Big data analysis and intelligent decision-making auxiliary optimization method and system for engineering cost

Abstract

This invention relates to data analysis technology, and specifically discloses a method and system for big data analysis and intelligent decision-making support optimization of engineering cost. This invention acquires records of the decision-making behaviors of multiple project participants from historical engineering cost data, extracts the strategic response behaviors of each participant after observing upstream events to identify response rules, and constructs a propagation chain with time delay and triggering parameters based on these response rules. This enables the characterization of the dynamic relationship and triggering sequence of strategy propagation among participants, solving the problem of existing methods neglecting the interaction between participants and achieving accurate modeling of the cost impact transmission process.

Description

Technical Field

[0001] This invention relates to data analysis technology, and more particularly to a method and system for big data analysis and intelligent decision support optimization in engineering cost estimation. Background Technology

[0002] Existing methods for predicting project costs primarily rely on static historical data or expert experience, making it difficult to capture the dynamic game-theoretic relationships among project stakeholders. When external factors such as material prices fluctuate, owners, contractors, suppliers, and other parties adjust their strategies based on their own interests. These adjustments propagate through the project stakeholder network, ultimately leading to a new cost equilibrium. Existing methods neglect the mutual influence between stakeholders, decision-making lags, and triggering mechanisms, resulting in significant prediction biases. Summary of the Invention

[0003] To overcome the problems in existing engineering cost forecasting, such as neglecting the mutual influence between subjects, decision-making time lag, and triggering mechanisms, which leads to large prediction deviations, this invention provides a method and system for engineering cost big data analysis and intelligent decision-making assistance optimization.

[0004] In a first aspect, the present invention provides a method for big data analysis and intelligent decision-making assistance optimization of engineering cost, including:

[0005] The decision-making behavior records of multiple project participants in the historical data of engineering cost are obtained. Based on the decision-making behavior records, the strategic response behavior of each participant after observing upstream events is extracted, and the response rules of each participant are identified. Based on the response rules of each participant, a propagation chain representing the strategy propagation relationship between the participants is constructed. Each directed edge of the propagation chain is labeled with a time delay parameter and a trigger parameter. In response to a target upstream event, the target upstream event is used as the initial impact. Based on the time delay parameters and trigger parameters in the propagation chain, the sequential adjustment process of each subject is simulated along the propagation chain to obtain a preliminary predicted multi-party equilibrium state. Based on the preliminary predicted multi-party equilibrium state, the expected value of acquiring target market information for improving the preliminary predicted multi-party equilibrium state is calculated. When the expected value exceeds the acquisition cost of the target market information, the parameters of the propagation chain are updated based on the target market information, and the sequential adjustment process is re-simulated to obtain a corrected multi-party equilibrium state. The cascaded prediction result of the engineering cost is then calculated. During project execution, actual cost data is collected and compared with the cascaded prediction results of the project cost to construct a cumulative deviation. When the cumulative deviation exceeds a preset deviation threshold, the response of each subject to different alternative intervention schemes is predicted based on the propagation chain, the feasibility of each alternative intervention scheme is evaluated, and the alternative intervention scheme with the highest feasibility is selected for execution.

[0006] According to a specific implementation method, the above optimization method involves acquiring decision-making behavior records of multiple project participants from historical engineering cost data, extracting the strategic response behavior of each participant after observing upstream events based on the decision-making behavior records, and identifying the response rules of each participant, including: Extract the decision timestamp sequence and strategy type sequence of each entity after the occurrence of the upstream event from the historical data of the project cost; A correlation analysis is performed on the strategy type sequence of each subject and the strategy type sequence of the corresponding upstream subject to extract the conditional probability distribution of the corresponding subject executing the corresponding response strategy after observing various upstream strategies; Based on the decision timestamp sequence, calculate the time interval distribution for each subject from observing the upstream strategy to executing the response strategy, and determine the time delay parameter based on the statistical characteristics of the time interval distribution; Based on the conditional probability distribution, identify the minimum upstream policy strength threshold required to trigger the response strategy, and use the minimum upstream policy strength threshold as the triggering parameter.

[0007] According to a specific implementation, in the above optimization method, in response to a target upstream event, the target upstream event is used as an initial impact. Based on the time delay parameters and triggering parameters in the propagation chain, the sequential adjustment process of each subject is simulated along the propagation chain to obtain a preliminary predicted multi-party equilibrium state, including: The intensity parameters and influence direction of the upstream event of the target are obtained and used as the initial impact event; Based on the time delay parameters in the propagation chain, the time delay at which each subject observes the policy changes of the upstream subject is determined, and the policy propagation time sequence between subjects is constructed. Identify one or more initial response entities directly affected by the initial impact event from the propagation chain, and determine an initial strategy adjustment scheme based on the response rules of the initial response entities; The initial strategy adjustment scheme is used as a new upstream strategy and input into the propagation chain. Based on the strategy propagation sequence and the triggering parameters in the propagation chain, the triggered downstream entities are identified round by round and their strategy adjustment schemes are calculated to form a sequential adjustment sequence. When the policy adjustment magnitude of all subjects in the sequential adjustment sequence is lower than the convergence threshold within consecutive preset rounds, the current policy state of each subject is extracted as the preliminary predicted multi-party equilibrium state.

[0008] According to a specific implementation, in the above optimization method, based on the preliminary predicted multi-party equilibrium state, the expected value of acquiring target market information for improving the preliminary predicted multi-party equilibrium state is calculated. When the expected value exceeds the acquisition cost of the target market information, the parameters of the propagation chain are updated based on the target market information, including: In the propagation chain, target entities whose response rule parameters can be modified by the target market information are identified; Along the propagation chain, traverse from the target entity downstream to identify the set of affected downstream entities; Based on the preliminary predicted multi-party equilibrium state, the sensitivity of the strategy adjustment of each entity in the downstream entity set to the error in the project cost prediction is calculated. The sensitivity of each downstream entity is weighted and summed, and the result is used as the expected value of the target market information. When the expected value is greater than the cost of acquiring the target market information, the target market information is acquired, and the target market information is used to correct the response rule parameters of the target entity.

[0009] According to a specific implementation method, the above optimization method involves weighted summation of the sensitivities corresponding to each downstream entity, including: Determine the influence range of each downstream entity in the propagation chain, wherein the influence range is calculated based on the number of other entities reachable from that downstream entity; The weight of each downstream entity is determined based on its influence range, where the weight is positively correlated with the influence range. Based on the determined weights, a weighted summation operation is performed on the sensitivity of each downstream entity.

[0010] According to a specific implementation method, in the above optimization method, during project execution, actual cost data is collected, and the actual cost data is compared with the cascaded prediction results of the project cost to construct a cumulative deviation, including: Actual cost data is collected at multiple time points during project execution, and the cascaded prediction results of project cost at the corresponding time points are extracted. Calculate the instantaneous deviation between the actual cost data and the cascaded prediction results of the project cost at each time point, and construct a deviation time series in chronological order; The amplitude features, duration features, and trend features of the deviation time series are extracted to construct a multi-dimensional time series feature vector; The multidimensional time-series feature vector is weighted and fused to obtain the cumulative deviation; wherein the weight configuration used in the weighted fusion is positively correlated with the duration feature and the trend feature.

