Artificial intelligence-based waste incineration power plant construction project management system

Abstract

The present application relates to the technical field of intelligent management of engineering projects, in particular to a garbage incineration power plant construction project management system based on artificial intelligence, comprising: a data acquisition module acquires drawing versions, construction progress logs, resource access records to form original project data, an intelligent analysis module analyzes the data in multiple dimensions, generates structured project status records and completes change intent recognition, progress mode extraction, supply risk deduction, a situation fusion module fuses multiple types of analysis results to generate a dynamic project comprehensive situation map, a decision scheduling module uses an improved Bayesian inference algorithm to output quantitative decision support information, generates an adaptive scheduling strategy for task order adjustment and resource reallocation, and converts it into executable instructions to be issued to the project execution terminal. The system realizes intelligent analysis and situation fusion of project data, and improves the automation and adaptability of project decision and scheduling.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for engineering projects, and in particular to an artificial intelligence-based management system for waste-to-energy plant construction projects. Background Technology

[0002] Waste-to-energy incineration plant construction projects often rely on manual methods to organize project data such as drawing versions, construction progress logs, and resource arrival records. Various types of information are stored in the form of scattered ledgers, and static project reports are formed by manual statistics and summarization. Design changes, construction progress, and material supply status are judged by human experience. Project decision-making and scheduling rely on manual formulation of plans and issuance of execution instructions.

[0003] The original project data lacks unified structure and integration. Design drawings, construction progress, and material supply information are independent and cannot form a sequential record. The underlying intentions of drawing changes, the operational patterns of construction progress, and potential risks in material supply cannot be identified and extracted through automated means. Project status can only achieve data stacking and display, failing to form a dynamic and integrated comprehensive situation. Conventional decision-making algorithms are unable to perform quantitative reasoning on multi-dimensional information. Project scheduling relies on manual adjustments to task arrangement and resource allocation, making it difficult to form adaptive optimization solutions. Scheduling strategies cannot be directly converted into standardized execution instructions.

[0004] Currently, there is a lack of structured sequence parsing for various types of raw project data, as well as targeted methods for intent recognition, pattern extraction, and risk extrapolation. There is also a lack of quantitative decision generation based on improved Bayesian inference, and an operational mechanism for adaptive scheduling of tasks and resources, and automatic issuance of instructions. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an artificial intelligence-based project management system for waste-to-energy incineration plants.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an artificial intelligence-based waste-to-energy plant construction project management system, comprising: The data acquisition module collects raw project data from the waste-to-energy plant construction project, including drawing versions, construction progress logs, and resource arrival records, forming a raw data set; The intelligent analysis module performs multi-dimensional analysis on the original data set to generate a structured project status record set including a design drawing change event sequence, a phased construction progress timeline, and a key material supply chain fluctuation sequence. It performs change intent identification on the design drawing change event sequence, progress pattern extraction on the phased construction progress timeline, and supply risk deduction on the key material supply chain fluctuation sequence. The situation fusion module inputs the change intent identification results, schedule pattern extraction results, and supply risk simulation results into the project situation fusion analyzer to generate a dynamic comprehensive project situation map. The decision scheduling module constructs a project decision support network based on the dynamic project comprehensive situation map. The project decision support network uses an improved Bayesian inference algorithm to output quantitative decision support information. Based on the quantitative decision support information, the project scheduling optimization engine generates an adaptive project scheduling strategy that includes task execution order adjustment schemes and resource reallocation schemes. The adaptive project scheduling strategy is translated into a set of executable project management instructions and sent to the corresponding project execution terminal.

[0007] As a further aspect of the present invention, the original project data collected from the waste-to-energy incineration plant construction project, including drawing versions, construction progress logs, and resource arrival records, forms an original data set, including: By deploying a data capture agent in the project design management system, version update events of design drawings and associated metadata are periodically captured as a drawing version data stream. By connecting to the project construction site's progress reporting system and monitoring video stream, structured construction log entries are obtained periodically, and the completion status of key process nodes in the video images is analyzed as a construction progress log data stream. By connecting the supplier management platform and the warehouse management system, the purchase order status, transportation trajectory and warehousing inspection records of key materials can be extracted in real time as a data stream for resource entry records; The drawing version data stream, construction progress log data stream, and resource arrival record data stream are aligned and integrated according to a unified time base, and stored in a temporary buffer database to form the original data set.

[0008] As a further aspect of the present invention, performing change intent identification on the design drawing change event sequence includes: Extract the drawing area code, change initiator identifier, and change description text associated with each change in the design drawing change event sequence; The drawing area codes are mapped to a preset factory functional area knowledge graph to locate the physical systems and process nodes affected by the changes. Natural language processing is performed on the change description text to identify the action verbs, technical parameters and constraints contained therein. Combined with the preset role weight of the change initiator's identifier, the potential impact intensity index of this change is calculated. By sequentially connecting the potential impact intensity index and the physical systems and process nodes affected by multiple changes, a change intent identification result reflecting the change propagation path and intensity evolution is generated.

[0009] As a further aspect of the present invention, the progress pattern extraction for the phased construction progress timeline includes: The construction progress timeline is divided into multiple construction phase segments based on preset key milestone nodes. The actual progress data within each construction phase segment is compared with the baseline plan data point by point to calculate the time series of progress deviation rate. Morphological analysis was performed on the time series of the schedule deviation rate to identify schedule deviation patterns with specific forms, including persistent lag, intermittent catch-up, and periodic fluctuations. The identified schedule deviation patterns are associated with the corresponding construction stages, weather records, and manpower input records to extract a set of potential correlation factors that lead to specific schedule deviation patterns, which are then used as the schedule pattern extraction results.

