Simulation Analysis Method for Metering Equipment Distribution Decisions

Abstract

This invention discloses a simulation analysis method for delivery decisions of metering equipment, belonging to the field of computer data processing and simulation technology. It includes: constructing a spatiotemporal knowledge dataset by fusing multi-source heterogeneous data; performing causal inference based on the dataset to establish a causal relationship graph; performing risk propagation deduction and causal link evolution simulation on the graph to generate a risk scenario set; generating counterfactual intervention strategies for the risk scenarios; performing quantitative analysis to output the optimal delivery strategy; and using the analysis results to update the causal relationship graph. This invention employs a data processing architecture that uses causal inference to construct a dynamic knowledge graph, enabling multi-step propagation deduction and link evolution simulation of risks in complex systems. This significantly improves the accuracy and robustness of the data processing system in making forward-looking decisions under uncertain environments.

Description

Technical Field

[0001] This invention relates to the field of computer data processing and simulation technology, and in particular to a simulation analysis method for decision-making in the delivery of metering equipment. Background Technology

[0002] Metering equipment, as a key terminal device in the power system used for electricity metering, data acquisition, and information management, directly affects the accuracy of power grid billing, line loss management, and customer service quality through its stable operation. The decision-making process for distributing metering equipment involves multiple stages, including new product installation, faulty replacement, and periodic rotation. Its core lies in how to rationally plan the warehousing layout and distribution routes of spare parts based on demand forecasts, inventory status, and operational pressure to ensure rapid response to operational needs while maintaining controllable costs.

[0003] Existing decision support technologies for the distribution of metering equipment typically rely on statistical analysis of historical data. For example, analyzing the geographical distribution and frequency of historical fault work orders can predict future spare parts demand, and this is combined with inventory management models to formulate warehousing plans. Some more advanced methods introduce geographic information systems for visualization analysis, or use simple regression models to predict fault trends, thereby providing some reference for distribution strategies.

[0004] However, the aforementioned existing technologies have significant limitations. Most are based on superficial correlation analysis of events, making it difficult to delve into the root causes of failures and the transmission relationships between different events, resulting in low prediction accuracy. Furthermore, these methods are typically static and cannot effectively address dynamic risks arising from the coupling of multiple factors such as environmental changes, equipment aging, and power grid load fluctuations. Decision-making solutions often lack foresight and adaptability to extreme situations, easily leading to resource waste or delayed emergency response. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a simulation analysis method for delivery decisions of metering equipment. It employs a data processing architecture that uses causal inference to construct a dynamic knowledge graph, enabling multi-step propagation and link evolution simulation of risks in complex systems. This significantly improves the accuracy and robustness of the data processing system in making forward-looking decisions under uncertain environments.

[0006] The above objectives can be achieved through the following approach:

[0007] A simulation analysis method for metering equipment delivery decisions includes collecting and fusing geographic information, historical maintenance work orders, real-time warehousing data, and external environmental data of the metering equipment to generate a spatiotemporal knowledge dataset; based on the spatiotemporal knowledge dataset, performing causal inference to mine the correlation between events and constructing a causal relationship graph; based on the causal relationship graph, using graph information propagation to perform multi-step risk propagation simulation, calculating and generating a risk quantification graph; based on the causal relationship graph and the risk quantification graph, simulating the dynamic evolution of causal links to generate an extended risk scenario set; for the extended risk scenario set, calculating the risk contribution of nodes, and filtering key risk nodes through sorting optimization to generate a counterfactual intervention strategy set; for the counterfactual intervention strategy set, performing cost-benefit quantification analysis, outputting metering equipment delivery strategy data, and feeding the quantification analysis results back to the causal relationship graph.

[0008] Optionally, the generation of the spatiotemporal knowledge dataset includes: dividing the geographic information into grid cells, binning the historical maintenance work orders according to fault type and time period, mapping the real-time warehousing data and the external environment data to the same grid, and constructing a three-dimensional sparse tensor index; performing tensor decomposition on the three-dimensional sparse tensor index and completing the missing data to separate spatiotemporal pattern factors and event factors; mapping the spatiotemporal pattern factors and the event factors to graph node features, and propagating information between adjacent nodes to generate a spatiotemporal knowledge dataset.

[0009] Optionally, the construction of the causal relationship graph includes: identifying causal candidate relationships for nodes and edges in the spatiotemporal knowledge dataset, detecting the sequence of events and spatiotemporal co-occurrence, and screening potential causal event pairs; based on the potential causal event pairs, performing causal verification and reinforcement, verifying the causal strength using multi-context comparison and counterfactual experiments, and generating weighted causal relationships; and performing topological organization and hierarchical layering according to the weighted causal relationships to form a causal relationship graph.

[0010] Optionally, the generation of weighted causal relationships includes: analyzing and extracting the co-occurrence of causal events based on the potential causal event pairs and the spatiotemporal knowledge dataset to obtain contextual co-occurrence factors; using the contextual co-occurrence factors as retrieval conditions to match and construct a simulated control group dataset from the spatiotemporal knowledge dataset; and analyzing the event occurrence overview and quantifying the causal intensity based on the simulated control group dataset to obtain weighted causal relationships.

[0011] Optionally, the method further includes: performing statistical co-occurrence analysis based on the spatiotemporal pattern factor and the event factor to generate a factor co-occurrence matrix; constructing a graph topology and calculating network centrality based on the weighted causal relationship to generate causal centrality vectors for nodes, forming a causal centrality vector set; and dynamically adjusting the information propagation weights in the risk propagation deduction by fusing the factor co-occurrence matrix and the causal centrality vector set.

