Engineering project risk supervision method, system and device and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and particularly provides an engineering project risk supervision method, system and device and a storage medium, and the method comprises the steps: constructing a corresponding monitoring agent for each risk type needing to be supervised; obtaining project progress information from a project management system, and calculating a cost hyperbranched rate based on project budget and actual expenditure; combining the project progress information, the cost hyperbranched rate information and the risk coefficient output by each monitoring agent to form an input state for a reinforcement learning model; and obtaining a regulation and control instruction which is generated by the reinforcement learning model based on the input state and is used for dynamically adjusting engineering resource allocation or a progress plan to cope with the risk. According to the method, the sudden risk response speed is increased, the compound risk misjudgment rate and the project hyperbranched probability are reduced, and the economic toughness of major engineering is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of artificial intelligence, and particularly relates to an engineering project risk supervision method, system, device and storage medium. BACKGROUND

[0002] Traditional engineering project risk management has three defects: 1. Static rough estimation defect: key parameters depend on historical mean, and cannot respond to real-time risk changes, such as typhoon attacks; 2. Risk fragmentation decision: climate, supply chain, compliance and other risk factors are processed by different departments in segments, and lack of collaborative quantification, such as the possible underestimation of the superimposed effect of steel price rise and rainy season; 3. Passive loss bearing: enterprises only passively provision for losses after the occurrence of risks, and have not established a dynamic adjustment mechanism in advance. SUMMARY

[0003] In view of the above defects of the prior art, the present application provides an engineering project risk supervision method, system, device and storage medium to solve the above technical problems.

[0004] In a first aspect, the present application provides an engineering project risk supervision method, comprising: constructing a corresponding monitoring agent for each risk type that needs to be supervised, the monitoring agent collecting data related to its corresponding risk type from a preset data source, and calculating a risk coefficient representing the comprehensive risk level of its corresponding risk type based on the data; obtaining engineering progress information from a project management system, and calculating a cost overrun rate based on project budget and actual expenditure; combining the engineering progress information, cost overrun rate information and risk coefficients output by each monitoring agent to form an input state for a reinforcement learning model; obtaining a control instruction generated by the reinforcement learning model based on the input state, the control instruction being used to dynamically adjust engineering resource allocation or progress plan to cope with risks.

[0005] In an optional embodiment, constructing a corresponding monitoring agent for each risk type that needs to be supervised comprises: constructing a climate agent for climate risk; constructing a supply chain agent for supply chain risk; constructing a regulation agent for industry regulation and compliance risk.

[0006] In an optional embodiment, constructing a climate agent for climate risk comprises: constructing a climate agent, receiving real-time meteorological data input, the meteorological data at least including wind speed and precipitation data; processing time-series meteorological data using a long short-term memory network, the input of the long short-term memory network being a meteorological data sequence within a time window; The future shutdown probability P is calculated by the output layer of the long short-term memory network, and the mathematical expression is: P = σ (W H + b), σ represents the Sigmoid activation function, H is the hidden state vector of the long short-term memory network at time t, W is the output layer weight matrix, and b is the output layer bias vector; The shutdown probability P is output as a climate risk coefficient.

[0007] In an optional embodiment, a supply chain agent is constructed for supply chain risk, including: The supply chain agent is constructed to obtain building material futures price data in real time; A mapping model of total project cost and building material price is established by using a gradient boosting tree regression algorithm, the input of the model is a historical building material price data set, and the output is a corresponding total project cost prediction value; The cost overrun sensitivity coefficient β is calculated by the mapping model c The mathematical essence is the partial derivative of the total project cost with respect to the target building material price:

[0008] Where Cost is the total project cost prediction value output by the mapping model, and Price is the real-time futures price of the target building material; The sensitivity coefficient β is output c As a supply chain cost risk coefficient.

[0009] In an optional embodiment, a regulation agent is constructed for industry regulatory and regulatory risk, including: The regulation agent receives the regulation text issued by the industry regulatory department; The regulation text is subjected to semantic analysis by using a Qwen-7B semantic analysis model, and a text risk score is output; An industry risk database is used to obtain an industry risk coefficient related to the regulation text; The compliance risk coefficient is obtained by calculating the product of the text risk score and the industry risk coefficient.

