A substation fire intelligent risk assessment model construction method and system

Abstract

The application discloses a substation fire intelligent risk assessment model construction method and system, through obtaining substation multi-source risk data, constructing an initial risk index system, mapping each risk factor in the key risk factor set into a Bayesian network node, establishing the causal relationship connection between each node, and constructing a Bayesian network structure model; a coupling strength matrix between each risk factor is constructed to represent the linkage influence relationship between each risk factor, the key parameters in the Bayesian network inference process are optimized by using a TC-CGO-BP algorithm, and an optimized Bayesian network structure model is obtained; a risk threshold is set, and when the fire risk probability value exceeds the risk threshold, an early warning signal is triggered. The efficiency and accuracy of the fire intelligent risk assessment are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent fire risk assessment technology, and in particular to a method and system for constructing an intelligent fire risk assessment model for substations. Background Technology

[0002] In recent years, with the continuous expansion of the power grid, the accelerated aging of equipment, and the frequent occurrence of extreme weather events, substation fire risks have exhibited significant characteristics such as multi-factor coupling and complex evolution processes. Currently, existing substation fire risk assessment methods rely heavily on expert subjective experience, and the weighting of indicators lacks objectivity and consistency. Most methods treat each risk factor as independent, failing to fully characterize equipment status and environmental factors. Traditional Bayesian network structures are fixed, and parameters are difficult to update dynamically, resulting in insufficient adaptive capabilities when dealing with real-time changes in operating conditions. These factors severely restrict the accuracy and efficiency of substation fire risk assessment. Summary of the Invention

[0003] The purpose of this invention is to solve the above-mentioned problems by designing a method and system for constructing an intelligent risk assessment model for substation fires.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned method for constructing an intelligent risk assessment model for substation fires: Acquire multi-source risk data of substations, construct an initial risk indicator system, conduct correlation analysis and information gain assessment on each risk factor in the initial risk indicator system, and screen a set of key risk factors. Each risk factor in the set of key risk factors is mapped to a Bayesian network node, causal relationships are established between the nodes, and a Bayesian network structure model is constructed. Construct a coupling strength matrix among various risk factors to characterize the interconnected influence among them. Based on the Bayesian network structure model, perform Bayesian network inference using the coupling strength matrix. The key parameters in the Bayesian network inference process are optimized using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model. Based on the optimized Bayesian network structure model, a corrected fire risk probability value is output, and a risk threshold is set. When the fire risk probability value exceeds the risk threshold, an early warning signal is triggered.

[0005] Furthermore, in the above-mentioned method for constructing an intelligent risk assessment model for substation fires, the steps of acquiring multi-source risk data of the substation, constructing an initial risk indicator system, performing correlation analysis and information gain assessment on each risk factor in the initial risk indicator system, and screening a set of key risk factors include: Multi-source risk data of substations are acquired to construct an initial risk indicator system covering four dimensions: equipment status, environmental factors, management level, and emergency response capabilities. The initial risk indicator system introduces lightning strike fire indicators and external fire spread indicators to characterize the impact of external disasters on the fire risk of substations.

[0006] Furthermore, in the above-mentioned method for constructing an intelligent risk assessment model for substation fires, the steps of acquiring multi-source risk data of the substation, constructing an initial risk indicator system, performing correlation analysis and information gain assessment on each risk factor in the initial risk indicator system, and screening a set of key risk factors include: Correlation analysis and information gain assessment are performed on each risk factor in the initial risk indicator system to screen out a set of key risk factors; and the continuous variables in the set of key risk factors are discretized and mapped to a finite state space suitable for Bayesian networks.

[0007] Furthermore, in the above-mentioned method for constructing an intelligent risk assessment model for substation fires, the steps of acquiring multi-source risk data of the substation, constructing an initial risk indicator system, performing correlation analysis and information gain assessment on each risk factor in the initial risk indicator system, and screening a set of key risk factors include: Methods for discretizing continuous variables include any one or more combinations of equal-width binning discretization, K-means clustering-based adaptive discretization, and soft discretization based on fuzzy membership functions.

