A multi-objective intelligent decision support system for construction of a continuous bridge support across a sea

Abstract

The application provides a kind of continuous bridge support construction multi-objective intelligent decision support system across sea, comprising: construction period model generation module obtains data to calculate project total construction period to form construction period target optimization model;Cost model generation module is used to generate cost target optimization model;Safety model generation module extracts and aggregates the minimum value of the output safety level of the immediately preceding working node to generate a safety level target optimization model;Multi-objective model integration module combines the three and generates an initial multi-objective equilibrium optimization model combined with constraint conditions;Self-adaptive penalty configuration module evaluates the violation degree and corrects to generate a revised multi-objective equilibrium optimization model;Cross variation search module establishes a guide to recombine parent generation to generate new individuals;Iterative optimization and decision output module evaluates individuals to select elite archives and outputs solution set.The application solves the problem that existing models are difficult to achieve multi-objective adaptive equilibrium optimization of construction period, cost and safety under the premise of quantifying safety risk cascade effect.

Description

Technical Field

[0001] This invention belongs to the technical field of computer-aided design and relates to a multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports. Background Technology

[0002] Large-scale civil engineering construction, especially the support construction projects in deep-water and complex meteorological environments such as cross-sea continuous bridges, often faces the challenge of coordinating and optimizing multiple objectives, including minimizing the construction period, reducing financial costs, and maximizing construction safety. Because cross-sea continuous bridges typically face short meteorological windows, high maritime logistics and equipment costs, and high safety coupling between successive construction processes, there are complex constraints and competitive relationships among these objectives. Increasing resource input to shorten the construction period often leads to higher direct project costs and can easily introduce additional safety risks due to overly rapid transitions between processes. The intertwined constraints among these multi-dimensional objectives make it difficult for conventional decision-making models to achieve adaptive equilibrium optimization.

[0003] For multi-objective construction management problems, common technical solutions in the industry mainly include single-objective optimization methods, weighted summation-based transformation methods, and manual iteration methods. Single-objective optimization methods focus on extracting a single core indicator as the optimization function, while setting other key factors as fixed boundary constraints. Weighted summation-based transformation methods rely on assigning preset static numerical weights to aggregate multi-dimensional objectives into a single algebraic computation problem for solution. Manual iteration methods rely on personnel to conduct experience-based trial calculations and resource allocation adjustments based on the construction schedule to achieve solution convergence.

[0004] The aforementioned existing technologies have significant technical limitations in practical applications due to their inherent model processing mechanisms. Single-objective optimization methods, employing fixed boundary constraints, struggle to reflect the dynamic interaction between objectives, easily leading to improvement in one objective while deteriorating overall performance in other dimensions. Weighted summation-based transformation methods, constrained by static weight allocation mathematical mechanisms, lack the ability to adaptively explore new solution frontiers in the face of changing engineering environments, resulting in poor resource allocation adaptability. Manual iterative methods, when dealing with complex construction networks with multi-level logical dependencies, suffer from low computational efficiency and weak global search capabilities due to their experience-driven trial-and-error process, and struggle to accurately quantify the cascading transmission effects of safety risks in specific processes. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports.

[0006] A multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports includes: The project duration model generation module is used to obtain the project's work processes and construct a network diagram, extract structural path data, and calculate the longest time node of the path data as the total project duration to generate a project duration target optimization model. The cost model generation module is used to obtain the direct cost value and calculate the total indirect cost value of the project in combination with the total project duration. It compares the total project duration with the planned duration to establish a time incentive penalty function, and then generates a cost target optimization model. The safety model generation module is used to assign inherent safety level indices to engineering project processes, extract and aggregate the output safety level minimum values ​​of preceding work nodes to generate a safety level target optimization model. The multi-objective model integration module is used to combine the schedule target optimization model, cost target optimization model and safety level target optimization model, and combine them with comprehensive constraints to generate an initial multi-objective equilibrium optimization model. The adaptive penalty configuration module is used to evaluate the total violation characteristics of individuals in the current population to the constraints, and outputs the population dynamic penalty coefficient to correct the initial multi-objective equilibrium optimization model, thereby generating the corrected multi-objective equilibrium optimization model. The crossover and mutation search module is used to traverse the parent individuals in the population, establish a guide reference in the non-dominated solution set space of the internal cache initialized by the non-dominated sorting of the previous generation population, and use the guide reference to recombine the parent individuals to generate new population individuals. The iterative optimization and decision output module is used to evaluate new population individuals using the modified multi-objective equilibrium optimization model, select elite individuals and load them into a fixed-capacity external archive, and output the non-dominated solution matrix in the fixed-capacity external archive as the objective solution set after satisfying the maximum number of iterations.

[0007] In a further embodiment of the present invention, the project duration model generation module is specifically used for: Extracting all structural path data by traversing the network graph using the critical path method; Define the time measurement parameters for each process in the structure path data. The time measurement parameters include the shortest duration, the actual duration, and the longest allowed duration. The longest time node for extracting structural path data from the network graph by comprehensively aligning time metric parameters is set as the total project duration, generating a project duration target optimization model.

[0008] In a further embodiment of the present invention, the cost model generation module is specifically used for: Obtain the direct cost of each independent process in the actual project, and multiply the preset indirect cost rate by the total project duration to calculate the total indirect cost of the project; Based on the total project duration and compared with the preset planned duration, a time incentive penalty function is established to quantify the deviation from the planned duration using preset delay penalty coefficients and early reward coefficients.

[0009] In a further embodiment of the present invention, the security model generation module is specifically used for: Extract the preset minimum safety level index requirements and independently assign an initial inherent safety level index to each specific construction procedure; By tracing back to the immediate preceding work node connected to the current calculation process, the impact value of the output safety level intervention characteristic propagated backward from the immediate preceding work node is deduced and calculated. The inherent safety level index and the minimum output safety level of the preceding working node are mathematically aggregated using a multiplicative function to generate a safety level target optimization model.

[0010] In a further embodiment of the present invention, the multi-objective model integration module is specifically used for: The optimization models for project duration, cost, and safety level are combined as a joint evaluation function set to drive the optimization algorithm. The minimum and maximum time limits for each process, the total project cost budget, and the safety engineering investment range are set as pre-defined comprehensive constraints. By applying comprehensive constraint filtering within the framework of the joint evaluation function group and defining the feasible solution set that satisfies all constraints, an initial multi-objective equilibrium optimization model is obtained.