[0010] In an optional embodiment, the reinforcement learning model includes: The state space includes engineering progress information, cost overrun rate information, and risk coefficients output by each monitoring agent; The action space includes a material cost weight coefficient for correcting cash flow parameters, a schedule delay impact coefficient, and an operation cycle adjustment instruction; The reward function is: R = α1 V a + α2 E + α3 R i , V aR represents the economic value accuracy reward, E represents the prediction error penalty, and R represents the prediction error penalty. i This represents the risk suppression penalty term, where α1, α2, and α3 are the dynamic weight coefficients of each term. An Actor-Critic dual-network structure using the PPO algorithm is employed, and the policy network is optimized through a probability ratio pruning mechanism with a pruning threshold set to 0.2.

[0011] In an optional implementation, the method further includes: The composite risk index is obtained by calculating the weighted sum of the various risk coefficients. If the composite risk index reaches a pre-set first threshold, the cash flow model is corrected based on the material cost weighting coefficient and the impact coefficient of construction period delay. If the compliance risk coefficient reaches the pre-set second threshold, the operation cycle of the project will be adjusted according to the operation cycle adjustment instruction.

[0012] Secondly, the present invention provides an engineering project risk monitoring system, comprising: The risk monitoring module is used to build a corresponding monitoring agent for each type of risk that needs to be regulated. The monitoring agent collects data related to its corresponding risk type from a preset data source and calculates a risk coefficient that characterizes the comprehensive risk level of its corresponding risk type based on the data. The information acquisition module is used to obtain project progress information from the project management system and calculate the cost overrun rate based on the project budget and actual expenditures. The state generation module is used to combine the project progress information, cost overrun information, and risk coefficients output by each monitoring agent to form the input state for the reinforcement learning model. The instruction generation module is used to obtain control instructions generated by the reinforcement learning model based on the input state, which are used to dynamically adjust the allocation of engineering resources or the schedule to cope with risks.

[0013] Thirdly, a device is provided, comprising: Memory, used to store project risk monitoring procedures; A processor is used to implement the steps of the project risk monitoring method as provided in the first aspect when executing the project risk monitoring procedure.

[0014] Fourthly, a computer-readable storage medium is provided, on which an engineering project risk monitoring program is stored, and when the engineering project risk monitoring program is executed by a processor, it implements the steps of the engineering project risk monitoring method provided in the first aspect.

[0015] The beneficial effects of this invention lie in the fact that the project risk monitoring method, system, equipment, and storage medium provided by this invention, through the construction of a multi-dimensional risk monitoring intelligent agent cluster, generate dynamic risk coefficients in real time, completely overcoming the shortcomings of traditional static parameters that are lagging behind. It collaboratively integrates fragmented risk factors such as climate and supply chain with project progress and cost data, constructing a reinforcement learning state space to accurately quantify the cumulative effects of compound risks (such as the resonance between extreme weather and rising building material prices). Based on the control instructions output by the PPO algorithm, it achieves proactive intervention in resource reallocation such as material scheduling and schedule adjustments, transforming passive loss-bearing into dynamic risk hedging. This method improves the response speed to sudden risks, reduces the misjudgment rate of compound risks and the probability of project cost overruns, and significantly enhances the economic resilience of major projects. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic flowchart of a method according to an embodiment of the present invention.

[0018] Figure 2 This is a schematic topological diagram of a method according to an embodiment of the present invention.

[0019] Figure 3 This is a schematic diagram illustrating the risk factor monitoring method of one embodiment of the present invention.

[0020] Figure 4 This is a flowchart illustrating the training process of a reinforcement learning model according to an embodiment of the present invention.

[0021] Figure 5 This is a flowchart illustrating the decision-making and execution process of a method according to an embodiment of the present invention.

[0022] Figure 6 This is a schematic block diagram of a system according to an embodiment of the present invention.

[0023] Figure 7 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0025] Unless otherwise defined, 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 invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0026] The project risk monitoring method provided in this embodiment of the invention is executed by computer equipment, and correspondingly, the project risk monitoring system runs on the computer equipment.

[0027] Figure 1 This is a schematic flowchart illustrating a method according to an embodiment of the present invention. Wherein, Figure 1 The implementing entity can be an engineering project risk monitoring system. Depending on different needs, the order of the steps in this flowchart can be changed, and some steps can be omitted.

