Visual operation and maintenance method for electric power communication network based on digital twinning

Abstract

The invention discloses an electric power communication network visual operation and maintenance method based on digital twinning, and the method comprises the steps: carrying out the coupling mapping of a global data mirror image and multiple sources, collecting data through a reconfigurable sensing array, and outputting a three-domain digital mirror image through time alignment; semantic topology reconstruction is carried out, and a dynamically growing semantic topology is generated by using a graph semantic engine; based on explainable causal chain fault tracing, an event-causal-countermeasure knowledge network is constructed; performing multi-target balancing operation and maintenance decision making, and screening an optimal strategy by using a three-dimensional Pareto model; immersive visualization and interactive intervention are performed, and a cross-platform engine is injected to realize multi-dimensional display. According to the method, through multi-source data accurate mapping, topology dynamic adaptation, causal interpretable traceability and scientific decision making, the problems of data islands, topology lagging and low decision making efficiency in existing operation and maintenance are solved, and the operation and maintenance accuracy and efficiency of the electric power communication network are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power systems and their automation, and specifically relates to the operation and maintenance direction of power communication networks, and particularly relates to a power communication network visual operation and maintenance method based on digital twinning. BACKGROUND

[0002] The existing power communication network operation and maintenance mainly relies on manual inspection, traditional monitoring systems and experiential decision-making, collects single-point data through dispersedly deployed sensors, and carries out operation and maintenance work in combination with static topology graphs and historical fault records, which to some extent supports the basic operation monitoring of the network.

[0003] However, the existing technology has significant defects: first, the multi-source data fusion and mapping precision is insufficient, the physical quantity collection lacks unified semantic coding, and different domain data (such as optical signals, radio frequency interference, and structural vibrations) have time and space asynchronization problems, which makes it difficult to accurately mirror the physical network state in the digital space and form "data islands"; second, the network topology dynamic adaptation capability is weak, the traditional topology graph is a static structure and cannot automatically evolve based on real-time operation data (such as node vulnerability and link quality drift), making it difficult to reflect the dynamic characteristics of network load changes and equipment aging; third, the fault tracing and decision-making scientificity is insufficient, the fault analysis relies on manual experience and lacks an interpretable causal reasoning mechanism, and the operation and maintenance decision-making does not balance the multi-objective balance of risk, cost and timeliness, resulting in low disposal efficiency and resource waste; fourth, the visualization and interactive experience is limited, the monitoring interface is mainly two-dimensional static display, lacks multi-dimensional data linkage and immersive interaction, and it is difficult to support operation and maintenance personnel to intuitively understand the network state and quickly intervene.

[0004] In view of the above defects, we propose a power communication network visual operation and maintenance method based on digital twinning. SUMMARY

[0005] In view of the above defects in the prior art, the purpose of the present application is to provide a wind farm short-term wind power prediction method based on a spatio-temporal attention mechanism, which solves the problems of missing physical constraints, insufficient dynamic adaptation and inefficient edge deployment in the existing power communication network operation and maintenance.

[0006] In order to achieve the above purpose, the summary of the present application adopts the following technical solutions:

[0007] A power communication network visual operation and maintenance method based on digital twinning, comprising the following steps:

[0008] S1. Global data mirroring and multi-source coupling mapping, a reconfigurable perception array is arranged at all nodes of the power communication network, the array converts the original physical quantity into a unified semantic tag stream in real time through a switchable topology bus, and outputs a spatio-temporal consistent three-domain global digital mirror after an edge time grid alignment algorithm;

[0009] S2. Semantic topology reconstruction of digital twin, taking the three-domain mirror of S1 as input, using the evolvable graph semantic engine to dynamically reconstruct the topology of the network, the output semantic topology is a self-growing dynamic graph;

[0010] S3. Fault tracing based on interpretable causal chain, on the basis of semantic topology reconstruction, build a ternary causal chain knowledge network of "event-cause-countermeasure", filter out the repair strategies highly matched with the root cause, form a one-to-one correspondence of "event-root cause-disposal";

[0011] S4. Multi-objective trade-off operation and maintenance decision engine, taking the "root cause-disposal" pair output by the ternary causal chain knowledge network as the candidate set, building a "risk-cost-time efficiency" three-dimensional Pareto decision model, screening the Pareto optimal operation and maintenance strategy;

[0012] S5. Immersive visualization and interactive intervention, inject the output results of semantic topology, ternary causal chain knowledge network and three-dimensional Pareto decision model into the cross-platform visualization engine.

[0013] Further, the array is composed of miniaturized light-electric-vibration three-mode sensitive units, programmable radio frequency front end and low-power consumption computing core.

[0014] Further, the three-domain digital mirror is logically divided into "visible light layer, radio frequency layer and structure layer", each domain realizes millisecond-level search through adaptive hash index, and cross-domain mapping is completed through scalable coupling matrix between three domains, ensuring that any physical event can be uniquely and instantaneously corresponded in digital space.

[0015] Further, the adaptive hash index function is:

[0016] H(x)=(a·x+b)modm;

[0017] x is the data characteristic value, a and b are dynamic parameters, and m is the index table length;

[0018] The element M of the scalable coupling matrix i,j The calculation formula is:

[0019] M i,j =Corr(D i ,D j );

[0020] D i and D j are physical quantity data sets of different domains respectively; Corr(·) is the Pearson correlation coefficient, which ensures that the physical event is uniquely corresponded in the digital space; M i,jThe value represents the degree of association between the coupling between the i-th domain and the j-th domain. Different combinations of i and j in the matrix can represent the coupling relationship between each two of the multiple domains, and the overall matrix can construct an association network of the multiple domains.

[0021] Further, the edge time grid alignment algorithm formula is:

[0022] Δt a =t l -t o ;

[0023] Wherein, t l is the local sampling time; t o is the clock synchronization deviation compensation value;

[0024] Further, the evolvable graph semantic engine has dynamic processing capability and can continuously reconstruct and update the network topology based on real-time data in the three-domain mirror.

[0025] Further, the method for outputting the semantic topology of the self-growing dynamic graph by using the evolvable graph semantic engine to dynamically reconstruct the network topology comprises:

[0026] Node semantic abstraction and modeling: each communication node is abstracted as a "function-state-vulnerability" triple; wherein the function is automatically annotated by the business flow characteristics, and the business role of the node in the network is clear; the state is generated by real-time multi-dimensional index compression, and the current running state of the node is quantitatively reflected; the vulnerability V is calculated by weighting the historical disturbance frequency, and the formula is:

[0027]

[0028] k represents the serial number of the historical disturbance index, which is used to distinguish different types or different time disturbance events; n represents the total number of historical disturbance indexes participating in the calculation, which covers all key disturbance records of the node within a certain period; A k represents the k-th historical disturbance index; ω k is the weight corresponding to the k-th disturbance; f is the vulnerability conversion function;

[0029] Link semantic vector construction and dynamic adjustment: a "carrying capacity-interference exposure-delay elasticity" three-axis vector is established for the optical fiber and wireless dual-mode link respectively; the link weight is self-adaptively adjusted according to the real-time link quality drift, so as to ensure the accurate description of the link state;

[0030] The formula for dynamically adjusting the link weight is:

[0031] W(t)=W0·(1+α·Q(t));

[0032] W(t) represents the real-time weight of the link at time t, which quantifies the importance or performance level of the link in the current network topology; W0 is the initial weight of the link, which is the reference weight value when the link is normally running and has no quality fluctuations; a is a sensitivity coefficient set according to the importance of the link; Q(t) is the real-time quality index of the link at t;

[0033] Topology evolution and attribute update, introduce a differentiable graph rewriting operator, when the vulnerability of any link exceeds the preset threshold, automatically insert logical relay or adjust the routing path under the constraint of maintaining network service connectivity, and synchronously update the geometry and physical attributes of the twin, so that the output semantic topology becomes a self-growing dynamic graph that adapts to the dynamic changes of the network;

[0034] The constraint condition for maintaining service connectivity is:

[0035] argmin∑ΔL op s.t Connectivity(G)=1;

[0036] argmin represents the optimization function, ΔL op is the length change of the oth path in the network before and after adjustment; Connectivity(G) represents the connectivity index of the network topology graph G; Connectivity(G)=1 indicates that the network is connected.