[0011] In a further embodiment of the present invention, the adaptive penalty configuration module is specifically used for: Identify the specific values ​​of the excesses of the current population individuals in the initial multi-objective equilibrium optimization model corresponding to the violation of various constraints of the project to be evaluated by the project, and calculate the total violation degree characteristics of the current entire population distribution ecology by weighted matching. Based on the preset penalty threshold, an adaptive constraint penalty strategy is activated for individuals that violate the comprehensive constraint conditions, and the population dynamic penalty coefficient is output. The initial multi-objective equilibrium optimization model is converted and corrected using the population dynamic penalty coefficient, and a corrected multi-objective equilibrium optimization model is generated.

[0012] In a further embodiment of the present invention, the crossover mutation search module is specifically used for: Retrieve and retrieve the current non-dominated solution set space in the internal cache state, and randomly select the feature individuals with the best level from it using the crowding degree descending extraction principle to establish them as the guide reference; Extract the parent individuals to be recombined, and use the guide reference to introduce a preset strategy mechanism to carry out sequence fragment mixing and mutation replacement recombination operations to generate new population individuals.

[0013] In a further embodiment of the present invention, the iterative optimization and decision output module is specifically used to evaluate and screen individuals in the new population, and includes: The modified multi-objective equilibrium optimization model was used to calculate and evaluate the fitness level vectors of individuals in the new population in batches. The elite individuals who are in the first priority non-dominant layer and belong to the strong feasible constraints are selected, sorted, and registered, and then loaded into the external archive with a fixed capacity to be established. The mutual dominance of new individuals merged into the fixed-capacity external archive and existing elite individuals in the fixed-capacity external archive is verified in order to identify and remove downgraded solutions.

[0014] In a further embodiment of the present invention, after performing the elimination of downgraded solutions, the iterative optimization and decision output module is also used for: Check if the number of individuals in the current fixed-capacity external archive exceeds the preset maximum capacity; When the number of individuals is detected to exceed the preset maximum capacity, the crowding distance metric is triggered to prune the last individuals to restore the reserved space in the fixed-capacity external archive.

[0015] In a further embodiment of the present invention, the iterative optimization and decision output module is also used for: Control the process of replacing parent individuals with new individuals in the population until the counter reaches the preset maximum number of iterations and then stops. Extract and display the full non-dominated solution matrix from the external archive of fixed capacity, and output it in a formatted format as the final target solution set to be selected.

[0016] In summary, the present invention has the following beneficial technical effects: 1. This invention constructs a schedule target optimization model and utilizes network graph structure path data based on the critical path method to extract and compare the cumulative time consumption of each path, thereby achieving quantitative identification of the decisive path and total time consumption under the current process duration configuration. This mechanism provides a logical mapping relationship between process time adjustments and total schedule changes for subsequent optimization operators, overcoming the logical chain breakage defect that easily occurs in traditional trial-and-error algorithms when retrospectively adjusting the time of a single process, and improving the accuracy of dynamic schedule evaluation.

[0017] 2. This invention constructs a safety level target optimization model and, using a multiplicative function incorporating a minimum value operator, aggregates the impact value of safety bottlenecks transmitted from preceding processes with the inherent safety level index of the current process. This simulates and quantifies the cascading risk transmission effect of the safety status of upstream processes spreading to downstream processes. This calculation method ensures that safety assessment indicators conform to the actual risk accumulation patterns in engineering projects, effectively avoiding the problem of illogical and meaningless geometrical overlap and amplification of single risks at the front end due to network topology cross-convergence. It provides numerical basis for intelligent optimization operations to filter potentially safety-limited combinations, helping to enhance the safety redundancy of the overall construction strategy.

[0018] 3. This invention introduces an evaluation mechanism based on adaptive constraint penalties when establishing a multi-objective model. It dynamically quantifies the total violation characteristics of candidate construction schemes against comprehensive constraints such as cost budget and bottom-line safety requirements, and adjusts the dynamic penalty coefficient of the population accordingly. This strategy guides the search process to spontaneously deviate from regions of severe constraint violations during the evaluation phase, while appropriately retaining candidate solutions that slightly exceed limits but possess advantages in other dimensions. While maintaining the global exploration space of the population, it optimizes the algorithm's convergence direction towards the feasible solution region.

[0019] 4. In the crossover and mutation search process, this invention employs a mechanism based on crowding-degree-based orientation extraction of guide references from the non-dominated solution set space. This guides new individuals to move towards the relatively sparse Pareto front distribution in the area to be explored. Combining exploration and utilization strategies helps maintain the diversity of the population distribution in the target space, reduces the probability of the optimization algorithm getting trapped in local optima, and outputs a full set of non-dominated solutions with high tolerance and distribution density. This provides reliable decision-making data support for multi-angle comparison of the final solution. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0021] Figure 1 This is a schematic diagram of the framework in the embodiments of this application.

[0022] Figure 2 This is a flowchart illustrating an embodiment of this application. Detailed Implementation

[0023] The following is in conjunction with the appendix Figures 1-2 A preferred description of the present invention is provided below.

[0024] See attached document Figures 1-2 This invention proposes a multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports, comprising the following modules: The project duration model generation module is used to obtain the project's work processes and construct a network diagram, extract structural path data, and calculate the longest time node of the path data as the total project duration to generate a project duration target optimization model. The cost model generation module is used to obtain the direct cost value and calculate the total indirect cost value of the project in combination with the total project duration. It compares the total project duration with the planned duration to establish a time incentive penalty function, and then generates a cost target optimization model. The safety model generation module is used to assign inherent safety level indices to engineering project processes, extract and aggregate the output safety level minimum values ​​of preceding work nodes to generate a safety level target optimization model. The multi-objective model integration module is used to combine the schedule target optimization model, cost target optimization model and safety level target optimization model, and combine them with comprehensive constraints to generate an initial multi-objective equilibrium optimization model. The adaptive penalty configuration module is used to evaluate the total violation characteristics of individuals in the current population to the constraints, and outputs the population dynamic penalty coefficient to correct the initial multi-objective equilibrium optimization model, thereby generating the corrected multi-objective equilibrium optimization model. The crossover and mutation search module is used to traverse the parent individuals in the population, establish a guide reference in the non-dominated solution set space of the internal cache initialized by the non-dominated sorting of the previous generation population, and use the guide reference to recombine the parent individuals to generate new population individuals. The iterative optimization and decision output module is used to evaluate new population individuals using the modified multi-objective equilibrium optimization model, select elite individuals and load them into a fixed-capacity external archive, and output the non-dominated solution matrix in the fixed-capacity external archive as the objective solution set after satisfying the maximum number of iterations.

[0025] In one embodiment of the present invention, the project duration model generation module is used to perform the following steps: Extracting all structural path data by traversing the network graph using the critical path method; Define the time measurement parameters for each process in the structure path data. The time measurement parameters include the shortest duration, the actual duration, and the longest allowed duration. The longest time node for extracting structural path data from the network graph by comprehensively aligning time metric parameters is set as the total project duration, generating a project duration target optimization model.

