Multi-source heterogeneous data processing and collaborative decision-making computing system and processing method for urban update intelligent construction management platform

Abstract

The invention relates to a multi-source heterogeneous data processing and collaborative decision-making computing system and a processing method of a city updating intelligent construction management platform, and aims to solve the problems of non-uniform data processing, decision lagging, multi-device collaboration and the like during electric digital data processing. The system constructs a dynamic risk field through access and processing of multi-source heterogeneous data, and uniformly represents different risk information and updates the risk information along with time. Based on the risk field, a causal relationship model between risk factors is constructed, the risk evolution trend is analyzed, and possible chain reactions are predicted to provide a basis for decision making. The system generates a decision result through the multi-agent collaborative decision module, the decision result is dynamically adjusted through the decision strategy adaptive optimization module, and continuous adaptation and optimization of construction site environment and operation requirements are achieved. The invention provides an intelligent and efficient construction site platform management solution, and the decision accuracy and the system robustness are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of data processing and decision-making operations, specifically to a multi-source heterogeneous data processing and collaborative decision-making computing system and method for an intelligent construction management platform for urban renewal. Background Technology

[0002] With the improvement of informatization and digitalization at construction sites, intelligent construction data processing systems for urban renewal scenarios are gradually being applied to construction sites. These systems typically connect to multiple data acquisition terminals, including environmental monitoring sensors, personnel positioning devices, and equipment operation status acquisition units such as ventilation and lighting systems, to acquire multi-source heterogeneous operational data and perform centralized processing.

[0003] Existing construction site information processing systems mostly employ rule-based judgment mechanisms based on threshold comparisons. Specifically, the system logically compares real-time collected environmental parameter data and personnel location data with preset thresholds. When the data meets preset trigger conditions, it executes the corresponding alarm logic or invokes predefined control commands. This approach is essentially a rule-based condition-triggered data processing model, where various physical quantities participate in the judgment process as independent variables, lacking a unified data state modeling mechanism.

[0004] In terms of risk assessment, some existing technologies introduce risk field or risk heat map models to visualize the risk distribution within the construction area, assisting managers in status identification. However, such risk field models are usually only used as data mapping results for the display layer or as the basis for alarm triggering. They do not construct a unified, computable state space model. Different types of risk data (temperature, humidity, gas concentration, personnel density, etc.) still participate in the control logic as independent input variables, and the system lacks a data processing and collaborative decision-making computing mechanism based on a unified state representation.

[0005] In risk analysis, some technical solutions introduce causal inference algorithms or root cause analysis models to perform offline computational analysis on historical data after an anomaly occurs, in order to identify the causal chains that led to the anomaly. Other technologies use causal inference structures in the interpretation layer of machine learning models to enhance the interpretability of the model output. However, these causal models typically exist as post-event analysis or interpretation tools and are not embedded in real-time data processing flows to participate in online decision-making computation. They cannot calculate and predict the evolution trend of risk states based on causal constraints, nor can they be used to limit or optimize the candidate solution space of real-time control strategies.

[0006] In summary, the existing technology has the following technical problems:

[0007] First, while existing systems may use risk fields for risk visualization or alarms, these risk fields are not used as a unified state representation and control input for the system. Various physical quantities and risk information still participate in decision-making separately as independent data streams, resulting in a lack of multi-device collaborative control capabilities in a unified state space. This can lead to conflicts in control decisions from different devices or unreasonable resource allocation.

[0008] Second, while existing systems may employ causal reasoning methods in post-event analysis or model interpretation, causal reasoning is not used for real-time risk situation evolution assessment and control decision generation. The system cannot determine whether the current risk state has an inevitable evolutionary trajectory based on causal relationships, cannot identify the structural causes and incidental disturbances of risk, and cannot constrain the decision space or support predictive control through causal reasoning, resulting in delayed and untargeted decision responses.

[0009] Third, existing systems lack a complete control loop that integrates the risk field as a unified state space, causal reasoning as the evolutionary logic, and multiple agents as decision-making entities. The data processing, risk analysis, decision generation, and feedback optimization stages of the system lack organic coupling, failing to form a proactive and predictive collaborative control paradigm. Summary of the Invention

[0010] The purpose of this invention is to provide a multi-source heterogeneous data processing and collaborative decision-making computing system and method for an intelligent construction management platform for urban renewal, which solves the technical problems in the prior art such as the risk field not being used as a unified state space, causal reasoning not participating in real-time decision generation, and the lack of a complete control closed loop.

[0011] A multi-source heterogeneous data processing and collaborative decision-making computing system for an intelligent construction management platform for urban renewal includes:

[0012] The multi-source data access and preprocessing module is used to receive and aggregate multi-source heterogeneous data from the urban renewal intelligent construction management platform. The multi-source heterogeneous data includes environmental data collected by an environmental sensor array, spatial location information of personnel collected by a personnel positioning device, and monitoring data collected by a drone inspection device.

[0013] The implementation of the multi-source data access and preprocessing module is as follows: As the system's data input module, it is responsible for establishing communication connections with various data acquisition devices within the platform, receiving multi-source heterogeneous data, and performing preliminary processing. Specifically, the multi-source data access and preprocessing module interfaces with different types of devices through different communication protocols: it communicates with environmental sensor arrays via wired or wireless sensor network protocols to acquire environmental parameters such as temperature, humidity, gas concentration, and light intensity; it communicates with personnel positioning devices via UWB or Wi-Fi positioning protocols to acquire personnel's three-dimensional spatial coordinates and movement trajectories; and it communicates with UAV inspection equipment via video streaming protocols and data transmission interfaces to acquire image, video, and recognition result data.

[0014] The multi-source data access and preprocessing module performs time synchronization, format conversion, and spatial mapping on the received heterogeneous data. Time synchronization unifies data with different sampling frequencies to the system's standard clock; format conversion converts data of different formats into the system's internal standard data structure; and spatial mapping maps location-dependent data to the platform's three-dimensional spatial coordinate system. The preprocessed data is then transmitted in a structured form to the risk field construction module.

[0015] The risk field construction module is used to construct a computable dynamic risk field in the spatial and temporal dimensions of the platform based on the multi-source heterogeneous data. The dynamic risk field serves as an intermediate state representation of the system, used to uniformly characterize the comprehensive risk state corresponding to different spatial locations and different objects, and is called by subsequent modules.

[0016] The dynamic risk field is stored using a data structure that combines a three-dimensional spatial field with a time series, and the risk intensity value of each spatial grid cell is continuously updated over time. The risk field data is stored in the system's shared memory area for subsequent modules to read and access in real time.

[0017] The risk situation analysis module is used to construct a causal relationship model between risk factors based on the dynamic risk field, and to analyze and infer the evolution direction and constraints of the risk state through the causal relationship model.

[0018] The risk situation analysis module configures quantitative parameters for each causal edge, including a causal influence coefficient and a time delay parameter. The causal influence coefficient represents the intensity of the impact of upstream risk factors on downstream risk factors; the time delay parameter represents the time interval from the occurrence of the causal influence to its manifestation. During operation, the module performs forward inference and backward tracing based on the current risk field state through a causal relationship model. Forward inference predicts the evolution trend of each risk factor within a future time window, determining which risk factors may further deteriorate; backward tracing identifies the root cause nodes leading to the current high-risk state. The inference results include the risk evolution direction, evolution constraints, and key causal paths, which guide the strategy selection of the collaborative decision-making calculation module.

[0019] The collaborative decision-making calculation module is used to generate multi-device collaborative decision-making results based on the risk situation analysis results, and output the collaborative decision-making results as operation instruction data to change or constrain the operating status or data processing flow of relevant equipment on the platform, thereby realizing the collaborative scheduling of platform operation behavior.

[0020] The collaborative decision-making computation module generates multi-device collaborative decisions based on the risk situation analysis results. This module contains multiple functional software agents, each corresponding to a specific operational objective, such as ventilation scheduling objectives, lighting control objectives, personnel evacuation objectives, and equipment maintenance scheduling objectives.

