Intelligent gas risk monitoring method and system based on multi-source data

Abstract

The invention relates to the technical field of gas pipe network safety monitoring, and discloses a gas risk intelligent monitoring method and system based on multi-source data, and the method comprises the steps: obtaining static attribute data, dynamic sensing data and future event data of a gas pipe network, and constructing a dynamic attribute graph; inputting the dynamic attribute graph into a time-space diagram neural network model, and predicting the physical failure probability of each node; calculating the time-varying failure influence of each node based on the future event data and the cost data; and calculating the value-at-risk of the node by combining the physical failure probability and the time-varying failure influence, and generating a maintenance decision through an optimization model based on the distribution of the value-at-risk. The system comprises a data acquisition module, a failure rate prediction module, a time-varying failure processing module, a value-at-risk calculation module and a maintenance decision generation module. According to the method, the time-varying influence function and the investment portfolio model are constructed, the risk is quantified prospectively, the maintenance decision is optimized, the problem of evaluation lag is solved, and the operation and maintenance efficiency and safety of the pipe network are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of gas pipe network safety monitoring, in particular to a gas risk intelligent monitoring method and system based on multi-source data. BACKGROUND

[0002] As a core component of urban lifeline engineering, the operation safety of gas pipe network plays a vital role in ensuring social stability and economic development. At present, the safety monitoring of gas pipe network mainly relies on the data acquisition and monitoring control system (SCADA), which monitors the state by obtaining real-time operation parameters such as pressure and flow of key nodes in the pipe network. At the same time, risk assessment is usually based on static attributes such as pipe material and pipe age, and historical failure data, and uses traditional statistical models or expert scoring methods to identify high-risk pipe sections.

[0003] However, the existing technology relies on historical and current static data for risk assessment, and fails to conduct forward-looking quantitative analysis on the dynamic time-varying characteristics of risk, especially the additional risk that may be caused by predictable future large-scale activities, urban planning or extreme weather events, resulting in a certain lag in risk assessment results.

[0004] In addition, when making maintenance decisions, it is usually based on simple risk level ranking or relies on the experience of operation and maintenance personnel, and lacks a systematic optimization framework to guide how to allocate maintenance resources optimally within the entire pipe network under the condition of limited budget and resources to maximize overall risk reduction.

[0005] Therefore, the present application provides a gas risk intelligent monitoring method and system based on multi-source data to solve the problems of the prior art. SUMMARY

[0006] In view of the deficiencies of the prior art, the present application provides a gas risk intelligent monitoring method and system based on multi-source data, which solves the problem that the influence of future events cannot be quantitatively analyzed in advance when conducting risk assessment of gas pipe network, and lacks a systematic optimization when making maintenance decisions.

[0007] To solve the above technical problems, the present application provides a gas risk intelligent monitoring method and system based on multi-source data.

[0008] The first aspect of the present application provides a gas risk intelligent monitoring method based on multi-source data, which comprises the following steps:

[0009] S1, obtaining static attribute data, dynamic sensing data, future event data and cost data of each node in the gas pipe network, and constructing a dynamic attribute graph based on the static attribute data, the dynamic sensing data and the future event data;

[0010] S2, input the dynamic attribute graph into a preset spatio-temporal graph neural network model, and predict the physical failure probability of each node in the gas pipe network through the spatio-temporal graph neural network model;

[0011] S3, weighting and summing the cost data of each node in the gas pipe network to obtain a basic influence value of each node, calculating a dynamic influence weight of each node in the gas pipe network based on the future event data and a dynamic influence weight calculation formula, and calculating and obtaining a time-varying failure influence of each node in the gas pipe network based on a time-varying failure influence calculation formula, the basic influence value and the dynamic influence weight, wherein the basic influence value represents the influence value of a node failure in the gas pipe network on the entire gas pipe network;

[0012] S4, combining the physical failure probability and the time-varying failure influence, calculating the risk value of each node in the gas pipe network, and forming a risk value distribution based on the risk value;

[0013] S5, generating a maintenance decision for the gas pipe network based on the risk value distribution.

[0014] Preferably, in step S1, the static attribute data includes pipe material, material code, pipe wall thickness, anticorrosion layer type, pipe diameter, laying year and geographic coordinates; the dynamic sensing data includes pressure, flow and temperature, and gas concentration; the future event data includes city planning information, large public activity information and weather warning information; and the cost data includes direct economic cost, indirect economic cost, social influence cost, environmental cost and safety cost.

[0015] The dynamic attribute graph is G sum =(V sum ,E sum ,X sum ), wherein V sum is a set of each node in the gas pipe network, E sum is a set of edges connecting each node composed of gas pipes, representing the connection relationship between each node, X sum is a node feature matrix for describing the static attributes, dynamic sensing data and future time characteristics of each node in the gas pipe network, which is extracted and obtained according to the static attribute data, the dynamic sensing data and the future event data, and sum is the number of nodes.