[0010] As a further aspect of the present invention, supply risk simulation is performed on the supply chain fluctuation sequence of the key materials, including: Analyze the supply chain fluctuation sequence of the key materials to obtain historical material order fulfillment rate, in-transit transportation time fluctuation data and supplier performance stability score; A supply chain topology network model is constructed with suppliers, logistics routes, and production batches as nodes, and the historical material order fulfillment rate, in-transit transportation time fluctuation data, and supplier performance stability score are used as the attributes and edge weights of the corresponding nodes. In the supply chain topology network model, material demand fluctuations and logistics disruption events are simulated, and the cascading impact of changes in the state of each node in the supply chain topology network model on the stability of material delivery in the entire network is calculated. Based on the simulation results, the identified supply chain bottlenecks, vulnerable logistics links, and their risk probabilities under the current fluctuations are output, forming the supply risk projection results.

[0011] As a further aspect of the present invention, the step of inputting the change intent identification result, schedule pattern extraction result, and supply risk simulation result into the project situation fusion analyzer to generate a dynamic project comprehensive situation map includes: Within the project situation fusion analyzer, feature projections are established for the change intent identification results, schedule pattern extraction results, and supply risk simulation results, respectively. Design an attention interaction mechanism across feature projections, and calculate the mutual attention weights among the feature projections of the change intention recognition result, the progress pattern extraction result, and the supply risk inference result. Based on the mutual attention weights, the feature projections of the change intent identification results, the progress pattern extraction results, and the supply risk simulation results are weighted, fused, and enhanced to generate a set of fused feature vectors with correlation. The fused feature vectors and their relationships are organized in a graph structure, where nodes represent the fused situational elements and edges represent the influence relationships and strengths between elements. The graph structure is updated in real time to form the dynamic project comprehensive situational map.

[0012] As a further aspect of the present invention, the improved Bayesian inference algorithm includes: Establish a dynamic Bayesian network structure with the comprehensive situational characteristics of the project as the observation variable; Initialize the prior probability distribution of each hidden node in the dynamic Bayesian network structure; During the reasoning process, the real-time feature data in the dynamic project comprehensive situation map is injected into the dynamic Bayesian network structure as new evidence. The real-time feature data includes design change frequency, schedule lag days and material supply delay warning level. Based on the conditional dependencies between nodes defined in the dynamic Bayesian network structure, the posterior probability distribution in the dynamic Bayesian network structure is iteratively approximated using the variational inference method. In each iteration of the approximation calculation, the state confidence of the hidden nodes associated with the key decision points is updated synchronously; When the change in the state confidence is less than the preset convergence threshold, the iteration stops, and the posterior probability distribution of the hidden nodes associated with the key decision point at this time is used as a measure of quantitative risk and opportunity to form the quantitative decision support information.

[0013] As a further aspect of the present invention, based on the quantitative decision support information, an adaptive project scheduling strategy including a task execution order adjustment scheme and a resource reallocation scheme is generated through a project scheduling optimization engine, including: In the project scheduling optimization engine, the project's task network, resource pool status, and the quantitative decision support information are taken as inputs; A multi-objective optimization model is established with the goal of minimizing the total project duration delay and resource conflicts; In the multi-objective optimization model, the quantitative decision support information is quantified into a time margin adjustment coefficient and a resource demand elasticity coefficient for a specific task. A heuristic search algorithm is used to solve the multi-objective optimization model. Under the premise of satisfying the task process logic constraints and total resource constraints, the algorithm explores the reordering space of task start time and the flow path of resources between different tasks. The new start and end times of all tasks are extracted from the solution results to form the task execution order adjustment scheme. At the same time, the changes in the allocation plan of various resources on the project timeline are extracted to form the resource reallocation scheme. Together, they constitute the adaptive project scheduling strategy.

[0014] As a further aspect of the present invention, the step of using a heuristic search algorithm to solve the multi-objective optimization model includes: Initialize a task scheduling sequence based on the current project baseline plan as the initial solution population; Define a fitness function, which is calculated by combining the total project duration, resource load balancing degree, and the buffering degree of high-risk tasks in the quantitative decision support information; For individuals in the initial solution population, new candidate solutions are generated by swapping task positions, inserting idle time, and adjusting resource allocation ratios. Calculate the fitness of the newly generated candidate solutions, and perform iterative population updates based on the fitness scores. During the population iteration process, a simulated annealing mechanism is introduced to accept solutions with poor fitness with a certain probability, thereby avoiding the search from getting trapped in local optima. When the preset maximum number of iterations is reached or the fitness improvement is continuously below the threshold, the search is terminated, and the individual with the highest fitness in the current population is taken as the final solution of the multi-objective optimization model.