[0028] like Figure 1 As shown, the method includes: S1. Construct a corresponding monitoring agent for each type of risk that needs to be regulated. The monitoring agent collects data related to its corresponding risk type from a preset data source and calculates a risk coefficient that characterizes the comprehensive risk level of its corresponding risk type based on the data. S2. Obtain project progress information from the project management system and calculate the cost overrun rate based on the project budget and actual expenditures; S3. Combine the project progress information, cost overrun information, and risk coefficients output by each monitoring agent to form the input state for the reinforcement learning model; S4. Obtain control instructions generated by the reinforcement learning model based on the input state, which are used to dynamically adjust the allocation of engineering resources or the schedule to cope with risks.

[0029] In one embodiment of the present invention, based on step S1, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0030] S101. Building Climate Smart Agents for Climate Risks.

[0031] Construct a climate intelligence agent to receive real-time meteorological data input, which includes at least wind speed and precipitation data; Long Short-Term Memory (LSTM) networks are used to process time-series meteorological data, where the input to the LSM network is a sequence of meteorological data within a time window. The probability of future shutdown, P, is calculated through the output layer of the Long Short-Term Memory Network. Its mathematical expression is: P=σ(W·H+ b), where σ represents the Sigmoid activation function, H is the hidden state vector of the Long Short-Term Memory Network at time t, W is the weight matrix of the output layer, and b is the bias vector of the output layer. The shutdown probability P is output as the climate risk coefficient.

[0032] S102. Build a supply chain intelligence agent to address supply chain risks.

[0033] Build a supply chain intelligence agent to obtain real-time building material futures price data; A gradient boosting tree regression algorithm is used to establish a mapping model between the total cost of an engineering project and the price of building materials. The input of the model is a historical building material price dataset, and the output is the corresponding predicted value of the total cost of the engineering project. The cost overrun sensitivity coefficient β is calculated using the mapping model. c Its mathematical essence is the partial derivative of the total cost of the engineering project with respect to the price of the target building materials:

[0034] Where Cost is the project cost forecast output by the mapping model, and Price is the real-time futures price of the target building material. Output the sensitivity coefficient β c As a risk factor for supply chain costs.

[0035] S103. Construct a regulatory intelligence body to address industry regulatory risks.

[0036] Receive regulatory texts issued by industry regulatory authorities through a regulatory intelligence agent; The Qwen-7B semantic analysis model was used to perform semantic analysis on the regulatory text, and a text risk score was output. The industry risk coefficients related to the aforementioned regulatory text are obtained based on the industry risk database. The compliance risk coefficient is obtained by multiplying the text risk score by the industry risk coefficient.

[0037] In one embodiment of the present invention, based on step S2, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0038] The acquisition of project progress information employs a multi-dimensional data aggregation strategy: Through the API interface provided by the project management system, core parameters such as planned start time, actual start time, planned completion percentage, and actual completion percentage of tasks at each level of the WBS (Work Breakdown Structure) are synchronized in real time. Combined with the Critical Path Method (CPM), progress deviation values ​​for milestone nodes are extracted, forming a three-dimensional progress dataset containing task level, time dimension, and completion indicators. For the calculation of cost overrun rate, a dynamic baseline adjustment mechanism is adopted: using the approved total project budget as the initial baseline, the project is decomposed into planned costs (BCWS) for each work package according to the WBS structure; by connecting to the expenditure ledger of the financial system, the actual costs (ACWP) of each work package are obtained, and earned value analysis is performed based on the budgeted value (BCWP) of completed work. The cost overrun rate calculation formula is optimized to: Cost Overrun Rate = (ΣACWP - ΣBCWP) / ΣBCWS × 100%, where the numerator uses a sliding window algorithm to eliminate interference from short-term abnormal expenditures, and the denominator is updated in real time based on engineering change orders to ensure that the calculation results reflect the stage-specific cost control status.

