy performance indicators after user interaction (such as adjusted performance indicators). The data is stored in the seventh part of the calibration database, triggering the update of hierarchical division rules and collaborative parameters.

[0126] Dynamic interface version management: Based on long-term operation data, automatically switch the visualization mode (such as switching from a detailed topology view to a simplified dashboard view) to adapt to different user roles (such as operations and maintenance personnel vs. management).

[0127] 7. System Adaptive Calibration Mechanism: A self-learning model is built based on long-term operational data to dynamically adjust the hierarchical division threshold and collaborative rule parameters. A closed-loop calibration process is designed to continuously optimize the accuracy of feature parsing and strategy generation by comparing the deviation between predictions and actual loads.

[0128] 7.1 Self-learning model construction and dynamic parameter updating: (1) Long-term operational data storage and feature extraction, data lake architecture: Long-term operational data is stored in time series, including load curve L(t) and hierarchical partitioning results. Strategy execution indicators Columnar storage (such as Parquet format) is used to support efficient querying.

[0129] Feature engineering: Extracting dynamic features from raw data, for example: Hierarchical stability index: Calculating hierarchical weights volatility : Strategy Failure Modes: Statistics on Strategy Failure Events The spatiotemporal distribution characteristics, such as the spatial clustering radius of high-frequency failure regions. .

[0130] (2) Incremental self-learning model, federated learning framework: Collaborative training of self-learning model across hierarchical devices under the premise of protecting privacy. For example, device-level model Aggregate global model by federated averaging (FedAvg) : where, is the local data set, is the loss function gradient.

[0131] Concept drift detection: Use ADWIN algorithm to dynamically detect data distribution changes. When the drift amplitude , trigger model retraining: where p(t) is the prediction probability distribution at the current time.

[0132] 7.2 Closed-loop calibration process design: (1) Prediction-actual deviation analysis, multi-dimensional error calculation: Compare predicted load with actual load , calculate time series error indicators: Absolute percentage error (MAPE): Dynamic time warping (DTW) distance: Measure the shape difference of load curve: (2) Dynamic adjustment of hierarchical division rules, adaptive update of threshold value: According to the deviation analysis results, dynamically adjust the hierarchical division threshold . For example, when , reduce the hierarchical division granularity: where, is the decay coefficient.

[0133] Topology reconstruction trigger: When a high-frequency failure area (such as ) is detected, trigger local topology optimization, such as merging adjacent inefficient levels.

[0134] (3) Collaborative rule parameter optimization, parameter adjustment based on reinforcement learning: Define state space , action space (Optimization target weight adjustment), reward function is: Update the policy network through the PPO algorithm to output the optimal parameter combination .

[0135] 7.3 Dynamic parameter optimization and policy iteration: (1) Online Bayesian optimization, target function modeling: model the policy generation accuracy as a Gaussian process (GP): Select the next optimization point by collecting the agent model (such as ExpectedImprovement) .

[0136] Parameter space reduction: use Markov Chain Monte Carlo (MCMC) sampling to reduce the search range of weight parameters : Where D is the historical calibration data.

[0137] (2) Hierarchical policy retraining, cross-level joint optimization: In a tree-like hierarchical structure, retrain the subtree model from bottom to top. For example, for frequently failed leaf nodes, use transfer learning to reuse parent node features: Where is the incremental update term.

[0138] Conflict resolution mechanism: when there is a conflict between policies across levels (such as the parent node requiring load reduction and the child node needing to maintain power supply), build a game theory model to solve the Nash equilibrium: 7.4 Calibration effect verification and feedback loop: (1) Virtual simulation stress testing, extreme scenario injection: simulate extreme weather (such as sudden temperature rise) or equipment failure in the digital twin environment to verify the robustness of the calibrated strategy: Multi-objective Pareto frontier analysis: compare the Pareto frontier of the strategy before and after calibration to quantify the trade-off optimization effect of energy efficiency-delay-stability: (2) Long-term performance decay monitoring, decay rate modeling: fit the decay curve of the performance indicators (such as ) of the calibrated strategy: When the decay rate , trigger the full retraining of the self-learning model in Section 7.

[0139] Cross-period data alignment: align the load feature distribution of different time periods, and test the data consistency through maximum mean discrepancy (MMD): wherein, is a kernel function.

[0140] The method provides an end-to-end visual multi-level load collaborative management scheme, realizes full-process automation from data processing, structure generation, model construction, strategy optimization to result display, and improves the intelligent level and management efficiency of the system. The management method has good data integration performance, strong dynamic adaptability, realizes multi-objective trade-off, realizes the intelligentization of load collaborative management, and greatly improves the adaptability of the load collaborative management strategy.

