Building construction safety risk assessment method and system

Abstract

The invention discloses a building construction safety risk assessment method and system, and relates to the technical field of building construction safety management, and the method comprises the steps: recognizing a plurality of risk factors affecting the construction safety based on multi-source construction data collected at a construction site, constructing a multi-layer fuzzy ANP network based on the fuzzy dependency relationship between the risk factors, and carrying out the construction safety assessment of the multi-layer fuzzy ANP network. Calculating the fuzzy weight of each risk factor; the multi-layer fuzzy ANP network is converted into a Bayesian graph, and initial conditional probability of each risk factor is generated by using fuzzy weight; constructing a failure propagation path set of the risk factors in the Bayesian network, and calculating the propagation probability and accumulated risk intensity of each path in combination with the preset severity level of each risk factor; and in combination with real-time construction data, dynamically correcting the conditional probability and the severity level in the Bayesian network, updating the sorting result of the failure paths and the risk factors, and outputting the risk assessment result of the current construction stage. The real-time assessment of the construction safety risk is realized, and the precision of the assessment result is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of building construction safety management, and in particular to a method and system for assessing building construction safety risks. Background Art

[0002] With the continuous expansion of construction projects and the increasing complexity of construction site environments, safety risk management during construction processes faces increasingly severe challenges. Traditional construction safety risk assessment methods are mostly based on static scoring models or empirical judgments, relying on manually formulated indicator systems and scoring rules. These methods fail to fully reflect the interdependencies and dynamic evolution of multiple risk factors. In actual construction, risk factors are often uncertain, multi-source, and contagious. Especially in situations involving the coordination of multiple trades, the interconnectedness of equipment, and frequent environmental disturbances, the activation of local risks can trigger systemic safety incidents through chain reactions.

[0003] In recent years, the analytic hierarchy process (AHP) and analytic network process (ANP) have been used to some extent in safety risk assessment. However, traditional ANP models lack the ability to characterize the fuzzy dependencies between risk factors and fail to reflect the impact of real-time data on risk status during the construction phase. Furthermore, while existing Bayesian networks have some causal modeling capabilities, most studies are limited to inferential analysis within static structures and fail to integrate multidimensional fuzzy weights. They also lack structural modeling and prioritization mechanisms for risk propagation pathways, making it difficult to support real-time decision-making and proactive early warning. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing construction risk assessment methods have problems such as insufficient accuracy, unclear structural expression and delayed response in factor dependency modeling, risk propagation path analysis and real-time dynamic updating.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for assessing construction safety risks, comprising:

[0007] Based on multi-source construction data collected at the construction site, multiple risk factors affecting construction safety are identified. A multi-layer fuzzy ANP network is constructed based on the fuzzy dependency relationships between the risk factors, and the fuzzy weights of each risk factor are calculated.

[0008] The multi-layer fuzzy ANP network is converted into a Bayesian graph, and the fuzzy weights are used to generate the initial conditional probabilities of each risk factor;

[0009] A set of failure propagation paths of risk factors is constructed in the Bayesian network, and the propagation probability and cumulative risk intensity of each path are calculated in combination with the preset severity level of each risk factor;

[0010] In combination with real-time construction data, the conditional probability and severity level in the Bayesian network are dynamically corrected, the ranking results of the failure paths and risk factors are updated, and the risk assessment results of the current construction stage are output.

[0011] As a preferred scheme of the building construction safety risk assessment method, the multi-source construction data includes environmental monitoring data, operation behavior data, equipment operation data, construction plan and progress data, and historical risk and accident records.

[0012] Based on the statistical characteristics of the multi-source construction data, a high-risk state of the construction site is identified; according to a preset mapping rule, the identified high-risk state is mapped to a standardized risk type, and a corresponding risk factor is generated.

[0013] As a preferred scheme of the building construction safety risk assessment method, the multi-layer fuzzy ANP network is constructed, including dividing the identified risk factors into a target layer, a risk category layer and a risk factor layer according to the overall evaluation target of construction safety; based on the construction task process and expert experience, a pair of risk factors with mutual influence is determined, a bidirectional connection relationship is established, and a multi-layer fuzzy ANP network is generated.

[0014] As a preferred scheme of the building construction safety risk assessment method, the calculation of the fuzzy weight of each risk factor includes: for any pair of risk factors with a dependency relationship, triangular fuzzy numbers are used for pairwise comparison to construct a fuzzy judgment matrix; based on the fuzzy judgment matrix, the fuzzy local weight of each risk factor is calculated using the fuzzy geometric mean method, and the local fuzzy weight vector is embedded into the corresponding sub-block structure of the super matrix of the multi-layer fuzzy ANP network to form an overall fuzzy super matrix.

[0015] After performing column normalization processing on the overall fuzzy super matrix to construct a column random fuzzy super matrix, the weighted update is performed through continuous power multiplication operation until the preset convergence precision threshold is reached, and the global fuzzy weight vector of each risk factor is obtained; the global fuzzy weight vector is de-fuzzied, the weighted average operation is performed on each fuzzy weight using the center method, the global fuzzy weight vector is converted into a determined value, and the converted value is normalized to obtain the fuzzy weight of each risk factor.