[0021] Each software agent is configured with an objective function and constraints corresponding to its operational goals. The objective function defines the optimization objective pursued by the agent, such as minimizing risk intensity, minimizing energy consumption, or minimizing personnel exposure time; the constraints define the limitations that the agent's decisions must meet, such as equipment capacity limitations, resource budget limitations, or safety specification limitations.

[0022] Based on the shared dynamic risk field state, each software agent generates preliminary decision-making schemes from its own maintained candidate strategy library, incorporating evolutionary predictions and causal constraint information provided by the risk situation analysis module. The module integrates the preliminary schemes of each agent through a coordination mechanism, resolves potential goal conflicts, and generates a globally optimized multi-device collaborative decision-making result.

[0023] Collaborative decision-making results are output in the form of operational instruction data, including specific equipment control parameters, execution time series, and expected effects. Operational instructions are issued to the relevant execution devices on the platform through the system's equipment control interface, changing the equipment's operating status or data processing flow.

[0024] The risk field construction module is used to uniformly represent environmental risks, structural risks, personnel activity risks, and equipment operation risks, and map them to the spatial area of ​​the platform in the form of risk intensity values, wherein the risk intensity values ​​are associated with at least one of the probability of risk occurrence, scope of impact, or duration.

[0025] The specific method by which the risk field construction module uniformly represents environmental risk, structural risk, personnel activity risk, and equipment operation risk is as follows:

[0026] Regarding environmental risks, the module extracts environmental anomaly indicators such as temperature deviation, humidity anomaly index, and harmful gas concentration ratio, and conducts a quantitative assessment based on the degree of impact of each indicator on personnel health and equipment safety.

[0027] Regarding structural risks, the module extracts structural health indicators such as wall crack width, roof deformation, and supporting structure displacement based on UAV image recognition results, and assesses the risk level according to structural safety standards.

[0028] Regarding the risk of personnel activity, the module calculates activity characteristics such as personnel density, personnel movement speed, and personnel stay duration based on personnel location data, and conducts risk assessment in conjunction with space capacity and evacuation capabilities.

[0029] Regarding equipment operation risks, the module monitors the operating parameters of ventilation equipment, lighting equipment, and power supply equipment, and calculates operating status indicators such as equipment load rate, failure rate, and performance degradation.

[0030] The module normalizes the different types of risk information mentioned above, mapping them to a unified numerical range. For each spatial grid cell, the module calculates the weighted sum of various risk components to obtain the comprehensive risk intensity value at that location. The risk intensity value is related to the probability of risk occurrence, the scope of influence, and the duration: the probability of risk occurrence affects the base value of risk intensity; the scope of influence determines the diffusion coefficient of risk in the spatial dimension; and the duration affects the decay rate of risk in the temporal dimension.

[0031] The dynamic risk field is a risk distribution model that is continuously updated over time. It is used to reflect the evolution trend of risk status in the time dimension and to provide status input for subsequent risk situation analysis and decision-making operations.

[0032] The dynamic risk field, as a risk distribution model that is continuously updated over time, has the following update mechanism:

[0033] The risk field construction module recalculates the risk intensity values ​​of all spatial grid cells according to a preset update cycle (e.g., every 30 seconds or every minute). During each update, the module obtains the latest data from the multi-source data access and preprocessing module, re-extracts the risk characteristic variables, and calculates a new risk intensity distribution based on the latest characteristic values.

[0034] The updated risk field data, together with historical risk field data, constitutes a time series, used to reflect the evolution trend of risk status over time. The system retains historical risk field data within a certain time window (e.g., data from the most recent 1 hour or 24 hours) for trend analysis and evolution prediction.

[0035] The time series of the dynamic risk field provides state input for subsequent risk situation analysis. The risk situation analysis module compares the risk field state at different times, calculates the time derivative of the risk intensity, and determines the rate of risk growth or decay, thereby identifying the direction of risk evolution.

[0036] Meanwhile, the dynamic risk field provides real-time state input to the collaborative decision-making computation module. Each software agent in the collaborative decision-making computation module shares the latest state of the same dynamic risk field, ensuring that each agent makes collaborative decisions based on a consistent understanding of the environment.

[0037] The risk situation analysis module constructs a causal relationship diagram between risk factors to describe the causal relationships between different risk factors and their influence paths. The causal relationship diagram is used to constrain the evolution path of risk states or limit the candidate decision space of the collaborative decision calculation module.

[0038] First, the module determines the causal relationships between risk factors based on domain knowledge and physical mechanisms. For example, according to fluid mechanics principles, a decrease in ventilation equipment capacity leads to poor air circulation, which in turn causes the accumulation of harmful gases; according to structural mechanics principles, continuous vibration or load can lead to structural fatigue and crack propagation.

[0039] Secondly, the module analyzes historical event data to uncover statistical correlations between risk factors. For risk factor pairs that frequently exhibit a sequential relationship in historical data, the module determines that a causal relationship may exist and verifies the directionality of the correlation through a causal inference algorithm.

[0040] The module establishes directed edges for confirmed causal relationships and assigns a causal influence coefficient to each edge. The influence coefficient, set through historical data fitting or expert experience, represents the expected change in downstream risk factors when upstream risk factors change by one unit. A causal relationship graph is used to constrain the evolution path of risk states. When predicting risk evolution, the module only considers the evolution direction of paths existing in the causal graph, excluding evolution possibilities that do not conform to causal logic. This constraint mechanism improves the reliability of risk prediction. The causal relationship graph is also used to limit the candidate decision space of the collaborative decision-making computation module. For key influence paths identified through causal reasoning, the collaborative decision-making computation module prioritizes strategies that can block or weaken the path, avoiding the selection of ineffective strategies that contradict causal logic.

[0041] The risk situation analysis module is used to distinguish between incidental risk states caused by instantaneous anomalies and cumulative risk states caused by long-term accumulated factors, and adjusts the weights or constraints of risk evolution analysis based on the distinction results.

[0042] The risk situation analysis module distinguishes between incidental and cumulative risk states in the following way: For each risk factor, the module analyzes its historical evolution trajectory. If a risk factor suddenly increases in a short period of time without obvious upstream causal drivers, the module determines that the risk is incidental, possibly caused by external random disturbances or measurement errors.

[0043] If a risk factor shows a continuous upward trend, and can be traced back to the upstream root cause of its continuous effect through causal reasoning, the module determines that the risk is a cumulative risk caused by long-term cumulative factors.

[0044] The module calculates the time stability index of risk factors. The stability index is calculated based on the time variance and autocorrelation coefficient of the risk value. Low stability corresponds to sporadic risk, while high stability corresponds to cumulative risk.

[0045] Based on the differentiation of risk characteristics, the module adjusts the weights or constraints of risk evolution analysis. For incidental risks, the module reduces their weight in evolution prediction to avoid overreacting to short-term fluctuations; for cumulative risks, the module increases their weight in evolution prediction and extends the prediction time window to fully assess their long-term impact.

[0046] The differentiation results are also passed to the collaborative decision-making calculation module. For incidental risks, the decision-making module adopts an observation and waiting strategy or low-cost, mild intervention; for cumulative risks, the decision-making module adopts a root cause elimination strategy or structural adjustment measures.

[0047] The collaborative decision-making computing module includes multiple functional software agents, each corresponding to different equipment operation or management objectives, and each configured with an objective function and constraints corresponding to the objective.

[0048] Based on the shared dynamic risk field state, each software agent generates the multi-device collaborative decision-making result through collaboration or game theory, according to its own objective function and constraints.

[0049] The collaborative decision-making computation module comprises multiple functional software agents that implement collaborative decision-making in the following manner:

[0050] Each software agent, as an independent decision-making unit, has a clearly defined operational objective. The objective function of the agent defines the direction of optimization it seeks. For example, the objective function of the ventilation scheduling agent is to minimize air quality risk, while the objective function of the lighting control agent is to minimize lighting energy consumption while ensuring minimum illuminance requirements.

[0051] Each agent is configured with constraints corresponding to its objectives. Constraints include hard constraints and soft constraints. Hard constraints are restrictions that must be met, such as the maximum power of the equipment or the minimum requirements of safety regulations; soft constraints are conditions that are preferred to be met, such as energy consumption budgets or comfort expectations.

