Construction whole process engineering consultation management method, device, equipment and medium

Abstract

The invention relates to a construction whole process engineering consultation management method and device, equipment and a medium. The method comprises the following steps: acquiring local prior update data of each participating object of a project; the local prior update data is used for representing dynamic propagation conditions among the risk factors; integrating the local prior update data of each participating object based on a preset participating object reputation weight parameter to obtain normal prior update data; updating a preset federated Bayesian network according to the normal prior update data to obtain a dynamic Bayesian network; the federated Bayesian network comprises nodes and connecting edges, the nodes are risk factors, and the connecting edges are used for representing standard prior propagation probabilities among the risk factors; performing risk propagation simulation on the dynamic Bayesian network to obtain an engineering risk propagation report; the engineering risk propagation report is used for representing risk factors existing in the project and influences caused by the risk factors. By adopting the method, the accuracy of whole-process engineering management of the building can be improved.

Description

Technical Field

[0001] This invention belongs to the field of whole-process engineering consulting, and in particular relates to a method, device, equipment and medium for whole-process construction engineering consulting management. Background Technology

[0002] With the development of whole-process engineering consulting technology, a risk prediction and management model based on federated learning has emerged. In this model, each participant, such as design units, construction units, and suppliers, can independently train a risk prediction model using local data, and then upload the updated model parameters to a central server for aggregation to build a global risk knowledge base.

[0003] In existing federated learning risk prediction and management models, a standard federated averaging algorithm is typically used for model aggregation. The process is as follows: Each participant first trains locally based on a unified initial model, using optimization algorithms such as stochastic gradient descent to minimize the local risk loss function and obtain local model parameter updates. Subsequently, each participant transmits its updated parameters to a central server. After collecting parameter updates from all participants, the central server performs a weighted average calculation of the model parameters based on the amount of data from each participant or a preset fixed weight, generating a new generation of global model. This global model is then redistributed to each participant as the starting point for the next round of training, and this process iterates and cycles to gradually optimize risk prediction performance.

[0004] However, existing risk prediction and management models for federated learning have significant security vulnerabilities: when malicious actors exist in a federated learning system, they can launch data poisoning attacks through carefully constructed local training data, polluting the global model; especially in risk prediction models such as Bayesian networks that are sensitive to prior knowledge, malicious actors can inject systematic biases to cause the model to continuously underestimate specific risk paths, while traditional parameter averaging and aggregation mechanisms lack effective anomaly detection and isolation capabilities, causing the pollution effect to accumulate and amplify continuously during the iteration process, ultimately producing high-confidence erroneous risk predictions, which seriously reduces the accuracy of construction process management. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, device, equipment, and medium for consulting and management of the entire construction process that can improve the accuracy of construction process management, addressing the aforementioned technical problems.

[0006] Firstly, this application provides a method for engineering consulting and management throughout the entire construction process, including:

[0007] Obtain local prior update data for each participant in the project; local prior update data is used to characterize the dynamic propagation among various risk factors.

[0008] Based on the preset reputation weight parameters of the participants, the local prior update data of each participant is integrated to obtain normal prior update data; the reputation weight parameters of the participants include the reputation weight of each participant.

[0009] The pre-defined federated Bayesian network is updated based on normal prior update data to obtain a dynamic Bayesian network. The federated Bayesian network includes nodes and connecting edges. The nodes are each risk factor, and the connecting edges are used to represent the standard prior propagation probability between each risk factor.

[0010] Risk propagation simulation is performed on dynamic Bayesian networks to obtain an engineering risk propagation report; the engineering risk propagation report is used to characterize the risk factors existing in the project and the impact of the risk factors.

[0011] Furthermore, the local prior update data is obtained through the following method:

[0012] Obtain local project data from participating entities; local project data is used to characterize the working status of participating entities; local project data includes sampling time labels.

[0013] Preprocess the local project data to obtain preprocessed project data;

[0014] Based on the preprocessed project data, the standard prior propagation probabilities between nodes in the federated Bayesian network are updated to obtain the local posterior propagation probabilities; and the local posterior propagation probabilities between nodes are concatenated to obtain the local posterior propagation parameters.

[0015] Based on the local posterior propagation parameters and the federated Bayesian network, differential privacy noise is injected into the local posterior propagation probability to obtain the noisy posterior propagation probability.

[0016] Extract the latest sampling timestamp from the local project data and determine it as the fresh sampling timestamp; and obtain the local prior update data based on the fresh sampling timestamp and the noisy posterior propagation probability.

[0017] Furthermore, based on preset participant reputation weight parameters, the local prior update data of each participant is integrated to obtain normal prior update data, including:

[0018] The weighted weight of each participant is calculated based on the reputation weight parameter of the participant and the fresh sampling timestamp;

[0019] For each node in the federated Bayesian network, the noisy posterior propagation probabilities of each participating object are integrated based on the weighted weights to obtain the global prior probability of each node.

[0020] Based on the preset propagation constraint rules, the global prior probability of each node is adjusted to obtain the complete global prior probability of each node.

[0021] Normal prior update data is composed of the complete global prior probabilities of each node.

[0022] Furthermore, for each node in the federated Bayesian network, based on weighted weights, the noisy posterior propagation probabilities of each participating object are integrated to obtain the global prior probability of each node, including:

[0023] For each node in the federated Bayesian network, obtain the noisy posterior propagation probability of each participating object, including the node, and form a candidate noisy posterior propagation probability set.

[0024] Based on the weighted weights of the participating objects corresponding to the denoised posterior propagation probabilities, the denoised posterior propagation probabilities in the candidate denoised posterior propagation probabilities set are combined with the corresponding weights to form denoised posterior propagation probability-weight pairs.

[0025] Based on the noisy posterior propagation probability-weight pair, the noisy posterior propagation probabilities are arranged in ascending order according to their corresponding weights to obtain the sorted propagation probability set;

[0026] The weighted probabilities in the sorted propagation probability set are stacked in ascending order until a preset weight stacking threshold is met. The denoised posterior propagation probability corresponding to the last weighted weight to be stacked is selected as the global prior probability of the node.

[0027] Furthermore, the method also includes:

[0028] For each participating object, based on the noisy posterior propagation probability of the participating object and the global prior probability of each node, the deviation of the participating object is quantified to obtain the deviation degree of each participating object;

[0029] The standard deviation of the deviation is calculated based on the deviation of each participating object;

[0030] Select participants whose deviation is greater than a preset deviation threshold to form an abnormal participant set; select participants whose deviation is less than the preset deviation threshold to form a normal participant set.

[0031] For each participant in the set of anomalous participants, based on the deviation threshold, the preset penalty factor coefficient, the deviation standard deviation, the participant's reputation weight, and the deviation, the following formula is used to update the participant's reputation weight:

[0032]

[0033] in, Participants Updated reputation weight, Participants Reputation weight, It is the penalty factor coefficient. Participants The degree of deviation It is the deviation threshold. It is the standard deviation of the deviation;

[0034] For each participant in the set of normal participants, based on the deviation threshold, the preset reward factor coefficient, the participant's deviation, and reputation weight, the reputation weight of the participant is updated using the following formula:

[0035]

[0036] in, Participants Updated reputation weight, Participants Reputation weight, It is the reward factor coefficient. Participants The degree of deviation It is the deviation threshold;

[0037] The updated reputation weight parameters of the participants are obtained based on the updated reputation weights of the participants in the abnormal participant set and the updated reputation weights of the participants in the normal participant set.