[0039] In one embodiment of the present invention, based on step S3, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0040] In the information fusion phase, data standardization is first performed: Min-max normalization is used to map various indicators in the project progress information to the [0,1] interval, with positive values ​​representing progress ahead and negative values ​​representing delays; cost overrun rates are standardized using Z-scores to eliminate the influence of dimensions; and the risk coefficients output by each monitoring agent (initial value range [0,10]) are uniformly mapped to the [-1,1] interval using linear transformation, achieving a positive and negative vector representation of risk levels. Subsequently, feature engineering optimization is performed: Strongly correlated features of progress, cost, and risk are screened using Pearson correlation coefficients to eliminate redundant variables; trend features (such as 3-period moving average progress deviation) are extracted from time-series progress data using a sliding window; and abrupt change features are extracted from the risk coefficient sequence using wavelet transform. The final reinforcement learning input state vector adopts a hierarchical structure: the bottom layer is the original feature matrix (containing 12-dimensional progress features, 3-dimensional cost features, and N-dimensional risk features, where N is the number of risk types), the middle layer is the feature interaction term (such as the product term of progress deviation and technical risk coefficient), and the top layer is the global statistics (such as feature variation coefficient and maximum risk percentage). These are input into the state space of deep reinforcement learning in tensor form to provide multi-scale feature support for intelligent decision-making.

[0041] In one embodiment of the present invention, based on step S4, the following will provide a possible embodiment and describe its specific implementation in a non-limiting manner.

[0042] Reinforcement learning models include: The state space includes project progress information, cost overrun information, and risk coefficients output by each monitoring agent. The action space includes material cost weighting coefficients, project delay impact coefficients, and operating cycle adjustment instructions used to correct cash flow parameters; The reward function is: R = α1·V a +α2·E+α3·R i V a R represents the economic value accuracy reward, E represents the prediction error penalty, and R represents the prediction error penalty. i This represents the risk suppression penalty term, where α1, α2, and α3 are the dynamic weight coefficients of each term. An Actor-Critic dual-network structure using the PPO algorithm is employed, and the policy network is optimized through a probability ratio pruning mechanism with a pruning threshold set to 0.2.

[0043] To improve the engineering applicability of the instructions, the original control instructions output by the model need to undergo feasibility verification and refined transformation. First, the physical feasibility of the instructions is verified through constraint satisfaction algorithms: resource allocation adjustments must meet the upper and lower limits of resource requirements for each work package (e.g., manpower allocation cannot exceed 15% of the available workforce), and schedule adjustments must conform to the sequential logical relationships between tasks (the rationality of the critical path is verified through topological sorting). For infeasible instructions, the Lagrange multiplier method is used for correction, satisfying the constraints while minimizing deviations from the original instructions. Second, engineering semantic transformation is performed: the corrected resource allocation ratio is mapped to a specific personnel transfer list, equipment leasing / retirement plan, and fund disbursement adjustment form; the number of days for schedule adjustments is transformed into a detailed process interleaving plan (e.g., increasing the start and end times of night shifts, and task splitting plans for parallel operations), along with verification of commencement conditions (e.g., material supply cycle, weather window matching degree).

[0044] Furthermore, the model introduces a dynamic exploration mechanism to adapt to the uncertainties of the engineering environment: during the training phase, adaptive noise is added to the Actor output through the Ornstein-Uhlenbeck process (the noise intensity decays exponentially with the number of training steps, with an initial value of 0.2 and a decay coefficient of 0.995) to balance exploration and utilization; during the deployment phase, the exploration probability is dynamically adjusted based on the coefficient of variation of the risk coefficient (when the risk fluctuation exceeds the threshold of 0.3, the exploration probability is increased to 0.3 to cope with sudden risks). At the same time, a feedback loop for the effect of instructions is established, using the actual engineering progress, cost, and risk data collected after the adjustment as new state inputs. Samples (s, a, r, s') are stored in an experience replay pool (capacity set to 10^6), and a priority experience replay mechanism (based on the absolute value of TD-error to determine the sampling weight) is used to optimize the efficiency of model parameter updates, enabling the adjustment instructions to have continuous self-optimization capabilities.

[0045] Based on the above embodiments, in one embodiment, the instructions output by the reinforcement learning model are further integrated and processed.