[0011] According to a specific implementation, in the above optimization method, when the cumulative deviation exceeds a preset deviation threshold, the response of each subject to different alternative intervention schemes is predicted based on the propagation chain, the feasibility of each alternative intervention scheme is evaluated, and the alternative intervention scheme with the highest feasibility is selected for execution, including: When the cumulative deviation exceeds a preset deviation threshold, at least one alternative intervention plan is generated, and the change in benefits for each subject after the alternative intervention plan is implemented is calculated. Based on changes in returns, the entities are divided into supporters and opponents; Based on the propagation chain, target supporting entities that have upstream influence over the opposing entities are identified from among the supporting entities and are used as the first-stage intervention targets. Starting with the first-stage intervention target, predict the impact of its decision adjustment along the propagation chain on downstream opposing entities, in order to identify the original opposing entities that have turned into supporters, and iteratively design subsequent stages to form a phased combined intervention path. For each phased combined intervention path, the intervention costs of each stage are summed up with the deviation correction effect in the final stage equilibrium state, and the combined intervention path that optimizes the cost-effectiveness ratio is selected for execution.

[0012] According to a specific implementation, in the above optimization method, determining the target supporting entity with upstream influence over the opposing entity from among the supporting entities, based on the propagation chain, includes: For the influence relationship between each supporting entity and each opposing entity in the propagation chain, the propagation path from the supporting entity to the opposing entity is extracted; wherein, the propagation path includes direct influence paths and indirect influence paths via intermediate entities; Calculate the influence intensity of each propagation path, and sum the influence intensity of each propagation path from a supporting entity to all opposing entities to obtain the comprehensive influence score of the supporting entity; Extract the time delay parameters of each side in the propagation chain, and calculate the average propagation time delay required for the policy change of each supporting subject to propagate to the opposing subject; The ratio of the comprehensive influence score to the average dissemination time lag is calculated as an intervention priority indicator, and the supporting entity with the highest intervention priority indicator is selected as the intervention target for the first stage.

[0013] According to one specific implementation, in the above optimization method, predicting the response of each subject to different alternative intervention schemes includes: For each stage of the phased combined intervention path, predict the response strategy of each subject based on the response rules of each subject; The response strategy is propagated along the propagation chain to downstream entities, and the counter-response strategy of the downstream entities is predicted to construct a multi-layered response game tree. The game tree is subjected to backward induction, tracing back from the leaf nodes to the root node to determine the strategy choices of each subject under the objective of maximizing profit, and the endpoint state of the strategy evolution path is extracted as the equilibrium state of that stage.

[0014] Secondly, the present invention provides a big data analysis and intelligent decision-making support optimization system for engineering cost, comprising: The first unit is used to acquire decision-making behavior records of multiple project participants in historical engineering cost data, extract the strategic response behavior of each participant after observing upstream events based on the decision-making behavior records, identify the response rules of each participant, and construct a propagation chain representing the strategy propagation relationship between each participant based on the response rules of each participant; wherein, each directed edge of the propagation chain is labeled with a time delay parameter and a trigger parameter. The second unit is used to respond to a target upstream event, taking the target upstream event as the initial impact, and simulating the sequential adjustment process of each subject along the propagation chain based on the time delay parameters and trigger parameters in the propagation chain to obtain a preliminary predicted multi-party equilibrium state; and, based on the preliminary predicted multi-party equilibrium state, calculating the expected value of acquiring target market information for improving the preliminary predicted multi-party equilibrium state, and when the expected value exceeds the acquisition cost of the target market information, updating the parameters of the propagation chain based on the target market information, resimulating the sequential adjustment process to obtain a corrected multi-party equilibrium state, and calculating the cascaded prediction result of the engineering cost; The third unit is used to collect actual cost data during project execution, compare the actual cost data with the cascaded prediction results of the project cost to construct a cumulative deviation; and when the cumulative deviation exceeds a preset deviation threshold, predict the response of each subject to different alternative intervention schemes based on the propagation chain, evaluate the feasibility of each alternative intervention scheme, and select the alternative intervention scheme with the highest feasibility for execution.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, by constructing a propagation chain with time delays and triggering parameters, accurately depicts the dynamic process of strategy propagation among project participants, significantly improving the accuracy of cost forecasting compared to traditional static forecasting methods. By calculating the expected value of target market information and comparing it with acquisition costs, it achieves economical decision-making in information acquisition, avoiding cost waste caused by ineffective information collection, while ensuring timely acquisition of key information for optimized forecasting. During the project execution phase, by constructing a cumulative deviation index that integrates comprehensive time-series characteristics, it can more sensitively identify systematic deviations rather than random fluctuations. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a method for big data analysis and intelligent decision support optimization of engineering cost provided in an embodiment of the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to specific embodiments. However, this should not be construed as limiting the scope of the present invention to the following embodiments; all technologies implemented based on the content of the present invention fall within the scope of the present invention.

[0018] Unless otherwise specified, the use of terms such as "first," "second," and "third" in the description of specific embodiments of the present invention is merely for distinguishing descriptions of identical or similar components and should not be construed as emphasizing or implying the relative importance of specific components. Furthermore, in the description of embodiments of the present invention, "several," "multiple," and "a number" represent at least two. The number can be any number, including 2, 3, 4, 5, 6, 7, 8, and 9, and can even exceed nine.

[0019] It is understood that, in this embodiment of the invention, project participants refer to stakeholders in an engineering construction project whose contractual relationships or business dealings affect the project cost, such as the owner, general contractor, subcontractors, material suppliers, and supervision units. When faced with upstream events, each stakeholder will make corresponding strategic adjustments based on their own interests and objectives, such as adjusting quotations, applying for extensions of the construction period, changing material brands, and reallocating resources.

[0020] Upstream events refer to external or internal events that can trigger strategic responses from project participants. They possess corresponding types, intensity parameters, and directions of impact. For example, upstream events include, but are not limited to, material price fluctuations, policy adjustment events (such as environmental production restrictions), and labor market price changes. When an upstream event is a material price fluctuation, its intensity parameter is the price fluctuation magnitude (expressed as a percentage), and its direction of impact is either upward or downward. When an upstream event is an environmental production restriction, its intensity parameter is the production restriction percentage (expressed as a percentage), and its direction of impact is negative (indicating increased costs or extended construction periods).

[0021] A propagation chain is a directed network graph data structure used to characterize the policy propagation relationships among project participants. In this network, nodes represent participating entities, and directed edges represent the direction of policy influence (from upstream entities to downstream entities). Each directed edge is labeled with a time delay parameter and a triggering parameter. The time delay parameter represents the time delay between an entity observing a policy change from an upstream entity and its own response policy; the triggering parameter represents the minimum threshold of the upstream policy strength required to trigger a response policy from an entity.

[0022] The sequential adjustment process refers to the dynamic process in which, starting from the initial impact event, each agent adjusts its own strategy sequentially (round by round) according to the time sequence and triggering conditions defined in the propagation chain. The multi-party equilibrium state refers to a state reached after multiple rounds of sequential adjustment where the adjustment magnitudes of the strategies of all agents in the propagation chain tend to stabilize and no longer undergo significant changes.

[0023] This invention can be applied to an engineering cost management platform or project management system. The system can be deployed on a cloud server or a local server cluster, communicating with the terminal devices (such as computers and mobile terminals) of various participating entities via a network to acquire data, perform analysis, and output decision support information.

