An Electrical Fire Risk Assessment Method Based on Dynamic Weight Adjustment

By using dynamic weight adjustment and a multi-dimensional evaluation index system, combined with the analytic hierarchy process, entropy weight method, fuzzy comprehensive evaluation and BP neural network, the timeliness and accuracy of electrical fire risk assessment are solved, and dynamic, objective and accurate assessment of electrical fire risk and identification of key factors are achieved.

CN123070698APending Publication Date: 2026-11-10STATE GRID SICHUAN ELECTRIC POWER CORP ELECTRIC POWER RES INST +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611225950.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-13
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing electrical fire risk assessment methods lack timeliness, fixed weights cannot reflect dynamic changes in parameters, multi-circuit index fusion processing is unreasonable, and assessment results are difficult to identify key disaster-causing factors, leading to discrepancies between assessment results and actual conditions.

Method used

An electrical fire risk assessment method with dynamic weight adjustment is adopted. The initial weights are determined by a combination weighting method, and a dynamic weighting strategy is constructed by combining the analytic hierarchy process and the entropy weighting method. Fuzzy comprehensive evaluation and BP neural network are used for evaluation, and key risk factors are identified by Bayesian back reasoning.

Benefits of technology

It enables dynamic, objective, and accurate assessment of electrical fire risks, reflects changes in the operating status of electrical circuits in real time, identifies key risk factors, provides targeted basis for fire prevention and control, and improves the timeliness and accuracy of assessment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123070698A_ABST
    Figure CN123070698A_ABST
Patent Text Reader

Abstract

This application relates to the field of electrical fire safety assessment technology, and in particular to an electrical fire risk assessment method based on dynamic weight adjustment. The method includes: constructing an electrical fire risk assessment index system, selecting assessment indicators from four dimensions: electrical factors, environmental factors, spatial factors, and human factors; determining the initial weights of the indicators using a combined weighting method combining the analytic hierarchy process (AHP) and the entropy weighting method; dynamically adjusting the weights based on real-time monitoring data of the indicators; performing risk-level weighted fusion processing on multi-circuit monitoring data; generating a training dataset using the fuzzy comprehensive evaluation method; determining the risk level using a BP neural network model; and identifying key disaster-causing factors based on Bayesian back-reasoning. This method achieves real-time and accurate assessment of electrical fire risk through dynamic weight adjustment, and by combining multi-circuit indicator fusion with neural network learning, it improves assessment accuracy, can trace key risk factors, and effectively reduces the false negative rate in electrical fire risk assessment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of electrical fire safety assessment technology, and in particular to an electrical fire risk assessment method based on dynamic weight adjustment. Background Technology

[0002] With the continuous rise in per capita electricity consumption, electrical fires have become the leading cause of all types of fires. According to statistics from the Fire and Rescue Bureau of the Ministry of Emergency Management, approximately 33.6% of fires with identified causes were due to electrical issues, primarily short circuits, overloads, and poor connections. Low-voltage power distribution networks are complex, large-scale, and suffer from weak operation and maintenance management, making electrical fire hazards particularly prominent and posing a significant challenge to fire prevention and control efforts.

[0003] Currently, scholars both domestically and internationally have conducted relevant research in the field of electrical fire risk assessment. Franca et al. proposed classifying electrical fire risk into five risk states based on a combination of three variables, providing a qualitative analysis of electrical fire risk. Ma Zichao et al. proposed an electrical fire risk assessment method based on hierarchical entropy and radar charts, using entropy weighting to correct the weights of the analytic hierarchy process (AHP). Su et al., based on the occurrence mechanism of electrical fires in high-rise buildings, proposed an electrical fire risk assessment model considering asymmetric proximity-based hybrid decision-making. Su Lei et al. used an improved Delphi method to obtain an assessment index system, combined with the interval TOPSIS method to identify risk factors.

[0004] However, current electrical fire risk assessment practices primarily employ static methods, lacking timeliness. Electrical fires are essentially dynamic processes involving the abnormal release of energy and the interaction of combustibles, exhibiting multifactorial and time-varying risk characteristics. Existing technologies fail to effectively integrate real-time operational data from electrical circuits, resulting in assessment results that lack timeliness and objectivity.

[0005] Furthermore, the weighting of indicators is often fixed, making it difficult to reflect the impact of dynamic parameter changes on risk. When certain electrical parameters, such as current, temperature, and leakage current, undergo abnormal changes, their relative importance in the overall risk assessment will change significantly. Fixed weights cannot capture such dynamic changes, leading to distorted assessment results.