[0012] Optionally, the calculation and generation of the risk quantification map includes: identifying risk source nodes based on the causal relationship map and initializing the risk state to form an initial risk vector set; performing iterative risk propagation deduction on the causal relationship map based on the initial risk vector set until propagation converges to generate a full node prognostic risk vector set; and performing weighted fusion based on the full node prognostic risk vector set to generate the risk quantification map.

[0013] Optionally, generating the extended risk scenario set includes: tracking and recording multiple consecutive diagnostic cycles based on the risk quantification map and the causal relationship map to obtain the map activation frequency and intensity distribution; analyzing and identifying causal links with sustained pressure and abnormal sensitivity based on the activation frequency and intensity distribution, and calculating the link vulnerability index; and using the link vulnerability index to weighted guide the dynamic evolution process of the simulated causal link to generate the extended risk scenario set.

[0014] Optionally, generating the counterfactual intervention strategy set includes: calculating the risk contribution distribution of nodes under different scenarios based on the extended risk scenario set and the full node prognostic risk vector set, to obtain a node risk contribution matrix; performing multi-objective ranking optimization based on the node risk contribution matrix, and selecting key risk nodes according to risk fluctuation amplitude and scenario sensitivity respectively; and mapping the key risk nodes to the causal relationship graph to analyze physical attributes and functional constraints to form a counterfactual intervention strategy set.

[0015] Optionally, the step of selecting key risk nodes based on risk fluctuation amplitude and situation sensitivity includes: calculating risk fluctuation gradient and sensitivity index based on the node risk contribution matrix to obtain a risk fluctuation matrix and a sensitivity index set; applying weighted sorting to the risk fluctuation matrix and the sensitivity index set respectively, and performing set cross-fusion of the sorting results to generate a comprehensive sorting list; and selecting key risk nodes according to priority order based on the comprehensive sorting list.

[0016] Optionally, the output metering equipment distribution strategy data includes: calculating the strategy lifecycle cost and power distribution reliability based on the counterfactual intervention strategy set and the node risk contribution matrix to obtain a strategy utility vector set; performing utility normalization processing and sorting based on the strategy utility vector set to output the metering equipment distribution strategy; and adjusting the causal edge weights in the causal relationship graph using the strategy utility vector set and the metering equipment distribution strategy data.

[0017] Compared with the prior art, the present invention has the following advantages:

[0018] 1. This invention constructs a spatiotemporal knowledge dataset, enabling deep fusion and correlation mining of multi-source heterogeneous data from geography, operations and maintenance, warehousing, and the environment. Compared to traditional fragmented data processing methods, this approach can capture complex coupling relationships between events in both time and space, such as the lagging correlation between specific climate patterns and equipment failure rates. This significantly improves the fundamental depth and data completeness of our understanding of system status, laying a solid foundation for subsequent accurate analysis.

[0019] 2. This invention achieves a scientific closed loop from risk identification to strategy generation by generating an expanded risk scenario set and using multi-objective optimization to screen key nodes. By proactively exploring the unknown risks that vulnerable links may cause, and combining risk volatility and scenario sensitivity to assess the importance of nodes, this method can accurately locate the weakest links with the greatest intervention value. The final set of hypothetical adjustment solutions is highly operable and cost-effective, transforming the decision-making process from relying on experience to relying on data-driven scientific optimization, thereby improving resource allocation efficiency and overall risk resistance.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the simulation analysis method for metering equipment delivery decision-making according to an embodiment of the present invention.

[0023] Figure 2This is a factor association clustering bubble diagram according to an embodiment of the present invention.

[0024] Figure 3 This is a Sankey diagram illustrating the causal link risk propagation and quantification in an embodiment of the present invention.

[0025] Figure 4 This is a composite correlation diagram of factor co-occurrence and risk contribution in an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figure 1 One embodiment of the present invention proposes a simulation analysis method for delivery decisions of metering equipment. It adopts a data processing architecture that constructs a dynamic knowledge graph based on causal inference, which can perform multi-step propagation and link evolution simulation of risks in complex systems, and significantly improve the accuracy and robustness of the data processing system in making forward-looking decisions under uncertain environments.

[0028] The method described in this embodiment specifically includes:

[0029] Collect and integrate geographic information of metering equipment, historical maintenance work orders, real-time warehouse data and external environmental data to generate a spatiotemporal knowledge dataset;

[0030] Based on the spatiotemporal knowledge dataset, causal inference is performed to mine the correlation between events and construct a causal relationship graph.

[0031] Based on the aforementioned causal relationship graph, multi-step risk propagation is performed using graph information propagation to calculate and generate a risk quantification graph.

[0032] Based on the causal relationship graph and the risk quantification graph, the dynamic evolution of causal links is simulated to generate an extended risk scenario set;

[0033] For the extended risk scenario set, the risk contribution of each node is calculated, and key risk nodes are screened through sorting optimization to generate a set of counterfactual intervention strategies;

[0034] For the set of counterfactual intervention strategies, a cost-benefit quantitative analysis is performed, and data on the distribution strategy of metering equipment is output. The results of the quantitative analysis are then fed back to the causal relationship graph.

[0035] The data processing architecture that uses causal inference to construct a dynamic knowledge graph can perform multi-step propagation and link evolution simulation of risks in complex systems, significantly improving the accuracy and robustness of the data processing system in making forward-looking decisions under uncertain environments.

[0036] Optionally, the generated spatiotemporal knowledge dataset includes:

[0037] The geographic information is divided into grid cells, the historical maintenance work orders are divided into buckets according to fault type and time period, the real-time warehouse data and the external environment data are mapped to the same grid, and a three-dimensional sparse tensor index is constructed.