[0046] The calculation of the composite risk index employs a hierarchical weighting mechanism to differentiate the various risk types in the overall assessment. First, a combined weighting strategy using the Delphi method and entropy weighting is employed to determine the weights of each risk coefficient: 5-8 experts in engineering management are invited to score the importance of n types of risks, such as technical risk, market risk, and compliance risk, using a scale of 1-9, thus forming a subjective weight matrix. Simultaneously, entropy is calculated to measure the information entropy of each risk coefficient in historical data; the lower the information entropy, the greater the impact of fluctuations in that risk type on the overall project, generating an objective weight vector. Finally, the weights are integrated using a weighted average method, with subjective weights accounting for 60% and objective weights accounting for 40%, denoted as ω1, ω2, ..., ω n Furthermore, the sum of the weights of each component is 1. The composite risk index is the sum of the products of each risk coefficient and its corresponding weight, where the risk coefficient for each risk category is standardized and ranges from 0 to 1. To enhance the index's sensitivity to dynamic changes, a risk coupling coefficient α is introduced. i (i≠j), used to correct for risk combinations with interactive effects, such as the coupling between supply chain risk and cost risk. The corrected composite risk index is the sum of the products of the original risk coefficients and their corresponding weights, plus the sum of the coupling terms between various risks. The coupling coefficient is determined by fitting the dependency structure between risks using a Copula function, and its value ranges from 0 to 0.2.

[0047] If the composite risk index reaches a pre-set first threshold, the cash flow model is corrected based on the material cost weighting coefficient and the impact coefficient of construction period delay.

[0048] The first threshold is set using a dynamic threshold calibration mechanism, determined by combining the characteristics of each stage of the project lifecycle and industry benchmarks: during the project planning phase, the threshold is set at 0.6, referencing the 90th percentile of historical risk trigger points for similar projects; during the construction phase, due to faster risk transmission, the threshold is lowered to 0.5; and during the operation phase, the threshold rises back to 0.65. When the composite risk index reaches or exceeds the first threshold, the two-dimensional correction process of the cash flow model is initiated. In the material cost dimension, the material cost weighting coefficient ranges from 0.3 to 0.7, determined by the proportion of material costs in the bill of materials (BOM). This coefficient is corrected in relation to the current material price fluctuation risk coefficient. The corrected coefficient is the original coefficient multiplied by (1 plus 0.5 times the material price fluctuation risk coefficient), and the corrected coefficient is substituted into the cost item of the cash flow forecasting model. For example, monthly material procurement expenditure is the sum of the product of the usage of each material, the benchmark unit price, and the corrected coefficient. In terms of project delay, the delay impact coefficient is obtained by multiplying the project delay sensitivity factor by the project delay risk coefficient. The project delay sensitivity factor is the project delay elasticity coefficient on the critical path, calculated using Monte Carlo simulation, and its value ranges from 1.2 to 2.5. The revised cash flow model considers the loss of the time value of money caused by project delays. Specifically, the adjusted net cash flow is the original net cash flow multiplied by an exponentially decaying term based on the delay impact coefficient and the expected number of delay days, with the discount rate set at 1.2 times the project financing cost. During the revision process, the material cost weighting coefficient and the delay impact coefficient are updated in real time using a 3-month sliding window to ensure that the cash flow forecast matches the dynamic changes in risk.

[0049] If the compliance risk coefficient reaches the pre-set second threshold, the operation cycle of the project will be adjusted according to the operation cycle adjustment instruction.

[0050] The second threshold is set separately to address the strong binding nature of compliance risks. It is determined using a combination of quantile analysis and expert judgment: based on a historical compliance risk event database, a critical risk coefficient triggering significant compliance penalties is calculated, typically using the 95th quantile, set at 0.75, and the final threshold is determined after review by the legal counsel team. When the compliance risk coefficient reaches or exceeds the second threshold, an operational cycle adjustment instruction is generated through a multi-objective optimization model to achieve a balance between compliance and economy. The objective function of the adjustment model is to minimize the cost increment caused by the extended operational cycle and maximize the reduction in compliance risk. The cost increment includes equipment idle fees, on-site personnel fees, etc., calculated on a daily basis. The reduction in compliance risk is measured by the expected decrease in the compliance risk coefficient, which is predicted based on regression analysis of historical rectification cases. The decision variable is the number of days for operational cycle adjustment, ranging from 0 to the maximum extension days allowed by the project contract. Constraints include meeting the new compliance inspection window after adjustment and the simultaneous extension of the lease term for key equipment. The final adjustment instructions are obtained through a genetic algorithm, outputting the optimal number of days for adjusting the operating cycle and a phased implementation plan. For example, internal rectification is completed in the first ΔT1 days, and a re-inspection is applied for in the next ΔT2 days. The sum of ΔT1 and ΔT2 is the total number of adjustment days. Tracking indicators for compliance risk mitigation are also included, such as rectification completion rate and regulatory feedback satisfaction.