[0016] As a preferred scheme of the building construction safety risk assessment method, wherein: the Bayesian graph comprises a set of graph nodes with risk factors as nodes; all pairs of risk factors with non-zero fuzzy weights are traversed to establish directed edges between corresponding graph nodes in the direction of the fuzzy weights;

[0017] The generating of the initial condition probability of each risk factor comprises: for a risk factor without parent nodes in the Bayesian directed acyclic graph, generating a prior probability through a linear scaling mapping function according to the corresponding value in the global fuzzy weight vector, the mapping function being ; wherein, represents a mapping adjustment coefficient, represents the fuzzy weight value of the corresponding risk factor, represents the prior probability value, represents that the upper limit of the limit probability is not more than 1; for a risk factor with parent nodes, based on the fuzzy weights between all parent nodes and the target risk factor, the condition probability value is generated by first using the center method to defuzzify and then using a weighted average method, and a condition probability table is constructed; the corresponding condition probability value is generated for each value combination of all parent nodes by enumerating all value combinations of the parent nodes; and the condition probability values under each condition are normalized to form the condition probability table.

[0018] As a preferred scheme of the building construction safety risk assessment method, wherein: the constructing of the failure propagation path set comprises taking a risk factor without parent nodes as a path starting point, traversing all graph nodes with directed edge connection relationship, and generating an ordered path sequence composed of multiple risk factors; each ordered path sequence is taken as a failure propagation path and is collected into the failure propagation path set.

[0019] The calculating of the propagation probability and the cumulative risk intensity of each path comprises: obtaining the preset severity level of each risk factor in each failure propagation path, and multiplying the condition probability values of all directed edges in the path to calculate the propagation probability of the path; the propagation probability and the severity level of each risk factor in the path are weighted and superimposed to obtain the cumulative risk intensity of the corresponding path; and the failure propagation path set is sorted in descending order according to the cumulative risk intensity of all paths, and the path priority result of the current stage is output.

[0020] As a preferred scheme of the building construction safety risk assessment method, wherein: the condition probability and severity level in the Bayesian network are corrected, the real-time construction data is corresponded to the established risk factor nodes in the Bayesian graph according to a preset mapping rule, and the condition probability table of the related risk factors is updated; on the basis of updating the condition probability table, the severity level of the corresponding risk factors is dynamically adjusted in combination with the risk factor content involved in the real-time construction data; based on the corrected condition probability table and severity level, the propagation probability and cumulative risk intensity of each path in the failure propagation path set are recalculated, and all paths are updated in descending order; and the dynamic risk assessment result of the current construction stage is output according to the sorting result.

[0021] As a preferred scheme of the building construction safety risk assessment system, wherein: the fuzzy network module is used to identify a plurality of risk factors affecting construction safety based on the multi-source construction data collected on the construction site, and a multi-layer fuzzy ANP network is constructed based on the fuzzy dependency relationship between the risk factors, and the fuzzy weight of each risk factor is calculated;

[0022] The Bayesian graph module is used to convert the multi-layer fuzzy ANP network into a Bayesian graph, and generate the initial condition probability of each risk factor by using the fuzzy weight;

[0023] The risk module is used to construct a failure propagation path set of the risk factors in the Bayesian network, and calculate the propagation probability and cumulative risk intensity of each path in combination with the preset severity level of each risk factor;

[0024] The correction module is used to dynamically correct the condition probability and severity level in the Bayesian network in combination with the real-time construction data, update the sorting result of the failure path and the risk factor, and output the risk assessment result of the current construction stage.

[0025] A computer device includes a memory and a processor, the memory stores a computer program, and the processor executes the computer program to implement the steps of the building construction safety risk assessment method.

[0026] A computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the building construction safety risk assessment method.

[0027] Beneficial effects of the present invention: The construction safety risk assessment method provided by the present invention collects multi-source construction data such as environmental monitoring, work behavior, and equipment status, and identifies high-risk factors in combination with task processes and expert experience; uses triangular fuzzy numbers to construct a risk factor dependency matrix, and adopts fuzzy geometric averaging and continuous power multiplication update algorithms to accurately obtain global fuzzy weights; constructs a directed acyclic graph containing node causal dependencies, and performs topological sorting and pruning operations to ensure that the Bayesian structure is stable and inferable; introduces defuzzification and linear scaling mechanisms to generate conditional probability tables, and realizes accurate causal quantification between risk factors; enumerates the departure paths of all starting nodes and multiplies them with conditional probabilities, calculates the cumulative risk intensity in combination with the severity level, and outputs the sorted risk propagation path; during the construction process, real-time observation data is used to correct the risk conditional probability and severity level, and dynamically updates the path priority, thereby realizing active safety assessment and risk pre-intervention in continuous construction stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0029] Figure 1 This is an overall flow chart of a construction safety risk assessment method provided by the first embodiment of the present invention. DETAILED DESCRIPTION

[0030] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0031] Example 1, with reference to Figure 1 , as one embodiment of the present invention, provides a method for assessing construction safety risks, comprising:

[0032] S1: Based on the multi-source construction data collected at the construction site, multiple risk factors affecting construction safety are identified. A multi-layer fuzzy ANP network is constructed based on the fuzzy dependency relationship between the risk factors, and the fuzzy weight of each risk factor is calculated.