[0006] Furthermore, the method for integrating multi-circuit indicators is unreasonable. In actual low-voltage distribution networks, there are multiple distribution circuits. Existing technologies typically use either the maximum value method or the average weight allocation method to integrate multi-circuit indicator parameters. The maximum value method tends to significantly amplify the risk status, distorting the assessment results; the average weight allocation method tends to lose information about circuits with potential risks, resulting in the overall system being assessed as normal even when risks exist.

[0007] Finally, the assessment results struggle to trace key hazard-causing factors. Most existing assessment methods only output risk levels and cannot further identify key indicators that lead to increased risk, making it difficult to provide targeted basis for subsequent fire prevention and control decisions.

[0008] In summary, existing electrical fire risk assessment methods have significant limitations in terms of dynamism, weight rationality, scientific integration of indicators, and identification of key factors. These limitations can easily lead to discrepancies between assessment results and actual conditions, and fail to meet the practical needs for precise prevention and control of electrical fire risks. Summary of the Invention

[0009] This invention provides an electrical fire risk assessment method based on dynamic weight adjustment, which enables dynamic, objective, and accurate assessment of electrical fire risks in low-voltage power distribution lines of buildings.

[0010] To achieve the above objectives, the present invention adopts the following technical solution:

[0011] This invention provides a method for assessing electrical fire risk based on dynamic weight adjustment, the method comprising the following steps:

[0012] Step S1: Construct an electrical fire risk assessment index system, selecting assessment indicators from four dimensions: electrical factors, environmental factors, spatial factors, and human factors;

[0013] Step S2: Determine the initial weights of each evaluation indicator using the combined weighting method, including determining the subjective weights of the indicators using the analytic hierarchy process, determining the objective weights of the indicators using the entropy weighting method, and combining the subjective and objective weights according to the principle of minimum information entropy to obtain the combined weights.

[0014] Step S3: Construct a dynamic weighting strategy for each evaluation indicator and dynamically adjust the weights based on real-time monitoring data of the indicators;

[0015] Step S4: Perform weighted fusion processing on the multi-loop monitoring data, assigning different weights to each indicator of each loop according to its risk level for weighted fusion;

[0016] Step S5: Use the fuzzy comprehensive evaluation method to comprehensively evaluate the fused index data and generate a training dataset;

[0017] Step S6: Construct and train the BP neural network model;

[0018] Step S7: Use the trained BP neural network model to conduct an electrical fire risk assessment;

[0019] Step S8: Identify key risk factors based on Bayesian reverse reasoning.

[0020] Preferably, step S2 uses the analytic hierarchy process (AHP) to determine the subjective weights of the indicators, including:

[0021] Construct a judgment matrix by using a 9-level scaling method, and establish the judgment matrix by comparing the importance of different elements in each level;

[0022] The judgment matrix is ​​normalized column by column;

[0023] Calculate the product of all elements in each row of the normalized matrix, find the nth root of the product and normalize it to obtain the eigenvector, which is the subjective weight vector.

[0024] Perform a consistency check and calculate the consistency ratio CR. When CR < 0.1, the matrix is ​​considered to have passed the consistency check.

[0025] Preferably, step S2 uses the entropy weight method to determine the objective weights of the indicators, including:

[0026] Construct an evaluation matrix, with m evaluation objects and n evaluation indicators;

[0027] The indicators are standardized, and different standardization formulas are used for positive and negative indicators.

[0028] Calculate the proportion of the i-th evaluation object under the j-th indicator to that indicator;

[0029] Calculate the entropy value of the j-th index;

[0030] Calculate the difference coefficient of the j-th indicator;

[0031] Calculate the entropy weight of the j-th index based on the difference coefficient.

[0032] Preferably, in step S2, the subjective weights and objective weights are combined according to the principle of minimum information entropy to obtain the combined weights. :

[0033]

[0034] in, Let the combined weight of the j-th indicator be , Let j be the subjective weight of the j-th indicator. Let j be the objective weight of the j-th indicator. The subjective weight of the k-th indicator. denoted as the objective weight of the k-th indicator, where n is the total number of indicators, and j and k are the indicator numbers, with values ​​ranging from 1, 2, ..., n.

[0035] Preferably, in step S3, for most indicators, when the monitored value exceeds the warning boundary value, the following variable weighting formula is used:

[0036] when hour:

[0037]

[0038] when hour:

[0039]

[0040] in, Let be the dynamic weight of the j-th indicator at time t; Let be the initial weight of the j-th indicator combination; α is the weight adjustment coefficient, with a value ranging from 0.5 to 2.0; Let be the warning boundary value for the j-th indicator; This represents the maximum allowed value for the j-th indicator.