[0008] Furthermore, in the above-mentioned method for constructing an intelligent risk assessment model for substation fires, the optimization of key parameters in the Bayesian network inference process using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model includes: Migration component analysis is used to map the original risk features to a common feature space, reducing the distribution differences between different data sources and enhancing the feature representation ability.

[0009] Furthermore, in the above-mentioned method for constructing an intelligent risk assessment model for substation fires, the optimization of key parameters in the Bayesian network inference process using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model includes: An improved population optimization algorithm (CGO) is used to globally optimize the initial weights and thresholds of a backpropagation neural network. The optimal combination of network parameters is searched through population initialization, fitness evaluation, and iterative update operations. The initial risk probability output by the Bayesian network inference and the mapped features are used as input to the BP network. The corrected risk probability output is obtained through forward propagation, and the network is trained using mean squared error as the loss function.

[0010] Further, in the above method for constructing an intelligent fire risk assessment model for a substation, the steps of outputting a corrected fire risk probability value according to the optimized Bayesian network structure model, setting a risk threshold, and triggering an early warning signal when the fire risk probability value exceeds the risk threshold include: A plurality of risk thresholds T1, T2 and T3 are set, where 0≤T1<T2<T3≤1. According to the threshold interval in which the fire risk probability value P falls, the risk levels are divided into four levels: low risk, medium risk, relatively high risk and high risk; when P≥T3, it is determined as high risk and the highest-level early warning is triggered.

[0011] Further, in an intelligent fire risk assessment model construction system for a substation, the risk assessment model construction system includes the following modules: An indicator establishment module, configured to obtain multi-source risk data of the substation, construct an initial risk indicator system, perform correlation analysis and information gain evaluation on each risk factor in the initial risk indicator system, and screen out a set of key risk factors; A model establishment module, configured to map each risk factor in the set of key risk factors to a Bayesian network node, establish causal connections between the nodes, and construct a Bayesian network structure model; A Bayesian inference module, configured to construct a coupling strength matrix between the risk factors to characterize the linkage influence relationship between the risk factors, and perform Bayesian network inference based on the Bayesian network structure model in combination with the coupling strength matrix; A parameter optimization module, configured to optimize key parameters in the Bayesian network inference process by using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model; A risk assessment module, configured to output a corrected fire risk probability value according to the optimized Bayesian network structure model, set a risk threshold, and trigger an early warning signal when the fire risk probability value exceeds the risk threshold.

[0012] Further, in the intelligent fire risk assessment model construction system for a substation, the parameter optimization module includes the following sub-modules: An optimization searching sub-module, configured to perform global optimization on the initial weights and thresholds of the BP neural network by using the improved group optimization algorithm CGO, and search for the optimal network parameter combination through operations of population initialization, fitness evaluation and iterative update, so as to prevent the BP network from falling into a local optimum; An input sub-module, configured to take the initial risk probability output by Bayesian network inference and the mapped features as the input of the BP network, obtain a corrected risk probability output through forward propagation calculation, and train the network by taking the mean square error as the loss function.

[0013] Further, in a substation fire intelligent risk assessment model construction system, the risk assessment module comprises the following sub-modules: A setting sub-module, configured to set a plurality of risk thresholds T1, T2 and T3, where 0≤T1<T2<T3≤1, and divide the risk level into four levels of low risk, medium risk, relatively high risk and high risk according to the threshold interval where the fire risk probability value P is located; when P≥T3, it is determined as high risk and the highest-level warning is triggered.