[0037] Further, the carrying capacity quantifies the current service bandwidth and data volume that the link can carry, reflecting the load capacity of the link; the interference exposure evaluates the degree of influence of external electromagnetic interference, environmental noise and other factors on the link, and the higher the value, the weaker the anti-interference ability; the delay elasticity represents the adaptability of the link to transmission delay fluctuations, and the higher the elasticity, the more stable the link can maintain service stability when the delay changes.

[0038] Further, the S3 comprises:

[0039] S311. Event capture and node encapsulation, based on the completed semantic topology, real-time monitoring of each node index in the network; set the baseline threshold judgment rule:

[0040] For any index, the baseline threshold T is calculated as the historical mean μ of the index plus 3 times the standard deviation σ, that is:

[0041] T=μ+3σ;

[0042] When the monitored index is greater than the baseline threshold T, it is determined that the index drift exceeds the baseline threshold, triggering the causal trigger; when the monitored index is less than the baseline threshold T, the abnormal direction is determined according to the characteristics of the index;

[0043] S312. Causal inference and contribution calculation, targeting the encapsulated event node, using an interpretable Bayesian causal discovery algorithm for causal inference, taking environmental disturbance, equipment aging, and business burst as potential parent nodes, and calculating the causal contribution of each parent node to the event through the Bayesian formula:

[0044]

[0045] E is the current event node; C i is the potential parent node; is the conditional probability of event E occurring when the parent node occurs; is the prior probability of the parent node (based on historical data statistics);

[0046] P(E) is the marginal probability of event E;

[0047] The calculation result is output in the form of natural language segments, so that the reasoning process and result can be understood by operation and maintenance personnel, and the key factors causing the fault are clear;

[0048] S313. Countermeasure mapping and repair strategy binding, based on the root cause obtained by causal inference, automatically binding repair strategies at the end of each causal chain; Repair strategies are derived from the operation and maintenance corpus distillation process, and high-value repair experience is selected from historical operation and maintenance data through the TF-IDF algorithm;

[0049] The formula for extracting high-value experience from historical operation and maintenance data through the TF-IDF algorithm is:

[0050]

[0051] TF(ω, d) is the term frequency of keyword ω in document set d, ω is the keyword or phrase selected from the operation and maintenance corpus that has a clear reference to the "root cause-repair strategy"; N is the total number of documents; DF(ω) is the number of documents containing the keyword ω;

[0052] The TF-IDF algorithm selects repair strategies that highly match the root cause, forming a one-to-one correspondence relationship between "events (encapsulated abnormal indicators)"- "root causes (causal inference results)"- "treatment (repair strategies)";

[0053] S314. Construction and update of ternary causal chain knowledge network, integrating event nodes created in the event capture stage, causal relationships and contribution degrees obtained by causal inference, and repair strategies bound by countermeasure mapping, to construct a "event-causal-countermeasure" ternary causal chain knowledge network; The knowledge network takes semantic topology as the basic framework, associates event nodes to the corresponding topological position, connects events and root causes through chain structure, and takes repair strategies as the treatment guide at the end of the chain.

[0054] When a new abnormal event is added in the network (the index exceeds the baseline threshold again), the steps S311-S313 are repeated, and the new event node, the causal reasoning result, and the repair strategy are integrated into the existing knowledge network. At the same time, the probability parameters in the knowledge network are iteratively updated based on the new data, continuously enriching the content of the knowledge network and improving the accuracy and comprehensiveness of fault tracing, so that it can better adapt to the dynamic changes of the network.

[0055] Further, in the "risk-cost-time" three-dimensional Pareto decision model:

[0056] Risk represents the risk dimension, and a risk diffusion simulation based on random walk is used to predict the expected loss of the fault on the network service availability in the next N time windows;

[0057] Cost represents the cost dimension, and an elastic resource pricing function is introduced to estimate the manpower, spare parts, transportation, and time overhead required to execute each treatment strategy in real time;

[0058] Time represents the time dimension, and a lightweight reinforcement learning agent is used to find the shortest closed-loop repair path while maintaining the stability of the network topology.

[0059] Further, the step of constructing the "risk-cost-time" three-dimensional Pareto decision model to screen the Pareto optimal operation and maintenance strategy includes:

[0060] S411. Candidate set determination and decision model construction, from the ternary causal chain knowledge network constructed in S3, extract all root cause-treatment pairs verified by causal reasoning as the candidate set for operation and maintenance decision; based on the extracted candidate set, construct a "risk-cost-time" three-dimensional Pareto decision model to screen the Pareto optimal solution;

[0061] The objective function of the three-dimensional Pareto decision model is defined as:

[0062]

[0063] Where R(A i ) is the risk value of strategy A i ; C(A i ) is the cost value; and T(A i ) is the repair time.

[0064] S412. Risk dimension evaluation and loss prediction, based on the semantic topology network structure in S2, a random walk model is used to simulate the fault diffusion process; the node connection relationship, device dependency, and environmental disturbance are used as diffusion influencing factors to define the fault state transition probability matrix; the future time is divided into The diffusion path and influence range of each time window are predicted through matrix iteration; the expected loss of each time window is calculated by combining the business importance weight, and the total risk value is obtained by summarizing;

[0065] The element in the failure state transition probability matrix represents the probability that the failure spreads from the node to the node , The probability value is obtained by statistical analysis of historical failure data:

[0066]

[0067] The expected loss calculation formula of different time windows is as follows:

[0068]

[0069] Where, B t is the set of affected businesses in the t window; ω b is the business weight; L b is the loss per unit time;

[0070] The total risk value L R is calculated as follows:

[0071]

[0072] P(t) is the probability that the t window failure affects the business;

[0073] S413. Cost dimension estimation and cost calculation, introduce an elastic resource pricing function, collect the supply and demand data of human resources, spare parts, transportation, time and other resources in real time, dynamically adjust the unit price of resources; for each disposal strategy, disassemble the resource types and quantities required for execution, calculate the total cost by combining the elastic pricing, and assist the operation and maintenance personnel to weigh the economic cost;

[0074] Elastic resource pricing function:

[0075] c q =c q0 ·α q ;

[0076] c q is the real-time elastic resource unit price; c q0 represents the basic unit price of a single elastic resource; α q is the supply and demand influence coefficient;

[0077] The total cost C(A i ) of elastic pricing is calculated as follows: q

[0078] ​

[0079] n iq For strategy A i The amount of resources q needs to be consumed; p is the total number of elastic resource types;

[0080] S414. Timeliness Dimension Optimization and Repair Path Finding: Taking the stability of the S2 semantic topology as the core constraint, a lightweight reinforcement learning agent is used to simulate the repair path selection process in a digital twin environment, and finally outputs the closed-loop repair path with the shortest time and the least disturbance to the network topology.

[0081] Furthermore, the step of simulating the repair path selection process in a digital twin environment using a lightweight reinforcement learning agent, and ultimately outputting the closed-loop repair path with the shortest time consumption and minimal disturbance to the network topology, includes:

[0082] S4141. Definition and Quantification of Topological Stability Constraints: Based on the semantic topological network structure constructed in S2, the stability constraints that the repair path must satisfy are clearly defined, providing environmental rules for reinforcement learning.