[0041] Preferably, in step S3, for indicators with abrupt change characteristics, the following variable weighting formula is used:

[0042] when ≥ hour:

[0043]

[0044] Wherein, β is the mutation regulation coefficient, with a value ranging from 1.0 to 3.0.

[0045] Preferably, in step S3, the dynamically adjusted weights are normalized:

[0046]

[0047] in, Let j be the normalized dynamic weight of the j-th index at time t, satisfying .

[0048] Preferably, in step S4, the risk level range of each indicator is divided into five levels {v1, v2, v3, v4, v5}, which correspond to the risk status: normal, basically normal, abnormal, risk, and alarm, respectively.

[0049] Determine the risk level weight corresponding to the indicator values ​​of each loop. :

[0050]

[0051] Calculate the fusion value of the j-th indicator. :

[0052] in, The value of the j-th indicator is obtained by weighted fusion of multiple circuits; N is the total number of distribution circuits; The monitoring value of the j-th indicator in the i-th loop; The risk level weight corresponds to the j-th indicator value of the i-th loop.

[0053] Preferably, step S5 employs the fuzzy comprehensive evaluation method, including:

[0054] Construct a membership function using a combination of trapezoidal and trigonometric functions;

[0055] Calculate the membership vector of each indicator and construct a fuzzy evaluation matrix;

[0056] Calculate the fuzzy comprehensive evaluation result B:

[0057]

[0058] in, This is used to assess the membership degree of an object belonging to the kth risk level.

[0059] Risk levels are determined based on the principle of maximum membership.

[0060] Preferably, the BP neural network structure constructed in step S6 includes an input layer, a first hidden layer, a second hidden layer, and an output layer.

[0061] Number of nodes in the first hidden layer Determined by the following empirical formula:

[0062]

[0063] Where m is the number of output layer nodes, and a is a constant between 1 and 10;

[0064] Number of nodes in the second hidden layer Determined by the following empirical formula:

[0065]

[0066] in, It is a constant between 1 and 10.

[0067] Preferably, in step S6, the network is trained using the backpropagation algorithm, and the outputs of the hidden layer and the output layer during the forward propagation process are:

[0068]

[0069] in, Let v be the connection weight from the v-th node in layer l (l-1) to the u-th node in layer l. Let f(·) be the threshold for the u-th node in the l-th layer, and f(·) be the activation function, using the Sigmoid function:

[0070]

[0071] The output layer uses the Softmax function to convert the output into a probability distribution:

[0072]

[0073] Cross-entropy is used as the loss function:

[0074]

[0075] Preferably, step S8, which identifies key risk factors based on Bayesian reverse reasoning, includes:

[0076] The risk level obtained from the assessment is input into the Bayesian network as evidence.

[0077] Assuming the electrical fire risk status is set to the highest risk level v5, we use Bayesian network for backpropagation to calculate the posterior probability of each indicator node:

[0078]

[0079] Wherein, Nodej is the j-th indicator node;

[0080] The indicators are ranked according to their posterior probability to identify the key factors that have the most significant impact on electrical fire risk.

[0081] Preferably, the electrical fire risk assessment system based on dynamic weight adjustment includes:

[0082] The data acquisition module is used to collect monitoring data of various evaluation indicators of each power distribution circuit in real time;

[0083] The data preprocessing module is used to clean, filter, and standardize the collected data.

[0084] The multi-loop fusion module is used to perform weighted fusion processing on multi-loop monitoring data;

[0085] The dynamic weight calculation module is used to calculate the dynamic weight of each indicator based on real-time monitoring data.

[0086] The risk assessment module has a built-in trained BP neural network model, which is used to output electrical fire risk assessment results based on the fused indicator data and dynamic weights.

[0087] The key factor identification module is used to identify key risk factors based on Bayesian network-based reverse reasoning.

[0088] The early warning output module is used to output risk levels and early warning information based on the assessment results;

[0089] The display module is used to display real-time monitoring data, dynamic weights, risk assessment results, and key factor information for each indicator.

[0090] Compared with the prior art, the present invention has the following beneficial effects:

[0091] 1. This invention constructs a multi-dimensional evaluation index system and uses a combined weighting strategy of analytic hierarchy process (AHP) and entropy weighting to determine the initial weights. It takes into account both the subjective judgment of expert experience and the objective information of sample data, forming a three-layer weight determination mechanism of subjective weighting, objective weighting and dynamic adjustment, making the weight allocation more reasonable.