[0014] The beneficial effects are as follows: by constructing a risk indicator system covering four dimensions of equipment status, environmental factors, management level and emergency response capability, the systematic integration and fine characterization of multi-source risk factors are realized, and the model's ability to describe complex fire scenarios is significantly enhanced. By improving the combined weighting method to optimize the distribution of indicator weights, the defect that traditional methods overly rely on subjective experience is effectively overcome, and the efficiency and accuracy of risk assessment are improved. By introducing a risk linkage mechanism and constructing a risk factor coupling strength matrix, the Bayesian network can effectively characterize the non-linear superposition and propagation effects among multiple factors, and overcome the deficiency that traditional methods ignore the linkage relationship between factors; the TC-CGO-BP algorithm is used to carry out global optimization and dynamic correction on model parameters, which greatly improves the fitting accuracy of complex non-linear relationships and the dynamic self-adaptation capability of the model. The assessment accuracy of the method of the present invention reaches 92.6%, which is about 3.9 percentage points higher than that of the traditional Bayesian network. Brief Description of the Drawings

[0015] Through reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The accompanying drawings are only for the purpose of illustrating the preferred embodiments, and are not to be regarded as a limitation to the present invention.

[0016] Figure 1 is a schematic diagram of the first embodiment of a method for constructing an intelligent substation fire risk assessment model in the embodiment of the present invention; Figure 2 is a schematic diagram of the second embodiment of the method for constructing an intelligent substation fire risk assessment model in the embodiment of the present invention. Detailed Description of the Embodiments

[0017] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not intended to limit the present invention.

[0018] The present invention will be specifically described below with reference to the accompanying drawings, Figure 1As shown, a method for constructing an intelligent risk assessment model for substation fires includes the following steps: Step 101: Obtain multi-source risk data of substations, construct an initial risk indicator system, conduct correlation analysis and information gain assessment on each risk factor in the initial risk indicator system, and screen the set of key risk factors. Specifically, in this embodiment Based on substation operation data, an initial feature set is constructed, comprising equipment state variables, environmental variables, and operational behavior variables. Let the original feature set be: (1); Where (x_i) represents the (i)th risk factor. To screen key variables, correlation analysis is introduced to measure the degree of association between the variables and fire risk.

[0019] The Pearson correlation coefficient is used to measure the correlation of continuous variables; its expression is as follows: (2); Where (y) represents the fire risk label variable, , These are the mean values. By setting a threshold ( The filter satisfies: (3); The variables are used as a preliminary set of key factors.

[0020] To further reduce redundancy, feature selection methods are introduced, such as evaluation based on information gain (IG): (4); The entropy function is defined as: (5); The conditional entropy is: (6); Finally, features with higher information gain are retained to construct an optimized set of risk factors: (7); Step 102: Map each risk factor in the set of key risk factors to a Bayesian network node, establish causal relationship connections between nodes, and construct a Bayesian network structure model. Specifically, in this embodiment; Building upon the substation fire risk indicator system established earlier, the four dimensions of risk factors—equipment, environment, management, and emergency response—are uniformly mapped to Bayesian network nodes. Specific indicators such as lightning strikes and fire spread outside the substation are introduced to form a multi-level set of risk nodes. Connections between different dimensions are established through causal relationships, thereby constructing a complete risk chain covering "external triggers—internal conditions—accident consequences."

[0021] Step 103: Construct a coupling strength matrix between each risk factor to characterize the linkage and influence relationship between each risk factor. Based on the Bayesian network structure model, perform Bayesian network inference in combination with the coupling strength matrix. Specifically, in this embodiment; In the structural modeling process, a combination of data-driven and expert experience is used: statistical dependencies between variables are analyzed based on historical operating data, while constraints on critical paths (such as "environment → equipment → fire" and "external disaster → propagation within the station") are imposed in conjunction with the substation operating mechanism to ensure that the network structure is both data-supported and conforms to engineering practice.

[0022] Bayesian networks, through the assumption of conditional independence, decompose the joint probability of a complex system into a product of local conditional probabilities: (13); This expression provides an efficient approach for the joint modeling of multidimensional risk factors and lays the foundation for subsequent risk linkage analysis.

[0023] Step 104: Optimize the key parameters in the Bayesian network inference process using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model; Specifically, in this embodiment; Building upon traditional Bayesian network inference, a risk linkage mechanism is introduced to characterize the coupling relationships between multiple risk factors. Simultaneously, the TC-CGO-BP algorithm is used to optimize key parameters in the model inference process, thereby improving the accuracy of risk assessment.

