Real-time evaluation and early warning system for engineering cost based on big data and reinforcement learning

Abstract

The application discloses a real-time evaluation and early warning system for engineering cost based on big data and reinforcement learning, comprising: a data processing module for processing data to generate historical engineering business data and real-time engineering business data; a historical chain construction module for connecting to generate a historical hidden cost debt suppression and rebound redemption chain library; a debt identification module for identifying current construction events and matching to generate hidden cost debt items; a suppression identification module for identifying and correlating to generate hidden cost debt suppression events; a rebound prediction module for generating hidden cost debt suppression and rebound prediction chains and cost change decomposition data; a reinforcement learning evaluation module for generating hidden cost debt release schemes and engineering cost dynamic evaluation results using the advantage actor-critic algorithm; and an early warning module for generating engineering cost early warning results. The application combines big data analysis and reinforcement learning to realize hidden cost debt identification, rebound prediction and real-time and accurate early warning.

Description

Technical Field

[0001] This invention relates to the field of intelligent assessment and early warning technology, and in particular to a real-time engineering cost assessment and early warning system based on big data and reinforcement learning. Background Technology

[0002] Current engineering cost assessments typically involve collecting data on project quantities, material consumption, equipment input, construction progress, and cost confirmation. This data is then compared to the planned cost, and historical project data is used to predict future cost changes. When the cost deviation exceeds a set range, the system generates an overspending alert to assist engineers in adjusting resource allocation, construction arrangements, and cost confirmation schedules.

[0003] However, existing technologies primarily focus on confirmed book costs, making it difficult to identify hidden cost burdens that have already been incurred but whose costs have not yet been recognized. When construction continues while the increase in book costs decreases, it is easy to misjudge the lag in cost recognition as a cost reduction. It is also difficult to track the evolution of the concentrated rebound after the suppression of hidden cost liabilities, resulting in a lack of long-term effect assessment for disposal actions and problems of delayed and distorted cost warnings.

[0004] Therefore, how to provide a real-time engineering cost assessment and early warning system based on big data and reinforcement learning is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a real-time engineering cost assessment and early warning system based on big data and reinforcement learning. This invention integrates big data analysis and reinforcement learning to achieve the identification of implicit cost debts, prediction of rebounds, and real-time accurate early warning.

[0006] The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to embodiments of the present invention includes: The data processing module is used to process historical engineering datasets and real-time engineering data streams, and generate historical engineering business data and real-time engineering business data. The historical chain construction module is used to generate historical implicit cost debt items, historical debt evolution data, and historical disposal feedback data based on historical engineering business data, and connect them to generate a historical implicit cost debt suppression and rebound realization chain library; The debt identification module is used to identify current construction events based on real-time engineering business data and match them with the historical implicit cost debt suppression and rebound realization chain library to generate implicit cost debt items; The suppression identification module is used to generate a book cost increment sequence based on real-time engineering business data, identify cost increment decline segments and cost suppression behaviors, and generate implicit cost debt suppression events by correlation. The rebound prediction module is used to generate a construction impact state sequence and match it with the historical implicit cost debt suppression and rebound realization chain library to generate implicit cost debt suppression and rebound prediction chain and cost change decomposition data. The reinforcement learning evaluation module is used to generate engineering cost status data based on implicit cost debt items, implicit cost debt suppression events, construction impact state sequences, implicit cost debt suppression and rebound prediction chains, and cost change decomposition data. It uses the dominant actor-commentator algorithm to generate implicit cost debt release plans and dynamic evaluation results of engineering costs. The early warning module is used to identify the risk of cost rebound by combining the dynamic evaluation results of engineering cost and generate early warning results for engineering cost.

[0007] Optionally, the data processing module includes: Deduplication, missing data completion, and data format standardization are performed on historical engineering datasets and real-time engineering data streams respectively to generate historical cleaned data and real-time cleaned data. The historical cleaning data and real-time cleaning data are aligned by time, matched by engineering object identifiers, and associated by construction procedures to generate historical engineering business data and real-time engineering business data.

[0008] Optionally, the history chain construction module includes: Identify historical construction events from historical engineering business data, and correlate these historical construction events with subsequent cost confirmation records to generate historical event cost association data; The time interval between the occurrence of a historical construction event and the formation of a cost confirmation record is defined as the historical debt evolution interval. The historical cost confirmation duration is obtained, and the cost burdens that have formed a construction impact but have not yet been confirmed are identified, thus generating historical implicit cost debt items. By performing time correlation analysis on changes in resource input, construction status, and costs within the historical debt evolution segment, historical debt suppression and historical debt rebound segments are generated, and these segments are then connected to generate historical debt evolution data. Identify disposal actions within the historical debt evolution phases and correlate them with historical debt evolution data before and after execution to generate historical disposal feedback data; Based on the project object, construction sequence, and time relationship, historical hidden cost debt items, historical debt evolution data, and historical disposal feedback data are connected to generate a historical hidden cost debt suppression and rebound realization chain, and a historical hidden cost debt suppression and rebound realization chain library is generated.

[0009] Optionally, the debt identification module includes: Identify the current construction event from real-time engineering business data, search for the cost confirmation record formed after the current construction event occurs according to the project object and construction procedure, identify the current construction event that has not yet formed a cost confirmation record as the current construction event to be matched, calculate the time interval between the current construction event occurrence time and the current time, and generate the cost confirmation missing duration; Based on the project object, construction procedure, and duration of missing cost confirmation, the current construction event to be matched is matched with the historical implicit cost debt suppression and rebound realization chain in the historical implicit cost debt suppression and rebound realization chain library, and candidate historical implicit cost debt suppression and rebound realization chains are screened. Align the current construction event to be matched with the historical construction events in the candidate historical implicit cost debt suppression and rebound realization chain according to the construction procedure and occurrence time, compare the project object, construction procedure and cost confirmation missing time, and determine the target historical implicit cost debt suppression and rebound realization chain; Map the historical implicit cost debt items in the target historical implicit cost debt suppression and rebound realization chain to the current construction event to be matched, and generate implicit cost debt items by combining the project object, construction procedure and cost confirmation missing time of the current construction event.

[0010] Optionally, the suppression recognition module includes: Based on the project object and construction procedure, the cumulative confirmed costs in the real-time project business data are differentiated by adjacent time points, and the cumulative confirmed cost differences are arranged in chronological order to generate a book cost increment sequence. The sequence of incremental book costs is divided into continuous segments, and the continuous segments in which the cost increment decreases continuously are identified as the initial cost increment decreasing segments. The changes in resource input, construction status, disposal actions, and cost changes during the period corresponding to the initial cost increment decrease segment are time-aligned. Based on the changes in construction status, the construction progress increment is calculated. The initial cost increment decrease segment with a continuous increase in construction progress increment is identified as the cost increment decrease segment. Construction behaviors where resource input, disposal action execution, or cumulative confirmed costs have not increased synchronously are identified, and cost suppression behaviors are generated. Based on the consistency of the project object, the correspondence of the construction procedures, and the overlap of the occurrence time, the segments of cost increase reduction, cost suppression behaviors, and implicit cost debt items are linked to generate implicit cost debt suppression events.