[0052] Each agent, sharing the same dynamic risk field state, independently generates a preliminary decision plan based on its own objective function and constraints. The preliminary plan includes the proposed control actions, the expected risk reduction effect, and the required resource consumption.

[0053] When initial plans from multiple agents may conflict, the module initiates a coordination mechanism. This mechanism includes a negotiation mode and a game-theoretic mode. In the negotiation mode, agents negotiate their respective actions and resource allocations through multiple rounds of communication to seek a Pareto optimal solution. In the game-theoretic mode, agents make policy choices based on a game theory framework to seek a Nash equilibrium.

[0054] The collaborative process considers causal constraint information provided by the risk situation analysis module. For key intervention paths determined through causal reasoning, relevant agents receive higher decision priority. The collaborative mechanism ensures that the final multi-device collaborative decision-making result achieves global optimization, avoiding suboptimal overall outcomes caused by local optima.

[0055] The system also includes a decision strategy adaptive optimization module, which is used to collect operational result feedback information after the execution of operational instruction data, and continuously optimize the multi-device collaborative decision results based on the feedback information.

[0056] The adaptive optimization module for decision-making strategies adjusts the decision preferences, strategy selection order, or collaboration methods of the software agent to achieve adaptive evolution of decision-making and operational strategies.

[0057] And a method for multi-source heterogeneous data processing and collaborative decision-making computation of an intelligent construction management platform for urban renewal, comprising the following steps:

[0058] Acquire multi-source heterogeneous data from the urban renewal intelligent construction management platform. The multi-source heterogeneous data includes environmental data collected by an environmental sensor array, spatial location information of personnel collected by a personnel positioning device, and monitoring data collected by a drone inspection device.

[0059] Based on the multi-source heterogeneous data, risk information from different data sources is fused and processed to construct a computable dynamic risk field in the spatial and temporal dimensions of the platform. The dynamic risk field serves as an intermediate state representation of the system and is used to uniformly characterize the comprehensive risk state corresponding to different spatial locations and different objects.

[0060] Based on the dynamic risk field, a causal relationship model between risk factors is constructed. The evolution direction and constraints of the risk state are analyzed and inferred through the causal relationship model. In the analysis and inference, a physical simulation model is introduced as a constraint condition for risk propagation to limit the evolution range of the risk state in the spatial and temporal dimensions.

[0061] Before generating multi-device collaborative decision results, it is determined whether the risk state meets the preset evolution conditions. If the evolution conditions are met, a strategy combination that meets the preset operation objectives is selected from multiple candidate operation strategies based on the risk situation analysis results to generate multi-device collaborative decision results.

[0062] The multi-device collaborative decision-making results are output as operational instruction data to change or constrain the operating status or data processing flow of relevant platform devices.

[0063] After the operation instruction data is executed, operation result feedback information is collected, and the subsequent decision-making operation process is adaptively adjusted based on the feedback information.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] I. Paradigm Innovation at the System Architecture Level: This invention constructs a new control closed-loop paradigm by using a dynamic risk field as the unified state space of the system, causal reasoning as the state evolution logic, and multiple agents as decision-making executors. The creativity of this paradigm lies not in the individual use of risk fields or causal reasoning techniques, but in the organic coupling of these three elements:

[0066] (1) The dynamic risk field is no longer just a visualization tool or alarm basis, but a unified state representation of a multi-device collaborative control system. Environmental risks, structural risks, personnel activity risks, and equipment operation risks from different data sources are mapped to the same risk field space, enabling decision-making agents of different devices to collaborate under a unified state language, avoiding equipment conflicts and unreasonable resource allocation caused by the independent participation of various physical quantities in decision-making in traditional methods.

[0067] (2) Causal reasoning is no longer used only for post-hoc root cause analysis or model explanation, but directly participates in the real-time risk situation evolution judgment and control strategy selection constraints. The system judges whether the risk state has the inevitability of evolution based on the causal relationship model, distinguishes between structural risks and occasional disturbances, and identifies key causal propagation paths, thereby supporting predictive decision-making and proactive intervention against root causes.

[0068] (3) Based on sharing the same risk field state and accepting causal reasoning constraints, the multi-agent collaborative decision-making mechanism generates a globally optimized strategy combination through negotiation or game theory, realizing a qualitative change from "passive response" to "active prediction" and from single point to collaborative control.

[0069] Second, compared to the independent judgment of each physical quantity in traditional systems, the risk field, as a unified state space, offers the following advantages in this invention: Different types of risk information (environment, structure, personnel, equipment) are uniformly represented as continuous risk intensity values, enabling the system to conduct comprehensive risk assessments within a unified framework, rather than making decentralized judgments. As a shared environment for multiple agents, the risk field provides a unified decision-making basis for agents with different operational objectives, allowing ventilation scheduling agents, lighting control agents, and personnel evacuation agents to collaborate based on consistent state cognition, avoiding decision-making conflicts. The spatial-temporal continuity of the risk field provides a foundation for risk evolution analysis and prediction. Compared to discrete threshold judgments, a continuous state space can more precisely characterize the gradual process and propagation path of risk.

[0070] Third, compared to traditional systems where causal reasoning is only used for post-hoc analysis or model explanation, this invention applies causal reasoning to real-time decision generation, achieving the following effects: The system can determine the inevitability of the evolution of a risk state. By calculating the cumulative influence between risk factors through a causal relationship model, the system can predict whether a certain risk state will naturally subside or continue to worsen, thereby avoiding over-response to occasional disturbances. The system can identify the root cause of a risk rather than its surface phenomena. When a risk is detected... When concentrations rise, the system uses causal reasoning to determine whether the cause is a ventilation equipment malfunction or excessive population density, thus intervening at the root cause rather than asymptomatic symptoms, improving the targeting and effectiveness of decision-making. The system can constrain the decision space through causal reasoning. For risk states that may trigger chain reactions, the system pre-identifies causal propagation paths and deploys preventative measures, achieving a shift from "post-event response" to "pre-event prevention." The combination of causal reasoning and physical simulation models limits the physical boundaries of risk evolution prediction, preventing reasoning results from violating physical laws and improving the reliability of predictions and the rationality of decisions.

[0071] This invention establishes a complete closed loop encompassing data acquisition, risk field construction, causal reasoning analysis, multi-agent decision-making, operational execution, and feedback optimization. This enables the adaptive optimization module to dynamically adjust risk field weight coefficients, causal model parameters, and agent decision preferences based on operational feedback, allowing the system to adapt to changes in different construction site scenarios and operational stages, thus enhancing its robustness. Furthermore, feedback information is used not only to adjust control parameters but also to update the causal relationship model, continuously optimizing the system's causal reasoning capabilities throughout the operational process, forming a closed-loop evolutionary mechanism of "data-model-decision-feedback." Essentially, this invention elevates the risk field from a visualization tool to a unified state space for the control system, transforms causal reasoning from a post-hoc analysis tool into a real-time decision constraint mechanism, and constructs a new paradigm of proactive, predictive, collaborative control for intelligent construction management platforms in urban renewal through multi-agent collaborative decision-making and adaptive feedback optimization. Attached Figure Description

[0072] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort. In the drawings:

[0073] Figure 1 This is a system architecture diagram of a multi-source heterogeneous data processing and collaborative decision-making computing system for an intelligent construction management platform for urban renewal.

[0074] Figure 2 This is a flowchart of a multi-source heterogeneous data processing and collaborative decision-making computation method for an intelligent construction management platform for urban renewal. Detailed Implementation

[0075] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0076] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0077] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0078] Example 1

[0079] See Figure 1 , Figure 2 This embodiment provides a specific implementation of a multi-source heterogeneous data processing and collaborative decision-making computing system for an intelligent construction management platform for urban renewal. The system is deployed in the server cluster of the construction site platform management platform. Through a modular software approach, it achieves real-time processing of multi-source data within the platform, dynamic modeling of risk states, causal reasoning analysis of risk evolution trends, and multi-device collaborative decision-making and operation scheduling based on the analysis results.