[0024] The following is combined with Figure 1 The flowchart shown below provides a detailed explanation of the engineering cost big data analysis and intelligent decision-making assistance optimization method provided in this embodiment of the invention. It should be noted that although this embodiment primarily uses material price fluctuations as an example of the target upstream event, the propagation chain construction logic, sequential adjustment simulation logic, information value assessment logic, and deviation monitoring and intervention logic involved in the above method steps are all general processing logics for the target upstream event, and their specific execution process is independent of the type of upstream event. When the target upstream event is of other types, it is only necessary to map this type of event to the corresponding intensity parameters and influence direction in the event acquisition step, and maintain consistency in subsequent processing steps to obtain the cascaded prediction results of engineering cost corresponding to this type of upstream event and the optimal intervention scheme.

[0025] Figure 1 This is a flowchart illustrating a method for big data analysis and intelligent decision support optimization of engineering cost, provided as an embodiment of the present invention. The method can be executed by the aforementioned engineering cost management platform or system, and includes: Step 101: Obtain decision-making behavior records of multiple project participants from historical engineering cost data, extract the strategic response behavior of each participant after observing upstream events based on the decision-making behavior records, and identify the response rules of each participant.

[0026] Specifically, historical project cost data can be obtained from a pre-built project management database. This database is stored in a relational structure, with one project record corresponding to each completed or under-construction project. Each project record contains a unique project identifier, a list of project participants, an upstream event sequence, and a decision timestamp sequence and strategy type sequence for each participant. The upstream event record includes the occurrence timestamp, event type, intensity parameter (such as price fluctuation range, expressed as a percentage), and direction of impact (such as increase or decrease). The decision timestamp sequence records the precise moment each participant made a decision, using a standard time format. The strategy type sequence records the type of strategy executed by the participant. Strategy types can be categorized as price adjustment, time extension, resource reallocation, claims, and plan changes, each carrying a quantified intensity parameter. For example, the intensity parameter for a price adjustment strategy is the proportion of the adjustment amount to the total contract price, and the intensity parameter for a time extension strategy is the number of extension days.

[0027] When extracting data, for each project, a preset observation time window (e.g., configurable to 90 days) is retrieved forward from the moment the material price change event occurs. All decision records for each subject within the window are sorted by timestamp, forming a decision timestamp sequence and strategy type sequence for that subject under that event. During the data cleaning phase, records with missing timestamps or incomplete strategy type labels are filtered out, and outliers (such as strategy strength parameters exceeding the 5th to 95th percentile range) are processed using quantile truncation to ensure data quality.

[0028] Furthermore, an association analysis is performed on the strategy type sequence of each subject and the strategy type sequence of the corresponding upstream subject to extract the conditional probability distribution of the corresponding subject executing the corresponding response strategy after observing various upstream strategies.

[0029] Understandably, in a project, the upstream and downstream relationships between various entities are predefined based on business processes. For example, the upstream entity of the general contractor is the owner, the upstream entity of the subcontractor is the general contractor, and the upstream entity of the material supplier can be both the general contractor and the subcontractor. A sliding time window is used to align the strategy sequence of the current entity with the strategy sequences of one or more of its upstream entities over time. For example, for a certain decision of the current entity, all strategies executed by the upstream entity within a certain time period (e.g., 30 days) are traced back in its decision records. By statistically analyzing a large amount of historical project data, a "strategy transition matrix" can be constructed. The rows of this matrix represent the types of upstream strategies, the columns represent the types of response strategies that the current entity may adopt, and each element value in the matrix represents the number of observations. Normalizing this matrix row by row yields a conditional probability distribution. This conditional probability distribution quantifies the certainty of strategy transmission and the entity's selection preferences. For example, if the probability distribution of a certain row in the matrix is ​​highly concentrated in a certain column (e.g., the probability value is greater than 0.9), it indicates that the upstream strategy will trigger that specific response strategy with high certainty.

[0030] Furthermore, based on the decision timestamp sequence, the time interval distribution from observing the upstream strategy to executing the response strategy for each subject is calculated, and the time delay parameter is determined according to the statistical characteristics of the time interval distribution.

[0031] Specifically, for each observed policy response event, the difference between the current subject's decision timestamp and the corresponding upstream subject's policy execution timestamp is calculated. This difference is a "time interval," which can be in days. All such time interval data are aggregated to form the time interval distribution for that subject against a specific upstream policy type. The median of this distribution is extracted as the time delay parameter for that edge. Using the median instead of the mean as the time delay parameter effectively avoids the interference of extreme delays caused by atypical reasons (such as holidays or process shutdowns) on the model parameters, making the simulation results more robust.

[0032] Furthermore, based on the conditional probability distribution, the minimum upstream policy strength threshold required to trigger the response strategy is identified, and the minimum upstream policy strength threshold is used as the triggering parameter.

[0033] In one possible implementation, for each upstream strategy type, the strength parameters of that strategy in historical data are sorted in ascending order and divided into several strength intervals (e.g., 10). The proportion of the current agent executing the response strategy within each strength interval is calculated, and a curve showing the response proportion changing with the strength of the upstream strategy is plotted. The strength interval where the response proportion first exceeds a preset threshold (e.g., 0.5) is identified, and the lower bound of this interval is used as the trigger parameter. If the curve does not have a significant jump point, the strength value corresponding to the maximum gradient point is used as the trigger parameter. The trigger parameter reflects the agent's sensitivity to upstream changes: the smaller the trigger parameter value, the more easily the agent reacts to small upstream fluctuations.

[0034] Step 102: Based on the response rules of each subject, construct a propagation chain representing the policy propagation relationship between the subjects. Each directed edge of the propagation chain is labeled with a time delay parameter and a trigger parameter.

[0035] Specifically, each participating entity in a project is treated as a node in the graph. Node attributes include a unique identifier for the entity, its role type, and a data structure (such as a hash table) encapsulating all response rules for that entity. When two entities are defined as having an upstream-downstream influence relationship, a directed edge is created between the nodes corresponding to these two entities. The starting point of this edge is the upstream entity node, and the ending point is the downstream entity node. The time delay parameters and trigger parameters extracted from the previous steps, corresponding to this entity relationship, are used as attribute values ​​and labeled on this directed edge. By traversing all entity pairs in all projects, a complete directed propagation network can be constructed; this network is called a "propagation chain." This propagation chain can be stored in the computer using an adjacency list data structure. Each node maintains an outgoing edge list and an incoming edge list. The outgoing edge list records downstream entities influenced by the upstream entity, and the incoming edge list records upstream entities that influence the node, facilitating rapid upstream-downstream traversal and simulation calculations in subsequent steps.

[0036] Through the above steps, the embodiments of the present invention can objectively and quantitatively construct a propagation chain model based on historical data, which reflects the complex game relationship and propagation dynamics characteristics among multiple parties in the field of engineering cost, providing a solid data foundation for subsequent dynamic prediction and decision optimization.

[0037] Step 201: In response to the upstream event of the target, take the upstream event of the target as the initial impact, and simulate the sequential adjustment process of each subject along the propagation chain according to the time delay parameter and triggering parameter in the propagation chain to obtain the preliminary predicted multi-party equilibrium state.

[0038] For example, the intensity parameter and direction of influence of the upstream event of the target are obtained and used as the initial impact event. The method of obtaining this parameter could be, for example, when the upstream event of the target is a fluctuation in material prices, the intensity parameter is the price fluctuation range (e.g., an increase of 10%), and the direction of influence is upward.