[0011] Optionally, the rebound prediction module includes: Starting from the time of occurrence of the hidden cost debt suppression event, the changes in resource input, construction status, disposal actions and cost changes of the corresponding project objects and construction procedures are obtained from the real-time engineering business data thereafter, and arranged according to the time of occurrence to generate a construction impact status sequence. Based on the correspondence of project objects, the connection of construction procedures, and the continuity of occurrence time, the implicit cost debt items, implicit cost debt suppression events, and construction impact status sequences are connected to generate current debt suppression data; The current debt suppression data is matched with the historical implicit cost debt suppression and rebound realization chain in the historical implicit cost debt suppression and rebound realization chain library. The construction impact status sequence is aligned with the historical debt evolution data in each historical implicit cost debt suppression and rebound realization chain to determine the construction progress. The occurrence order, duration and magnitude of resource input changes, construction status changes and cost changes are compared to determine the target historical implicit cost debt suppression and rebound realization chain and the target matching position. Obtain the chain segment following the target matching position in the target's historical implicit construction cost debt suppression and rebound realization chain. Adjust the occurrence time and amount of cost changes in the chain segment according to the duration ratio and change ratio between the construction impact state sequence and the chain segment. Connect the current debt suppression data to generate an implicit construction cost debt suppression and rebound prediction chain. By mapping the cost changes in the implicit cost debt suppression and rebound prediction chain to the cost changes resulting from the release of implicit cost debt, the cost changes resulting from the removal of cost suppression behavior, and the cost changes resulting from the continued impact of construction, cost change decomposition data is generated.

[0012] Optionally, the reinforcement learning evaluation module includes: Based on the project object, construction procedure, and occurrence time, the implicit cost debt items, implicit cost debt suppression events, construction impact state sequence, implicit cost debt suppression and rebound prediction chain, and cost change decomposition data are matched to generate project cost status data. The project cost status data is matched with the historical implicit cost debt items and historical debt evolution data in the historical implicit cost debt suppression and rebound realization chain library. The matching historical implicit cost debt suppression and rebound realization chain is determined according to the debt formation location, debt suppression duration, construction impact change order and cost change magnitude, and the corresponding historical disposal feedback data is obtained. Identify disposal actions from historical disposal feedback data, and correlate each disposal action with historical implicit construction cost debts before and after execution, historical debt evolution data and cost changes to generate disposal action feedback data and historical status transition data; The dominant actor-commentator algorithm is adopted, which uses the engineering cost status data as the state and the disposal actions in the disposal action set as the actions. Based on the disposal action feedback data, the single cost return is generated, and combined with the status after the disposal action is executed in the historical state transition data and the single cost return of the corresponding disposal action, the long-term cost return of each disposal action is iteratively updated. Based on the long-term cost return of each disposal action, and combined with the cost occurrence time and cost change decomposition data in the implicit cost debt suppression and rebound prediction chain, the execution order and execution time of the disposal actions are determined, an implicit cost debt release plan is generated, and the disposal actions, execution order and execution time in the implicit cost debt release plan are mapped to the implicit cost debt suppression and rebound prediction chain. The occurrence time and change amount of the corresponding cost changes are adjusted to generate the dynamic evaluation result of the project cost.

[0013] Optionally, the early warning module includes: By correlating the suppression of hidden construction cost debt with the rebound prediction chain, the decomposition data of cost changes, the release plan of hidden construction cost debt, and the dynamic evaluation results of engineering cost, data on the identification of cost rebound is generated. Identify the chain segment where costs continuously increase after the execution of disposal actions, and generate cost rebound segments by combining cost change decomposition data; By comparing the cost rebound segment with the predicted cost changes for the corresponding time period, the risk of cost rebound can be determined. Based on the project objects, construction procedures, occurrence time, and cost changes corresponding to the risk of cost rebound, project cost early warning results are generated.

[0014] The beneficial effects of this invention are: This invention cleanses, aligns, matches, and associates construction object identifiers with historical engineering datasets and real-time engineering data streams. It unifies data scattered across quantity records, material usage records, equipment operation records, construction status records, and cost confirmation records under the same engineering object and construction procedure, forming engineering business data that continuously reflects the relationship between construction activities and cost changes. This reduces time misalignment and object mismatch between different data sources, ensuring that subsequent cost assessments no longer rely solely on isolated accounting cost data. It also improves the correlation between construction process data and cost confirmation data, providing a complete and continuous data foundation for identifying hidden cost risks.

[0015] This invention identifies historical implicit cost liabilities based on the temporal relationship between historical construction events and subsequent cost confirmation records. These liabilities have already been incurred but their costs have not yet been fully confirmed. The invention connects historical debt suppression segments, historical debt rebound segments, and historical disposal feedback data to form a chain library of historical implicit cost liability suppression and rebound realization. This chain library can reconstruct the evolution of unconfirmed costs from their formation, temporary suppression, subsequent rebound, to disposal response. It overcomes the problem of traditional cost assessments that only compare confirmed costs with planned costs and struggle to track unrecorded cost burdens, enabling the system to identify potential cost pressures before formal cost confirmation.

[0016] This invention further identifies cost-suppression behavior from the abnormal state of continuously decreasing book cost increments while construction progress continues to increase, and associates the cost increment decrease segment, cost-suppression behavior, and implicit cost debt items as implicit cost debt suppression events. By aligning the current construction impact state sequence with the historical implicit cost debt suppression and rebound realization chain to match the construction progress and evolution process, it is possible to predict the occurrence time and cost changes of subsequent implicit cost debt rebounds, and decompose the predicted costs into cost changes of implicit cost debt release, cost suppression behavior removal, and the continued formation of construction impacts, thereby improving the interpretability of cost change sources.

[0017] This invention employs a dominant actor-commentator algorithm, combining project cost status data, historical disposal feedback data, and historical state transition data to update the long-term cost returns of different disposal actions. Based on this, it generates implicit cost debt release plans and dynamic project cost assessment results. Compared to methods that select disposal actions solely based on current cost deviations, this invention simultaneously considers the ongoing impact of disposal actions on subsequent cost rebounds, changes in implicit cost debt, and cost fluctuations. It also identifies continuous cost increases that persist even after the release plan is implemented, generating early warning results corresponding to the project object, construction sequence, occurrence time, and cost change amount. This improves the real-time nature, foresight, and accuracy of early warnings in project cost assessment. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overall flowchart of the real-time engineering cost assessment and early warning system based on big data and reinforcement learning proposed in this invention; Figure 2 This diagram illustrates the construction of a historical implicit cost debt suppression and rebound realization chain library in the real-time engineering cost assessment and early warning system based on big data and reinforcement learning proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-2 A real-time engineering cost assessment and early warning system based on big data and reinforcement learning includes: The data processing module is used to acquire historical engineering datasets and real-time engineering data streams, and to perform data cleaning, time alignment, engineering object identification matching, and construction procedure association, respectively, to generate historical engineering business data and real-time engineering business data. The historical chain construction module is used to identify historical construction events and cost confirmation records of the same project object after the occurrence of historical construction events based on historical engineering business data. It combines changes in resource input, construction status, disposal actions and cost changes during the period to generate historical implicit cost debt items, historical debt evolution data and historical disposal feedback data. It connects historical implicit cost debt items, historical debt evolution data and historical disposal feedback data to generate a historical implicit cost debt suppression and rebound realization chain library. The debt identification module is used to identify current construction events based on real-time engineering business data. It matches current construction events that have not yet formed cost confirmation records with the historical implicit cost debt suppression and rebound realization chain library to generate implicit cost debt items. The suppression identification module is used to generate a book cost increment sequence based on real-time engineering business data, identify the cost increment decrease segment in the book cost increment sequence and the corresponding cost suppression behavior, and associate the cost increment decrease segment, cost suppression behavior and implicit cost debt items to generate implicit cost debt suppression events. The rebound prediction module is used to generate a construction impact status sequence based on real-time engineering business data after the occurrence of implicit cost debt suppression events. It matches implicit cost debt items, implicit cost debt suppression events, and construction impact status sequences with the historical implicit cost debt suppression and rebound realization chain library to generate implicit cost debt suppression and rebound prediction chains and cost change decomposition data. The reinforcement learning evaluation module is used to generate project cost status data based on implicit cost debt items, implicit cost debt suppression events, construction impact state sequences, implicit cost debt suppression and rebound prediction chains, and cost change decomposition data. It obtains historical disposal feedback data that matches the project cost status data from the historical implicit cost debt suppression and rebound realization chain library, generates a set of disposal actions, and uses the dominant actor-commentator algorithm to update the long-term cost return of each disposal action, generating implicit cost debt release plans and dynamic project cost evaluation results. The early warning module is used to identify the risk of cost rebound based on the prediction chain of hidden cost debt suppression and rebound, cost change decomposition data, hidden cost debt release plan and dynamic evaluation results of engineering cost, and generate early warning results for engineering cost.