[0016] Preferably, in step S2, the step of inputting the dynamic attribute graph into a preset spatio-temporal graph neural network model and predicting the physical failure probability of each node in the gas pipe network through the spatio-temporal graph neural network model includes:

[0017] acquire static attribute data, dynamic sensing data and future event data of each node at time t based on the dynamic attribute graph and sensing devices of each node in the gas pipe network;

[0018] set a reference risk coefficient of each node according to dynamic sensing data and future event data of each node at time t, and acquire the physical failure probability based on the spatio-temporal graph neural network model, dynamic sensing data, future event data and reference risk coefficient of each node at time t;

[0019] The spatio-temporal graph neural network model is as follows:

[0020] In the formula, P fail (I,t) is the physical failure probability of node i at time t, λ0(t) is the reference risk coefficient of each node, which is set according to dynamic sensing data and future event data of each node, W l is the value of dynamic sensing data of node i at time t, b l is the value of future event data of node i at time t, is the value of static attribute data of node i at time t.

[0021] Preferably, the step S3 further comprises:

[0022] weighting and summing cost data of each node in the gas pipe network to obtain a basic influence value of each node, calculating a dynamic influence weight of each node in the gas pipe network based on the future event data and a dynamic influence weight calculation formula, and calculating and obtaining a time-varying failure influence of each node in the gas pipe network based on a time-varying failure influence calculation formula, the basic influence value and the dynamic influence weight, wherein the basic influence value represents an influence value of failure of a node in the gas pipe network on the entire gas pipe network;

[0023] S3-1, weighting and summing direct economic cost, indirect economic cost, social influence cost, environmental cost and safety cost in the cost data of each node based on a basic influence value calculation formula to obtain a basic influence value of each node;

[0024] S3-2, calculating a dynamic influence weight of each node based on the future event data and a dynamic influence weight calculation formula, wherein the dynamic influence weight calculation formula is as follows:

[0025]

[0026] In the formula, w(i,t) is the dynamic influence weight of each node, E tis a future event set known at time t, e is a future event in the future event set, I(·) is an indicator function, L e is an influence range of future event e e (t) is an influence intensity function of future event e at time t

[0027] S3-3, based on the time-varying failure influence calculation formula, the basic influence value and the dynamic influence weight, calculating and obtaining the time-varying failure influence of each node, the time-varying failure influence calculation formula is:

[0028] C impact (i,t)=w(i,t)×C base (i);

[0029] In the formula, C impact (i,t) is the time-varying failure influence of node i at time t, w(i,t) is the dynamic influence weight of node i at time t, C base (i) is the basic influence value of node i.

[0030] Preferably, in step S3-1, the basic influence value calculation formula is:

[0031] C base (i)=αC dir (i)+βC ind (i)+γC soc (i)+δC env (i);

[0032] In the formula, C base (i) is the basic influence value of node i, C dir (i), C ind (i), C soc (i) and C env (i) are respectively the quantified values of the direct economic cost, indirect economic cost, social influence cost, environmental cost and safety cost corresponding to the failure of node i; α, β, γ, δ are respectively the pre-set normalized weight coefficients corresponding to the direct economic cost, indirect economic cost, social influence cost, environmental cost and safety cost, used to balance the weights of different cost types.

[0033] Preferably, in step S4, the step of combining the physical failure probability with the time-varying failure influence to calculate the risk value of each node in the gas pipe network, and forming a risk value distribution based on the risk value further comprises:

[0034] Based on the risk value calculation formula, using the physical failure probability and the time-varying failure impact as inputs, the risk value of each node in the gas pipeline network is calculated and obtained. The risk value distribution dynamically changes based on the risk values ​​of each node at different times, wherein the risk value distribution varies with time and the risk value of each node. The risk value calculation formula is as follows:

[0035] R V (i,t)=P fail (i,t)×C impact (i,t);

[0036] In the formula, R V (i,t) represents the risk value of node i at time t, P fail (i,t) represents the physical failure probability of node i at time t, C impact (i,t) represents the time-varying failure effect of node i at time t.

[0037] Preferably, in step S5, the step of generating maintenance decisions for the gas pipeline network based on the risk value distribution includes:

[0038] Based on the risk value distribution and expected risk calculation formula, the expected risk value of each node is calculated in real time. All nodes are sorted in descending order of expected risk value. The expected risk value of each node is compared with the expected risk threshold. The risk status of each node is determined based on the comparison result. The maintenance decision is determined based on the risk level of each node.

[0039] The expected risk threshold is pre-set based on the risk influencing factors and cost data of each node, and the expected risk calculation formula is R. i =P fail (i,t)×R V (i,t), where R is the value of t. i Let R be the expected risk value of node i at time t. V (i,t) represents the risk value of node i at time t, P fail (i,t) represents the physical failure probability of node i at time t;

[0040] If the expected risk value R i If the expected risk value R is greater than or equal to the expected risk threshold, then the expected risk value R is... i The corresponding node i at time t is marked as a high-risk node;

[0041] If the expected risk value R i If the value is less than the expected risk threshold, then the expected risk value R is... i The risk status of the corresponding node i at time t is marked as a low-risk node.

[0042] Preferably, in step S5, the step of generating the maintenance decision for the gas pipeline network based on the risk value distribution comprises:

[0043] taking the preset maintenance budget as an investment principal, and taking all possible maintenance actions as candidate financial assets, wherein each candidate financial asset has a corresponding maintenance cost and an expected risk value reduction amount;

[0044] applying a modern portfolio theory model to calculate a plurality of optimal maintenance portfolio schemes under the constraint of the maintenance budget and aiming to maximize the expected risk value reduction amount;

[0045] taking the optimal maintenance portfolio scheme as the maintenance decision.