[0015] As a further aspect of the present invention, the adaptive project scheduling strategy is translated into a set of executable project management instructions, including: The task execution order adjustment scheme in the adaptive project scheduling strategy is analyzed and decomposed into task start time change notification and task prerequisite relationship update notification for specific construction teams or subcontractors. The resource reallocation scheme in the adaptive project scheduling strategy is analyzed and decomposed into material delivery plan change notifications for specific warehouses or suppliers and manpower allocation instructions for different types of work. Based on the standard application programming interface protocol of project management software, the task start time change notification, task prerequisite relationship update notification, material distribution plan change notification, and manpower allocation instruction are formatted into standardized data packets that can be directly read and executed by the project execution terminal. All standardized data packets are sorted and packaged according to their effective time to form a complete set of project management instructions.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The raw data set consisting of drawing versions, construction progress logs, and resource arrival records is analyzed from multiple dimensions to form three types of structured project status records: design drawing change event sequences, phased construction progress timelines, and key material supply chain fluctuation sequences. Change intent identification is performed for change event sequences, progress pattern extraction is performed for construction progress timelines, and supply risk simulation is performed for supply chain fluctuation sequences. Scattered unstructured data is transformed into time-coherent structured information. The internal logic of design changes, the changing characteristics of construction progress, and the fluctuation trends of material supply are fully extracted. The status of each dimension of the project is continuously traceable, and the correlation between different types of project data is clearly presented.

[0017] By integrating multi-dimensional analysis results to form a dynamic comprehensive project situation map, an improved Bayesian inference algorithm is used to construct a project decision support network and output quantitative decision support information. Through the project scheduling optimization engine, task execution order adjustment schemes and resource reallocation schemes are generated to form an adaptive project scheduling strategy. The strategy is transformed into standardized executable instructions and issued to the corresponding project execution terminals. Multi-dimensional situation information forms an objective decision reference through quantitative inference. Task arrangement and resource allocation are aligned with the real-time project situation. The scheduling scheme is no longer dominated by human experience. The matching degree between instruction form and execution terminal is improved. Project scheduling realizes the transformation from manual passive adjustment to automated adaptive optimization. Attached Figure Description

[0018] Figure 1 This is a sequence diagram of the AI-based waste-to-energy plant construction project management system described in this invention. Figure 2 A flowchart illustrating the process of identifying change intent; Figure 3 A flowchart illustrating the process of extracting data in a progress mode. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0021] See Figure 1 This invention provides an artificial intelligence-based project management system for waste-to-energy incineration plants, specifically including: The data acquisition module collects raw project data from the project, including drawing versions, construction progress logs, and resource arrival records, forming a raw data set. The intelligent analysis module performs multi-dimensional analysis on this set, generating a structured project status record set. It also identifies change intent in the sequence of design drawing change events, extracts progress patterns from the phased construction progress timeline, and performs supply risk simulation on the supply chain fluctuation sequence of key materials. The situation fusion module inputs the above identification, extraction, and simulation results into the project situation fusion analyzer to generate a dynamic comprehensive project situation map. Based on this situation map, the decision scheduling module uses an improved Bayesian inference algorithm of the project decision support network to output quantitative decision support information. It then generates an adaptive project scheduling strategy through the project scheduling optimization engine and translates it into a set of executable project management instructions, which are then sent to the project execution terminal.

[0022] In one embodiment of the present invention, a data capture agent deployed in the design management system periodically captures version update events and associated metadata of design drawings, forming a drawing version data stream; by connecting to the progress reporting system and monitoring video stream, structured construction log entries are periodically obtained and the completion status of key process nodes in the video images is analyzed, forming a construction progress log data stream; by connecting to the supplier management platform and warehouse management system, the purchase order status, transportation trajectory, and warehousing inspection records of key materials are extracted in real time, forming a resource entry record data stream; the three data streams are aligned and integrated according to a unified time base and stored in a temporary buffer database to form the original data set. (See also...) Figure 2The process involves extracting the drawing area code, change initiator identifier, and change description text for each change in the design drawing change event sequence; mapping the drawing area code to a pre-defined factory functional area knowledge graph to locate the physical systems and process nodes affected by the change; performing natural language processing on the change description text to identify action verbs, technical parameters, and constraints; and calculating the potential impact intensity index of this change based on the pre-defined role weight of the change initiator identifier; and then sequentially linking the potential impact intensity indices of multiple changes with their affected physical systems and process nodes to generate a change intent identification result that reflects the change propagation path and intensity evolution.

[0023] In practical implementation, the data capture agent deployed in the project design management system polls the drawing library every 24 hours to capture design drawing version update events and associated metadata, including version number, modification time, and the employee number of the person making the modification, forming a drawing version data stream; the on-site progress reporting system uploads the daily construction log entries at 18:00 every day, and the monitoring camera captures the process images of the boiler installation area every 30 minutes. The image recognition unit analyzes whether the key process nodes in the images are in the welding completed state or hoisted in place state, forming a construction progress log data stream; the supplier management platform pushes the purchase order status updates of key materials in real time, and the warehouse management system records the material arrival time and appearance inspection results, forming a resource entry record data stream; the three types of data streams are aligned with a unified timestamp and stored in a temporary buffer database to form the original data set. In practical implementation, the intelligent analysis module extracts the boiler area code "BOILER_A_SECTION_01", the change initiator identifier "DESIGN_LEAD_03", and the change description text "increase the number of flue gas pipe supports to 12 sets" corresponding to a single change in the design drawing change event sequence. The boiler area code is mapped to the factory functional area knowledge graph to locate that the change affects the pipe support structure of the flue gas outlet section of the boiler body. The change description text is segmented and subjected to dependency parsing analysis to identify the action verb "increase", the technical parameter "12 sets", and the constraint condition "must be completed before the anti-corrosion layer construction". Combined with the design manager's weight value of 0.85 corresponding to the change initiator identifier, the potential impact intensity index of this change is calculated to be 0.72. The potential impact intensity indices of three changes within five days (0.72, 0.58, and 0.81) and their affected flue gas purification system nodes are connected in chronological order to generate a change intent identification result reflecting the change propagation path and intensity evolution.