[0080] The system receives real-time data streams from different data sources on the construction site platform through a multi-source data access and preprocessing module. Specifically, the environmental sensor array collects environmental parameters such as temperature, humidity, gas concentration, and light intensity at a sampling frequency of once per minute; the personnel positioning device obtains the three-dimensional spatial coordinates of personnel within the platform using UWB ultra-wideband technology, updating at a frequency of ten times per second; and the UAV inspection equipment collects video image data and infrared thermal imaging data during inspection flights, and extracts structural anomaly features and equipment operating status information through an image recognition module.

[0081] In this embodiment, the system adopts a hierarchical data processing architecture, with each functional module communicating through a unified data bus. The system uses a dynamic risk field as the core intermediate state representation, performing spatiotemporal alignment, feature extraction, and fusion processing on data from different data sources, and completing the modeling and updating of risk states within a unified state space. Instead of directly controlling and judging the original physical quantities independently, each module performs collaborative calculations by sharing the state vector of the dynamic risk field, thereby achieving decoupling and collaboration between the data layer, model layer, and decision layer.

[0082] Due to differences in sampling frequency, data format, and time base among the aforementioned data sources, the system performs the following preprocessing operations in the multi-source data access and preprocessing module: First, all timestamps are uniformly mapped to the system's standard clock, and linear interpolation is used to time-align data from different frequencies; second, personnel spatial location information and UAV monitoring results are mapped to the platform's three-dimensional spatial grid model, with a grid size set to 1 meter × 1 meter × 1 meter; finally, sliding window filtering is performed on the sensor data, with a window length of 5 minutes, to eliminate transient noise interference. The preprocessed data is then transmitted to the risk field construction module in the form of structured data packets.

[0083] The risk field construction module transforms the preprocessed multi-source data into a unified risk state representation. In this embodiment, the risk field is defined as a combination structure of a three-dimensional spatial field and a time series, where each spatial grid cell corresponds to a risk intensity value, which is dynamically updated over time.

[0084] Specifically, the system uses environmental risk, structural risk, personnel activity risk, and equipment operation risk as four basic risk characteristics. For environmental risk, the system extracts indicators such as the degree of temperature exceeding the standard, the concentration of harmful gases, and abnormal humidity; for structural risk, the system extracts features such as wall cracks, roof deformation, and displacement of supporting structures based on UAV image recognition results; for personnel activity risk, the system assesses it based on parameters such as personnel density, movement speed, and duration of stay; and for equipment operation risk, the system monitors the degree of deviation of operating parameters of ventilation equipment, lighting equipment, and power supply equipment.

[0085] The system provides each spatial grid cell The risk intensity value R(i, j, k, t) at time t is calculated as follows:

[0086] ;

[0087] in, As an environmental risk component, As a structural risk component, Assess the risk level of personnel activities. Assign risk components to equipment operation; For the corresponding weight coefficients, satisfying The numerical range of each risk component is normalized to the interval [0, 1], where 0 represents no risk and 1 represents extremely high risk. The weighting coefficients are initially set based on historical accident data and expert experience, and are dynamically adjusted during subsequent adaptive optimization.

[0088] In order to dynamically adjust the weighting coefficients based on historical data and expert experience In this embodiment, the following strategy is adopted:

[0089] a) Initial settings: Weighting coefficients at system startup The settings are based on historical accident data and expert experience. For example:

[0090] ; ; ; ;

[0091] b) Adaptive Adjustment: Feedback Mechanism: During system operation, as real-time data accumulates, the weighting coefficients... It will adaptively adjust through a feedback mechanism. Whenever the system detects a historical risk event (such as equipment failure, personnel gathering, etc.), it will dynamically optimize the weight coefficients through a Bayesian update method to ensure the model's adaptability to the current construction site environment.

[0092] In this invention, Bayesian updating is used to adjust the weight coefficients of each risk factor to improve the accuracy of the system's risk assessment. Weight coefficients The update is based on the following formula:

[0093] ;

[0094] in:

[0095] Represents given data After (real-time feedback data), weighting The posterior distribution of; Indicates a given weight In the case of observed data Likelihood; Indicates weight The prior distribution can usually be initialized using historical data. In this embodiment, the key steps of Bayesian update are as follows:

[0096] Initialization: Initial weighting coefficients are determined based on historical accident data and expert experience. Configure the settings. For example, based on historical risk assessment, initialize the weights as follows: .

[0097] Real-time data acquisition: Whenever the system receives new sensor data or a new safety event occurs (such as equipment failure or changes in personnel density), the data will be input into the system through the multi-source data access and preprocessing module.

[0098] Update weighting coefficients: The system updates the weighting coefficients using the newly collected data. Assuming At a certain point in time The weight coefficients are updated based on the collected feedback data by calculating the posterior distribution:

[0099] ;

[0100] in, Indicates at time The adjustment amount of the weighting coefficients is based on the current risk assessment results and historical data feedback of the system.

[0101] Subsequently, the system further optimizes the weight coefficients using gradient descent. This ensures the accuracy and timeliness of decision-making results, especially in complex and dynamically changing construction site environments.

[0102] Gradient descent is used to optimize weight coefficients by automatically adjusting the values ​​of each weight by minimizing the loss function.

[0103] First, define the loss function: The system defines a loss function. This function measures the deviation between the predicted risk state of the system and the actual outcome. The loss function can be expressed as:

[0104] ;

[0105] in, : Represents the system's loss function, which measures the error between predicted risk and actual risk. The number of data points represents the risk assessment results of the construction site at different times. For the construction site at the first The actual risk intensity value at any given moment is derived from real-time sensor data. For the system based on the current weight coefficients ( The predicted risk intensity value calculated.

[0106] Then, gradient calculation is performed, where the system calculates the gradient of the loss function with respect to each weight:

[0107] ;

[0108] Where represents the loss function with respect to the weights. The gradient measures the weights. The impact of adjusting the loss function. Indicates the predicted risk intensity Weights The derivative of the weight indicates the weight. How do changes affect risk prediction results?

[0109] The purpose of gradient calculation is to find the direction of adjustment for each weight coefficient. By calculating the derivative of the loss function with respect to each weight, the system can determine how to adjust the weight coefficients so that the loss function gradually decreases.

[0110] The loss function calculates the sum of squares of the differences between the actual risk and the predicted risk; the smaller the difference, the more accurate the system's prediction.

[0111] Finally, gradient descent is performed to update the weight coefficients based on the calculated gradient.

[0112] ;

[0113] Where: represents the updated weight coefficient. This represents the weighting coefficient at the current moment. The learning rate controls the size of each update step. Generally, a smaller learning rate ensures stable convergence but results in slower updates; a larger learning rate speeds up convergence but may lead to instability. : Represents the gradient, indicating the sensitivity of the loss function to the weight coefficients. In each update, gradient descent updates the weights by subtracting the gradient multiplied by the learning rate. The weights are gradually adjusted in the direction of reducing the loss function until an optimal solution is found.

[0114] The constructed dynamic risk field serves as an intermediate state representation of the system and is stored in a shared memory area for subsequent use by the risk situation analysis module and the collaborative decision-making calculation module. The risk field is updated every 30 seconds, forming a time series. It is used to reflect the evolution of risk status.

[0115] The risk situation analysis module constructs a causal relationship model between risk factors based on a dynamic risk field. This model represents the causal relationship paths between different risk factors in the form of a directed acyclic graph. In this embodiment, the system establishes a causal relationship graph containing 15 main risk factor nodes by analyzing historical risk event records and physical constraint rules.

[0116] For example, there is a causal edge between the "ventilation equipment failure" node and the "increased concentration of harmful gases" node, and the "increased concentration of harmful gases" node further points to the "personal health risk" node; the "rainstorm weather" node points to the "flooding" node and the "abnormal humidity" node, and the "abnormal humidity" node in turn points to the "electrical equipment risk" node. The system configures propagation strength parameters and time delay parameters for each causal edge to quantify the degree and timing of causal influence.

[0117] During operation, when the system detects that a certain risk factor value exceeds a preset threshold, the situation analysis module performs forward reasoning through a causal relationship graph to predict potentially activated downstream risk factors and their evolution time windows. Simultaneously, the system performs reverse causal reasoning to identify the upstream root causes leading to the current risk state. This bidirectional causal reasoning result is used to constrain the decision space: for risk states traceable to a clear root cause, the system prioritizes intervention strategies targeting the root cause; for risk states that may trigger chain reactions, the system pre-deploys preventative measures to block causal propagation paths.