[0141] Embodiment 2 In this embodiment, a dynamic aggregate visual multi-level load collaborative management system is adopted, which includes a data processing and integration module, a dynamic aggregate generation module, a hierarchical load modeling module, a collaborative strategy optimization module, and a strategy evaluation and visualization module. The data processing and integration module is used to obtain power system operation data, user energy consumption behavior data and environmental parameters, and to structure process the three types of data to obtain a structured test data set. The dynamic aggregate generation module is used to divide the levels according to the resource characteristics and spatial distribution characteristics in the structured test data set, generate a tree or network topology structure, and generate a dynamic aggregate. The hierarchical load modeling module is used to extract time domain and frequency domain characteristic parameters of each level of load in the dynamic aggregate, and construct a hierarchical load behavior model based on the characteristic parameters. The collaborative strategy optimization module is used to construct a multi-objective optimization function based on the dynamic aggregate topology and the hierarchical load behavior model, combined with the constraint conditions, and generate a collaborative management strategy based on the multi-objective optimization function. The strategy evaluation and visualization module is used to optimize the collaborative management strategy using three mechanisms of real-time feedback, virtual verification and long-term learning, and to construct a hierarchical visualization mapping rule, and to output the execution results of the dynamic aggregate, the hierarchical load behavior model and the collaborative management strategy as graphical expressions using the visualization mapping rule.

[0142] In the system, each module is driven by data, functions are progressive, and closed-loop optimization is the core logic, forming a close one-way flow and bidirectional feedback connection, as follows: Basic data flow link: the data processing and integration module provides the core input for the system, and the structured test data set output by the module is directly transmitted to the dynamic aggregate generation module to support topology structure division and dynamic aggregate generation; the dynamic aggregate topology output by the dynamic aggregate generation module is used together with the time domain / frequency domain characteristic parameters extracted by the hierarchical load modeling module as the input of the collaborative strategy optimization module to provide data and model basis for multi-objective optimization function construction; the collaborative management strategy generated by the collaborative strategy optimization module is finally transmitted to the strategy evaluation and visualization module for subsequent optimization and graphical output.

[0143] Closed-loop optimization feedback link: the strategy evaluation and visualization module reverses the deviation data in strategy execution to the collaborative strategy optimization module through real-time feedback and virtual verification mechanism, realizes dynamic adjustment of strategy, and at the same time, the historical optimization experience accumulated by the long-term learning mechanism is fed back to the hierarchical load modeling module to update the parameters of the hierarchical load behavior model, improve the prediction accuracy of the model, and form a closed loop of data-model-strategy-evaluation-optimization.

[0144] The specific implementation method of the system is the same as that of embodiment 1; the data processing and integration module realizes standardized integration of multi-source heterogeneous data (power operation, user behavior, environmental parameters), avoids the problem of data island; the hierarchical load modeling module combines time / frequency domain characteristics and hierarchical structure to accurately describe the operation law of different levels of load, and can dynamically iterate the model through long-term learning to adapt to the change of load characteristics; the collaborative strategy optimization module balances the core needs of delay, stability and other needs based on a multi-objective optimization function, generates a scientific management strategy, which is more suitable for actual power grid operation scenarios than traditional single-target strategies.

[0145] The dynamic aggregation generation module clearly presents the hierarchy and spatial correlation of the load through tree / network topology, which facilitates managers to quickly locate key load nodes; the strategy evaluation and visualization module converts abstract models and strategies into graphical expressions, reducing the technical threshold of managers and reducing decision-making time; the closed-loop optimization mechanism reduces the frequency of manual intervention, reduces the cost of operation and maintenance personnel, and avoids the risk of power grid caused by manual decision deviation.

[0146] Real-time feedback and virtual verification mechanism can correct strategy deviation in time, effectively suppress the influence of load fluctuation on power grid frequency and voltage, and improve system stability; accurate load behavior model can optimize power distribution, reduce redundant power supply, and reduce energy loss; in addition, visual output facilitates the cooperation of multi-subjects such as power grid dispatching center and user side management platform, helps to realize the integration of source, network, load and storage, and promotes the transformation of power system to low carbon and intelligent.

[0147] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for dynamic aggregate visualization and multi-level load collaborative management, characterized in that, The method steps include: S1. Acquire power system operation data, user energy consumption behavior data, and environmental parameters, and perform structured processing on these three types of data to obtain a structured test dataset; S2. Based on the resource characteristics and spatial distribution features in the structured test dataset, perform hierarchical partitioning to generate tree or network topology structures and generate dynamic aggregates. S3. Extract the time-domain and frequency-domain characteristic parameters of the load at each level in the dynamic aggregate, and construct a hierarchical load behavior model based on the characteristic parameters; S4. Based on the dynamic aggregate topology and hierarchical load behavior model, and combined with constraints, a multi-objective optimization function is constructed, and a collaborative management strategy is generated based on the multi-objective optimization function; S5. Optimize collaborative management strategies by utilizing three mechanisms: real-time feedback, virtual verification, and long-term learning. Construct hierarchical visualization mapping rules and use these rules to output the execution results of dynamic aggregates, hierarchical load behavior models, and collaborative management strategies as graphical representations.