[0024] First, TC (TransferComponent) is used to map and reconstruct the original risk features, enhancing their expressive power and making data from different sources more separable in a unified feature space. Let the original feature be (X) and the mapped feature be (Z), then: (14) in, This represents the feature mapping function.

[0025] Building upon this, CGO (Culture Group Optimization) is used to globally optimize the parameters of the BP neural network. Through population initialization, fitness evaluation, and iterative updates, the optimal combination of weights and thresholds is searched, thereby preventing the BP network from getting trapped in local optima.

[0026] The output of the BP network is used to perform nonlinear corrections on the inference results of the Bayesian network. The basic mapping relationship is as follows: (15); in, (a) represents network weights, and (b) represents the bias term. This is the activation function.

[0027] The CGO algorithm is used to optimize the parameters of the BP network, causing the error function to gradually converge. (16); Under the risk linkage mechanism, the state change of a certain risk node will be propagated through the Bayesian network structure, and the propagation result will be dynamically corrected by combining the nonlinear fitting capability of the BP network, thereby realizing the integrated modeling of "probabilistic reasoning + nonlinear optimization".

[0028] After completing Bayesian network inference and TC-CGO-BP optimization, the model outputs an optimized fire risk probability, which is used for risk level classification and early warning triggering.

[0029] Let the initial risk probability obtained from Bayesian network inference be... The output after TC-CGO-BP optimization is Then it can be expressed as: (17); in, This represents a nonlinear mapping function constructed using a BP network. This is the set of parameters optimized by CGO.

[0030] A threshold determination mechanism is introduced during the risk warning process. When the optimized risk probability exceeds a set threshold (T), a warning signal is triggered. (18); Alarm=1 indicates that an early warning has been triggered.

[0031] Step 105: Based on the optimized Bayesian network structure model, output the corrected fire risk probability value, set a risk threshold, and trigger an early warning signal when the fire risk probability value exceeds the risk threshold.

[0032] Specifically, in this embodiment; In actual operation, the risk factors in substations are not independent of each other, but rather exhibit significant coupling and linkage relationships. For example, equipment overheating may be affected by both increased ambient temperature and increased load, further leading to insulation aging and even fire risks. Therefore, it is necessary to model the interactions between risk factors.

[0033] Let the set of risk factors be... By introducing the coupling strength matrix Describe the influence relationships between factors, among which Indicates risk factors right The extent of the impact.

[0034] When considering the interconnectedness of risks, the relatively independent probabilistic relationships between nodes in traditional Bayesian networks need to be extended to reflect the dynamic impact of interconnected factors on the probability distribution.

[0035] During model inference, when the observations of certain nodes change, their impact propagates to other related nodes through interconnected relationships, thereby altering the probability distribution of the entire network. At this point, the conditional probability of a node no longer depends solely on its direct parent node but is also modulated by interconnected factors.

[0036] In terms of form, the linkage effect can be regarded as a modification of the original conditional probability, so that the posterior probability of the node can dynamically reflect the changes in the system state.

[0037] Based on the Bayesian network inference results, the posterior probability of the target fire risk node can be obtained. , where (F) represents a fire event and (E) represents observed evidence. This probability reflects the likelihood of a fire occurring in the system under the current operating conditions.

[0038] Building upon this, a risk level classification mechanism is introduced to map continuous probabilities to discrete risk levels, facilitating engineering applications and management decisions. Multiple risk thresholds are set. The risks are divided into different levels, and the judgment rules are as follows: (19); Among them, each threshold The settings can be made based on historical data statistics or expert experience to ensure that the grading results match the actual operating scenario.

[0039] Methodology for constructing a comprehensive risk index; Since a single risk node cannot fully reflect the overall risk level of the system, it is necessary to comprehensively evaluate the inference results of multiple nodes. This study uses a weighted fusion method to construct a Composite Risk Index (CRI) to achieve a unified expression of multidimensional risk information.