[0118] To limit the evolution of risk states in both spatial and temporal dimensions, the system introduces physical simulation models as constraints. For example, for the risk of hazardous gas diffusion, the system calls the fluid dynamics simulation module to calculate the diffusion rate and impact range of the gas within the platform. The simulation results are used to correct the risk evolution predictions obtained through causal reasoning, preventing the reasoning results from violating physical laws.

[0119] The collaborative decision-making computation module comprises multiple functional software agents, each corresponding to different operational objectives such as ventilation scheduling, lighting control, personnel evacuation guidance, and equipment maintenance scheduling. Each software agent is configured with an objective function and constraints corresponding to its objective. For example, the objective function of the ventilation scheduling agent is to minimize air quality risk, with constraints including ventilation equipment power limits and energy consumption budgets; the objective function of the personnel evacuation guidance agent is to minimize personnel exposure time, with constraints including evacuation passage capacity and personnel movement speed limits.

[0120] Before generating collaborative decisions, the system first determines whether the current risk status meets the decision triggering conditions. Specifically, when the risk intensity value of any spatial grid exceeds 0.6, or the time derivative of the risk intensity value is greater than 0.05 / minute, the system determines that the decision operation process needs to be initiated.

[0121] Based on the shared dynamic risk field state, each software agent generates preliminary decision-making schemes from its own maintained candidate strategy library, using evolutionary predictions and causal constraint information provided by the risk situation analysis module. Candidate strategies include, but are not limited to: increasing ventilation power, activating backup lighting, sending personnel evacuation instructions, and adjusting equipment inspection routes. The system evaluates the risk reduction effect, resource consumption cost, and feasibility of each candidate strategy.

[0122] Because the goals of different agents may conflict, the system employs a negotiation mechanism to generate collaborative decision-making results. In this embodiment, the system sets up a central coordinator. Each agent submits a preliminary plan and its priority to the coordinator. The coordinator, based on the global risk situation and resource constraints, solves for the Pareto optimal strategy combination using a multi-objective optimization algorithm. The final determined strategy combination is output in the form of operational instruction data, including specific equipment control parameters, execution time series, and expected performance indicators. Because the goals of different agents may conflict, the system employs a negotiation mechanism to generate collaborative decision-making results. In this embodiment, the system sets up a central coordinator as the scheduling hub for multi-agent negotiation. The specific steps of the negotiation process are as follows:

[0123] Based on the current dynamic risk field state and risk situation analysis results, each agent in S1 selects a preliminary decision plan from its respective candidate strategy library. Each preliminary plan includes the following elements: a description of the proposed control action, a list of required resources, the expected risk reduction, and the estimated execution cost.

[0124] Each agent in S2 calculates the priority of its preliminary plan. The priority calculation comprehensively considers the following factors: the severity of the current corresponding risk factor, the critical path weight of the risk factor in the causal relationship graph, the rate of risk evolution, and historical execution effect feedback. The priority is represented by a normalized value, ranging from [0, 1], with a larger value indicating a more urgent plan.

[0125] Each agent in S3 submits a preliminary plan data packet to the central coordinator. The data packet uses a standardized format and includes fields such as agent identifier, plan content, priority value, resource requirement vector, and expected effect vector.

[0126] After receiving all the solutions submitted by the agents, the S4 central coordinator constructs a global optimization problem:

[0127] Specifically, the coordinator first counts all the preliminary proposals submitted so far. Assume there are N agents submitting N proposals, denoted as... The coordinator defines decision variables for each scenario. ,in Representation scheme Selected to execute Representation scheme It will not be executed.

[0128] The coordinator constructs a global objective function. In this embodiment, weighted benefit maximization is used as the optimization objective:

[0129] ;

[0130] Let i be the priority weight of the i-th option. Let the expected risk reduction effect value be the i-th option. Let be the execution cost of the i-th scheme, and λ be the cost penalty coefficient. The objective function represents maximizing the weighted risk reduction effect under cost constraints.

[0131] The coordinator also constructs constraints. These constraints include resource constraints and logical constraints:

[0132] Resource constraints are represented as:

[0133] ;

[0134] in For the plan The demand for the j-th type of resource, For the plan The total available resources for resource type j. This constraint ensures that the total resource requirements of all selected options do not exceed the available resources.

[0135] Time conflict constraint is expressed as: if the solution and plan If the same device needs to be used within overlapping time periods, then This means that the two plans cannot be executed simultaneously.

[0136] Logical mutual exclusion constraints are expressed as: if the solution and plan Logically mutually exclusive (e.g., one requires the device to be turned on, and the other requires the device to be turned off), then .

[0137] S5 coordinator detects conflicts between schemes:

[0138] The coordinator uses the following method for conflict detection: For the detection of resource contention conflicts, the coordinator establishes a resource demand matrix. Where row index i corresponds to the scheme and column index j corresponds to the resource type. The coordinator calculates the total resource requirement assuming all schemes are executed simultaneously. If there exists a resource j such that... If so, a resource contention conflict is determined. The coordinator further identifies the set of schemes competing for the resource: .

[0139] For detecting target adversarial conflicts, the coordinator analyzes the effect interactions between schemes. Specifically, the coordinator maintains an effect interaction matrix. ,in Representation scheme Regarding the plan The impact coefficient of the effect. If , indicating the scheme The implementation will weaken the plan The effect is then assessed, and at this point, it is determined that there is a conflict of objectives between the two solutions. The interaction coefficient is set based on historical data statistics or expert knowledge. For example, if historical data shows that the "activate ventilation" solution is 30% less effective in "densely populated" conditions, then the corresponding interaction coefficient is set to -0.3.

[0140] For detecting logical mutual exclusion conflicts, the coordinator checks the logical consistency of the scheme content. The coordinator extracts the relevant device objects and target states for each scheme and constructs a state change list. If two schemes require different target states for the same device object (e.g., scheme...), the coordinator will detect these conflicts. Device A needs to be turned on. Solution If device A is required to be turned off, then a logical mutual exclusion conflict is determined to exist.

[0141] The coordinator records all detected conflicts as a conflict set. Each conflicting element It includes information such as the type of conflict, the set of solutions involved, and the severity of the conflict.

[0142] For detected conflicts, the coordinator adopts the following resolution strategy:

[0143] For resolving resource contention conflicts: the coordinator first allocates resources based on the priority of the solutions. For the set A of solutions competing for resource j... j The coordinator is based on priority. The resources are sorted from highest to lowest priority. The coordinator uses a priority allocation algorithm: it prioritizes the resource requirements of high-priority solutions, and when the remaining resources are insufficient to meet the requirements of subsequent solutions, the subsequent solutions are marked as delayed or degraded for execution.

[0144] For a scenario where a downgrade is acceptable, the coordinator calculates the downgrade plan.

[0145] For resolving adversarial conflicts, the coordinator uses a multi-objective optimization algorithm to find the Pareto optimal policy combination. Specifically, the coordinator defines the multiple adversarial objectives as a vector objective function:

[0146] ;

[0147] in Let i be the objective function of the i-th agent. This is a vector of decision variables.

[0148] The coordinator uses a weighted summation method to transform a multi-objective problem into a single-objective problem. Weight vector. Based on the priority settings of each scheme, the following conditions are met. The single objective function is:

[0149] ;

[0150] The coordinator considers the effect interaction matrix M as a modification to the objective function. The modified objective function is:

[0151] ;

[0152] This amendment reflects the synergistic or antagonistic effects between the options, i.e., when When, it indicates the plan The implementation will weaken the plan The effect; when When the value is greater than 0, it indicates that there is a synergistic enhancement effect between the two schemes.

[0153] The coordinator obtains the optimal decision variables by solving the corrected optimization problem. .for The coordinator further optimizes the execution order of the scheme. If the scheme is detected... and A causal dependency exists, which is identified through a causal relationship graph, and the process is executed first. Can be improved If the effect is the same, then the execution order will be adjusted to... Prior to .