[0046] Preferably, in step S5, the step of generating the maintenance decision for the gas pipeline network based on the risk value distribution further comprises:

[0047] establishing a reinforcement learning model, wherein the action space of the reinforcement learning model is to select from the optimal maintenance portfolio scheme, the state space of the reinforcement learning model is the risk value distribution, and the reward function of the reinforcement learning model is determined based on the risk value reduction amount and the maintenance cost;

[0048] learning a long-term maintenance strategy through the reinforcement learning model, and selecting a final maintenance decision from the optimal maintenance portfolio scheme based on the long-term maintenance strategy.

[0049] The second aspect of the present application provides a gas risk intelligent monitoring system based on multi-source data, comprising a data acquisition module, a failure rate prediction module, a time-varying failure processing module, a risk value calculation module and a maintenance decision generation module;

[0050] wherein,

[0051] the data acquisition module is configured to acquire static attribute data, dynamic sensing data, future event data and cost data of each node in the gas pipeline network, and construct a dynamic attribute graph based on the static attribute data, the dynamic sensing data and the future event data;

[0052] the failure rate prediction module is configured to input the dynamic attribute graph into a preset spatio-temporal graph neural network model, and predict the physical failure probability of each node in the gas pipeline network through the spatio-temporal graph neural network model;

[0053] The time-varying failure processing module is used to perform weighted summation of cost data of each node in the gas pipeline network to obtain the basic impact value of each node, calculate the dynamic impact weight of each node in the gas pipeline network based on the future event data and the dynamic impact weight calculation formula, and calculate and obtain the time-varying failure impact of each node in the gas pipeline network based on the time-varying failure impact calculation formula, the basic impact value and the dynamic impact weight. The basic impact value represents the impact value of the failure of one node in the gas pipeline network on the entire gas pipeline network.

[0054] The risk value calculation module is used to combine the physical failure probability and the time-varying failure impact to calculate the risk value of each node in the gas pipeline network, and form a risk value distribution based on the risk value.

[0055] The maintenance decision generation module is used to generate maintenance decisions for the gas pipeline network based on the risk value distribution.

[0056] This invention provides a method and system for intelligent monitoring of gas risk based on multi-source data. It has the following beneficial effects:

[0057] 1. This invention, by introducing future event data and constructing time-varying failure impacts, can proactively quantify the comprehensive risks of pipeline network nodes at specific future points in time. This method ensures that risk assessment results no longer rely solely on the current physical state of the pipeline network, but incorporate the potential impacts of future external events, thereby improving the timeliness and comprehensiveness of risk assessment and enabling the identification of potential risks with high impact at specific future points in time.

[0058] 2. This invention optimizes maintenance decisions by introducing a modern portfolio theory model, mapping maintenance budgets and actions to investment capital and financial assets, respectively. Under budget constraints, it can solve for the optimal maintenance portfolio scheme with the goal of maximizing risk reduction, providing a quantitative and systematic optimization method for gas pipeline network resource allocation. Compared to the traditional sequential disposal method, this improves the utilization efficiency of maintenance resources.

[0059] 3. This invention achieves dynamic updates to the pipeline network model by setting a closed-loop self-correction step, injecting executed maintenance decisions as action events back into the dynamic attribute graph. This mechanism enables subsequent risk prediction to take into account the state changes caused by executed maintenance interventions, giving the system adaptive capabilities and allowing for continuous optimization of the accuracy of the prediction model and decision-making strategies. Attached Figure Description

[0060] Figure 1 This is a flowchart of the method of the present invention;

[0061] Figure 2 This is a system architecture diagram of the present invention.

[0062] The module includes: 10. Data acquisition module; 20. Failure rate prediction module; 30. Time-varying failure handling module; 40. Risk value calculation module; 50. Maintenance decision generation module; and 60. Closed-loop correction module. Detailed Implementation

[0063] The technical solutions in 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, and 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.

[0064] See attached document Figure 1 -Appendix Figure 2 , attached Figure 1 This is a flowchart of the method according to the present invention; Figure 2 This is a system architecture diagram of the present invention.

[0065] This invention provides a gas risk intelligent monitoring system based on multi-source data, which can be deployed on one or more servers. The server includes a processor, a memory connected to the processor, and a network interface. The network interface is used to communicate with external data sources, including but not limited to SCADA systems of gas pipeline networks, IoT sensing devices, meteorological data platforms, and urban management data centers. The memory stores a computer program, which, when executed by the processor, implements the method provided by this invention.

[0066] In one embodiment, the system can be functionally divided into: a data acquisition module 10, a failure rate prediction module 20, a time-varying failure processing module 30, a risk value calculation module 40, a maintenance decision generation module 50, and a closed-loop correction module 60.

[0067] The present invention provides a method for intelligent monitoring of gas risk based on multi-source data, which may include the following steps:

[0068] S1. Obtain static attribute data, dynamic sensing data, and future event data of each node in the gas pipeline network, and construct a dynamic attribute graph based on the static attribute data, dynamic sensing data, and future event data.