[0026] The processor executing a specific algorithm retrieves the engineering breakdown structure (EBSD) table of the bridge composite support construction project, stored in a local or cloud database. A bridge composite support construction project's work steps refer to the smallest manageable work units with clearly defined inputs, outputs, and execution entities, broken down from a complex bridge support project using an EBSD structure, such as foundation excavation, abutment concrete pouring, and pole erection. This table digitally records all independent construction steps of the project and their prerequisite dependencies.

[0027] The processor constructs a directed acyclic graph (DAG), or network graph, based on the project decomposition structure table and using graph theory data structures. In this embodiment, the network graph is a visualization tool for project management, specifically using an activity node graph representation. In this network graph, each process is abstracted as a node, and the logical relationship of the construction sequence between processes is represented as directed edges between nodes. Subsequently, the processor traverses the network graph using a depth-first search or breadth-first search algorithm, starting from the unique project start node, enumerating and recording all paths leading to the unique project end node, forming a set of structural path data. It should be noted that structural path data refers to the set of all process sequences from the start node to the end node of the network graph without duplicate nodes. Let's assume... Let be the total number of all independent structural paths in the network graph. Representing the A structural path is a series of interconnected process nodes. A set that consists of.

[0028] For each process node in the network graph, the processor loads and allocates three core time metrics from the associated engineering parameter database: these are the minimum physical or technologically permissible durations for that process. The ideal construction period achievable under unlimited resources or specific conditions is determined based on the technical specifications of the work process, the physicochemical reaction time of materials (such as the curing period of concrete), or the fastest construction record under extreme resource input. It is typically determined by the project's technical department based on historical data statistics and industry standards. Assuming a low-risk, conventional construction scenario, its value is 0.8 times the standard construction period; it also constitutes a variable for subsequent decision-making. The lower bound.

[0029] Secondly, the estimated actual duration under actual constraints and resource allocation. This parameter serves as a core decision variable in the optimization process, and its value range is strictly limited to [specific range]. and In between. Finally, there is the maximum permissible duration required by the contract or regulations. Exceeding this timeframe will trigger penalty clauses or lead to quality risks. Its setting is based on various factors such as contractual deadlines, buffer periods with subsequent processes, and cost control. For example, in cross-sea construction, the necessary work stoppage period before typhoons or high tides is typically 1.2 to 1.5 times the standard construction period, thus constituting a decision variable. The upper boundary.

[0030] The processor analyzes the network graph based on the critical path method's computational criteria, considering any given set of actual durations for each process. The total time for each structural path is obtained by summing the actual durations of all processes along that path. The maximum value among the total times for all paths is then identified and defined as the total project duration under the current schedule. As a supplement, the total project duration... This is the duration of the longest path among all structural paths; this path is the critical path of the project. Its mathematical expression is as follows:

[0031] Accordingly, the processor establishes a mathematical model with the goal of minimizing the total project duration, using the actual duration of each process as the decision variable, and constrained by the shortest duration and the longest allowable duration, thereby generating a project duration optimization model. The model is then output to memory as a structured data object for later use.

[0032] It should be noted that the project schedule target optimization model can be represented as follows: .

[0033] In one embodiment of the present invention, a specific implementation scenario is provided: Assume a simplified bridge support construction project includes five procedures, namely A (e.g., foundation and steel pipe pile construction), B (e.g., offshore pier construction), C (e.g., approach bridge support erection), D (e.g., main bridge continuous beam formwork construction), and E (e.g., full bridge closure). Their dependencies are as follows: A is the initial procedure; B and C both depend on the completion of A; D depends on the completion of B; and E depends on the completion of C and D.

[0034] The processor first constructs a network graph, forming two structural path data from start to finish. Path one is... Path 2 is Next, the processor loads preset time metrics, in days, for each process. The shortest duration for process A is... The maximum allowed duration is 2. It is 4. Process B. It is 4. It is 7. Process C. It is 5. It is 8. Process D. It is 3. It is 5. Process E. It is 2. The value is 3. In a specific optimization iteration, it is assumed that the model allocates the actual duration for each process. They are respectively: , , , , .

[0035] Then, the time consumption of each structural path was calculated. The total time consumption of path one was... Days. The total time for Path 2 is... The processor compares the time taken by the two paths and takes the maximum value as the total project duration under this scheme. At this point, path one becomes the critical path. Ultimately, the objective function... Decision variables and its constraints As a structured data package, it generates a project schedule target optimization model.

[0036] In one embodiment of the present invention, the cost model generation module is configured to perform the following steps: Obtain the direct cost of each independent process in the actual project, and multiply the preset indirect cost rate by the total project duration to calculate the total indirect cost of the project; Based on the total project duration and compared with the preset planned duration, a time incentive penalty function is established to quantify the deviation from the planned duration using preset delay penalty coefficients and early reward coefficients.

[0037] Based on the schedule target optimization model, the processor continues to execute the model building program to generate a cost target optimization model. The processor addresses each individual process in the network diagram. It queries the internally associated bill of quantities database using its unique identifier, retrieves and loads the pre-defined direct cost values ​​for each independent process. It should be noted here that the direct cost amount... It refers to something used directly in the process. The costs of materials (such as strong anti-corrosion coatings), labor, and construction machinery shifts (such as offshore operation shifts for floating cranes) are derived from the statistical averages of project budget documents or historical project cost databases.

[0038] Based on the global parameters set in memory, namely the indirect cost rate per unit of time. This rate covers expenses such as project management personnel salaries, office expenses, and depreciation of temporary facilities, which are not directly attributable to a single process but continue to occur over time. The value is typically set at 0.1 to 0.5 million yuan per day, based on corporate accounting standards and the specific project's management structure. The processor will use this rate... Multiply this by the total project duration T derived from the schedule target optimization model to calculate the total indirect cost of the project as the duration changes. Its calculation formula can be expressed as: .

[0039] To establish a dynamic time-based reward and penalty mechanism, the processor retrieves a pre-defined planned duration from a contract or project charter. As a benchmark, the planned duration is a key milestone clearly defined in the project contract, serving as a fixed reference value. The total project duration is calculated by comparing this target duration with the target duration optimization model. With the preset planned construction period The processor enables a time-incentive penalty function for quantization. .

[0040] When the total project duration exceeds the preset planned duration, the function will multiply the excess time by a preset delay penalty factor. If the project's total duration does not reach the preset planned duration, the function will multiply the amount of time saved by a preset advance bonus coefficient. This creates a negative cost incentive.