[0092] 2. This invention introduces a dynamic weighting strategy, which dynamically adjusts the weight of each indicator in the risk assessment based on real-time monitoring data, so that the assessment results can reflect the changes in the operating status of electrical circuits in real time. This overcomes the shortcomings of the existing static assessment methods in terms of timeliness and realizes dynamic assessment of electrical fire risk.

[0093] 3. This invention proposes a multi-loop indicator fusion method based on risk level weighting. It assigns different weights to each indicator in each loop according to the risk level of each loop and performs weighted fusion. This avoids the problem of amplifying local risks by the maximum value method and overcomes the defect of the average value method in overwhelming hidden danger information, thereby improving the rationality of multi-loop indicator fusion.

[0094] 4. This invention uses Bayesian network back reasoning to calculate the posterior probability of each indicator under a given risk level, thereby accurately identifying the key factors that have the most significant impact on electrical fire risk, achieving accurate identification of key risk factors, and providing targeted basis for subsequent fire prevention and control decisions.

[0095] 5. This invention uses fuzzy comprehensive evaluation to generate training datasets and BP neural networks for learning and prediction. The combination of the two fully leverages the advantages of fuzzy comprehensive evaluation in processing fuzzy information and the self-learning and strong generalization capabilities of BP neural networks, resulting in high evaluation accuracy and strong generalization ability. Attached Figure Description

[0096] Figure 1 A flowchart illustrating the steps of the method provided by this invention;

[0097] Figure 2 A schematic diagram of the system architecture of the method provided by the present invention. Detailed Implementation

[0098] The terms "first," "second," and "third," etc., used in this application specification, claims, and drawings are used to distinguish different objects, not to limit a specific order.

[0099] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0100] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the related technologies is given first:

[0101] Current electrical fire risk assessments mainly rely on the following technical means: First, regular manual inspections, where professionals use infrared thermometers, clamp meters, and other equipment to inspect key parts such as distribution cabinets and cable joints; second, automatic alarm devices based on fixed thresholds, which monitor residual current and temperature parameters in real time by installing residual current and temperature-measuring electrical fire monitoring detectors in the power distribution circuit, triggering an alarm when the monitored values ​​exceed preset thresholds; and third, power quality measurement equipment, used to measure parameters such as harmonic content and three-phase imbalance.

[0102] Research has revealed the following shortcomings in existing technologies: most assessment methods are static, lacking timeliness and failing to reflect dynamic changes in electrical parameters; fixed weights are used for index weight allocation, making it difficult to reflect changes in risk contribution when parameters undergo abnormal changes; multi-circuit index fusion processing uses either the maximum value method or the average value allocation method, with the former easily amplifying local risks and the latter easily losing hidden danger information; and the assessment results only output the risk level, failing to identify key disaster-causing factors.

[0103] To address the above problems, this invention provides an electrical fire risk assessment method based on dynamic weight adjustment. Please refer to [link / reference]. Figure 1 and Figure 2 .

[0104] S1: Constructing an electrical fire risk assessment index system

[0105] Data from the electrical system to be evaluated was collected, and evaluation indicators were selected from four dimensions: electrical factors, environmental factors, spatial factors, and human factors. Electrical factors included 11 secondary indicators: electric arc, current, leakage current, voltage, resistance, conductor insulation, conductor temperature, contact temperature, load, equipment usage status, and abnormal time. Environmental factors included 4 secondary indicators: combustible material type, ambient temperature, ambient humidity, and other environmental factors. Spatial factors included 3 secondary indicators: safety distance, ventilation and heat dissipation conditions, and other spatial factors. Human factors included 3 secondary indicators: installation compliance, usage compliance, and other human factors.

[0106] In one specific embodiment, quantitative indicators such as current, leakage current, voltage, conductor temperature, contact temperature, abnormal time, ambient temperature, ambient humidity, and safety distance are collected in real time by deploying corresponding sensors. Current data is collected through current transformers, leakage current through residual current transformers, temperature data through temperature sensors, and ambient temperature and humidity through environmental monitoring equipment. Qualitative indicators such as conductor insulation, equipment usage condition, types of combustible materials, ventilation and heat dissipation conditions, installation compliance, and usage compliance are scored and assigned values ​​by professionals based on on-site inspection results.