[0051] In one specific embodiment, please refer to Figure 2 The data perception layer, serving as the system foundation, collects multi-source data in real time via API interfaces, covering weather data, commodity futures prices from the Shanghai Futures Exchange, and government policy text libraries. The intelligent agent decision-making layer deploys climate agents, supply chain agents, and policy agents to achieve parallel feature extraction. The dynamic output layer, based on the Proximal Policy Optimization (PPO) algorithm, outputs economic decision-making instructions, forming a complete data processing and decision-making chain. The specific technical solution is as follows: 1. Multi-agent collaborative module, see reference Figure 3 .

[0052] The climate agent takes meteorological data such as wind speed and precipitation as input, processes it using a Long Short-Term Memory (LSTM) network, and calculates the result using the formula P. s =σ(W·[X t X t−1 ,…,X t−n +b) Output the probability of future shutdown P s , where X tLet be the meteorological feature vector at time t, σ be the sigmoid activation function, W be the weight matrix in the LSTM model, and b be the bias vector. The supply chain agent receives futures price data for commodities such as rebar and cement, and calculates the cost overrun sensitivity coefficient β using the Gradient Boosting Tree (GBDT) regression algorithm. c Its calculation logic is as follows: Where Cost refers to the total cost of the project, and Price refers to the price of a specific building material. The policy agent processes the government gazette text, using the Qwen-7B semantic analysis model to extract the keyword co-occurrence network, thereby outputting the compliance risk increment ΔR. p =fQwen(Text)×K risk Where fQwen is the text risk score (0-1) output by the Qwen-7B model, Text is the input government bulletin text, and K risk This represents the industry risk coefficient. The three intelligent agents work collaboratively to generate a composite risk index, Irisk, where Irisk = αP. s +θβ c +γΔR p The weighting coefficients α, θ, and γ are dynamically updated based on the sliding window covariance, accurately quantifying the combined impact of multiple risk factors.

[0053] 2. Enhance the learning decision engine, refer to Figure 4 .

[0054] This invention defines a well-defined state space, action space, and reward function. The state space consists of a three-dimensional vector S. t =[P s ,β c ΔR p The system comprises: action space outputs core control commands for the economic model, including cash flow parameter adjustments (material cost weight adjustment ±15%, construction delay impact coefficient ±20%) and NPV / IRR calculation assumption updates (operation cycle expansion ±1 year). The reward function focuses on economic value stability, including rewards for value accuracy, penalties for prediction errors, and penalties for risk mitigation. An Actor-Critic dual-network structure optimization strategy using the PPO algorithm is employed. The loss function incorporates a probability ratio pruning mechanism (threshold 0.2) to ensure stability, completing the risk impact-to-value reassessment closed loop within 24 hours, thus continuously optimizing the decision-making strategy.

[0055] 3. Dynamic output interface, see reference. Figure 5 .

[0056] The system is equipped with a threshold triggering mechanism and a safety constraint module. When the composite risk index I... risk When the value is greater than τ1, a cash flow model reconstruction instruction is generated, initiating a joint correction of the material cost weight Δwm and the project delay coefficient Δdp; when the compliance risk increment ΔR...p When the value exceeds τ2, an operational cycle adjustment instruction is triggered (ΔT∈{−1,0,1} years), synchronously updating the residual value rate of the underlying assets of REITs. The safety constraint module provides dual assurance of decision reliability through an action amplitude limiter (material weight correction amount |Δwm|<=0.15, construction period impact coefficient correction amount |Δdp|<=0.2) and an outlier filter (intercepting abnormal actions based on historical data 3σ boundary detection).

[0057] In terms of data processing, a 64-dimensional engineering risk feature space is constructed through unified encoding of heterogeneous data. Numerical data is standardized and PCA is used for dimensionality reduction, while text data is embedded using BERT-128D. In terms of deployment, containerized agent deployment is adopted, with each agent independently encapsulated in a Docker container to achieve resource allocation isolation. In terms of hardware adaptability, the LSTM model is quantized to INT8 precision, supporting deployment on edge servers at construction sites, thereby improving the system's flexibility and practicality.

[0058] In some embodiments, the project risk monitoring system may include multiple functional modules composed of computer program segments. The computer programs for each program segment in the project risk monitoring system may be stored in the memory of a computer device and executed by at least one processor to perform (see details). Figure 1 (Description) The function of risk supervision in engineering projects.