[0021] In this embodiment, the data processing module includes: Deduplication, missing data completion, and data format standardization are performed on historical engineering datasets and real-time engineering data streams respectively to generate historical cleaned data and real-time cleaned data. The historical cleaning data and real-time cleaning data are aligned by time, matched by engineering object identifiers, and associated by construction procedures to generate historical engineering business data and real-time engineering business data.

[0022] In this embodiment, the history chain construction module includes: Identify historical construction events from historical engineering business data, and match historical construction events with subsequent cost confirmation records according to project object, construction procedure and occurrence time to generate historical event cost association data; Specifically, the historical engineering business data includes 31,840 records of engineering completion, material usage, equipment operation, construction status, and cost confirmation from 126 projects. These records are categorized by project object and construction procedure and arranged by occurrence time. The engineering completion, material usage, and equipment operation records corresponding to the same construction process under the same project object and construction procedure are linked together to form historical construction events, resulting in a total of 7,620 historical construction events. Then, the measurement confirmation and settlement records with the same project object and construction procedure are searched from the data after each historical construction event to obtain 6,954 sets of correspondences between historical construction events and cost confirmation records. Based on historical event cost correlation data, the time interval between the occurrence of a historical construction event and the formation of a cost confirmation record is defined as the historical debt evolution interval, and the duration of the historical debt evolution interval is defined as the historical cost confirmation duration. Cost burdens that have formed construction impacts but have not yet been confirmed within the historical debt evolution interval are identified, and historical implicit cost debt items are generated. Specifically, between the occurrence of a historical construction event and the formation of the cost confirmation record, labor hours, material usage, equipment operation time, completed work volume, and cost changes are read. Combined with the corresponding labor unit price, material unit price, equipment shift unit price, and work volume unit price, labor costs of 42,000 yuan, material costs of 106,000 yuan, and equipment costs of 18,000 yuan are obtained. Other construction costs of 20,000 yuan are obtained from the cost changes, resulting in a total of 186,000 yuan of incurred construction costs. Subtracting the cumulative confirmed costs of 112,000 yuan during the same period, 74,000 yuan of unconfirmed costs are obtained. At the same time, the 10-day duration of the historical debt evolution period is determined as the historical cost confirmation period. The unconfirmed costs, historical cost confirmation period, project object, construction procedure, and occurrence time are connected to generate historical implicit cost debt items. The changes in resource input, construction status, and costs within the historical debt evolution segment are correlated over time to generate historical debt suppression segments and historical debt rebound segments. The historical debt suppression segments and historical debt rebound segments are then connected to generate historical debt evolution data. Specifically, the changes in resource input, construction status, and costs within the historical debt evolution segment are arranged according to daily time intervals. For example, the construction progress of a certain construction process increased from 41.2% to 58.6% over 14 consecutive days, the labor input increased from 1260 man-hours to 1778 man-hours, and the daily increase in cumulative confirmed costs decreased from 28,000 yuan to 7,000 yuan. This 14-day data is connected to form the historical debt suppression segment. In the following 9 days, the cumulative confirmed costs increased from 213,000 yuan to 284,000 yuan. This 9-day data is connected to form the historical debt rebound segment. Then, the historical debt suppression segment and the historical debt rebound segment are connected according to the time of occurrence to generate historical debt evolution data. Identify the disposal actions performed within the historical debt evolution period, and correlate the disposal actions with the historical debt evolution data before and after the execution to generate historical disposal feedback data; Specifically, taking the execution time of the disposal action as the center, the changes in construction status and costs for 14 days before and after the execution are obtained respectively. For a certain segment, the construction progress increased from 57.8% to 63.4% in the 14 days before the execution of the cost confirmation action, and the cumulative confirmed cost increased by 16,000 yuan. In the 14 days after the execution, the construction progress increased from 63.4% to 69.1%, and the cumulative confirmed cost increased by 55,000 yuan. The disposal action is matched with the changes in construction status and costs before and after the execution in terms of time to generate historical disposal feedback data. Based on the correspondence of project objects, the connection of construction procedures, and the chronological order, historical hidden cost debt items, historical debt evolution data, and historical disposal feedback data are connected to generate a historical hidden cost debt suppression and rebound realization chain. The historical hidden cost debt suppression and rebound realization chain is then summarized to generate a historical hidden cost debt suppression and rebound realization chain library. Specifically, taking historical implicit construction cost debts as the head of the chain, the corresponding historical debt suppression segments, historical debt rebound segments, and historical disposal feedback data are sequentially connected according to the occurrence time. Each chain records the project object, construction procedure, implicit construction cost debt amount, suppression duration, rebound duration, rebound cost change, and disposal action. The data corresponding to 6954 historical construction events are connected to form 4286 historical implicit construction cost debt suppression and rebound realization chains. The historical implicit construction cost debt suppression and rebound realization chain library is then generated by summarizing the historical implicit construction cost debt suppression and rebound realization chain.