[0107] S2: Determine the initial weights of each evaluation indicator using a combined weighting method.

[0108] Step S2.1: Determine the subjective weights of the indicators using the Analytic Hierarchy Process (AHP).

[0109] Construct the judgment matrix. Using a 9-level scale, construct the judgment matrix by comparing the importance of each element at the same level. Let the evaluation matrix be A:

[0110]

[0111] in, This indicates the importance of the i-th indicator relative to the j-th indicator. , , .

[0112] Normalize the judgment matrix A column by column:

[0113]

[0114] Where n is the order of the judgment matrix.

[0115] Calculate the product of all elements in each row of the normalized matrix. :

[0116]

[0117] beg nth root :

[0118]

[0119] Will Normalization yields the eigenvectors :

[0120]

[0121] This is the subjective weight vector of each indicator.

[0122] Perform a consistency check. Calculate the largest eigenvalue. :

[0123]

[0124] Calculate the consistency index (CI):

[0125]

[0126] Calculate the consistency ratio (CR):

[0127]

[0128] Wherein, RI is the random consistency index. When CR < 0.1, the judgment matrix passes the consistency test.

[0129] Step S2.2: Determine the objective weights of the indicators using the entropy weight method.

[0130] Construct evaluation matrix B. Given m evaluation objects and n evaluation indicators, construct the matrix as follows:

[0131]

[0132] Standardize the indicators. For positive indicators (higher values ​​indicate higher risk):

[0133]

[0134] For negative indicators (the smaller the indicator value, the higher the risk):

[0135]

[0136] Calculate the proportion of the i-th evaluation object to the j-th indicator. :

[0137]

[0138] Calculate the entropy value of the j-th index. :

[0139]

[0140] Calculate the coefficient of difference for the j-th indicator. :

[0141]

[0142] Calculate the entropy weight of the j-th index :

[0143]

[0144] This is the objective weight vector of each indicator.

[0145] Step S2.3: Combine the subjective weights and objective weights according to the principle of minimum information entropy to obtain the combined weights. :

[0146]

[0147] in, Let the combined weight of the j-th indicator be , Let j be the subjective weight of the j-th indicator. Let be the objective weight of the j-th indicator, and n be the total number of indicators.

[0148] S3: Constructing a dynamic weighting strategy for each evaluation indicator

[0149] Let the real-time monitoring value of the j-th indicator at time t be... The risk warning boundary value corresponding to this indicator is The maximum allowed value is .

[0150] For most indicators, their importance in risk assessment should be appropriately increased when the monitored value exceeds the warning threshold. The following variable weighting formula should be used:

[0151] when hour:

[0152]

[0153] when hour:

[0154]

[0155] in, Let the combined weight of the j-th indicator be , Let j be the subjective weight of the j-th indicator. Let j be the objective weight of the j-th indicator. The subjective weight of the k-th indicator. denoted as the objective weight of the k-th indicator, where n is the total number of indicators, and j and k are the indicator numbers, with values ​​ranging from 1, 2, ..., n.

[0156] For indices with abrupt changes (such as voltage), the following weighting formula is used:

[0157] when hour:

[0158]

[0159] in, This is the mutation regulation coefficient, with a value ranging from 1.0 to 3.0.

[0160] The dynamically adjusted weights are then normalized.

[0161]

[0162] in, Let j be the normalized dynamic weight of the j-th index at time t, satisfying .

[0163] In one specific embodiment, taking the current index as an example, the warning boundary value of the current is set to 32A, and the maximum allowable value is 63A. When the real-time monitored current value is 40A, the weighting adjustment coefficient... If we take 1.0, then:

[0164]

[0165] After normalization, the dynamic weight of this current index was adjusted from the initial 0.187 to 0.235, reflecting its increased importance in risk assessment when the current is abnormal.

[0166] S4: Perform weighted fusion processing on multi-loop monitoring data.

[0167] There are N power distribution circuits, and the monitoring value of the j-th indicator in the i-th circuit is... .

[0168] The risk level ranges for each indicator are divided into five levels. These correspond to the following risk states: Normal, Basically Normal, Abnormal, Risk, and Alarm.

[0169] Determine the risk level weight corresponding to the indicator values ​​of each loop. :

[0170]

[0171] Calculate the fusion value of the j-th indicator. :

[0172]

[0173] in, The value of the j-th indicator is obtained by weighted fusion of multiple circuits; N is the total number of distribution circuits; The monitoring value of the j-th indicator in the i-th loop; The risk level weight corresponds to the j-th indicator value of the i-th loop.