2. The dynamic aggregate visualization multi-level load collaborative management method according to claim 1, characterized in that, The power system operation data in S1 includes node voltage, frequency, line power, and equipment status; User energy consumption behavior data includes user-side load curves; environmental parameters include temperature, humidity, and light intensity; the structured processing includes: converting the above three types of data into a unified tensor representation, detecting and cleaning outliers, unifying the data format, modeling the spatial correlation of nodes, supplementing missing data, verifying the correlation of data, and checking the integrity of data.

3. The dynamic aggregate visualization multi-level load collaborative management method according to claim 1, characterized in that, The specific process of S2 includes: Based on a structured test dataset, quantitative indicators and spatial distribution characteristics of resource characteristics are established. Among them, the quantitative indicators of resource characteristics include controllability indicators and response timeliness indicators; the spatial distribution characteristics include communication cost and electrical correlation. A tree-like hierarchy is generated based on resource characteristic quantitative indicators and spatial distribution features. The node with the highest controllability index and the best geographical centrality is selected as the root node of the entire hierarchy. Based on communication cost and electrical correlation, nodes other than the root node are assigned to subtrees through K-means clustering. Based on the tree structure, long-range connections across levels are added by reconnection probability to form a network structure; Once the hierarchical structure is established, tasks and information transfer on it must follow the task allocation mechanism and information transfer protocol. The task allocation mechanism is as follows: calculate the load index of each node; when allocating tasks, first sort the nodes according to the load index, and prioritize nodes with load indices less than or equal to a preset load threshold; then update the node priority sorting according to real-time response timeliness indicators. The information transmission protocol specifically includes intra-layer communication rules, cross-layer communication rules, and information compression rules. The intra-layer communication rules stipulate that nodes at the same level use a broadcast mode, with intra-layer communication bandwidth accounting for 70% of the total bandwidth resource. The cross-layer communication rules stipulate that nodes at different levels use a unicast mode, with cross-layer communication bandwidth accounting for 30% of the total bandwidth resource, and the cross-layer communication latency must be less than or equal to 100ms. ms .

4. The dynamic aggregate visualization multi-level load collaborative management method according to claim 1, characterized in that, The time-domain and frequency-domain characteristic parameters of each load level extracted in step S3 include steady-state fluctuation patterns and sudden change response characteristics. The steady-state fluctuation patterns include steady-state fluctuation amplitude and periodic intensity; the sudden change response characteristics include sudden change amplitude and sudden change recovery time. The frequency domain feature parameters include frequency band energy distribution and frequency domain energy entropy; and the time domain and frequency domain feature parameters also include correlation weight indicators, which include cross-level correlation and spatial correlation weight.

5. The dynamic aggregate visualization multi-level load collaborative management method according to claim 4, characterized in that, The hierarchical load behavior model in S3 is a hierarchical structure, which includes a first level and a second level. In the first level, a device-level load behavior model is established to describe the start-up and shutdown rules of a single device. In the second level, a hierarchical aggregation model is constructed by superimposing the first level and introducing a synergistic effect term. In this hierarchical load behavior model, the autoregressive moving average model is used to predict the load recovery trend after abrupt changes at time scales of minutes and below, while the LSTM network is used to capture nonlinear time-series dependencies at time scales of hours and above.

6. The dynamic aggregate visualization multi-level load collaborative management method according to claim 1, characterized in that, The multi-objective optimization function in S4 includes a response delay penalty term and a system stability penalty term; the specific formula of the multi-objective optimization function is as follows: ； ； ； in, To optimize decision variables; and These are the weighting coefficients; In response to the delay penalty, the greater the delay, the higher the penalty value; Total number of devices; k For equipment serial number; The current time; For the first k Response time of each device; The exponential decay coefficient; This is a penalty item for system stability. and These are the weighting coefficients; For time t System frequency deviation at that time; For time t The voltage at that time is within acceptable limits; This is the system's rated voltage; This is the voltage deviation term.

7. The dynamic aggregate visualization multi-level load collaborative management method according to claim 6, characterized in that, The collaborative management strategy in S4 is a recursive strategy, specifically including a first decision level and a second decision level, and the collaborative management strategy also includes a negotiation mechanism based on game theory; The inputs to the first decision level are global load forecasts and system constraints, and a global strategy is generated based on these inputs. The goal of this global strategy is to minimize the system stability penalty term; then, this global strategy is transformed into a set of inter-level collaborative instructions and output. The input to the second decision level is the hierarchy weight and local constraints. After receiving the inter-hierarchical collaborative instruction set, the second decision level combines the input to decompose the task into sub-levels, thereby calculating the local policy for each device or sub-region. Finally, the local policy is converted into device-level operation instructions and output. A negotiation mechanism based on game theory: When there is a conflict between instructions at different levels, a non-cooperative game model is constructed, and the optimal strategy is finally solved through Nash equilibrium.