[0038] Specifically, this step aims to unify diverse and structurally varied raw data into a high-dimensional, structured data container, laying the foundation for subsequent deep pattern mining. A processing unit first performs spatial discretization, dividing the geographically covered area into a series of standardized grid cells. Simultaneously, historical maintenance work orders are preprocessed, aggregated and bucketed based on the clearly defined fault or business type and occurrence time within their content. Subsequently, the processing unit precisely maps real-time warehouse data and external environmental data to their corresponding spatiotemporal coordinates, i.e., specific grid cells and time buckets. Through these steps, a three-dimensional sparse tensor index with spatial, temporal, and event dimensions is constructed. Mathematically, this data structure is a high-dimensional array with most elements being zero, thus exhibiting "sparse" characteristics.

[0039] For the three-dimensional sparse tensor index, tensor decomposition is performed and missing data is filled in to separate the spatiotemporal pattern factor and event factor.

[0040] Specifically, this step aims to uncover potential correlation patterns and fill in missing information from sparse and incomplete data using a higher-order factorization model. This process decomposes the original tensor into a core tensor and factor matrices for each dimension, capturing more complex interactions between factors than traditional decomposition methods. An exemplary decomposition can be represented by the following formula:

[0041] ,

[0042] in, For three-dimensional sparse tensor indexing; This is the core tensor, whose elements represent the interaction weights between different factors; , , These are factor matrices representing the three dimensions of space, time, and event. This represents the n-modulo product of a tensor and a matrix. In this model, the factor matrix... and The column vectors together constitute the spatiotemporal pattern factor; the factor matrix The column vectors constitute the event factors. The processing device solves this model using, for example, a higher-order singular value decomposition optimization algorithm, during reconstruction. During the process, the missing data was filled in, and the core tensor and various factor matrices were successfully separated, such as... Figure 2 As shown, the figure uses the size and color of bubbles in the central area to demonstrate the quantitative correlation between different spatiotemporal pattern factors and event factors. The top and right sides of the figure use clustering tree diagrams to demonstrate the cluster categories automatically formed by these factors based on their correlations, which are used to reveal the deep internal structure of the data.

[0043] The spatiotemporal pattern factors and event factors are mapped to graph node features, and information is propagated between adjacent nodes to generate a spatiotemporal knowledge dataset.

[0044] Specifically, this step aims to transform the abstract factors extracted in the previous step into a more information-rich graph structure data that can be used for subsequent causal inference. The processing device first creates an initial node for each spatiotemporal pattern factor and event factor, using the corresponding factor vector as the initial feature of that node. Then, to ensure that the node's features include contextual information about its local environment, the processing device performs an information propagation operation between nodes representing geographically or temporally adjacent nodes. This operation, through an aggregation and update mechanism, weightedly integrates the feature information of neighboring nodes into the feature representation of the central node. Through this process, the features of each node are enriched and enhanced, ultimately forming a graph structure containing rich node features—the spatiotemporal knowledge dataset.

[0045] Optionally, constructing the causal relationship graph includes:

[0046] Causal candidate relationships are identified for nodes and edges in the spatiotemporal knowledge dataset to detect event sequence and spatiotemporal co-occurrence and screen potential causal event pairs.

[0047] Specifically, this step aims to initially filter out potentially causal relationships from the massive relationships contained in the spatiotemporal knowledge dataset, thereby improving the efficiency of subsequent verification. A processing device traverses all event node pairs in the spatiotemporal knowledge dataset, filtering them using two core criteria. The first is to detect event sequence, meaning the timestamp of the causal event must be earlier than the result event, and the time interval between the two must fall within a preset window that conforms to business logic. The second is to detect spatiotemporal co-occurrence, which assesses the proximity or correlation strength of two events in the geographic and temporal dimensions, for example, by calculating their mutual information values ​​in the dataset. Only event pairs that simultaneously meet the conditions of temporal sequence and spatiotemporal co-occurrence are identified and filtered as potential causal event pairs.

[0048] Based on the potential causal event pairs, causal verification and reinforcement are performed, and the strength of causality is verified by multi-scenario comparison and counterfactual experiments to generate weighted causal relationships.

[0049] Specifically, this step aims to extract genuine causality from the correlations identified in the previous step, which is crucial for constructing a reliable causal graph. For each potential causal event pair, the processing device performs rigorous causal verification and reinforcement. This process first utilizes multi-contextual comparison, analyzing whether the conditional probability of the result event remains stable under different background contexts within a spatiotemporal knowledge dataset. Subsequently, counterfactual experiments are performed to quantify the causal strength by constructing a simulated control group dataset and comparing the difference in the probability of the result event occurring in the real context and the simulated control group. The magnitude of this probability difference is used to quantify the strength of the causal relationship, ultimately generating a series of weighted causal relationships with direction and credibility weights.

[0050] Based on the weighted causal relationships, topological organization and hierarchical layering are performed to form a causal relationship graph.

[0051] Specifically, this step aims to organize the discrete causal relationships verified in the previous step into a structured global network. The processing device first performs topological sorting on all weighted causal relationships. This sorting process includes pruning, removing edges with causal strength below a preset threshold to ensure the reliability of the graph. Subsequently, topological sorting and hierarchical layering are performed to ensure the entire graph is a Directed Acyclic Graph (DAG), which conforms to the logic of irreversible causal relationships. In this way, all event nodes are naturally divided into different levels, forming a clear propagation path from the root cause event to the final result event. The weighted directed network with a hierarchical topological structure obtained after this sorting and layering is the final causal graph.