[0023] In this embodiment, the debt identification module includes: Identify the current construction event from real-time engineering business data, search for the cost confirmation record formed after the current construction event occurs according to the project object and construction procedure, identify the current construction event that has not yet formed a cost confirmation record as the current construction event to be matched, calculate the time interval between the current construction event occurrence time and the current time, and generate the cost confirmation missing duration; Specifically, the current construction event is formed by connecting the completion records of the project quantity, material usage records, and equipment operation records in the real-time project business data according to the project object and construction procedure. A certain current construction event occurred on the 18th construction day, which included 320 cubic meters of completed work, 326 cubic meters of material usage, and 42 hours of equipment operation. As of the 27th construction day, the most recently collected data in the real-time project data stream, no cost confirmation record has been formed. This event is identified as a current construction event to be matched, and the cost confirmation missing time of 9 days is generated by subtracting the 18th construction day from the 27th construction day. Based on the project object, construction procedure, and duration of missing cost confirmation, the current construction event to be matched is matched with the historical implicit cost debt suppression and rebound realization chain in the historical implicit cost debt suppression and rebound realization chain library, and candidate historical implicit cost debt suppression and rebound realization chains are screened. Specifically, the project object, construction procedure, and historical cost confirmation duration are read from 4,286 historical implicit construction cost debt suppression and rebound realization chains. The current construction event to be matched is compared with the project object of each historical chain. If they are the same, it is recorded as 0 and if they are different, it is recorded as 1. The construction procedure is compared. If they are the same, it is recorded as 0 and if they are different, it is recorded as 1. Then, the absolute difference between the 9-day cost confirmation missing duration and the cost confirmation duration of each historical chain is divided by the maximum historical cost confirmation duration of 42 days to obtain the duration difference. The three values ​​are added together to generate the candidate distance, and the top 20 historical chains are selected according to the candidate distance from smallest to largest. Align the current construction event to be matched with the historical construction events in the candidate historical implicit cost debt suppression and rebound realization chain according to the construction procedure and occurrence time, compare the project object, construction procedure and cost confirmation missing time, and determine the target historical implicit cost debt suppression and rebound realization chain; Specifically, the occurrence time of the current construction event to be matched is converted into its time position relative to the start time of the corresponding construction procedure. The same method is used to process the historical construction events in the 20 candidate historical chains. The differences in project objects, construction procedures, occurrence time positions, and the differences between the missing cost confirmation time and the historical cost confirmation time are calculated respectively. Then, the normalized differences of each item are added together. The current construction event occurs on the 18th day after the start of the construction procedure and the cost confirmation time is 9 days. The historical chain with the smallest total difference corresponds to the cost confirmation time on the 17th and 10th days after the start of the construction procedure. This historical chain is determined as the target historical implicit cost debt suppression and rebound realization chain. Map the historical implicit cost debt items in the target historical implicit cost debt suppression and rebound realization chain to the current construction event to be matched, and generate implicit cost debt items by combining the project object, construction procedure and cost confirmation missing time of the current construction event; Specifically, the historical chain library contains multiple historical implicit construction cost debt items. One of these historical implicit construction cost debt items is 74,000 yuan. Another historical implicit construction cost debt item corresponding to the target historical implicit construction cost debt suppression and rebound realization chain is 68,000 yuan. Its historical cost confirmation duration is 10 days. The cost confirmation missing duration for the current construction event to be matched is 9 days. Dividing 9 days by 10 days gives a duration ratio of 0.9. Then, multiplying 68,000 yuan by 0.9 gives 61,200 yuan. The 61,200 yuan item is then connected with the project object, construction procedure, and cost confirmation missing duration of the current construction event to be matched to generate an implicit construction cost debt item.

[0024] In this embodiment, the suppression recognition module includes: Based on the project object and construction procedure, the cumulative confirmed costs in the real-time project business data are differentiated by adjacent time points, and the cumulative confirmed cost differences are arranged in chronological order to generate a book cost increment sequence. The sequence of incremental book costs is divided into continuous segments, and the continuous segments in which the cost increment decreases continuously are identified as the initial cost increment decreasing segments. The changes in resource input, construction status, disposal actions, and cost changes during the period corresponding to the initial cost increment decrease segment are time-aligned. Based on the changes in construction status, the construction progress increment is calculated. The initial cost increment decrease segment with a continuous increase in construction progress increment is identified as the cost increment decrease segment. Construction behaviors where resource input, disposal action execution, or cumulative confirmed costs have not increased synchronously are identified, and cost suppression behaviors are generated. Specifically, the cumulative completed work volume is read from the changes in construction status, the difference in work volume between adjacent dates is calculated, and the difference in work volume is divided by the contract work volume of the corresponding construction process to generate the construction progress increment. If a certain initial cost increment decrease segment lasts for 10 days, the construction progress increases from 46.8% to 59.2%, the resource input remains at 48 to 55 man-hours per day, and the cumulative daily confirmed cost increment decreases from 26,000 yuan to 8,000 yuan. During this period, the number of disposal actions is 0. This initial cost increment decrease segment is identified as the cost increment decrease segment, and the corresponding relationship between the continuous increase in construction progress, the continuous occurrence of resource input, the non-execution of disposal actions, and the decrease in the cumulative confirmed cost increment is used to generate cost suppression behavior. Based on the consistency of the project object, the correspondence of the construction procedures, and the overlap of the occurrence time, the segments of cost increase reduction, cost suppression behavior, and implicit cost debt items are associated to generate implicit cost debt suppression events; Specifically, the 10-day period of decreasing cost increment, the corresponding cost suppression behavior, and the implicit cost debt of RMB 61,200 are connected according to the same project object, the same construction process, and the overlapping time of occurrence. The start and end times of the cost increment decrease period, the changes in resource input and expenses corresponding to the cost suppression behavior, and the amount of the implicit cost debt are recorded. The data are combined into an implicit cost debt suppression event that lasts for 10 days and corresponds to a debt amount of RMB 61,200.