[0033] First, obtain multi-source construction data sets from multiple types of sensing terminals deployed on the construction site, including: environmental monitoring data (such as temperature and humidity, noise, PM2.5 concentration, etc.); work behavior data (such as personnel positioning trajectory, work frequency, violation behavior record, etc.); equipment operation data (such as the operation condition of tower crane, excavator, hoist, etc.); construction plan and progress data (such as the current stage task type, execution deviation rate); historical risk and accident records (including the trigger factor and consequence label of typical cases).

[0034] Based on the above data, statistical characteristic indicators (mean, range, coefficient of variation, etc.) are calculated to construct a high-risk state detection model. If multiple indicators exceed the preset threshold in succession, the system maps this state to a specific standardized risk type. For example: if the temperature is >40℃ and the personnel work time is more than 3 hours, it is mapped to "high temperature fatigue" risk; if the equipment vibration amplitude is abnormal + cumulative working hours exceed 500 hours, it is mapped to "mechanical failure risk".

[0035] The risk type set formed after mapping is denoted as , which is used as the input of risk factors in subsequent modeling. Among them, represents an independent construction safety risk factor, represents the total number of risk factors.

[0036] To calculate the relative importance of each risk factor, a multi-layer fuzzy ANP network is constructed, with the following structure division:

[0037] Target layer: overall risk assessment of construction safety.

[0038] Middle layer: risk categories (such as personnel, environment, equipment, organization and management).

[0039] Bottom layer: risk factors (i.e. identified ).

[0040] Introduce mutual connection relationship between risk factors to form an interconnected network. For factor pair with influence relationship, triangular fuzzy numbers are used for pairwise comparison to construct fuzzy judgment matrix , where each element is:

[0041]

[0042] where represents the triangular fuzzy judgment value of risk factor relative to risk factor ; represents the lower limit of the judgment value, i.e. the least possible importance; represents the median of the judgment value, i.e. the most likely importance; Indicates the upper limit of the judgment value, that is, the highest possible degree of importance.

[0043] The risk factor is calculated using the fuzzy geometric mean method. The fuzzy local weight vector , the calculation formula is:

[0044]

[0045] in, Represents risk factors The fuzzy local weight of Indicates the Take the product of all fuzzy judgment values ​​in the row; Represents the total number of risk factors.

[0046] Embed all local fuzzy weight vectors into the corresponding positions of the fuzzy super matrix to construct the super matrix . Then, the super matrix Perform column normalization to obtain the column random fuzzy super matrix . Perform successive exponentiation operations on this matrix: Until the preset convergence threshold is met ,Right now , get the final global fuzzy weight vector In order to make the weights available for the subsequent probability construction in the Bayesian network, the fuzzy numbers need to be defuzzified. Calculate its clarity value , and then for all Perform normalization to obtain the normalized , and finally output the standardized fuzzy weight vector , as the input basis for subsequent probabilistic network models.

[0047] in, Indicates the The fuzzy supermatrix of the iteration (appears in the third formula) Indicates the The fuzzy super matrix of the iteration; represents the normalized column random fuzzy supermatrix; ε represents the convergence accuracy threshold; , w represents the clear value corresponding to the fuzzy triangular number; Respectively represent the lower limit, median and upper limit of the fuzzy number; Represents the normalized The certainty weight of each risk factor; represents the normalized fuzzy weight vector of all risk factors; Indicates the certainty weight value corresponding to each risk factor.

[0048] In the process of construction safety assessment, the common practice is to rely on single source sensor data or expert experience scoring, which not only limits the identification of risk factor dimensions, but also has the problem of slow response to risk state. In view of this limitation, the current method first introduces a multi-source data parallel collection mechanism, integrating five types of data content such as environmental monitoring, behavior recognition, equipment status, progress plan and accident history. Different types of data are collected on a unified time axis to avoid information silos and improve the time resolution and coverage of risk factor identification. Through statistical feature extraction and risk event matching rules, typical risk states can be dynamically extracted from the original data and mapped to a standardized risk factor set, ensuring the quantification and standardization of risk factors.

[0049] Traditional multi-level analysis structure (such as AHP, hierarchical fuzzy method) mostly uses one-way hierarchical structure, which cannot reflect the interaction and feedback influence between risk factors. This scheme uses a three-layer structure to construct a risk factor network, corresponding to the evaluation target layer, risk category layer and basic factor layer, and identifies the influence direction between risk factors through expert judgment to build a two-way dependent structure. Under this structure, risk factors not only receive information from the upper layer, but also feedback their influence, realizing the ring-like coupling representation of the risk network. Further mapping the above structure to a fuzzy ANP network introduces a super matrix form to express the influence path between multiple layers, overcoming the limitations of traditional linear hierarchical models in modeling complex relationships between factors.