[0039] Based on the time delay parameters of each edge in the propagation chain, the time delay at which each subject observes the policy change of the upstream subject is determined, and a policy propagation time sequence is constructed between subjects. The policy propagation time sequence table can be implemented using a priority queue data structure. Each element in the queue is an event object, which includes the source subject identifier, the target subject identifier, the upstream policy content to be propagated, and the expected arrival time. The expected arrival time equals the source subject's policy execution time plus the time delay parameter labeled on the directed edge. The queue is arranged in order of expected arrival times.

[0040] Furthermore, one or more initial response entities directly affected by the initial shock event are identified from the propagation chain. In the node attributes of the propagation chain, material supplier nodes are marked as initial response entities directly affected by material price fluctuation events. Their initial strategy adjustment schemes are determined based on the response rules of these initial response entities. Specifically, the conditional probability distribution of the entity is queried according to the direction of influence, and the response strategy type is determined through probability sampling. The strength parameter of the response strategy can be calculated based on the strength parameter of the initial shock event and the strategy strength coefficient of the entity. The strategy strength coefficient is obtained from historical data statistics, for example, with a default value of 0.8. The initial strategy adjustment scheme includes the strategy type, strategy strength, and execution time.

[0041] Furthermore, the initial strategy adjustment scheme is input into the propagation chain as a new upstream strategy. Based on the strategy propagation timing and the triggering parameters in the propagation chain, the downstream entities that are triggered are identified round by round, and their strategy adjustment schemes are calculated to form a sequential adjustment sequence.

[0042] Specifically, the initial strategy adjustment plan of the material supplier ("increase the price by 8%) is treated as a new upstream strategy event, and all outgoing edges originating from that material supplier node are traversed. For each downstream entity (e.g., the general contractor) pointed to by an outgoing edge, the strength of the upstream strategy (8%) is compared with the triggering parameter marked on that edge (e.g., the trigger threshold for the general contractor to raise prices for suppliers is 5%). Since 8% is greater than 5%, the downstream entity (general contractor) is triggered. The triggered event (containing upstream strategy information) along with its expected arrival time (determined by the time delay parameter) is inserted into the strategy propagation sequence table.

[0043] The simulation clock advances in fixed steps (e.g., 0.1 days). At each simulation moment, all events with expected arrival times less than or equal to the current moment are retrieved from the strategy propagation timeline and processed one by one. For a triggered downstream entity (general contractor), its own response rules are queried. If the entity simultaneously receives strategies from multiple upstream entities (e.g., observing both supplier price increases and owner requests for time-cutting), this embodiment first performs a weighted sum of the strengths of all upstream strategies; the weights can be based on preset rules or learned from historical data. Then, based on the accumulated upstream impact, its response strategy type is determined by querying its conditional probability distribution (e.g., the general contractor might choose to "file a claim with the owner"). The strength of its response strategy can also be calculated by multiplying the accumulated upstream strength by its own strategy strength coefficient. The execution time of this response strategy is the current observation time plus the entity's decision delay. This response strategy then propagates as a new upstream strategy event to its downstream entities.

[0044] In this embodiment of the invention, the strategy adjustment schemes of all subjects generated in each round of triggering are recorded in a list in chronological order, and this list is the sequential adjustment sequence.

[0045] Furthermore, when the policy adjustment magnitude of all subjects in the sequential adjustment sequence is lower than the convergence threshold within consecutive preset rounds, the current policy state of each subject is extracted as the preliminary predicted multi-party equilibrium state.

[0046] Specifically, after each round of simulation calculation, all agents in the propagation chain are traversed, and their policy adjustment magnitude in the current round relative to the previous round is calculated. The policy adjustment magnitude can be defined as: |Current Round Policy Strength - Previous Round Policy Strength| / Previous Round Policy Strength. If the percentage change in a subject's policy strength is less than a preset convergence threshold (e.g., 1%), then the subject is considered to have reached stability in this round. This embodiment of the invention maintains a counter to record the number of consecutive rounds in which all subjects in the propagation chain satisfy the convergence condition. When the counter reaches a preset value (e.g., 3 consecutive rounds), it is determined that the entire propagation network has reached dynamic equilibrium, and the simulation process is terminated. At this point, the set of current policy states (policy type and policy strength) of each subject in this embodiment of the invention represents the preliminary predicted multi-party equilibrium state.

[0047] Through the above simulation process, the embodiments of the present invention can predict the chain reaction caused by the upstream event of the target in a quantitative way among the various participants in the project, and accurately find the stable point reached by the game of interests of all parties, providing a more accurate basis for the early cost prediction of the project than traditional static estimation.

[0048] Step 202: Based on the preliminary predicted multi-party equilibrium state, calculate the expected value of obtaining target market information for improving the preliminary predicted multi-party equilibrium state. When the expected value exceeds the cost of obtaining the target market information, update the parameters of the propagation chain based on the target market information, re-simulate the sequential adjustment process to obtain the corrected multi-party equilibrium state, and calculate the cascade prediction result of the engineering cost.

[0049] In one possible implementation, the available market information includes changes in material supplier capacity, anticipated policy adjustments, regional demand fluctuations, and price trends of alternative materials. Each type of market information is assigned an information type identifier, acquisition cost, and credibility score.

[0050] Furthermore, target entities within the propagation chain are identified that can have their response rule parameters modified by information from the target market. Modification relationships can be established through a mapping table between information types and entity roles. For example, information on changes in material supplier capacity corresponds to modifying the strategy strength coefficient and trigger parameters of the material supplier node; information on anticipated policy adjustments corresponds to modifying the conditional probability distribution of the owner node.

[0051] Next, the process involves traversing the propagation chain from the target entity downstream to identify the set of affected downstream entities. The traversal can employ a breadth-first search, with a maximum depth of, for example, 5 levels. The set of affected downstream entities is stored as a list, containing the downstream entity identifier, propagation path length, and propagation delay time.

[0052] Furthermore, based on the preliminary predicted multi-party equilibrium state, the sensitivity of the strategy adjustment of each entity in the downstream entity set to the error in the project cost prediction is calculated.

[0053] Specifically, for each downstream entity, assuming a unit disturbance (e.g., 1%) in its strategy intensity, the project cost forecast is recalculated. The project cost forecast equals the baseline cost plus the sum of cost changes caused by the strategy intensity of all entities. The cost change is obtained from the cost impact coefficient table based on the entity's role and strategy type. The difference between the project cost forecast before and after the disturbance is divided by the unit disturbance magnitude to obtain the sensitivity of that downstream entity.

[0054] Furthermore, the sensitivity of each downstream entity is weighted and summed, and the result is used as the expected value of the target market information.

[0055] Optionally, when performing weighted summation, the influence range of each downstream entity in the propagation chain is determined, where the influence range is calculated based on the number of other entities reachable from that downstream entity. A weight is determined based on the influence range of each downstream entity, with the weight positively correlated with the influence range; for example, the weight equals the downstream entity's influence range divided by the sum of the influence ranges of all downstream entities. Based on the determined weights, a weighted summation is performed on the sensitivity of each downstream entity. The expected value represents the expected reduction in engineering cost prediction error after acquiring this market information. For example, if a downstream entity's influence range includes 10 other entities, and the total influence range of all affected downstream entities is 100, then that entity's weight can be set to 0.1. Finally, based on the determined weights, a weighted summation is performed on the sensitivity of each downstream entity. In this way, the sensitivity of entities with greater influence in the network (able to influence more people) is assigned a higher weight, making the calculation of expected value more reflective of the global improvement potential of market information on the overall prediction accuracy of the network. The result of this weighted sum is the expected value of this type of market information, expressed in monetary units, and is directly linked to the potential reduction in forecast error.