[0025] In this embodiment, the rebound prediction module includes: Starting from the time of occurrence of the hidden cost debt suppression event, the changes in resource input, construction status, disposal actions and cost changes of the corresponding project objects and construction procedures are obtained from the real-time engineering business data thereafter, and arranged according to the time of occurrence to generate a construction impact status sequence. Specifically, taking the occurrence time of the implicit cost debt suppression event as the starting point, the changes in resource input, construction status, disposal actions and cost changes corresponding to the same project object and construction process within the following 35 days are continuously obtained. Resource input record, construction status record, disposal action record and cost change record are generated daily, resulting in a total of 140 sets of time series data, which are then arranged in order of occurrence time to generate a construction impact status sequence. Based on the correspondence of project objects, the connection of construction procedures, and the continuity of occurrence time, the implicit cost debt items, implicit cost debt suppression events, and construction impact status sequences are connected to generate current debt suppression data; Specifically, the implicit construction cost debt of RMB 61,200 is taken as the starting node of the current debt suppression data. The implicit construction cost debt suppression event lasting 10 days is taken as the event node to record the segment of cost increase reduction, cost suppression behavior and debt amount. Then, it is connected to 140 sets of construction impact status records formed over 35 consecutive days from the time of the event. The event node only records the start and end time of the event and the related objects. The construction impact status sequence records the changes in resource input, construction status, disposal actions and cost changes on each date, thereby avoiding duplicate writing of real-time engineering business data during the event period and generating the current debt suppression data. The current debt suppression data is matched with the historical implicit cost debt suppression and rebound realization chain in the historical implicit cost debt suppression and rebound realization chain library. The construction impact status sequence is aligned with the historical debt evolution data in each historical implicit cost debt suppression and rebound realization chain to determine the construction progress. The occurrence order, duration and magnitude of resource input changes, construction status changes and cost changes are compared to determine the target historical implicit cost debt suppression and rebound realization chain and the target matching position. Specifically, the current debt suppression data is first matched with 4,286 historical implicit cost debt suppression and rebound realization chains according to the project object and construction procedure. Then, the difference between the implicit cost debt amount of 61,200 yuan and each historical implicit cost debt amount, as well as the difference between the 10-day suppression duration and the duration of each historical debt suppression segment, are calculated. The current construction impact state sequence and each historical debt evolution data are mapped to a unified construction progress coordinate. Dynamic time warping is used to calculate the cumulative difference of resource input changes, construction status changes, and cost changes. The difference of debt amount, suppression duration, and cumulative difference are normalized and added together. The resulting difference is distributed between 8.72 and 31.46. The historical chain with a difference of 8.72 is determined as the target historical implicit cost debt suppression and rebound realization chain. The 64% construction progress position corresponding to the end of the construction impact state sequence in the dynamic time warping path is determined as the target matching position. Obtain the chain segment following the target matching position in the target's historical implicit construction cost debt suppression and rebound realization chain. Adjust the occurrence time and amount of cost changes in the chain segment according to the duration ratio and change ratio between the construction impact state sequence and the chain segment. Connect the current debt suppression data to generate an implicit construction cost debt suppression and rebound prediction chain. Specifically, the chain segment following the 64% construction progress position in the target historical implicit construction cost debt suppression and rebound realization chain is obtained. This chain segment lasts for 28 days with a cumulative cost change of 320,000 yuan. The construction impact status sequence lasts for 35 days with a cost change of 164,000 yuan. The cost change of the corresponding historical chain matching segment is 200,000 yuan. 35 divided by 28 gives a duration ratio of 1.25, and 164,000 divided by 20 gives a change ratio of 0.82. The time interval of each cost change in the subsequent chain segment is multiplied by 1.25, and the cost change is multiplied by 0.82 to obtain an adjustment chain segment that lasts for 35 days with a cumulative cost change of 262,400 yuan. This adjustment chain segment is then connected to the current debt suppression data to generate an implicit construction cost debt suppression and rebound prediction chain. The cost changes in the implicit cost debt suppression and rebound prediction chain are respectively mapped to the cost changes caused by the release of implicit cost debt, the cost changes caused by the removal of cost suppression behavior, and the cost changes caused by the continued impact of construction, generating cost change decomposition data; Specifically, the cost changes resulting from the release of implicit construction cost debt are defined as the cost changes caused by the conversion of unconfirmed costs corresponding to implicit construction cost debt items into confirmed costs in the prediction chain. Starting from the starting position of the predicted cost change of 262,400 yuan, the cost increments are accumulated sequentially. The portion accumulated up to 61,200 yuan of implicit construction cost debt is corresponding to the cost changes resulting from the release of implicit construction cost debt. The 54,800 yuan that increases after the end of the cost suppression behavior is corresponding to the cost changes resulting from the lifting of the cost suppression behavior. The remaining 146,400 yuan is corresponding to the cost changes resulting from the continued impact of construction. The amounts, occurrence times, and construction procedures of the three types of cost changes are connected to generate cost change decomposition data.

[0026] In this embodiment, the reinforcement learning evaluation module includes: Based on the project object, construction procedure, and occurrence time, the implicit cost debt items, implicit cost debt suppression events, construction impact state sequence, implicit cost debt suppression and rebound prediction chain, and cost change decomposition data are matched to generate project cost status data. Specifically, the following steps are taken: 61,200 yuan of debt amount, 9 days of missing cost confirmation duration and associated occurrence time are obtained from the implicit cost debt items. The time interval between the occurrence time and the start time of the corresponding construction procedure is determined as the debt formation position. The 10-day debt suppression duration is obtained from the implicit cost debt suppression event. The order of occurrence of resource input changes, construction status changes, disposal actions and cost changes is obtained from the construction impact status sequence. Then, the cost change range of 262,400 yuan and the proportion of the three types of cost changes are obtained from the implicit cost debt suppression and rebound prediction chain and cost change decomposition data. All data are normalized and arranged in a fixed order to generate project cost status data. The project cost status data is matched with the historical implicit cost debt items and historical debt evolution data in the historical implicit cost debt suppression and rebound realization chain library. The matching historical implicit cost debt suppression and rebound realization chain is determined according to the debt formation location, debt suppression duration, construction impact change order and cost change magnitude, and the corresponding historical disposal feedback data is obtained. Specifically, the debt formation location, 10-day debt suppression duration, construction impact change order, and 262,400 yuan cost change range in the project cost status data are normalized and calculated with the corresponding data in each historical implicit cost debt item and historical debt evolution data. Then, the differences of each item are added together to form the historical chain matching distance. The matching distance of 4,286 historical chains is distributed between 0.084 and 1.826. The top 30 historical chains are obtained according to the matching distance from small to large, and 186 historical disposal feedback data are obtained from the historical chains. Identify disposal actions from historical disposal feedback data, and correlate each disposal action with historical implicit construction cost debts before and after execution, historical debt evolution data and cost changes to generate disposal action feedback data and historical status transition data; Specifically, the affected engineering objects, construction procedures, execution content, and execution time of the disposal actions were read from 186 historical disposal feedback data. Disposal actions with the same construction procedures and execution content were merged to form 6 types of disposal actions and generate disposal action sets. Then, the debt amount, cost confirmation duration, debt formation location, debt suppression duration, order of changes in construction impact, and cost change magnitude were obtained from the historical implicit cost debt items and historical debt evolution data before and after the execution of the disposal actions. The same normalization method and data arrangement order were used to generate the engineering cost status before execution and the engineering cost status after execution. The engineering cost status before execution, disposal actions, engineering cost status after execution, and cost changes were connected to generate 186 sets of historical status transition data. The dominant actor-commentator algorithm is adopted, which uses the engineering cost status data as the state and the disposal actions in the disposal action set as the actions. Based on the disposal action feedback data, the single cost return is generated, and combined with the status after the disposal action is executed in the historical state transition data and the single cost return of the corresponding disposal action, the long-term cost return of each disposal action is iteratively updated. Specifically, both the actor network and the commentator network employ a two-layer fully connected structure containing 64 and 32 neurons, respectively. The actor network outputs the selection probabilities of six types of actions, while the commentator network outputs the pre-execution state value and the post-execution state value. The reduction in cost rebound amount, the reduction in implicit cost debt, and the reduction in cost fluctuation are normalized, multiplied by 0.5, 0.3, and 0.2, respectively, and added together to generate a single cost return. The post-execution state value is multiplied by 0.9 and added to the single cost return to generate a long-term cost return. The long-term cost return is then subtracted from the pre-execution state value to generate an advantage value. 500 iterations are performed using 186 sets of historical state transition data. The advantage value is used to update the parameters of the actor network, and the difference between the long-term cost return and the pre-execution state value is used to update the parameters of the commentator network. Based on the long-term cost return of each disposal action, combined with the cost occurrence time and cost change decomposition data in the implicit cost debt suppression and rebound prediction chain, the execution order and execution time of the disposal actions are determined, an implicit cost debt release plan is generated, and the disposal actions, execution order and execution time in the implicit cost debt release plan are mapped to the implicit cost debt suppression and rebound prediction chain, the occurrence time and change amount of the corresponding cost changes are adjusted, and a dynamic evaluation result of the project cost is generated. Specifically, the current project cost status data is combined with the feedback data of the six types of disposal actions. Six sets of post-execution project cost statuses are generated based on the historical status changes before and after each disposal action. The commentator network outputs the status value of each of the six post-execution project cost statuses. The single cost return corresponding to each disposal action is then added to 0.9 times the post-execution status value to obtain the long-term cost returns of 0.82, 0.71, 0.58, 0.46, 0.31, and 0.19 for the six disposal actions under the current project cost status. The execution order of the disposal actions is determined according to the long-term cost returns from largest to smallest. The execution time is determined by combining the cost occurrence time in the implicit cost debt suppression and rebound prediction chain with the cost source in the cost change decomposition data, generating an implicit cost debt release plan. The cost change ratio corresponding to each disposal action is then mapped to the corresponding time period in the prediction chain, adjusting the predicted cost change from 262,400 yuan to 181,600 yuan, generating a dynamic project cost assessment result.

