CN-121124034-B - Dynamic polymer visual multi-level load collaborative management method

CN121124034BCN 121124034 BCN121124034 BCN 121124034BCN-121124034-B

Abstract

The invention relates to a dynamic polymer visualization multi-level load collaborative management method. The method comprises the steps of firstly obtaining power system operation, user energy behavior and environment parameters, carrying out structural processing to obtain a data set, generating a dynamic aggregate according to resource characteristics and space distribution characteristic hierarchical division in the data set, then extracting time domain and frequency domain characteristic parameters of loads of each hierarchical layer, constructing a hierarchical load behavior model, then constructing a multi-objective optimization function based on dynamic aggregate topology, the hierarchical load behavior model and constraint conditions to generate a collaborative management strategy, optimizing the strategy by means of real-time feedback, virtual verification and a long-term learning mechanism, constructing a hierarchical visual mapping rule, and graphically outputting the dynamic aggregate, the hierarchical load behavior model and a strategy execution result. Compared with the prior art, the method has the advantages of good data integration performance, strong dynamic adaptability, realization of multi-objective balance and the like.

Inventors

ZHAO JIANLI
ZHANG RUI
WANG JUN
ZHENG QINGRONG
WANG QING
Zhu Shana
LU YI
Xiang Jiani

Assignees

国网上海市电力公司

Dates

Publication Date: 20260508
Application Date: 20251114

Claims (9)