[0024] In some embodiments, the data capture agent is configured in a reactive trigger mode, immediately capturing change records upon detecting a drawing version submission action, without waiting for a fixed period; the monitoring camera automatically triggers high-definition recording after key process nodes are completed, and the image recognition unit identifies the grate robotic arm positioning markers, supplementing the real-time details of the construction progress log data stream. It is understood that the natural language processing of the change description text also includes professional terminology normalization, such as uniformly mapping "supports and hangers" and "support structures" to the standard term "pipeline support"; the formula for calculating the potential impact intensity index is:

[0025] in: Indicates the potential impact intensity index. It changes the role weight corresponding to the initiator's identifier. It is the intensity score of action verbs extracted from the change description text. It is the coefficient for the variation range of technical parameters. and These are the standardized mapping functions of verb intensity and parameter magnitude, respectively.

[0026] Understandably, the data alignment process uses the international standard time UTC format to store timestamps, avoiding data misalignment caused by differences in on-site time zones. The temporary buffer database is equipped with a rolling cleanup strategy, retaining the original data set for the most recent thirty days, and archiving records exceeding the time limit to a long-term repository. Optionally, the calculation results of the potential impact intensity index of the change event sequence are stored in the form of a time series array, where each element contains the change time, impact index value, and the ID of the affected functional area node. Optionally, the plant functional area knowledge graph pre-sets the typical subsystem topology of the waste incineration power plant, including four major sections: incineration line, waste heat boiler, flue gas purification, and turbine island. After the changed area code matches the node path in the graph, it can be directly associated with upstream and downstream process nodes.

[0027] In one embodiment of the present invention, see [reference] Figure 3The project divides the construction progress timeline into multiple construction phase segments based on preset key milestone nodes. The actual progress data within each segment is compared point-by-point with the baseline plan data to calculate the progress deviation rate time series. Morphological analysis is performed on this series to identify progress deviation patterns such as continuous lag, intermittent catching up, and periodic fluctuations. The identified patterns are then linked to corresponding construction phases, weather records, and manpower input records to extract a set of potential correlation factors leading to specific progress deviation patterns, which serve as the progress pattern extraction results. The project also analyzes the supply chain fluctuation sequence of key materials to obtain historical material order fulfillment rates, in-transit transportation time fluctuation data, and supplier performance stability scores. A supply chain topology network model is constructed with suppliers, logistics routes, and production batches as nodes, using the above data as attributes and edge weights for the corresponding nodes. Material demand fluctuations and logistics disruption events are simulated in this model to calculate the cascading impact of node state changes on the stability of network material delivery. Based on the simulation results, the project outputs supply chain bottleneck nodes, vulnerable logistics links, and their risk probabilities under current fluctuations, forming a supply risk projection result.

[0028] In practice, the phased construction progress timeline is divided into weeks, with key milestones including the main chimney capping, boiler hydrostatic test, and turbine cylinder installation. Within the main chimney capping phase, the actual progress data is the daily completed pouring height, while the baseline plan data is the planned pouring height. After comparing each point, the daily progress deviation rate is calculated, forming a progress deviation rate time series. Morphological analysis of the progress deviation rate time series identifies a continuous lag pattern where the actual value is lower than the planned value for four consecutive days, and an intermittent catch-up pattern where the actual value exceeds the planned value by 20% for the following two days. The continuous lag pattern is associated with the corresponding stage's rainstorm weather records and concrete mixing plant failure records, extracting weather factors and equipment failure factors as a set of potential related factors leading to lag, which serve as the progress pattern extraction result. In practical implementation, the key material supply chain fluctuation sequence includes a historical order fulfillment rate of 82% for refractory bricks, a transit time fluctuation range of ±3 days, and a fulfillment stability score of 76 for supplier A. A supply chain topology network model is constructed, with nodes including supplier A, road transport route L1, and the 2025-06 production batch. The historical order fulfillment rate is assigned to the supplier A node, and the transit time fluctuation data is assigned to the edge weights of road transport route L1. The supplier fulfillment stability score is used as the reliability attribute of supplier A node. The network model simulates a 50% surge in refractory brick demand, calculating the cascading impact of order backlog caused by supplier A's insufficient capacity on the transit time of road transport route L1, resulting in a 35% increase in the delay probability of route L1. A 48-hour interruption event is simulated for road transport route L1, calculating that the inventory depletion time of the production batch node is 5 days earlier. The results show that supplier A is the bottleneck node in the supply chain, road transport route L1 is a vulnerable logistics link, and their risk probabilities under the current fluctuation are 28% and 42%, respectively, forming the supply risk projection results.

[0029] In some embodiments, the morphological analysis of the schedule deviation rate time series employs the sliding window differencing method with a window length of 7 days to identify a persistent lag pattern where the deviation rate exceeds -10% for three consecutive days. The simulation process for the key material supply chain fluctuation sequence is set to run ten times repeatedly, and the average state change value of the network nodes is taken as the cascading effect result. The schedule deviation rate calculation formula can be understood as follows:

[0030] in: This represents the schedule deviation rate at time t. This represents the actual progress value at time t. This represents the baseline planned progress value at time t.