[0038] Furthermore, the pre-defined federated Bayesian network is updated based on normal prior update data to obtain a dynamic Bayesian network, including:

[0039] Based on the normal prior update data, the nodes and the connection edges between the nodes in the updated federated Bayesian network are obtained to form the first federated Bayesian network.

[0040] Based on the expectation-maximization algorithm, the connection edges between nodes in the first federated Bayesian network are adjusted to obtain the second federated Bayesian network.

[0041] Based on the second federated Bayesian network, the conditional probability distribution parameters between each node in the federated Bayesian network are adjusted to generate an optimized parameter set;

[0042] The optimized parameter set is populated into the second federated Bayesian network to obtain the dynamic Bayesian network.

[0043] Furthermore, risk propagation simulations were performed on the dynamic Bayesian network to obtain an engineering risk propagation report, including:

[0044] Based on dynamic Bayesian networks, the marginal probabilities of each risk factor are analyzed to obtain the integrated risk probability distribution.

[0045] Based on the integrated risk probability distribution, the propagation path between various risk factors is analyzed to obtain the risk path analysis results.

[0046] Based on the risk path analysis results, intervention analysis is performed on the propagation path to obtain risk propagation simulation results; the risk propagation simulation results are used to characterize the influence range of risk factors in dynamic Bayesian networks.

[0047] Based on the results of risk propagation simulation, the impact of each propagation path is quantified to obtain a risk impact assessment matrix; the risk impact assessment matrix includes expected loss value, value at risk, and conditional value at risk.

[0048] Based on preset language organization rules, the risk impact assessment matrix is ​​transformed to obtain an engineering risk communication report.

[0049] Secondly, this application also provides a construction whole-process engineering consulting management device, comprising:

[0050] The data acquisition module is used to acquire local prior update data from each participant in the project; the local prior update data is used to characterize the dynamic propagation among various risk factors.

[0051] The data prior module is used to integrate the local prior update data of each participant based on the preset participant reputation weight parameters to obtain normal prior update data; the participant reputation weight parameters include the reputation weight of each participant;

[0052] The data update module is used to update the preset federated Bayesian network based on normal prior update data to obtain a dynamic Bayesian network. The federated Bayesian network includes nodes and connecting edges. The nodes are each risk factor, and the connecting edges are used to represent the standard prior propagation probability between each risk factor.

[0053] The report generation module is used to simulate risk propagation on dynamic Bayesian networks and obtain an engineering risk propagation report. The engineering risk propagation report is used to characterize the risk factors existing in the project and the impact of the risk factors.

[0054] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the construction whole-process engineering consulting management methods described in the first aspect of this application.

[0055] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for engineering consulting and management throughout the entire construction process.

[0056] The aforementioned method, apparatus, equipment, and medium for whole-process construction engineering consulting management involves acquiring local prior update data from each participant in the project. This local prior update data is used to characterize the dynamic propagation of various risk factors. Based on preset participant reputation weight parameters, the local prior update data of each participant is integrated to obtain normal prior update data. The participant reputation weight parameters include the reputation weight of each participant. A preset federated Bayesian network is updated according to the normal prior update data to obtain a dynamic Bayesian network. The federated Bayesian network includes nodes and connecting edges, where nodes represent various risk factors and connecting edges characterize the standard prior propagation probability between various risk factors. Risk propagation simulation is performed on the dynamic Bayesian network to obtain an engineering risk propagation report. The engineering risk propagation report is used to characterize the risk factors present in the project and the impact of these risk factors. By combining the collaborative concept of federated learning with the causal reasoning capabilities of Bayesian networks, this approach effectively aggregates dispersed professional knowledge while fully protecting the data privacy and trade secrets of all parties. Furthermore, through multi-layered protection mechanisms, it significantly improves the robustness of risk prediction models in scenarios involving untrustworthy participants, thereby enhancing the accuracy, foresight, and risk resistance capabilities of whole-process construction engineering management. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 A flowchart illustrating a construction full-process engineering consulting management method provided in one embodiment of this application;

[0059] Figure 2 A schematic diagram of a construction whole-process engineering consulting and management device provided in one embodiment of this application;

[0060] Figure 3 This is a schematic diagram of the structure of a computer device for a construction whole-process engineering consulting and management method provided in one embodiment of this application. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] In one embodiment, such as Figure 1As shown, a method for engineering consulting and management throughout the entire construction process is provided. This embodiment illustrates the method by applying it to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps S101-S104, wherein:

[0063] S101, Obtain local prior update data for each participant in the project; local prior update data is used to characterize the dynamic propagation among various risk factors.

[0064] Specifically, the participants refer to all entities providing professional services in the entire construction engineering consulting process, including but not limited to design firms, construction companies, supervision firms, and suppliers. Local prior update data is the adjustment amount of the propagation probability among various risk factors, used to describe the dynamic changes in the causal relationships between risk factors. Risk factors are used to characterize the types of risks present in a construction project, including but not limited to "design changes," "material supply delays," and "construction safety accidents." For example, the local prior update data can be in the form of a matrix, where each element corresponds to the updated value of a certain conditional probability parameter in a Bayesian network.

[0065] S102, based on the preset reputation weight parameters of the participants, integrate the local prior update data of each participant to obtain normal prior update data; the reputation weight parameters of the participants include the reputation weight of each participant.

[0066] Specifically, the terminal integrates the local prior update data of each participant based on preset participant reputation weight parameters to obtain normal prior update data. The preset participant reputation weight parameters are a preset, dynamically maintained set of parameters, in which a reputation weight is assigned to each participant. The reputation weight is used to characterize the credibility of the participant. For example, the initial value of the reputation weight can be set equally or based on the participant's qualifications, and can be calculated and updated based on the participant's performance in historical federated learning rounds (such as the quality and stability of updates provided, whether it is detected as anomaly, etc.). This embodiment does not further limit the setting of these preset participant reputation weight parameters.

[0067] S103. Update the preset federated Bayesian network based on the normal prior update data to obtain the dynamic Bayesian network. The federated Bayesian network includes nodes and connecting edges. The nodes are each risk factor, and the connecting edges are used to represent the standard prior propagation probability between each risk factor.

[0068] Specifically, the terminal integrates the purified normal prior update data from all participating parties into a unified, dynamic risk prediction model. The federated Bayesian network is a predefined probabilistic graphical model framework. Its nodes represent various risk factors in a construction project, such as "design changes," "material supply delays," and "construction safety accidents." The connecting edges represent the inherent, industry-consensus-compliant causal relationships between these risk factors. The weights on the edges are the standard prior propagation probabilities, constructed based on historical project data and domain expert knowledge, used to characterize the propagation probability of any risk factor determined based on historical project data and domain expert knowledge to another risk factor. The update process essentially integrates the local knowledge and new evidence contained in the "normal prior update data" provided by each participant into this pre-defined federated Bayesian network, resulting in a dynamic Bayesian network. The pre-defined federated Bayesian network can be constructed based on the various risk factors and propagation probabilities existing in actual work.

[0069] S104, perform risk propagation simulation on dynamic Bayesian network to obtain engineering risk propagation report; engineering risk propagation report is used to characterize the risk factors existing in the project and the impact of the risk factors.