[0059] In this embodiment, the project risk monitoring system can be divided into multiple functional modules based on its functions, such as... Figure 6 As shown. The module referred to in this invention is a series of computer program segments that can be executed by at least one processor and perform a fixed function, and is stored in memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0060] The risk monitoring module is used to build a corresponding monitoring agent for each type of risk that needs to be regulated. The monitoring agent collects data related to its corresponding risk type from a preset data source and calculates a risk coefficient that characterizes the comprehensive risk level of its corresponding risk type based on the data. The information acquisition module is used to obtain project progress information from the project management system and calculate the cost overrun rate based on the project budget and actual expenditures. The state generation module is used to combine the project progress information, cost overrun information, and risk coefficients output by each monitoring agent to form the input state for the reinforcement learning model. The instruction generation module is used to obtain control instructions generated by the reinforcement learning model based on the input state, which are used to dynamically adjust the allocation of engineering resources or the schedule to cope with risks.

[0061] Figure 7 The project risk monitoring method provided in the embodiments of this application can be applied to equipment. Those skilled in the art will understand that the equipment structure involved in the embodiments of this invention does not constitute a limitation on the equipment. Equipment may include more or fewer components than illustrated, or combine certain components, or have different component arrangements. In the embodiments of this invention, equipment includes, but is not limited to, laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Equipment may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the embodiments of this application described and / or claimed herein.

[0062] The device 700 may include a processor 710, a memory 720, and a communication unit 730. These components communicate via one or more buses. Those skilled in the art will understand that the server structure shown in the figures does not constitute a limitation of the present invention. It may be a bus topology or a star topology, and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0063] The memory 720 can be used to store execution instructions of the processor 710. The memory 720 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. When the execution instructions in the memory 720 are executed by the processor 710, the device 700 is able to perform some or all of the steps in the above method embodiments.

[0064] The processor 710 serves as the control center of the storage device, connecting various parts of the electronic device via various interfaces and lines. It executes software programs and / or modules stored in the memory 720, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of integrated circuits (ICs), such as a single packaged IC or multiple packaged ICs with the same or different functions connected together. For example, the processor 710 may consist only of a central processing unit (CPU). In this embodiment of the invention, the CPU may have a single processing core or include multiple processing cores.

[0065] The communication unit 730 is used to establish a communication channel, enabling the storage device to communicate with other devices. It can receive user data sent by other devices or send user data to other devices.

[0066] The present invention also provides a computer storage medium, wherein the computer storage medium may store a program, which, when executed, may include some or all of the steps provided in the embodiments of the present invention. The storage medium may be a magnetic disk, an optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0067] Those skilled in the art will clearly understand that the techniques in the embodiments of the present invention can be implemented using software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solutions in the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, or any other medium capable of storing program code. It includes several instructions to cause a computer device (which may be a personal computer, a server, or a second device, network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0068] The same or similar parts between the various embodiments in this specification can be referred to mutually. In particular, the device embodiments are basically similar to the method embodiments, so the description is relatively simple, and the relevant parts can be referred to the description in the method embodiments.

[0069] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or modules may be electrical, mechanical, or other forms.

[0070] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0072] Although the present invention has been described in detail with reference to the accompanying drawings and preferred embodiments, the present invention is not limited thereto. Various equivalent modifications or substitutions can be made to the embodiments of the present invention by those skilled in the art without departing from the spirit and essence of the invention, and such modifications or substitutions should all be within the scope of the present invention. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should also be covered within the protection scope of the present invention.

Claims

1. A method for risk monitoring in engineering projects, characterized in that, include: For each type of risk that needs to be regulated, a corresponding monitoring agent is constructed. The monitoring agent collects data related to its corresponding risk type from a preset data source and calculates a risk coefficient that characterizes the comprehensive risk level of its corresponding risk type based on the data. Obtain project progress information from the project management system and calculate the cost overrun rate based on the project budget and actual expenditures; The project progress information, cost overrun information, and risk coefficients output by each monitoring agent are combined to form the input state for the reinforcement learning model. Obtain control instructions generated by the reinforcement learning model based on the input state, which are used to dynamically adjust the allocation of engineering resources or the schedule to cope with risks.

2. The method according to claim 1, characterized in that, For each type of risk requiring oversight, a corresponding monitoring agent will be constructed, including: Building climate intelligence agents for climate risks; Building supply chain intelligence to address supply chain risks; Build a regulatory intelligence system to address regulatory risks in the industry.