[0052] Optionally, the generation of weighted causal relationships includes:

[0053] Based on the potential causal event pairs and the spatiotemporal knowledge dataset, the co-occurrence of causal events is analyzed and extracted to obtain the contextual accompaniment factor;

[0054] Specifically, its core lies in precisely quantifying the causal strength between potential causal event pairs through simulated controlled experiments. The processing device first analyzes, for each potential causal relationship, other environmental or state characteristics that typically accompany the occurrence of causal event C in the spatiotemporal knowledge dataset. These co-occurring features are contextual concomitant factors, which are confounding variables that may affect the outcome. For example, if event C is "abnormal cable temperature in a certain area," association rule analysis might reveal that it frequently occurs simultaneously with the factors "high summer temperatures" and "high industrial load."

[0055] Using the contextual accompanying factors as search criteria, a simulation control group dataset is matched and constructed from the spatiotemporal knowledge dataset;

[0056] Specifically, this step aims to eliminate confounding bias and is crucial to the entire causal strength quantification method. The processing device uses the contextual co-occurrence factors extracted in the previous step as search criteria to match across the entire spatiotemporal knowledge dataset. All samples that satisfy the contextual conditions and where the causal event C actually occurred are assigned to the "experimental group." Simultaneously, all samples that satisfy the same contextual conditions but where the causal event C did not occur are selected to form the simulated control group dataset. In this way, we ensure that the experimental and control groups are statistically comparable under all known important background conditions, with the only significant difference being whether the causal event C occurred. This conceptually simulates a randomized controlled trial in scientific research.

[0057] Based on the simulated control group dataset, the event occurrence overview is analyzed and the causal strength is quantified to obtain weighted causal relationships.

[0058] Specifically, this step aims to quantify the final causal strength based on the constructed experimental and control groups. To obtain a more robust estimate than simple risk difference, this invention employs a dual robust estimation method. This method integrates a propensity score model and an outcome regression model, ensuring an unbiased causal effect estimate even if one model is misspecificated. An exemplary dual robust estimation model for quantifying causal strength can be expressed by the following formula:

[0059] ,

[0060] in, This is a dual robust estimate of the average causal effect, i.e., the causal strength in this invention; The total number of samples; As an indicator variable, it is 1 if the causal event in sample i occurs, and 0 otherwise; This is the outcome variable; it is 1 if the outcome event occurs in sample i, and 0 otherwise. The contextual accompanying factor vector corresponding to sample i; For factor-based A score indicating a tendency to predict the occurrence of causal events; and Based on factors, under the two conditions of the occurrence and non-occurrence of the causal event, respectively. The regression prediction of the probability of the outcome event occurring. Calculated... The value is assigned a corresponding causal relationship as its weight, ultimately forming a weighted causal relationship.

[0061] Optionally, the method further includes:

[0062] Based on the spatiotemporal pattern factors and the event factors, a statistical co-occurrence analysis is performed to generate a factor co-occurrence matrix;

[0063] Specifically, this step aims to uncover static, latent correlation patterns from the underlying factors of the spatiotemporal knowledge dataset. A processing device performs statistical co-occurrence analysis on the separated spatiotemporal pattern factors and event factors. Specifically, it analyzes the co-occurrence of these factors in historical data, constructing a symmetric factor co-occurrence matrix. The element values ​​in this matrix, for example, represent the Pearson correlation coefficients of factors related to a spatial pattern and factors related to an event co-occurring within the same spatiotemporal unit. This factor co-occurrence matrix reveals the static correlation strength between different underlying features; for example, a specific climate pattern factor may be highly correlated with a certain equipment failure event factor.

[0064] Based on the weighted causal relationship, a graph topology is constructed and network centrality is calculated to generate causal centrality vectors for nodes, forming a causal centrality vector set.

[0065] Specifically, this step aims to quantify the importance of each node in the overall causal network topology based on weighted causal relationships. The processing unit calculates the causal centrality of each node in the graph. Causal centrality is a comprehensive metric, and its constituent causal centrality vector can include multiple dimensions: for example, out-degree centrality, in-degree centrality, and betweenness centrality. The causal centrality vectors of all nodes constitute the causal centrality vector set, which profoundly reflects the topological importance of each node in the dynamic causal transmission network.

[0066] By integrating the factor co-occurrence matrix and the causal centrality vector set, the information propagation weights in the risk propagation simulation are dynamically adjusted.

[0067] Specifically, this step is the core of achieving adaptive risk propagation model. During risk propagation simulation, the propagation strength of information from node u to node v is no longer fixed, but dynamically calculated using a function that integrates static association and dynamic topological importance. For example, the dynamic propagation weight from node u to node v... It can be represented as:

[0068] ,

[0069] in, The propagation weights are dynamically adjusted. The original causal weights obtained from the weighted causal relationship; and These are the out-degree and in-degree centrality of nodes u and v, respectively, obtained from the causal centrality vector set; This is the hyperbolic tangent function, used for nonlinear smoothing of the effect of centrality; It is the co-occurrence strength between the underlying factors most relevant to nodes u and v, obtained from the factor co-occurrence matrix; It is a natural exponential function, used to amplify the effect of co-occurrence intensity; and These are adjustable hyperparameters that control the influence of topological importance and factor co-occurrence. In this way, the propagation weights can be dynamically adjusted to accurately reflect static feature correlations and dynamic topological importance, enabling risk propagation simulations to more accurately model the concentrated outbreaks and rapid spread of risks in the real world.