[0050] To achieve high-precision weight quantization, fuzzy mathematical methods are embedded in the structure. By introducing triangular fuzzy numbers to construct pairwise judgment matrices, the relative importance between each pair of factors is described as an uncertain interval with upper and lower limits, preserving the fuzziness of expert judgment. In the calculation process, fuzzy geometric mean method is used to extract local fuzzy weight vectors, which are further embedded into super matrix sub-blocks to complete the fuzzy modeling of the overall network. To ensure global consistency, column normalization and continuous power multiplication methods are introduced for multiple rounds of iterative calculation of the network structure, and finally stable fuzzy weight results are obtained. In the final stage, the central method is used to de-fuzzify the fuzzy weights, converting them into deterministic probability values that can be used for subsequent Bayesian modeling. Compared with conventional quantitative scoring methods, this processing method can more fully preserve the nonlinear influence relationships between complex risk factors, and has strong mathematical consistency and convergence verification standards, significantly improving the stability and precision of risk weight evaluation.

[0051] S2: Convert the multi-level fuzzy ANP network to a Bayesian graph, and use the fuzzy weights to generate the initial conditional probability of each risk factor.

[0052] Take each of the risk factor sets ​ A set of nodes is constructed as nodes in the graph. All pairs of risk factors with non-zero fuzzy weights in the fuzzy ANP network are traversed , and if there is a dependency relationship, a directed edge is established between the nodes and in the direction of the weight . To ensure that the Bayesian graph satisfies the Directed Acyclic Graph (DAG) structure, topological sorting and loop detection operations need to be performed on the above graph structure. If a loop is detected, the edges from upper layer factors to lower layer factors are preferentially retained, and for the loop paths formed between nodes in the same layer, pruning processing is performed to ensure the legality of the graph structure.

[0053] For the constructed Bayesian graph structure, the initial condition probability needs to be generated for each risk factor node , including the following two cases:

[0054] 1. Prior probability calculation of risk factors without parent nodes: for risk factors without parent nodes , the prior probability is generated by the linear scaling mapping function from the global fuzzy weight after defuzzification :

[0055]

[0056] wherein represents the prior probability of the risk factor ; represents the global fuzzy weight after defuzzification of the risk factor ; represents the mapping adjustment coefficient; represents limiting the mapping value within the interval [0, 1] to avoid exceeding the legal range of probability.

[0057] 2. Construction of conditional probability table for risk factors with parent nodes: for risk factors with parent nodes , the conditional probability needs to be calculated in combination with the fuzzy weights between all parent nodes and the risk factor. The process is as follows:

[0058] For each parent-child risk factor pair , take its fuzzy weight , and perform defuzzification processing using the center method to obtain the defuzzification value . Based on the influence of all parent nodes on the risk factor , the weighted average method is used to generate the conditional probability value Parents). Enumerate all parent node state combinations to construct the conditional probability table structure, and perform normalization processing on each group of conditional probabilities to ensure that the sum of the conditional probabilities under each state is 1.

[0059] wherein, represents a risk factor to a risk factor fuzzy weight; represents a weight value after defuzzification processing; represents a target risk factor; Parents represents a joint state of all parent nodes; represents a conditional probability value of a risk factor under the condition of a given parent node state; represents a certain state value of a parent node .

[0060] The existing risk assessment method often relies on artificial judgment or static structure design when constructing a Bayesian network, which is easy to lead to graph structure redundancy or loop, and cannot effectively support subsequent causal reasoning. The method has generated a fuzzy weight relationship of risk factors in the previous stage, and uses the non-zero dependency relationship as the basis for graph edge construction to realize automatic derivation of the structure. Taking the risk factor set as the graph node, the directed edge is established according to the fuzzy weight direction, and after the preliminary graph structure is constructed, the structure legality is verified through topological sorting and loop detection. For the detected loop, the influence direction from the upper layer to the lower layer is preferentially retained, and the same layer or downward non-causal significance connection is automatically pruned to ensure that the final graph structure meets the mathematical requirements of directed acyclic graph. The structure standardization process provides a verifiable basic framework for subsequent probability propagation and causal reasoning, and improves the overall engineering usability and logical closure of the model.

[0061] For the risk factors with parent nodes, a complete conditional probability table needs to be constructed to support Bayesian reasoning. First, obtain the fuzzy weight between all parent nodes and the target node, use the center method to defuzzify the fuzzy weight into specific values, and then generate the conditional probability results under different inputs according to the combination state of each parent node using weighted average method. Enumerate all state combinations of the parent nodes, calculate the conditional probability value of the target node for each combination, and perform normalization processing on all values to ensure that the probability sum under each condition is 1, thereby forming a complete CPT table structure that meets the specifications. Such processing not only retains important features in fuzzy information, but also can accommodate complex dependency relationships of any number of parent nodes, improving the adaptability and logical consistency of risk propagation modeling.