[0083] Critical link protection constraints: links carrying core business functions are marked as "critical links," and repair paths must not include or traverse critical links.

[0084] Network connectivity constraints require that the network topology connectivity index remain unchanged after path adjustment, i.e., satisfy: Connectivity(G) = 1 to avoid network segmentation or isolated nodes due to path selection;

[0085] The state space of reinforcement learning is defined as the combination of fault states of network nodes; the action space of reinforcement learning is defined as the repair operations between nodes.

[0086] Node vulnerability constraints and path prioritization avoid highly vulnerable nodes to reduce the risk of secondary failures during the repair process;

[0087] Highly vulnerable nodes are those in S2 with vulnerability V > 0.8;

[0088] S4142. Reinforcement learning environment modeling: Constructing a reinforcement learning environment to simulate the dynamic process of repair path selection. The core elements include state space, action space, and digital twin interaction mechanism.

[0089] The state space consists of the current node location, the location of the faulty node, the distribution of critical links, and the node vulnerability; the action space defines the agent's movement operations between nodes, that is, moving from the current node to its adjacent node;

[0090] Digital twin interaction: The agent obtains feedback by interacting with the digital twin environment, which simulates the state of movement time, link load changes, topology disturbances, etc. in real time, providing a basis for reward calculation;

[0091] S4143. Reward function design and state transition learning, design the reward function to guide the agent to learn the optimal strategy of "short path + low disturbance", and realize path optimization through iterative update of state-action value;

[0092] The reward function R represents the comprehensive path time consumption and topology disturbance, and the formula is:

[0093] R = -T(P) - λ·P stab ;

[0094] T(P) is the total time consumption of path P; P stab is the topology disturbance value, P stab > 0 when passing through key links or nodes with high vulnerability, otherwise 0; λ is the disturbance penalty coefficient, and λ > 0 when the key business scenario takes a larger value;

[0095] The state transition learning adopts Q-Learning algorithm to update the state-action value, and the formula is:

[0096] Q(s,z)←Q(s,z)+α′[R+γmax z′ Q(s′,z′)-Q(s,z)];

[0097] Q(s,z) is the state-action value, which represents the long-term cumulative reward expectation of selecting action z under the current network state s; α' is the learning rate; γ is the discount factor; s' is the new state to which the network topology is transferred after executing action z; a' is all possible actions; max z′ Q(s′,z′) is the optimal action value in the new state;

[0098] S4144. Shortest closed loop path optimization and output, taking the closed loop path from "operation starting point → fault node → operation starting point" as the optimization goal, and through the iterative exploration of the path space by the reinforcement learning agent, finally outputting the optimal path P * that meets the shortest path length and minimum topology disturbance;

[0099] Shortest path length:

[0100]

[0101] L(P * ) represents the total length of the path; d(v i ,v i+1 ) is the physical distance or time cost from node v i to v i+1 , and m' is the number of path nodes;

[0102] Minimum topology disturbance, path P *No critical link, through the node vulnerability V <0.6 and network connectivity index remains Connectivity(G) = 1;

[0103] Optimal path objective function:

[0104]

[0105] L critical Critical link, that is, the link carrying core business; L(P) is the path link set.

[0106] Further, the S5 comprises:

[0107] S511. The output data of the semantic topology, the ternary causal chain knowledge network and the three-dimensional Pareto decision model are standardized and integrated in format and injected into a cross-platform visualization engine; the platform visualization engine is started based on a lightweight WebGL framework, loads basic rendering components through responsive layout adaptation to PC, tablet and mobile devices and the like, and ensures consistency of visualization effects of different terminals;

[0108] S512. Three-dimensional scene layered rendering and state visualization, the WebGL framework based on physical rendering (PBR) is adopted to perform layered rendering on power communication network elements, and the link state is presented through color band and particle flow in a dual mode;

[0109] Node rendering: the nodes such as towers and substations are modeled at a scale of 1:500, and the material mapping physical properties;

[0110] Node state is associated with vulnerability V, when V > 0.8, the node edge prompts failure warning, when 0.5≤V≤0.8, it prompts risk attention, and when V < 0.5, it prompts normal state;

[0111] Link rendering:

[0112] The optical fiber link is presented in a cylindrical model, and the diameter is positively correlated with the "carrying capacity"; the wireless link is presented in a dashed line model, and the line width is positively correlated with the signal strength;

[0113] Two-dimensional presentation of link state:

[0114] The color band is mapped to color based on the link health degree, and the link health degree is:

[0115] H = 0.4 (1-interference exposure) + 0.6 x latency elasticity value;

[0116] Particle flow: the particle generation rate is positively correlated with the real-time data transmission rate of the link, the flow velocity is positively correlated with the latency elasticity value, and the business transmission state is dynamically displayed;

[0117] S513. Two-dimensional billboard design and multi-view interactive, design "risk heat-cost curve-residual life" three-view interactive panel, realize multi-dimensional visualization and interactive trigger of decision data;

[0118] Three-view design:

[0119] Risk heat map, S2 semantic topology as base map, node color depth and risk value L R Positive correlation, intuitive presentation of network risk distribution;

[0120] Cost curve, horizontal axis for disposal strategies in candidate set, vertical axis for cost, each strategy cost marked with polyline, Pareto optimal solution marked with red dot;

[0121] The vertical axis cost is calculated according to the elastic pricing total cost calculation formula;

[0122] Residual life view, based on device runtime and vulnerability V, predict residual life:

[0123] L 剩余 =(1-t / design life)×(1-V);

[0124] Residual life is displayed in the form of progress bar + percentage;

[0125] Interactive logic, user clicks any view element, system triggers three responses:

[0126] Three-dimensional scene automatically focuses on the corresponding physical node and highlights the associated link;

[0127] Pop-up causal chain pop-up window, display S3 event nodes, root cause contribution and repair strategy text;

[0128] Synchronize cost curve and residual life view, locate the strategy cost and device life data corresponding to the node, and mark the association with dashed line;

[0129] S514. Multi-modal interaction and closed-loop verification execution, multi-modal interaction such as voice interaction, gesture interaction, touch screen interaction, etc.; If resource gap Δn>0, the system pop-up window prompts "need to allocate Δn spare parts", and automatically updates S4 cost model; Where, Δn=n n -n s , n n is the number of spare parts required for repair, n s is the current inventory;

[0130] After the operation is completed, the system records the trajectory data, synchronously writes back to S3 digital twin, updates the strategy execution record of knowledge network, corrects the node state of S2 topology to ensure that the digital space and the physical system state are consistent.

[0131] In summary, by adopting the technical scheme, the beneficial technical effects of the present application are as follows:

[0132] The present application realizes systematic improvement for the defects of the prior art by the full-flow technical scheme of full-domain data mirroring, dynamic topology reconstruction, intelligent fault tracing, multi-target decision and immersive visualization, and has the following beneficial effects:

[0133] First, the precise mapping of physical network and digital space is realized, and the problems of multi-source data fusion and space-time synchronization are solved. Through the reconfigurable perception array to collect light-electricity-vibration multi-mode physical quantities, the edge time grid alignment and three-domain coupling mapping (visible light layer, radio frequency layer, structure layer) are performed, combined with adaptive hash index and scalable coupling matrix, to ensure that any physical event can be uniquely and instantaneously corresponded in the digital space, breaking the "data island" and providing a precise digital twin basis for subsequent operation and maintenance.

[0134] Second, the dynamic adaptation and fault tracing capability of network topology is improved, and the problems of static topology lag and inefficient fault analysis are solved. Through the evolvable graph semantic engine, based on the node "function-state-vulnerability" triple and link dynamic weight adjustment, the topology self-growth and real-time update are realized; relying on the "event-cause-countermeasure" three-cause-chain knowledge network, combined with Bayesian causal reasoning and TF-IDF experience screening, the corresponding relationship of "event-root cause-disposal" is formed, which greatly improves the accuracy and dynamic adaptability of fault tracing.