[0031] It is understandable that the node attributes of the supply chain topology network model also include the upper limit of the warehouse node capacity, and the edge weight includes the transportation cost coefficient; when simulating interruption events, the availability of alternative logistics routes and the switching delay time are given priority. Optionally, the association binding process of the progress pattern extraction excludes shift data that has not changed in the manpower input records, and only retains the records that overlap with the deviation pattern time; optionally, the output format of the supply risk simulation results is a list of triples, each triple containing the bottleneck node name, vulnerable link identifier, and risk probability value.

[0032] In one embodiment of the present invention, feature projections are established within the project situation fusion analyzer for the change intent identification result, the schedule pattern extraction result, and the supply risk projection result, respectively; an attention interaction mechanism across feature projections is designed to calculate the mutual attention weights of the three among the feature projections; based on these weights, the feature projections of the three are weighted and fused and information is enhanced to generate a set of fused feature vectors with correlations; the vectors and their correlations are organized in a graph structure, where nodes represent the fused situation elements, edges represent the influence relationships and strengths between elements, and the graph structure is updated in real time to form a dynamic comprehensive project situation map. A dynamic Bayesian network structure is established with the comprehensive project situation characteristics as the observation variable; the prior probability distribution of each hidden node is initialized; real-time feature data from the dynamic comprehensive project situation map is injected into the network as new evidence, including the frequency of design changes, the number of days of schedule delay, and the warning level of material supply delay; the posterior probability distribution is iteratively approximated by variational inference based on the conditional dependencies between nodes; the state confidence of the hidden nodes associated with the key decision points is updated synchronously in each iteration; the iteration stops when the change in state confidence is less than the preset convergence threshold, and the posterior probability distribution of the hidden nodes associated with the key decision points at this time is used as a measure of quantitative risk and opportunity to form quantitative decision support information.

[0033] In practical implementation, the project situation fusion analyzer establishes a 128-dimensional feature projection for the change intent identification result, a 64-dimensional feature projection for the schedule pattern extraction result, and a 96-dimensional feature projection for the supply risk simulation result. The attention interaction mechanism across feature projections calculates the attention weight of the change intent identification result feature projection to the schedule pattern extraction result feature projection as 0.62, the attention weight to the supply risk simulation result feature projection as 0.51, and the attention weight to the supply risk simulation result feature projection as 0.78. Based on the mutual attention weights, the three types of feature projections are weighted and fused to generate a set of fused feature vectors containing 32 elements. Each element in the vector corresponds to a fused situation element, and the correlation coefficient between elements constitutes the influence strength between elements. The fused feature vectors are organized in a graph structure, with nodes labeled as "design change diffusion," "schedule lag accumulation," and "material supply disruption risk," and edges labeled as "change exacerbates lag" (weight 0.68) and "lag amplifies risk" (weight 0.73). The graph structure is updated every 15 minutes to form a dynamic comprehensive project situation map. In practical implementation, the observed variables of the dynamic Bayesian network structure include design change frequency (times / week), schedule lag days (days), and material supply delay warning level (levels 1-5); the hidden nodes include "design stability", "construction rhythm", and "supply chain health"; the prior probability distribution of the design stability node at initialization is {normal: 0.75, fluctuation: 0.25}; real-time feature data extracted from the dynamic project comprehensive situation map (design change frequency 4 times / week, schedule lag days 3 days, material supply delay warning level 3) are injected into the network as new evidence; based on the conditional dependencies between nodes, the posterior probability distribution is iteratively calculated using variational inference methods. After the third iteration, the state confidence of the design stability node changes from 0.69 to 0.71, and after the fifth iteration, the change is less than 0.02; after stopping the iteration, the posterior probability distribution of the hidden nodes associated with the key decision point "total schedule risk" is {low: 0.22, medium: 0.53, high: 0.25}, which is output as quantitative decision support information.

[0034] In some embodiments, the attention weights across feature projections are calculated using a scaled dot product, with the key vector dimension set to 16 and the query vector dimension set to 16; the variational inference process of the dynamic Bayesian network structure has a learning rate of 0.05 and a convergence threshold of 0.015. The weighted fusion calculation formula can be understood as follows:

[0035] in: Represents the fused feature vector. It is the first The normalized attention weights corresponding to the class results It is the linear transformation matrix corresponding to the feature projection. It is the first For the feature projection of the class results, please refer to Table 1.

[0036] Table 1: Update Record of Hidden Node State Confidence of Dynamic Bayesian Network ; It is understandable that the edge weight update rule of the dynamic project comprehensive situation map adopts an exponential decay smoothing strategy, where the new weight is equal to 0.9 times the original weight plus 0.1 times the newly calculated weight; in the variational inference process, the conditional probability table of hidden nodes predefines that for every increase of 1 design change per week, the probability of the "fluctuation" state of the design stability node increases by 0.08. Optionally, the correlation of the fused feature vectors is stored in an adjacency matrix, where the matrix row and column indices correspond to the situation element numbers, and the matrix element values ​​are the influence strength; optionally, the output format of the quantitative decision support information is a JSON object, containing the risk level probability distribution and confidence update timestamp for each key decision point.