[0070] Optionally, the calculation and generation of the risk quantification map includes:

[0071] Based on the causal relationship graph, risk source nodes are identified and risk states are initialized to form an initial risk vector set;

[0072] Specifically, this process aims to establish accurate, multi-dimensional initial boundary conditions for risk propagation simulation. A processing device first identifies risk source nodes in a causal graph. These nodes typically represent externally independent input events or endogenous, root cause events without upstream causes, represented in the graph topology as nodes with zero in-degree. After identification, the processing device initializes a risk state for each risk source node. This state is constructed as a multi-dimensional initial risk state vector, whose dimensions may include probability of occurrence, magnitude of impact, and time decay coefficient. These initial values ​​are set based on historical data statistics or external predictive data sources.

[0073] Based on the initial risk vector set, the risk propagation deduction is iteratively performed on the causal relationship graph until the propagation converges, generating a full node prognostic risk vector set.

[0074] Specifically, this step aims to simulate how initial risks propagate and evolve step-by-step within a causal network. The processing device performs iterative risk propagation calculations on the causal graph based on the initial risk vector set. In each iteration, the risk state vector of each node in the graph is updated according to the risk state vectors of all its upstream neighbors and the weights of the causal edges connecting them. A nonlinear propagation function is used to simulate the amplification or attenuation effect of risk during propagation. Notably, the weights of the connecting edges are dynamically adjusted during the process, making the deduction process highly context-adaptive. This iterative process continues until the risk state vectors of all nodes in the entire graph no longer change significantly, i.e., convergence is achieved. When propagation converges, each node obtains a stable vector representing its future risk status; the set of all these vectors constitutes the overall prognostic risk vector set for all nodes.

[0075] A risk quantification map is generated by weighted fusion based on the full-node prognostic risk vector set.

[0076] Specifically, this step aims to transform the multidimensional and complex risk vectors obtained in the previous step into a single, intuitive, and quantifiable risk indicator for decision-makers. The processing unit performs a weighted fusion operation on each vector in the full-node prognostic risk vector set. This process assigns different weights to different dimensions in the risk state vectors according to a preset business strategy, then performs a weighted summation to calculate a comprehensive risk index. Finally, the processing unit assigns this comprehensive risk index to the corresponding node in the graph, forming a visualized risk heatmap. This final causal relationship graph, where each node is assigned a quantified risk value, is the risk quantification graph, as shown below. Figure 3 As shown, this diagram, presented in the form of a stream, comprehensively illustrates how initial risk originates from the risk source node on the left, passes through the causal transmission event nodes in the middle, and finally converges and is quantified and allocated to the target risk node on the right. The width of the stream in the diagram is proportional to the risk magnitude, revealing the main propagation paths and key convergence points of systemic risk in the causal network.

[0077] Optionally, generating the extended risk scenario set includes:

[0078] Based on the risk quantification map and the causal relationship map, multiple consecutive diagnostic cycles are tracked and recorded to obtain the map activation frequency and intensity distribution.

[0079] Specifically, this process aims to go beyond static risk assessment by exploring potential risk scenarios caused by link vulnerabilities through deep learning of historical dynamics. A processing device first tracks the activation status of a risk quantification graph over consecutive, multiple diagnostic cycles. Within each diagnostic cycle, the processing device records edges in the causal relationship graph where the risk value exceeds a preset activation threshold, thus obtaining the activation frequency of each link. Simultaneously, it records the intensity of risk transmission during each activation. These historical records are compiled into a graph activation frequency and intensity distribution, which includes the frequency with which each causal link is activated.

[0080] Based on the activation frequency and intensity distribution, causal links of continuous stress and abnormal sensitivity are analyzed and identified, and link vulnerability indices are calculated.

[0081] Specifically, this step aims to identify vulnerable causal links most likely to cause cascading effects from historical dynamics. The processing unit calculates a link vulnerability index for each causal link based on the frequency and intensity distribution of activation patterns, combined with the link's importance in the network topology. This index is a comprehensive quantitative value representing the link's vulnerability, uncertainty, and potential downstream impact. An exemplary model for calculating this index can be expressed by the following formula:

[0082] ,

[0083] in, This is a vulnerability indicator for link l; and These are the normalized historical activation frequency and average activation intensity of the link, respectively. It is the standard deviation of the activation strength of the link, used to characterize its sensitivity or volatility; It is the out-degree centrality of the target node v to which the link is pointed, obtained from the causal centrality vector set, and is used to quantify the downstream topological influence of the link. and These are preset weighting coefficients. This formula combines the historical performance of a link with its topological importance in the network, thereby enabling more accurate identification of critical and vulnerable links.

[0084] Using the aforementioned link vulnerability index, the dynamic evolution process of the simulated causal link is weighted and guided to generate an extended risk scenario set.

[0085] Specifically, this step aims to leverage the vulnerable link information identified in the previous step to generate more challenging future risk scenarios. When the processing device performs this exploratory simulation of the dynamic evolution of causal links, its process is guided by a weighted index of link vulnerability. Specifically, when random disturbances or simulated link failures are introduced into the simulation, the probability of each link being selected as the disturbance target is proportional to the value of its link vulnerability index. This weighted guidance mechanism causes the simulation process to tend to extrapolate along the weakest links, thereby efficiently exploring those rare but potentially catastrophic "black swan" or "gray rhino" risks. Through multiple simulations with this weighted guidance, a diverse, high-risk, and highly valuable set of extended risk scenarios is ultimately generated.