[0070] Specifically, the terminal uses dynamic Bayesian networks to simulate risk propagation, simulating the probability of occurrence and impact of various risk factors, and integrating this simulation into an engineering risk propagation report. Risk propagation simulation refers to using probabilistic reasoning algorithms to calculate the probability distribution changes of other risk factor states given the occurrence of certain risk factors or the current state of the project, thereby predicting the possible propagation paths and scope of impact of risks within the project network.

[0071] This embodiment provides a method for full-process engineering consulting and management in construction. It acquires local prior update data representing the local knowledge updates of each participating entity in a distributed environment, and purifies the data source through deviation detection to obtain normal prior update data. This normal prior update data is then integrated and updated into a unified, pre-defined federated Bayesian network to obtain a dynamic Bayesian network. Finally, the dynamic Bayesian network is used for risk inference and report generation, resulting in an engineering risk propagation report. By combining the collaborative concept of federated learning with the causal reasoning capabilities of Bayesian networks, it effectively aggregates dispersed professional knowledge while fully protecting the data privacy and trade secrets of all parties. Furthermore, through multi-layered protection mechanisms, it significantly improves the robustness of the risk prediction model in scenarios involving untrusted participants, thereby enhancing the accuracy, foresight, and risk resistance capabilities of full-process engineering management in construction.

[0072] In one embodiment, the local prior update data is obtained through the following method:

[0073] S201, Obtain local project data of the participating objects; local project data is used to characterize the working status of the participating objects; local project data has sampling time labels.

[0074] Specifically, local project data refers to the raw data generated and collected by each participating entity during its business activities. This data specifically characterizes the working status of the participating entity. For example, when the participating entity is a construction unit, local project data includes, but is not limited to, construction logs, schedule completion rates, safety accident records, and daily work hour records; when the participating entity is a design unit, local project data includes, but is not limited to, design drawing change records, calculation sheets, and review comments.

[0075] S202, preprocess the local project data to obtain preprocessed project data.

[0076] Specifically, the terminal preprocesses local project data to obtain preprocessed project data. This involves sequentially performing missing value completion, feature extraction, and standardization on the local project data. For example, the terminal can use interpolation to complete missing values, obtaining complete local project data. Optionally, the terminal performs feature extraction on the local project data to extract features relevant to risk prediction, such as extracting "average weekly overtime hours" from daily work hour records as a risk feature reflecting construction fatigue. Feature processing can employ corresponding feature extraction methods depending on the participating entities and the local project data. The terminal performs feature extraction on the complete local project data to obtain project feature data. For example, the terminal can use Z-score standardization to standardize the project feature data, obtaining preprocessed project data.

[0077] S203. Based on the preprocessed project data, update the standard prior propagation probabilities between nodes in the federated Bayesian network to obtain the local posterior propagation probabilities; and concatenate the local posterior propagation probabilities between nodes to obtain the local posterior propagation parameters.

[0078] Specifically, the terminal uses the preprocessed project data of the participating objects to update its understanding of risk propagation patterns. The standard prior propagation probability is an initially set probability parameter in the federated Bayesian network, based on general knowledge or historical data. The update process follows a Bayesian learning framework, treating the preprocessed project data as newly observed "evidence," and calculating the posterior probability using Bayes' theorem. For example, for a node in the network called "construction delay," whose parent nodes are "design change" and "severe weather," the update process uses the frequency of simultaneous occurrences of "design change" and "construction delay" in local data to adjust the conditional probability between the two. The local posterior propagation probability is the updated conditional probability parameter incorporating the preprocessed project data of the participating object. The terminal also concatenates all updated local posterior propagation probabilities in the federated Bayesian network using the preprocessed project data of the participating object in a preset order to obtain the local posterior propagation parameter belonging to that participating object. For example, the preset order can be in ascending order of node numbers in the federated Bayesian network; this embodiment does not further limit the setting of this preset order.

[0079] S204, based on the local posterior propagation parameters and the federated Bayesian network, performs differential privacy noise injection on the local posterior propagation probability to obtain the noisy posterior propagation probability.

[0080] Specifically, differential privacy is a rigorous and quantifiable definition of privacy protection. Its core idea is to add carefully controlled random noise to the computational results, making the output insensitive to changes in any single record in the dataset. The noise injection operation is performed on the local posterior propagation parameters. For each local posterior propagation probability in the local posterior propagation parameters of the participating object, the terminal samples a random noise value from the Laplace distribution based on its sensitivity and a preset privacy budget, and adds it to the local posterior propagation probability to obtain the corresponding noisy posterior propagation probability. For example, the preset privacy budget is used to control the strength of privacy protection and can be set according to actual work. Sensitivity can be obtained by calculating the maximum change in the parameter that a single data point can cause.

[0081] S205, extract the latest sampling timestamp from the local project data and determine it as the fresh sampling timestamp; and obtain the local prior update data based on the fresh sampling timestamp and the noisy posterior propagation probability.

[0082] Specifically, the fresh sampling timestamp is determined by scanning the entire local project data and finding the maximum value (i.e., the latest time point) among all sampled timestamps. The terminal integrates the fresh sampling timestamp of the participating entity with the noisy posterior propagation probability to obtain the local prior update data. For example, in the local project data of a participating entity that is a construction unit, if the latest record is "Safety Inspection Report of November 15, 2023", then "2023-11-15" is extracted as the fresh sampling timestamp.

[0083] This embodiment provides a method for managing the entire construction process consulting services. Starting with acquiring local project data from each participating entity, the method undergoes data preprocessing, Bayesian learning-based parameter updates, noise injection under differential privacy protection, and extraction and appending of fresh sampling timestamps before encapsulation. This forms standardized local prior update data representing the corresponding participating entities. It scientifically extracts local evidence representing participating entities based on risk models and fundamentally protects the business secrets and original data privacy of each participating entity through differential privacy technology. This solves the trust problem in cross-organizational data collaboration and provides reliable input data for building secure, reliable, and efficient distributed risk prediction.

[0084] In one embodiment, based on a preset participant reputation weight parameter, the local prior update data of each participant is integrated to obtain normal prior update data, including:

[0085] S301, calculate the weighted weight of each participating object based on the reputation weight parameter of the participating object and the fresh sampling timestamp.

[0086] Specifically, the terminal uses the following formula based on the reputation weight of each participant and the fresh sampling timestamp: The weighted weight corresponding to the participating object is calculated. Participants The weighted weights. Participants Reputation weight. It is the time decay coefficient, which can be set according to the actual work. This is the current time. Participants Fresh sampling timestamp.

[0087] S302, for each node in the federated Bayesian network, integrates the noisy posterior propagation probabilities of each participating object based on weighted weights to obtain the global prior probability of each node.

[0088] Specifically, for each node in the federated Bayesian network, the terminal collects the denoised posterior propagation probabilities provided by all participating objects to the node (i.e., the conditional probability parameters of the node under different configurations of its parent node) based on the denoised posterior propagation probabilities of each participating object, and fuses these denoised posterior propagation probability values ​​according to the weighted weights of the corresponding participating objects to obtain the global prior probability of each node.

[0089] S303, based on preset propagation constraint rules, adjusts the global prior probability of each node to obtain the complete global prior probability of each node.