[0056] Furthermore, when the expected value exceeds the cost of acquiring the target market information, the target market information is acquired, and the response rule parameters of the target entity are modified using the target market information. During modification, the strategy strength coefficient can be directly replaced; for trigger parameters, a weighted average of the modified value and the credibility score can be used; for conditional probability distributions, the corresponding probability vector is adjusted according to the modified value and re-normalized.

[0057] Specifically, the expected value of each type of market information is compared with its acquisition cost. The actual acquisition process (such as automatic purchase or API call) is only triggered when the expected value exceeds the acquisition cost. After acquiring the information, the response rule parameters for the target entity are adjusted based on the information content and credibility score. For example, for trigger parameters, a weighted average of the suggested correction value in the information and the original trigger parameter can be taken, with the weight determined by the credibility score. The correction formula is: New trigger parameter = Corrected value × Credibility score + Original trigger parameter × (1 - Credibility score).

[0058] After parameter correction, the target subject with the corrected parameters is used as a new starting point, and the aforementioned sequential adjustment simulation process is executed again until a new convergence state is reached, thus obtaining the corrected multi-party equilibrium state. Based on this corrected state, the cascaded prediction result of the project cost is recalculated. This result is the final cost prediction value after considering the incremental information of high-value markets.

[0059] The corrected multi-party equilibrium state is stored in a structured data format, with each entity corresponding to a record containing entity identifier, strategy type, strategy strength, and final execution time. The cascaded cost prediction equals the sum of cost changes caused by the strategy strengths of all entities in the corrected equilibrium state, plus the baseline cost. The baseline cost is obtained by multiplying the quantities of each item in the bill of quantities by their corresponding unit prices. Cost changes are calculated and obtained from the cost impact coefficient table based on entity roles, strategy types, and strategy strengths. The interaction effect of cost changes between entities is considered during the summation. If multiple entities have conflicting strategy types, the minimum cost increment principle is used to adjust the summation result. The cascaded cost prediction result includes predicted values, prediction confidence intervals, and decomposition of influencing factors. The prediction confidence interval is generated through Monte Carlo simulation based on the correction range of the correction parameters, and the number of simulations can be configured to 1000.

[0060] Through the above steps, this invention not only improves the accuracy of predictions, but more importantly, it introduces an economic decision-making process based on information value. It can intelligently determine which information is worth acquiring and which information is too costly to acquire, thereby avoiding unnecessary spending on invalid or low-value information and achieving an optimal balance between improved prediction accuracy and information acquisition costs.

[0061] Step 301: During project execution, collect actual cost data and compare the actual cost data with the cascaded prediction results of the project cost to construct a cumulative deviation.

[0062] In one possible implementation, multiple data acquisition nodes are pre-set during the project planning phase. These acquisition nodes can correspond to key project milestones, such as "completion of foundation construction," "topping out of the main structure," "completion of electromechanical installation," and "final acceptance." When the project management process reaches the corresponding node, this embodiment of the invention automatically triggers the data acquisition program.

[0063] Furthermore, actual cost data is collected at multiple time points during project execution, and the cascaded prediction results of project costs at the corresponding time points are extracted.

[0064] Specifically, actual cost data can be read in real time from the enterprise's financial system or project management system, including payments already made, confirmed but unpaid accounts payable, and expenses accrued based on contracts and completed work. Based on the actual trigger time of the data collection node, the cascaded cost forecast value that best matches that time is retrieved from the forecast result database. If the times are not perfectly consistent, linear interpolation can be used to calculate the forecast value at that moment. In one possible implementation, if the difference between the actual trigger time and the timestamp of the closest forecast record exceeds a time tolerance threshold (configurable to 24 hours with an adjustable range of 6 to 72 hours), linear interpolation is performed on the two forecast records to obtain the forecast value at the corresponding time point. During linear interpolation, the previous forecast value is added to the time difference ratio multiplied by the change in forecast value. The time difference ratio = (actual trigger time - previous timestamp) ÷ (next timestamp - previous timestamp), and the change in forecast value = next forecast value - previous forecast value. The interpolated forecast value is rounded to an integer and used as the cascaded cost forecast value corresponding to that data collection node.

[0065] Furthermore, the instantaneous deviation between the actual cost data and the cascaded prediction results of the project cost at each time point is calculated, and the deviation time series is constructed in chronological order.

[0066] The formula for calculating instantaneous deviation is: Instantaneous Deviation = Actual Cost Data - Predicted Value. A positive instantaneous deviation indicates an actual cost overrun, while a negative value indicates cost savings. The instantaneous deviation value calculated for each node and its occurrence timestamp are stored sequentially in an array to form a deviation time series.

[0067] Furthermore, the amplitude features, duration features, and trend features of the deviation time series are extracted to construct a multi-dimensional time series feature vector.

[0068] Specifically, magnitude characteristics can include the absolute value of the largest instantaneous deviation in the sequence and the average of the absolute values ​​of all instantaneous deviations. Duration characteristics can be calculated by analyzing the longest continuous interval in the deviation time series where the sign (positive or negative) remains unchanged. Trend characteristics can be obtained by performing a least-squares linear fit on each point in the deviation time series (with time as the independent variable and deviation as the dependent variable), and the slope of the fitted line is the trend characteristic.

[0069] A multidimensional time-series feature vector is constructed by combining amplitude features, duration features, and trend features. This vector is stored as a one-dimensional array with a length of 4, where the elements are, in order, the maximum absolute deviation, the mean absolute deviation, the duration feature, and the trend feature. To eliminate dimensional differences, the feature vector is normalized. The normalized value is calculated as (feature value - historical minimum value of that feature) ÷ (historical maximum value of that feature - historical minimum value of that feature). The normalized feature value ranges from 0 to 1, accurate to three decimal places.

[0070] Furthermore, the multidimensional time-series feature vectors are weighted and fused to obtain the cumulative bias. The weighting configuration used in the weighted fusion is positively correlated with the duration feature and the trend feature.

[0071] For example, a set of weights can be pre-configured: maximum deviation magnitude weight 0.20, average deviation magnitude weight 0.15, duration feature weight 0.35 (adjustable range 0.25 to 0.45), and trend feature weight 0.30 (adjustable range 0.20 to 0.40). The sum of these two weights accounts for 65% of the total weight, reflecting a positive correlation. The sum of the four weights is normalized to 1. The weighted fusion result is the cumulative deviation, calculated as: Cumulative Deviation = Normalized Maximum Absolute Deviation × Corresponding Weight + Normalized Average Absolute Deviation × Corresponding Weight + Normalized Duration Feature × Corresponding Weight + Normalized Trend Feature × Corresponding Weight. The cumulative deviation is a value between 0 and 1, accurate to three decimal places. The closer the value is to 1, the more severe the deviation between the actual cost and the prediction, comprehensively reflecting the degree, trend, and persistence of the deviation between the actual cost and the prediction. This design makes the embodiments of the present invention more sensitive to continuously expanding rather than sporadic deviations.

[0072] Step 302: When the cumulative deviation exceeds the preset deviation threshold, predict the response of each subject to different alternative intervention schemes based on the propagation chain, evaluate the feasibility of each alternative intervention scheme, and select the alternative intervention scheme with the highest feasibility for execution.