3. The method according to claim 2, characterized in that, Building climate smart agents for climate risk, including: Construct a climate intelligence agent to receive real-time meteorological data input, which includes at least wind speed and precipitation data; Long Short-Term Memory (LSTM) networks are used to process time-series meteorological data, where the input to the LSM network is a sequence of meteorological data within a time window. The probability of future shutdown P is calculated through the output layer of the Long Short-Term Memory Network. Its mathematical expression is: P=σ(W·H+b), where σ represents the Sigmoid activation function, H is the hidden state vector of the Long Short-Term Memory Network at time t, W is the weight matrix of the output layer, and b is the bias vector of the output layer. The shutdown probability P is output as the climate risk coefficient.

4. The method according to claim 2, characterized in that, Building supply chain intelligence agents to address supply chain risks, including: Build a supply chain intelligence agent to obtain real-time building material futures price data; A gradient boosting tree regression algorithm is used to establish a mapping model between the total cost of an engineering project and the price of building materials. The input of the model is a historical building material price dataset, and the output is the corresponding predicted value of the total cost of the engineering project. The cost overrun sensitivity coefficient β is calculated using the mapping model. c Its mathematical essence is the partial derivative of the total cost of the engineering project with respect to the price of the target building materials: Where Cost is the project cost forecast output by the mapping model, and Price is the real-time futures price of the target building material. Output the sensitivity coefficient β c As a risk factor for supply chain costs.

5. The method according to claim 2, characterized in that, To build a regulatory intelligence agent for industry regulatory risks, including: Receive regulatory texts issued by industry regulatory authorities through a regulatory intelligence agent; The Qwen-7B semantic analysis model was used to perform semantic analysis on the regulatory text, and a text risk score was output. The industry risk coefficients related to the aforementioned regulatory text are obtained based on the industry risk database. The compliance risk coefficient is obtained by multiplying the text risk score by the industry risk coefficient.

6. The method according to claim 5, characterized in that, The reinforcement learning model includes: The state space includes project progress information, cost overrun information, and risk coefficients output by each monitoring agent. The action space includes material cost weighting coefficients, project delay impact coefficients, and operating cycle adjustment instructions used to correct cash flow parameters; The reward function is: R = α1·V a +α2·E+α3·R i V a R represents the economic value accuracy reward, E represents the prediction error penalty, and R represents the prediction error penalty. i This represents the risk suppression penalty term, where α1, α2, and α3 are the dynamic weight coefficients of each term. An Actor-Critic dual-network structure using the PPO algorithm is employed, and the policy network is optimized through a probability ratio pruning mechanism with a pruning threshold set to 0.

2.

7. The method according to claim 6, characterized in that, The method further includes: The composite risk index is obtained by calculating the weighted sum of the various risk coefficients. If the composite risk index reaches a pre-set first threshold, the cash flow model is corrected based on the material cost weighting coefficient and the impact coefficient of construction period delay. If the compliance risk coefficient reaches the pre-set second threshold, the operation cycle of the project will be adjusted according to the operation cycle adjustment instruction.

8. A project risk monitoring system, characterized in that, include: The risk monitoring module is used to build a corresponding monitoring agent for each type of risk that needs to be regulated. The monitoring agent collects data related to its corresponding risk type from a preset data source and calculates a risk coefficient that characterizes the comprehensive risk level of its corresponding risk type based on the data. The information acquisition module is used to obtain project progress information from the project management system and calculate the cost overrun rate based on the project budget and actual expenditures. The state generation module is used to combine the project progress information, cost overrun information, and risk coefficients output by each monitoring agent to form the input state for the reinforcement learning model. The instruction generation module is used to obtain control instructions generated by the reinforcement learning model based on the input state, which are used to dynamically adjust the allocation of engineering resources or the schedule to cope with risks.

9. A risk monitoring device for engineering projects, characterized in that, include: Memory, used to store project risk monitoring procedures; A processor is configured to implement the steps of the project risk monitoring method as described in any one of claims 1-7 when executing the project risk monitoring procedure.

10. A computer-readable storage medium storing a computer program, characterized in that, The readable storage medium stores an engineering project risk monitoring program, which, when executed by a processor, implements the steps of the engineering project risk monitoring method as described in any one of claims 1-7.

Images