[0090] Specifically, the preset propagation constraint rules are used to characterize the hard and soft constraints among various risk factors. The terminal checks whether the global prior probabilities of each node obtained in S302 violate these preset rules. If they violate them, corrections are made according to the corresponding propagation constraint rules. For example, the preset propagation constraint rules can be set based on the experience of domain experts, physical laws, or logical relationships. This embodiment does not further limit the setting of these preset propagation constraint rules, and their form can be: ,in It is the first one A hard constraint is used to ensure the mathematical completeness and logical consistency of the probability model. If the sum of the global prior probabilities of the aggregated nodes is not 1, the terminal will automatically perform normalization to scale it to a sum of 1. , It is the first One soft rule is used to incorporate professional knowledge and common sense from the construction field into the model, guiding the probability parameters to adjust in a direction consistent with empirical understanding. If the aggregation results show that the probability of accidents increases after safety investment, the terminal will adjust it to a reasonable range within a certain tolerance, such as forcing it to decrease slightly or remain unchanged. Another soft rule may involve the rationality of risk propagation, such as "the impact of extreme weather on outdoor construction should be greater than its impact on indoor construction."

[0091] S304, based on the complete global prior probabilities of each node, forms the normal prior update data.

[0092] Specifically, the terminal concatenates the complete global prior probabilities of each node to obtain normal prior update data.

[0093] This embodiment provides a method for managing the entire construction process consulting services. It dynamically calculates the weighted weights of each participating object using reputation weights and timestamps. A weighted robust aggregation algorithm is used to integrate the probabilities of each node, obtaining the global prior probability of each node. Furthermore, pre-defined propagation constraint rules are introduced to correct the aggregation results at the domain knowledge level, resulting in complete global prior probabilities for each node that conform to industry logic and common sense. These complete global prior probabilities are then integrated into normal prior update data. This method solves the data poisoning and model contamination problems faced by federated learning in untrusted environments, improves the robustness, accuracy, and rationality of the global risk prediction model, and provides reliable input data for generating high-quality risk propagation reports.

[0094] In one embodiment, for each node in the federated Bayesian network, the noisy posterior propagation probabilities of each participating object are integrated based on weighted weights to obtain the global prior probability of each node, including:

[0095] S401: For each node in the federated Bayesian network, obtain the noisy posterior propagation probability of each participating object, including the node, and form a candidate noisy posterior propagation probability set.

[0096] Specifically, for a given node in a federated Bayesian network, the terminal iterates through the local prior update data of all participating objects. It extracts the noisy posterior propagation probability associated with that node from each participating object. These probability parameters describe the probability of the node occurring in different states of its parent node from the participant's local perspective. The noisy posterior propagation probabilities provided by all participating objects for this node are then integrated to form a candidate noisy posterior propagation probability set for that node.

[0097] S402, based on the weighted weights of the participating objects corresponding to the denoised posterior propagation probability, form a denoised posterior propagation probability-weight pair by combining the denoised posterior propagation probability in the candidate denoised posterior propagation probability set with the corresponding weighted weights.

[0098] Specifically, the weighted weight is the weighted weight of each participating object calculated according to the aforementioned S301. It is a scalar value that reflects the overall credibility of the corresponding participating object in this update. For each noisy posterior propagation probability in the candidate noisy posterior propagation probability set, the terminal finds the participating object that provides the noisy posterior propagation probability, obtains its corresponding weighted weight, and combines the noisy posterior propagation probability and the corresponding weighted weight into a tuple to obtain the noisy posterior propagation probability-weight pair.

[0099] S403, based on the noisy posterior propagation probability-weight pair, arranges the noisy posterior propagation probabilities in ascending order of their corresponding weights to obtain the sorted propagation probability set.

[0100] Specifically, the terminal iterates through all noisy posterior propagation probability-weight pairs obtained in S402. Based on the weight values ​​in each pair, it sorts the noisy posterior propagation probabilities in ascending order of their respective weights, resulting in a sorted propagation probability set. In this sorted propagation probability set, the probabilities from low-weight (i.e., low-confidence) participants are ranked higher, while the probability estimates from high-weight (i.e., high-confidence) participants are ranked lower.

[0101] S404: The weighted probabilities in the sorted propagation probability set are stacked in ascending order until a preset weight stacking threshold is met. The denoised posterior propagation probability corresponding to the last weighted weight to be stacked is selected as the global prior probability of the node.

[0102] Specifically, the preset weight stacking threshold means finding a position where the sum of the weights corresponding to all probability values ​​less than or equal to that position first reaches or exceeds the weight stacking threshold. The preset weight stacking threshold is set to 0.5 by default, indicating that a position is found where the sum of the weights corresponding to all probability values ​​less than or equal to that position first reaches or exceeds half of the total weight. The terminal starts from the head of the sorted propagation probability set (already arranged in ascending order of weight) and sequentially accumulates the weighted weights corresponding to each noisy posterior propagation probability. During the accumulation process, it checks in real time whether the current accumulated weight sum has reached the preset threshold (e.g., 0.5). Once the accumulated sum reaches or exceeds the threshold, the accumulation stops, and the noisy posterior propagation probability corresponding to the last weighted weight added to the accumulation is selected as the global prior probability of that node.

[0103] This embodiment provides a method for managing the entire construction process consulting services. It collects the noisy posterior propagation probabilities of all participating objects for each node; associates each noisy posterior propagation probability with the weighted weight of its source to obtain a noisy posterior propagation probability-weight pair; sorts these probabilities by weight to organize the decision sequence, resulting in an ordered propagation probability set; and determines the global prior probability of a node by accumulating weights to locate the median. This method effectively tolerates and filters out malicious or low-quality anomalous probability updates provided by some participants. As long as the total weight of these anomalous updates does not exceed half, they cannot decisively influence the final global prior probability. This significantly enhances the stability and security of the federated Bayesian network in the face of data poisoning attacks or non-malicious data bias, ensuring the reliability of the global risk model.

[0104] In one embodiment, the method further includes:

[0105] S501: For each participating object, based on the noisy posterior propagation probability of the participating object and the global prior probability of each node, the deviation of the participating object is quantified to obtain the deviation degree of each participating object.

[0106] Specifically, the terminal compares the noisy posterior propagation probability of each participating object with the global prior probability element-wise, and calculates a covariance matrix based on the noisy posterior propagation probabilities of all participating objects. This matrix describes the correlation between probability updates of different nodes. The terminal then uses the Mahalanobis distance formula to calculate the distance between each participating object and the global consensus vector, obtaining the deviation of each participating object. Mahalanobis distance is a statistical distance metric that considers the covariance structure of a dataset. It is used to calculate the difference between a point and a distribution, and its advantage lies in its ability to consider correlations between different dimensions and its insensitivity to scale.

[0107] S502, based on the deviation of each participating object, calculate the standard deviation of the deviation.

[0108] Specifically, the standard deviation of deviation is the sample standard deviation of the deviation values ​​of all participants. It is calculated as follows: first, calculate the average of all deviations; then, calculate the sum of the squares of the differences between each deviation and the average; divide this sum by (number of participants - 1) to obtain the variance; finally, take the square root to obtain the standard deviation. The standard deviation reflects the consistency of the update behavior of each participant in this round. If the standard deviation is small, it indicates that the updates of most participants are close to the global consensus, indicating high overall consistency; if the standard deviation is large, it indicates significant differences in updates among participants, potentially indicating obvious abnormal behavior.