[0154] When a satisfactory solution cannot be found through weighted summation, the coordinator adopts... Constraint method. This method transforms n-1 objectives into constraints, optimizing only one primary objective. For example, it focuses on the highest priority objective. As the primary objective, other objectives must meet the following requirements. ,in This represents the minimum acceptable value for the i-th objective.

[0155] For resolving logical mutual exclusion conflicts: the coordinator evaluates the overall benefits of the mutual exclusion schemes. For logical mutual exclusion schemes... The coordinator calculates the benefit-cost ratio for each option; it selects the option with the higher benefit-cost ratio to add to the execution set, while excluded options are added to the candidate queue. Options in the candidate queue can be re-evaluated and executed when conditions are met (such as the mutual exclusion condition being lifted or the execution plan being completed).

[0156] The coordinator also considers root cause information provided by causal reasoning. If a certain solution... If the targeted risk factor is identified as the root cause of other risks through causal attribution, the scheme gains additional priority weighting. Priority weight updated to , where β is the root cause priority coefficient, usually set to 0.2-0.5.

[0157] The S7 coordinator generates the final collaborative decision results. These results include: the set of approved strategies, the execution time window for each strategy, the resource allocation to each strategy, the execution order or parallel relationship between strategies, and the expected global risk reduction effect.

[0158] The S8 coordinator converts collaborative decision-making results into operational instruction data. This operational instruction data is formatted according to equipment type and control interface specifications, including: equipment identifier, control parameters, execution timestamp, duration, and expected performance indicators. The operational instruction data is then distributed to the relevant execution equipment in the capsule warehouse via the system's equipment control interface, triggering changes in equipment operating status or data.

[0159] Operational instructions are sent to relevant equipment on the platform, such as ventilation, lighting, and personnel positioning systems, through the system's device control interface, directly altering the equipment's operating status. Simultaneously, some instructions affect the data processing flow itself, such as adjusting sensor sampling frequencies, modifying anomaly detection thresholds, and activating specific monitoring modes.

[0160] After the operational instructions are executed, the system continuously monitors changes in the risk status within the platform and collects operational result feedback information after 5 minutes. The feedback information includes: the actual decrease in risk intensity value, equipment response time, actual resource consumption, and personnel feedback.

[0161] The adaptive optimization module for decision-making compares the feedback information with the expected results before the decision is executed, and calculates the decision bias. If the actual risk reduction is less than 80% of the expected value, the system determines that the decision is ineffective and records the failure scenario characteristics of the solution. Based on the accumulated feedback data, the system adjusts the decision preference parameters of each software agent through a reinforcement learning mechanism, such as increasing the selection weight of a certain type of strategy, modifying the cost coefficient in strategy evaluation, and updating the preconditions for strategy execution.

[0162] In addition, the system adjusts the weighting coefficients in the risk field construction module based on feedback information. If historical feedback indicates that structural risks could lead to severe consequences, the system will automatically upgrade its strategy. The numerical values ​​make the risk field more sensitive to structural anomalies. Weight adjustment employs a gradient descent method, with the optimization objective being to minimize the cumulative sum of squared decision biases.

[0163] Through the above feedback and adaptive adjustment process, the system forms a complete closed loop of data acquisition, risk modeling, situation analysis, decision-making and operation, and effect evaluation, realizing continuous optimization and intelligent management of the operation process of the urban renewal intelligent construction management platform.

[0164] Example 2

[0165] This embodiment provides another implementation of a multi-source heterogeneous data processing and collaborative decision-making computing system and method for an intelligent construction management platform for urban renewal. Compared with Embodiment 1, this embodiment focuses on strengthening the role of causal reasoning mechanisms in risk situation analysis. By constructing a quantitative causal influence model among risk factors, it achieves the prediction of risk evolution trends and directly constrains the selection space of decision-making and operational strategies based on the causal reasoning results, thereby improving the pertinence and effectiveness of decision-making.

[0166] The system receives real-time data streams from environmental sensor arrays, personnel positioning devices, and drone inspection equipment via a multi-source data access and preprocessing module. The environmental sensor array collects data including temperature, humidity, and... Environmental parameters such as concentration and PM2.5 concentration; personnel positioning devices obtain the real-time location coordinates and movement trajectory of personnel within the platform; drone inspection equipment obtains structural health status and equipment operation status information through visual recognition and infrared detection.

[0167] During the data preprocessing stage, the system maps the aforementioned multi-source data to a unified spatiotemporal reference framework and extracts risk-related characteristic variables from it. Specifically, the system defines 12 categories of risk characteristic variables, including: temperature deviation, humidity anomaly index, ratio of harmful gas concentrations, population density, duration of personnel stay, structural deformation, crack propagation rate, ventilation equipment load rate, lighting equipment failure rate, power supply voltage fluctuation amplitude, ground water depth, and external weather level. Each characteristic variable is normalized and mapped to the interval [0, 1], where 0 represents a normal state and 1 represents an extreme abnormal state.

[0168] The system constructs a dynamic risk field based on the aforementioned risk characteristic variables. The risk field is represented by a combination of a three-dimensional spatial grid and a time series, with each spatial grid cell corresponding to a set of risk characteristic vectors. The system weights and fuses each risk characteristic to generate a comprehensive risk intensity value for each spatial location at each time point. This risk field serves as an intermediate state representation of the system and is stored in a shared data area for subsequent use by the causal inference module and the collaborative decision-making calculation module. The risk field is updated every minute, forming a time series data stream.

[0169] The core technical feature of this embodiment is to construct a causal relationship model between risk factors and to make quantitative inferences about the evolution trend of risks based on this model.

[0170] The system first constructs a directed graph of causal relationships G=(V, E), where V is the set of risk factor nodes, containing the aforementioned 12 types of risk characteristic variables; E is the set of causal edges, and each directed edge... This indicates that risk factor i has a causal impact on risk factor j. The establishment of causal relationships is based on physical mechanism analysis and historical event correlation mining. For example, the "ventilation equipment load rate" node points to the "harmful gas concentration ratio" node, indicating that a decrease in ventilation capacity will lead to the accumulation of harmful gases; the "external weather level" node points to the "ground water depth" node and the "humidity anomaly index" node; the "humidity anomaly index" node further points to the "power supply voltage fluctuation range" node, reflecting the impact of a humid environment on electrical equipment.

[0171] To achieve quantitative reasoning, the system assigns a causal influence coefficient to each causal edge. and time delay parameters The causal impact coefficient represents the intensity of the influence of upstream risk factors on downstream risk factors, with a value range of [0, 1]. The time delay parameter represents the time interval required for the causal impact to manifest from its occurrence, in minutes.

[0172] During operation, the system uses the risk characteristic state vector at the current time t. Predicting the next moment using a causal relationship model The evolution values ​​of each risk factor. Specifically, for risk factor j, its value at time... Predicted value Calculate as follows:

[0173] ;

[0174] The summation iterates through all upstream nodes pointing to node j. ; Upstream risk factors The state value before the delay time τ; The trigger threshold for upstream risk factors, when Below this threshold, it is considered that no significant causal effect is produced; This is the Heaviside step function, which takes a value of 1 when the independent variable is greater than 0, and a value of 0 otherwise. It is used to filter weak causal influences. This calculation expression reflects the cumulative effect of risk evolution: the growth of downstream risk depends on the weighted sum of the effects of multiple upstream risk factors.

[0175] The system will predict the value Compared with current observations Compare, if the predicted increment If the value is greater than 0.1, risk factor j is determined to have an evolutionary trend and is marked as a "potential evolutionary risk". The system further identifies the key upstream root cause nodes leading to this evolutionary trend by tracing back through a causal graph.

[0176] To prevent causal reasoning from violating physical constraints, the system introduces a physical simulation model as evolution boundary conditions. For example, for the evolution prediction of "harmful gas concentration ratio," the system calls the fluid diffusion simulation module to calculate the upper limit of gas diffusion and the decay rate in a closed space. If the predicted value obtained from causal reasoning exceeds the reachable range of the physical simulation, the system corrects the predicted value to the physical boundary value. Similarly, for the evolution prediction of "structural deformation," the system calls the structural mechanics simulation module to verify whether the deformation conforms to the laws of material mechanics, avoiding unreasonable evolution predictions.