1. A dynamic polymer visualization multi-level load collaborative management method is characterized by comprising the following steps: S1, acquiring operation data of a power system, user energy behavior data and environmental parameters, and carrying out structural processing on the three types of data to obtain a structural test data set; s2, carrying out hierarchical division according to the resource characteristics and the spatial distribution characteristics in the structured test data set to generate a tree structure, and adding long-range connection crossing the hierarchy through reconnection probability on the basis of the tree structure to form a network structure to generate a dynamic aggregate; S3, extracting time domain and frequency domain characteristic parameters of each level of load in the dynamic aggregate, and constructing a layered load behavior model based on the characteristic parameters; S4, constructing a multi-objective optimization function based on the dynamic polymer topology and the layered load behavior model and combining constraint conditions, and generating a collaborative management strategy based on the multi-objective optimization function; S5, optimizing the collaborative management strategy by using three mechanisms of real-time feedback, virtual verification and long-term learning, constructing a layered visual mapping rule, and outputting the execution results of the dynamic aggregate, the layered load behavior model and the collaborative management strategy into graphical expression by using the visual mapping rule; The hierarchical load behavior model in the S3 is of a hierarchical structure, the model comprises a first level and a second level, the first level is used for establishing a device-level load behavior model to describe a single device start-stop rule, the second level is used for establishing a hierarchical aggregation model by superposing the first level and introducing a synergistic effect item to model interaction among devices, in the hierarchical load behavior model, a time scale of a minute level and below adopts an autoregressive sliding average model to predict a load recovery trend after mutation, and a time scale of an hour level and above adopts an LSTM network to capture nonlinear time sequence dependence.
2. The dynamic polymer visualization multi-level load collaborative management method according to claim 1 is characterized in that the power system operation data in S1 comprises node voltage, frequency, line power and equipment state, the user behavior data comprises a user side load curve, the environment parameters comprise temperature, humidity and illumination intensity, and the structuring process comprises the steps of converting the three types of data into unified tensor representation, detecting and cleaning abnormal values, unified data formats, modeling node space relevance, supplementing missing data, verifying relevance of the data and verifying completeness of the data.
3. The dynamic polymer visualization multi-level load collaborative management method according to claim 1, wherein the specific process of S2 comprises: establishing resource characteristic quantization indexes and spatial distribution characteristics based on a structured test data set, wherein the resource characteristic quantization indexes comprise controllability indexes and response timeliness indexes; Generating a tree-shaped level based on the quantitative index of the resource characteristics and the spatial distribution characteristics, selecting a node with the highest controllable index and optimal geographic centrality as a root node of the whole level structure, and dividing nodes except the root node into subtrees through K-means clustering according to the communication cost and the electrical association degree; On the basis of a tree structure, a network structure is formed by adding cross-level long-range connection through reconnection probability; After the hierarchical structure is established, when the task is operated and information is transmitted on the hierarchical structure, the hierarchical structure follows a task allocation mechanism and an information transmission protocol; The task allocation mechanism comprises the steps of calculating the load index of each node, and when the task is allocated, firstly, carrying out node priority ordering according to the load index, and preferentially selecting nodes with the load index smaller than or equal to a preset load threshold value; The information transfer protocol specifically comprises intra-layer communication rules, cross-layer communication rules and information compression rules, wherein the intra-layer communication rules adopt a broadcast mode among nodes at the same level, the bandwidth of intra-layer communication accounts for 70% of the total bandwidth resource, the cross-layer communication rules adopt a unicast mode among nodes at upper and lower levels, the bandwidth of cross-layer communication accounts for 30% of the total bandwidth resource, and the time delay of the cross-layer communication is less than or equal to 100ms.
4. The method for collaborative management of dynamic polymer visualization multi-level loads according to claim 1, wherein the time domain and frequency domain feature parameters of each level load extracted in the step S3 comprise steady-state fluctuation rules and abrupt response characteristics, the steady-state fluctuation rules comprise steady-state fluctuation amplitudes and periodic intensities, the abrupt response characteristics comprise abrupt amplitude and abrupt recovery time, the frequency domain feature parameters comprise frequency band energy distribution and frequency domain energy entropy, and the time domain and frequency domain feature parameters further comprise relevance weight indexes, wherein the relevance weight indexes comprise cross-level relevance and spatial relevance weights.
5. The method for collaborative management of dynamic polymer visualization multi-level loads according to claim 1, wherein the multi-objective optimization function in S4 includes a response delay penalty term and a system stability penalty term, and the specific formula of the multi-objective optimization function is: ; ; ; Wherein, the To optimize decision variables; And Is a weight coefficient; for response delay penalty term, the larger the delay, the higher the penalty value; K is the number of the devices; is the current time; response time for the kth device; is an exponential decay coefficient; penalty items for system stability; And Is a weight coefficient; the system frequency deviation at time t; is the qualified voltage value at time t; rated voltage of the system; is a voltage deviation term.
6. The method for collaborative management of a dynamic aggregate visualization multi-level load according to claim 5, wherein the collaborative management policy in S4 is a recursive policy, specifically comprising a first decision level and a second decision level, and further comprising a negotiation mechanism based on game theory; The input of the first decision level is global load prediction and system constraint, and a global strategy is generated based on the input, wherein the target of the global strategy is to minimize a system stability penalty item; The input of the second decision level is the level weight and the local constraint, after the collaborative instruction set among the levels is received, the second decision level combines the input decomposition task to the sub-level, so that the local strategy is calculated for each device or sub-area, and finally the local strategy is converted into the device-level operation instruction and output; And a negotiation mechanism based on game theory, namely constructing a non-cooperative game model when instructions conflict among the layers, and finally solving an optimal strategy through Nash equilibrium.
7. The method for collaborative management of dynamic polymer visualization multi-level loads according to claim 6, wherein the real-time feedback in S5 is that, for current measurement, multi-dimensional feedback data is collected in real time, the multi-dimensional feedback data comprises instruction completion rate, response delay distribution and load deviation, performance degradation detection is carried out based on the multi-dimensional feedback data, when the instruction completion rate is smaller than a preset completion rate threshold or the response delay distribution is larger than a preset delay distribution threshold, performance degradation is judged, abnormal feedback data points are detected by adopting an isolated algorithm, an abnormal score is calculated, and a strategy optimization process is carried out, wherein the strategy optimization process comprises the steps of adopting online gradient descent update strategy parameters, global strategy adjustment based on a federal learning framework and optimization of local strategies based on reinforcement learning; the virtual verification is that a simulation model of the power system containing disturbance scenes is constructed, the current strategy is injected into the simulation model, and Monte Carlo simulation is operated Secondly, counting the failure probability of the strategy, and if the failure probability of the strategy is higher than a preset failure probability threshold, carrying out the optimization process of the strategy, wherein the optimization process of the strategy comprises the steps of adding a robustness penalty term in an optimization target, optimizing the strategy based on countermeasure training, and finally solving a countermeasure sample through projection gradient descent; The method comprises the steps of constructing a database, storing real-time operation data according to time sequences, calculating the fluctuation rate of hierarchical weights and the error between predicted load and actual load based on the real-time operation data, constructing an incremental self-learning model, utilizing cross-hierarchical equipment to cooperatively train the self-learning model based on a federal learning frame, utilizing the self-learning model to complete the strategy optimization process, constructing a closed loop calibration and optimization process, periodically comparing the deviation between the predicted load and the actual load, adjusting a hierarchical division threshold according to the deviation, utilizing reinforcement learning to conduct parameter adjustment based on the real-time operation data, updating the strategy through a PPO algorithm, fitting a decay curve of the calibrated strategy performance index based on the real-time operation data, conducting cross-period data alignment, aligning load characteristic distribution of different time periods, checking data consistency through maximum mean value difference, and utilizing the incremental self-learning model to conduct the strategy optimization process when the decay rate at a certain position of the decay curve is larger than a preset decay rate threshold or data consistency is smaller than a preset consistency threshold.
8. The method for collaborative management of dynamic aggregate visualization multi-level loads according to claim 1, wherein the visualization mapping rules in S5 specifically include an aggregate topology mapping rule, a load state mapping rule, and a policy execution result mapping rule; The polymer topology mapping rule is used for outputting a dynamic polymer as a graphical expression, and specifically comprises the steps of dynamically rendering nodes and edges by adopting a force-directed graph or a hierarchical layout algorithm based on a hierarchical division result of the dynamic polymer, wherein the attributes of the nodes comprise node size and node color, the node size maps hierarchical weights, the response timeliness score of the node color codes the nodes, the attributes of the edges comprise edge thickness and a virtual solid line, the edge thickness represents the occupation rate of communication bandwidth, and the virtual solid line is used for distinguishing cooperative rules among hierarchies; The load state mapping rule is used for outputting a layered load behavior model into a graphical representation, and comprises a time sequence load curve layering and overlapping process and a sudden change event highlighting process, wherein when the time sequence load curve layering and overlapping process, the time sequence load curve layering and overlapping display is carried out according to a layering aggregation load curve and a device load curve, transparency layering is adopted; The policy execution result mapping rule is used for outputting an execution result of a collaborative management policy into a graphical representation, and specifically adopts a multi-objective optimization index dashboard and a policy execution thermodynamic diagram, wherein the multi-objective optimization index dashboard dynamically displays delay penalty and stability indexes of a current policy, compares a preset reference threshold with an actual value after the current policy is executed by adopting an annular progress bar or thermodynamic diagram, and the policy execution thermodynamic diagram is used for policy execution intensity matrix in an aggregate topological diagram Sang Diejia.
9. The method for collaborative management of dynamic polymer visualization multi-level loads according to claim 8, wherein the specific formula of the policy execution intensity matrix is: Wherein, the The method comprises the steps of determining a policy execution intensity value of a topological position of an ith, j in the topology of an aggregate at a moment t, wherein N is the total number of devices participating in policy execution in the aggregate; Weights for device k; Real-time active power of device k at time t; Is the rated maximum active power of the device.