[0086] Optionally, the generation of the counterfactual intervention strategy set includes:

[0087] Based on the extended risk scenario set and the full node prognostic risk vector set, the risk contribution distribution of different scenarios of the node is calculated to obtain the node risk contribution matrix;

[0088] Specifically, this step aims to accurately attribute and quantify the specific contribution of each potential risk source to the overall risk from a series of complex future scenarios. A processing unit analyzes the difference between each scenario in the expanded risk scenario set and the baseline risk state, calculating the risk contribution of each node in that scenario using a contribution allocation algorithm, such as Shapley value decomposition based on game theory. This process is repeated for all scenarios, ultimately generating a node risk contribution matrix, where rows represent different risk nodes, columns represent different projected scenarios, and the element values ​​in the matrix represent the risk contribution of a specific node in a specific scenario.

[0089] Based on the node risk contribution matrix, multi-objective sorting optimization is performed to select key risk nodes according to risk fluctuation amplitude and situation sensitivity, respectively.

[0090] Specifically, this step aims to move beyond a single risk value ranking and screen for nodes with the greatest intervention value from a more complex perspective. The processing device performs multi-objective ranking optimization on the node risk contribution matrix. This optimization process includes two core dimensions: First, risk volatility, which calculates the variance or standard deviation of each node's risk contribution across all scenarios to identify nodes that are highly unstable and potentially explosive. Second, scenario sensitivity, which identifies nodes whose contribution only spikes dramatically in a few extremely high-risk scenarios; these nodes are key vulnerabilities in responding to "black swan" events. By performing weighted ranking or Pareto front analysis on these two dimensions, a set of key risk nodes that excel in both dimensions is ultimately selected.

[0091] Based on the aforementioned key risk nodes, the physical attributes and functional constraints of the causal relationship graph are mapped to form a set of counterfactual intervention strategies.

[0092] Specifically, this step aims to transform abstract key risk nodes into concrete, actionable response strategies. The processing device maps each key risk node back to a causal relationship graph to analyze the real physical entities or events it represents and retrieves its physical attributes and functional constraints from the spatiotemporal knowledge dataset. For example, a key risk node might correspond to "material delivery from supplier A," with functional constraints including "minimum order quantity" and "average delivery cycle." Based on these constraints, the processing device constructs counterfactual intervention strategies by simulating adjustments to these constraints. For example, for the aforementioned node, specific strategies such as "increasing the minimum order quantity by 20%" or "activating alternative supplier B" can be generated. All these concrete, simulateable adjustment schemes generated to address key risks collectively constitute a set of counterfactual intervention strategies, such as... Figure 4 As shown, this figure uses a heatmap of the co-occurrence intensity among different spatiotemporal event factors as its core, and comprehensively displays the risk contribution relationship between these underlying factors and the key risk nodes on the right. The different styles of the connecting lines in the figure represent the magnitude, positive or negative, and statistical significance of the risk contribution, thus achieving a comprehensive visualization of potential patterns and key risk sources within a single view.

[0093] Optionally, the selection of key risk nodes based on risk fluctuation magnitude and situation sensitivity includes:

[0094] Based on the node risk contribution matrix, the risk fluctuation gradient and sensitivity index are calculated to obtain the risk fluctuation matrix and sensitivity index set.

[0095] Specifically, this step aims to transform complex risk contribution data into two clear, ranking-optimized core evaluation dimensions. A processing unit first calculates the risk volatility gradient based on the node risk contribution matrix. This gradient is obtained by calculating the variance or standard deviation of each row's risk contribution value under different scenarios, measuring the instability and potential explosiveness of the node's contribution. The results constitute the risk volatility matrix. Simultaneously, the processing unit calculates the node's sensitivity index. This index focuses on the node's performance in extremely high-risk scenarios, obtained by identifying the scenarios where each node's risk contribution ranks highest and calculating its average contribution value under these scenarios. This index reveals which nodes are vulnerable under specific stress scenarios, and the results constitute a sensitivity index set.

[0096] The risk volatility matrix and the sensitivity index set are respectively weighted and sorted, and the sorting results are combined and cross-fused to generate a comprehensive sorting list;

[0097] Specifically, this step aims to merge the evaluation results obtained from the two different dimensions in the previous step into a unified, final priority ranking. The processing device first performs weighted sorting on the risk volatility matrix and the sensitivity index set, generating two independent node priority lists. Then, it performs a set cross-fusion process on these two ranking results. This fusion process can employ various strategies; for example, taking the intersection of the top N nodes in both lists to filter out the most critical nodes that simultaneously possess high volatility and high sensitivity; or using a counting-based voting mechanism to combine the ranking positions of each node in both lists to generate a final comprehensive score. Through this fusion process, a comprehensive ranking list that fully reflects the criticality of the nodes is ultimately formed.

[0098] Based on the comprehensive sorting list, key risk nodes are selected in order of priority.

[0099] Specifically, this step involves making the final critical node selection decision based on the comprehensive ranking generated in the previous step. The processing device selects nodes according to the comprehensive ranking list, from highest to lowest priority. The selection process can be based on a preset quantity threshold or a score threshold. This threshold can also be dynamically adjusted based on currently available intervention resources or desired risk reduction targets. These selected nodes are those deemed to have the greatest impact on overall risk and require the highest priority for attention and intervention under multi-dimensional assessment. They collectively constitute the critical risk nodes and serve as the direct basis for subsequently developing specific intervention strategies.

[0100] Optionally, the output metering device delivery strategy data includes:

[0101] Based on the counterfactual intervention strategy set and the node risk contribution matrix, the strategy lifecycle cost and power distribution reliability are calculated to obtain the strategy utility vector set.