[0069] S2. Input the dynamic attribute map into the preset spatiotemporal graph neural network model, and predict the physical failure probability of each node in the gas pipeline network through the spatiotemporal graph neural network model.

[0070] S3. Based on future event data and preset cost data, calculate the time-varying failure impact of each node in the gas pipeline network.

[0071] S4. Combining the physical failure probability and the impact of time-varying failure, calculate the risk value of each node in the gas pipeline network, and form a risk value distribution based on the risk value.

[0072] S5. Based on the risk value distribution, generate maintenance decisions for the gas pipeline network.

[0073] See attached document Figure 1 -Appendix Figure 2 , attached Figure 1 This is a flowchart of the method according to the present invention; Figure 2 This is a system architecture diagram of the present invention. The following is a description of the appendix. Figure 1 With appendix Figure 2 The specific steps in the method shown will be explained in detail.

[0074] Regarding step S1, constructing a dynamic property graph.

[0075] The main entity responsible for this step is the data acquisition module 10. The goal of this step is to integrate multi-source heterogeneous information to build a unified, dynamic data structure for subsequent risk prediction and value calculation.

[0076] First, the data acquisition module 10 obtains raw data from multiple preset external data sources through the system's network interface. These data sources are divided into three categories:

[0077] The first category is static attribute data, which is mainly obtained from the gas company's asset management database or geographic information system (GIS). Static attribute data includes the inherent attributes of each physical entity in the pipeline network, specifically the material code, nominal pipe diameter, pipe wall thickness, anti-corrosion layer type, laying year, and precise geographic coordinates of the pipe section.

[0078] The second category is dynamic sensing data, which is mainly acquired in real time from the gas pipeline network monitoring and data acquisition (SCADA) system and IoT sensor terminals deployed in key locations. Dynamic sensing data includes pressure, flow rate, and temperature at various monitoring points in the pipeline network at specific timestamps, as well as readings from gas concentration sensors installed around the pipeline or in valve wells.

[0079] The third category is future event data, obtained from urban management data centers, public affairs information release platforms, or third-party meteorological service interfaces. Future event data can be structured (such as calendar entries containing time, location, and event type) or unstructured (such as notices of urban planning or security notices for large-scale sporting events), covering urban road planning, arrangements for large-scale public events, or early warning information for extreme weather (such as rainstorms and geological disasters) in the future.

[0080] After acquiring the raw data, the data acquisition module 10 performs preprocessing operations on the data. These preprocessing operations include:

[0081] Data cleaning is used to identify and correct outliers and fill in missing values ​​in the original data.

[0082] Timestamp alignment is used to ensure that dynamic data obtained from different data sources are accurately synchronized over time.

[0083] Data normalization, through specific mathematical transformations (such as min-max normalization), scales characteristic data with different physical dimensions and numerical ranges into a unified numerical interval, such as [0, 1], to facilitate subsequent model processing.

[0084] After preprocessing, the system performs a graph construction operation, abstracting the entire gas pipeline network system into a dynamic attribute graph G that changes over time. t In this graph, each independent physical entity, such as a pipe segment, a valve, a pressure regulating station, or a monitoring point, is defined as a node v, and all nodes constitute the node set V. The physical connections or fluid connectivity between nodes are defined as edges e, and all edges constitute the edge set E. For each node v, the system concatenates its preprocessed static attribute data, dynamic sensing data, and parsed future event data to form a multidimensional feature vector X. t (v). The eigenvectors of all nodes in the network at time t together constitute the dynamic attribute matrix X at that time. t Thus, a dynamic attribute graph G is constructed that structurally reflects the physical topology and integrates multi-source dynamic information in terms of attributes. t =(V,E,X) t This graph will serve as the basic data structure for all subsequent computational steps.

[0085] Regarding step S2, predict the probability of physical failure.

[0086] The main component performing this step is the failure rate prediction module 20. The goal of this step is to use a deep learning model, based on the constructed dynamic attribute graph, to predict the probability of physical failure of each node in the gas pipeline network at a future point in time.

[0087] Failure rate prediction module 20 receives the dynamic attribute graph sequence {G} generated in step S1, which has a preset historical time length T. t-T+1 ,...,G t The sequence serves as input. It comprehensively records the evolution of the network's topology and node attributes over a past period.

[0088] The input sequence is fed into a pre-trained spatiotemporal graph neural network model (STGNN). This model is designed to simultaneously capture and process the spatial and temporal dependencies present in gas pipeline network data. In one specific embodiment, the STGNN model consists of multiple stacked spatiotemporal convolutional blocks. Each spatiotemporal convolutional block contains a spatial convolutional module and a temporal convolutional module.

[0089] The spatial convolution module is used to capture the spatial dependencies brought about by the pipeline network topology. This module can be specifically implemented as a graph convolutional network (GCN) or a graph attention network (GAT). At each time step, it generates a node representation that can incorporate local topological information by aggregating the feature information of each node and its first-order or multi-order neighbor nodes.