[0109] S503: Select participants whose deviation is greater than a preset deviation threshold to form an abnormal participant set; select participants whose deviation is less than a preset deviation threshold to form a normal participant set.

[0110] Specifically, a preset deviation threshold is a key parameter used to delineate the boundary between "normal" and "abnormal" behavior. This threshold can be determined based on the distribution of historical deviation data (such as the mean plus a certain number of standard deviations), through simulation experiments, or set by domain experts based on their tolerance for abnormal behavior. This embodiment does not further limit the setting of this preset deviation threshold. The terminal compares the deviation of each participant with this deviation threshold. All participants with deviations greater than the threshold are classified into the abnormal participant set, meaning that their updates are considered unreliable or malicious. All participants with deviations less than or equal to the threshold are classified into the normal participant set, meaning that their update behavior is acceptable and consistent with the mainstream consensus.

[0111] S504, for each participant in the abnormal participant set, based on the deviation threshold, the preset penalty factor coefficient, the deviation standard deviation, the participant's reputation weight, and the deviation, update the participant's reputation weight using the following formula:

[0112]

[0113] in, Participants Updated reputation weight, Participants Reputation weight, It is the penalty factor coefficient. Participants The degree of deviation It is the deviation threshold. It is the standard deviation of the deviation.

[0114] Specifically, for each participant in the set of abnormal participants, the terminal updates its reputation weight using a formula. Among these, the participant... Updated reputation weight It will vary depending on the participants. The deviation increases, but decreases exponentially, thus significantly reducing its influence in future rounds of federated learning. Participants Reputation weight It can be obtained based on preset participant reputation weight parameters. (Penalty factor coefficient) This is used to control the overall intensity of punishment and can be set according to the actual work situation. Participants deviation This can be obtained from S501. Deviation threshold. This is the preset deviation threshold mentioned in S503. Deviation standard deviation. It is the standard deviation of the deviation calculated by S502.

[0115] S505, for each participant in the set of normal participants, based on the deviation threshold, the preset reward factor coefficient, the participant's deviation, and reputation weight, the reputation weight of the participant is updated using the following formula:

[0116]

[0117] in, Participants Updated reputation weight, Participants Reputation weight, It is the reward factor coefficient. Participants The degree of deviation It is the deviation threshold.

[0118] Specifically, for each participant in the set of normal participants, the terminal uses a formula to update its reputation weight. Among these, the participants... Updated reputation weight It increases linearly with the proportion of deviations exceeding the deviation threshold. (Participating entities) Reputation weight The reward factor coefficient can be obtained based on preset participant reputation weight parameters. Used to control the overall strength of rewards, it is usually set conservatively to prevent the weight from expanding too quickly, but it can also be set according to the actual work. Participants deviation This can be obtained from S501. Deviation threshold. It is the preset deviation threshold described in S503.

[0119] S506. Based on the updated reputation weights of participants in the abnormal participant set and the updated reputation weights of participants in the normal participant set, the updated participant reputation weight parameters are obtained.

[0120] Specifically, the terminal updates the reputation weight of the participants in the abnormal participant group and the normal participant group, and updates the reputation weight of the corresponding participant in the participant reputation weight parameter to obtain the updated participant reputation weight parameter.

[0121] This embodiment provides a method for managing the entire construction process consulting services. It precisely quantifies the deviation behavior of each participant to obtain the deviation degree of each participant; calculates the standard deviation of the overall deviation to perceive the consensus environment of the current round, and obtains the deviation standard deviation; it also distinguishes between normal and abnormal behavior based on preset thresholds; and designs reputation update formulas with exponential penalties and linear rewards for abnormal and normal participants respectively, obtaining updated participant reputation weight parameters. This creates a sound incentive mechanism, enabling participants who continuously provide high-quality, consensus-compliant updates to accumulate reputation and gain greater influence; simultaneously, it imposes sufficiently deterrent penalties on abnormal behavior commensurate with its severity, effectively eliminating sources of "noise" and "malice." This improves the long-term robustness, stability, and collaborative efficiency of the federated learning framework in open, competitive, or partially untrustworthy environments.

[0122] In one embodiment, a preset federated Bayesian network is updated based on normal prior update data to obtain a dynamic Bayesian network, including:

[0123] S601, based on normal prior update data, updates each node and the connection edges between each node in the federated Bayesian network to obtain the first federated Bayesian network.

[0124] Specifically, the terminal fills the corresponding nodes in the pre-defined federated Bayesian network with the complete global prior probabilities of each node included in the normal prior update data, replacing the old parameter values. This stage typically does not change the network structure; that is, the node set and connection edges remain unchanged. Through this step, the network's knowledge content (parameters) is refreshed to reflect the latest collected data, resulting in the first federated Bayesian network.

[0125] S602, based on the expectation-maximization algorithm, adjusts the connection edges between nodes in the first federated Bayesian network to obtain the second federated Bayesian network.

[0126] Specifically, the Expectation-Maximization (EM) algorithm is an iterative algorithm for maximum likelihood estimation in cases with latent variables or missing data. In Bayesian network structure learning, the Structure-EM algorithm is a commonly used method. It alternately performs two steps: the E-step (expectation step), which calculates the expectation of missing data or latent variables based on existing parameters within the current network structure; and the M-step (maximization step), which searches for a network structure that maximizes the expected likelihood based on the E-step result (i.e., adding, removing, or changing the direction of edges). Through iteration, the algorithm searches for a better network topology. Adjusting connections may reveal new risk propagation paths (adding edges), negate some previously assumed weak relationships (removing edges), or correct causal directions (reversing edges). The network obtained after optimizing the structure using the EM algorithm is called the second federated Bayesian network. Its structure is superior to the first network and can more accurately represent the actual dependencies between risk factors.

[0127] S603, based on the second federated Bayesian network, adjusts the conditional probability distribution parameters between nodes in the federated Bayesian network to generate an optimized parameter set.

[0128] Specifically, after obtaining the second federated Bayesian network, it is necessary to relearn the optimal parameters that best match it. The change in network structure means that the dependencies of node conditional probabilities have changed (e.g., the set of parent nodes of a node has changed). Therefore, it is necessary to re-estimate the conditional probability distribution parameters of all nodes under the new structure, based on existing or new data (in this context, this can be viewed as the information distribution implicitly represented by normal prior updated data). For example, methods such as maximum likelihood estimation or Bayesian estimation can be used for this calculation. This process generates a new set of conditional probability parameters for each node in the second federated Bayesian network that best matches the structure. The set of all these new parameters constitutes the optimized parameter set.

[0129] S604. The optimized parameter set is filled into the second federated Bayesian network to obtain the dynamic Bayesian network.

[0130] Specifically, the terminal assigns the optimized parameter set obtained in S603 to the corresponding positions in the second federated Bayesian network. Specifically, the terminal fills the conditional probability parameters of each node in the network into the corresponding node conditional probability table, resulting in a dynamic Bayesian network whose structure and parameters have been collaboratively optimized, is rich in the latest knowledge, and can be immediately used for probabilistic inference.