[0037] In one embodiment of the present invention, the project task network, resource pool status, and quantitative decision support information are taken as inputs to the project scheduling optimization engine; a multi-objective optimization model is established with the goal of minimizing the total project duration delay and resource conflicts; in this model, the quantitative decision support information is quantified into the time margin adjustment coefficient and resource demand elasticity coefficient for specific tasks; a heuristic search algorithm is used to solve the model, exploring the rescheduling space of task start times and the flow path of resources among different tasks under the premise of satisfying the task process logic constraints and total resource constraints; the new plan start and end times of all tasks are extracted from the solution results to form a task execution order adjustment scheme, and the allocation plan changes of various resources on the project timeline are extracted to form a resource reallocation scheme, the two constituting an adaptive project scheduling strategy. Initialize the task scheduling sequence based on the current baseline plan as the initial solution population; define a fitness function that combines the total project duration, resource load balancing, and the buffering degree of high-risk tasks in the quantitative decision support information; generate new candidate solutions by swapping task positions, inserting idle time, and adjusting resource allocation ratios for individuals in the population; calculate the fitness of candidate solutions and iteratively update the population according to their quality; introduce a simulated annealing mechanism to accept solutions with poor fitness with a certain probability to avoid local optima; terminate the search when the maximum number of iterations is reached or the fitness improvement is continuously lower than the threshold, and take the individual with the highest fitness in the current population as the final solution.

[0038] In practical implementation, the project task network received by the project scheduling optimization engine contains 32 key tasks. The current resource pool shows a total of 60 steelworkers and 3 tower cranes. The quantitative decision support information includes a high-risk marker (probability 0.52) for the "T12 flue installation" task. The objective function of the multi-objective optimization model is set to minimize the total project duration delay and the difference in daily peak load of steelworkers. The quantitative decision support information is quantified as a time margin adjustment coefficient of +3 days and a resource demand elasticity coefficient of 1.3 for the "T12 flue installation" task. A genetic algorithm is used as the heuristic search algorithm. The initial solution population contains 20 chromosomes, each representing a task scheduling sequence. The algorithm is designed to meet the requirements of "incinerator masonry..." Under the constraints of the process logic that "the furnace steel structure must be completed" and the total resource constraint that the number of steelworkers should not exceed 60, the task start time rearrangement space was explored through cross-mutation operations. It was found that by postponing the "T12 flue installation" by 2 days, the "T15 induced draft fan installation" could be started 1 day earlier. Finally, the solution extracted that the start time of the "T12 flue installation" in the new plan was adjusted from day 45 to day 47, and the start time of the "T15 induced draft fan installation" was adjusted from day 49 to day 48, forming a task execution order adjustment scheme. At the same time, it was extracted that the number of steelworkers allocated from day 46 to day 48 was reduced from the original plan of 25 to 18, forming a resource reallocation scheme. The two constitute an adaptive project scheduling strategy.

[0039] In some embodiments, the initial solution population is generated using a serial scheduling generation scheme, and the chromosomes are encoded as a sequence of task numbers; the calculation scope of resource load balancing is limited to the daily utilization variance of critical tasks and large equipment. It can be understood that the fitness function is calculated as follows:

[0040] in: Indicates fitness. This indicates the total number of days the project was delayed. This indicates the degree of resource load balancing (standard deviation). This represents the total number of buffer days for high-risk tasks. , , These are the weighting coefficients for the corresponding items, see Table 2.

[0041] Table 2: Adjustment of Task Time and Resources for Adaptive Project Scheduling Strategy ; It is understandable that the crossover operation in the heuristic search process uses a two-point crossover method, and the mutation operation randomly swaps the positions of adjacent tasks; the initial temperature of the simulated annealing mechanism is 100, the cooling coefficient is set to 0.95, and solutions with a fitness decrease of no more than 5 are allowed. Optionally, the task execution order adjustment scheme is stored in Gantt chart data format, including the unique ID of each task, the start and end times of the old and new tasks, and the predecessor relationship change marker; optionally, the output fields of the resource reallocation scheme include resource type, date range, original allocation quantity, and new allocation quantity, used to generate a resource histogram comparison view.

[0042] In one embodiment of the present invention, the task execution order adjustment scheme in the adaptive project scheduling strategy is analyzed and decomposed into task start time change notifications and task prerequisite relationship update notifications for specific construction teams or subcontractors; the resource reallocation scheme is analyzed and decomposed into material distribution plan change notifications for specific warehouses or suppliers and manpower allocation instructions for different types of work; the above notifications and instructions are formatted into standardized data packets that can be directly read and executed by the project execution terminal according to the standard application programming interface protocol of the project management software; all standardized data packets are sorted and packaged according to their effective time to form a complete set of management instructions.