[0174] In one specific embodiment, taking a low-voltage power distribution system in a shopping mall as an example, the system includes 4 distribution rooms and 103 shops. Taking current indicators as an example, the current values ​​of the 103 shops are distributed at 10A.

[0175] Within the 45A range. According to the risk level classification standard, current <16A corresponds to level V1 (q=1), 16-32A corresponds to level V2 (q=2), and 32-45A corresponds to level V3 (q=3). The fused current value is calculated using the above weighted fusion formula. This value comprehensively reflects the current risk status of all distribution circuits. Compared with the traditional maximum value method, this method assigns different weights to each circuit according to its risk level, which highlights the impact of high-risk circuits while retaining information about low-risk circuits, thus avoiding the problems of risk amplification or information loss.

[0176] S5: Use the fuzzy comprehensive evaluation method to comprehensively evaluate the fused indicator data.

[0177] Determine the comment set These correspond to the following risk states: Normal, Basically Normal, Abnormal, Risk, and Alarm.

[0178] Construct the membership function. Use a membership function that combines trapezoidal and trigonometric functions. Membership function for level v1:

[0179]

[0180] Membership function for level v2:

[0181]

[0182] The membership functions for levels v3, v4, and v5 are constructed similarly. The membership vectors for each index are then calculated. Construct the fuzzy evaluation matrix R:

[0183]

[0184] Calculate the fuzzy comprehensive evaluation result B:

[0185]

[0186] in, , This is used to assess the membership degree of an object belonging to the kth risk level.

[0187] Risk level is determined based on the principle of maximum membership:

[0188]

[0189] Select multiple sets of indicator data covering different working conditions, repeat the above process, and generate a training dataset containing m sets of input-output pairs. ,in Let p be the index data vector. This is the corresponding risk level encoding vector.

[0190] S6: Construct a BP neural network model and train it.

[0191] Construct a backpropagation (BP) neural network structure, including an input layer, a first hidden layer, a second hidden layer, and an output layer. The number of nodes in the input layer equals the number of evaluation metrics, n. The number of nodes in the first hidden layer... Determined by the following empirical formula:

[0192]

[0193] Where m is the number of output layer nodes, and a is a constant between 1 and 10. (Number of nodes in the second hidden layer) Determined by the following empirical formula:

[0194]

[0195] Where b is a constant between 1 and 10. The number of output layer nodes is equal to the number of risk levels, i.e., m = 5.

[0196] Initialize network parameters by using random numbers within the range (0,1) to initialize the connection weights and thresholds between layers. Normalize the training data using the min-max normalization method.

[0197]

[0198] The network is trained using the backpropagation algorithm. During the forward propagation process, the outputs of the hidden and output layers are:

[0199]

[0200] in, The sigmoid function is used as the activation function.

[0201]

[0202] The output layer uses the Softmax function to convert the output into a probability distribution:

[0203]

[0204] Cross-entropy is used as the loss function:

[0205]

[0206] During backpropagation, gradient descent is used to update weights and thresholds. Training stops and the trained BP neural network model is saved when the training reaches the preset maximum number of iterations or the loss function value falls below the preset threshold.

[0207] S7: Using a trained BP neural network model for electrical fire risk assessment

[0208] Real-time monitoring data of each evaluation indicator for each power distribution circuit is collected. The multi-circuit monitoring data is weighted and fused according to the method in step S4 to obtain the fused indicator data. The dynamic weight of each indicator is calculated according to the method in step S3. The fused indicator data is input into a trained BP neural network model, and the model outputs the probability distribution of each risk level. The risk level with the highest probability is taken as the current electrical fire risk assessment result.

[0209] In one specific embodiment, measured data from a shopping mall's low-voltage power distribution system are used as an example. The fused index data vector is collected. The dynamic weight vector is calculated. When the above data is input into a BP neural network model, the probability distribution output by the model is as follows: , , , , Based on the principle of maximum membership, the risk level v3 (abnormal) with the highest probability was selected as the assessment result. This result indicates that there are certain electrical fire hazards in the low-voltage power distribution system of the shopping mall, which is basically consistent with the on-site inspection results (abnormal phenomena such as overheated terminals, damaged wires, and poor contact in the power distribution room).