[0090] The temporal convolution module is used to learn the evolution of each node's attributes over time. This module can be specifically implemented as a gated recurrent unit (GRU) or a long short-term memory network (LSTM), which receives the node representation sequence processed by the spatial convolution module and captures the dynamic change pattern of the sequence in the time dimension.

[0091] The model training process is an offline supervised learning process. The training dataset consists of a large number of historical dynamic attribute graph sequences and corresponding real failure event records. The real failure event records are processed into binary labels (e.g., 1 represents failure, 0 represents no failure). During training, the system uses a binary cross-entropy loss function to measure the difference between the model's predicted probabilities and the true labels, and continuously adjusts the model's network parameters through backpropagation and an optimizer (such as the Adam optimizer) until the loss function converges.

[0092] After training, the model is deployed to perform forward inference. The failure rate prediction module 20 feeds the latest dynamic attribute map sequence into the trained model, which then calculates and outputs a prediction vector. Each element of this vector is defined as the physical failure probability P of the corresponding node i at a future preset time point t+Δt. fail (i, t + Δt). This probability value is the basic input for subsequent calculations of Value at Risk.

[0093] Regarding step S3, calculate the impact of time-varying failure.

[0094] The time-varying failure handling module 30 is responsible for executing this step. The goal of this step is to calculate the comprehensive impact of a physical failure on each node in the gas pipeline network. This impact is dynamic and changes over time, reflecting the potential effects of future external events. The calculation process is broken down into two parts: the calculation of the basic impact value and the calculation of the dynamic impact weight.

[0095] The first part involves the calculation of the basic impact value. The basic impact value represents a relatively fixed quantified loss caused by a node failing under normal conditions. The time-varying failure handling module 30 calculates the basic impact value C for each node i based on a preset cost database. base (i). Cost data specifically includes four categories: direct economic costs C dir It quantifies the material and labor costs required for emergency repairs; indirect economic costs C ind It quantifies the production or operational losses caused by gas outages to downstream industrial or commercial users due to pipeline failure; social impact cost C soc It quantifies the impact on the normal operation of critical public service institutions such as hospitals and schools; environmental and safety costs C env It quantifies the pollution of the surrounding environment and the potential threat to public safety posed by gas leaks.

[0096] The formula for calculating the basic influence value is:

[0097] C base (i)=αC dir (i)+βC ind (i)+γC soc (i)+δC env (i);

[0098] In the formula, C base (i) represents the base influence value of node i; C dir (i), C ind (i), C soc (i) and C env (i) represents the quantified values ​​of the four types of costs corresponding to the failure of node i; α, β, γ, and δ are pre-set normalized weighting coefficients used to balance different cost types.

[0099] The second part involves the calculation of the dynamic impact weight. The dynamic impact weight is a time-varying, dimensionless coefficient used to reflect the amplification effect of known future events on the failure consequences of a specific node. The time-varying failure processing module 30 incorporates a future event parsing engine, which is responsible for processing the future event data acquired in step S1. This engine parses unstructured or structured event information (such as text announcements and schedules) into unified structured event parameters, primarily including the geographical scope L of the event's impact. e and the time-varying influence intensity function λ e (t).

[0100] Based on the parsed event parameters, the system calculates the dynamic influence weight w(i,t) of each node i at time t. The formula for calculating the dynamic influence weight is as follows:

[0101]

[0102] In the formula, E t Let L be the set of future events known at time t; let e be a future event in the set of future events; let I(·) be the indicator function; L e Let L be the range of influence of future event e. When the geographical location of node i is within the range of influence of future event e... e When the value is within a certain range, its value is 1; otherwise, it is 0. e (t) is the influence intensity function of future event e at time t, and its value reflects the severity of the event's influence.

[0103] Finally, the time-varying failure processing module 30 multiplies the two calculation results to obtain the final time-varying failure impact. The calculation formula for the time-varying failure impact is:

[0104] C impact (i,t)=w(i,t)×C base (i);

[0105] In the formula, C impact (i,t) represents the time-varying failure effect of node i at time t; C base (i) represents the base influence value of node i. The output C of this step is... impact (i,t) is a value that can accurately quantify the consequences of failure of each node at different time points, and is used as the input for the next step.

[0106] Regarding step S4, calculate the value at risk and form a value at risk distribution.

[0107] The entity responsible for executing this step is the Risk Value Calculation Module 40. The goal of this step is to combine the failure probability and failure consequences calculated in the previous steps to generate a quantitative indicator that comprehensively reflects the overall risk of all parts of the gas pipeline network.

[0108] The Risk Value Calculation Module 40 receives two core input data items: the first is the physical failure probability P for each node i at a future time t, output by the Failure Rate Prediction Module 20 in step S2. fail (i,t); the second term is the time-varying failure impact C output by the time-varying failure processing module 30 in step S3, which is the time-varying failure impact C on the same node i at the same time t. impact (i,t).

[0109] After acquiring the input data, the Value at Risk (VaR) calculation module 40 performs a VaR calculation operation on each node i. This operation is achieved by multiplying the node's physical failure probability by its time-varying failure impact. The VaR calculation formula is:

[0110] R V(i,t)=P fail (i,t)×C impact (i,t);

[0111] In the formula, R V (i,t) represents the risk value of node i at time t; P fail (i,t) represents the physical failure probability of node i at time t; C impact (i,t) represents the time-varying failure effect of node i at time t.