[0043] In practical implementation, the task execution order adjustment scheme in the adaptive project scheduling strategy is analyzed. This scheme includes changing the original planned start time of task "Steel Frame Hoisting G07" to August 12, 2025, and the new planned start time to August 14, 2025. It also changes task "Flue Prefabrication Y05" from its original immediate predecessor task "Steel Frame Hoisting G07" to having no immediate predecessor dependency. This adjustment is further broken down into a notification regarding the task start time change for the third construction team responsible for steel frame hoisting, containing the task number G07, the original start time, the new start time, and the effective date of August 10, 2025. Finally, a notification regarding the updated task prerequisites for the second subcontractor responsible for flue prefabrication is also provided, containing the task number Y05. Remove the dependency of the immediate predecessor task G07; analyze the resource reallocation scheme in the adaptive project scheduling strategy, which includes reducing the average daily cement delivery volume from the original plan of 120 tons to 100 tons and adjusting the number of welders from the original allocation of 25 to 18; decompose the above adjustments into a material delivery plan change notification for the central warehouse, which includes the material code CEM425, the delivery date range of August 12, 2025 to August 18, 2025, and the new delivery volume of 100 tons / day, as well as a welder manpower allocation instruction for the human resources department, which includes the job code WELDER, the allocation date of August 13, 2025, the original number of workers 25, the new number of workers 18, and the allocation direction from the main plant area to the boiler area.

[0044] In some embodiments, the task prerequisite update notification also includes scenarios involving new dependencies, such as the task "boiler hydrostatic test" adding a preceding task "pipeline insulation completed"; the material distribution plan change notification supports a multi-warehouse collaboration mode, splitting distribution instructions to backup warehouses when the central warehouse's inventory is insufficient. It can be understood that the formatting formula for the standardized data packet is:

[0045] in: Represents standardized data packets. It is a packet header containing the instruction type, version number, and target terminal ID. It is a JSON serialized instruction content entity. It is a 32-bit cyclic redundancy check code. This indicates a byte concatenation operation.

[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. An AI-based project management system for waste-to-energy incineration plants, characterized in that: The system includes: The data acquisition module collects raw project data from the waste-to-energy plant construction project, including drawing versions, construction progress logs, and resource arrival records, forming a raw data set; The intelligent analysis module performs multi-dimensional analysis on the original data set to generate a structured project status record set including a design drawing change event sequence, a phased construction progress timeline, and a key material supply chain fluctuation sequence. It performs change intent identification on the design drawing change event sequence, progress pattern extraction on the phased construction progress timeline, and supply risk deduction on the key material supply chain fluctuation sequence. The situation fusion module inputs the change intent identification results, schedule pattern extraction results, and supply risk simulation results into the project situation fusion analyzer to generate a dynamic comprehensive project situation map. The decision scheduling module constructs a project decision support network based on the dynamic project comprehensive situation map. The project decision support network uses an improved Bayesian inference algorithm to output quantitative decision support information. Based on the quantitative decision support information, the project scheduling optimization engine generates an adaptive project scheduling strategy that includes task execution order adjustment schemes and resource reallocation schemes. The adaptive project scheduling strategy is translated into a set of executable project management instructions and sent to the corresponding project execution terminal.

2. The artificial intelligence-based waste-to-energy plant project management system according to claim 1, characterized in that, The original project data collected from the waste-to-energy plant construction project, including drawing versions, construction progress logs, and resource arrival records, forms an original data set, including: By deploying a data capture agent in the project design management system, version update events of design drawings and associated metadata are periodically captured as a drawing version data stream. By connecting to the project construction site's progress reporting system and monitoring video stream, structured construction log entries are obtained periodically, and the completion status of key process nodes in the video images is analyzed as a construction progress log data stream. By connecting the supplier management platform and the warehouse management system, the purchase order status, transportation trajectory and warehousing inspection records of key materials can be extracted in real time as a data stream for resource entry records; The drawing version data stream, construction progress log data stream, and resource arrival record data stream are aligned and integrated according to a unified time base, and stored in a temporary buffer database to form the original data set.

3. The artificial intelligence-based waste-to-energy plant project management system according to claim 1, characterized in that, Perform change intent identification on the sequence of design drawing change events, including: Extract the drawing area code, change initiator identifier, and change description text associated with each change in the design drawing change event sequence; The drawing area codes are mapped to a preset factory functional area knowledge graph to locate the physical systems and process nodes affected by the changes. Natural language processing is performed on the change description text to identify the action verbs, technical parameters and constraints contained therein. Combined with the preset role weight of the change initiator's identifier, the potential impact intensity index of this change is calculated. By sequentially connecting the potential impact intensity index and the physical systems and process nodes affected by multiple changes, a change intent identification result reflecting the change propagation path and intensity evolution is generated.

4. The artificial intelligence-based waste-to-energy plant project management system according to claim 1, characterized in that, The progress pattern extraction for the aforementioned phased construction progress timeline includes: The construction progress timeline is divided into multiple construction phase segments based on preset key milestone nodes. The actual progress data within each construction phase segment is compared point by point with the baseline plan data to calculate the time series of progress deviation rate. Morphological analysis was performed on the time series of the schedule deviation rate to identify schedule deviation patterns with specific forms, including persistent lag, intermittent catch-up, and periodic fluctuations. The identified schedule deviation patterns are associated with the corresponding construction stages, weather records, and manpower input records to extract a set of potential correlation factors that lead to specific schedule deviation patterns, which are then used as the schedule pattern extraction results.