[0073] Furthermore, when the cumulative deviation exceeds a preset deviation threshold, at least one alternative intervention plan is generated, and the change in benefits for each subject after the alternative intervention plan is implemented is calculated.

[0074] The preset deviation threshold can be configured to 0.60, with an adjustable range of 0.50 to 0.80.

[0075] Specifically, alternative solutions are retrieved from a predefined intervention solution library. This library may include options such as "replacing with low-cost alternative materials," "negotiating to shorten non-critical path project timelines," "reallocating internal construction resources," and "initiating contract term revision negotiations." For each solution, the economic benefits or losses for each stakeholder (owner, contractor, etc.) are calculated based on a pre-defined benefit-impact matrix. For example, a material replacement solution might increase the owner's procurement costs (reduced benefits) but reduce the general contractor's construction difficulty (increased benefits).

[0076] Furthermore, based on changes in returns, the entities are divided into supporters and opponents. Entities with positive changes in returns are labeled as supporters, and those with negative changes are labeled as opponents.

[0077] Furthermore, based on the propagation chain, a target supporting entity that has upstream influence over the opposing entity is identified from among the supporting entities as the first-stage intervention target.

[0078] In one possible implementation, the process of determining the target supporting entity can be performed as follows: For the influence relationship between each supporting entity and each opposing entity in the propagation chain, the propagation path from the supporting entity to the opposing entity is extracted; wherein, the propagation path includes direct influence paths and indirect influence paths via intermediate entities. A depth-first search algorithm can be used to discover all possible paths. Next, the influence strength of each propagation path is calculated, and the influence strengths of all propagation paths from a supporting entity to all opposing entities are summed to obtain the comprehensive influence score of the supporting entity. The influence strength of a direct influence path is equal to the response coefficient on that edge (which can be extracted from the conditional probability distribution); the influence strength of an indirect influence path is equal to the product of the response coefficients of each edge on the path. Finally, the time delay parameters of each edge in the propagation chain are extracted, and the average propagation time delay required for the strategy change of each supporting entity to propagate to the opposing entity is calculated; and the ratio of the comprehensive influence score to the average propagation time delay is calculated as the intervention priority index, and the supporting entity with the highest intervention priority index is selected as the intervention target for the first stage. The larger the index value, the stronger the influence on more opposing entities in a shorter time when intervening in that supporting entity, and the better the efficiency and effect of the intervention.

[0079] Furthermore, after determining the first-stage intervention target, starting from the first-stage intervention target, the impact of its decision adjustment along the propagation chain on downstream opposing entities is predicted, so as to identify the original opposing entities that have turned into supporters, and to iteratively design subsequent stages to form a phased combined intervention path.

[0080] Specifically, a simulated intervention incentive (e.g., providing compensation in exchange for cooperation) is applied to the target supporter, and their potential new strategies are predicted based on their response rules. This new strategy is then treated as an upstream event and propagated downstream along the chain, observing which initially opposing downstream entities will change their payoff status due to the new strategy (e.g., from negative to positive payoff). These entities that change their stance are labeled as "former opposing entities that have become supporters." Subsequently, the supporters from the first stage are merged with these transformed supporters as the joint driving force for the second stage, and the above process is repeated until no new entities transform or the maximum number of iterations is reached. This design creates a phased combined intervention path that includes multiple intervention stages and gradually dismantles the opposing forces.

[0081] For each of the aforementioned phased combined intervention paths, it is necessary to further accurately predict its final effect after execution. In one possible implementation, this prediction process may include: predicting the response strategy of each subject based on its response rules for each stage of the phased combined intervention path; propagating the response strategy along the propagation chain to downstream subjects and predicting the counter-response strategies of downstream subjects to construct a multi-layered response game tree; using backward induction on the game tree, tracing back from the leaf nodes to the root node to determine the strategy selection of each subject under the objective of maximizing profit, and extracting the endpoint state of the strategy evolution path as the equilibrium state of that stage.

[0082] Specifically, the root node of the game tree represents the state before the intervention phase begins. Starting from the root node, the first-level child nodes represent all possible response strategies that the intervention target entity might choose (e.g., cooperation, partial cooperation, or non-cooperation). Each child node then generates the next level of child nodes, representing the counter-response strategies that downstream entities might adopt after observing the upstream strategy. This extends in a tree-like structure, and the depth of the tree can be set as needed (e.g., 3 levels). The leaf nodes of the tree represent all possible game endgames. When using backward induction, calculations are performed starting from the leaf nodes: assuming the game reaches a certain leaf node, what is the final payoff for each entity? Then, tracing back one level, for an intermediate node, the entity it represents, when faced with multiple optional child nodes (corresponding to their different strategy choices), will rationally choose the child node that maximizes its own payoff. This maximized payoff value is taken as the payoff value of the intermediate node, and other non-optimal branches are pruned. Through layer-by-layer backtracking calculations from the leaf to the root, the unique optimal strategy path starting from the root node can finally be determined, and the endpoint state of this path is the most likely equilibrium state to be reached by the intervention in this phase. This state describes the final strategy choices and outcomes for each entity.

[0083] Furthermore, for each phased combined intervention path, the intervention costs of each stage are summed up with the deviation correction effect at the final equilibrium state, and the combined intervention path that optimizes the cost-effectiveness ratio is selected for execution. Intervention costs include incentive expenditures and implementation costs at each stage. The deviation correction effect is the reduction in cumulative deviation compared to before the intervention, calculated after implementing the intervention plan. The cost-effectiveness ratio equals the total intervention cost divided by the deviation correction effect. This embodiment of the invention selects the plan with the smallest ratio, which means achieving the most significant deviation correction effect at the lowest cost, and issues this as the final execution instruction to the project management team.

[0084] Through the above process, this embodiment can not only provide timely warnings when cost deviations occur, but also intelligently design a precise intervention path with the lowest cost and least resistance based on a deep understanding of the complex game relationship between the project participants, thereby efficiently and smoothly bringing the project cost back to the expected track.

[0085] In one possible implementation, when predicting the responses of each agent to different alternative intervention schemes, the game tree is constructed and solved for each stage of the phased combined intervention path using the following steps: First, create a root node. The data state corresponding to the root node includes the current strategy state and current profit value of each subject at the start of the current phase. Set this root node as the current node.

[0086] Secondly, a recursive node expansion process is executed. For a specific subject corresponding to the current node (starting from the intervention target subject, and proceeding sequentially in the propagation chain as decision-making subjects), the conditional probability distribution in the subject's response rules is read. All candidate response strategy types with probabilities greater than a preset threshold (e.g., 0.05) are extracted from this conditional probability distribution. For each candidate response strategy type, a child node is created for the current node. Each child node inherits the data state of its parent node and applies the changes in strategy state and payoff resulting from the subject selecting the corresponding response strategy. The payoff change is equal to the change in the subject's strategy strength multiplied by the payoff impact coefficient per unit of strategy strength.

[0087] Third, for each newly created child node, it is treated as the new current node, and its immediate downstream entity in the propagation chain is selected as the decision-making entity for the next round. The above node expansion process is repeated until the preset maximum depth of the game tree (e.g., three layers) is reached, or the current node has no unexpanded downstream entities in the propagation chain. The nodes generated at this point are called leaf nodes. The data state of the leaf nodes represents the final strategy state and cumulative payoff of each entity after evolving along the game path.