[0112] The result of the formula is R. V (i,t) is a comprehensive indicator that uniformly quantifies the probability of failure of a specific node at a specific point in time and the comprehensive consequences caused by its failure.

[0113] After calculating the risk value for all nodes in the pipeline network, the collection of risk values ​​for all nodes together constitutes a dynamically changing risk value distribution over time. This distribution is formally a data structure that maps risk values ​​to each node in the pipeline network, representing the comprehensive risk level of each location in the network at the current moment and within a certain future period. This risk value distribution serves as the direct basis for subsequent maintenance decision optimization and is passed to step S5 as the final output.

[0114] Regarding step S5, generating maintenance decisions

[0115] The main entity executing this step is the maintenance decision generation module 50. The goal of this step is to calculate and generate maintenance decisions for the gas pipeline network based on the risk value distribution generated in step S4, using a systematic optimization method. In a specific embodiment, this step is implemented through a two-stage decision framework.

[0116] The first stage involves applying Modern Portfolio Theory (MPT) to generate a set of short-term optimal maintenance solutions. In this stage, the system constructs the pipeline maintenance problem as a financial portfolio optimization problem. Specifically, the preset total maintenance budget B is set as the investment principal; all feasible and specific maintenance actions (such as inspecting, repairing, or replacing specific pipeline sections) are defined as a set of alternative financial assets, where each maintenance action a... j Each corresponds to a specific maintenance cost c(a) j and the expected reduction in risk value ΔR after implementation. V (a j ).

[0117] The maintenance decision generation module 50 solves a 0-1 integer programming problem with the goal of maximizing the total expected risk value reduction of all selected maintenance actions, while ensuring that the total cost of all selected maintenance actions does not exceed the budget B. By solving this optimization problem, one or more optimal maintenance portfolio schemes can be obtained under different budget levels, each scheme being a specific set of maintenance actions.

[0118] The second stage involves applying reinforcement learning (RL) to optimize the long-term maintenance strategy. To ensure that the decision is not only optimal in the current moment but also has long-term benefits, the system establishes a reinforcement learning model. This model is defined as a Markov decision process (MDP), whose specific elements include:

[0119] State-Space S: Defined as the risk value distribution generated in step S4 that reflects the overall health status of the pipeline network.

[0120] Action Space A: Defined as the selection from the set of optimal maintenance portfolio solutions generated in the first stage. Each action of the reinforcement learning agent is the selection of a specific maintenance solution to execute.

[0121] Reward Function R: This function is determined based on the total risk value reduction resulting from the selected maintenance plan and the total cost of the plan. The reward function is designed to encourage agents to choose plans that offer high risk reduction and cost-effectiveness.

[0122] The reinforcement learning agent is trained in a simulation environment centered on a dynamic attribute graph. During training, the agent selects an action (a maintenance plan) based on its current state (risk-value distribution). After the environment executes the action, it updates the dynamic attribute graph and calculates the new state and corresponding reward, which is then fed back to the agent. Through a continuous cycle of "state-action-reward-new state," the agent learns and converges to an optimal long-term maintenance strategy π. * .

[0123] After training, the optimal policy π * It is used for actual decision-making. The decision generation module 50 maintains the latest risk value distribution as input to the trained policy π. * The strategy directly outputs the optimal portfolio maintenance plan that should be executed in this state. This plan is used as the final maintenance decision and can be passed to step S6 to perform closed-loop correction.

[0124] Regarding step S6, perform closed-loop self-calibration.

[0125] This step is a preferred embodiment, and its execution entity is the closed-loop correction module 60. The goal of this step is to feed back maintenance decision information that has been implemented in the real world to the system and update the system's basic data model to form a closed-loop workflow that can adaptively adjust according to intervention measures.

[0126] Once a maintenance decision generated in step S5 is actually executed, the relevant execution confirmation information is fed back to the system. The closed-loop correction module 60 first receives this feedback information and performs parameterization processing to generate a structured "action event". For example, if the executed decision is "the pipe section numbered SG-113 was replaced on June 22, 2025, with the new material being PE100", then the generated action event will contain the following fields: operation type (replacement), action node ID (SG-113), execution timestamp (2025-06-22), and updated attribute information (material: PE100, laying year: 2025).

[0127] Subsequently, the system uses this structured "action event" to update the dynamic attribute graph database built and maintained in step S1. The closed-loop correction module 60 queries and locates the corresponding node in the dynamic attribute graph based on the action node ID in the action event.

[0128] After locating the node, the system modifies the corresponding dimension values ​​in the node's feature vector. Specifically, the system updates the feature value of "laying year" in the node's feature vector to the current year (2025), updates the feature value of "pipe material" to the code corresponding to the new material PE100, and adds a new maintenance record in a feature dimension used to record maintenance history.

[0129] By performing this step, the system's fundamental data structure—the dynamic attribute graph—can accurately reflect the changes in the physical world's state caused by maintenance intervention. In subsequent calculation cycles, all predictions (step S2) and decisions (step S5) made by the system will be based on this updated graph, which better reflects the current true state of the pipeline network, thereby achieving closed-loop operation and adaptive adjustment of the entire monitoring and decision-making method.