5. The artificial intelligence-based waste-to-energy plant project management system according to claim 1, characterized in that, Perform supply risk simulation on the supply chain fluctuation sequence of the aforementioned key materials, including: Analyze the supply chain fluctuation sequence of the key materials to obtain historical material order fulfillment rate, in-transit transportation time fluctuation data and supplier performance stability score; A supply chain topology network model is constructed with suppliers, logistics routes, and production batches as nodes, and the historical material order fulfillment rate, in-transit transportation time fluctuation data, and supplier performance stability score are used as the attributes and edge weights of the corresponding nodes. In the supply chain topology network model, material demand fluctuations and logistics disruption events are simulated, and the cascading impact of changes in the state of each node in the supply chain topology network model on the stability of material delivery in the entire network is calculated. Based on the simulation results, the identified supply chain bottlenecks, vulnerable logistics links, and their risk probabilities under the current fluctuations are output, forming the supply risk projection results.

6. The artificial intelligence-based waste-to-energy plant construction project management system according to claim 1, characterized in that, The process of inputting the change intent identification results, schedule pattern extraction results, and supply risk simulation results into the project situation fusion analyzer to generate a dynamic comprehensive project situation map includes: Within the project situation fusion analyzer, feature projections are established for the change intent identification results, schedule pattern extraction results, and supply risk simulation results, respectively. Design an attention interaction mechanism across feature projections, and calculate the mutual attention weights among the feature projections of the change intention recognition result, the progress pattern extraction result, and the supply risk inference result. Based on the mutual attention weights, the feature projections of the change intent identification results, the progress pattern extraction results, and the supply risk simulation results are weighted, fused, and enhanced to generate a set of fused feature vectors with correlation. The fused feature vectors and their relationships are organized in a graph structure, where nodes represent the fused situational elements and edges represent the influence relationships and strengths between elements. The graph structure is updated in real time to form the dynamic project comprehensive situational map.

7. The artificial intelligence-based waste-to-energy plant project management system according to claim 1, characterized in that, The improved Bayesian inference algorithm includes: Establish a dynamic Bayesian network structure with the comprehensive situational characteristics of the project as the observation variable; Initialize the prior probability distribution of each hidden node in the dynamic Bayesian network structure; During the reasoning process, the real-time feature data in the dynamic project comprehensive situation map is injected into the dynamic Bayesian network structure as new evidence. The real-time feature data includes design change frequency, schedule lag days and material supply delay warning level. Based on the conditional dependencies between nodes defined in the dynamic Bayesian network structure, the posterior probability distribution in the dynamic Bayesian network structure is iteratively approximated using the variational inference method. In each iteration of the approximation calculation, the state confidence of the hidden nodes associated with the key decision points is updated synchronously; When the change in the state confidence is less than the preset convergence threshold, the iteration stops, and the posterior probability distribution of the hidden nodes associated with the key decision point at this time is used as a measure of quantitative risk and opportunity to form the quantitative decision support information.

8. The artificial intelligence-based waste-to-energy plant construction project management system according to claim 1, characterized in that, Based on the quantitative decision support information, an adaptive project scheduling strategy, including task execution order adjustment schemes and resource reallocation schemes, is generated through the project scheduling optimization engine, including: In the project scheduling optimization engine, the project's task network, resource pool status, and the quantitative decision support information are taken as inputs; A multi-objective optimization model is established with the goal of minimizing the total project duration delay and resource conflicts; In the multi-objective optimization model, the quantitative decision support information is quantified into a time margin adjustment coefficient and a resource demand elasticity coefficient for a specific task. A heuristic search algorithm is used to solve the multi-objective optimization model. Under the premise of satisfying the task process logic constraints and total resource constraints, the algorithm explores the reordering space of task start time and the flow path of resources between different tasks. The new start and end times of all tasks are extracted from the solution results to form the task execution order adjustment scheme. At the same time, the changes in the allocation plan of various resources on the project timeline are extracted to form the resource reallocation scheme. Together, they constitute the adaptive project scheduling strategy.

9. The artificial intelligence-based waste-to-energy plant project management system according to claim 8, characterized in that, The step of solving the multi-objective optimization model using a heuristic search algorithm includes: Initialize a task scheduling sequence based on the current project baseline plan as the initial solution population; Define a fitness function, which is calculated by combining the total project duration, resource load balancing degree, and the buffering degree of high-risk tasks in the quantitative decision support information; For individuals in the initial solution population, new candidate solutions are generated by swapping task positions, inserting idle time, and adjusting resource allocation ratios. Calculate the fitness of the newly generated candidate solutions, and perform iterative population updates based on the fitness scores. During the population iteration process, a simulated annealing mechanism is introduced to accept solutions with poor fitness with a certain probability, thereby avoiding the search from getting trapped in local optima. When the preset maximum number of iterations is reached or the fitness improvement is continuously below the threshold, the search is terminated, and the individual with the highest fitness in the current population is taken as the final solution of the multi-objective optimization model.

10. The artificial intelligence-based waste-to-energy plant project management system according to claim 1, characterized in that, The adaptive project scheduling strategy is translated into a set of executable project management instructions, including: The task execution order adjustment scheme in the adaptive project scheduling strategy is analyzed and decomposed into task start time change notification and task prerequisite relationship update notification for specific construction teams or subcontractors. The resource reallocation scheme in the adaptive project scheduling strategy is analyzed and decomposed into material delivery plan change notifications for specific warehouses or suppliers and manpower allocation instructions for different types of work. Based on the standard application programming interface protocol of project management software, the task start time change notification, task prerequisite relationship update notification, material distribution plan change notification, and manpower allocation instruction are formatted into standardized data packets that can be directly read and executed by the project execution terminal. All standardized data packets are sorted and packaged in chronological order of their effective dates to form a complete set of project management instructions.

Images