[0062] S3: Constructing a set of failure propagation paths of risk factors in the Bayesian network, combining the preset severity level of each risk factor, calculating the propagation probability and cumulative risk intensity of each path.

[0063] The system performs directed path traversal on all graph nodes in the Bayesian network graph with the risk factor without parent node as the path starting point, and generates an ordered path sequence composed of multiple risk factors in topological order. Each path extending from the starting node to the downstream node along the directed edge is regarded as an effective failure propagation path, and all paths form a set denoted as:

[0064]

[0065] wherein each path is an ordered sequence composed of a group of risk factors:

[0066]

[0067] wherein, represents the th node belonging to the risk factor set; represents the failure propagation path represents the total number of failure paths in the path set; represents the th risk factor node appearing in the path represents the directed edge between nodes in the Bayesian network; represents the number of risk factors contained in the path.

[0068] For each path , the overall propagation probability thereof is calculated according to the conditional probability values of all directed edges in the path. Let the conditional probability between adjacent risk factor pairs in the path be:

[0069]

[0070] the propagation probability of the path is expressed as:

[0071]

[0072] wherein, represents the propagation probability of the path ; represents the continuous multiplication symbol; represents the conditional probability of the th edge (from ); is the index of the current edge in the path, from 1 to ; represents the conditional probability; represents the current node; represents the successor node. ​​

[0073] Further combine the preset severity levels of each risk factor in the path. Let the severity level of each risk factor in the path be Then the cumulative risk intensity of the path is Expressed as:

[0074]

[0075] Wherein , represents the severity level of the risk factor , the closer to 1 means the more serious; represents the cumulative risk intensity of the path.

[0076] All paths are sorted in descending order as the path priority ranking results of the current stage, which are used for subsequent risk intervention and scheduling decisions.

[0077] Existing risk reasoning systems usually only focus on static probability inference at the node level, lack of systematic mining of risk propagation paths, making it difficult for the evaluation results to reflect the cumulative effect of causal chains. To avoid this problem, a path enumeration mechanism is designed with the parentless factor as the starting point, which automatically identifies all factor-level connection sequences by traversing all directed edges in the Bayesian graph. Each path consists of multiple nodes, forming an ordered set of propagation paths, each path representing a potential risk chain of influence. This mechanism makes risk evolution no longer limited to single-node state changes, but reflects systemic risk through the structure and state changes of the entire path, solving the evaluation dilemma of fragmented causality and one-sided reasoning in the past.

[0078] After the path set is established, the occurrence probability of each path needs to be evaluated to identify potential high-risk chain structures. In terms of calculation logic, each path contains multiple directed connection edges, according to the conditional probability table constructed in the Bayesian graph, the conditional probability value of each connection edge is extracted, and the multiplication operation is performed in the order of the path, thus obtaining the propagation probability of the entire path. This structured propagation calculation method is more holistic than traditional node-level probability analysis, and can truly reflect the joint effect of multi-risk factor interaction propagation, avoiding the distortion or isolated conclusions that may be caused by single-edge probability inference.

[0079] ​​On the basis of the propagation probability calculation, the cumulative risk intensity evaluation model is further constructed by combining the severity level of each risk factor in the path. The risk intensity of each path is determined by the propagation probability and the severity of each risk node. The two are effectively combined by using the segmented weighting method to reflect whether the risk passes through the high-risk factor node in the propagation process and whether the overall risk has a superposition effect. After completing the intensity evaluation of all paths, the path set is sorted in descending order to obtain the risk propagation priority result, which provides a basis for dynamic risk assessment in the construction phase. This mechanism can superimpose the severity dimension on the basis of propagation quantification, realize the transition of risk assessment from "possibility" to "impact", and make the final output result more in line with the decision-making needs of the engineering site.

[0080] S4: Combine real-time construction data to dynamically correct the conditional probability and severity level in the Bayesian network, update the ranking results of failure paths and risk factors, and output the risk assessment results of the current construction phase.

[0081] The data set collected in real time on the construction site is defined as:

[0082]

[0083] wherein each contains the risk factor at time . represents the real-time data set collected on the construction site; represents a single real-time observation data at time . represents the total real-time observation time;

[0084] Based on the mapping rule, the corresponds to the risk factor set:

[0085]

[0086] The real-time state set is obtained:

[0087]

[0088] wherein represents the current observation state of factor . represents the risk factor state set corresponding to the real-time construction state.

[0089] For each risk factor with a parent node, the conditional probability table is corrected. If the original conditional probability value is , then is calculated by normalization from the following formula:

[0090]

[0091] wherein, represents the risk factor the first condition state in the modified probability value; represents the frequency of the risk factor in the historical data in the first state; represents the adjustment coefficient of the conditional probability; represents the scoring function for the current state ; and represents the total number of states of the risk factor .

[0092] The original severity level of the risk factor is dynamically adjusted to calculate a new severity

[0093]

[0094] wherein, is a severity adjustment coefficient represents the dynamically adjusted severity level of the risk factor ; and represents the original preset severity level of the risk factor . represents the impact score function of the risk factor based on the real-time data set .