[0040] Suppose there are (n) critical risk nodes in the Bayesian network, and their corresponding posterior probabilities are... The comprehensive risk index can then be expressed as: (20); in, The weights of each risk node satisfy the following: (twenty one); Weight The risk factors can be determined based on their importance, historical impact, or expert assignment. By introducing a weighting mechanism, the model can highlight the impact of key risk factors.

[0041] Furthermore, the comprehensive risk index can be mapped to risk level ranges to achieve a unified assessment standard. A threshold needs to be set. ,but: (twenty two); Using the above method, the results of multi-node reasoning can be integrated into a single indicator, enabling a quantitative expression of the overall fire risk of the substation, thereby providing a basis for dispatching decisions and risk warnings.

[0042] Its beneficial effects lie in the fact that by constructing a risk indicator system covering four dimensions—equipment status, environmental factors, management level, and emergency response capabilities—it achieves systematic integration and refined characterization of multi-source risk factors, significantly enhancing the model's ability to depict complex fire scenarios. By improving the combined weighting method and optimizing the allocation of indicator weights, it effectively overcomes the shortcomings of traditional methods that overly rely on subjective experience, improving the efficiency and accuracy of risk assessment. By introducing a risk linkage mechanism and constructing a risk factor coupling strength matrix, the Bayesian network can effectively characterize the nonlinear superposition and propagation effects between multiple factors, overcoming the shortcomings of traditional methods that ignore the linkage relationships between factors. Combined with the TC-CGO-BP algorithm for global optimization and dynamic correction of model parameters, it significantly improves the fitting accuracy of complex nonlinear relationships and the model's dynamic adaptive capability. The evaluation accuracy of the method in this invention reaches 92.6%, an improvement of approximately 3.9 percentage points compared to traditional Bayesian networks.

[0043] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and equivalents.

Claims

1. A method for constructing an intelligent risk assessment model for substation fires, characterized in that, The risk assessment model construction method includes the following steps: Acquire multi-source risk data of substations, construct an initial risk indicator system, conduct correlation analysis and information gain assessment on each risk factor in the initial risk indicator system, and screen a set of key risk factors. Each risk factor in the set of key risk factors is mapped to a Bayesian network node, causal relationships are established between the nodes, and a Bayesian network structure model is constructed. Construct a coupling strength matrix among various risk factors to characterize the interconnected influence among them. Based on the Bayesian network structure model, perform Bayesian network inference using the coupling strength matrix. The key parameters in the Bayesian network inference process are optimized using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model. Based on the optimized Bayesian network structure model, a corrected fire risk probability value is output, and a risk threshold is set. When the fire risk probability value exceeds the risk threshold, an early warning signal is triggered.

2. The method for constructing an intelligent risk assessment model for substation fires as described in claim 1, characterized in that, The process involves acquiring multi-source risk data from substations, constructing an initial risk indicator system, performing correlation analysis and information gain assessment on each risk factor within the initial risk indicator system, and screening a set of key risk factors, including: Multi-source risk data of substations are acquired to construct an initial risk indicator system covering four dimensions: equipment status, environmental factors, management level, and emergency response capabilities. The initial risk indicator system introduces lightning strike fire indicators and external fire spread indicators to characterize the impact of external disasters on the fire risk of substations.

3. The method for constructing an intelligent risk assessment model for substation fires as described in claim 2, characterized in that, The process involves acquiring multi-source risk data from substations, constructing an initial risk indicator system, performing correlation analysis and information gain assessment on each risk factor within the initial risk indicator system, and screening a set of key risk factors, including: Correlation analysis and information gain assessment are performed on each risk factor in the initial risk indicator system to screen out a set of key risk factors; and the continuous variables in the set of key risk factors are discretized and mapped to a finite state space suitable for Bayesian networks.