[0131] This embodiment provides a method for whole-process engineering consulting and management in construction. It updates basic parameters using purified global prior probabilities and introduces an expectation-maximization algorithm to perform data-driven optimization search of the network topology to discover more realistic risk propagation paths, resulting in a second federated Bayesian network. Furthermore, it relearns the optimal probability parameters matching the optimized new structure, obtaining an optimized parameter set. Finally, it integrates the optimized parameter set and the second federated Bayesian network into a complete model, resulting in a dynamic Bayesian network. This method overcomes the limitations of traditional federated learning, which only updates parameters without optimizing the structure. It adaptively discovers and characterizes unique, non-prior-defined risk causal relationships in projects, enhancing the expressive power and accuracy of risk prediction models and improving the precision, foresight, and risk resilience of whole-process engineering management in construction.

[0132] In one embodiment, a dynamic Bayesian network performs risk propagation simulation to obtain an engineering risk propagation report, including:

[0133] S701, based on a dynamic Bayesian network, analyzes the marginal probabilities of each risk factor to obtain an integrated risk probability distribution.

[0134] Specifically, the dynamic Bayesian network is the model constructed in step S103, containing a complete structure and parameters. Marginal probability analysis refers to calculating the probability of each node (risk factor) in the dynamic Bayesian network being in each possible state, given some known information (i.e., evidence, such as the knowledge that a "design change" has occurred) or no evidence at all, using the connection tree algorithm. The connection tree algorithm is an efficient algorithm for precise inference in Bayesian networks. It calculates the marginal or conditional probabilities of any node in the network by converting the Bayesian network into a data structure called a connection tree (or clique tree) and performing message passing on this tree. The terminal uses the connection tree algorithm to compile the nodes and connecting edges in the dynamic Bayesian network into a connection tree, where each node (called a clique) contains a set of variables from the original network. The conditional probability table of each node in the dynamic Bayesian network is then initialized with the potential function of the corresponding clique in the connection tree. If known evidence exists (such as the occurrence of certain risk factors), the evidence is injected into the connection tree. The message-passing algorithm is then executed on the linked tree to propagate probability information throughout the tree and achieve global consistency. By querying the clique corresponding to each risk factor (node) in the linked tree, the marginal probability of that risk factor in all possible states can be calculated. For example, the resulting integrated risk probability distribution is a set of probability distributions containing the marginal probability of each risk factor; for instance, the probability of "supplier delay" is 0.15, and the probability of "construction safety accident" is 0.03.

[0135] S702, based on the integrated risk probability distribution, analyzes the propagation path between various risk factors and obtains the risk path analysis results.

[0136] Specifically, risk path analysis aims to identify the critical propagation sequences from the risk source (root cause) to the risk sink (ultimate impact). The terminal uses graph traversal algorithms to find all directed paths from the source node to the sink node, identifying the propagation paths of the risk. It then evaluates and ranks the overall "accessibility" or risk flow magnitude of the paths by combining the conditional probabilities of each node along the path. Graph traversal algorithms are a fundamental class of algorithms in graph theory, used to systematically access all nodes in a graph or explore paths between nodes; common algorithms include depth-first search and breadth-first search. For example, the result of risk path analysis can be a list of critical risk paths, containing the risk factors included in the propagation path and their corresponding impact scores. .in, This path is characterized as the first in the risk path analysis results. One transmission path.

[0137] S703, based on the risk path analysis results, conducts intervention analysis on the propagation path to obtain risk propagation simulation results; the risk propagation simulation results are used to characterize the influence range of risk factors in dynamic Bayesian networks.

[0138] Specifically, intervention analysis is a crucial concept in causal inference, and its mathematical tool is "do-calculus." It simulates actively forcibly setting a risk factor to a specific state (e.g., by strengthening management "do(safety investment = high)") and observing the impact of this intervention on the posterior probabilities of other risk factors in the network, thus obtaining the risk propagation simulation results. Do-calculus is a set of causal inference rules proposed by Judea Pearl, which allows the calculation of the intervention distribution (i.e., P(Y|do(X))) from the observed distribution in a causal graph. It simulates actively forcibly setting a risk factor to a specific state (e.g., by strengthening management "do(safety investment = high)") and then observing the impact of this intervention on the posterior probabilities of other risk factors in the network, particularly key target nodes (such as project duration and cost). For example, by performing such intervention simulations on nodes on the critical path identified in the risk path analysis results, the terminal can evaluate the effectiveness of various risk response measures. The risk propagation simulation results quantitatively demonstrate how intervention behaviors change the propagation process and scope of impact of risks. For example, "Increasing safety investment can reduce the probability of safety accidents by X%, thereby reducing the probability of project delays by Y%."

[0139] S704, based on the results of risk propagation simulation, quantifies the impact of each propagation path to obtain a risk impact assessment matrix; the risk impact assessment matrix includes expected loss value, risk value, and conditional risk value.

[0140] Specifically, the risk impact assessment matrix is ​​a structured table where rows may represent different risk scenarios or propagation paths, and columns represent different quantitative impact indicators. The quantitative impact indicators in the risk impact assessment matrix include expected loss, value at risk (VaR), and conditional value at risk (VaR). Expected loss represents the product of the probability of an adverse consequence occurring under a propagation path and the severity of that consequence, providing an average expected value of the risk impact. For any propagation path, the expected loss is obtained by determining the adverse consequences (such as increased costs) caused by that path, calculating the probability of that consequence occurring based on simulation results, and multiplying the probability by the estimated consequence. VaR represents the maximum loss that the propagation path may cause at a given confidence level (e.g., 95%). For any propagation path, the VaR is obtained by finding a value among all possible loss values ​​obtained through Monte Carlo simulations such that the proportion of all loss values ​​less than or equal to that value reaches a specified confidence level (e.g., 95%). Conditional value at risk represents the average loss if the loss exceeds the VaR. It measures the average degree of loss in extreme cases. The conditional value of risk for any propagation path is obtained by filtering out all scenarios in the Monte Carlo simulation results where the loss exceeds VaR, and then calculating the average of the loss values ​​in these scenarios.

[0141] S705, based on preset language organization rules, transforms the risk impact assessment matrix to obtain an engineering risk communication report.

[0142] Specifically, the preset language organization rules are a set of templates or natural language generation rules that define how to map the structured data (numbers, indicators) in the matrix into coherent text descriptions, conclusions, and recommendations that conform to engineering report specifications. The terminal, based on the preset language organization rules, converts the risk impact assessment matrix into an engineering risk propagation report that is easily understood by project managers. For example, the preset language organization rules can be set according to the report format requirements in actual work; this embodiment does not further limit the setting of these preset language organization rules.

[0143] This embodiment provides a comprehensive construction engineering consulting and management method. By constructing a full-chain risk simulation and decision support process, from basic probability analysis to path mining, intervention simulation, impact quantification, and report generation, it fully leverages the powerful capabilities of dynamic Bayesian networks in risk prediction. This enhances the expressive power and accuracy of the risk prediction model, thereby improving the precision, foresight, and risk resilience of comprehensive construction engineering management.