[0102] Specifically, this step aims to conduct a comprehensive and quantitative evaluation of the various intervention strategies generated in the previous step, designed to address key risks, in order to find the optimal solution. A processing device calculates the utility of each strategy in the counterfactual intervention strategy set based on two core dimensions. The first is the strategy lifecycle cost, which includes not only the direct logistical and human costs of strategy implementation but also long-term holding costs such as adjusting inventory levels. The second is distribution reliability, a function of a reduced global risk value calculated by re-performing risk propagation on the causal relationship graph after applying the intervention strategy. To balance the conflicting objectives of cost and reliability, the processing device employs Multi-Attribute Utility Theory (MAUT) to calculate the comprehensive utility score for each strategy. An exemplary utility calculation model can be represented by the following formula:

[0103] ,

[0104] in, The overall utility score of intervention strategy s; The improvement in power distribution reliability resulting from implementing this strategy; The lifecycle cost of this strategy; and These are decision weight coefficients representing the importance of reliability and cost, respectively. It is a natural exponential function. This is a risk aversion coefficient, and the index term is used to nonlinearly model the marginal benefit of reliability improvement. The cost, reliability, and final utility scores of all strategies are integrated to form a strategy utility vector set.

[0105] Based on the aforementioned strategy utility vector set, the utility is normalized and sorted to output the metering equipment delivery strategy.

[0106] Specifically, this step aims to make a final, unique decision from numerous candidate strategies. The processing device normalizes and sorts the strategy utility vector set generated in the previous step, selecting the strategy with the highest overall utility score as the optimal solution for this decision. Subsequently, the processing device parses and transforms this selected, conceptually optimal strategy into machine-readable, directly executable metering equipment delivery strategy data, which includes specific material allocation instructions, vehicle route planning data, inventory threshold adjustment parameters, and execution time nodes.

[0107] The weights of causal edges in the causal relationship graph are adjusted using the strategy utility vector set and the metering equipment delivery strategy data.

[0108] Specifically, this step is the final link in the entire learning loop of this invention, aiming to learn and evolve from this decision-making simulation. The processing device uses the evaluation results of all candidate policies contained in the policy utility vector set to update the weights of the causal relationship graph. Specifically, for causal links that have been proven to be effectively intervened in, their weights are increased or an attribute label indicating "effectively interveneable" is added to that link. In this way, the causal relationship graph not only records the causal relationships between events, but also gradually learns and accumulates meta-knowledge about "how to effectively deal with these causal relationships," thereby making more accurate risk projections and more intelligent decisions in subsequent diagnostic cycles.

[0109] To verify the feasibility and effectiveness of this invention, it was applied to the intelligent operation center of metering equipment in a large city power grid company to address cascading failures caused by complex factors. The company aims to use the method of this invention to achieve dynamic simulation and forward-looking optimization of metering equipment delivery decisions.

[0110] In this embodiment, historical maintenance work orders, geographic information, real-time warehousing, and external environment data of the company were first collected and integrated. The multi-source data was constructed into a three-dimensional sparse tensor index, and spatiotemporal pattern factors and event factors were separated through tensor decomposition, ultimately generating a spatiotemporal knowledge dataset.

[0111] Subsequently, by screening potential causal event pairs and conducting counterfactual experiments using a simulated control group dataset, a global causal relationship graph was successfully verified and constructed. In one simulation, an external risk source input of "a strong typhoon will make landfall in the next week" was received. Risk propagation was performed on the graph, and the link vulnerability index of "typhoon → large-scale power outage → concentrated repair requests" was identified as extremely high. Therefore, when generating the extended risk scenario set, the extreme case of link saturation was simulated in detail.

[0112] For this scenario set, a node risk contribution matrix was calculated, and by ranking the risk fluctuation amplitude and scenario sensitivity in multiple objectives, "the inventory of Type B electricity meters in coastal Warehouse C" was selected as the key risk node. Based on this, a set of counterfactual intervention strategies was generated, including "urgently transferring 500 Type B electricity meters from inland Warehouse E to Warehouse C".

[0113] Finally, a cost-benefit quantitative analysis was conducted on the strategy set, calculating the strategy utility vector set. The strategy with the highest utility was selected to form the final metering equipment distribution strategy data for execution. Simultaneously, the successful experience of the "cross-regional emergency allocation" strategy, verified as efficient in this simulation, was fed back into the causal relationship graph, completing the learning loop.

[0114] Table 1. Data on Causal Relationship Discovery and Link Vulnerability Analysis

[0115]

[0116] Table 2. Key Risk Node Screening and Counterfactual Intervention Strategy Data Table

[0117]

[0118] Table 3 Comparison of the effects before and after the implementation of the delivery strategy

[0119]

[0120] As can be seen from the data in Tables 1 to 3 above, the method of the present invention demonstrates significant advantages in the decision-making process for metering equipment distribution. Table 1 shows that the system can uncover risk links with high causal strength and high vulnerability. Table 2 shows that the system can accurately locate key risk nodes and generate targeted adjustment plans. The comparative data in Table 3 fully demonstrates the practical application effect of the present invention: through forward-looking distribution decisions, spare parts fulfillment rate and fault repair efficiency are significantly improved, overall costs are significantly reduced, and the system can achieve self-learning and evolution through feedback.

[0121] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or parameters that can be superimposed in the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. These are conventional technical methods and will not be elaborated further. The electrical connections between the various units described above do not necessarily represent direct or indirect connections; any indirect connection method is applicable to the embodiments of this invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of this invention.