[0130] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent monitoring of gas risk based on multi-source data, characterized in that, The method includes the following steps: S1. Obtain static attribute data, dynamic sensing data, future event data, and cost data of each node in the gas pipeline network, and construct a dynamic attribute map based on the static attribute data, the dynamic sensing data, and the future event data; S2. Input the dynamic attribute graph into a preset spatiotemporal graph neural network model, and predict the physical failure probability of each node in the gas pipeline network through the spatiotemporal graph neural network model. S3. The cost data of each node in the gas pipeline network are weighted and summed to obtain the basic impact value of each node. The dynamic impact weight of each node in the gas pipeline network is calculated based on the future event data and the dynamic impact weight calculation formula. The time-varying failure impact of each node in the gas pipeline network is calculated and obtained based on the time-varying failure impact calculation formula, the basic impact value and the dynamic impact weight. The basic impact value represents the impact value of the failure of one node in the gas pipeline network on the entire gas pipeline network. S4. Combining the physical failure probability with the time-varying failure impact, calculate the risk value of each node in the gas pipeline network, and form a risk value distribution based on the risk value. S5. Based on the risk value distribution, generate maintenance decisions for the gas pipeline network.

2. The intelligent monitoring method for gas risk based on multi-source data according to claim 1, characterized in that, In step S1, the static attribute data includes pipe material, material code, pipe wall thickness, anti-corrosion layer type, pipe diameter, laying year, and geographical coordinates; the dynamic sensing data includes pressure, flow rate, temperature, and gas concentration; the future event data includes urban planning information, large-scale public event information, and weather warning information; and the cost data includes direct economic costs, indirect economic costs, social impact costs, environmental costs, and safety costs. The dynamic attribute graph is G. sum =(V sum E sum X sum ), where V sum E is the set of all nodes in the gas pipeline network. sum Let X be the set of edges connecting the nodes formed by gas pipelines, representing the connection relationships between the nodes. sum The node feature matrix describes the static attributes, dynamic sensing data, and future time features of each node in the gas pipeline network. It is extracted from the static attribute data, the dynamic sensing data, and the future event data, and sum is the number of nodes.

3. The intelligent monitoring method for gas risk based on multi-source data according to claim 1, characterized in that, In step S2, the step of inputting the dynamic attribute graph into a preset spatiotemporal graph neural network model and predicting the physical failure probability of each node in the gas pipeline network through the spatiotemporal graph neural network model includes: Based on the dynamic attribute map and the sensing devices of each node in the gas pipeline network, the static attribute data, dynamic sensing data and future event data of each node at time t are obtained in real time. The baseline risk coefficient of each node is pre-set based on the dynamic sensing data and future event data of each node at time t, and the physical failure probability is obtained based on the spatiotemporal graph neural network model, as well as the dynamic sensing data, future event data, and baseline risk coefficient of each node at time t. The spatiotemporal graph neural network model is as follows: In the formula, P fail (I,t) represents the physical failure probability of node i at time t, λ0(t) is the baseline risk coefficient for each node, which is set based on the dynamic sensing data and future event data of each node, W l Let b be the value of the dynamic sensing data of node i at time t. l Let i be the value of the future event data for node i at time t. Let be the value of the static attribute data of node i at time t.

4. The intelligent monitoring method for gas risk based on multi-source data according to claim 1, characterized in that, Step S3 further includes: The basic impact value of each node is obtained by weighted summation of the cost data of each node in the gas pipeline network. The dynamic impact weight of each node in the gas pipeline network is calculated based on the future event data and the dynamic impact weight calculation formula. The time-varying failure impact of each node in the gas pipeline network is calculated and obtained based on the time-varying failure impact calculation formula, the basic impact value and the dynamic impact weight. The basic impact value represents the impact value of the failure of one node in the gas pipeline network on the entire gas pipeline network. S3-1, based on the basic impact value calculation formula, the basic impact value of each node is obtained by weighted summation of the direct economic cost, indirect economic cost, social impact cost, environmental cost and safety cost in the cost data of each node. S3-2, Calculate the dynamic influence weight of each node based on the future event data and the dynamic influence weight calculation formula, wherein the dynamic influence weight calculation formula is: In the formula, w(i,t) represents the dynamic influence weight of each node, and E t Let L be the set of future events known at time t, e be a future event in the set of future events, I(·) be an indicator function, and L be a future event in the set of future events. e λ represents the scope of influence of future event e. e (t) is the influence intensity function of future event e at time t; S3-3, the time-varying failure impact of each node is calculated and obtained based on the time-varying failure impact calculation formula, the basic impact value, and the dynamic impact weight. The time-varying failure impact calculation formula is as follows: C impact (i,t)=w(i,t)×C base (i); In the formula, C impact (i,t) represents the time-varying failure impact of node i at time t, w(i,t) represents the dynamic impact weight of node i at time t, and C base (i) represents the basic influence value of node i.