8. The dynamic aggregate visualization multi-level load collaborative management method according to claim 7, characterized in that, The real-time feedback in S5 is as follows: for the current measurement, multi-dimensional feedback data is collected in real time, including instruction completion rate, response delay distribution and load deviation; performance degradation detection is performed based on multi-dimensional feedback data. When the instruction completion rate is less than the preset completion rate threshold or the response delay distribution is greater than the preset delay distribution threshold, it is determined to be performance degradation. An isolated algorithm is used to detect abnormal feedback data points, calculate the abnormal score, and optimize the strategy. The optimization process of the policy at this time includes: updating the policy parameters using online gradient descent, adjusting the global policy based on the federated learning framework, and optimizing the local policy based on reinforcement learning. Virtual verification involves constructing a power system simulation model that includes disturbance scenarios, injecting the current strategy into the simulation model, and running a Monte Carlo simulation. Next, the failure probability of the strategy is statistically analyzed. If the failure probability of the strategy is higher than the preset failure probability threshold, the strategy optimization process is carried out. At this time, the strategy optimization process is as follows: a robustness penalty term is added to the optimization objective, the strategy is optimized based on adversarial training, and finally the adversarial examples are solved by projective gradient descent. The long-term learning process involves: constructing a data lake to store real-time operational data in time series; calculating the volatility of hierarchical weights and the error between predicted and actual loads based on the real-time operational data; constructing an incremental self-learning model, using a federated learning framework to collaboratively train the self-learning model across hierarchical devices, and using this self-learning model to complete the policy optimization process; constructing a closed-loop calibration and optimization process, periodically comparing the deviation between predicted and actual loads, adjusting the hierarchical division threshold based on the deviation, and using reinforcement learning to adjust parameters based on real-time operational data, updating the policy through the PPO algorithm; fitting the decay curve of the calibrated policy performance indicators based on real-time operational data, performing cross-cycle data alignment, aligning the load characteristic distribution of different time periods, and verifying data consistency through the maximum mean difference; when the decay rate at a certain point on the decay curve exceeds a preset decay rate threshold or the data consistency is less than a preset consistency threshold, using the aforementioned incremental self-learning model to perform the policy optimization process.

9. The dynamic aggregate visualization multi-level load collaborative management method according to claim 1, characterized in that, The visualization mapping rules in S5 specifically include aggregate topology mapping rules, load status mapping rules, and strategy execution result mapping rules; The aggregate topology mapping rule is used to output a graphical representation of the dynamic aggregate. Specifically, based on the hierarchical partitioning result of the dynamic aggregate, a force-directed graph or hierarchical layout algorithm is used to dynamically render nodes and edges. Node attributes include node size and node color; node size maps to hierarchical weights, and node color encodes the node's response timeliness score. Edge attributes include edge thickness and solid / dashed lines; edge thickness represents communication bandwidth utilization, and solid / dashed lines are used to distinguish inter-hierarchical collaboration rules. Node dragging and scaling are supported, as is triggering local topology detail expansion. The load status mapping rule is used to output the hierarchical load behavior model as a graphical representation, including the process of layering and overlaying time-series load curves and the process of highlighting abrupt events. When layering and overlaying time-series load curves, the load curves are aggregated according to the hierarchy and displayed on top of the equipment load curves, with transparency layering. When highlighting abrupt events, the abrupt response events are marked with pulse waveforms and associated with the corresponding equipment status labels. The strategy execution result mapping rule is used to output the execution result of the collaborative management strategy as a graphical representation, specifically using a multi-objective optimization index dashboard and a strategy execution heatmap. The multi-objective optimization index dashboard dynamically displays the delay penalty and stability index of the current strategy, using a circular progress bar or heatmap, and compares the preset reference threshold with the actual value after the current strategy is executed. The strategy execution heatmap is superimposed on the aggregate topology map with a strategy execution intensity matrix.

10. A dynamic aggregate visualization multi-level load collaborative management method according to claim 9, characterized in that, The specific formula for the strategy execution strength matrix is ​​as follows: in, Let be the policy execution intensity value of the i-th and j-th topological positions in the aggregate topology at time t; N is the total number of devices participating in policy execution in the aggregate. Let k be the weight of device k; Let be the real-time active power of device k at time t; This refers to the equipment's rated maximum active power.

Images