[0177] The collaborative decision-making computation module generates targeted operational strategies based on the results of causal reasoning analysis. In this embodiment, the system's decision logic considers not only the current risk intensity but also the causal evolution trend and root cause identification.

[0178] The system maintains a candidate operational strategy library, with each strategy corresponding to a set of equipment control instructions or process adjustment operations. For example, the "enhanced ventilation strategy" includes operations such as increasing the speed of ventilation equipment and opening backup exhaust vents; the "personnel evacuation strategy" includes operations such as sending evacuation notices, activating safety passage lighting, and closing access control to hazardous areas; and the "equipment maintenance strategy" includes operations such as suspending the operation of specific equipment, initiating fault diagnosis programs, and dispatching maintenance personnel.

[0179] When generating decisions, the system prioritizes strategies that can weaken key causal paths. Specifically, for risk factor j marked as "potential evolutionary risk," the system identifies its main set of upstream root cause nodes through a causal graph. The system evaluates the inhibitory effect of candidate strategies on these root cause nodes. The inhibitory effect is quantified as the expected decrease in the state value of the root cause node after strategy execution. The system prioritizes the strategy combination with the largest inhibitory effect, maximizing the cumulative causal impact. This significantly reduces risk, thereby blocking the risk evolution path.

[0180] For risk states where the root cause cannot be traced back to a clear one, or where the root cause is outside the system's control (such as external weather), the system will instead choose direct intervention strategies targeting downstream risk factors, such as strengthening protective measures, increasing monitoring frequency, and pre-deploying emergency resources.

[0181] The generated decision-making scheme is output in the form of operational instruction data, including specific equipment control parameters, execution time nodes, and expected mitigation effects of causal impact. Operational instructions are issued to various execution devices on the platform through the equipment control interface, changing their operating status or data processing mode.

[0182] After the operational instructions are executed, the system continuously monitors the actual changes in the risk status and collects operational result feedback information after a preset time window. The feedback information includes: the actual evolution trajectory of each risk factor, the extent of risk reduction after strategy implementation, and the actual time delay of causal impact propagation.

[0183] The adaptive optimization module for decision-making compares the actual evolutionary trajectory with the predicted trajectory based on causal inference and calculates the prediction error. If a certain causal edge... The predicted results show a persistent deviation from the actual observations, and the system determines the influence coefficient of this causal edge. or time delay parameter Adjustments are needed. The system uses the least squares method to update the causal parameters, with the optimization objective being to minimize the cumulative sum of squares of historical prediction errors.

[0184] Furthermore, the system evaluates the effectiveness of causal reasoning-driven decision-making strategies based on feedback information. If key causal impact paths are not effectively weakened after strategy execution, or if downstream risk factors continue to evolve, the system records the failure characteristics of the decision-making scenario and adjusts the priority ranking or trigger threshold of strategy selection.

[0185] Through the aforementioned causal reasoning-enhanced analysis and adaptive adjustment process, the system achieves a deep understanding and predictive ability of risk evolution mechanisms, thereby generating more targeted and forward-looking decision-making and operational solutions, forming a technological closed loop of data-driven, causal reasoning, and intelligent decision-making.

[0186] Example 3

[0187] This embodiment provides a comparative verification method for a multi-source heterogeneous data processing and collaborative decision-making computing system of an intelligent construction management platform for urban renewal. The purpose of this embodiment is to: construct two system architectures, "without causal reasoning" and "with causal reasoning," and under the same construction site scenario and data conditions, compare the performance differences of the two systems in risk situation analysis and decision-making operations, thereby verifying the effect of the causal reasoning mechanism on improving decision accuracy, response timeliness, and resource utilization efficiency.

[0188] Both systems share the same multi-source data access and preprocessing module, receiving multi-source heterogeneous data from the construction site platform. Data sources include: temperature and humidity sensors, gas concentration sensors, and light sensors deployed within the platform; personnel positioning devices based on UWB technology; and drone inspection equipment equipped with visual recognition and infrared detection functions.

[0189] The data preprocessing workflow remains consistent across both systems. The system performs time alignment on data from different sampling frequencies, maps spatial location information to a unified 3D grid coordinate system, and performs filtering and noise reduction on the sensor data. After preprocessing, the system extracts 12 categories of risk characteristic variables, including temperature deviation, humidity anomaly index, etc. Concentration ratio, population density, duration of residence, structural crack width, wall displacement, ventilation equipment load rate, lighting equipment failure rate, power supply voltage fluctuation, ground water depth, and external weather level. All characteristic variables were normalized and mapped to the [0, 1] interval.

[0190] The non-causal reasoning system determines risk status by comparing risk characteristic values ​​with preset thresholds. The system assigns a fixed threshold to each risk characteristic variable; for example, the temperature deviation threshold is set to 0.7. The concentration ratio threshold is set to 0.6, and the personnel density threshold is set to 0.8, etc. When a certain characteristic value exceeds its corresponding threshold, the system determines that the risk corresponding to that characteristic is in a state of alert.

[0191] During the decision generation phase, the non-causal reasoning system employs predefined "feature-policy" mapping rules. The system maintains a decision rule table, with each rule in the form of "If feature X exceeds threshold Y, then execute policy Z". For example, rule 1: "If..." "If the concentration ratio is >0.6, then activate the enhanced ventilation strategy"; Rule 2: "If the personnel density is >0.8, then send a personnel dispersal recommendation"; Rule 3: "If the lighting equipment failure rate is >0.5, then activate the backup lighting system".

[0192] When multiple risk features simultaneously exceed a threshold, the non-causal inference system prioritizes them based on the magnitude of their exceedance and selects the strategy with the highest priority. This system does not analyze the correlation between risk features, determine whether a particular risk is caused by other risks, or predict future risk trends. Decisions are based on a snapshot of the data at the current moment, making it a passive, reactive decision-making model.

[0193] After extracting risk features, the causal reasoning system constructs a directed graph of causal relationships containing 12 risk factor nodes. The causal edges in the graph are determined based on physical mechanisms and historical event associations. For example, the "ventilation equipment load rate" node points to... The "concentration ratio" node has a causal influence coefficient of 0.75; the "external weather level" node points to the "surface water depth" node and the "humidity anomaly index" node, with influence coefficients of 0.85 and 0.65, respectively; the "humidity anomaly index" node points to the "power supply voltage fluctuation" node, with an influence coefficient of 0.60.

[0194] Based on the current state values ​​of each risk characteristic, the system performs forward inference using a causal graph to calculate the evolution trend of each risk factor within a future time window. For risk factor j, its predicted increment at the next time step is calculated as follows:

[0195] ;

[0196] in The causal influence coefficient. This represents the current value of upstream risk factors. This is the trigger threshold. If calculated as follows... The system determines that risk factor j has an evolutionary trend.

[0197] Simultaneously, the system performs reverse causal reasoning to identify the root cause of the current high-risk state. By tracing the reverse path in the causal graph, the system locates the key upstream root cause nodes.

[0198] During the decision-making phase, the causal reasoning system not only considers the current risk state but also selects strategies based on evolutionary trend predictions and root cause localization results. Specifically, for risk factors with evolutionary trends, the system prioritizes intervention strategies that can weaken their upstream root causes; for risk states traceable to clear root causes, the system intervenes at the root cause's non-representational level.

[0199] This embodiment involves a 30-day comparative test conducted within an intelligent platform at a construction site. During the test, both systems ran simultaneously, receiving the same real-time data streams but generating decision commands independently. The system administrator alternately adopted the decision schemes from both systems based on the actual situation and recorded the execution results.

[0200] Test Scenario 1: On the morning of the 5th day, continuous rainfall occurred, and the ground water depth gradually increased, causing the humidity anomaly index to rise to 0.65. The non-causal inference system immediately activated the dehumidifier after the humidity anomaly index exceeded the threshold of 0.6. However, because it did not anticipate that the continued increase in humidity might cause power supply voltage fluctuations, the system did not take preventative measures. 40 minutes later, the power supply voltage fluctuation index rose to 0.72, triggering the electrical equipment protection mechanism and causing some equipment to temporarily shut down. The causal inference system, after detecting the increased external weather level and the increase in ground water, predicted through causal inference that the humidity anomaly would further affect the power supply system. This system activated the dehumidifier in advance when the humidity index reached 0.55 (below the threshold of 0.6) and simultaneously reduced the load distribution of the power supply system, avoiding equipment downtime caused by voltage fluctuations. The comparative results show that the causal inference system's decision-making lead time was 15 minutes, avoiding equipment downtime losses.