[0210] S8: Identifying Key Risk Factors Based on Bayesian Backward Reasoning

[0211] The risk level assessed in step S7 is used as evidence input into the Bayesian network. The structure of the Bayesian network is consistent with the hierarchical structure of the assessment index system: the target layer node (electrical fire risk) is the top-level parent node, the first-level index nodes (electrical factors, environmental factors, spatial factors, human factors) are the intermediate-level nodes, and the second-level index nodes (each assessment index) are the bottom-level child nodes.

[0212] The electrical fire risk status is set to the highest risk level. Backward inference based on Bayesian networks is used to calculate the posterior probability of each indicator node:

[0213]

[0214] in, This refers to the j-th indicator node; In a risky state Under the condition that the j-th indicator is in The posterior probability of a state.

[0215] The indicators are ranked according to their posterior probability to identify the key factors that have the most significant impact on electrical fire risk.

[0216] In one specific embodiment, the electrical fire risk state is set to v5 (alarm), and backpropagation is performed based on a Bayesian network. The backpropagation results show that electrical factors are the core drivers of the electrical fire risk state, and their sensitivity is significantly higher than that of human factors, environmental factors, and spatial factors. Among electrical factors, electric arc and leakage current are the most sensitive indicators. Therefore, in daily electrical fire prevention and control work, the focus should be on controlling these three core sensitive factors: electric arc, leakage current, and abnormal current.

[0217] This invention also provides an electrical fire risk assessment system based on dynamic weight adjustment, comprising: a data acquisition module for real-time acquisition of monitoring data of each assessment indicator of each power distribution circuit; a data preprocessing module for cleaning, filtering, and standardizing the acquired data; a multi-circuit fusion module for weighted fusion processing of multi-circuit monitoring data according to the method in step S4; a dynamic weight calculation module for calculating the dynamic weight of each indicator based on the real-time monitoring data according to the method in step S3; a risk assessment module with a built-in trained BP neural network model for outputting electrical fire risk assessment results based on the fused indicator data and dynamic weights; a key factor identification module for identifying key risk factors based on Bayesian network back reasoning; an early warning output module for outputting risk levels and early warning information based on the assessment results; and a display module for displaying the real-time monitoring data, dynamic weights, risk assessment results, and key factor information of each indicator.

[0218] In one specific embodiment, the data acquisition module includes sensors deployed in each power distribution circuit, including current transformers, voltage transformers, leakage current transformers, temperature sensors, and arc sensors. The data preprocessing module filters, denoises, and standardizes the acquired data. The multi-circuit fusion module calculates the fusion value based on the risk level of each indicator in each circuit. The dynamic weight calculation module calculates and normalizes the dynamic weights based on the real-time monitoring values ​​of each indicator. The BP neural network model built into the risk assessment module, trained in step S6, can output risk assessment results based on the input fused indicator data and dynamic weights. The key factor identification module, with its built-in Bayesian network model, can identify key risk factors based on back-reasoning. The early warning output module outputs the risk level (normal, basically normal, abnormal, risk, alarm) and corresponding early warning information based on the assessment results. The display module visually displays the real-time monitoring data of each indicator, dynamic weight allocation, risk assessment results, and key factor information.

[0219] This invention achieves dynamic, objective, and accurate assessment of electrical fire risks through the aforementioned technical solutions. Initial weights are determined using a combined weighting method, balancing subjective judgment from expert experience with objective information from sample data. A dynamic weighting strategy adjusts weights based on real-time monitoring data, ensuring the assessment results reflect changes in the operational status of electrical circuits in real time. A multi-circuit index fusion method based on risk level weighting avoids the problem of amplifying local risks using the maximum value method while overcoming the deficiency of the average value method in obscuring hazard information. Key risk factors are identified through Bayesian back-reasoning, providing targeted support for subsequent fire prevention and control decisions.

[0220] 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 illustrative of the principles of 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 claims and their equivalents.

Claims

1. A method for assessing electrical fire risk based on dynamic weight adjustment, characterized in that, The method includes: An electrical fire risk assessment index system was constructed, and assessment indicators were selected from four dimensions: electrical factors, environmental factors, spatial factors, and human factors. The initial weights of each evaluation indicator are determined by a combination weighting method, including the use of the analytic hierarchy process (AHP) to determine the subjective weights of the indicators, the use of the entropy weight method to determine the objective weights of the indicators, and the combination of the subjective and objective weights to obtain the combined weights based on the principle of minimum information entropy. Construct a dynamic weighting strategy for each evaluation indicator and dynamically adjust the weights based on real-time monitoring data of the indicators; The multi-loop monitoring data is weighted and fused, and different weights are assigned to each indicator of each loop for weighted fusion according to the risk level of each loop. The fuzzy comprehensive evaluation method is used to comprehensively evaluate the fused index data and generate a training dataset. Construct and train a BP neural network model; Using a trained BP neural network model to assess electrical fire risk; Identifying key risk factors based on Bayesian reverse reasoning.