4. The method for constructing an intelligent risk assessment model for substation fires as described in claim 3, characterized in that, The process involves acquiring multi-source risk data from substations, constructing an initial risk indicator system, performing correlation analysis and information gain assessment on each risk factor within the initial risk indicator system, and screening a set of key risk factors, including: Methods for discretizing continuous variables include any one or more combinations of equal-width binning discretization, K-means clustering-based adaptive discretization, and soft discretization based on fuzzy membership functions.

5. The method for constructing an intelligent risk assessment model for substation fires as described in claim 1, characterized in that, The optimization of key parameters in the Bayesian network inference process using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model includes: Migration component analysis is used to map the original risk features to a common feature space, reducing the distribution differences between different data sources and enhancing the feature representation ability.

6. The method for constructing an intelligent risk assessment model for substation fires as described in claim 5, characterized in that, The optimization of key parameters in the Bayesian network inference process using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model includes: The improved group optimization algorithm CGO is used to perform global optimization on the initial weights and thresholds of the BP neural network, and the optimal network parameter combination is searched through operations of population initialization, fitness evaluation and iterative update; The initial risk probability output by Bayesian network inference and the mapped features are taken as the input of the BP network, the corrected risk probability output is obtained through forward propagation calculation, and the network is trained with the mean square error as the loss function.

7. The method for constructing an intelligent risk assessment model for substation fires as described in claim 1, characterized in that, The step of outputting the corrected fire risk probability value according to the optimized Bayesian network structure model, setting a risk threshold, and triggering an early warning signal when the fire risk probability value exceeds the risk threshold includes: A plurality of risk thresholds T1, T2 and T3 are set, where 0≤T1<T2<T3≤1, and according to the threshold interval where the fire risk probability value P is located, the risk levels are divided into four levels: low risk, medium risk, relatively high risk and high risk; when P≥T3, it is determined as high risk and the highest-level early warning is triggered.

8. A system for constructing an intelligent risk assessment model for substation fires, characterized in that, The risk assessment model construction system includes the following modules: An indicator establishment module, configured to acquire multi-source risk data of a substation, construct an initial risk indicator system, perform correlation analysis and information gain evaluation on each risk factor in the initial risk indicator system, and screen out a key risk factor set; A model establishment module, configured to map each risk factor in the key risk factor set to a Bayesian network node, establish causal relationship connections between the nodes, and construct a Bayesian network structure model; A Bayesian inference module, configured to construct a coupling strength matrix between the risk factors to characterize the linkage influence relationship between the risk factors, and perform Bayesian network inference based on the Bayesian network structure model in combination with the coupling strength matrix; A parameter optimization module, configured to optimize key parameters in the Bayesian network inference process by using the TC-CGO-BP algorithm to obtain an optimized Bayesian network structure model; A risk assessment module, configured to output a corrected fire risk probability value according to the optimized Bayesian network structure model, set a risk threshold, and trigger an early warning signal when the fire risk probability value exceeds the risk threshold.

9. The intelligent risk assessment model construction system for substation fires as described in claim 8, characterized in that, The parameter optimization module includes the following sub-modules: An optimization seeking sub-module, configured to use the improved group optimization algorithm CGO to perform global optimization on the initial weights and thresholds of the BP neural network, search for the optimal network parameter combination through operations of population initialization, fitness evaluation and iterative update, and prevent the BP network from falling into a local optimum; An input sub-module, configured to take the initial risk probability output by Bayesian network inference and the mapped features as the input of the BP network, obtain the corrected risk probability output through forward propagation calculation, and train the network with the mean square error as the loss function.

10. The intelligent risk assessment model construction system for substation fires as described in claim 8, characterized in that, The risk assessment module includes the following sub-modules: A setting sub-module, configured to set a plurality of risk thresholds T1, T2 and T3, where 0≤T1<T2<T3≤1, and divide the risk levels into four levels of low risk, medium risk, relatively high risk and high risk according to the threshold interval where the fire risk probability value P is located; When P≥T3, it is determined as high risk and the highest-level early warning is triggered.

Images