[0201] Test Scenario 2: On the afternoon of the 12th day, the population density index rose to 0.82. The concentration ratio rose to 0.68. The non-causal reasoning system simultaneously activated the personnel dispersal strategy and the enhanced ventilation strategy according to the rules. However, because the ventilation equipment load rate was already high at 0.88, the enhanced ventilation command triggered an overload alarm. The causal reasoning system identified the problem through causal reasoning. The elevated concentration was caused by excessive personnel density, not by ventilation equipment malfunction. The system prioritized personnel dispersal, and once the personnel density decreased to 0.65, The concentration naturally decreased to 0.52, eliminating the need to activate enhanced ventilation. Comparative results show that the causal reasoning system avoided the risk of equipment overload and reduced energy consumption by 30%.

[0202] During the 30-day test, the non-causal reasoning system generated 286 decision instructions, of which 72 were overreactions or false alarms, achieving a decision accuracy rate of 74.8% and an average response time of 2.3 minutes. The causal reasoning system generated 198 decision instructions, with only 18 being overreactions, achieving a decision accuracy rate of 90.9% and an average response time of 1.8 minutes. The causal reasoning system reduced the number of decisions by 30.8%, indicating that it reduced repetitive decisions through root cause intervention; the decision accuracy rate improved by 16.1 percentage points, and the response time was shortened by 21.7%.

[0203] Furthermore, regarding resource consumption, the non-causal reasoning system activated the enhanced ventilation mode 47 times and the backup lighting system 23 times during the test, with a total energy consumption of 1850 kWh. The causal reasoning system, through proactive prevention and root cause intervention, activated the enhanced ventilation mode 29 times and the backup lighting system 9 times, with a total energy consumption of 1280 kWh, representing a 30.8% reduction in energy consumption.

[0204] The comparative test results above show that the causal reasoning system, by constructing a causal relationship model between risk factors, can predict the trend of risk evolution and identify the root cause, thus outperforming the non-causal reasoning system in terms of decision-making accuracy, response timeliness, and resource utilization efficiency.

Claims

1. A multi-source heterogeneous data processing and collaborative decision-making computing system for an intelligent construction management platform for urban renewal, characterized in that, include: The multi-source data access and preprocessing module is used to receive and aggregate multi-source heterogeneous data, including environmental data collected by an environmental sensor array, spatial location information of personnel collected by a personnel positioning device, and monitoring data collected by a drone inspection device. The risk field construction module is used to construct a computable dynamic risk field in the spatial and temporal dimensions of the platform based on the multi-source heterogeneous data. The dynamic risk field serves as an intermediate state representation of the system, used to uniformly characterize the comprehensive risk state corresponding to different spatial locations and different objects, and is called by subsequent modules. The risk situation analysis module is used to construct a causal relationship model between risk factors based on the dynamic risk field, and to analyze and infer the evolution direction and constraints of the risk state through the causal relationship model. The collaborative decision-making calculation module is used to generate multi-device collaborative decision-making results based on the risk situation analysis results, and output the collaborative decision-making results as operation instruction data to change or constrain the operating status or data processing flow of relevant equipment on the platform, thereby realizing the collaborative scheduling of platform operation behavior.

2. The multi-source heterogeneous data processing and collaborative decision-making computing system for an urban renewal intelligent construction management platform according to claim 1, characterized in that: The risk field construction module is used to uniformly represent environmental risks, structural risks, personnel activity risks, and equipment operation risks, and map them to the spatial area of ​​the platform in the form of risk intensity values, wherein the risk intensity values ​​are associated with at least one of the probability of risk occurrence, scope of impact, or duration.

3. The multi-source heterogeneous data processing and collaborative decision-making computing system of an urban renewal intelligent construction management platform according to claim 1, characterized in that: The dynamic risk field is a risk distribution model that is continuously updated over time. It is used to reflect the evolution trend of risk status in the time dimension and to provide status input for subsequent risk situation analysis and decision-making operations.

4. The multi-source heterogeneous data processing and collaborative decision-making computing system for an urban renewal intelligent construction management platform according to claim 1, characterized in that: The risk situation analysis module is used to distinguish between incidental risk states caused by instantaneous anomalies and cumulative risk states caused by long-term accumulated factors, and adjusts the weights or constraints of risk evolution analysis based on the distinction results; the risk situation analysis module describes the causal relationship between different risk factors and their influence paths by constructing a causal relationship diagram between risk factors, and the causal relationship diagram is used to constrain the evolution path of risk states or limit the candidate decision space of the collaborative decision calculation module.

5. The multi-source heterogeneous data processing and collaborative decision-making computing system for an urban renewal intelligent construction management platform according to claim 4, characterized in that: The causal edges in the causal relationship graph are configured with causal influence coefficients and time delay parameters. The causal influence coefficients are used to quantify the intensity of the influence of upstream risk factors on downstream risk factors, and the time delay parameters are used to characterize the time interval from the occurrence of causal influence to its manifestation.

6. The multi-source heterogeneous data processing and collaborative decision-making computing system for an intelligent construction management platform for urban renewal according to claim 1, characterized in that: The collaborative decision-making computation module includes multiple functional software agents, each corresponding to different equipment operation or management objectives, and each configured with an objective function and constraints corresponding to the objective.

7. The multi-source heterogeneous data processing and collaborative decision-making computing system for an urban renewal intelligent construction management platform according to claim 6, characterized in that: Based on sharing the same dynamic risk field state, the software agents generate the multi-device collaborative decision-making results through collaboration or game theory, according to their respective objective functions and constraints.

8. The multi-source heterogeneous data processing and collaborative decision-making computing system for an intelligent construction management platform for urban renewal according to claim 1, characterized in that: The system also includes a decision strategy adaptive optimization module, which is used to collect operational result feedback information after the execution of operational instruction data, and to continuously optimize the multi-device collaborative decision results based on the feedback information.

9. The multi-source heterogeneous data processing and collaborative decision-making computing system for an urban renewal intelligent construction management platform according to claim 8, characterized in that: The adaptive optimization module for decision-making strategies adjusts the decision preferences, strategy selection order, or collaboration methods of the software agent to achieve adaptive evolution of decision-making and operational strategies.

10. A method for multi-source heterogeneous data processing and collaborative decision-making computation in an intelligent construction management platform for urban renewal, characterized in that, Includes the following steps: S1 acquires multi-source heterogeneous data from the urban renewal intelligent construction management platform. The multi-source heterogeneous data includes environmental data collected by an environmental sensor array, spatial location information of personnel collected by a personnel positioning device, and monitoring data collected by a drone inspection device. Based on the multi-source heterogeneous data, S2 fuses risk information from different data sources and constructs a computable dynamic risk field in the spatial and temporal dimensions of the platform. The dynamic risk field serves as an intermediate state representation of the system and is used to uniformly characterize the comprehensive risk state corresponding to different spatial locations and different objects. S3 constructs a causal relationship model between risk factors based on the dynamic risk field, and analyzes and infers the evolution direction and constraints of the risk state through the causal relationship model. In the analysis and inference, a physical simulation model is introduced as a constraint condition for risk propagation to limit the evolution range of the risk state in the spatial and temporal dimensions. Before generating multi-device collaborative decision results, S4 determines whether the risk state meets preset evolution conditions. If the evolution conditions are met, based on the risk situation analysis results, it selects a combination of strategies that meet preset operational objectives from multiple candidate operational strategies to generate multi-device collaborative decision results. S5 outputs the multi-device collaborative decision-making results as operational instruction data, which is used to change or constrain the operating status or data processing flow of relevant platform devices. S6 collects operational result feedback information after the operational instruction data is executed, and makes adaptive adjustments to the subsequent decision-making and operational processes based on the feedback information.

Images