[0135] Third, the scientific nature and interaction efficiency of operation and maintenance decision are optimized, and the problems of unbalanced decision and inefficient visualization are solved. Through the "risk-cost-time" three-dimensional Pareto decision model, based on random walk risk simulation, elastic resource pricing and reinforcement learning path optimization, the optimal operation and maintenance strategy is screened to realize reasonable resource allocation; combined with the cross-platform visualization engine, through three-dimensional layered rendering, multi-view linkage (risk heat, cost curve, remaining life) and multi-modal interaction, the immersive display and closed-loop intervention of network state are realized, ensuring the real-time consistency of digital space and physical system state, and significantly improving the operation and maintenance efficiency and the scientific nature of decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0136] Figure 1 A flowchart of a wind farm short-term wind power prediction method based on a spatio-temporal attention mechanism. DETAILED DESCRIPTION

[0137] In order to make the purpose, technical scheme and advantages of the present application clearer and more understandable, the present application will be further described in detail in combination with embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0138] As Figure 1A wind farm short-term wind power prediction method based on spatio-temporal attention mechanism is shown:

[0139] S1. Global data mirroring and multi-source coupling mapping;

[0140] S111. Reconfigurable sensing array deployment and composition, reconfigurable sensing array is deployed at global nodes of power communication network (towers, cable joints, substation communication racks, distribution terminals), the array is composed of miniaturized light-electric-vibration three-mode sensitive units, programmable radio frequency front end and low-power computing core, realizing multi-dimensional physical quantity acquisition;

[0141] S112. Raw data encoding and time alignment, reconfigurable sensing array encodes raw physical quantities (optical power, radio frequency spectrum, structural micro-vibration, temperature and humidity, electromagnetic interference) into unified semantic tag stream in real time through switchable topology bus, and then eliminates phase drift caused by different sampling clocks and transmission delays through edge time grid alignment algorithm, generating global digital mirror with spatio-temporal consistency;

[0142] The formula of time grid alignment algorithm is:

[0143] Δt a =t l -t o ;

[0144] Where, t l is the local sampling time; t o is the clock synchronization deviation compensation value;

[0145] S113. Three-domain digital mirror construction and mapping, the global digital mirror is logically divided into "visible light layer, radio frequency layer, structure layer" three domains, millisecond-level search is realized in the domain through adaptive hash index, cross-domain mapping is completed between three domains through extensible coupling matrix, ensuring that any physical event can be uniquely and instantaneously corresponding in digital space;

[0146] The adaptive hash index function is: H(x) = (a x + b) mod m, x is the data feature value, a and b are dynamic parameters, and m is the index table length;

[0147] The calculation formula of the element M i,j of the extensible coupling matrix is:

[0148] M i,j = Corr(D i , D j );

[0149] D i and D j are physical quantity data sets of different domains; Corr(·) is Pearson correlation coefficient, ensuring that physical events are uniquely corresponding in digital space; Mi,j The coupling correlation value between the ith domain and the jth domain; different i, j combinations (rows and columns) in the matrix can present the coupling relationship between two domains (such as the visible light domain, the radio frequency domain, the structure domain, etc. in power communication), and the whole matrix can construct a correlation network of multi-domain coordination.

[0150] S2. Semantic topology reconstruction of digital twin:

[0151] S211. Engine start, the three-domain digital mirror generated in S1 step as the core input, realize millisecond-level search within the domain through adaptive hash index, complete cross-domain mapping through scalable coupling matrix, ensure the unique correspondence of physical events in digital space; start the evolvable graph semantic engine, which has dynamic processing capability and can continuously reconstruct and update the network topology based on real-time data in the three-domain mirror;

[0152] S212. Node semantic abstraction and modeling, abstract each communication node as a "function-state-vulnerability" triple; where the function is automatically labeled by the business flow characteristics, and the business role (such as data forwarding, signal processing, etc.) that the node undertakes in the network is clear; the state is generated by real-time multi-dimensional indicators (such as optical power, temperature, vibration, etc.), which quantitatively reflect the current running state of the node; the vulnerability V is calculated by weighting the historical disturbance frequency, the formula is:

[0153]

[0154] k represents the serial number of the historical disturbance indicator, which is used to distinguish different types or different time of disturbance events; n represents the total number of historical disturbance indicators participating in the calculation, covering all key disturbance records of the node within a certain period; A k represents the kth historical disturbance indicator (such as the number of failures, performance fluctuation value); ω k is the weight corresponding to the kth disturbance; f is the vulnerability conversion function (mapping the disturbance indicator to the [0, 1] interval);

[0155] S213. Link semantic vector construction and dynamic adjustment, respectively establish "carrying capacity-interference exposure-latency elasticity" three-axis vector for optical fiber and wireless dual-mode link; the link weight is adaptively adjusted with the real-time link quality drift, ensuring accurate characterization of link state;

[0156] The formula for dynamic adjustment of link weight is:

[0157] W(t)=W0·(1+α·Q(t));

[0158] W(t) represents the real-time weight of the link at time t, which quantifies the importance or performance level of the link in the current network topology; W0 is the initial weight of the link, which is the baseline weight value when the link is running normally and there is no quality fluctuation; a is a sensitivity coefficient set according to the importance of the link; Q(t) is the real-time quality index of the link at time t (range [-1, 1], positive number indicates quality improvement, negative number indicates quality decline);

[0159] The carrying capacity quantifies the current carrying capacity of the link for traffic bandwidth and data volume, reflecting the load capacity of the link;

[0160] The interference exposure evaluates the degree of influence of external electromagnetic interference, environmental noise and other factors on the link, and the higher the value, the weaker the anti-interference ability;

[0161] The delay elasticity represents the adaptability of the link to transmission delay fluctuation, and the higher the elasticity, the more stable the link can maintain the service stability when the delay changes;

[0162] S214. Topology evolution and attribute update, introduce a differentiable graph rewriting operator, when the vulnerability of any link exceeds the preset threshold, automatically insert logical relays or adjust routing paths under the constraint of maintaining network service connectivity, and synchronously update the geometry and physical attributes of the twin, so that the output semantic topology becomes a dynamic graph that can grow itself, adapting to the dynamic changes of the network;

[0163] The constraint condition for maintaining service connectivity is:

[0164] argminΣΔL op s.t Connectivity(G)=1;

[0165] argmin represents the optimization function, ΔL op is the length change amount of the oth path in the network before and after adjustment; Connectivity(G) represents the connectivity index of the network topology graph G, which is a key parameter in graph theory to measure whether the network is connected; Connectivity(G)=1 indicates that the network is connected, that is, there is at least one valid path between any two nodes for service transmission; If the connectivity index is less than 1, there are isolated nodes or the network is divided into multiple sub-networks, which will cause service interruption; This constraint ensures that the network can still normally carry services after topology adjustment, and will not cause communication interruption risk due to optimization operation.