[0088] Fourth, after all leaf nodes are generated, a bottom-up backward induction process is performed on the game tree. This process starts from the parent node closest to the leaf node and proceeds upwards layer by layer until the root node. For each non-leaf node, the identifier of the decision-maker corresponding to that node is first read. Then, all child nodes of that node are obtained. The payoff value of the decision-maker is extracted from each child node. The child node with the maximum payoff value is marked as the optimal choice branch for that decision-maker, and this maximum payoff value is used as the payoff value of the parent node in that decision-maker dimension. For the payoff values ​​of other non-decision-makers in the parent node, the weighted average of the corresponding payoff values ​​of all its child nodes is taken, with the weight being the probability of each child node being selected. The calculated comprehensive payoff value is stored in the parent node.

[0089] Fifth, when the reverse induction process traces back to the root node, the path formed by the optimal choice branch along each layer of labels starting from the root node is extracted. The leaf node data state corresponding to the endpoint of this path is the most likely equilibrium state to be reached at this stage under the rational decisions of all parties.

[0090] Based on the above technical solution, this invention obtains the decision-making behavior records of multiple project participants in the historical data of engineering cost and extracts the strategic response behavior of each participant after observing upstream events to identify response rules. Based on the response rules, a propagation chain with time delay parameters and triggering parameters is constructed, which can characterize the dynamic relationship and triggering sequence of strategy propagation among participants. This solves the problem of existing methods ignoring the interaction between participants and realizes accurate modeling of the cost impact transmission process.

[0091] Furthermore, by taking the upstream event of the target as the initial impact and simulating the sequential adjustment process of each subject along the propagation chain until a preliminary multi-party equilibrium state is reached, the present invention can predict the cascading transmission results of strategy adjustments among the subjects, avoid the limitations of static prediction, and improve the rationality of cost prediction.

[0092] Furthermore, this invention calculates the expected value of target market information in improving the initial forecast and updates the propagation chain parameters when the value exceeds the cost to obtain the corrected equilibrium state and cascaded forecast results. It can selectively optimize based on information value, avoid the blindness of information acquisition, and achieve a balance between forecast accuracy and information acquisition cost.

[0093] Furthermore, this invention collects actual cost data during project execution and compares it with cascaded prediction results to construct cumulative deviations. When the deviation exceeds a threshold, it predicts the subject's response to the intervention plan based on the propagation chain and selects the most feasible plan for execution. This enables intervention decision-making based on the propagation relationship between subjects, reduces the risk of blind intervention, and achieves dynamic response and controllable correction of cost deviations.

[0094] Based on the same inventive concept, embodiments of the present invention also provide an engineering cost big data analysis and intelligent decision-making support optimization system. This system includes a first unit, a second unit, and a third unit.

[0095] The first unit is used to acquire decision-making behavior records of multiple project participants in historical engineering cost data, extract the strategic response behavior of each participant after observing upstream events based on the decision-making behavior records, identify the response rules of each participant, and construct a propagation chain representing the strategy propagation relationship between each participant based on the response rules of each participant; wherein, each directed edge of the propagation chain is labeled with a time delay parameter and a trigger parameter.

[0096] The second unit is used to respond to a target upstream event. Taking the target upstream event as the initial impact, it simulates the sequential adjustment process of each subject along the propagation chain based on the time delay parameters and trigger parameters in the propagation chain to obtain a preliminary predicted multi-party equilibrium state. Based on the preliminary predicted multi-party equilibrium state, it calculates the expected value of acquiring target market information to improve the preliminary predicted multi-party equilibrium state. When the expected value exceeds the acquisition cost of the target market information, it updates the parameters of the propagation chain based on the target market information, re-simulates the sequential adjustment process to obtain a corrected multi-party equilibrium state, and calculates the cascaded prediction result of the engineering cost.

[0097] The third unit is used to collect actual cost data during project execution, compare the actual cost data with the cascaded prediction results of the project cost to construct a cumulative deviation; when the cumulative deviation exceeds a preset deviation threshold, predict the response of each subject to different alternative intervention schemes based on the propagation chain, evaluate the feasibility of each alternative intervention scheme, and select the alternative intervention scheme with the highest feasibility for execution.

[0098] In this embodiment, the functions of the first unit, the second unit, and the third unit specifically correspond to the relevant steps in the above method embodiments, and will not be repeated here. Each unit in this system can be deployed as a software module on a cloud server, and its function can be implemented by one or more processors executing the corresponding program code.

[0099] It is understood that those skilled in the art will recognize that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the invention. In the several embodiments provided by the present 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 example, the division of units is merely 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 mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0100] The units described as separate components may or may not be physically separate. The components shown as units 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.

[0101] In addition, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module.

[0102] In the above embodiments, the functions of each functional unit can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs (DVDs)), or semiconductor media (e.g., solid-state drives (SSDs)).

[0103] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0104] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for big data analysis and intelligent decision support optimization of engineering cost, characterized in that, include: The decision-making behavior records of multiple project participants in the historical data of engineering cost are obtained. Based on the decision-making behavior records, the strategic response behavior of each participant after observing upstream events is extracted, and the response rules of each participant are identified. Based on the response rules of each participant, a propagation chain representing the strategy propagation relationship between the participants is constructed. Each directed edge of the propagation chain is labeled with a time delay parameter and a trigger parameter. In response to a target upstream event, the target upstream event is used as the initial impact. Based on the time delay parameters and trigger parameters in the propagation chain, the sequential adjustment process of each subject is simulated along the propagation chain to obtain a preliminary predicted multi-party equilibrium state. Based on the preliminary predicted multi-party equilibrium state, the expected value of acquiring target market information for improving the preliminary predicted multi-party equilibrium state is calculated. When the expected value exceeds the acquisition cost of the target market information, the parameters of the propagation chain are updated based on the target market information, and the sequential adjustment process is re-simulated to obtain a corrected multi-party equilibrium state. The cascaded prediction result of the engineering cost is then calculated. During project execution, actual cost data is collected and compared with the cascaded prediction results of the project cost to construct a cumulative deviation. When the cumulative deviation exceeds a preset deviation threshold, the response of each subject to different alternative intervention schemes is predicted based on the propagation chain, the feasibility of each alternative intervention scheme is evaluated, and the alternative intervention scheme with the highest feasibility is selected for execution.

2. The engineering cost big data analysis and intelligent decision-making support optimization method according to claim 1, characterized in that, Obtain decision-making behavior records of multiple project participants from historical engineering cost data. Based on these records, extract the strategic response behavior of each participant after observing upstream events, and identify the response rules for each participant, including: Extract the decision timestamp sequence and strategy type sequence of each entity after the occurrence of the upstream event from the historical data of the project cost; A correlation analysis is performed on the strategy type sequence of each subject and the strategy type sequence of the corresponding upstream subject to extract the conditional probability distribution of the corresponding subject executing the corresponding response strategy after observing various upstream strategies; Based on the decision timestamp sequence, calculate the time interval distribution for each subject from observing the upstream strategy to executing the response strategy, and determine the time delay parameter based on the statistical characteristics of the time interval distribution; Based on the conditional probability distribution, identify the minimum upstream policy strength threshold required to trigger the response strategy, and use the minimum upstream policy strength threshold as the triggering parameter.