[0095] For each path , the propagation probability is recalculated

[0096]

[0097] The cumulative risk intensity is recalculated

[0098]

[0099] wherein, is the global ambiguity weight after de-ambiguating and normalizing; represents the propagation probability of the risk path ; and represents the first ordered failure propagation path; represents the node to the modified conditional probability of the risk factor; representing the cumulative risk intensity of the path representing the modified severity level of the risk factor in the path.

[0100] Finally, all the values are sorted in descending order to form a risk path priority sequence and output as the dynamic risk assessment of the current stage.

[0101] The traditional Bayesian network generates a static conditional probability table based on historical data at the beginning of modeling. Once the construction environment changes, the network inference result is easily distorted. To improve this problem, a conditional probability updating mechanism linked with real-time construction data is introduced. According to the pre-prepared risk factor mapping rules, the current collected construction parameters are matched to the corresponding nodes in the network, and the conditional probability table modification operation is automatically triggered. This modification operation does not depend on manual intervention, but adjusts the probability distribution in each state in proportion to the difference between historical and current data, ensuring that the network structure has adaptive adjustment capability when new information is input, so that the risk assessment result always fits the actual situation on site.

[0102] In the traditional model, the severity level of the risk factor is often fixed and set before the assessment, which cannot reflect the dynamic evolution characteristics of the risk state. To address this deficiency, an automatic adjustment mechanism for the severity level is introduced. When new risk characteristics appear in real-time data or the state of the identified risk factor intensifies, the system will increase the level value of the corresponding factor according to the pre-set severity level update rule, and vice versa. This adjustment mechanism is executed in conjunction with the conditional probability table to avoid logical errors caused by input updates but static levels, improving the synchronization and accuracy of the risk output results and the site state.

[0103] Once the conditional probability table and the severity level are updated, the system will immediately recalculate the probability multiplication and intensity weighting for all failure propagation paths, automatically calculating new path propagation values and risk intensity values. The evaluation results of all paths will be updated in the sorting mechanism to form a new round of risk sorting results that are completely aligned with the real-time input state. This processing method ensures that each data refresh receives complete closed-loop feedback at the output end, enhancing the continuity and timeliness of the evaluation system and making the risk warning proactive and visible, providing more accurate data support for risk control on the construction site.

[0104] Embodiment 2, which is an embodiment of the present application, provides a method for evaluating the safety risk of building construction. In order to verify the beneficial effects of the present application, economic benefit calculation and simulation experiments are used for scientific demonstration.

[0105] ​The actual application test was carried out in a subway construction project. The project is located in the core area of the city, with complex construction environment, various risk factors and great representativeness. First of all, based on the environmental monitoring equipment installed on the site (such as PM2.5, wind speed, humidity sensor), operation behavior tracking camera, equipment operation log collection module, construction progress record platform and accident history database, 30 days of multi-source construction data were collected. The data collection frequency is unified as 10 minutes once, and about 4320 complete data records are formed.

[0106] Subsequently, based on the above data, eight types of high-frequency risk states were identified through statistical analysis method, such as "high temperature and humidity + continuous operation" "night concrete pouring + equipment frequent start-stop" and other scenes, and according to the mapping rules formulated by experts, they were classified into standardized risk factors, such as environment class ( ), operation behavior class ( ), equipment state class ( ), plan delay class ( ) and so on.

[0107] Based on these standardized risk factors, a three-layer fuzzy ANP network was constructed, in which the target layer was "total target of construction safety", the risk category layer included "environmental risk" and "behavioral risk", and the bottom layer was the specific risk factor. By introducing the expert team to use triangular fuzzy numbers for pairwise judgment, a fuzzy judgment matrix was constructed, and fuzzy geometric mean method and continuous power multiplication algorithm were used to process the super matrix, and finally the global fuzzy weight of each risk factor was obtained. The fuzzy weight is de-fuzzied and normalized, and then input into the Bayesian network to construct the initial structure.

[0108] On this basis, the fuzzy weight non-zero value is used to establish the directed edge of the Bayesian graph, and the initial structured model is obtained. The prior probability of the factor without parent node is generated by linear scaling function, and the conditional probability table of the node with parent node is calculated by center method and weighted average. In order to verify the propagation effect, the system traverses the network structure, constructs 18 failure paths starting from the source node based on DFS algorithm, and calculates the propagation probability and cumulative risk intensity of each path in turn.

[0109] In the third stage, the real-time data updating module is introduced, and the risk factor conditional probability table is dynamically updated based on the construction data of the past 7 days (a total of 1680 records). Through the actual occurrence of "pumping equipment abnormal stop during night concrete pouring process", the corresponding probability of and is dynamically adjusted, and the path ranking and intensity are recalculated, and the final construction stage evaluation result is output.

[0110] Fuzzy weight of high temperature and humidity environment: 0.63, Fuzzy weight of (job fatigue): 0.58, Fuzzy weight of (frequent start-stop of equipment): 0.42, Fuzzy weight of (progress lag): 0.31.