[0041] In this embodiment, the delay penalty coefficient With advance reward coefficient These are incentive clauses in the contract, set based on the potential economic losses or gains the owner may incur due to delayed or early delivery. For example, in bridge projects, delays can significantly impact traffic. The value is usually set relatively high, while The incentive penalty function is relatively low. The specific calculation formula is as follows:

[0042] Finally, the processor sums up the direct cost values ​​of all processes, adds them to the total indirect cost value of the project, and calculates the result of the time incentive penalty function to obtain the total project cost. The complete expression:

[0043] Building on this, the processor further introduces a maximum budget limit for the total project cost read from the project's financial planning documents. As an insurmountable hard constraint, and also the project's investment red line, it is determined by the owner's funding plan. The processor uses the expression for minimizing the total project cost as the objective, and the maximum budget limit as the constraint, ultimately integrating them into a complete mathematical optimization model, that is, generating the objective function as follows: The cost objective optimization model can be expressed as:

[0044] The processor then temporarily stores it in memory for later integration.

[0045] For example, continuing with the previous scenario, the total project duration... The actual duration of each process is 14 days. They are respectively sky, sky, sky, sky, Now, let's introduce the economic parameters for this step. Assume that, based on a database query, the direct cost values ​​for each process are... (Unit: 10,000 yuan) are as follows: , , , , The total direct cost is then... Ten thousand yuan. The set indirect cost rate per unit construction period. 10,000 yuan / day.

[0046] Therefore, the total indirect cost of the project for Ten thousand yuan. Assuming the planned construction period is preset in the contract. Heaven, delay penalty coefficient 10,000 yuan / day, advance bonus coefficient Ten thousand yuan / day. Processor computation time incentive penalty function value. The amount of 10,000 yuan indicates that a bonus of 20,000 yuan will be awarded for early completion.

[0047] Finally, the processor sums the results to arrive at the total project cost. Ten thousand yuan. If the maximum budget limit for the total project cost... If it is 1.5 million yuan, then The constraints are satisfied. The processor has thus generated the target as... , constraint is The cost target optimization model.

[0048] In one embodiment of the present invention, the security model generation module is configured to perform the following steps: Extract the preset minimum safety level index requirements and independently assign an initial inherent safety level index to each specific construction procedure; By tracing back to the immediate preceding work node connected to the current calculation process, the impact value of the output safety level intervention characteristic propagated backward from the immediate preceding work node is deduced and calculated. The inherent safety level index and the minimum output safety level of the preceding working node are mathematically aggregated using a multiplicative function to generate a safety level target optimization model.

[0049] The processor extracts the explicitly specified minimum safety level index requirements from the project's safety procedures or technical specifications. This index serves as a rigid constraint benchmark. It should be noted that this requirement can be determined comprehensively based on national standards such as the "Unified Technical Specification for Construction Safety," combined with the project's importance level and environmental sensitivity, and its value range is usually between 0.85 and 0.95.

[0050] Subsequently, the processor traverses each process node in the network graph. Based on a pre-set database of process safety risk levels, an initial inherent safety level index is assigned to each independent construction process. This index represents the theoretical safety state of the process when it is free from external interference. It should be understood that... This is a priori value derived from expert assessment or historical accident data statistics. It quantifies the inherent danger level of a specific work process (such as "working at height" or "deep foundation pit excavation"). The closer the value is to 1, the higher the inherent safety. Assuming standard operating procedures are followed, the general ground work process... It can be set to 0.98, while complex hoisting procedures involving special equipment may be set to 0.92.

[0051] To quantify the transmission effect of safety risks between contextual processes, the processor performs topological sorting of the network graph and calculates the actual output safety level for each process in sequence. This value combines the inherent risks of the process with the actual safety assessment value after the transmission of risks from upstream. For any currently calculated process... The processor first retrieves all its immediate predecessor working nodes in the network graph in reverse order (defined as a set). Based on the calculated output safety level value of the preceding working node. Its representative process The final safety status upon completion will be used as a safety risk input to affect all subsequent processes. The processor will deduce the cascading negative impacts that these nodes collectively exert on the current process, i.e., calculate the comprehensive output safety level intervention characteristic impact value.

[0052] This process is accomplished through a multiplicative function containing a minimum operator. This function mathematically aggregates the inherent safety level index of the current process with the combined safety impact transmitted from all preceding processes, ultimately generating the actual output safety level of the current process. Based on the barrel effect and the principle of risk transmission bottlenecks, to avoid the meaningless geometrical amplification of a single front-end risk due to network topology cross-convergence, the calculation is performed as follows:

[0053] For a starting operation that has no immediate preceding work nodes, For an empty set, define its immediate preceding security impact terms at this moment. The value of is 1, thus making All safety indices are dimensionless values ​​between 0 and 1.

[0054] In this embodiment, the significance of this function lies in the fact that the safety risks transmitted from the preceding process should follow a cumulative law of continuous attenuation. This represents the constraint imposed on the current node by the weakest safety vulnerability among all preceding processes. This calculation method ensures that if any high-risk (low safety index) node exists upstream, the safety level of the current process's output will be penalized and reduced, thus obtaining a true output safety level that reflects the objective reality of the project.

[0055] The processor must ensure that the output safety level of all processes exceeds the minimum safety level index requirement. Set as constraints, with the goal of maximizing the overall safety level of the project. Maximizing the overall safety level of the project is usually defined as all The lowest output safety level in each process is used to improve the overall safety and robustness of the project, thereby generating a representation as follows: Safety level target optimization model:

[0056] The model is also stored in memory as structured data, awaiting further integration.

[0057] For example, continuing with the aforementioned project case involving processes A, B, C, D, and E, the processor first sets the minimum safety level index requirements. Next, an inherent safety level index is assigned to each process, with the following values ​​set: , , , , The processor performs calculations in topological order. Operation A has no preceding nodes, and its output safety level is [not specified]. The value is greater than 0.85, thus satisfying the constraint.

[0058] The immediate predecessor of process B is A, and its output safety level is The value is greater than 0.85. The immediate predecessor node of process C is A, and its output safety level is... The value is greater than 0.85. The immediate predecessor node of process D is B, and its output safety level is... The value is greater than 0.85. The immediate predecessor nodes of process E are C and D, and its output safety level is... This value is also greater than 0.85. At this point, the safety levels of all processes meet the requirements. Under the current scheme, the overall safety level of the project is the minimum output safety level of all processes, which is 0.9033. The processor then generates the safety level target optimization model. Its goal is to maximize this minimum value.