[0027] In this embodiment, the early warning module includes: Based on the project object, construction procedure and occurrence time, the implicit cost debt suppression and rebound prediction chain, cost change decomposition data, implicit cost debt release plan and project cost dynamic evaluation results are matched to generate cost rebound identification data; Specifically, based on the same project object and the same construction procedure, the implicit cost debt suppression and rebound prediction chain with a cumulative cost change of 262,400 yuan, the cost change decomposition data corresponding to the three types of cost changes, the implicit cost debt release plan corresponding to the three disposal actions and their execution time, and the project cost dynamic assessment results with a cumulative cost change of 181,600 yuan are mapped to a unified time axis. The original predicted cost changes, disposal action execution time, adjusted cost changes, and cost sources are matched according to the time axis position to generate cost rebound identification data. Based on the cost rebound identification data, the execution time in the implicit cost debt release plan is mapped to the dynamic evaluation result of the project cost, the chain segment of continuous cost increase after the execution time is identified, and the corresponding cost change is determined by combining the cost change decomposition data to generate cost rebound segments. Specifically, the execution times of the three disposal actions in the implicit cost debt release plan are sequentially mapped to the dynamic evaluation results of the project cost. The execution time of the earliest disposal action is used as the starting point for arranging cost changes. The execution times of the other two disposal actions are marked on a unified time axis. The cost changes after the starting point are differentially analyzed on adjacent dates according to the daily time interval. The continuously positive cost differences are connected according to the occurrence time to form a continuous cost increase chain segment, thereby avoiding the repeated calculation of cost changes on the same date. Then, the corresponding cost source is determined by combining the cost change decomposition data. The cost increase of 6 consecutive days is added together to obtain 96,000 yuan, generating a cost rebound segment. By comparing the cost rebound segment with the cost changes in the corresponding time period in the suppression of implicit cost debt and the rebound prediction chain, the cost increase that continues to exist after the implementation of the implicit cost debt release plan is identified as the cost rebound risk. Specifically, the predicted cost change of 140,000 yuan within 6 days, which is the same as the cost rebound segment, is obtained from the implicit cost debt suppression and rebound prediction chain. This is compared with the cost change of 96,000 yuan in the dynamic assessment results of the project cost. The residual rebound ratio of 68.57% is obtained by dividing 96,000 yuan by 140,000 yuan. The cost change of 96,000 yuan, the residual rebound ratio of 68.57%, and the 6-day duration are then connected to identify the 96,000 yuan cost increase that continues to exist after the implementation of the implicit cost debt release plan as the cost rebound risk. Based on the project objects, construction procedures, occurrence time and cost changes corresponding to the risk of cost rebound, generate project cost early warning results; Specifically, the corresponding project object, construction procedure, rebound start time, cost change of 96,000 yuan, and 6-day duration are obtained from the cost rebound risk. The cost rebound risks are arranged from largest to smallest according to the cost change. The project objects with a cost change of 96,000 yuan are arranged to the corresponding early warning sequence position. Then, the project object, construction procedure, rebound start time, cost change, duration, and residual rebound ratio are combined to generate the project cost early warning result.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a real-time cost assessment and early warning scenario at a large public building construction site. This construction site suffers from asynchronous issues in labor input, material usage, equipment operation, construction progress, and cost confirmation. Traditional cost management primarily relies on confirmed costs to judge cost changes. When construction continues while cost confirmation lags, the increase in book costs may temporarily decrease, easily leading to the misconception that costs are under control. However, unconfirmed labor, material, and equipment costs have already been incurred, and subsequent concentrated confirmation will result in a significant cost rebound.

[0029] The system accesses historical project datasets and real-time project data streams, cleaning, aligning, matching project object identifiers, and associating construction procedures for project completion records, material usage records, equipment operation records, construction status records, cost confirmation records, and action records. This generates historical and real-time project business data. The historical project business data contains 31,840 business records from 126 projects. From this data, the system identifies 7,620 historical construction events and obtains 6,954 sets of correspondences between historical construction events and cost confirmation records.

[0030] In a certain historical construction event, labor costs were 42,000 yuan, material costs were 106,000 yuan, equipment costs were 18,000 yuan, and other construction costs were 20,000 yuan, with total incurred costs amounting to 186,000 yuan. The cumulative confirmed costs during the same period were 112,000 yuan. Subtracting the two, we get 74,000 yuan of unconfirmed costs. The cost confirmation period is 10 days. The system connects the unconfirmed costs, project objects, construction procedures, occurrence time, and cost confirmation period to generate a historical implicit cost liability item.

[0031] The system continued to analyze resource input, construction status, and cost changes following this historical construction event. Over 14 consecutive days, the construction progress increased from 41.2% to 58.6%, and labor input increased from 1260 man-hours to 1778 man-hours. However, the daily increase in cumulative confirmed costs decreased from 28,000 yuan to 7,000 yuan. The system identified this period as a historical debt suppression segment. In the following 9 days, the cumulative confirmed costs increased from 213,000 yuan to 284,000 yuan. The system identified this period as a historical debt rebound segment and connected the historical debt suppression segment, historical debt rebound segment, and historical disposal feedback data to form 4286 chains of historical implicit construction cost debt suppression and rebound realization.

[0032] During real-time application, the system identified a construction event on the 18th construction day after the start of a certain construction process. This event corresponds to 320 cubic meters of completed work, 326 cubic meters of material usage, and 42 hours of equipment operation. However, no cost confirmation record was generated by the 27th construction day, representing a 9-day gap in cost confirmation. The system matched this construction event with the historical chain database, determining that the historical cost confirmation period for the target historical chain was 10 days, and the historical implicit cost debt amounted to 68,000 yuan. Mapping this to the 9-day and 10-day timeframes, the current implicit cost debt amount was calculated to be 61,200 yuan.

[0033] The system performed a differential analysis of real-time cumulative confirmed costs across adjacent dates. It was found that the construction progress increased from 46.8% to 59.2% over 10 consecutive days, with daily labor input remaining between 48 and 55 man-hours. However, the daily increase in cumulative confirmed costs decreased from 26,000 yuan to 8,000 yuan, and no action was taken during this period. This indicates that the decrease in cost increase did not represent a reduction in construction costs, but rather a lag in cost confirmation. The system correlated the segment of cost increase decrease, cost suppression behavior, and a hidden cost debt of 61,200 yuan, generating a hidden cost debt suppression event.