3. The engineering cost big data analysis and intelligent decision-making support optimization method according to claim 1, characterized in that, In response to an upstream event, the upstream event is used as the initial shock. Based on the time delay parameters and triggering parameters in the propagation chain, the sequential adjustment process of each entity is simulated along the propagation chain to obtain a preliminary predicted multi-party equilibrium state, including: The intensity parameters and influence direction of the upstream event of the target are obtained and used as the initial impact event; Based on the time delay parameters in the propagation chain, the time delay at which each subject observes the policy changes of the upstream subject is determined, and the policy propagation time sequence between subjects is constructed. Identify one or more initial response entities directly affected by the initial impact event from the propagation chain, and determine an initial strategy adjustment scheme based on the response rules of the initial response entities; The initial strategy adjustment scheme is used as a new upstream strategy and input into the propagation chain. Based on the strategy propagation sequence and the triggering parameters in the propagation chain, the triggered downstream entities are identified round by round and their strategy adjustment schemes are calculated to form a sequential adjustment sequence. When the policy adjustment magnitude of all subjects in the sequential adjustment sequence is lower than the convergence threshold within consecutive preset rounds, the current policy state of each subject is extracted as the preliminary predicted multi-party equilibrium state.

4. The engineering cost big data analysis and intelligent decision-making support optimization method according to claim 1, characterized in that, Based on the preliminary predicted multi-party equilibrium state, the expected value of acquiring target market information for improving the preliminary predicted multi-party equilibrium state is calculated. When the expected value exceeds the cost of acquiring the target market information, the parameters of the propagation chain are updated based on the target market information, including: In the propagation chain, target entities whose response rule parameters can be modified by the target market information are identified; Along the propagation chain, traverse from the target entity downstream to identify the set of affected downstream entities; Based on the preliminary predicted multi-party equilibrium state, the sensitivity of the strategy adjustment of each entity in the downstream entity set to the error in the project cost prediction is calculated. The sensitivity of each downstream entity is weighted and summed, and the result is used as the expected value of the target market information. When the expected value is greater than the cost of acquiring the target market information, the target market information is acquired, and the target market information is used to correct the response rule parameters of the target entity.

5. The engineering cost big data analysis and intelligent decision-making support optimization method according to claim 4, characterized in that, The sensitivity of each downstream entity is weighted and summed, including: Determine the influence range of each downstream entity in the propagation chain, wherein the influence range is calculated based on the number of other entities reachable from that downstream entity; The weight of each downstream entity is determined based on its influence range, where the weight is positively correlated with the influence range. Based on the determined weights, a weighted summation operation is performed on the sensitivity of each downstream entity.

6. The method for big data analysis and intelligent decision-making support optimization of engineering cost according to claim 1, characterized in that, During project execution, actual cost data is collected and compared with the cascaded project cost prediction results to construct a cumulative deviation, including: Actual cost data is collected at multiple time points during project execution, and the cascaded prediction results of project cost at the corresponding time points are extracted. Calculate the instantaneous deviation between the actual cost data and the cascaded prediction results of the project cost at each time point, and construct a deviation time series in chronological order; The amplitude features, duration features, and trend features of the deviation time series are extracted to construct a multi-dimensional time series feature vector; The multidimensional time-series feature vector is weighted and fused to obtain the cumulative deviation; wherein the weight configuration used in the weighted fusion is positively correlated with the duration feature and the trend feature.

7. The method for big data analysis and intelligent decision-making support optimization of engineering cost according to claim 1, characterized in that, When the cumulative deviation exceeds a preset deviation threshold, the responses of each subject to different alternative intervention schemes are predicted based on the propagation chain, the feasibility of each alternative intervention scheme is evaluated, and the alternative intervention scheme with the highest feasibility is selected for execution, including: When the cumulative deviation exceeds a preset deviation threshold, at least one alternative intervention plan is generated, and the change in benefits for each subject after the alternative intervention plan is implemented is calculated. Based on changes in returns, the entities are divided into supporters and opponents; Based on the propagation chain, target supporting entities that have upstream influence over the opposing entities are identified from among the supporting entities and are used as the first-stage intervention targets. Starting with the first-stage intervention target, predict the impact of its decision adjustment along the propagation chain on downstream opposing entities, in order to identify the original opposing entities that have turned into supporters, and iteratively design subsequent stages to form a phased combined intervention path. For each phased combined intervention path, the intervention costs of each stage are summed up with the deviation correction effect in the final stage equilibrium state, and the combined intervention path that optimizes the cost-effectiveness ratio is selected for execution.

8. The engineering cost big data analysis and intelligent decision-making support optimization method according to claim 7, characterized in that, Based on the propagation chain, target supporting entities with upstream influence over the opposing entity are identified from among the supporting entities, including: For the influence relationship between each supporting entity and each opposing entity in the propagation chain, the propagation path from the supporting entity to the opposing entity is extracted; wherein, the propagation path includes direct influence paths and indirect influence paths via intermediate entities; Calculate the influence intensity of each propagation path, and sum the influence intensity of each propagation path from a supporting entity to all opposing entities to obtain the comprehensive influence score of the supporting entity; Extract the time delay parameters of each side in the propagation chain, and calculate the average propagation time delay required for the policy change of each supporting subject to propagate to the opposing subject; The ratio of the comprehensive influence score to the average dissemination time lag is calculated as an intervention priority indicator, and the supporting entity with the highest intervention priority indicator is selected as the intervention target for the first stage.

9. The method for big data analysis and intelligent decision-making support optimization of engineering cost according to claim 7, characterized in that, Predicting the responses of various stakeholders to different alternative intervention options, including: For each stage of the phased combined intervention path, predict the response strategy of each subject based on the response rules of each subject; The response strategy is propagated along the propagation chain to downstream entities, and the counter-response strategy of the downstream entities is predicted to construct a multi-layered response game tree. The game tree is subjected to backward induction, tracing back from the leaf nodes to the root node to determine the strategy choices of each subject under the objective of maximizing profit, and the endpoint state of the strategy evolution path is extracted as the equilibrium state of that stage.

10. A big data analysis and intelligent decision-making support optimization system for engineering cost, characterized in that, include: The first unit is used to acquire decision-making behavior records of multiple project participants in historical engineering cost data, extract the strategic response behavior of each participant after observing upstream events based on the decision-making behavior records, identify the response rules of each participant, and construct a propagation chain representing the strategy propagation relationship between each participant based on the response rules of each participant; wherein, each directed edge of the propagation chain is labeled with a time delay parameter and a trigger parameter. The second unit is used to respond to a target upstream event, taking the target upstream event as the initial impact, and simulating the sequential adjustment process of each subject along the propagation chain based on the time delay parameters and trigger parameters in the propagation chain to obtain a preliminary predicted multi-party equilibrium state; and, based on the preliminary predicted multi-party equilibrium state, calculating the expected value of acquiring target market information for improving the preliminary predicted multi-party equilibrium state, and when the expected value exceeds the acquisition cost of the target market information, updating the parameters of the propagation chain based on the target market information, resimulating the sequential adjustment process to obtain a corrected multi-party equilibrium state, and calculating the cascaded prediction result of the engineering cost; The third unit is used to collect actual cost data during project execution, compare the actual cost data with the cascaded prediction results of the project cost to construct a cumulative deviation; and when the cumulative deviation exceeds a preset deviation threshold, predict the response of each subject to different alternative intervention schemes based on the propagation chain, evaluate the feasibility of each alternative intervention scheme, and select the alternative intervention scheme with the highest feasibility for execution.

Images