[0059] In one embodiment of the present invention, the multi-objective model integration module is used to perform the following steps: The optimization models for project duration, cost, and safety level are combined as a joint evaluation function set to drive the optimization algorithm. The minimum and maximum time limits for each process, the total project cost budget, and the safety engineering investment range are set as pre-defined comprehensive constraints. By applying comprehensive constraint filtering within the framework of the joint evaluation function group and defining the feasible solution set that satisfies all constraints, an initial multi-objective equilibrium optimization model is obtained.

[0060] The three independent single-objective optimization models mentioned above are integrated into a unified multi-objective optimization framework. First, the processor integrates the schedule target optimization model... Cost target optimization model and safety level target optimization model By combining these three objective functions into a single vectorized evaluation function, we can form a set of joint evaluation functions. It is understandable that the joint evaluation function set... This is a method that maps multidimensional inputs (schedule vectors) to multidimensional outputs (schedule, cost, and safety target value vectors). The function set takes the actual duration vector of each process as input and sets the decision variables... The vector formed by the actual duration of all processes can simultaneously output a three-dimensional target value vector containing the total project duration, total project cost, and overall project safety level.

[0061] It should be noted that during the integration process... By taking the negative sign, the maximization problem is uniformly transformed into a minimization problem commonly used in optimization algorithms. The objective function of multi-objective optimization is expressed as:

[0062] Simultaneously, the various boundary conditions for bridge construction are systematically aggregated to form comprehensive constraints. Specifically, the processor logically aggregates constraints such as the minimum and maximum allowable durations set for each process, the maximum budget limit for the total project cost, and the specified minimum safety level index requirements. These comprehensive constraints are a collection of constraints from all the aforementioned single-objective models, collectively defining the effective region in the search space.

[0063] Finally, the processor combines the joint evaluation function set with the comprehensive constraints, mathematically defining a process guided by minimizing the multidimensional objective vector, where all solutions must satisfy all constraints. The optimization problem involves all combinations of decision variables that satisfy the comprehensive constraints, which are then included in the feasible solution set. The scope. In this embodiment, the feasible solution set. Specifically defined as:

[0064] This definition allows us to construct an initial multi-objective equilibrium optimization model that includes a complete objective function and constraint boundaries. This is the fully defined multi-objective optimization problem described above, encompassing the objective to be optimized and the rules that must be followed. It should be understood that the dimensions remain independent in this multi-objective representation and are ultimately processed by the optimization algorithm. The processor stores this model in its cache as a standardized data structure, providing a clear solution objective and search space for subsequent algorithmic optimization.

[0065] For example, the processor continues to construct the initial multi-objective equilibrium optimization model based on the sampled calculation results from the aforementioned steps. In the aforementioned example, when the decision variable vector... At that time, the processor had already calculated the total project duration. Today, the total project cost The target value is 10,000 yuan, and the overall security level of the project (determined by the weakest link) is 0.9033. The processor combines these three values ​​into a target vector. To unify it into a minimization problem, the processor represents it as follows: .

[0066] Next, the processor verifies whether the decision variable belongs to the feasible solution set. First, it checks the schedule constraint: exist Inside, exist Inside, exist Inside, exist Inside, exist Within this period, all project durations meet the constraints. Next, check the cost constraints: the calculated total cost. The amount of 10,000 yuan is less than the maximum budget limit for the total project cost. The amount is 10,000 yuan, meeting the constraints. Finally, the safety constraints were checked: the output safety level index of all processes was higher than the minimum safety level index requirement. The constraints are satisfied. Since all constraints are satisfied, the decision vector... This is a feasible solution. The processor will eventually contain the objective function. The complete definition of all the above constraints is encapsulated into an initial multi-objective equilibrium optimization model, which is then used as input to prepare for passing to the improved optimization algorithm for solving.

[0067] In one embodiment of the present invention, the adaptive penalty configuration module is configured to perform the following steps: Identify the specific values ​​of the excesses of the current population individuals in the initial multi-objective equilibrium optimization model corresponding to the violation of various constraints of the project to be evaluated by the project, and calculate the total violation degree characteristics of the current entire population distribution ecology by weighted matching. Based on the preset penalty threshold, an adaptive constraint penalty strategy is activated for individuals that violate the comprehensive constraint conditions, and the population dynamic penalty coefficient is output. The initial multi-objective equilibrium optimization model is converted and corrected using the population dynamic penalty coefficient, and a corrected multi-objective equilibrium optimization model is generated.

[0068] The runtime environment for subsequent optimization algorithms is configured, and a dynamic evaluation mechanism capable of handling constraint conflicts is established. The processor first executes a population initialization procedure. Based on a preset population size parameter, the extreme value of which is an integer determining the parallel search width of the algorithm, and given the complexity of the engineering problem, its value is typically set between 100 and 300. A set of random initial solutions, i.e., the parent population, is generated within the decision space corresponding to the feasible solution set defined by the initial multi-objective equilibrium optimization model. Each individual represents a specific vector of actual process durations, and its generation process follows the constraints of the shortest and longest allowed durations for each process. Simultaneously, the processor loads the algorithm's start / stop control parameters, i.e., the maximum number of iterations, which is the main parameter controlling the algorithm's running depth, typically set to 200 to 500 iterations.

[0069] Subsequently, for each individual in the population The processor begins the constraint violation assessment. It reads each comprehensive constraint defined in the initial multi-objective equilibrium optimization model, including the maximum budget limit for the total project cost and the minimum safety level index requirement. For any given entity, the processor uses its duration vector to calculate the corresponding total project cost and the output safety level of each process, and compares this with the constraint boundaries to obtain the specific value of each constraint exceeding its limit. For example, for cost constraints, the out-of-limit value is... .

[0070] To ensure dimensional consistency in multidimensional calculations and eliminate numerical overload caused by unit differences, specific values ​​exceeding the limits are addressed. The processor extracts the corresponding constraint reference values. Such as the total project cost Or the minimum safety index requirement Perform ratio conversion. The scaled dimensionless out-of-limit deviation rate is then normalized using preset weights. By performing a weighted summation, the processor obtains the total violation feature corresponding to that individual. Among them, normalized weights The sum of these values ​​is 1, and their specific values ​​reflect the decision-maker's emphasis on different constraints. Because ratio scaling is used, this helps to truly balance constraint violations of different dimensions; and the final calculated pure numerical total violation characteristic... This combines multi-dimensional constraint violations into a single scalar metric, facilitating uniform punishment. The specific calculation formula is as follows:

[0071] Next, the processor enables an adaptive constraint penalty strategy. It compares the total violation characteristic calculated for each individual with the system's preset penalty threshold. A comparison is needed. It should be understood that the penalty threshold... Used to distinguish between minor and serious violations, this is the system's upper tolerance limit for constraint violations. Its setting is based on experience, and can be, for example, 1% of the typical cost constraint exceedance value. If the threshold is not exceeded, a smaller base penalty factor is assigned to it. ( (This is a constant greater than 0, for example, set to 10); if the threshold is exceeded, a significantly increased penalty value is calculated using an exponential growth function, subject to the exponential growth factor. The nonlinear growth intensity of the penalty is controlled, typically ranging from 0.5 to 2.0, to reflect the rate at which the penalty intensity increases with the degree of violation. This generates a unique population dynamic penalty coefficient for each individual. Its specific judgment and calculation logic can be expressed as a formula:

[0072] The processor uses a population dynamics penalty coefficient to modify the initial multi-objective equilibrium optimization model. Specifically, a dimensional mapping vector with a baseline magnitude of the objective dimension is introduced. This corresponds to the true upper limit of tolerance for each objective item. The total violation feature... With population dynamics penalty coefficient The product of these factors serves as a comprehensive penalty factor to map the dimensional vector. Perform scalar stretching to generate a penalty vector carrying engineering units, which is then added to the original objective function vector. This is applied to each component, thus forming a penalized fitness evaluation model, i.e., generating a modified multi-objective equilibrium optimization model. This uniformly increases the overall objective function value of the default solution, making it worse across all dimensions. The calculation formula is as follows:

[0073] The modified model makes solutions that violate constraints disadvantaged in fitness evaluation, and the disadvantage worsens sharply with the degree of violation, thereby guiding the search process to converge toward the region of feasible solutions.

[0074] For example, following the setup of the previous steps, assume a maximum budget limit for the total project cost. The minimum safety level index requirement is 10,000 yuan. Now, a single individual is randomly selected from the initial population. The corresponding project duration vector is [4, 6, 8, 5, 3]. The model yields a total project duration of T = 18 days, a total project cost of C = 1.04 million yuan, and an overall safety level of 0.9033. This individual is the only one to violate the cost constraint, and its specific excess value is... Ten thousand yuan.

[0075] Introducing Constraint Reference Parameters Normalization is performed. Since no safety constraints were triggered, the corresponding term is 0. Assume the weight of the cost constraint. Then the dimensionless total violation characteristic of this individual The set penalty threshold This means tolerating a 2% over-limit, which is the base penalty factor. Exponential growth factor .because The processor employs an exponential penalty strategy to calculate the population dynamic penalty coefficient for that individual. .

[0076] Finally, the processor generates a revised multi-objective equilibrium optimization model. For this individual... The original target vector is [18, 104, -0.9033]. Let the preset upper order of magnitude parameter for dimension mapping vector extraction be... First, calculate the comprehensive penalty factor as follows: The resulting penalty vector is Its corrected target vector is For minimizing the safety metric -S, this correction operation worsens its value from an advantageous -0.9033 to -0.4953. At this point, the three objective dimensions are not only uniformly penalized for violating constraints but also retain their original physical meaning. The processor outputs this corrected objective vector to the subsequent filtering module, which marks it as a degraded solution according to the non-dominated sorting rule and performs a removal operation.

[0077] In one embodiment of the present invention, the crossover mutation search module is configured to perform the following steps: Retrieve and retrieve the current non-dominated solution set space in the internal cache state, and randomly select the feature individuals with the best level from it using the crowding degree descending extraction principle to establish them as the guide reference; Extract the parent individuals to be recombined, and use the guide reference to introduce a preset strategy mechanism to carry out sequence fragment mixing and mutation replacement recombination operations to generate new population individuals.

[0078] First, iterate through every individual in the current population, i.e., the parent individuals. This represents the surviving members of the population at the end of the previous iteration, with each individual encoding a complete process duration scheme. By generating new child individuals for each parent individual, a child population entity array equal in size to the parent population is ultimately formed, which is the new population output.

[0079] When generating offspring for a specific parent individual, the processor first retrieves the external archive of the current target solution set, i.e., the current non-dominated solution set space, which consists of all solutions not dominated by any other solution and represents an approximation of the Pareto optimal front. All non-dominated solutions are then retrieved. The processor calculates the crowding degree of each individual in the non-dominated solution set and sorts them in descending order. Subsequently, the processor extracts characteristic individuals from the top-ranked non-dominated solutions using either roulette wheel selection or random selection, establishing them as guide references for this genetic evolution. It is an elite individual selected from the non-dominated solution set, whose role is to guide the population to evolve towards the optimal region. Next, the processor copies the decision variable vector of the parent individual, i.e., the ontology sequence, which is the position vector of the individual, and in this project, the vector of the actual duration of each process, into a computational copy.

[0080] Based on the algorithm's pre-defined exploration-utilization balance strategy, usually determined by a random number or the ratio of the current iteration number to the maximum iteration number, the processor determines whether the current iteration should perform the exploration phase (sequence fragment mixing) or the utilization phase (mutation, replacement, and recombination).

[0081] If the exploration phase is executed, the processor randomly selects another individual from the parent population as the target for simulation. At this point, the attraction information of the guide reference will be discarded to encourage the population to perform a global divergent search. This is achieved through vector differencing and random scaling operations, involving random numbers drawn from a uniform distribution of [0, 1]. and an integer randomly selected from {1, 2} The computational copy is updated to generate new locations that tend to explore in the direction of a random target.

[0082] If the exploitation phase is executed, the processor simulates the random step size of Levy's flight trajectory. This guides the computational copy to perform a precise local search in the direction of the guide's reference point. The Levy flight term is influenced by the Levy index. The value of the control is usually determined based on a large number of numerical experiments, and generally ranges from 0.3 to 1.99. In this embodiment, it is set to 1.5. Calculation step size. It also involves random numbers drawn from the standard normal distribution. and and related constants .

[0083] By alternating between these two strategies, the processor updates the computational replicas, generating a new population of individuals with a higher probability of achieving better performance. That is, in the first The new offspring generated in the next iteration are stored in the offspring population entity array. The specific iterative formula for the above search rule can be uniformly expressed as:

[0084] Among them, the Levi Flight Project and its internal constants The calculation method is as follows:

[0085]

[0086] This process is repeated until all parent individuals have completed their reproductive operations.

[0087] For example, suppose the current processor is processing a parent instance. Its ontology sequence (schedule vector) is a feasible solution in the multi-objective model integration module. The processor first selects a guide reference in the current non-dominated solution set space. Assuming that after crowding sorting and random selection, the selected... The project duration vector is The processor then determines the current action. Assuming the current iteration is in the exploration phase, since this phase does not require the use of facilitator information for iteration, the processor simply randomly selects another entity as the target. Its construction period vector is The processor draws random numbers from a preset probability distribution. Assuming that in this calculation... random integer The processor performs calculations based on the formula used in the exploration phase: first, it calculates the difference term; Then multiply by a random scaling factor. : Finally, add the duration vector of the child individual to the duration vector of the parent individual: Since some time parameters exceeded the minimum construction constraints, the processor immediately triggered a boundary absorption operation, which forcibly reset the operations with durations less than the minimum to the minimum duration limit value (i.e., operation A is corrected to 2, operation E is corrected to 2), and operations with durations greater than the maximum allowable duration are reset to the upper limit value. After boundary safety correction, a new population of individuals that finally passed the upper and lower limit checks of the project duration are generated and passed to subsequent steps to evaluate their performance.