[0166] S3. Fault tracing based on interpretable causal chain:

[0167] S311. Event capture and node encapsulation, based on the completed semantic topology (built by S2, containing node "function-state-vulnerability" triplets and link semantic vectors), real-time monitoring of each node indicator in the network (such as node state dimensions of optical power, temperature, vulnerability dimensions of historical disturbance derived values, etc.). Set baseline threshold decision rules:

[0168] For any indicator, the baseline threshold T is calculated as the historical mean μ of the indicator plus 3 times the standard deviation σ, i.e.:

[0169] T = μ + 3σ;

[0170] When the indicator is greater than the baseline threshold T, it is determined that the indicator drift exceeds the baseline threshold, triggering the causal trigger; when the indicator is less than the baseline threshold T, the abnormal direction is determined according to the characteristics of the indicator;

[0171] S312. Causal reasoning and contribution calculation, taking the encapsulated event node as the target, using an interpretable Bayesian causal discovery algorithm for causal reasoning, with environmental disturbances (such as wind speed, humidity in meteorological data, corresponding to the environmental impact factors of nodes in the topology), equipment aging (historical disturbance accumulation value in the vulnerability dimension of the node, quantified by the formula reflecting the degree of equipment aging), business surges (traffic mutation of the node function dimension carrying business, based on business flow feature labeling value judgment) as potential parent nodes, and calculating the causal contribution of each parent node to the event through the Bayesian formula:

[0172]

[0173] E is the current event node; is the potential parent node; is the conditional probability of event E occurring when the parent node occurs; is the prior probability of the parent node (based on historical data statistics);

[0174] P(E) is the marginal probability of event E;

[0175] The calculation result is output in the form of a natural language segment (such as "environmental electromagnetic interference contribution 65%"), so that the reasoning process and results can be understood by operation and maintenance personnel, and the key factors of fault initiation are clear;

[0176] S313. The countermeasure mapping is bound to the repair strategy, relying on the attribute information of nodes and links in the semantic topology (such as device type, vulnerability, and bearing service), and automatically binding the repair strategy for each end of the causal chain (i.e. the link after the explicit root cause). The repair strategy is derived from the operation and maintenance corpus distillation process: high-value repair experience is selected from historical operation and maintenance data through the TF-IDF algorithm; the TF-IDF algorithm selects a repair strategy that highly matches the root cause, forming a one-to-one correspondence relationship between "event (abnormal index encapsulation) - root cause (causal reasoning result) - treatment (repair strategy)", and providing direct operation guidance for fault handling;

[0177] The formula for extracting high-value experience from historical operation and maintenance data through the TF-IDF algorithm is:

[0178]

[0179] TF(ω, d) is the term frequency of keyword ω in document set d, ω is a keyword or phrase selected from operation and maintenance corpus (text converted from historical operation and maintenance data) that has a clear reference to "root cause-repair strategy"; for example, in the causal chain "electromagnetic interference event-transformer magnetic field leakage-deploy electromagnetic shield", "electromagnetic interference", "transformer magnetic field leakage", and "electromagnetic shield" can all be keywords, corresponding to event type, root cause, and repair action, respectively; N is the total number of documents; DF(ω) is the number of documents containing keyword ω;

[0180] S314. Construction and update of ternary causal chain knowledge network, integrating event nodes created in the event capture stage, causal relationships and contribution degrees obtained through causal reasoning, and repair strategies bound by countermeasure mapping, to construct a "event-causal-countermeasure" ternary causal chain knowledge network. The knowledge network is based on a semantic topology framework, which associates event nodes to corresponding topology locations, causal relationships connect events and root causes through a chain structure, and repair strategies serve as disposal instructions at the end of the chain;

[0181] When a new abnormal event (the index exceeds the baseline threshold again) is added to the network, repeat steps S311-S313 to integrate the new event node, causal reasoning result, and repair strategy into the existing knowledge network. At the same time, based on the new data, the probability parameters in the knowledge network are iteratively updated, continuously enriching the content of the knowledge network and improving the accuracy and comprehensiveness of fault tracing, so that it can better adapt to the dynamic changes of the network.

[0182] S4. Multi-objective trade-off operation and maintenance decision engine:

[0183] S411. Candidate set determination and decision model construction, from the three-element causal chain knowledge network constructed in S3, extract all root cause-treatment pairs verified by causal reasoning as the candidate set of operation and maintenance decisions; for example, if there is a treatment strategy "firmware upgrade" triggered by the root cause "equipment aging" in the knowledge network, then (equipment aging, firmware upgrade) is a candidate pair; based on the extracted candidate set, a "risk-cost-time" three-dimensional Pareto decision model is constructed to screen out the Pareto optimal solution (i.e. the solution set that cannot improve one target while not deteriorating other targets);

[0184] The objective function of the three-dimensional Pareto decision model is defined as:

[0185]

[0186] Where, R(A i ) is the risk value of strategy A i ; C(A i ) is the cost value; T(A i ) is the repair time;

[0187] S412. Risk dimension evaluation and loss prediction, based on the semantic topology network structure in S2, a random walk model is used to simulate the fault diffusion process; the node connection relationship (such as optical cable link), device dependence (such as master-slave association), and environmental disturbance (such as electromagnetic interference) are taken as diffusion influencing factors to define the fault state transition probability matrix; the future time is divided into time windows, and the diffusion path and influence range of each time window are predicted by matrix iteration; combined with the business importance weight (critical business such as relay protection weight is higher), the expected loss of each time window is calculated, and the total risk value is obtained by summarizing;

[0188] The element in the fault state transition probability matrix represents the probability of fault diffusion from node to node , The probability value is obtained by statistical analysis of historical fault data:

[0189]

[0190] The expected loss calculation formula of different time windows is:

[0191]

[0192] Where, B t is the set of affected businesses in t window; ω b is the business weight; L b is the loss per unit time;

[0193] The total risk value L RThe calculation formula is:

[0194]

[0195] P(t) is the probability of t window fault affecting service;

[0196] S413. Cost dimension estimation and cost calculation, introduce elastic resource pricing function, real-time collection of supply and demand data of human resources, spare parts, transportation, time and other resources (such as current spare parts inventory, operation and maintenance personnel scheduling, traffic conditions), dynamically adjust the unit price of resources; For each treatment strategy, disassemble the required resource type and quantity, combine the elastic pricing to calculate the total cost, and assist the operation and maintenance personnel to weigh the economic cost;

[0197] Elastic resource pricing function:

[0198] c q =c q0 ·α q ;

[0199] c q Real-time elastic resource unit price; c q0 represents the basic unit price of a single elastic resource; α q is the supply and demand influence coefficient;

[0200] The total cost C(A i ) of elastic pricing q The calculation formula is:

[0201]

[0202] n iq is the number of resources q consumed by strategy A i ; p is the total number of elastic resource types;

[0203] S414. Time dimension optimization and repair path finding, taking the stability of the S2 semantic topology as the core constraint, simulating the repair path selection process in the digital twin environment through a lightweight reinforcement learning agent, and finally outputting a closed-loop repair path with the shortest time consumption and the smallest disturbance to the network topology;

[0204] S4141. Topology stability constraint definition and quantification, based on the semantic topology network structure constructed by S2, the stability constraints that the repair path must satisfy are determined, and the environment rules are provided for reinforcement learning:

[0205] Key link protection constraint, mark the link carrying core services (such as relay protection signal, dispatching instruction) as "key link", and the repair path cannot contain or pass through the key link;

[0206] Network connectivity constraint, the connectivity index of network topology should remain unchanged after path adjustment, i.e. Connectivity(G) = 1 to avoid network segmentation or isolated nodes caused by path selection;

[0207] The state space of reinforcement learning is defined as the combination of fault states of network nodes (such as which nodes have failed and which have been repaired); the action space of reinforcement learning is defined as the repair operation between nodes;

[0208] Node vulnerability constraint, preferentially avoid high vulnerability nodes (nodes with vulnerability V > 0.8 in S2) in path selection to reduce the risk of secondary failure in the repair process;

[0209] S4142. Reinforcement learning environment modeling, build a reinforcement learning environment to simulate the dynamic process of repair path selection, the core elements include state space, action space and digital twin interaction mechanism;

[0210] The state space is composed of the current location node, the location of the failed node, the distribution of critical links, and the vulnerability of the node; the action space defines the movement operation of the agent between nodes, i.e. moving from the current node to its adjacent node;

[0211] Digital twin interaction: the agent obtains feedback by interacting with the digital twin environment, which simulates the state of movement time consumption, link load change, topology disturbance in real time, and provides the basis for reward calculation;