2. The method according to claim 1, characterized in that, The method of determining the subjective weights of indicators using the analytic hierarchy process includes: constructing a judgment matrix by using a 9-level scaling method to establish the judgment matrix by comparing the importance of different elements at each level; normalizing the judgment matrix column by column; calculating the product of all elements in each row of the normalized matrix, finding the nth root of the product and normalizing it to obtain the eigenvector, which is the subjective weight vector; and performing a consistency test by calculating the consistency ratio CR. When CR < 0.1, the judgment matrix passes the consistency test.

3. The method according to claim 1, characterized in that, The method of determining the objective weight of indicators using entropy weight includes: constructing an evaluation matrix with m evaluation objects and n evaluation indicators; standardizing the indicators, using different standardization formulas for positive and negative indicators; calculating the proportion of the i-th evaluation object under the j-th indicator; calculating the entropy value of the j-th indicator; calculating the difference coefficient of the j-th indicator; and calculating the entropy weight of the j-th indicator based on the difference coefficient.

4. The method according to claim 3, characterized in that, The subjective and objective weights are combined according to the principle of minimum information entropy to obtain the combined weight. : in, Let the combined weight of the j-th indicator be , Let j be the subjective weight of the j-th indicator. Let j be the objective weight of the j-th indicator. The subjective weight of the k-th indicator. is the objective weight of the k-th indicator, n is the total number of indicators, and j and k are the indicator numbers, with values ​​ranging from 1, 2, ..., n.

5. The method according to claim 1, characterized in that, The dynamic adjustment of weights based on real-time monitoring data of the indicators includes: When monitoring value When this is the case, the following variable weight formula is used: When monitoring value hour, in, Let j be the dynamic weight of the j-th index at time t. Let α be the initial weight of the j-th indicator combination, and let α be the weight adjustment coefficient, with a value ranging from 0.5 to 2.

0. Let be the warning boundary value for the j-th indicator; This represents the maximum allowed value for the j-th indicator.

6. The method according to claim 1, characterized in that, For indicators with mutation characteristics, when the monitored value ≥ When this is the case, the following variable weight formula is used: Wherein, β is the mutation regulation coefficient, with a value ranging from 1.0 to 3.

0.

7. The method according to claim 1, characterized in that, The dynamically adjusted weights are then normalized. in, Let j be the normalized dynamic weight of the j-th index at time t, satisfying .

8. The method according to claim 1, characterized in that, The weighted fusion processing of multi-loop monitoring data includes: dividing the risk level range of each indicator into five levels {v1, v2, v3, v4, v5}, corresponding to risk states: normal, basically normal, abnormal, risk, and alarm; and determining the risk level weight corresponding to the indicator value of each loop. : Calculate the fusion value of the j-th indicator. : in, The value of the j-th indicator is the weighted fusion value after multiple circuits, where N is the total number of distribution circuits. Let j be the monitoring value of the j-th indicator in the i-th loop. The risk level weight corresponds to the j-th indicator value of the i-th loop.

9. The method according to claim 1, characterized in that, The fuzzy comprehensive evaluation method includes: constructing a membership function, using a combination of trapezoidal and trigonometric functions; calculating the membership vector of each index; constructing a fuzzy evaluation matrix; and calculating the fuzzy comprehensive evaluation result B. in, To assess the membership degree of the object to the k-th risk level, the risk level is determined based on the principle of maximum membership.

10. The method according to claim 1, characterized in that, The construction and training of the BP neural network model includes: constructing the BP neural network structure, including an input layer, a first hidden layer, a second hidden layer, and an output layer; the number of nodes in the first hidden layer... Determined by the following empirical formula: Where m is the number of output layer nodes, and a is a constant between 1 and 10; the number of nodes in the second hidden layer. Determined by the following empirical formula: Where b is a constant between 1 and 10; the network is trained using the backpropagation algorithm, and the outputs of the hidden layer and the output layer during the forward propagation process are: in, Let v be the connection weight from the v-th node in layer l (l-1) to the u-th node in layer l. Let f(·) be the threshold for the u-th node in the l-th layer, and f(·) be the activation function, using the Sigmoid function: The output layer uses the Softmax function to convert the output into a probability distribution: Cross-entropy is used as the loss function: 。