[0088] In one embodiment of the present invention, since continuous numerical values ​​(including decimals and out-of-bounds negative values) generated by variations in the purely mathematical environment lack executable physical meaning in actual project scheduling, and to ensure that the correction process does not cause the model to fail, the processor introduces an independently running engineering semantic discretization and feasible region repair program here: First, the engineering precision discretization operation is performed: Given that the actual scheduling of bridge projects often uses 0.5 days or 1 day as the minimum resource allocation unit, the processor uses rounding and other nearest-neighbor algorithms to map the initial feature vector containing irrational decimals into the actual engineering variables. Assuming that the minimum scheduling precision of the project is preset to 0.5 days, the discretized schedule vector will be [1.5, 2.5, 3.0, 2.0, 1.5].

[0089] Secondly, bidirectional boundary absorption protection is implemented to correct illegal solutions: the processor extracts the preset shortest duration of each process in the structural path data. With the longest allowed time The corresponding hard boundary constraints are applied, and the discretized vectors are checked one by one. Values ​​below the lower red limit are forcibly shifted upwards. Values ​​exceeding the upper limit will be cut off and throttled. For example, based on the aforementioned settings, the minimum durations for processes A through E are set to 2, 4, 5, 3, and 2 days, respectively. Therefore, for the discretized vector [1.5, 2.5, 3.0, 2.0, 1.5], the processor will forcibly reposition all values ​​below the constraint lower limit to the corresponding lower limit value. After repair and reconstruction, a valid sub-generation duration vector is generated. [2.0, 4.0, 5.0, 3.0, 2.0].

[0090] Finally, it should be noted that the decision variables in this algorithm model are the static relative durations of each process, rather than absolute start and end times with calendar coordinates. Therefore, boundary-independent absorption of a single parameter will not cause network logic deadlocks or topological breaks due to strong predecessor / successor dependencies. This legitimate, complete, independent duration vector, after being passed to the underlying layer, will be re-inferred forward by the multi-objective model in conjunction with network logic to reflect the actual changes in the overall project duration. After the above corrections, the final output of the new population effectively balances multi-dimensional search performance and engineering physical feasibility, and is then fed into the subsequent evaluation module for operation.

[0091] In one embodiment of the present invention, the iterative optimization and decision output module is used to perform the following steps: The modified multi-objective equilibrium optimization model was used to calculate and evaluate the fitness level vectors of individuals in the new population in batches. The elite individuals who are in the first priority non-dominant layer and belong to the strong feasible constraints are selected, sorted, and registered, and then loaded into the external archive with a fixed capacity to be established. The mutual dominance of new individuals merged into the fixed-capacity external archive and existing elite individuals in the fixed-capacity external archive is verified in order to identify and remove downgraded solutions.

[0092] First, for each individual in the new population, the modified multi-objective equilibrium optimization model generated in the adaptive penalty configuration module is called one by one to calculate the three-dimensional fitness level vector of each new individual. That is, the three-dimensional feature vector of each new individual, including the project period, cost, and safety evaluation value, and corrected by the penalty function, is calculated.

[0093] The processor merges the parent population with the new population to form a temporary large population. The processor then executes a fast non-dominated sorting algorithm on this large population, dividing it into multiple hierarchical levels ordered by dominance. The first non-dominated level contains all optimal solutions and is the current best approximation of the Pareto front, where no individual is completely outperformed by any other individual in this level across all objectives. Within the first non-dominated level, the processor further selects elite individuals that fully satisfy the strong feasible constraints of the comprehensive constraints. These individuals not only perform well in the objective space but also fully comply with all engineering constraints in the decision space.

[0094] The processor loads these elite individuals into a fixed-capacity storage structure, namely a fixed-capacity external archive. This archive serves as a reservoir of elite solutions, with a preset size, such as 100, to store all the optimal solutions found by the algorithm so far. Next, the processor executes an archive maintenance procedure. Whenever a new elite individual is added to the fixed-capacity external archive, the processor immediately performs pairwise dominance verification between all new individuals in the archive and existing elite individuals in the fixed-capacity external archive, that is, it determines whether solution A is not inferior to solution B on all objectives, and is superior to solution B on at least one objective. If it is found that an old solution in the archive is dominated by a newly added solution, the old solution will be determined as a downgraded solution, that is, the old solution replaced by the newly found better solution, and removed from the archive.

[0095] After completing the dominance update, the processor checks the current capacity of the archive. If it finds that the number of individuals exceeds the preset fixed capacity, that is, if it detects that the number of elite individuals in the archive outside the fixed capacity exceeds the preset maximum capacity, the processor triggers the crowding distance metric to calculate the crowding distance of all existing individuals in the archive. That is, it calculates the average distance between each solution and its neighboring solutions at the same level in its objective function space, which is used to evaluate the sparsity of the solution distribution. The processor restores the reserved archive space by pruning the last individuals with the smallest crowding distance. These individuals are usually located in areas with high solution density.

[0096] Based on the non-dominated sorting results and crowding distance, the processor selects the same number of individuals as the parent population from the merged large population to form the new generation parent population, completing one full iteration. This iteration process is repeated until the iteration counter reaches the preset maximum number of iterations. This variable is a key parameter controlling the total running time of the algorithm and has been defined in the adaptive penalty configuration module. The processor stops calculating and outputs the full set of non-dominated solutions stored in the fixed-capacity external archive in a formatted matrix. This full set of non-dominated solutions is usually an N-row, 3-column matrix, where N is the number of final solutions in the archive, each row represents an optimal construction plan, and the three columns correspond to its construction period, cost, and safety indicators, respectively, to form the final set of target solutions for final decision-making.

[0097] For example, suppose that in one iteration, the processor evaluates a new population individual whose time vector is... Calculations show that this individual is a feasible solution, and its objective vector is... The processor compares it with existing solutions in a fixed-capacity external archive. Assume there exists a solution A in the archive with an objective vector of... Since the new solution outperforms solution A in all three objectives—time (14.1 < 14.5), cost (95 < 98), and safety (-0.88 < -0.87, i.e., a higher level of safety)—solution A is dominated by the new solution, is classified as a downgraded solution, and is removed from the archive. The new solution is then added to a fixed-capacity external archive.