[0111] Prior probability of (equipment failure): 0.63x0.95=0.60.

[0112] In the case of =1, the conditional probability is: 0.72. Failure path

[0113] ={ → → } with a propagation probability P=0.60x0.72x0.68=0.29. Severity level S of each risk factor:

[0114] =0.7, =0.8, =0.6. Cumulative risk intensity F=0.29x(0.7+0.8+0.6)=0.62.

[0115] After real-time data correction,

[0116] the conditional probability is adjusted to 0.84, and the path propagation probability is raised to 0.36, corresponding to =0.36x(0.7+0.8+0.6)=0.73. From the experimental data, it can be seen that in the initial model construction stage, the proposed multi-layer fuzzy ANP structure can effectively integrate expert knowledge and multi-source data, not only ensuring the completeness of risk factor identification, but also improving the adaptability of the model in complex field environments. Through the normalization processing of fuzzy weights and the introduction of mapping functions, the problem of difficulty in directly using original fuzzy quantities in Bayesian reasoning is solved, realizing the quantitative transition from expert judgment to probability model, and showing significant innovation.

[0117] In the propagation path modeling stage, through the calculation of the failure cumulative intensity of the path level, the local risk can be aggregated to the system risk evaluation level. Compared with the traditional method which only outputs the node-level risk intensity, the present scheme provides a global safety evaluation result for the path, and the technical effect is more instructive.

[0118]

[0119] ​Especially in the dynamic updating module, the correction mechanism of conditional probability and severity level by real-time data makes the evaluation model reflect the instant state of the construction site. Data comparison shows that after the occurrence of equipment abnormalities, the path propagation probability and risk intensity both increase significantly, and the system successfully identifies the path as a high-priority risk path after reordering, which reflects the response capability and adjustability of the model to unexpected events.

[0120] Compared with the prior art which only relies on static models or expert scoring, the present application forms a dynamically evolving safety evaluation structure by fusing fuzzy ANP and Bayesian network, which not only improves the prediction accuracy, but also has higher implementation and site adaptability, and has substantial technical progress and creativity.

[0121] Embodiment 3, as an embodiment of the present application, provides a construction safety risk evaluation system, comprising: a fuzzy network module, which identifies a plurality of risk factors affecting construction safety based on a plurality of construction data collected on the construction site, and constructs a multi-layer fuzzy ANP network based on the fuzzy dependency relationship between the risk factors, and calculates the fuzzy weight of each risk factor.

[0122] A Bayesian graph module converts the multi-layer fuzzy ANP network into a Bayesian graph, and generates the initial conditional probability of each risk factor using the fuzzy weight.

[0123] A risk module constructs a set of failure propagation paths of the risk factors in the Bayesian network, calculates the propagation probability and cumulative risk intensity of each path in combination with the preset severity level of each risk factor.

[0124] A correction module dynamically corrects the conditional probability and severity level in the Bayesian network in combination with real-time construction data, updates the ranking results of the failure paths and risk factors, and outputs the risk evaluation results of the current construction stage.

[0125] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application or the parts that essentially contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product, which is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method of each embodiment of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0126] The logic and / or steps represented in the flow diagrams or otherwise described herein, for example, can be considered as a sequence of executable instructions, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this specification, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be for example but not limited to the following: an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, device, or propagation medium.

[0127] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed, and stored in a computer memory in a form that can be later executed by a computer. In some embodiments, the computer-readable medium can be non-transitory.

[0128] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies, which are known in the art, can be used to realize the logic having logic gates for implementing the logic functions on data signals: discrete logic circuits having logic gates for implementing the logic functions on data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like. It should be noted that the above-mentioned embodiments are merely intended to illustrate the technical solutions of the present application but not to limit the same, and although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the spirit and scope of the present application, and all such modifications and equivalents should be encompassed in the scope of the claims of the present application.

[0129] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A construction safety risk assessment method, characterized in that: include: Based on multi-source construction data collected at the construction site, multiple risk factors affecting construction safety are identified. A multi-layer fuzzy ANP network is constructed based on the fuzzy dependency relationships between the risk factors, and the fuzzy weights of each risk factor are calculated. The multi-layer fuzzy ANP network is converted into a Bayesian graph, and the fuzzy weights are used to generate the initial conditional probabilities of each risk factor; Constructing a set of failure propagation paths of risk factors in the Bayesian network, and calculating the propagation probability and cumulative risk intensity of each path in combination with a preset severity level of each risk factor; Combined with real-time construction data, the conditional probability and severity level in the Bayesian network are dynamically modified, the ranking results of failure paths and risk factors are updated, and the risk assessment results of the current construction stage are output.

2. The construction safety risk assessment method according to claim 1, wherein: The multi-source construction data includes environmental monitoring data, operation behavior data, equipment operation data, construction plan and progress data, and historical risk and accident records; Based on the statistical characteristics of the multi-source construction data, high-risk conditions at the construction site are identified; according to preset mapping rules, the identified high-risk conditions are mapped to standardized risk types, and corresponding risk factors are generated.