[0144] In the aforementioned method for whole-process construction engineering consulting and management, local prior update data of each participant in the project is acquired. This local prior update data is used to characterize the dynamic propagation of various risk factors. Based on preset participant reputation weight parameters, the local prior update data of each participant is integrated to obtain normal prior update data. The participant reputation weight parameters include the reputation weight of each participant. A preset federated Bayesian network is updated according to the normal prior update data to obtain a dynamic Bayesian network. The federated Bayesian network includes nodes and connecting edges, where nodes represent various risk factors and connecting edges are used to characterize the standard prior propagation probability between various risk factors. Risk propagation simulation is performed on the dynamic Bayesian network to obtain an engineering risk propagation report. The engineering risk propagation report is used to characterize the risk factors existing in the project and the impact of these risk factors. By combining the collaborative concept of federated learning with the causal reasoning ability of Bayesian networks, this method effectively aggregates dispersed professional knowledge while fully protecting the data privacy and trade secrets of all parties. Through multi-layered protection mechanisms, it significantly improves the robustness of the risk prediction model in scenarios with untrustworthy participants, thereby enhancing the accuracy, foresight, and risk resistance capabilities of whole-process construction engineering management.

[0145] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0146] Based on the same inventive concept, this application also provides a building whole-process engineering consulting management device for implementing the above-mentioned building whole-process engineering consulting management method. The solution provided by this device is similar to the solution described in the above-described method. Therefore, the specific limitations of one or more embodiments of the building whole-process engineering consulting management device provided below can be found in the limitations of the building whole-process engineering consulting management method described above, and will not be repeated here.

[0147] In one exemplary embodiment, such as Figure 2 As shown, a construction whole-process engineering consulting management device 200 is provided, comprising:

[0148] The data acquisition module 201 is used to acquire local prior update data of each participant in the project; the local prior update data is used to characterize the dynamic propagation among various risk factors.

[0149] The data prior module 202 is used to integrate the local prior update data of each participant based on the preset participant reputation weight parameters to obtain normal prior update data; the participant reputation weight parameters include the reputation weight of each participant;

[0150] The data update module 203 is used to update the preset federated Bayesian network based on normal prior update data to obtain a dynamic Bayesian network. The federated Bayesian network includes nodes and connecting edges. The nodes are each risk factor, and the connecting edges are used to represent the standard prior propagation probability between each risk factor.

[0151] The report generation module 204 is used to perform risk propagation simulation on the dynamic Bayesian network and obtain an engineering risk propagation report; the engineering risk propagation report is used to characterize the risk factors existing in the project and the impact of the risk factors.

[0152] Furthermore, the data acquisition module is also used for:

[0153] Obtain local project data from participating entities; local project data is used to characterize the working status of participating entities; local project data includes sampling time labels.

[0154] Preprocess the local project data to obtain preprocessed project data;

[0155] Based on the preprocessed project data, the standard prior propagation probabilities between nodes in the federated Bayesian network are updated to obtain the local posterior propagation probabilities; and the local posterior propagation probabilities between nodes are concatenated to obtain the local posterior propagation parameters.

[0156] Based on the local posterior propagation parameters and the federated Bayesian network, differential privacy noise is injected into the local posterior propagation probability to obtain the noisy posterior propagation probability.

[0157] Extract the latest sampling timestamp from the local project data and determine it as the fresh sampling timestamp; and obtain the local prior update data based on the fresh sampling timestamp and the noisy posterior propagation probability.

[0158] Furthermore, the data prior module includes:

[0159] The weighted calculation unit is used to calculate the weighted weight of each participating object based on the reputation weight parameter of the participating object and the fresh sampling timestamp;

[0160] The global prior calculation unit is used to integrate the noisy posterior propagation probabilities of each participating object based on weighted weights for each node in the federated Bayesian network, so as to obtain the global prior probability of each node.

[0161] The complete posterior calculation unit is used to adjust the global prior probability of each node based on the preset propagation constraint rules to obtain the complete global prior probability of each node.

[0162] The normal prior update unit is used to form normal prior update data based on the complete global prior probabilities of each node.

[0163] Furthermore, the global prior computation unit is also used for:

[0164] For each node in the federated Bayesian network, obtain the noisy posterior propagation probability of each participating object, including the node, and form a candidate noisy posterior propagation probability set.

[0165] Based on the weighted weights of the participating objects corresponding to the denoised posterior propagation probabilities, the denoised posterior propagation probabilities in the candidate denoised posterior propagation probabilities set are combined with the corresponding weights to form denoised posterior propagation probability-weight pairs.

[0166] Based on the noisy posterior propagation probability-weight pair, the noisy posterior propagation probabilities are arranged in ascending order according to their corresponding weights to obtain the sorted propagation probability set;

[0167] The weighted probabilities in the sorted propagation probability set are stacked in ascending order until a preset weight stacking threshold is met. The denoised posterior propagation probability corresponding to the last weighted weight to be stacked is selected as the global prior probability of the node.

[0168] Furthermore, the device also includes a reputation weight update unit for:

[0169] For each participating object, based on the noisy posterior propagation probability of the participating object and the global prior probability of each node, the deviation of the participating object is quantified to obtain the deviation degree of each participating object;

[0170] The standard deviation of the deviation is calculated based on the deviation of each participating object;

[0171] Select participants whose deviation is greater than a preset deviation threshold to form an abnormal participant set; select participants whose deviation is less than the preset deviation threshold to form a normal participant set.

[0172] For each participant in the set of anomalous participants, based on the deviation threshold, the preset penalty factor coefficient, the deviation standard deviation, the participant's reputation weight, and the deviation, the following formula is used to update the participant's reputation weight:

[0173]

[0174] in, Participants Updated reputation weight, Participants Reputation weight, It is the penalty factor coefficient. Participants The degree of deviation It is the deviation threshold. It is the standard deviation of the deviation;

[0175] For each participant in the set of normal participants, based on the deviation threshold, the preset reward factor coefficient, the participant's deviation, and reputation weight, the reputation weight of the participant is updated using the following formula:

[0176]

[0177] in, Participants Updated reputation weight, Participants Reputation weight, It is the reward factor coefficient. Participants The degree of deviation It is the deviation threshold;

[0178] The updated reputation weight parameters of the participants are obtained based on the updated reputation weights of the participants in the abnormal participant set and the updated reputation weights of the participants in the normal participant set.

[0179] Furthermore, the data update module is also used for:

[0180] Based on the normal prior update data, the nodes and the connection edges between the nodes in the updated federated Bayesian network are obtained to form the first federated Bayesian network.

[0181] Based on the expectation-maximization algorithm, the connection edges between nodes in the first federated Bayesian network are adjusted to obtain the second federated Bayesian network.

[0182] Based on the second federated Bayesian network, the conditional probability distribution parameters between each node in the federated Bayesian network are adjusted to generate an optimized parameter set;

[0183] The optimized parameter set is populated into the second federated Bayesian network to obtain the dynamic Bayesian network.

[0184] Furthermore, the report generation module is also used for:

[0185] Based on dynamic Bayesian networks, the marginal probabilities of each risk factor are analyzed to obtain the integrated risk probability distribution.

[0186] Based on the integrated risk probability distribution, the propagation path between various risk factors is analyzed to obtain the risk path analysis results.

[0187] Based on the risk path analysis results, intervention analysis is performed on the propagation path to obtain risk propagation simulation results; the risk propagation simulation results are used to characterize the influence range of risk factors in dynamic Bayesian networks.

[0188] Based on the results of risk propagation simulation, the impact of each propagation path is quantified to obtain a risk impact assessment matrix; the risk impact assessment matrix includes expected loss value, value at risk, and conditional value at risk.

[0189] Based on preset language organization rules, the risk impact assessment matrix is ​​transformed to obtain an engineering risk communication report.