[0098] At this point, assuming the archive capacity is full, reaching 100 solutions, the processor will immediately activate the congestion distance metric to calculate the congestion distance of all 101 solutions in the archive. Suppose that after calculation, a solution B is found to have a very dense surrounding area, resulting in the smallest congestion distance value of 0.02. The congestion distances of all other solutions are greater than this value. The processor will then select solution B to be pruned, deleting it from the archive and restoring the archive size to 100. This iterative process continues until the preset number of iterations reaches 500. At this point, the processor stops computation and outputs the final 100 non-dominated solutions in the archive. Multiple sets of optimal construction scheme combinations are then formatted and output using the full non-dominated solution matrix, for example, an output vector of [14.0, 81.5, -0.91], thereby generating decision support instructions covering parameters such as total project duration, project cost, and safety level.

[0099] Each of the modules can be implemented in whole or in part through software, hardware, or a combination thereof. It supports hardware embedded in or independent of the processor in the computer device, and also supports software stored in the memory of the computer device, so that the processor can call and execute the operations corresponding to each of the above modules.

[0100] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports, characterized in that, include: The project duration model generation module is used to obtain the project's work processes and construct a network diagram, extract structural path data, and calculate the longest time node of the path data as the total project duration to generate a project duration target optimization model. The cost model generation module is used to obtain the direct cost value and calculate the total indirect cost value of the project in combination with the total project duration. It compares the total project duration with the planned duration to establish a time incentive penalty function, and then generates a cost target optimization model. The safety model generation module is used to assign inherent safety level indices to engineering project processes, extract and aggregate the output safety level minimum values ​​of preceding work nodes to generate a safety level target optimization model. The multi-objective model integration module is used to combine the schedule target optimization model, cost target optimization model and safety level target optimization model, and combine them with comprehensive constraints to generate an initial multi-objective equilibrium optimization model. The adaptive penalty configuration module is used to evaluate the total violation characteristics of individuals in the current population to the constraints, and outputs the population dynamic penalty coefficient to correct the initial multi-objective equilibrium optimization model, thereby generating the corrected multi-objective equilibrium optimization model. The crossover and mutation search module is used to traverse the parent individuals in the population, establish a guide reference in the non-dominated solution set space of the internal cache initialized by the non-dominated sorting of the previous generation population, and use the guide reference to recombine the parent individuals to generate new population individuals. The iterative optimization and decision output module is used to evaluate new population individuals using the modified multi-objective equilibrium optimization model, select elite individuals and load them into a fixed-capacity external archive, and output the non-dominated solution matrix in the fixed-capacity external archive as the objective solution set after satisfying the maximum number of iterations.

2. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The project duration model generation module is specifically used for: Extracting all structural path data by traversing the network graph using the critical path method; Define the time measurement parameters for each process in the structure path data. The time measurement parameters include the shortest duration, the actual duration, and the longest allowed duration. The longest time node for extracting structural path data from the network graph by comprehensively aligning time metric parameters is set as the total project duration, generating a project duration target optimization model.

3. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The cost model generation module is specifically used for: Obtain the direct cost of each independent process in the actual project, and multiply the preset indirect cost rate by the total project duration to calculate the total indirect cost of the project; Based on the total project duration and compared with the preset planned duration, a time incentive penalty function is established to quantify the deviation from the planned duration using preset delay penalty coefficients and early reward coefficients.

4. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The security model generation module is specifically used for: Extract the preset minimum safety level index requirements and independently assign an initial inherent safety level index to each specific construction procedure; By tracing back to the immediate preceding work node connected to the current calculation process, the impact value of the output safety level intervention characteristic propagated backward from the immediate preceding work node is deduced and calculated. The inherent safety level index and the minimum output safety level of the preceding working node are mathematically aggregated using a multiplicative function to generate a safety level target optimization model.

5. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The multi-objective model integration module is specifically used for: The optimization models for project duration, cost, and safety level are combined as a joint evaluation function set to drive the optimization algorithm. The minimum and maximum time limits for each process, the total project cost budget, and the safety engineering investment range are set as pre-defined comprehensive constraints. By applying comprehensive constraint filtering within the framework of the joint evaluation function group and defining the feasible solution set that satisfies all constraints, an initial multi-objective equilibrium optimization model is obtained.

6. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The adaptive penalty configuration module is specifically used for: Identify the specific values ​​of the excesses of the current population individuals in the initial multi-objective equilibrium optimization model corresponding to the violation of various constraints of the project to be evaluated by the project, and calculate the total violation degree characteristics of the current entire population distribution ecology by weighted matching. Based on the preset penalty threshold, an adaptive constraint penalty strategy is activated for individuals that violate the comprehensive constraint conditions, and the population dynamic penalty coefficient is output. The initial multi-objective equilibrium optimization model is converted and corrected using the population dynamic penalty coefficient, and a corrected multi-objective equilibrium optimization model is generated.

7. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The crossover and mutation search module is specifically used for: Retrieve and retrieve the current non-dominated solution set space in the internal cache state, and randomly select the feature individuals with the best level from it using the crowding degree descending extraction principle to establish them as the guide reference; Extract the parent individuals to be recombined, and use the guide reference to introduce a preset strategy mechanism to carry out sequence fragment mixing and mutation replacement recombination operations to generate new population individuals.

8. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The iterative optimization and decision output module is specifically used to evaluate and screen individuals in the new population, and includes: The modified multi-objective equilibrium optimization model was used to calculate and evaluate the fitness level vectors of individuals in the new population in batches. The elite individuals who are in the first priority non-dominant layer and belong to the strong feasible constraints are selected, sorted, and registered, and then loaded into the external archive with a fixed capacity to be established. The mutual dominance of new individuals merged into the fixed-capacity external archive and existing elite individuals in the fixed-capacity external archive is verified in order to identify and remove downgraded solutions.

9. A multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 8, characterized in that, After performing the removal of downgraded solutions, the iterative optimization and decision output module is also used for: Check if the number of individuals in the current fixed-capacity external archive exceeds the preset maximum capacity; When the number of individuals is detected to exceed the preset maximum capacity, the crowding distance metric is triggered to prune the last individuals to restore the reserved space in the fixed-capacity external archive.

10. The multi-objective intelligent decision support system for the construction of cross-sea continuous bridge supports according to claim 1, characterized in that, The iterative optimization and decision output module is also used for: Control the process of replacing parent individuals with new individuals in the population until the counter reaches the preset maximum number of iterations and then stops. Extract and display the full non-dominated solution matrix from the external archive of fixed capacity, and output it in a formatted format as the final target solution set to be selected.

Images