3. The construction safety risk assessment method according to claim 2, wherein: The multi-layer fuzzy ANP network is constructed, which includes dividing the identified risk factors into a target layer, a risk category layer and a risk factor layer according to the overall assessment goal of construction safety; Based on the construction task process and expert experience, the pairs of risk factors that affect each other are determined, bidirectional connection relationships are established, and a multi-layer fuzzy ANP network is generated.

4. The construction safety risk assessment method according to claim 3, wherein: The fuzzy weight calculation of each risk factor includes: for any risk factor pair with a dependency relationship, using triangular fuzzy numbers to perform pairwise comparison to construct a fuzzy judgment matrix; based on the fuzzy judgment matrix, using a fuzzy geometric mean method to calculate the fuzzy local weight of each risk factor, and embedding the local fuzzy weight vector into the corresponding sub-block structure of the supermatrix of the multi-layer fuzzy ANP network to form an overall fuzzy supermatrix; The overall fuzzy supermatrix is ​​subjected to column normalization processing. After the column random fuzzy supermatrix is ​​constructed, it is weighted updated through continuous power multiplication operations until a preset convergence accuracy threshold is reached, thereby obtaining the global fuzzy weight vector of each risk factor. The global fuzzy weight vector is defuzzified, and a weighted average operation is performed on each fuzzy weight using the center method. The global fuzzy weight vector is converted into a definite numerical value, and the converted numerical value is normalized to obtain the fuzzy weight of each risk factor.

5. The construction safety risk assessment method according to claim 4, wherein: The Bayesian graph includes constructing a graph node set with risk factors as graph nodes; Traversing all risk factor pairs with non-zero fuzzy weights, and establishing directed edges between corresponding graph nodes in the direction of the fuzzy weights; The generating of the initial conditional probability of each risk factor includes: for the risk factor that does not have a parent node in the Bayesian directed acyclic graph, generating a priori probability by a linear scaling mapping function according to the corresponding value in the global fuzzy weight vector, the mapping function is ;in, represents the mapping adjustment coefficient, represents the fuzzy weight value of the corresponding risk factor, represents the prior probability value, Indicates that the upper limit of the probability does not exceed 1; for risk factors with parent nodes, based on the fuzzy weights between all parent nodes and the target risk factor, the center method is first used to perform defuzzification, and then the weighted average method is used to generate conditional probability values, and a conditional probability table is constructed; by enumerating the value combinations of all parent nodes, a corresponding conditional probability value is generated for each value combination; all conditional probability values ​​under each group of conditions are normalized to form the conditional probability table.

6. The construction safety risk assessment method according to claim 5, wherein: Constructing the failure propagation path set includes taking a risk factor without a parent node as a path starting point, traversing all graph nodes with directed edge connections, and generating an ordered path sequence consisting of multiple risk factors; treating each ordered path sequence as a failure propagation path, and aggregating the results into the failure propagation path set; Calculating the propagation probability and cumulative risk intensity of each path includes obtaining a preset severity level of each risk factor in each failure propagation path and multiplying the conditional probability values ​​of all directed edges in the path to calculate the propagation probability of the path; The transmission probability is weighted and superimposed with the severity level of each risk factor in the path to obtain the cumulative risk intensity of the corresponding path; The failure propagation path set is sorted in descending order according to the cumulative risk intensity of all paths, and the path priority result of the current stage is output.

7. The construction safety risk assessment method according to claim 6, wherein: Correcting the conditional probability and severity level in the Bayesian network includes, according to a preset mapping rule, mapping the real-time construction data to the risk factor nodes established in the Bayesian graph, and updating the conditional probability table of the relevant risk factors; on the basis of updating the conditional probability table, dynamically adjusting the severity level of the corresponding risk factor in combination with the risk factor content involved in the real-time construction data; based on the corrected conditional probability table and severity level, recalculating the propagation probability and cumulative risk intensity of each path in the failure propagation path set, and updating all paths in descending order; and outputting the dynamic risk assessment result of the current construction stage according to the sorting result.

8. A system using the construction safety risk assessment method according to any one of claims 1 to 7, characterized in that: The fuzzy network module identifies multiple risk factors that affect construction safety based on multi-source construction data collected at the construction site, constructs a multi-layer fuzzy ANP network based on the fuzzy dependency relationships between the risk factors, and calculates the fuzzy weight of each risk factor; The Bayesian graph module converts the multi-layer fuzzy ANP network into a Bayesian graph and generates the initial conditional probability of each risk factor using fuzzy weights; A risk module constructs a set of failure propagation paths of risk factors in the Bayesian network, and calculates the propagation probability and cumulative risk intensity of each path in combination with a preset severity level of each risk factor; The correction module combines real-time construction data to dynamically correct the conditional probabilities and severity levels in the Bayesian network, update the ranking results of failure paths and risk factors, and output the risk assessment results of the current construction stage.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for assessing construction safety risks according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for assessing construction safety risks according to any one of claims 1 to 7 are implemented.

Images