[0212] S4143. Reward function design and state transition learning, design the reward function to guide the agent to learn the optimal strategy of "short path + low disturbance", and realize path optimization through iterative update of state-action value;

[0213] The reward function R represents the combination of path time consumption and topology disturbance, the formula is:

[0214] R = -T(P) - λ·P stab ;

[0215] T(P) is the total time consumption of path P; P stab is the topology disturbance value, P stab > 0 when passing through critical links or high vulnerability nodes, otherwise 0; λ is the disturbance penalty coefficient, λ > 0, take a larger value in critical business scenarios;

[0216] State transition learning uses Q-Learning algorithm to update state-action value, the formula is:

[0217] Q(s,z)←Q(s,z)+α′[R+γmax z′ Q(s′,z′)-Q(s,z)];

[0218] Q(s,z) is the state-action value, representing the long-term cumulative reward expectation of selecting action z under the current network state s; a' is the learning rate; g is the discount factor; s' is the new state to which the network topology moves after performing action z; a' is all possible actions; max z′ Q(s',z') is the optimal action value in the new state;

[0219] S4144. Shortest closed loop path optimization and output, taking the closed loop path from "operation starting point-fault node-operation starting point" as the optimization goal, the path space is iteratively explored by the reinforcement learning agent, and finally the optimal path P satisfying the shortest path length and the minimum topology disturbance is output * ;

[0220] Shortest path length:

[0221]

[0222] L(P * ) represents the total length of the path; d(v i ,v i+1 ) is the physical distance or time cost from node v i to v i+1 , and m' is the number of path nodes;

[0223] Minimum topology disturbance, path P * does not pass through the critical link, and the node vulnerability degree is V < 0.6 and the network connectivity index remains Connectivity(G) = 1;

[0224] Optimal path objective function:

[0225]

[0226] L critical is the critical link, i.e. the link marked to carry core business; L(P) is the path link set.

[0227] S5. Immersive visualization and interactive intervention:

[0228] S511. Format standardization and integration of output data of semantic topology, ternary causal chain knowledge network, and three-dimensional Pareto decision model, and injection into cross-platform visualization engine; the platform visualization engine is started based on the lightweight WebGL framework, and through responsive layout adaptation to PC, tablet, mobile terminal and other devices, basic rendering components (lighting, material, camera control) are loaded to ensure the consistency of visualization effects on different terminals;

[0229] S512. Three-dimensional scene layered rendering and state visualization, using a physical rendering (PBR) based WebGL framework, layered rendering of power communication network elements, link state presented by color band and particle flow double way;

[0230] Node rendering: nodes such as towers and substations are modeled at a scale of 1:500, with physical properties mapped to materials (metal towers use metal PBR materials, and optical cable joints use high-gloss materials).

[0231] Node state associated vulnerability V, V>0.8 node edge flicker red (fault warning), 0.5≤V≤0.8 flicker yellow (risk attention), V<0.5 display green (normal state);

[0232] Link rendering:

[0233] Optical fiber link is presented in a cylindrical model, with a diameter positively correlated with "carrying capacity"; wireless link is presented in a dashed model, with line width positively correlated with signal strength;

[0234] Two-dimensional presentation of link state:

[0235] Color band is based on link health mapping color, link health:

[0236] H=0.4(1-interference exposure)+0.6×delay elasticity value;

[0237] Particle flow: particle generation rate is positively correlated with real-time data transmission rate of the link (the higher the rate, the greater the particle density), flow speed is positively correlated with delay elasticity value (the higher the elasticity, the faster the particle speed), dynamically displaying the state of business transmission;

[0238] S513. Two-dimensional board design and multi-view interactive, design "risk heat-cost curve-residual life" three-view interactive panel, realize multi-dimensional visualization and interactive trigger of decision data;

[0239] Three-view design:

[0240] Risk heat map, with S2 semantic topology as the base map, node color depth positively correlated with risk value L R (L R The higher, the closer the color to dark red), intuitively presenting the risk distribution of the whole network;

[0241] Cost curve, horizontal axis for candidate disposal strategies (such as "firmware upgrade" "replace module"), vertical axis for cost (calculated according to the elasticity pricing total cost formula), with broken line graph to mark the cost of each strategy, Pareto optimal solution marked with a red dot;

[0242] Remaining life view, based on device runtime and vulnerability V, to predict remaining life:

[0243] L 剩余 = (1 - t / design life) x (1 - V);

[0244] The remaining life is displayed in the form of a progress bar + percentage (such as "Optical module remaining life: 68%").

[0245] Linkage interaction logic, the user clicks any view element (such as the red node of the risk heat map), the system triggers a triple response:

[0246] The three-dimensional scene automatically focuses on the corresponding physical node (through coordinate positioning), and highlights the associated link;

[0247] Pop-up causal chain pop-up window, display S3 event nodes, root cause contribution (such as "Device aging contribution 70%") and repair strategy text;

[0248] Synchronize the cost curve and the remaining life view, locate the strategy cost and device life data corresponding to the node, and mark the association with a dashed line;

[0249] S513. Multi-modal interaction and closed-loop verification execution, multi-modal interaction voice interaction, gesture interaction, touch screen interaction, etc.; If the resource gap Δn > 0, the system pop-up window prompts "need to allocate Δn spare parts", and automatically updates the S4 cost model; Where, Δn = n n -n s , n n is the number of spare parts required for repair, n s is the current inventory;

[0250] After the operation is completed, the system records the track data (operation time, object ID, resource change amount), synchronously writes back to the S3 digital twin, updates the strategy execution record of the knowledge network, and corrects the node state of the S2 topology to ensure that the digital space and the physical system state are consistent.

[0251] The above describes a preferred embodiment of the invention content, and does not limit the invention content. Any modification, equivalent replacement and improvement made within the spirit and principle of the invention content shall be included in the protection scope of the invention content.

Claims

1. A digital-twin-based power communication network visual operation and maintenance method, characterized in that, Comprise the following steps: S1. Global data mirror and multi-source coupling mapping, reconfigurable perception array is laid out at global nodes of power communication network, array encodes original physical quantity into unified semantic tag stream through switchable topology bus in real time, and outputs space-time consistent three-domain global digital mirror after edge time grid alignment algorithm; S2. Semantic topology reconstruction of digital twin, input three-domain mirror of S1, dynamic topology reconstruction is carried out on network by using evolvable graph semantic engine, and output semantic topology is a dynamic graph which can grow by itself; S3. Fault tracing based on interpretable causal chain, on the basis of semantic topology reconstruction, "event-cause-countermeasure" ternary causal chain knowledge network is constructed, repair strategy highly matched with root cause is screened out, and "event-root cause-disposal" one-to-one correspondence is formed; S4. Multi-objective trade-off operation and maintenance decision engine, "root cause-disposal" pair output by ternary causal chain knowledge network is taken as candidate set, "risk-cost-time efficiency" three-dimensional Pareto decision model is constructed, and Pareto optimal operation and maintenance strategy is screened out; S5. Immersive visualization and interactive intervention, output results of semantic topology, ternary causal chain knowledge network and three-dimensional Pareto decision model are injected into cross-platform visualization engine.

2. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, The array is composed of miniaturized light-electric-vibration three-mode sensitive units, programmable radio frequency front end and low-power consumption computing core; the three-domain digital mirror is logically divided into "visible light layer, radio frequency layer and structure layer", millisecond-level search is realized in each domain through adaptive hash index, cross-domain mapping is completed between three domains through scalable coupling matrix, and it is ensured that any physical event can be uniquely and instantaneously corresponded in digital space.