5. The intelligent monitoring method for gas risk based on multi-source data according to claim 4, characterized in that, In step S3-1, the formula for calculating the basic influence value is: C base (i)=αC dir (i)+βC ind (i)+γC soc (i)+δC env (i); In the formula, C base (i) represents the base influence value of node i, C dir (i), C ind (i), C soc (i) and C env (i) represents the quantified values ​​of direct economic cost, indirect economic cost, social impact cost, environmental cost, and safety cost corresponding to the failure of node i; α, β, γ, and δ are pre-set normalized weight coefficients corresponding to direct economic cost, indirect economic cost, social impact cost, environmental cost, and safety cost, respectively, used to balance the weights of different cost types.

6. The intelligent monitoring method for gas risk based on multi-source data according to claim 1, characterized in that, In step S4, the step of calculating the risk value of each node in the gas pipeline network by combining the physical failure probability and the time-varying failure effect, and forming a risk value distribution based on the risk value, further includes: Based on the risk value calculation formula, using the physical failure probability and the time-varying failure impact as inputs, the risk value of each node in the gas pipeline network is calculated and obtained. The risk value distribution dynamically changes based on the risk values ​​of each node at different times, wherein the risk value distribution varies with time and the risk value of each node. The risk value calculation formula is as follows: R V (i,t)=P fail (i,t)×C impact (i,t); In the formula, R V (i,t) represents the risk value of node i at time t, P fail (i,t) represents the physical failure probability of node i at time t, C impact (i,t) represents the time-varying failure effect of node i at time t.

7. The intelligent monitoring method for gas risk based on multi-source data according to claim 1, characterized in that, Step S5, which involves generating maintenance decisions for the gas pipeline network based on the risk value distribution, includes the following steps: Based on the risk value distribution and expected risk calculation formula, the expected risk value of each node is calculated in real time. All nodes are sorted in descending order of expected risk value. The expected risk value of each node is compared with the expected risk threshold. The risk status of each node is determined based on the comparison result. The maintenance decision is determined based on the risk level of each node. The expected risk threshold is pre-set based on the risk influencing factors and cost data of each node, and the expected risk calculation formula is R. i =P fail (i,t)×R V (i,t), where R is the value of t. i Let R be the expected risk value of node i at time t. V (i,t) represents the risk value of node i at time t, P fail (i,t) represents the physical failure probability of node i at time t; If the expected risk value R i If the expected risk value R is greater than or equal to the expected risk threshold, then the expected risk value R is... i The corresponding node i at time t is marked as a high-risk node; If the expected risk value R i If the value is less than the expected risk threshold, then the expected risk value R is... i The risk status of the corresponding node i at time t is marked as a low-risk node.

8. The intelligent monitoring method for gas risk based on multi-source data according to claim 7, characterized in that, Step S5, which involves generating maintenance decisions for the gas pipeline network based on the risk value distribution, includes the following steps: The pre-set maintenance budget is used as the investment principal, and all possible maintenance actions are used as alternative financial assets, each of which has a corresponding maintenance cost and expected reduction in risk value. By applying modern portfolio theory models, and under the constraint of the maintenance budget, multiple optimal maintenance portfolio schemes are calculated with the objective of maximizing the expected reduction in value at risk. The optimal maintenance portfolio scheme is used as the maintenance decision.

9. The intelligent monitoring method for gas risk based on multi-source data according to claim 7, characterized in that, In step S5, the step of generating maintenance decisions for the gas pipeline network based on the risk value distribution also includes: A reinforcement learning model is established, wherein the action space of the reinforcement learning model is the selection from the optimal maintenance portfolio scheme, the state space of the reinforcement learning model is the risk value distribution, and the reward function of the reinforcement learning model is determined based on the risk value reduction and maintenance cost. The reinforcement learning model learns a long-term maintenance strategy, and based on the long-term maintenance strategy, the final maintenance decision is selected from the optimal maintenance portfolio scheme.

10. A system for implementing the intelligent gas risk monitoring method based on multi-source data as described in claim 1, characterized in that, The system includes a data acquisition module, a failure rate prediction module, a time-varying failure handling module, a risk value calculation module, and a maintenance decision generation module. in, The data acquisition module is used to acquire static attribute data, dynamic sensing data, future event data and cost data of each node in the gas pipeline network, and to construct a dynamic attribute map based on the static attribute data, the dynamic sensing data and the future event data. The failure rate prediction module is used to input the dynamic attribute map into a preset spatiotemporal graph neural network model, and predict the physical failure probability of each node in the gas pipeline network through the spatiotemporal graph neural network model. The time-varying failure processing module is used to perform weighted summation of cost data of each node in the gas pipeline network to obtain the basic impact value of each node, calculate the dynamic impact weight of each node in the gas pipeline network based on the future event data and the dynamic impact weight calculation formula, and calculate and obtain the time-varying failure impact of each node in the gas pipeline network based on the time-varying failure impact calculation formula, the basic impact value and the dynamic impact weight. The basic impact value represents the impact value of the failure of one node in the gas pipeline network on the entire gas pipeline network. The risk value calculation module is used to combine the physical failure probability and the time-varying failure impact to calculate the risk value of each node in the gas pipeline network, and form a risk value distribution based on the risk value. The maintenance decision generation module is used to generate maintenance decisions for the gas pipeline network based on the risk value distribution.

Images