[0190] In one embodiment, such as Figure 3 A computer device is provided, comprising:

[0191] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301: the memory stores application code executable by at least one of the processors, the application code being executed by at least one of the processors to enable at least one of the processors to perform a construction process engineering consulting management method as described above.

[0192] Computer equipment may also include: sensor 303.

[0193] The processor 301, memory 302 and sensor 303 can be connected via a bus or other means, with the bus being an example in the figure.

[0194] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0195] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0196] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for managing engineering consulting throughout the entire construction process, characterized in that, The method includes: Obtain local prior update data for each participant in the project; the local prior update data is used to characterize the dynamic propagation among various risk factors. Based on preset participant reputation weight parameters, the local prior update data of each participant is integrated to obtain normal prior update data; the participant reputation weight parameters include the reputation weight of each participant. The preset federated Bayesian network is updated based on the normal prior update data to obtain a dynamic Bayesian network; the federated Bayesian network includes nodes and connecting edges, the nodes are each of the risk factors, and the connecting edges are used to characterize the standard prior propagation probability between each of the risk factors. Risk propagation simulation is performed on the dynamic Bayesian network to obtain an engineering risk propagation report; the engineering risk propagation report is used to characterize the risk factors existing in the project and the impact caused by the risk factors.

2. The method according to claim 1, characterized in that, The local prior update data is obtained through the following method: Obtain the local project data of the participating object; the local project data is used to characterize the working status of the participating object; the local project data is labeled with a sampling time. The local project data is preprocessed to obtain preprocessed project data; Based on the preprocessed project data, the standard prior propagation probability between each node in the federated Bayesian network is updated to obtain the local posterior propagation probability; and the local posterior propagation probabilities between each node are concatenated to obtain the local posterior propagation parameter. Based on the local posterior propagation parameters and the federated Bayesian network, differential privacy noise is injected into the local posterior propagation probability to obtain the noisy posterior propagation probability. Extract the latest sampling timestamp from the local project data and determine it as the fresh sampling timestamp; and obtain the local prior update data based on the fresh sampling timestamp and the noisy posterior propagation probability.

3. The method according to claim 2, characterized in that, The method, based on preset participant reputation weight parameters, integrates the local prior update data of each participant to obtain normal prior update data, including: Based on the reputation weight parameter of the participating object and the fresh sampling timestamp, calculate the weighted weight of each participating object; For each node in the federated Bayesian network, the noisy posterior propagation probabilities of each participating object are integrated based on the weighted weights to obtain the global prior probability of each node. Based on preset propagation constraint rules, the global prior probabilities of each node are adjusted to obtain the complete global prior probabilities of each node. The normal prior update data is composed of the complete global prior probabilities of each node.

4. The method according to claim 3, characterized in that, For each node in the federated Bayesian network, based on the weighted weights, the noisy posterior propagation probabilities of each participating object are integrated to obtain the global prior probability of each node, including: For each node in the federated Bayesian network, the noisy posterior propagation probability of each participating object, including the node, is obtained to form a candidate noisy posterior propagation probability set. Based on the weighted weight of the participating object corresponding to the denoised posterior propagation probability, the denoised posterior propagation probability and the corresponding weight in the candidate denoised posterior propagation probability set are combined to form a denoised posterior propagation probability-weight pair; Based on the noisy posterior propagation probability-weight pair, the noisy posterior propagation probabilities are arranged in ascending order according to the corresponding weights to obtain the sorted propagation probability set; The weighted weights in the sorted propagation probability set are stacked in ascending order until a preset weight stacking threshold is met. The denoised posterior propagation probability corresponding to the last weighted weight participating in the stacking is selected as the global prior probability of the node.

5. The method according to claim 3, characterized in that, The method further includes: For each participating object, based on the noisy posterior propagation probability of the participating object and the global prior probability of each node, the deviation of the participating object is quantified to obtain the deviation degree of each participating object; Based on the deviation of each of the aforementioned participating objects, the standard deviation of the deviation is calculated; The participating objects whose deviation is greater than a preset deviation threshold are selected to form an abnormal participating object set; the participating objects whose deviation is less than the preset deviation threshold are selected to form a normal participating object set. For each participant in the set of abnormal participants, based on the deviation threshold, the preset penalty factor coefficient, the deviation standard deviation, the participant's reputation weight, and the deviation, the reputation weight of the participant is updated using the following formula: in, Participants Updated reputation weight, Participants Reputation weight, It is the penalty factor coefficient. Participants The degree of deviation It is the deviation threshold. It is the standard deviation of the deviation; For each participant in the set of normal participants, based on the deviation threshold, the preset reward factor coefficient, the participant's deviation, and the reputation weight, the reputation weight of the participant is updated using the following formula: in, Participants Updated reputation weight, Participants Reputation weight, It is the reward factor coefficient. Participants The degree of deviation It is the deviation threshold; Based on the updated reputation weight of the participating objects in the abnormal participating object set and the updated reputation weight of the participating objects in the normal participating object set, the updated reputation weight parameter of the participating object is obtained.

6. The method according to claim 3, characterized in that, The step of updating the preset federated Bayesian network based on the normal prior update data to obtain a dynamic Bayesian network includes: Based on the normal prior update data, update each node in the federated Bayesian network and the connection edges between each node to obtain the first federated Bayesian network. Based on the expectation-maximization algorithm, the connection edges between the nodes in the first federated Bayesian network are adjusted to obtain the second federated Bayesian network. Based on the second federated Bayesian network, the conditional probability distribution parameters among the nodes in the federated Bayesian network are adjusted to generate an optimized parameter set; The optimized parameter set is filled into the second federated Bayesian network to obtain the dynamic Bayesian network.

7. The method according to claim 1, characterized in that, The process of simulating risk propagation in the dynamic Bayesian network to obtain an engineering risk propagation report includes: Based on the dynamic Bayesian network, the marginal probabilities of each risk factor are analyzed to obtain the integrated risk probability distribution. Based on the integrated risk probability distribution, the propagation path between each risk factor is analyzed to obtain the risk path analysis results. Based on the risk path analysis results, intervention analysis is performed on the propagation path to obtain risk propagation simulation results; the risk propagation simulation results are used to characterize the influence range of the risk factor in the dynamic Bayesian network. Based on the risk propagation simulation results, the impact of each propagation path is quantified to obtain a risk impact assessment matrix; the risk impact assessment matrix includes expected loss value, risk value, and conditional risk value. Based on preset language organization rules, the risk impact assessment matrix is ​​transformed to obtain the engineering risk propagation report.

8. A construction whole-process engineering consulting and management device, characterized in that, The device includes: The data acquisition module is used to acquire local prior update data of each participant in the project; the local prior update data is used to characterize the dynamic propagation among various risk factors. The data prior module is used to integrate the local prior update data of each participating object based on the preset reputation weight parameters of the participating objects to obtain normal prior update data; the reputation weight parameters of the participating objects include the reputation weight of each participating object; The data update module is used to update the preset federated Bayesian network according to the normal prior update data to obtain a dynamic Bayesian network; the federated Bayesian network includes nodes and connecting edges, the nodes are each of the risk factors, and the connecting edges are used to characterize the standard prior propagation probability between each of the risk factors; The report generation module is used to perform risk propagation simulation on the dynamic Bayesian network to obtain an engineering risk propagation report; the engineering risk propagation report is used to characterize the risk factors existing in the project and the impact caused by the risk factors.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images