3. The power communication network visual operation and maintenance method based on digital twinning according to claim 2, characterized in that, The adaptive hash index function is: H(x) = (a*x+b)modm; x is data characteristic value, a and b are dynamic parameters, and m is index table length; Extensible coupling matrix element M i,j The calculation formula is: M i,j = Corr(D i , D j ); D i , D j are physical quantity data sets of different domains respectively; Corr(·) is a Pearson correlation coefficient, ensuring that the physical event corresponds uniquely in the digital space; M i,j represents the coupling correlation degree value between the ith domain and the jth domain; different combinations of i and j in the matrix can present the coupling relationship between each two of the multi-domain, and the correlation network of multi-domain cooperation can be constructed through the whole matrix; The edge time grid alignment algorithm formula is: Δt a = t l - t o ; Wherein, t l is the local sampling time; t o is the clock synchronization deviation compensation value.

4. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, The evolvable graph semantic engine has dynamic processing capacity, can continuously reconstruct and update network topology based on real-time data in three-domain mirror, and the method comprises the following steps: Node semantic abstraction and modeling, each communication node is abstracted into a "function-state-vulnerability" triplet; wherein, the function is automatically labeled by business flow characteristics, and the business role of the node in the network is clear; the state is generated by real-time multi-dimensional index compression, and the current running state of the node is quantitatively reflected; the vulnerability V is calculated by weighting the historical disturbance frequency, and the formula is: k represents the serial number of the historical disturbance index, used to distinguish different types or different time of disturbance events; n represents the total number of historical disturbance indexes participating in the calculation, covering all key disturbance records of the node within a certain period; A k represents the kth historical disturbance index; ω k is the weight corresponding to the kth disturbance; f is the vulnerability conversion function; Link semantic vector construction and dynamic adjustment, "bearing capacity-interference exposure-delay elasticity" three-axis vector is established for optical fiber and wireless dual-mode link respectively; link weight is adaptively adjusted with real-time link quality drift, and accurate description of link state is ensured; Link weight dynamic adjustment formula is: W(t) = W0*(1+alpha*Q(t)); W(t) represents the real-time weight of the link at time t, which quantifies the importance or performance level of the link in the current network topology; W0 is the initial weight of the link, which is the baseline weight value when the link is running normally and there is no quality fluctuation; a is a sensitivity coefficient set according to the importance of the link; Q(t) is the real-time quality index of the link at t; Topology evolution and attribute update, introduce a differentiable graph rewriting operator, when the vulnerability of any link exceeds the preset threshold, automatically insert logical relay or adjust the routing path under the constraint of maintaining network service connectivity, and synchronously update the geometry and physical attributes of the twin, so that the output semantic topology becomes a self-growing dynamic graph that adapts to the dynamic changes of the network; The constraint condition for maintaining service connectivity is: argmin∑ΔL op s.t Connectivity(G) = 1; argmin represents an optimization function, AL op is the length variation of the oth path in the network before and after adjustment; Connectivity(G) represents the connectivity index of the network topology graph G; Connectivity(G) = 1 indicates that the network is connected.

5. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, The carrying capacity quantifies the current carrying capacity of the link, reflecting the load capacity of the link; the interference exposure evaluates the degree of influence of external electromagnetic interference and environmental noise factors on the link, and the higher the value, the weaker the anti-interference ability; the delay elasticity represents the adaptability of the link to transmission delay fluctuations, and the higher the elasticity, the more stable the link can maintain in the case of delay change.

6. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, The S3 includes: S311. Event capture and node encapsulation, based on the completed semantic topology, real-time monitoring of each node index in the network; set the baseline threshold judgment rule: For any index, the baseline threshold T is calculated as the historical mean μ of the index plus 3 times the standard deviation σ, that is: T = μ + 3σ; When the monitored index is greater than the baseline threshold T, it is determined that the index drift exceeds the baseline threshold, triggering the causal trigger; when the monitored index is less than the baseline threshold T, the abnormal direction is determined according to the characteristics of the index; S312. Causal reasoning and contribution calculation, taking the encapsulated event node as the target, using the interpretable Bayesian causal discovery algorithm for causal reasoning, taking environmental disturbance, equipment aging, and business burst as potential parent nodes, and calculating the causal contribution of each parent node to the event through the Bayesian formula: E is the current event node; C i is the th potential parent node; is the parent node occurs when event E occurs; P(E|C) is the conditional probability that event E occurs when event C occurs; is the prior probability (based on historical data statistics) of the parent node ​ P(E) is the marginal probability of event E; The calculation result is output in the form of natural language fragments, so that the reasoning process and result can be understood by the operation and maintenance personnel, and the key factors of fault initiation are clear; S313. Countermeasure mapping and repair strategy binding, based on the root cause obtained by causal reasoning, automatically binding repair strategies at the end of each causal chain; the repair strategy is derived from the operation and maintenance corpus distillation process, and high-value repair experience is selected from historical operation and maintenance data through the TF-IDF algorithm; The formula for extracting high-value experience from historical operation and maintenance data through the TF-IDF algorithm is: TF(ω, d) is the term frequency of keyword ω in document set d, ω is a keyword or phrase selected from the operation and maintenance corpus that has a clear reference to "root cause-repair strategy"; N is the total number of documents; DF(ω) is the number of documents containing keyword ω; The TF-IDF algorithm selects repair strategies that highly match the root cause, forming a one-to-one correspondence between "event-root cause-treatment"; S314. Three-element causal chain knowledge network construction and update, integrate the event nodes created in the event capture stage, the causal relationship and contribution degree obtained by causal reasoning, and the repair strategy bound by countermeasure mapping, to construct a "event-causal-countermeasure" three-element causal chain knowledge network; the knowledge network takes semantic topology as the basic framework, associates the event nodes to the corresponding topological position, connects the events and root causes through chain structure for causal relationship, and takes the repair strategy as the disposal guide at the end of the chain; When a new abnormal event is added to the network, repeat steps S311-S313 to integrate the new event node, causal reasoning result, and repair strategy into the existing knowledge network; at the same time, based on the new data, iteratively update the probability parameters in the knowledge network, continuously enrich the content of the knowledge network, and improve the accuracy and comprehensiveness of fault tracing, so that it can better adapt to the dynamic changes of the network.

7. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, In the "risk-cost-time" three-dimensional Pareto decision model: Risk represents the risk dimension, adopts risk diffusion simulation based on random walk to predict the expected loss of the fault on the overall network service availability in the next N time windows; Cost represents the cost dimension, introduces an elastic resource pricing function to estimate the manpower, spare parts, transportation and time cost required to execute each disposal strategy in real time; Time represents the time dimension, through a lightweight reinforcement learning agent, the shortest closed-loop repair path is found under the premise of maintaining the stability of the network topology.

8. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, The steps of constructing the "risk-cost-time" three-dimensional Pareto decision model and screening the Pareto optimal operation and maintenance strategy include: S411. Candidate set determination and decision model construction, from the three-element causal chain knowledge network constructed in S3, extract all root cause-disposal pairs that have passed causal reasoning verification as the candidate set for operation and maintenance decision; based on the extracted candidate set, construct a "risk-cost-time" three-dimensional Pareto decision model to screen out the Pareto optimal solution; The objective function of the three-dimensional Pareto decision model is defined as: where R(A i ) is the risk value for policy A i ; C(A i ) is the cost value; and T(A i ) is the repair time S412. Risk dimension evaluation and loss prediction, based on S2 semantic topology network structure, using random walk model to simulate fault diffusion process; taking node connection relationship, equipment dependence, environmental disturbance as diffusion influencing factors, define fault state transition probability matrix; divide future time into time windows, predict diffusion path and influence range of each time window through matrix iteration; combine with business importance weight, calculate expected loss of each time window, and summarize to get total risk value; elements in the failure state transition probability matrix representing the probability of a failure spreading from a node to a node , The probability values are obtained by statistical analysis of historical failure data: The expected loss calculation formula for different time windows is: where B t is the set of affected services for t window; ω b is the service weight; L b is the loss per unit time; Total risk value L R Calculation formula: P(t) is the probability of the t-window fault affecting the service; S413. Cost dimension estimation and cost calculation, introduce an elastic resource pricing function to collect supply and demand data of manpower, spare parts, transportation and time resources in real time, and dynamically adjust the unit price of resources; for each disposal strategy, decompose the required resource types and quantities, calculate the total cost by combining the elastic pricing, and assist operation and maintenance personnel in weighing the economic cost; Elastic resource pricing function: c q = c q0 · α q ; c q real-time elastic resource unit price; c q0 denotes the base unit price of a single elastic resource; α q is the supply-demand influence coefficient; The total cost C(A i ) q The calculation formula is: n iq For strategy A i the number of resources q to be consumed; p is the total number of elastic resource types; S414. Time dimension optimization and repair path finding, taking the stability of the semantic topology in S2 as the core constraint, simulating the repair path selection process in the digital twin environment through a lightweight reinforcement learning agent, and finally outputting the closed-loop repair path with the shortest time consumption and the smallest disturbance to the network topology; The steps of simulating the repair path selection process in the digital twin environment through a lightweight reinforcement learning agent, and finally outputting the closed-loop repair path with the shortest time consumption and the smallest disturbance to the network topology include: S4141. Topology stability constraint definition and quantification, based on the semantic topology network structure constructed in S2, clearly define the stability constraints that the repair path must satisfy, and provide environmental rules for reinforcement learning: Key link protection constraint, mark the link carrying core service as "key link", the repair path cannot contain or cross the key link; Network connectivity constraint, the network topology connectivity index should remain unchanged after path adjustment, i.e. Connectivity(G) = 1 to avoid network segmentation or isolated nodes caused by path selection; The state space of reinforcement learning is defined as the combination of fault states of network nodes; the action space of reinforcement learning is defined as the repair operation between nodes; Node vulnerability constraint, the path preferentially avoids high vulnerability nodes to reduce the risk of secondary failure in the repair process; High vulnerability nodes are nodes with vulnerability V > 0.8 in S2; S4142. Reinforcement learning environment modeling, build a reinforcement learning environment to simulate the dynamic process of repair path selection, the core elements include state space, action space and digital twin interaction mechanism; The state space is composed of the current location node, the location of the fault node, the distribution of key links and the node vulnerability; the action space defines the movement operation between nodes, i.e. moving from the current node to its adjacent node; Digital twin interaction: the agent obtains feedback by interacting with the digital twin environment, which simulates the movement time, link load change and topology disturbance state in real time to provide the basis for reward calculation; S4143. Reward function design and state transition learning, design the reward function to guide the agent to learn the optimal strategy of "short path + low disturbance", and update the state-action value through iteration to achieve path optimization; The reward function R represents the comprehensive path time consumption and topology disturbance, and the formula is: R = -T(P) - λ - P stab ; T(P) is the total time consumption of path P; P stab is the topology perturbation value, P stab > 0 when passing through a critical link or a node with high vulnerability, otherwise 0; λ is the perturbation penalty coefficient, λ > 0, and a larger value is taken in a critical service scenario; The state transition learning updates the state-action value using the Q-Learning algorithm, and the formula is: Q(s, z) <- Q(s, z) + a'[R + γmax z′ Q(s', z') - Q(s, z)]; Q(s,z) is the state-action value, representing the long-term cumulative reward expectation of choosing action z at the current network state s; a' is the learning rate; g is the discount factor; s' is the new state to which the network topology moves after performing action z; a' is all possible actions; max z′ Q(s',z') is the optimal action value at the new state; S4144. Shortest closed-loop path optimization and output, taking the closed-loop path from "operation starting point → fault node → operation starting point" as the optimization goal, exploring the path space through reinforcement learning agent iteration, and finally outputting the optimal path P that meets the shortest path length and the smallest topology disturbance * ; Shortest path length: L(P * ) denotes the total length of the path; d(v i ,v i+1 ) is the physical distance or time cost from node v i to v i+1 , and m' is the number of nodes in the path. Topology perturbation is minimal, path P * Not through the key link, through the node vulnerability degree is less than 0.6 and network connectivity index remains connectivity(G)=1; Optimal path objective function: Connectivity(G) = 1; L critical L is the set of links in the path; L(P) is the set of links in the path.

9. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, The S5 comprises: S511. Format standardization and integration of the output data of semantic topology, triadic causal chain knowledge network and three-dimensional Pareto decision model, and injection into a cross-platform visualization engine; the platform visualization engine is started based on a lightweight WebGL framework, and through responsive layout adaptation to PC, tablet and mobile devices, loads basic rendering components to ensure consistency of visualization effects on different terminals; S512. Three-dimensional scene layered rendering and state visualization, using a WebGL framework based on physical rendering (PBR) to perform layered rendering on power communication network elements, and presenting link states through color band and particle flow in a dual manner; Node rendering: the tower and substation nodes are modeled at a scale of 1:500, and the material mapping is physically mapped; Node state association vulnerability V, when V > 0.8, the node edge prompts a failure warning, when 0.5 ≤ V ≤ 0.8, it prompts a risk attention, and when V < 0.5, it prompts a normal state; Link rendering: Fiber links are presented in a cylindrical model, and the diameter is positively correlated with the "carrying capacity"; wireless links are presented in a dashed model, and the line width is positively correlated with the signal strength; Two-dimensional presentation of link state: Color band is based on link health degree mapping color, and the link health degree is: H = 0.4 (1 - interference disturbance exposure) + 0.6 x delay elasticity value; ​ ​ Particle Flow: The particle generation rate is positively correlated with the real-time data transmission rate of the link, and the flow speed is positively correlated with the latency elasticity value, dynamically displaying the service transmission status; S513. Two-dimensional dashboard design and multi-view linkage interaction: Design a three-view linkage panel of "risk heat map - cost curve - remaining life" to realize multi-dimensional visualization and interactive triggering of decision data; Three-view design: Risk heat map, with S2 semantic topology as the base map, node color depth and risk value L R Positive correlation, intuitively presents the risk distribution of the whole network; The cost curve has the horizontal axis representing the disposal strategies in the candidate set and the vertical axis representing the cost. The cost of each strategy is marked with a line graph, and the Pareto optimal solution is marked with a red dot. S513. Multi-modal interaction and closed-loop verification execution, multi-modal interaction voice interaction, gesture interaction, touch screen interaction; if resource gap Δn > 0, the system pop-up window prompts "need to allocate Δn spare parts", and automatically updates the S4 cost model; wherein, Δn = n n -n s , n n is the number of spare parts required for repair, n s is the current inventory; After the operation is completed, the system records the trajectory data, synchronously writes back to the S3 digital twin, updates the policy execution records of the knowledge network, and corrects the node status of the S2 topology to ensure that the digital space is consistent with the physical system status.

10. The power communication network visual operation and maintenance method based on digital twinning according to claim 1, characterized in that, The vertical axis cost is calculated according to the aforementioned formula for calculating the total cost of flexible pricing. The remaining lifetime view, based on device runtime and vulnerability V, predicts the remaining lifetime: L 剩余 = (1 - t / design life) x (1 - V); Remaining lifespan is displayed as a progress bar plus a percentage. The interactive logic triggers a triple response when a user clicks on any view element: The 3D scene automatically focuses on the corresponding physical node and highlights the associated links; A causal chain pop-up window appears, displaying the S3 event nodes, root cause contribution, and repair strategy text; Synchronously update the cost curve and remaining life view, locate the strategy cost and equipment life data corresponding to the node, and mark the relationship with dashed lines.

Images