Description

Dynamic polymer visual multi-level load collaborative management method Technical Field The invention relates to the technical field of power system load management, in particular to a dynamic polymer visualization multi-level load collaborative management method. Background Currently, load management of a power system generally faces core challenges such as low collaborative efficiency caused by data heterogeneity, difficulty in adapting to dynamic resource characteristics in static hierarchy division, insufficient optimization capability of multi-objective strategies and the like. In the prior art, a traditional load management system (such as centralized control based on SCADA) is difficult to realize linkage analysis of user behaviors and environmental parameters due to single data source and lack of cross-dimensional information fusion capability, a distributed management scheme (such as a hierarchical demand response system) based on fixed hierarchy division is limited by a stiff topological structure, hierarchy relation cannot be dynamically adjusted according to equipment controllability and response timeliness, so that resource cooperative efficiency is low, a single target optimization strategy (such as energy efficiency balance or delay suppression only) is easy to sink into local optimum in a complex scene and lacks a multi-target weighing mechanism, in addition, the existing system generally lacks dynamic calibration capability, policy generation depends on a static parameter model, and hierarchy division threshold and cooperative rule cannot be optimized through real-time feedback. The current prior art scheme includes: And in the dynamic aggregate management technology, part of researches adopt a clustering algorithm to divide the load aggregate, but only static division of the hierarchy is performed based on historical load characteristics, so that the hierarchy division is disjointed with actual running conditions. The multi-objective optimization strategy is generated by combining targets such as energy efficiency, delay and the like into a single objective function through a weighted summation method in the prior art, but the strategy is stiff due to fixed weight, and the multi-objective dynamic game requirement under a disturbance scene cannot be adapted. The prior method analyzes the load rule through time domain statistics (such as mean value and variance), but ignores frequency domain energy entropy and cross-level relevance weight index, and is difficult to quantify mutation response characteristics and inter-level coupling strength. The invention patent with publication number CN120013088A discloses a dynamic collaborative management method and a system for a multi-level power market user file, but only carries out the association management between user portraits and a power consumption mode model, the resource characteristic quantization standard and the dynamic topology generation logic are not combined with a collaborative management mechanism of multi-objective optimization and real-time feedback calibration, and a full-flow closed-loop system covering data processing, dynamic aggregation and load modeling to strategy optimization is not constructed. The scheme only expands around the association of the user portraits and the power consumption modes, an integration mechanism is not established for the heterogeneous problem of data, multi-source information is split, data support cannot be provided for cross-dimensional load evolution analysis, the hierarchical division depends on static classification of user types, dynamic adjustment logic based on resource characteristics such as equipment controllability, response timeliness and the like is lacking, real-time change of equipment states and power consumption requirements in an electric power system is difficult to adapt, hierarchical instruction transmission redundancy or failure is easy to occur, meanwhile, policy optimization only focuses on power supply policy flexibility, dynamic game among multiple targets such as response delay, system stability and the like is not considered, calibration mechanisms such as virtual verification, long-term learning and the like are not introduced, and policy robustness is insufficient under disturbance scenes such as load mutation, equipment failure and the like. In summary, in the prior art, multi-source information is split (such as lack of correlation between power system operation data and user behavior data) due to data heterogeneity, a cross-dimensional load evolution model is difficult to build, so that collaborative decisions depend on local optimal strategies, meanwhile, static hierarchical division (such as fixed tree topology) cannot dynamically adapt to resource characteristic changes such as equipment controllability, response timeliness and the like, so that instruction transmission redundancy or failure among hierarchies is caused, in addition, single-objective optimizatio