[0034] The system acquires information on resource input changes, construction status changes, handling actions, and cost changes within 35 days of the event, generates a construction impact status sequence, and matches it with a historical chain database. The chain segment following the matching position in the target historical chain lasts 28 days, with a cumulative cost change of 320,000 yuan. After adjustments based on the duration and cost change magnitude of the current construction impact status sequence, a 35-day implicit cost debt suppression and rebound prediction chain is generated, with a cumulative cost change of 262,400 yuan.

[0035] The system further breaks down the predicted cost change of 262,400 yuan into three parts: 61,200 yuan is the cost change caused by the conversion of previously unconfirmed costs into confirmed costs; 54,800 yuan is the cost change caused by the removal of cost suppression measures; and 146,400 yuan is the cost change caused by the continued impact of construction. This allows cost management personnel to clearly identify the specific sources of subsequent cost increases.

[0036] The system retrieves 186 historical disposal feedback data from the matched historical chain and merges the disposal actions into 6 categories. The dominant actor-commentator algorithm is used to calculate the long-term cost returns for the 6 categories of disposal actions, which are 0.82, 0.71, 0.58, 0.46, 0.31, and 0.19, respectively. Based on the long-term cost returns, the system determines the execution order and timing of the disposal actions, generates a hidden cost debt release plan, and adjusts the predicted cumulative cost change from 262,400 yuan to 181,600 yuan, a reduction of 80,800 yuan.

[0037] After the action was taken, the system continued to monitor cost changes and identified a cost rebound segment that lasted for six consecutive days, with a cost change of 96,000 yuan. The original predicted cost change for the same six days in the prediction chain was 140,000 yuan, and the residual rebound rate after the action was 68.57%. The system combined the project object, construction procedure, rebound start time, cost change, duration, and residual rebound rate to generate a project cost early warning result.

[0038] During a 90-day application period, the system identified 37 current construction events for which cost confirmation had not yet been completed. Of these, 29 were identified as hidden cost liabilities before a significant increase in book costs, representing an advance identification time of 5 to 12 days. After adopting the hidden cost liability release scheme, the predicted cumulative cost change for the relevant construction procedures was adjusted from RMB 1.864 million to RMB 1.479 million, and the duration of cost rebound was shortened from an average of 9.3 days to 6.1 days. This demonstrates that the present invention can identify hidden cost liabilities before formal cost confirmation, predict subsequent cost rebounds, and determine more appropriate actions, thereby improving the real-time performance, accuracy, and early warning foreseeability of project cost assessment.

[0039] Table 1. Comparison of the overall performance of different engineering cost assessment and early warning methods

[0040] The above comparative data is based on statistics of 37 missing cost recognition events formed within 90 consecutive days. The cumulative cost recognition threshold method mainly relies on changes in already formed book costs, and risks can usually only be identified after a significant increase in cumulative recognized costs. Therefore, only 11 events were identified in advance, with an advance identification rate of 29.73%. The earned value analysis method adds a comparison between construction progress and cost deviations, increasing the number of early identifications to 16. However, it still lacks the ability to directly identify situations where the construction impact has already occurred but the costs have not yet been recognized.

[0041] Time series forecasting predicts subsequent costs by analyzing historical cost trends, identifying up to 22 items in advance, with an early identification rate of 59.46%, which is an improvement over the previous two methods. However, this method mainly relies on the cost series itself. When resource input and construction progress continue to increase while the increase in book costs decreases, it is easy to extend the abnormal downward trend caused by the lag in cost recognition, thereby affecting the accuracy of predicting the amount and timing of subsequent rebounds.

[0042] This invention identified 29 hidden cost liabilities in advance, achieving an early identification rate of 78.38%. This is 48.65 percentage points, 35.14 percentage points, and 18.92 percentage points higher than the cumulative confirmed cost threshold method, earned value analysis method, and time series forecasting method, respectively. The average early warning time reached 8.4 days, which is 6.3 days, 5.0 days, and 2.7 days longer than the three comparison methods, respectively. This provides more time for cost verification, segmented confirmation, and review of construction records.

[0043] In terms of cost rebound prediction, the rebound amount prediction error of this invention is 8.70%, which is lower than 28.60% of the cumulative recognized cost threshold method, 21.40% of the earned value analysis method, and 14.80% of the time series prediction method. This is because this invention not only analyzes changes in book costs, but also incorporates the amount of implicit cost debt, the duration of missing cost recognition, the duration of debt suppression, the order of changes in construction impact, and the magnitude of cost changes into the project cost status, so that the prediction results can reflect the continuous process of cost formation, temporary suppression, and subsequent concentrated recognition.

[0044] In terms of the disposal effect, the cumulative cost change corresponding to this invention is RMB 1.479 million, which is RMB 385,000 lower than the cumulative confirmed cost threshold method, RMB 269,000 lower than the earned value analysis method, and RMB 134,000 lower than the time series forecasting method. The average duration of cost rebound is shortened to 6.1 days, and the cost change of rebound over 6 consecutive days is reduced to RMB 96,000. This indicates that the disposal plan generated by this invention can reduce the magnitude of concentrated cost release and shorten the duration of cost rebound.

[0045] The aforementioned performance improvement mainly stems from the continuous correlation between the historical implicit cost debt suppression and rebound realization chain library and the historical cost formation, debt suppression, rebound realization, and disposal feedback process. This invention predicts subsequent cost changes by matching the current construction impact state with the historical debt evolution process, and uses the dominant actor-commentator algorithm to evaluate the long-term cost return of different disposal actions, thereby determining the appropriate execution sequence and execution time for the current project cost state, and improving the accuracy of implicit cost debt identification, cost rebound prediction, and real-time early warning.

[0046] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time engineering cost assessment and early warning system based on big data and reinforcement learning, characterized in that, include: The data processing module is used to process historical engineering datasets and real-time engineering data streams, and generate historical engineering business data and real-time engineering business data. The historical chain construction module is used to generate historical implicit cost debt items, historical debt evolution data, and historical disposal feedback data based on historical engineering business data, and connect them to generate a historical implicit cost debt suppression and rebound realization chain library; The debt identification module is used to identify current construction events based on real-time engineering business data and match them with the historical implicit cost debt suppression and rebound realization chain library to generate implicit cost debt items; The suppression identification module is used to generate a book cost increment sequence based on real-time engineering business data, identify cost increment decline segments and cost suppression behaviors, and generate implicit cost debt suppression events by correlation. The rebound prediction module is used to generate a construction impact state sequence and match it with the historical implicit cost debt suppression and rebound realization chain library to generate implicit cost debt suppression and rebound prediction chain and cost change decomposition data. The reinforcement learning evaluation module is used to generate engineering cost status data based on implicit cost debt items, implicit cost debt suppression events, construction impact state sequences, implicit cost debt suppression and rebound prediction chains, and cost change decomposition data. It uses the dominant actor-commentator algorithm to generate implicit cost debt release plans and dynamic evaluation results of engineering costs. The early warning module is used to identify the risk of cost rebound by combining the dynamic evaluation results of engineering cost and generate early warning results for engineering cost.

2. The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to claim 1, characterized in that, The data processing module includes: Deduplication, missing data completion, and data format standardization are performed on historical engineering datasets and real-time engineering data streams respectively to generate historical cleaned data and real-time cleaned data. The historical cleaning data and real-time cleaning data are aligned by time, matched by engineering object identifiers, and associated by construction procedures to generate historical engineering business data and real-time engineering business data.