[0122] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method of simulation analysis of metering equipment distribution decisions, characterized in that, The method comprises: Collect and fuse the geographic information, historical operation and maintenance work orders, real-time warehouse data and external environment data of the metering equipment to generate a spatio-temporal knowledge dataset; wherein the generation of the spatio-temporal knowledge dataset comprises: dividing the geographic information into grid units, dividing the historical operation and maintenance work orders into buckets according to fault types and time periods, mapping the real-time warehouse data and the external environment data into the same grid to construct a three-dimensional sparse tensor index; for the three-dimensional sparse tensor index, tensor decomposition is performed and missing data is completed to separate spatio-temporal pattern factors and event factors; the spatio-temporal pattern factors and the event factors are mapped into graph node features, and information is propagated between adjacent nodes to generate a spatio-temporal knowledge dataset; Based on the spatio-temporal knowledge dataset, the correlation between events is mined by causal inference to construct a causal relationship graph; Based on the causal relationship graph, multi-step risk propagation deduction is performed by using graph information propagation to calculate and generate a risk quantification graph; Based on the causal relationship graph and the risk quantification graph, the dynamic evolution of causal links is simulated to generate an extended risk scenario set; wherein the generation of the extended risk scenario set comprises: based on the risk quantification graph and the causal relationship graph, tracking and recording a plurality of continuous diagnosis cycles to obtain graph activation frequency and intensity distribution; based on the activation frequency and the intensity distribution, the causal links of continuous pressure and sensitivity anomalies are analyzed and identified, and a link vulnerability index is calculated; the link vulnerability index is used to guide the simulation of the dynamic evolution of the causal links, and an extended risk scenario set is generated; For the extended risk scenario set, the risk contribution degree of the node is calculated, and the key risk nodes are screened by sorting optimization to generate a counterfactual intervention strategy set; For the counterfactual intervention strategy set, cost-benefit quantification analysis is performed to output metering equipment distribution strategy data, and the quantification analysis result is fed back to the causal relationship graph.

2. The simulation analysis method of metering equipment dispensing decisions according to claim 1, characterized in that, The construction of the causal relationship graph comprises: Causal candidate relationship identification is performed on the nodes and edges in the spatio-temporal knowledge dataset to detect event sequence and spatio-temporal co-occurrence, and potential causal event pairs are screened; Based on the potential causal event pairs, causal verification and strengthening are performed, multi-scenario comparison and counterfactual experiment are used to verify the causal strength, and a weighted causal relationship is generated; According to the weighted causal relationship, topological arrangement and hierarchical layering are performed to form a causal relationship graph.

3. The simulation analysis method of metering equipment dispensing decisions according to claim 2, characterized in that, The generation of the weighted causal relationship comprises: Based on the potential causal event pairs and the spatio-temporal knowledge dataset, reason event co-occurrence is analyzed and extracted to obtain scenario accompanying factors; Using the scenario accompanying factors as a search condition, a simulated control group dataset is matched and constructed from the spatio-temporal knowledge dataset; Based on the simulated control group dataset, event occurrence overview is analyzed and causal strength is quantified to obtain a weighted causal relationship.

4. The simulation analysis method of a metering equipment dispensing decision according to claim 2, characterized by, The method further comprises: Based on the spatio-temporal pattern factors and the event factors, statistical co-occurrence analysis is performed to generate a factor co-occurrence matrix; Based on the weighted causal relationship, graph topology is constructed and network centrality is calculated to generate a causal centrality vector for the node, and a causal centrality vector set is formed; The factor co-occurrence matrix is fused with the causal centrality vector set to dynamically adjust the information propagation weight in the risk propagation deduction.

5. The simulation analysis method of metering equipment dispensing decisions according to claim 1, characterized in that, The calculation and generation of the risk quantification graph include: Based on the causal relationship graph, a risk source node is identified and a risk state is initialized to form an initial risk vector set; Based on the initial risk vector set, the risk propagation deduction is iterated on the causal relationship graph until the propagation converges, generating a full-node prognosis risk vector set; Based on the full-node prognosis risk vector set, a weighted fusion is performed to generate a risk quantification graph.

6. The simulation analysis method of a metering equipment dispensing decision according to claim 5, characterized in that, The generation of the counterfactual intervention strategy set includes: Based on the extended risk scenario set and the full-node prognosis risk vector set, the risk contribution degree distribution of nodes in different scenarios is calculated to obtain a node risk contribution degree matrix; Based on the node risk contribution degree matrix, multi-objective sorting optimization is performed to filter out key risk nodes according to risk fluctuation amplitude and scenario sensitivity respectively; Based on the key risk nodes, the counterfactual intervention strategy set is mapped to the causal relationship graph to analyze physical properties and functional constraints.

7. The simulation analysis method of a metering equipment dispensing decision according to claim 6, characterized in that, The filtering out of key risk nodes according to risk fluctuation amplitude and scenario sensitivity respectively includes: Based on the node risk contribution degree matrix, risk fluctuation gradient and sensitivity indicators are calculated to obtain a risk fluctuation matrix and a sensitivity indicator set; The risk fluctuation matrix and the sensitivity indicator set are respectively applied to weighted sorting, and the sorting results are cross-fused to generate a comprehensive sorting list; Based on the comprehensive sorting list, key risk nodes are selected according to priority order.

8. The simulation analysis method of a metering device dispensing decision according to claim 6, characterized in that, The output of the metering equipment distribution strategy data includes: Based on the counterfactual intervention strategy set and the node risk contribution degree matrix, the strategy life cycle cost and the power distribution reliability are calculated to obtain a strategy utility vector set; Based on the strategy utility vector set, utility normalization processing and sorting are performed to output a metering equipment distribution strategy; The strategy utility vector set and the metering equipment distribution strategy data are used to adjust the causal edge weight in the causal relationship graph.

Images