3. The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to claim 1, characterized in that, The historical chain construction module includes: Identify historical construction events from historical engineering business data, and correlate these historical construction events with subsequent cost confirmation records to generate historical event cost association data; The time interval between the occurrence of a historical construction event and the formation of a cost confirmation record is defined as the historical debt evolution interval. The historical cost confirmation duration is obtained, and the cost burdens that have formed a construction impact but have not yet been confirmed are identified, thus generating historical implicit cost debt items. By performing time correlation analysis on changes in resource input, construction status, and costs within the historical debt evolution segment, historical debt suppression and historical debt rebound segments are generated, and these segments are then connected to generate historical debt evolution data. Identify disposal actions within the historical debt evolution phases and correlate them with historical debt evolution data before and after execution to generate historical disposal feedback data; Based on the project object, construction sequence, and time relationship, historical hidden cost debt items, historical debt evolution data, and historical disposal feedback data are connected to generate a historical hidden cost debt suppression and rebound realization chain, and a historical hidden cost debt suppression and rebound realization chain library is generated.

4. The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to claim 1, characterized in that, The debt identification module includes: Identify the current construction event from real-time engineering business data, search for the cost confirmation record formed after the current construction event occurs according to the project object and construction procedure, identify the current construction event that has not yet formed a cost confirmation record as the current construction event to be matched, calculate the time interval between the current construction event occurrence time and the current time, and generate the cost confirmation missing duration; Based on the project object, construction procedure, and duration of missing cost confirmation, the current construction event to be matched is matched with the historical implicit cost debt suppression and rebound realization chain in the historical implicit cost debt suppression and rebound realization chain library, and candidate historical implicit cost debt suppression and rebound realization chains are screened. Align the current construction event to be matched with the historical construction events in the candidate historical implicit cost debt suppression and rebound realization chain according to the construction procedure and occurrence time, compare the project object, construction procedure and cost confirmation missing time, and determine the target historical implicit cost debt suppression and rebound realization chain; Map the historical implicit cost debt items in the target historical implicit cost debt suppression and rebound realization chain to the current construction event to be matched, and generate implicit cost debt items by combining the project object, construction procedure and cost confirmation missing time of the current construction event.

5. The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to claim 1, characterized in that, The suppression recognition module includes: Based on the project object and construction procedure, the cumulative confirmed costs in the real-time project business data are differentiated by adjacent time points, and the cumulative confirmed cost differences are arranged in chronological order to generate a book cost increment sequence. The sequence of incremental book costs is divided into continuous segments, and the continuous segments in which the cost increment decreases continuously are identified as the initial cost increment decreasing segments. The changes in resource input, construction status, disposal actions, and cost changes during the period corresponding to the initial cost increment decrease segment are time-aligned. Based on the changes in construction status, the construction progress increment is calculated. The initial cost increment decrease segment with a continuous increase in construction progress increment is identified as the cost increment decrease segment. Construction behaviors where resource input, disposal action execution, or cumulative confirmed costs have not increased synchronously are identified, and cost suppression behaviors are generated. Based on the consistency of the project object, the correspondence of the construction procedures, and the overlap of the occurrence time, the segments of cost increase reduction, cost suppression behaviors, and implicit cost debt items are linked to generate implicit cost debt suppression events.

6. The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to claim 1, characterized in that, The rebound prediction module includes: Starting from the time of occurrence of the hidden cost debt suppression event, the changes in resource input, construction status, disposal actions and cost changes of the corresponding project objects and construction procedures are obtained from the real-time engineering business data thereafter, and arranged according to the time of occurrence to generate a construction impact status sequence. Based on the correspondence of project objects, the connection of construction procedures, and the continuity of occurrence time, the implicit cost debt items, implicit cost debt suppression events, and construction impact status sequences are connected to generate current debt suppression data; The current debt suppression data is matched with the historical implicit cost debt suppression and rebound realization chain in the historical implicit cost debt suppression and rebound realization chain library. The construction impact status sequence is aligned with the historical debt evolution data in each historical implicit cost debt suppression and rebound realization chain to determine the construction progress. The occurrence order, duration and magnitude of resource input changes, construction status changes and cost changes are compared to determine the target historical implicit cost debt suppression and rebound realization chain and the target matching position. Obtain the chain segment following the target matching position in the target's historical implicit construction cost debt suppression and rebound realization chain. Adjust the occurrence time and amount of cost changes in the chain segment according to the duration ratio and change ratio between the construction impact state sequence and the chain segment. Connect the current debt suppression data to generate an implicit construction cost debt suppression and rebound prediction chain. By mapping the cost changes in the implicit cost debt suppression and rebound prediction chain to the cost changes resulting from the release of implicit cost debt, the cost changes resulting from the removal of cost suppression behavior, and the cost changes resulting from the continued impact of construction, cost change decomposition data is generated.

7. The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to claim 1, characterized in that, The reinforcement learning evaluation module includes: Based on the project object, construction procedure, and occurrence time, the implicit cost debt items, implicit cost debt suppression events, construction impact state sequence, implicit cost debt suppression and rebound prediction chain, and cost change decomposition data are matched to generate project cost status data. The project cost status data is matched with the historical implicit cost debt items and historical debt evolution data in the historical implicit cost debt suppression and rebound realization chain library. The matching historical implicit cost debt suppression and rebound realization chain is determined according to the debt formation location, debt suppression duration, construction impact change order and cost change magnitude, and the corresponding historical disposal feedback data is obtained. Identify disposal actions from historical disposal feedback data, and correlate each disposal action with historical implicit construction cost debts before and after execution, historical debt evolution data and cost changes to generate disposal action feedback data and historical status transition data; The dominant actor-commentator algorithm is adopted, which uses the engineering cost status data as the state and the disposal actions in the disposal action set as the actions. Based on the disposal action feedback data, the single cost return is generated, and combined with the status after the disposal action is executed in the historical state transition data and the single cost return of the corresponding disposal action, the long-term cost return of each disposal action is iteratively updated. Based on the long-term cost return of each disposal action, and combined with the cost occurrence time and cost change decomposition data in the implicit cost debt suppression and rebound prediction chain, the execution order and execution time of the disposal actions are determined, an implicit cost debt release plan is generated, and the disposal actions, execution order and execution time in the implicit cost debt release plan are mapped to the implicit cost debt suppression and rebound prediction chain. The occurrence time and change amount of the corresponding cost changes are adjusted to generate the dynamic evaluation result of the project cost.

8. The real-time engineering cost assessment and early warning system based on big data and reinforcement learning according to claim 1, characterized in that, The early warning module includes: By correlating the suppression of hidden construction cost debt with the rebound prediction chain, the decomposition data of cost changes, the release plan of hidden construction cost debt, and the dynamic evaluation results of engineering cost, data on the identification of cost rebound is generated. Identify the chain segment where costs continuously increase after the execution of disposal actions, and generate cost rebound segments by combining cost change decomposition data; By comparing the cost rebound segment with the predicted cost changes for the corresponding time period, the risk of cost rebound can be determined. Based on the project objects, construction procedures, occurrence time, and cost changes corresponding to the risk of cost rebound, project cost early warning results are generated.

Images